diff --git "a/log/log-train-2023-02-05-17-58-35-3" "b/log/log-train-2023-02-05-17-58-35-3" new file mode 100644--- /dev/null +++ "b/log/log-train-2023-02-05-17-58-35-3" @@ -0,0 +1,24969 @@ +2023-02-05 17:58:35,366 INFO [train.py:973] (3/4) Training started +2023-02-05 17:58:35,367 INFO [train.py:983] (3/4) Device: cuda:3 +2023-02-05 17:58:35,412 INFO [train.py:992] (3/4) {'best_train_loss': inf, 'best_valid_loss': inf, 'best_train_epoch': -1, 'best_valid_epoch': -1, 'batch_idx_train': 0, 'log_interval': 50, 'reset_interval': 200, 'valid_interval': 3000, 'feature_dim': 80, 'subsampling_factor': 4, 'warm_step': 2000, 'env_info': {'k2-version': '1.23.3', 'k2-build-type': 'Debug', 'k2-with-cuda': True, 'k2-git-sha1': '3b81ac9686aee539d447bb2085b2cdfc131c7c91', 'k2-git-date': 'Thu Jan 26 20:40:25 2023', 'lhotse-version': '1.9.0.dev+git.97bf4b0.dirty', 'torch-version': '1.10.0+cu102', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'surt', 'icefall-git-sha1': 'b3d0d34-dirty', 'icefall-git-date': 'Sat Feb 4 14:53:48 2023', 'icefall-path': '/exp/draj/mini_scale_2022/icefall', 'k2-path': '/exp/draj/mini_scale_2022/k2/k2/python/k2/__init__.py', 'lhotse-path': '/exp/draj/mini_scale_2022/lhotse/lhotse/__init__.py', 'hostname': 'r7n07', 'IP address': '10.1.7.7'}, 'world_size': 4, 'master_port': 12354, 'tensorboard': True, 'num_epochs': 30, 'start_epoch': 1, 'start_batch': 0, 'exp_dir': PosixPath('pruned_transducer_stateless7_streaming/exp/v1'), 'bpe_model': 'data/lang_bpe_500/bpe.model', 'base_lr': 0.05, 'lr_batches': 5000, 'lr_epochs': 3.5, 'context_size': 2, 'prune_range': 5, 'lm_scale': 0.25, 'am_scale': 0.0, 'simple_loss_scale': 0.5, 'seed': 42, 'print_diagnostics': False, 'inf_check': False, 'save_every_n': 2000, 'keep_last_k': 10, 'average_period': 200, 'use_fp16': True, 'num_encoder_layers': '2,2,2,2,2', 'feedforward_dims': '768,768,768,768,768', 'nhead': '8,8,8,8,8', 'encoder_dims': '256,256,256,256,256', 'attention_dims': '192,192,192,192,192', 'encoder_unmasked_dims': '192,192,192,192,192', 'zipformer_downsampling_factors': '1,2,4,8,2', 'cnn_module_kernels': '31,31,31,31,31', 'decoder_dim': 512, 'joiner_dim': 512, 'short_chunk_size': 50, 'num_left_chunks': 4, 'decode_chunk_len': 32, 'full_libri': True, 'manifest_dir': PosixPath('data/manifests'), 'max_duration': 500, 'bucketing_sampler': True, 'num_buckets': 30, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'on_the_fly_feats': False, 'shuffle': True, 'drop_last': True, 'return_cuts': True, 'num_workers': 2, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True, 'input_strategy': 'PrecomputedFeatures', 'blank_id': 0, 'vocab_size': 500} +2023-02-05 17:58:35,412 INFO [train.py:994] (3/4) About to create model +2023-02-05 17:58:36,053 INFO [zipformer.py:402] (3/4) At encoder stack 4, which has downsampling_factor=2, we will combine the outputs of layers 1 and 3, with downsampling_factors=2 and 8. +2023-02-05 17:58:36,065 INFO [train.py:998] (3/4) Number of model parameters: 20697573 +2023-02-05 17:58:51,146 INFO [train.py:1013] (3/4) Using DDP +2023-02-05 17:58:51,428 INFO [asr_datamodule.py:420] (3/4) About to get the shuffled train-clean-100, train-clean-360 and train-other-500 cuts +2023-02-05 17:58:52,644 INFO [asr_datamodule.py:224] (3/4) Enable MUSAN +2023-02-05 17:58:52,645 INFO [asr_datamodule.py:225] (3/4) About to get Musan cuts +2023-02-05 17:58:54,366 INFO [asr_datamodule.py:249] (3/4) Enable SpecAugment +2023-02-05 17:58:54,366 INFO [asr_datamodule.py:250] (3/4) Time warp factor: 80 +2023-02-05 17:58:54,366 INFO [asr_datamodule.py:260] (3/4) Num frame mask: 10 +2023-02-05 17:58:54,366 INFO [asr_datamodule.py:273] (3/4) About to create train dataset +2023-02-05 17:58:54,366 INFO [asr_datamodule.py:300] (3/4) Using DynamicBucketingSampler. +2023-02-05 17:58:54,386 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-05 17:58:56,573 INFO [asr_datamodule.py:316] (3/4) About to create train dataloader +2023-02-05 17:58:56,573 INFO [asr_datamodule.py:430] (3/4) About to get dev-clean cuts +2023-02-05 17:58:56,574 INFO [asr_datamodule.py:437] (3/4) About to get dev-other cuts +2023-02-05 17:58:56,575 INFO [asr_datamodule.py:347] (3/4) About to create dev dataset +2023-02-05 17:58:56,929 INFO [asr_datamodule.py:364] (3/4) About to create dev dataloader +2023-02-05 17:59:06,777 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-05 17:59:11,983 INFO [train.py:901] (3/4) Epoch 1, batch 0, loss[loss=7.184, simple_loss=6.498, pruned_loss=6.838, over 7431.00 frames. ], tot_loss[loss=7.184, simple_loss=6.498, pruned_loss=6.838, over 7431.00 frames. ], batch size: 17, lr: 2.50e-02, grad_scale: 2.0 +2023-02-05 17:59:11,983 INFO [train.py:926] (3/4) Computing validation loss +2023-02-05 17:59:24,177 INFO [train.py:935] (3/4) Epoch 1, validation: loss=6.888, simple_loss=6.229, pruned_loss=6.575, over 944034.00 frames. +2023-02-05 17:59:24,178 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6171MB +2023-02-05 17:59:31,384 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=3.87 vs. limit=2.0 +2023-02-05 17:59:37,909 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-05 17:59:48,688 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=5.39 vs. limit=2.0 +2023-02-05 17:59:55,491 INFO [train.py:901] (3/4) Epoch 1, batch 50, loss[loss=1.185, simple_loss=1.048, pruned_loss=1.223, over 7546.00 frames. ], tot_loss[loss=2.18, simple_loss=1.971, pruned_loss=2.007, over 368825.90 frames. ], batch size: 18, lr: 2.75e-02, grad_scale: 0.25 +2023-02-05 17:59:56,195 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:00:07,031 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3845, 4.3773, 4.3454, 4.3800, 4.3797, 4.3874, 4.3868, 4.3708], + device='cuda:3'), covar=tensor([0.0022, 0.0027, 0.0016, 0.0022, 0.0032, 0.0022, 0.0017, 0.0014], + device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0013, 0.0013, 0.0014, 0.0014, 0.0013, 0.0014, 0.0013], + device='cuda:3'), out_proj_covar=tensor([8.6881e-06, 9.0252e-06, 8.9078e-06, 8.9464e-06, 8.9258e-06, 8.9722e-06, + 8.8412e-06, 8.7866e-06], device='cuda:3') +2023-02-05 18:00:11,300 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-05 18:00:13,765 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:00:28,700 INFO [train.py:901] (3/4) Epoch 1, batch 100, loss[loss=1.077, simple_loss=0.9161, pruned_loss=1.267, over 7532.00 frames. ], tot_loss[loss=1.643, simple_loss=1.464, pruned_loss=1.614, over 644635.68 frames. ], batch size: 18, lr: 3.00e-02, grad_scale: 0.0625 +2023-02-05 18:00:28,817 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:00:31,818 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=10.81 vs. limit=2.0 +2023-02-05 18:00:32,055 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-05 18:00:32,813 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.087e+01 6.689e+01 1.862e+02 6.030e+02 6.185e+04, threshold=3.723e+02, percent-clipped=0.0 +2023-02-05 18:01:00,489 INFO [train.py:901] (3/4) Epoch 1, batch 150, loss[loss=0.9989, simple_loss=0.8501, pruned_loss=1.077, over 8143.00 frames. ], tot_loss[loss=1.407, simple_loss=1.238, pruned_loss=1.436, over 859443.32 frames. ], batch size: 22, lr: 3.25e-02, grad_scale: 0.0625 +2023-02-05 18:01:34,597 INFO [train.py:901] (3/4) Epoch 1, batch 200, loss[loss=0.9575, simple_loss=0.8116, pruned_loss=0.9806, over 8245.00 frames. ], tot_loss[loss=1.265, simple_loss=1.102, pruned_loss=1.301, over 1026898.35 frames. ], batch size: 22, lr: 3.50e-02, grad_scale: 0.125 +2023-02-05 18:01:37,990 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.848e+01 5.119e+01 6.630e+01 8.708e+01 3.236e+02, threshold=1.326e+02, percent-clipped=1.0 +2023-02-05 18:01:59,644 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=13.39 vs. limit=5.0 +2023-02-05 18:02:05,438 INFO [train.py:901] (3/4) Epoch 1, batch 250, loss[loss=0.8919, simple_loss=0.7493, pruned_loss=0.8904, over 7788.00 frames. ], tot_loss[loss=1.178, simple_loss=1.018, pruned_loss=1.203, over 1158228.87 frames. ], batch size: 19, lr: 3.75e-02, grad_scale: 0.125 +2023-02-05 18:02:14,834 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-05 18:02:21,590 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9300, 2.9300, 2.9300, 2.9300, 2.9300, 2.9300, 2.9300, 2.9300], + device='cuda:3'), covar=tensor([0.0002, 0.0003, 0.0001, 0.0001, 0.0002, 0.0004, 0.0002, 0.0001], + device='cuda:3'), in_proj_covar=tensor([0.0015, 0.0014, 0.0014, 0.0015, 0.0014, 0.0015, 0.0014, 0.0014], + device='cuda:3'), out_proj_covar=tensor([9.6301e-06, 9.5557e-06, 9.5193e-06, 9.2880e-06, 9.6126e-06, 9.4880e-06, + 9.5885e-06, 9.2433e-06], device='cuda:3') +2023-02-05 18:02:22,944 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-05 18:02:37,912 INFO [train.py:901] (3/4) Epoch 1, batch 300, loss[loss=0.9556, simple_loss=0.7941, pruned_loss=0.9397, over 7643.00 frames. ], tot_loss[loss=1.119, simple_loss=0.9593, pruned_loss=1.132, over 1259830.68 frames. ], batch size: 19, lr: 4.00e-02, grad_scale: 0.25 +2023-02-05 18:02:42,338 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=306.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:02:42,566 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=21.59 vs. limit=5.0 +2023-02-05 18:02:42,690 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.041e+01 5.570e+01 7.201e+01 9.677e+01 1.807e+02, threshold=1.440e+02, percent-clipped=6.0 +2023-02-05 18:02:47,399 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=314.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:02:57,561 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=9.81 vs. limit=5.0 +2023-02-05 18:03:10,250 INFO [train.py:901] (3/4) Epoch 1, batch 350, loss[loss=0.9643, simple_loss=0.794, pruned_loss=0.9303, over 8029.00 frames. ], tot_loss[loss=1.077, simple_loss=0.9168, pruned_loss=1.078, over 1341268.89 frames. ], batch size: 22, lr: 4.25e-02, grad_scale: 0.25 +2023-02-05 18:03:36,858 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.51 vs. limit=5.0 +2023-02-05 18:03:42,305 INFO [train.py:901] (3/4) Epoch 1, batch 400, loss[loss=1.002, simple_loss=0.8226, pruned_loss=0.9335, over 8357.00 frames. ], tot_loss[loss=1.055, simple_loss=0.89, pruned_loss=1.04, over 1410011.35 frames. ], batch size: 24, lr: 4.50e-02, grad_scale: 0.5 +2023-02-05 18:03:44,608 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=405.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:03:45,457 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.847e+01 5.714e+01 6.661e+01 8.261e+01 1.252e+02, threshold=1.332e+02, percent-clipped=0.0 +2023-02-05 18:03:55,287 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=421.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:04:11,522 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=445.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:04:15,520 INFO [train.py:901] (3/4) Epoch 1, batch 450, loss[loss=0.9513, simple_loss=0.7764, pruned_loss=0.8665, over 8048.00 frames. ], tot_loss[loss=1.035, simple_loss=0.8677, pruned_loss=1.006, over 1454719.88 frames. ], batch size: 20, lr: 4.75e-02, grad_scale: 0.5 +2023-02-05 18:04:45,727 INFO [train.py:901] (3/4) Epoch 1, batch 500, loss[loss=0.8925, simple_loss=0.7237, pruned_loss=0.7972, over 7657.00 frames. ], tot_loss[loss=1.016, simple_loss=0.8461, pruned_loss=0.9694, over 1485566.27 frames. ], batch size: 19, lr: 4.99e-02, grad_scale: 1.0 +2023-02-05 18:04:47,786 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.79 vs. limit=2.0 +2023-02-05 18:04:49,477 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 5.283e+01 6.268e+01 7.626e+01 9.977e+01 2.238e+02, threshold=1.525e+02, percent-clipped=10.0 +2023-02-05 18:05:16,922 INFO [train.py:901] (3/4) Epoch 1, batch 550, loss[loss=0.9715, simple_loss=0.7991, pruned_loss=0.8117, over 8791.00 frames. ], tot_loss[loss=0.9993, simple_loss=0.8289, pruned_loss=0.9331, over 1513812.58 frames. ], batch size: 40, lr: 4.98e-02, grad_scale: 1.0 +2023-02-05 18:05:22,228 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=560.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:05:22,436 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=7.56 vs. limit=5.0 +2023-02-05 18:05:34,666 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=580.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:05:39,265 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=586.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:05:47,855 INFO [train.py:901] (3/4) Epoch 1, batch 600, loss[loss=0.8942, simple_loss=0.7433, pruned_loss=0.7089, over 8467.00 frames. ], tot_loss[loss=0.9847, simple_loss=0.8156, pruned_loss=0.8941, over 1538658.31 frames. ], batch size: 25, lr: 4.98e-02, grad_scale: 1.0 +2023-02-05 18:05:51,160 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 5.986e+01 8.101e+01 1.064e+02 1.512e+02 3.340e+02, threshold=2.128e+02, percent-clipped=22.0 +2023-02-05 18:05:51,948 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=608.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:05:57,495 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-05 18:06:15,538 INFO [train.py:901] (3/4) Epoch 1, batch 650, loss[loss=0.786, simple_loss=0.6605, pruned_loss=0.5932, over 7716.00 frames. ], tot_loss[loss=0.9604, simple_loss=0.7963, pruned_loss=0.8466, over 1552882.00 frames. ], batch size: 18, lr: 4.98e-02, grad_scale: 1.0 +2023-02-05 18:06:20,640 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=658.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:06:31,063 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=677.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:06:44,410 INFO [train.py:901] (3/4) Epoch 1, batch 700, loss[loss=0.761, simple_loss=0.64, pruned_loss=0.5613, over 5925.00 frames. ], tot_loss[loss=0.9358, simple_loss=0.7778, pruned_loss=0.799, over 1568995.66 frames. ], batch size: 13, lr: 4.98e-02, grad_scale: 1.0 +2023-02-05 18:06:45,076 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:06:46,413 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.18 vs. limit=2.0 +2023-02-05 18:06:48,196 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.046e+02 2.109e+02 3.132e+02 4.412e+02 1.990e+03, threshold=6.264e+02, percent-clipped=73.0 +2023-02-05 18:07:14,476 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=749.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:07:15,374 INFO [train.py:901] (3/4) Epoch 1, batch 750, loss[loss=0.7555, simple_loss=0.6401, pruned_loss=0.5379, over 7784.00 frames. ], tot_loss[loss=0.9125, simple_loss=0.7608, pruned_loss=0.7546, over 1581164.66 frames. ], batch size: 19, lr: 4.97e-02, grad_scale: 1.0 +2023-02-05 18:07:25,623 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-05 18:07:26,836 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=773.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:07:32,338 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-05 18:07:43,637 INFO [train.py:901] (3/4) Epoch 1, batch 800, loss[loss=0.7988, simple_loss=0.6766, pruned_loss=0.5597, over 8138.00 frames. ], tot_loss[loss=0.8848, simple_loss=0.7405, pruned_loss=0.7095, over 1586213.52 frames. ], batch size: 22, lr: 4.97e-02, grad_scale: 2.0 +2023-02-05 18:07:43,939 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=6.18 vs. limit=5.0 +2023-02-05 18:07:46,610 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.447e+02 2.528e+02 3.354e+02 4.455e+02 1.086e+03, threshold=6.708e+02, percent-clipped=4.0 +2023-02-05 18:07:51,296 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=816.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:08:05,568 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=841.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:08:11,170 INFO [train.py:901] (3/4) Epoch 1, batch 850, loss[loss=0.7196, simple_loss=0.6155, pruned_loss=0.4867, over 7524.00 frames. ], tot_loss[loss=0.8583, simple_loss=0.7215, pruned_loss=0.6675, over 1595253.51 frames. ], batch size: 18, lr: 4.96e-02, grad_scale: 2.0 +2023-02-05 18:08:22,425 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=864.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:08:22,899 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=865.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:08:33,836 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=5.64 vs. limit=5.0 +2023-02-05 18:08:42,863 INFO [train.py:901] (3/4) Epoch 1, batch 900, loss[loss=0.7063, simple_loss=0.6008, pruned_loss=0.4764, over 7241.00 frames. ], tot_loss[loss=0.8349, simple_loss=0.7047, pruned_loss=0.6307, over 1598131.81 frames. ], batch size: 16, lr: 4.96e-02, grad_scale: 2.0 +2023-02-05 18:08:46,415 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.847e+02 3.070e+02 3.818e+02 4.702e+02 7.623e+02, threshold=7.636e+02, percent-clipped=5.0 +2023-02-05 18:08:55,930 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=924.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:08:58,991 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=930.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:09:10,094 INFO [train.py:901] (3/4) Epoch 1, batch 950, loss[loss=0.6854, simple_loss=0.5888, pruned_loss=0.4482, over 7969.00 frames. ], tot_loss[loss=0.8125, simple_loss=0.6889, pruned_loss=0.5968, over 1606047.71 frames. ], batch size: 21, lr: 4.96e-02, grad_scale: 2.0 +2023-02-05 18:09:10,759 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=952.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:09:26,439 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-05 18:09:29,016 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=5.36 vs. limit=5.0 +2023-02-05 18:09:37,668 INFO [train.py:901] (3/4) Epoch 1, batch 1000, loss[loss=0.751, simple_loss=0.6417, pruned_loss=0.4903, over 8259.00 frames. ], tot_loss[loss=0.7902, simple_loss=0.6728, pruned_loss=0.5655, over 1608023.85 frames. ], batch size: 24, lr: 4.95e-02, grad_scale: 2.0 +2023-02-05 18:09:40,945 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.960e+02 3.215e+02 4.159e+02 4.799e+02 1.770e+03, threshold=8.319e+02, percent-clipped=6.0 +2023-02-05 18:09:52,920 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1029.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:09:53,905 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-05 18:09:59,204 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1039.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:10:02,591 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1045.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:10:05,088 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-05 18:10:05,577 INFO [train.py:901] (3/4) Epoch 1, batch 1050, loss[loss=0.8104, simple_loss=0.692, pruned_loss=0.524, over 8239.00 frames. ], tot_loss[loss=0.7725, simple_loss=0.6601, pruned_loss=0.5396, over 1613711.50 frames. ], batch size: 24, lr: 4.95e-02, grad_scale: 2.0 +2023-02-05 18:10:07,201 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1054.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:10:14,079 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1067.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:10:33,060 INFO [train.py:901] (3/4) Epoch 1, batch 1100, loss[loss=0.6373, simple_loss=0.5579, pruned_loss=0.3912, over 7404.00 frames. ], tot_loss[loss=0.7517, simple_loss=0.6454, pruned_loss=0.5127, over 1617354.67 frames. ], batch size: 17, lr: 4.94e-02, grad_scale: 2.0 +2023-02-05 18:10:36,094 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.854e+02 3.463e+02 4.480e+02 5.452e+02 1.232e+03, threshold=8.959e+02, percent-clipped=3.0 +2023-02-05 18:10:43,739 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1120.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:10:56,863 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1145.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:10:59,903 INFO [train.py:901] (3/4) Epoch 1, batch 1150, loss[loss=0.656, simple_loss=0.5699, pruned_loss=0.4047, over 6418.00 frames. ], tot_loss[loss=0.7344, simple_loss=0.633, pruned_loss=0.4905, over 1615603.45 frames. ], batch size: 14, lr: 4.94e-02, grad_scale: 2.0 +2023-02-05 18:11:01,598 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-05 18:11:02,660 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.7594, 3.8524, 3.8558, 1.9348, 3.8243, 4.0600, 3.7095, 3.9535], + device='cuda:3'), covar=tensor([0.0441, 0.0563, 0.0483, 0.1643, 0.0540, 0.0465, 0.0508, 0.0497], + device='cuda:3'), in_proj_covar=tensor([0.0038, 0.0032, 0.0035, 0.0048, 0.0037, 0.0035, 0.0038, 0.0039], + device='cuda:3'), out_proj_covar=tensor([2.5354e-05, 2.2330e-05, 2.2442e-05, 3.5247e-05, 2.4419e-05, 2.3636e-05, + 2.5796e-05, 2.5344e-05], device='cuda:3') +2023-02-05 18:11:05,790 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6254, 1.2649, 2.0076, 1.8245, 1.3255, 1.1424, 1.3392, 1.6682], + device='cuda:3'), covar=tensor([0.5609, 1.4936, 0.4080, 0.4683, 1.0158, 1.1708, 1.1662, 0.6630], + device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0074, 0.0046, 0.0051, 0.0074, 0.0078, 0.0081, 0.0063], + device='cuda:3'), out_proj_covar=tensor([3.3258e-05, 5.2402e-05, 2.6211e-05, 2.8161e-05, 4.8208e-05, 5.1034e-05, + 4.9756e-05, 3.8707e-05], device='cuda:3') +2023-02-05 18:11:11,766 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1171.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:11:27,730 INFO [train.py:901] (3/4) Epoch 1, batch 1200, loss[loss=0.5851, simple_loss=0.5283, pruned_loss=0.3364, over 8124.00 frames. ], tot_loss[loss=0.7159, simple_loss=0.6197, pruned_loss=0.4687, over 1609697.87 frames. ], batch size: 22, lr: 4.93e-02, grad_scale: 4.0 +2023-02-05 18:11:30,965 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.986e+02 3.424e+02 4.173e+02 5.178e+02 8.029e+02, threshold=8.346e+02, percent-clipped=0.0 +2023-02-05 18:11:32,143 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1209.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:11:53,794 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 +2023-02-05 18:11:56,782 INFO [train.py:901] (3/4) Epoch 1, batch 1250, loss[loss=0.7161, simple_loss=0.6337, pruned_loss=0.4233, over 8503.00 frames. ], tot_loss[loss=0.7029, simple_loss=0.6108, pruned_loss=0.4516, over 1609417.85 frames. ], batch size: 28, lr: 4.92e-02, grad_scale: 4.0 +2023-02-05 18:12:05,165 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.7196, 3.8904, 3.8571, 2.3739, 3.6634, 4.0576, 3.6041, 3.5916], + device='cuda:3'), covar=tensor([0.0377, 0.0442, 0.0362, 0.1046, 0.0447, 0.0313, 0.0438, 0.0534], + device='cuda:3'), in_proj_covar=tensor([0.0047, 0.0040, 0.0042, 0.0062, 0.0046, 0.0041, 0.0045, 0.0046], + device='cuda:3'), out_proj_covar=tensor([3.1671e-05, 2.7687e-05, 2.6536e-05, 4.6501e-05, 3.0072e-05, 2.6935e-05, + 3.0494e-05, 2.9634e-05], device='cuda:3') +2023-02-05 18:12:21,181 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1295.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:12:24,257 INFO [train.py:901] (3/4) Epoch 1, batch 1300, loss[loss=0.5912, simple_loss=0.5375, pruned_loss=0.3334, over 7917.00 frames. ], tot_loss[loss=0.6889, simple_loss=0.6012, pruned_loss=0.4348, over 1612269.35 frames. ], batch size: 20, lr: 4.92e-02, grad_scale: 4.0 +2023-02-05 18:12:24,450 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1301.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:12:27,420 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.398e+02 3.917e+02 4.747e+02 6.152e+02 9.080e+02, threshold=9.493e+02, percent-clipped=1.0 +2023-02-05 18:12:34,741 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:12:36,266 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1323.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:12:36,766 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1324.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:12:37,911 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1326.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:12:51,947 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1348.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:12:53,366 INFO [train.py:901] (3/4) Epoch 1, batch 1350, loss[loss=0.7066, simple_loss=0.6087, pruned_loss=0.4288, over 8724.00 frames. ], tot_loss[loss=0.6764, simple_loss=0.5927, pruned_loss=0.4198, over 1612336.91 frames. ], batch size: 34, lr: 4.91e-02, grad_scale: 4.0 +2023-02-05 18:13:22,443 INFO [train.py:901] (3/4) Epoch 1, batch 1400, loss[loss=0.6738, simple_loss=0.5878, pruned_loss=0.3996, over 7810.00 frames. ], tot_loss[loss=0.668, simple_loss=0.5872, pruned_loss=0.4088, over 1613389.43 frames. ], batch size: 20, lr: 4.91e-02, grad_scale: 4.0 +2023-02-05 18:13:25,824 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.139e+02 3.466e+02 4.520e+02 5.912e+02 1.396e+03, threshold=9.040e+02, percent-clipped=6.0 +2023-02-05 18:13:38,639 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0012, 1.2284, 2.5925, 1.4526, 1.6353, 1.6792, 1.1084, 1.4847], + device='cuda:3'), covar=tensor([0.8715, 1.0867, 0.1508, 0.5266, 0.6881, 0.5120, 0.7594, 0.8337], + device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0076, 0.0041, 0.0062, 0.0080, 0.0058, 0.0075, 0.0086], + device='cuda:3'), out_proj_covar=tensor([4.8778e-05, 5.1714e-05, 2.3044e-05, 4.0066e-05, 5.4528e-05, 3.7801e-05, + 4.7841e-05, 5.8963e-05], device='cuda:3') +2023-02-05 18:13:49,601 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2023-02-05 18:13:50,939 INFO [train.py:901] (3/4) Epoch 1, batch 1450, loss[loss=0.5099, simple_loss=0.4714, pruned_loss=0.2783, over 7258.00 frames. ], tot_loss[loss=0.6556, simple_loss=0.5786, pruned_loss=0.3956, over 1611615.53 frames. ], batch size: 16, lr: 4.90e-02, grad_scale: 4.0 +2023-02-05 18:13:51,608 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1452.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:13:54,969 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-05 18:14:21,307 INFO [train.py:901] (3/4) Epoch 1, batch 1500, loss[loss=0.648, simple_loss=0.5863, pruned_loss=0.3634, over 8107.00 frames. ], tot_loss[loss=0.6475, simple_loss=0.5731, pruned_loss=0.3862, over 1607894.44 frames. ], batch size: 23, lr: 4.89e-02, grad_scale: 4.0 +2023-02-05 18:14:24,737 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.084e+02 4.059e+02 4.884e+02 5.820e+02 1.191e+03, threshold=9.769e+02, percent-clipped=4.0 +2023-02-05 18:14:29,250 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1515.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:14:50,474 INFO [train.py:901] (3/4) Epoch 1, batch 1550, loss[loss=0.5724, simple_loss=0.5287, pruned_loss=0.3116, over 7962.00 frames. ], tot_loss[loss=0.6389, simple_loss=0.5672, pruned_loss=0.3768, over 1604494.35 frames. ], batch size: 21, lr: 4.89e-02, grad_scale: 4.0 +2023-02-05 18:15:08,671 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1580.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:15:10,913 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1584.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:15:20,766 INFO [train.py:901] (3/4) Epoch 1, batch 1600, loss[loss=0.5907, simple_loss=0.5379, pruned_loss=0.3268, over 8294.00 frames. ], tot_loss[loss=0.6352, simple_loss=0.5651, pruned_loss=0.3711, over 1613402.63 frames. ], batch size: 23, lr: 4.88e-02, grad_scale: 8.0 +2023-02-05 18:15:24,013 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1605.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:15:24,960 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.844e+02 4.893e+02 6.465e+02 8.597e+02 2.177e+03, threshold=1.293e+03, percent-clipped=12.0 +2023-02-05 18:15:33,782 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.0205, 5.2160, 4.9740, 2.1578, 4.8972, 5.3830, 4.6630, 4.9411], + device='cuda:3'), covar=tensor([0.0420, 0.0367, 0.0301, 0.1949, 0.0269, 0.0478, 0.0319, 0.0370], + device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0054, 0.0060, 0.0097, 0.0060, 0.0057, 0.0061, 0.0058], + device='cuda:3'), out_proj_covar=tensor([4.5513e-05, 3.7738e-05, 4.0761e-05, 7.0688e-05, 4.0835e-05, 3.9221e-05, + 4.1937e-05, 3.8797e-05], device='cuda:3') +2023-02-05 18:15:37,781 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1629.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:15:38,284 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1630.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:15:50,690 INFO [train.py:901] (3/4) Epoch 1, batch 1650, loss[loss=0.592, simple_loss=0.5418, pruned_loss=0.3249, over 8503.00 frames. ], tot_loss[loss=0.6289, simple_loss=0.5618, pruned_loss=0.3636, over 1611015.38 frames. ], batch size: 26, lr: 4.87e-02, grad_scale: 8.0 +2023-02-05 18:16:21,955 INFO [train.py:901] (3/4) Epoch 1, batch 1700, loss[loss=0.6116, simple_loss=0.5544, pruned_loss=0.3385, over 8253.00 frames. ], tot_loss[loss=0.6192, simple_loss=0.5564, pruned_loss=0.3538, over 1614723.67 frames. ], batch size: 24, lr: 4.86e-02, grad_scale: 8.0 +2023-02-05 18:16:25,341 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.633e+02 4.287e+02 5.230e+02 6.455e+02 2.107e+03, threshold=1.046e+03, percent-clipped=2.0 +2023-02-05 18:16:51,240 INFO [train.py:901] (3/4) Epoch 1, batch 1750, loss[loss=0.5433, simple_loss=0.5023, pruned_loss=0.2938, over 8137.00 frames. ], tot_loss[loss=0.6133, simple_loss=0.5525, pruned_loss=0.3477, over 1613089.86 frames. ], batch size: 22, lr: 4.86e-02, grad_scale: 8.0 +2023-02-05 18:17:18,074 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1796.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:17:21,116 INFO [train.py:901] (3/4) Epoch 1, batch 1800, loss[loss=0.5784, simple_loss=0.5364, pruned_loss=0.3114, over 8472.00 frames. ], tot_loss[loss=0.6067, simple_loss=0.5488, pruned_loss=0.3411, over 1612614.09 frames. ], batch size: 25, lr: 4.85e-02, grad_scale: 8.0 +2023-02-05 18:17:24,721 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.688e+02 4.554e+02 5.596e+02 6.733e+02 1.418e+03, threshold=1.119e+03, percent-clipped=4.0 +2023-02-05 18:17:26,909 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.01 vs. limit=2.0 +2023-02-05 18:17:44,339 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.83 vs. limit=2.0 +2023-02-05 18:17:52,109 INFO [train.py:901] (3/4) Epoch 1, batch 1850, loss[loss=0.607, simple_loss=0.5556, pruned_loss=0.3307, over 8510.00 frames. ], tot_loss[loss=0.6013, simple_loss=0.5456, pruned_loss=0.3357, over 1616464.31 frames. ], batch size: 28, lr: 4.84e-02, grad_scale: 8.0 +2023-02-05 18:17:55,086 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1856.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:17:55,654 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9151, 1.2690, 0.8573, 2.2682, 1.2871, 1.4380, 1.3075, 1.9423], + device='cuda:3'), covar=tensor([0.2760, 0.6968, 1.4274, 0.0945, 0.6216, 0.4200, 0.7004, 0.1784], + device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0210, 0.0311, 0.0135, 0.0214, 0.0201, 0.0248, 0.0177], + device='cuda:3'), out_proj_covar=tensor([1.1794e-04, 1.4496e-04, 2.0223e-04, 9.1296e-05, 1.5002e-04, 1.3122e-04, + 1.6409e-04, 1.1045e-04], device='cuda:3') +2023-02-05 18:18:06,765 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1875.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:18:13,302 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1886.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:18:14,331 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1888.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:18:21,902 INFO [train.py:901] (3/4) Epoch 1, batch 1900, loss[loss=0.5793, simple_loss=0.5301, pruned_loss=0.3152, over 7969.00 frames. ], tot_loss[loss=0.5939, simple_loss=0.5413, pruned_loss=0.3289, over 1617514.19 frames. ], batch size: 21, lr: 4.83e-02, grad_scale: 8.0 +2023-02-05 18:18:25,466 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.326e+02 4.483e+02 5.242e+02 7.443e+02 2.270e+03, threshold=1.048e+03, percent-clipped=7.0 +2023-02-05 18:18:27,930 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1911.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:18:27,943 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1911.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:18:30,293 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3864, 1.6197, 4.2089, 1.4383, 2.3730, 2.6044, 1.1403, 1.9916], + device='cuda:3'), covar=tensor([0.4306, 0.5133, 0.0442, 0.3024, 0.3365, 0.2670, 0.4589, 0.4210], + device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0097, 0.0051, 0.0077, 0.0108, 0.0086, 0.0097, 0.0112], + device='cuda:3'), out_proj_covar=tensor([6.2742e-05, 6.6057e-05, 2.9482e-05, 4.9955e-05, 7.4022e-05, 6.0976e-05, + 6.2921e-05, 7.5834e-05], device='cuda:3') +2023-02-05 18:18:37,743 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1928.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:18:45,037 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-05 18:18:52,614 INFO [train.py:901] (3/4) Epoch 1, batch 1950, loss[loss=0.5505, simple_loss=0.5188, pruned_loss=0.2912, over 8358.00 frames. ], tot_loss[loss=0.5879, simple_loss=0.5377, pruned_loss=0.3236, over 1617586.26 frames. ], batch size: 24, lr: 4.83e-02, grad_scale: 8.0 +2023-02-05 18:18:55,553 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-05 18:19:00,392 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8594, 1.3773, 2.1877, 1.3456, 2.0076, 2.9135, 2.8510, 2.5565], + device='cuda:3'), covar=tensor([0.4364, 0.5856, 0.0986, 0.5193, 0.2389, 0.0733, 0.0508, 0.0723], + device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0184, 0.0101, 0.0181, 0.0144, 0.0089, 0.0082, 0.0102], + device='cuda:3'), out_proj_covar=tensor([1.1815e-04, 1.3210e-04, 6.1524e-05, 1.1884e-04, 1.0173e-04, 5.6628e-05, + 4.7520e-05, 6.2277e-05], device='cuda:3') +2023-02-05 18:19:05,743 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1973.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:19:11,336 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-05 18:19:23,720 INFO [train.py:901] (3/4) Epoch 1, batch 2000, loss[loss=0.6118, simple_loss=0.5473, pruned_loss=0.3382, over 6986.00 frames. ], tot_loss[loss=0.5834, simple_loss=0.5354, pruned_loss=0.3193, over 1615443.51 frames. ], batch size: 72, lr: 4.82e-02, grad_scale: 8.0 +2023-02-05 18:19:27,541 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.734e+02 4.600e+02 5.655e+02 7.771e+02 1.691e+03, threshold=1.131e+03, percent-clipped=5.0 +2023-02-05 18:19:50,362 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2043.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:19:56,662 INFO [train.py:901] (3/4) Epoch 1, batch 2050, loss[loss=0.4603, simple_loss=0.46, pruned_loss=0.2303, over 8234.00 frames. ], tot_loss[loss=0.5748, simple_loss=0.5304, pruned_loss=0.3124, over 1614412.53 frames. ], batch size: 22, lr: 4.81e-02, grad_scale: 8.0 +2023-02-05 18:20:05,737 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3255, 1.3984, 2.1804, 1.3407, 2.3645, 2.4101, 2.4312, 2.0753], + device='cuda:3'), covar=tensor([0.2587, 0.2606, 0.0508, 0.2719, 0.0799, 0.0393, 0.0344, 0.0461], + device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0181, 0.0095, 0.0168, 0.0143, 0.0084, 0.0080, 0.0098], + device='cuda:3'), out_proj_covar=tensor([1.1779e-04, 1.2906e-04, 5.8209e-05, 1.1144e-04, 1.0200e-04, 5.3951e-05, + 4.7212e-05, 6.1994e-05], device='cuda:3') +2023-02-05 18:20:21,052 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2088.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:20:29,073 INFO [train.py:901] (3/4) Epoch 1, batch 2100, loss[loss=0.628, simple_loss=0.558, pruned_loss=0.349, over 7087.00 frames. ], tot_loss[loss=0.5683, simple_loss=0.5269, pruned_loss=0.307, over 1614361.90 frames. ], batch size: 71, lr: 4.80e-02, grad_scale: 16.0 +2023-02-05 18:20:32,716 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.532e+02 4.654e+02 5.875e+02 8.240e+02 2.515e+03, threshold=1.175e+03, percent-clipped=11.0 +2023-02-05 18:20:55,902 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3680, 5.6484, 4.7931, 1.6381, 4.6091, 5.1359, 4.6366, 4.7301], + device='cuda:3'), covar=tensor([0.0367, 0.0297, 0.0320, 0.2793, 0.0376, 0.0420, 0.0649, 0.0398], + device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0077, 0.0088, 0.0140, 0.0080, 0.0070, 0.0099, 0.0079], + device='cuda:3'), out_proj_covar=tensor([6.0274e-05, 6.3256e-05, 5.9619e-05, 9.9716e-05, 5.5193e-05, 5.1718e-05, + 7.1618e-05, 5.3931e-05], device='cuda:3') +2023-02-05 18:21:01,657 INFO [train.py:901] (3/4) Epoch 1, batch 2150, loss[loss=0.4852, simple_loss=0.4863, pruned_loss=0.2421, over 7927.00 frames. ], tot_loss[loss=0.5585, simple_loss=0.521, pruned_loss=0.2997, over 1610118.68 frames. ], batch size: 20, lr: 4.79e-02, grad_scale: 16.0 +2023-02-05 18:21:11,753 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2167.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:21:29,909 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2192.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:21:35,018 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2200.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:21:35,572 INFO [train.py:901] (3/4) Epoch 1, batch 2200, loss[loss=0.6085, simple_loss=0.5605, pruned_loss=0.3282, over 8554.00 frames. ], tot_loss[loss=0.5507, simple_loss=0.5164, pruned_loss=0.2938, over 1611845.39 frames. ], batch size: 34, lr: 4.78e-02, grad_scale: 16.0 +2023-02-05 18:21:39,335 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.979e+02 3.885e+02 5.100e+02 6.280e+02 1.293e+03, threshold=1.020e+03, percent-clipped=3.0 +2023-02-05 18:21:46,995 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2219.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:21:55,784 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2232.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:22:07,883 INFO [train.py:901] (3/4) Epoch 1, batch 2250, loss[loss=0.496, simple_loss=0.4792, pruned_loss=0.2564, over 7970.00 frames. ], tot_loss[loss=0.5437, simple_loss=0.5122, pruned_loss=0.2887, over 1613475.54 frames. ], batch size: 21, lr: 4.77e-02, grad_scale: 16.0 +2023-02-05 18:22:41,035 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2299.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:22:42,094 INFO [train.py:901] (3/4) Epoch 1, batch 2300, loss[loss=0.4354, simple_loss=0.4306, pruned_loss=0.2201, over 7397.00 frames. ], tot_loss[loss=0.5441, simple_loss=0.514, pruned_loss=0.2879, over 1621204.64 frames. ], batch size: 17, lr: 4.77e-02, grad_scale: 16.0 +2023-02-05 18:22:45,954 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.442e+02 5.272e+02 6.513e+02 7.975e+02 1.884e+03, threshold=1.303e+03, percent-clipped=9.0 +2023-02-05 18:22:51,195 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2315.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:22:56,962 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2324.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:23:03,216 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2334.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:23:09,703 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2344.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:23:12,276 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2347.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:23:14,708 INFO [train.py:901] (3/4) Epoch 1, batch 2350, loss[loss=0.5159, simple_loss=0.5122, pruned_loss=0.2598, over 8318.00 frames. ], tot_loss[loss=0.5389, simple_loss=0.5107, pruned_loss=0.2842, over 1614903.40 frames. ], batch size: 26, lr: 4.76e-02, grad_scale: 16.0 +2023-02-05 18:23:19,256 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2358.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:23:23,687 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 +2023-02-05 18:23:26,044 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2369.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:23:46,437 INFO [train.py:901] (3/4) Epoch 1, batch 2400, loss[loss=0.4616, simple_loss=0.4667, pruned_loss=0.2283, over 8190.00 frames. ], tot_loss[loss=0.5346, simple_loss=0.5083, pruned_loss=0.2809, over 1619645.01 frames. ], batch size: 23, lr: 4.75e-02, grad_scale: 16.0 +2023-02-05 18:23:50,348 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.591e+02 4.467e+02 5.905e+02 7.151e+02 1.301e+03, threshold=1.181e+03, percent-clipped=0.0 +2023-02-05 18:24:20,801 INFO [train.py:901] (3/4) Epoch 1, batch 2450, loss[loss=0.5529, simple_loss=0.5258, pruned_loss=0.29, over 8479.00 frames. ], tot_loss[loss=0.5318, simple_loss=0.5065, pruned_loss=0.2789, over 1615220.29 frames. ], batch size: 25, lr: 4.74e-02, grad_scale: 16.0 +2023-02-05 18:24:21,025 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5283, 1.2835, 1.7627, 1.6575, 1.4324, 1.7893, 0.9447, 1.6747], + device='cuda:3'), covar=tensor([0.1109, 0.0741, 0.0655, 0.0810, 0.1067, 0.0555, 0.2128, 0.0936], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0083, 0.0080, 0.0085, 0.0096, 0.0074, 0.0125, 0.0109], + device='cuda:3'), out_proj_covar=tensor([6.8589e-05, 5.7500e-05, 5.3330e-05, 6.0851e-05, 6.9431e-05, 4.9919e-05, + 9.2098e-05, 8.0544e-05], device='cuda:3') +2023-02-05 18:24:52,768 INFO [train.py:901] (3/4) Epoch 1, batch 2500, loss[loss=0.537, simple_loss=0.5068, pruned_loss=0.2835, over 7821.00 frames. ], tot_loss[loss=0.527, simple_loss=0.5036, pruned_loss=0.2755, over 1615758.41 frames. ], batch size: 20, lr: 4.73e-02, grad_scale: 16.0 +2023-02-05 18:24:56,549 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.099e+02 5.238e+02 6.448e+02 8.237e+02 1.660e+03, threshold=1.290e+03, percent-clipped=6.0 +2023-02-05 18:25:25,593 INFO [train.py:901] (3/4) Epoch 1, batch 2550, loss[loss=0.5541, simple_loss=0.5228, pruned_loss=0.2927, over 8471.00 frames. ], tot_loss[loss=0.5245, simple_loss=0.5023, pruned_loss=0.2736, over 1617792.86 frames. ], batch size: 25, lr: 4.72e-02, grad_scale: 16.0 +2023-02-05 18:25:27,057 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.7906, 1.3193, 5.1770, 2.4909, 4.9442, 4.3059, 4.6395, 4.4317], + device='cuda:3'), covar=tensor([0.0127, 0.3588, 0.0226, 0.1213, 0.0212, 0.0238, 0.0317, 0.0372], + device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0201, 0.0088, 0.0115, 0.0098, 0.0094, 0.0104, 0.0114], + device='cuda:3'), out_proj_covar=tensor([4.3529e-05, 1.2420e-04, 5.7263e-05, 7.7078e-05, 5.5557e-05, 5.2458e-05, + 6.3758e-05, 6.8137e-05], device='cuda:3') +2023-02-05 18:25:38,471 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2571.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:25:51,086 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2590.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:25:54,854 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2596.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:25:57,886 INFO [train.py:901] (3/4) Epoch 1, batch 2600, loss[loss=0.4265, simple_loss=0.4226, pruned_loss=0.2152, over 7254.00 frames. ], tot_loss[loss=0.5182, simple_loss=0.4993, pruned_loss=0.2687, over 1619749.52 frames. ], batch size: 16, lr: 4.71e-02, grad_scale: 16.0 +2023-02-05 18:25:59,371 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2603.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:26:01,610 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.388e+02 4.352e+02 5.534e+02 7.344e+02 1.370e+03, threshold=1.107e+03, percent-clipped=3.0 +2023-02-05 18:26:05,385 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.01 vs. limit=5.0 +2023-02-05 18:26:06,868 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2615.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:26:15,229 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2628.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:26:21,137 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 +2023-02-05 18:26:31,161 INFO [train.py:901] (3/4) Epoch 1, batch 2650, loss[loss=0.4979, simple_loss=0.503, pruned_loss=0.2465, over 8476.00 frames. ], tot_loss[loss=0.5142, simple_loss=0.4977, pruned_loss=0.2655, over 1617672.68 frames. ], batch size: 29, lr: 4.70e-02, grad_scale: 16.0 +2023-02-05 18:26:31,333 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3625, 1.1462, 4.2103, 2.2496, 3.9024, 3.5498, 3.5841, 3.6872], + device='cuda:3'), covar=tensor([0.0140, 0.3543, 0.0190, 0.1033, 0.0229, 0.0247, 0.0335, 0.0308], + device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0210, 0.0089, 0.0117, 0.0102, 0.0095, 0.0107, 0.0115], + device='cuda:3'), out_proj_covar=tensor([4.3958e-05, 1.2911e-04, 5.7165e-05, 7.9091e-05, 5.8958e-05, 5.2637e-05, + 6.5615e-05, 6.8482e-05], device='cuda:3') +2023-02-05 18:27:03,817 INFO [train.py:901] (3/4) Epoch 1, batch 2700, loss[loss=0.5234, simple_loss=0.5027, pruned_loss=0.2721, over 8343.00 frames. ], tot_loss[loss=0.5096, simple_loss=0.4942, pruned_loss=0.2626, over 1618722.89 frames. ], batch size: 24, lr: 4.69e-02, grad_scale: 16.0 +2023-02-05 18:27:04,571 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2702.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:27:05,219 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2703.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:27:08,314 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.214e+02 4.351e+02 5.311e+02 6.408e+02 1.471e+03, threshold=1.062e+03, percent-clipped=4.0 +2023-02-05 18:27:31,461 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5180, 1.6228, 1.1530, 1.8551, 1.5725, 1.5459, 1.3527, 1.4834], + device='cuda:3'), covar=tensor([0.1525, 0.1525, 0.2726, 0.0825, 0.2290, 0.1706, 0.3239, 0.1650], + device='cuda:3'), in_proj_covar=tensor([0.0111, 0.0105, 0.0145, 0.0087, 0.0126, 0.0106, 0.0153, 0.0111], + device='cuda:3'), out_proj_covar=tensor([7.9057e-05, 7.3684e-05, 9.7280e-05, 6.0446e-05, 9.0541e-05, 7.2948e-05, + 1.0652e-04, 7.4015e-05], device='cuda:3') +2023-02-05 18:27:32,088 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7568, 1.4946, 2.9518, 1.6669, 2.1490, 3.6152, 3.4082, 3.2476], + device='cuda:3'), covar=tensor([0.2688, 0.2998, 0.0377, 0.2592, 0.1440, 0.0219, 0.0257, 0.0375], + device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0216, 0.0115, 0.0195, 0.0168, 0.0094, 0.0092, 0.0120], + device='cuda:3'), out_proj_covar=tensor([1.4425e-04, 1.5460e-04, 8.7834e-05, 1.3309e-04, 1.3085e-04, 6.7281e-05, + 6.7995e-05, 8.2497e-05], device='cuda:3') +2023-02-05 18:27:37,287 INFO [train.py:901] (3/4) Epoch 1, batch 2750, loss[loss=0.4934, simple_loss=0.4769, pruned_loss=0.255, over 8086.00 frames. ], tot_loss[loss=0.5068, simple_loss=0.4927, pruned_loss=0.2605, over 1617502.09 frames. ], batch size: 21, lr: 4.68e-02, grad_scale: 16.0 +2023-02-05 18:28:11,556 INFO [train.py:901] (3/4) Epoch 1, batch 2800, loss[loss=0.5253, simple_loss=0.514, pruned_loss=0.2683, over 8679.00 frames. ], tot_loss[loss=0.5064, simple_loss=0.4929, pruned_loss=0.26, over 1613714.26 frames. ], batch size: 39, lr: 4.67e-02, grad_scale: 16.0 +2023-02-05 18:28:15,256 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.340e+02 3.916e+02 4.898e+02 6.530e+02 2.276e+03, threshold=9.797e+02, percent-clipped=2.0 +2023-02-05 18:28:21,893 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2817.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:28:28,487 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.73 vs. limit=5.0 +2023-02-05 18:28:38,536 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2842.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:28:44,024 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 +2023-02-05 18:28:44,180 INFO [train.py:901] (3/4) Epoch 1, batch 2850, loss[loss=0.515, simple_loss=0.5178, pruned_loss=0.2561, over 8352.00 frames. ], tot_loss[loss=0.5062, simple_loss=0.4932, pruned_loss=0.2597, over 1619738.62 frames. ], batch size: 24, lr: 4.66e-02, grad_scale: 16.0 +2023-02-05 18:29:18,789 INFO [train.py:901] (3/4) Epoch 1, batch 2900, loss[loss=0.5044, simple_loss=0.4964, pruned_loss=0.2562, over 8712.00 frames. ], tot_loss[loss=0.5067, simple_loss=0.4935, pruned_loss=0.26, over 1615680.85 frames. ], batch size: 39, lr: 4.65e-02, grad_scale: 16.0 +2023-02-05 18:29:22,676 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.417e+02 4.413e+02 5.664e+02 7.338e+02 1.737e+03, threshold=1.133e+03, percent-clipped=8.0 +2023-02-05 18:29:48,922 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-05 18:29:52,158 INFO [train.py:901] (3/4) Epoch 1, batch 2950, loss[loss=0.5386, simple_loss=0.5289, pruned_loss=0.2741, over 8244.00 frames. ], tot_loss[loss=0.5021, simple_loss=0.4904, pruned_loss=0.257, over 1615322.51 frames. ], batch size: 24, lr: 4.64e-02, grad_scale: 16.0 +2023-02-05 18:29:54,924 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2955.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:30:25,905 INFO [train.py:901] (3/4) Epoch 1, batch 3000, loss[loss=0.508, simple_loss=0.498, pruned_loss=0.259, over 8338.00 frames. ], tot_loss[loss=0.5012, simple_loss=0.4902, pruned_loss=0.2561, over 1616735.87 frames. ], batch size: 25, lr: 4.63e-02, grad_scale: 16.0 +2023-02-05 18:30:25,905 INFO [train.py:926] (3/4) Computing validation loss +2023-02-05 18:30:40,786 INFO [train.py:935] (3/4) Epoch 1, validation: loss=0.4518, simple_loss=0.5106, pruned_loss=0.1966, over 944034.00 frames. +2023-02-05 18:30:40,787 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6178MB +2023-02-05 18:30:44,896 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.692e+02 4.264e+02 5.642e+02 7.781e+02 1.743e+03, threshold=1.128e+03, percent-clipped=6.0 +2023-02-05 18:31:07,322 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3037.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:31:13,911 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3047.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:31:16,515 INFO [train.py:901] (3/4) Epoch 1, batch 3050, loss[loss=0.4737, simple_loss=0.4838, pruned_loss=0.2318, over 8427.00 frames. ], tot_loss[loss=0.4963, simple_loss=0.4873, pruned_loss=0.2527, over 1613437.01 frames. ], batch size: 27, lr: 4.62e-02, grad_scale: 16.0 +2023-02-05 18:31:30,880 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3073.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:31:47,515 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3098.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:31:48,194 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4441, 0.8959, 2.6349, 1.4056, 1.6479, 1.4928, 0.8810, 2.7416], + device='cuda:3'), covar=tensor([0.0938, 0.0699, 0.0266, 0.0683, 0.0958, 0.0558, 0.0952, 0.0322], + device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0080, 0.0068, 0.0090, 0.0079, 0.0079, 0.0100, 0.0073], + device='cuda:3'), out_proj_covar=tensor([6.8089e-05, 5.2254e-05, 4.5432e-05, 6.6716e-05, 5.4866e-05, 5.1492e-05, + 6.8975e-05, 4.7165e-05], device='cuda:3') +2023-02-05 18:31:49,294 INFO [train.py:901] (3/4) Epoch 1, batch 3100, loss[loss=0.4929, simple_loss=0.4809, pruned_loss=0.2524, over 7661.00 frames. ], tot_loss[loss=0.4963, simple_loss=0.4866, pruned_loss=0.253, over 1612463.03 frames. ], batch size: 19, lr: 4.61e-02, grad_scale: 16.0 +2023-02-05 18:31:53,107 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.570e+02 4.257e+02 6.045e+02 8.311e+02 2.838e+03, threshold=1.209e+03, percent-clipped=13.0 +2023-02-05 18:32:23,218 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 +2023-02-05 18:32:24,771 INFO [train.py:901] (3/4) Epoch 1, batch 3150, loss[loss=0.3752, simple_loss=0.3925, pruned_loss=0.179, over 7432.00 frames. ], tot_loss[loss=0.4929, simple_loss=0.4844, pruned_loss=0.2507, over 1609144.24 frames. ], batch size: 17, lr: 4.60e-02, grad_scale: 16.0 +2023-02-05 18:32:32,236 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3162.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:32:47,673 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3186.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:32:57,082 INFO [train.py:901] (3/4) Epoch 1, batch 3200, loss[loss=0.4077, simple_loss=0.4101, pruned_loss=0.2026, over 7420.00 frames. ], tot_loss[loss=0.4904, simple_loss=0.4829, pruned_loss=0.249, over 1610326.04 frames. ], batch size: 17, lr: 4.59e-02, grad_scale: 16.0 +2023-02-05 18:33:00,911 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.241e+02 4.232e+02 5.266e+02 6.948e+02 2.778e+03, threshold=1.053e+03, percent-clipped=2.0 +2023-02-05 18:33:32,107 INFO [train.py:901] (3/4) Epoch 1, batch 3250, loss[loss=0.4871, simple_loss=0.4681, pruned_loss=0.253, over 7696.00 frames. ], tot_loss[loss=0.4905, simple_loss=0.4825, pruned_loss=0.2493, over 1607104.94 frames. ], batch size: 18, lr: 4.58e-02, grad_scale: 16.0 +2023-02-05 18:34:04,438 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3299.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:34:05,630 INFO [train.py:901] (3/4) Epoch 1, batch 3300, loss[loss=0.5583, simple_loss=0.5398, pruned_loss=0.2883, over 8328.00 frames. ], tot_loss[loss=0.4855, simple_loss=0.4799, pruned_loss=0.2455, over 1610314.07 frames. ], batch size: 25, lr: 4.57e-02, grad_scale: 16.0 +2023-02-05 18:34:05,848 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3301.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:34:08,973 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3306.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:34:09,426 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.757e+02 4.334e+02 5.638e+02 7.160e+02 2.697e+03, threshold=1.128e+03, percent-clipped=10.0 +2023-02-05 18:34:39,416 INFO [train.py:901] (3/4) Epoch 1, batch 3350, loss[loss=0.4671, simple_loss=0.4582, pruned_loss=0.238, over 7934.00 frames. ], tot_loss[loss=0.4839, simple_loss=0.4795, pruned_loss=0.2442, over 1611936.14 frames. ], batch size: 20, lr: 4.56e-02, grad_scale: 16.0 +2023-02-05 18:35:01,953 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3381.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:35:14,980 INFO [train.py:901] (3/4) Epoch 1, batch 3400, loss[loss=0.5149, simple_loss=0.5083, pruned_loss=0.2608, over 8355.00 frames. ], tot_loss[loss=0.4807, simple_loss=0.4778, pruned_loss=0.2418, over 1611166.00 frames. ], batch size: 24, lr: 4.55e-02, grad_scale: 16.0 +2023-02-05 18:35:19,028 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.486e+02 3.960e+02 5.068e+02 6.311e+02 1.481e+03, threshold=1.014e+03, percent-clipped=3.0 +2023-02-05 18:35:23,821 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3414.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:35:26,545 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3418.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:35:36,490 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.18 vs. limit=5.0 +2023-02-05 18:35:43,706 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3443.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:35:48,746 INFO [train.py:901] (3/4) Epoch 1, batch 3450, loss[loss=0.4703, simple_loss=0.4873, pruned_loss=0.2267, over 8252.00 frames. ], tot_loss[loss=0.4775, simple_loss=0.4753, pruned_loss=0.2398, over 1613960.23 frames. ], batch size: 24, lr: 4.54e-02, grad_scale: 16.0 +2023-02-05 18:35:57,466 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.0587, 0.9868, 3.0017, 1.4951, 2.6993, 2.4075, 2.6305, 2.5472], + device='cuda:3'), covar=tensor([0.0289, 0.3590, 0.0286, 0.1334, 0.0459, 0.0505, 0.0452, 0.0547], + device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0254, 0.0110, 0.0146, 0.0127, 0.0127, 0.0124, 0.0141], + device='cuda:3'), out_proj_covar=tensor([5.2483e-05, 1.4952e-04, 7.1997e-05, 9.8125e-05, 7.6931e-05, 7.7708e-05, + 7.8837e-05, 9.0307e-05], device='cuda:3') +2023-02-05 18:35:59,195 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.87 vs. limit=2.0 +2023-02-05 18:36:05,372 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.14 vs. limit=2.0 +2023-02-05 18:36:21,034 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3496.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:36:24,212 INFO [train.py:901] (3/4) Epoch 1, batch 3500, loss[loss=0.4919, simple_loss=0.4975, pruned_loss=0.2432, over 8189.00 frames. ], tot_loss[loss=0.4783, simple_loss=0.4761, pruned_loss=0.2402, over 1610577.23 frames. ], batch size: 23, lr: 4.53e-02, grad_scale: 16.0 +2023-02-05 18:36:28,198 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.305e+02 4.405e+02 5.773e+02 7.537e+02 2.537e+03, threshold=1.155e+03, percent-clipped=7.0 +2023-02-05 18:36:28,428 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8372, 2.0743, 1.7721, 2.7862, 1.4673, 1.4576, 1.6959, 2.0704], + device='cuda:3'), covar=tensor([0.1596, 0.1713, 0.1666, 0.0305, 0.2483, 0.2105, 0.2352, 0.1554], + device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0234, 0.0224, 0.0150, 0.0292, 0.0269, 0.0313, 0.0228], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2023-02-05 18:36:36,234 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-05 18:36:57,810 INFO [train.py:901] (3/4) Epoch 1, batch 3550, loss[loss=0.433, simple_loss=0.4401, pruned_loss=0.2129, over 8281.00 frames. ], tot_loss[loss=0.4757, simple_loss=0.4746, pruned_loss=0.2384, over 1607346.62 frames. ], batch size: 23, lr: 4.51e-02, grad_scale: 16.0 +2023-02-05 18:37:02,112 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3557.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:37:07,182 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3564.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:37:19,184 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3582.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:37:33,301 INFO [train.py:901] (3/4) Epoch 1, batch 3600, loss[loss=0.3964, simple_loss=0.4228, pruned_loss=0.185, over 7543.00 frames. ], tot_loss[loss=0.4834, simple_loss=0.4792, pruned_loss=0.2439, over 1603604.70 frames. ], batch size: 18, lr: 4.50e-02, grad_scale: 16.0 +2023-02-05 18:37:37,961 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.853e+02 4.660e+02 6.337e+02 8.772e+02 4.832e+03, threshold=1.267e+03, percent-clipped=11.0 +2023-02-05 18:37:54,512 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.12 vs. limit=5.0 +2023-02-05 18:38:06,596 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3650.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:38:07,050 INFO [train.py:901] (3/4) Epoch 1, batch 3650, loss[loss=0.5058, simple_loss=0.4666, pruned_loss=0.2725, over 7259.00 frames. ], tot_loss[loss=0.4788, simple_loss=0.4765, pruned_loss=0.2405, over 1606863.65 frames. ], batch size: 16, lr: 4.49e-02, grad_scale: 16.0 +2023-02-05 18:38:19,634 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3670.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:38:23,765 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6597, 1.5858, 2.1951, 1.6964, 1.6458, 2.2428, 0.7692, 1.2260], + device='cuda:3'), covar=tensor([0.0718, 0.0691, 0.0477, 0.0554, 0.0752, 0.0268, 0.1755, 0.0973], + device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0125, 0.0108, 0.0112, 0.0126, 0.0095, 0.0161, 0.0126], + device='cuda:3'), out_proj_covar=tensor([9.2370e-05, 9.6245e-05, 7.7311e-05, 8.1404e-05, 9.4306e-05, 6.5028e-05, + 1.2332e-04, 1.0062e-04], device='cuda:3') +2023-02-05 18:38:31,953 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4851, 1.3170, 2.7870, 1.2494, 2.0346, 3.2257, 2.9703, 2.7078], + device='cuda:3'), covar=tensor([0.2028, 0.2463, 0.0386, 0.2674, 0.1042, 0.0181, 0.0244, 0.0501], + device='cuda:3'), in_proj_covar=tensor([0.0213, 0.0240, 0.0134, 0.0231, 0.0176, 0.0097, 0.0099, 0.0137], + device='cuda:3'), out_proj_covar=tensor([1.6267e-04, 1.7835e-04, 1.1150e-04, 1.6325e-04, 1.4670e-04, 7.7567e-05, + 8.5671e-05, 1.0787e-04], device='cuda:3') +2023-02-05 18:38:36,829 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3694.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:38:37,567 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3695.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:38:40,429 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-05 18:38:41,127 INFO [train.py:901] (3/4) Epoch 1, batch 3700, loss[loss=0.483, simple_loss=0.4798, pruned_loss=0.2431, over 8088.00 frames. ], tot_loss[loss=0.4806, simple_loss=0.4779, pruned_loss=0.2416, over 1602276.50 frames. ], batch size: 21, lr: 4.48e-02, grad_scale: 16.0 +2023-02-05 18:38:45,134 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.178e+02 4.586e+02 6.278e+02 1.050e+03 3.437e+03, threshold=1.256e+03, percent-clipped=14.0 +2023-02-05 18:38:46,738 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([0.3200, 0.8058, 0.9220, 0.1005, 0.5847, 0.7859, 0.0838, 0.9193], + device='cuda:3'), covar=tensor([0.0716, 0.0502, 0.0372, 0.0966, 0.0526, 0.0612, 0.0952, 0.0369], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0077, 0.0069, 0.0096, 0.0074, 0.0081, 0.0101, 0.0074], + device='cuda:3'), out_proj_covar=tensor([6.7548e-05, 5.0918e-05, 4.7601e-05, 7.3518e-05, 5.4900e-05, 5.6439e-05, + 7.2397e-05, 4.8028e-05], device='cuda:3') +2023-02-05 18:39:17,454 INFO [train.py:901] (3/4) Epoch 1, batch 3750, loss[loss=0.4552, simple_loss=0.4448, pruned_loss=0.2328, over 7432.00 frames. ], tot_loss[loss=0.4767, simple_loss=0.4752, pruned_loss=0.2391, over 1600809.10 frames. ], batch size: 17, lr: 4.47e-02, grad_scale: 16.0 +2023-02-05 18:39:18,342 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3752.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:39:27,126 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3765.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:39:35,214 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3777.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:39:51,681 INFO [train.py:901] (3/4) Epoch 1, batch 3800, loss[loss=0.4524, simple_loss=0.4562, pruned_loss=0.2243, over 7809.00 frames. ], tot_loss[loss=0.4752, simple_loss=0.4744, pruned_loss=0.238, over 1604076.45 frames. ], batch size: 20, lr: 4.46e-02, grad_scale: 16.0 +2023-02-05 18:39:55,874 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.457e+02 5.389e+02 6.979e+02 9.091e+02 1.609e+03, threshold=1.396e+03, percent-clipped=5.0 +2023-02-05 18:40:12,368 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2023-02-05 18:40:16,371 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.06 vs. limit=2.0 +2023-02-05 18:40:27,875 INFO [train.py:901] (3/4) Epoch 1, batch 3850, loss[loss=0.4851, simple_loss=0.4695, pruned_loss=0.2504, over 7648.00 frames. ], tot_loss[loss=0.4745, simple_loss=0.4739, pruned_loss=0.2375, over 1608124.55 frames. ], batch size: 19, lr: 4.45e-02, grad_scale: 16.0 +2023-02-05 18:40:46,564 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-05 18:41:01,009 INFO [train.py:901] (3/4) Epoch 1, batch 3900, loss[loss=0.4231, simple_loss=0.4424, pruned_loss=0.2019, over 7926.00 frames. ], tot_loss[loss=0.4717, simple_loss=0.4716, pruned_loss=0.2359, over 1606511.92 frames. ], batch size: 20, lr: 4.44e-02, grad_scale: 16.0 +2023-02-05 18:41:05,008 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.102e+02 5.552e+02 7.100e+02 9.321e+02 1.906e+03, threshold=1.420e+03, percent-clipped=2.0 +2023-02-05 18:41:05,745 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3908.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:41:17,811 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 +2023-02-05 18:41:23,676 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-05 18:41:25,613 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.82 vs. limit=5.0 +2023-02-05 18:41:29,965 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3944.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:41:35,349 INFO [train.py:901] (3/4) Epoch 1, batch 3950, loss[loss=0.4516, simple_loss=0.4625, pruned_loss=0.2203, over 8664.00 frames. ], tot_loss[loss=0.4691, simple_loss=0.4699, pruned_loss=0.2341, over 1604118.43 frames. ], batch size: 34, lr: 4.43e-02, grad_scale: 16.0 +2023-02-05 18:42:10,921 INFO [train.py:901] (3/4) Epoch 1, batch 4000, loss[loss=0.5744, simple_loss=0.5315, pruned_loss=0.3086, over 7270.00 frames. ], tot_loss[loss=0.4693, simple_loss=0.4704, pruned_loss=0.2341, over 1606912.34 frames. ], batch size: 73, lr: 4.42e-02, grad_scale: 8.0 +2023-02-05 18:42:15,524 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.262e+02 4.572e+02 5.687e+02 7.371e+02 1.820e+03, threshold=1.137e+03, percent-clipped=4.0 +2023-02-05 18:42:24,572 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4021.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:42:25,219 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2858, 2.7686, 2.0520, 2.1679, 2.0988, 2.2991, 1.8875, 2.6389], + device='cuda:3'), covar=tensor([0.1656, 0.1247, 0.2139, 0.1052, 0.2230, 0.1409, 0.3212, 0.1415], + device='cuda:3'), in_proj_covar=tensor([0.0218, 0.0157, 0.0254, 0.0155, 0.0224, 0.0187, 0.0257, 0.0197], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2023-02-05 18:42:25,843 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4023.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:42:36,380 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4038.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:42:42,710 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4046.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:42:46,095 INFO [train.py:901] (3/4) Epoch 1, batch 4050, loss[loss=0.4018, simple_loss=0.4305, pruned_loss=0.1865, over 8353.00 frames. ], tot_loss[loss=0.4712, simple_loss=0.4722, pruned_loss=0.235, over 1613746.64 frames. ], batch size: 24, lr: 4.41e-02, grad_scale: 8.0 +2023-02-05 18:43:17,413 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 +2023-02-05 18:43:22,347 INFO [train.py:901] (3/4) Epoch 1, batch 4100, loss[loss=0.405, simple_loss=0.4341, pruned_loss=0.188, over 8339.00 frames. ], tot_loss[loss=0.4699, simple_loss=0.4719, pruned_loss=0.234, over 1616676.81 frames. ], batch size: 25, lr: 4.40e-02, grad_scale: 8.0 +2023-02-05 18:43:26,892 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.479e+02 4.889e+02 6.474e+02 8.616e+02 2.054e+03, threshold=1.295e+03, percent-clipped=5.0 +2023-02-05 18:43:56,550 INFO [train.py:901] (3/4) Epoch 1, batch 4150, loss[loss=0.4579, simple_loss=0.4725, pruned_loss=0.2216, over 8103.00 frames. ], tot_loss[loss=0.4679, simple_loss=0.4709, pruned_loss=0.2324, over 1613922.48 frames. ], batch size: 23, lr: 4.39e-02, grad_scale: 8.0 +2023-02-05 18:43:58,161 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4153.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 18:44:02,268 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4159.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:44:33,523 INFO [train.py:901] (3/4) Epoch 1, batch 4200, loss[loss=0.5005, simple_loss=0.4998, pruned_loss=0.2506, over 8493.00 frames. ], tot_loss[loss=0.4642, simple_loss=0.4688, pruned_loss=0.2299, over 1616705.34 frames. ], batch size: 28, lr: 4.38e-02, grad_scale: 8.0 +2023-02-05 18:44:38,301 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.396e+02 4.057e+02 5.109e+02 6.409e+02 1.525e+03, threshold=1.022e+03, percent-clipped=2.0 +2023-02-05 18:44:44,312 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-05 18:45:04,964 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-05 18:45:07,067 INFO [train.py:901] (3/4) Epoch 1, batch 4250, loss[loss=0.5085, simple_loss=0.4898, pruned_loss=0.2636, over 8325.00 frames. ], tot_loss[loss=0.4673, simple_loss=0.4703, pruned_loss=0.2321, over 1613724.42 frames. ], batch size: 25, lr: 4.36e-02, grad_scale: 8.0 +2023-02-05 18:45:26,610 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4279.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:45:33,325 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4288.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:45:42,864 INFO [train.py:901] (3/4) Epoch 1, batch 4300, loss[loss=0.4008, simple_loss=0.4249, pruned_loss=0.1883, over 7804.00 frames. ], tot_loss[loss=0.4631, simple_loss=0.4674, pruned_loss=0.2294, over 1615967.44 frames. ], batch size: 20, lr: 4.35e-02, grad_scale: 8.0 +2023-02-05 18:45:43,733 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1187, 1.1677, 1.8024, 0.4781, 1.2362, 1.0674, 0.2185, 1.4673], + device='cuda:3'), covar=tensor([0.0649, 0.0447, 0.0232, 0.1269, 0.0755, 0.0627, 0.0958, 0.0438], + device='cuda:3'), in_proj_covar=tensor([0.0111, 0.0085, 0.0072, 0.0111, 0.0081, 0.0098, 0.0106, 0.0081], + device='cuda:3'), out_proj_covar=tensor([7.8775e-05, 5.7395e-05, 5.0760e-05, 8.8134e-05, 6.3493e-05, 6.9266e-05, + 7.8165e-05, 5.6099e-05], device='cuda:3') +2023-02-05 18:45:45,736 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4304.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:45:47,031 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4306.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:45:48,885 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.647e+02 4.666e+02 6.207e+02 8.078e+02 1.600e+03, threshold=1.241e+03, percent-clipped=6.0 +2023-02-05 18:46:03,145 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 +2023-02-05 18:46:18,293 INFO [train.py:901] (3/4) Epoch 1, batch 4350, loss[loss=0.4551, simple_loss=0.4397, pruned_loss=0.2352, over 7724.00 frames. ], tot_loss[loss=0.4617, simple_loss=0.466, pruned_loss=0.2287, over 1609611.51 frames. ], batch size: 18, lr: 4.34e-02, grad_scale: 8.0 +2023-02-05 18:46:37,364 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-05 18:46:38,879 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6405, 1.6203, 1.8618, 1.5654, 1.1890, 1.9319, 0.3957, 1.0336], + device='cuda:3'), covar=tensor([0.0837, 0.0542, 0.0464, 0.0484, 0.0690, 0.0400, 0.1796, 0.0948], + device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0116, 0.0109, 0.0121, 0.0125, 0.0092, 0.0164, 0.0139], + device='cuda:3'), out_proj_covar=tensor([1.1179e-04, 9.4067e-05, 8.2953e-05, 8.9767e-05, 9.7724e-05, 6.7258e-05, + 1.2693e-04, 1.1050e-04], device='cuda:3') +2023-02-05 18:46:46,376 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 +2023-02-05 18:46:52,945 INFO [train.py:901] (3/4) Epoch 1, batch 4400, loss[loss=0.4757, simple_loss=0.482, pruned_loss=0.2347, over 8667.00 frames. ], tot_loss[loss=0.4592, simple_loss=0.4645, pruned_loss=0.227, over 1610456.93 frames. ], batch size: 34, lr: 4.33e-02, grad_scale: 8.0 +2023-02-05 18:46:54,525 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4403.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:46:57,924 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.833e+02 4.338e+02 5.789e+02 7.262e+02 1.136e+03, threshold=1.158e+03, percent-clipped=0.0 +2023-02-05 18:46:58,935 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4409.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:47:18,598 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4434.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 18:47:21,208 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-05 18:47:22,739 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1680, 1.4458, 1.6320, 1.6274, 1.3910, 1.1210, 1.4276, 1.4584], + device='cuda:3'), covar=tensor([0.2512, 0.1040, 0.0670, 0.0691, 0.0877, 0.1508, 0.1121, 0.0924], + device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0107, 0.0078, 0.0088, 0.0112, 0.0125, 0.0131, 0.0127], + device='cuda:3'), out_proj_covar=tensor([1.0147e-04, 6.1722e-05, 4.2985e-05, 5.0462e-05, 6.2690e-05, 6.9145e-05, + 7.3832e-05, 6.8300e-05], device='cuda:3') +2023-02-05 18:47:29,979 INFO [train.py:901] (3/4) Epoch 1, batch 4450, loss[loss=0.3802, simple_loss=0.3963, pruned_loss=0.182, over 7792.00 frames. ], tot_loss[loss=0.4567, simple_loss=0.4626, pruned_loss=0.2254, over 1609605.10 frames. ], batch size: 19, lr: 4.32e-02, grad_scale: 8.0 +2023-02-05 18:48:04,127 INFO [train.py:901] (3/4) Epoch 1, batch 4500, loss[loss=0.4292, simple_loss=0.4578, pruned_loss=0.2002, over 8487.00 frames. ], tot_loss[loss=0.4565, simple_loss=0.4624, pruned_loss=0.2253, over 1608119.81 frames. ], batch size: 28, lr: 4.31e-02, grad_scale: 8.0 +2023-02-05 18:48:05,594 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4503.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:48:09,058 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.251e+02 4.383e+02 5.863e+02 8.313e+02 2.632e+03, threshold=1.173e+03, percent-clipped=9.0 +2023-02-05 18:48:15,316 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-05 18:48:41,806 INFO [train.py:901] (3/4) Epoch 1, batch 4550, loss[loss=0.431, simple_loss=0.4616, pruned_loss=0.2003, over 8132.00 frames. ], tot_loss[loss=0.4517, simple_loss=0.4597, pruned_loss=0.2218, over 1612268.19 frames. ], batch size: 22, lr: 4.30e-02, grad_scale: 8.0 +2023-02-05 18:49:16,702 INFO [train.py:901] (3/4) Epoch 1, batch 4600, loss[loss=0.5922, simple_loss=0.5499, pruned_loss=0.3172, over 7013.00 frames. ], tot_loss[loss=0.4541, simple_loss=0.4615, pruned_loss=0.2233, over 1613326.42 frames. ], batch size: 71, lr: 4.29e-02, grad_scale: 8.0 +2023-02-05 18:49:17,597 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2421, 1.2883, 3.7493, 1.8567, 1.9269, 4.4822, 3.6133, 4.2359], + device='cuda:3'), covar=tensor([0.1548, 0.2179, 0.0240, 0.2095, 0.1116, 0.0177, 0.0494, 0.0269], + device='cuda:3'), in_proj_covar=tensor([0.0222, 0.0246, 0.0134, 0.0237, 0.0177, 0.0103, 0.0102, 0.0142], + device='cuda:3'), out_proj_covar=tensor([1.7746e-04, 1.9206e-04, 1.1864e-04, 1.7895e-04, 1.5860e-04, 8.5862e-05, + 9.3240e-05, 1.2081e-04], device='cuda:3') +2023-02-05 18:49:21,483 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.209e+02 3.983e+02 5.037e+02 6.922e+02 1.236e+03, threshold=1.007e+03, percent-clipped=2.0 +2023-02-05 18:49:28,486 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4618.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:49:37,336 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6534, 1.0653, 3.0744, 1.3886, 2.2791, 3.2938, 3.0573, 2.9676], + device='cuda:3'), covar=tensor([0.1693, 0.2402, 0.0268, 0.2412, 0.0833, 0.0260, 0.0342, 0.0411], + device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0257, 0.0139, 0.0245, 0.0182, 0.0107, 0.0105, 0.0147], + device='cuda:3'), out_proj_covar=tensor([1.8115e-04, 1.9991e-04, 1.2458e-04, 1.8525e-04, 1.6297e-04, 8.9528e-05, + 9.6566e-05, 1.2484e-04], device='cuda:3') +2023-02-05 18:49:51,614 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4650.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:49:52,218 INFO [train.py:901] (3/4) Epoch 1, batch 4650, loss[loss=0.4842, simple_loss=0.4869, pruned_loss=0.2407, over 8338.00 frames. ], tot_loss[loss=0.4532, simple_loss=0.4611, pruned_loss=0.2226, over 1617142.11 frames. ], batch size: 26, lr: 4.28e-02, grad_scale: 8.0 +2023-02-05 18:49:59,116 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4659.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:50:16,189 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4684.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:50:20,057 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.87 vs. limit=2.0 +2023-02-05 18:50:24,978 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4384, 1.1985, 4.2914, 2.1621, 3.9350, 3.6203, 3.6401, 3.7549], + device='cuda:3'), covar=tensor([0.0136, 0.3393, 0.0207, 0.0908, 0.0252, 0.0233, 0.0287, 0.0282], + device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0300, 0.0139, 0.0171, 0.0153, 0.0153, 0.0146, 0.0164], + device='cuda:3'), out_proj_covar=tensor([6.8514e-05, 1.7204e-04, 8.7556e-05, 1.1607e-04, 9.0235e-05, 9.4652e-05, + 9.1922e-05, 1.0867e-04], device='cuda:3') +2023-02-05 18:50:27,580 INFO [train.py:901] (3/4) Epoch 1, batch 4700, loss[loss=0.5101, simple_loss=0.5145, pruned_loss=0.2528, over 8583.00 frames. ], tot_loss[loss=0.4513, simple_loss=0.4597, pruned_loss=0.2214, over 1613778.32 frames. ], batch size: 34, lr: 4.27e-02, grad_scale: 8.0 +2023-02-05 18:50:27,811 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3746, 1.7367, 1.2179, 1.8996, 1.1913, 1.3140, 1.3730, 1.9100], + device='cuda:3'), covar=tensor([0.1238, 0.0818, 0.1726, 0.0612, 0.1375, 0.1283, 0.1586, 0.0673], + device='cuda:3'), in_proj_covar=tensor([0.0260, 0.0184, 0.0290, 0.0191, 0.0263, 0.0222, 0.0292, 0.0231], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 18:50:32,372 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.254e+02 4.576e+02 5.443e+02 6.674e+02 1.320e+03, threshold=1.089e+03, percent-clipped=4.0 +2023-02-05 18:51:00,649 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2962, 1.4070, 1.2978, 1.1343, 1.7126, 1.1987, 0.9821, 1.7004], + device='cuda:3'), covar=tensor([0.1792, 0.2293, 0.2401, 0.2372, 0.1135, 0.2503, 0.1753, 0.1174], + device='cuda:3'), in_proj_covar=tensor([0.0268, 0.0283, 0.0268, 0.0272, 0.0281, 0.0256, 0.0262, 0.0258], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:3') +2023-02-05 18:51:01,882 INFO [train.py:901] (3/4) Epoch 1, batch 4750, loss[loss=0.506, simple_loss=0.4794, pruned_loss=0.2663, over 6983.00 frames. ], tot_loss[loss=0.4488, simple_loss=0.4573, pruned_loss=0.2201, over 1611730.71 frames. ], batch size: 72, lr: 4.26e-02, grad_scale: 8.0 +2023-02-05 18:51:12,296 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4765.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:51:21,697 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-05 18:51:23,849 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-05 18:51:37,814 INFO [train.py:901] (3/4) Epoch 1, batch 4800, loss[loss=0.4584, simple_loss=0.4727, pruned_loss=0.222, over 8311.00 frames. ], tot_loss[loss=0.4472, simple_loss=0.4561, pruned_loss=0.2191, over 1610027.99 frames. ], batch size: 25, lr: 4.25e-02, grad_scale: 8.0 +2023-02-05 18:51:42,622 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.690e+02 4.367e+02 5.327e+02 7.244e+02 1.939e+03, threshold=1.065e+03, percent-clipped=6.0 +2023-02-05 18:52:11,418 INFO [train.py:901] (3/4) Epoch 1, batch 4850, loss[loss=0.3715, simple_loss=0.3933, pruned_loss=0.1748, over 7527.00 frames. ], tot_loss[loss=0.446, simple_loss=0.4548, pruned_loss=0.2186, over 1613187.64 frames. ], batch size: 18, lr: 4.24e-02, grad_scale: 8.0 +2023-02-05 18:52:13,500 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-05 18:52:21,868 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7745, 2.0975, 1.2495, 2.1323, 1.8783, 1.4905, 1.5341, 2.2072], + device='cuda:3'), covar=tensor([0.1322, 0.0750, 0.1833, 0.0679, 0.1303, 0.1419, 0.2166, 0.0846], + device='cuda:3'), in_proj_covar=tensor([0.0265, 0.0186, 0.0299, 0.0202, 0.0276, 0.0227, 0.0298, 0.0229], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 18:52:27,480 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4874.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:52:47,404 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4899.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:52:48,546 INFO [train.py:901] (3/4) Epoch 1, batch 4900, loss[loss=0.4258, simple_loss=0.4261, pruned_loss=0.2128, over 6811.00 frames. ], tot_loss[loss=0.4436, simple_loss=0.4532, pruned_loss=0.217, over 1611133.20 frames. ], batch size: 15, lr: 4.23e-02, grad_scale: 8.0 +2023-02-05 18:52:53,371 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.332e+02 4.394e+02 5.447e+02 6.722e+02 1.310e+03, threshold=1.089e+03, percent-clipped=5.0 +2023-02-05 18:53:05,360 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1999, 1.7535, 1.7964, 1.5331, 1.1822, 1.8844, 0.3765, 0.7652], + device='cuda:3'), covar=tensor([0.1014, 0.0573, 0.0387, 0.0390, 0.0780, 0.0411, 0.1418, 0.0843], + device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0111, 0.0103, 0.0125, 0.0123, 0.0093, 0.0166, 0.0138], + device='cuda:3'), out_proj_covar=tensor([1.1032e-04, 9.3566e-05, 8.1136e-05, 9.3620e-05, 1.0139e-04, 7.2355e-05, + 1.2957e-04, 1.1106e-04], device='cuda:3') +2023-02-05 18:53:08,705 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5192, 2.1428, 2.2529, 2.5243, 1.9505, 1.2716, 1.7603, 1.9830], + device='cuda:3'), covar=tensor([0.2030, 0.0808, 0.0531, 0.0338, 0.0706, 0.1096, 0.0887, 0.0933], + device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0118, 0.0085, 0.0097, 0.0127, 0.0138, 0.0143, 0.0150], + device='cuda:3'), out_proj_covar=tensor([1.1315e-04, 6.8332e-05, 4.8496e-05, 5.4958e-05, 7.0348e-05, 7.9129e-05, + 8.0833e-05, 8.2202e-05], device='cuda:3') +2023-02-05 18:53:22,696 INFO [train.py:901] (3/4) Epoch 1, batch 4950, loss[loss=0.6146, simple_loss=0.5586, pruned_loss=0.3353, over 8658.00 frames. ], tot_loss[loss=0.4437, simple_loss=0.4533, pruned_loss=0.2171, over 1613159.43 frames. ], batch size: 34, lr: 4.21e-02, grad_scale: 8.0 +2023-02-05 18:53:59,103 INFO [train.py:901] (3/4) Epoch 1, batch 5000, loss[loss=0.518, simple_loss=0.5004, pruned_loss=0.2678, over 8559.00 frames. ], tot_loss[loss=0.4451, simple_loss=0.4542, pruned_loss=0.218, over 1608068.41 frames. ], batch size: 49, lr: 4.20e-02, grad_scale: 8.0 +2023-02-05 18:54:04,640 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.658e+02 4.358e+02 5.438e+02 7.182e+02 1.797e+03, threshold=1.088e+03, percent-clipped=3.0 +2023-02-05 18:54:13,641 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5021.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:54:30,644 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5046.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 18:54:33,891 INFO [train.py:901] (3/4) Epoch 1, batch 5050, loss[loss=0.3744, simple_loss=0.4032, pruned_loss=0.1728, over 7659.00 frames. ], tot_loss[loss=0.4422, simple_loss=0.4525, pruned_loss=0.216, over 1609364.37 frames. ], batch size: 19, lr: 4.19e-02, grad_scale: 8.0 +2023-02-05 18:54:50,671 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-05 18:55:08,927 INFO [train.py:901] (3/4) Epoch 1, batch 5100, loss[loss=0.4948, simple_loss=0.4774, pruned_loss=0.2561, over 8032.00 frames. ], tot_loss[loss=0.4438, simple_loss=0.4531, pruned_loss=0.2172, over 1608441.31 frames. ], batch size: 22, lr: 4.18e-02, grad_scale: 8.0 +2023-02-05 18:55:13,606 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.507e+02 4.431e+02 5.257e+02 6.582e+02 1.311e+03, threshold=1.051e+03, percent-clipped=2.0 +2023-02-05 18:55:45,846 INFO [train.py:901] (3/4) Epoch 1, batch 5150, loss[loss=0.4459, simple_loss=0.4596, pruned_loss=0.2161, over 8645.00 frames. ], tot_loss[loss=0.4422, simple_loss=0.4521, pruned_loss=0.2162, over 1608889.44 frames. ], batch size: 39, lr: 4.17e-02, grad_scale: 8.0 +2023-02-05 18:56:06,777 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.87 vs. limit=2.0 +2023-02-05 18:56:19,015 INFO [train.py:901] (3/4) Epoch 1, batch 5200, loss[loss=0.4951, simple_loss=0.4811, pruned_loss=0.2546, over 8646.00 frames. ], tot_loss[loss=0.4442, simple_loss=0.4534, pruned_loss=0.2175, over 1615358.50 frames. ], batch size: 34, lr: 4.16e-02, grad_scale: 8.0 +2023-02-05 18:56:23,577 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.039e+02 3.937e+02 5.264e+02 6.479e+02 1.558e+03, threshold=1.053e+03, percent-clipped=7.0 +2023-02-05 18:56:25,138 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.7091, 1.1557, 4.3840, 2.1882, 4.0944, 3.7211, 3.8879, 3.7898], + device='cuda:3'), covar=tensor([0.0108, 0.3279, 0.0219, 0.1063, 0.0268, 0.0254, 0.0263, 0.0329], + device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0300, 0.0141, 0.0180, 0.0158, 0.0158, 0.0144, 0.0165], + device='cuda:3'), out_proj_covar=tensor([6.6723e-05, 1.7103e-04, 8.9388e-05, 1.1870e-04, 9.2658e-05, 9.6908e-05, + 8.9898e-05, 1.0832e-04], device='cuda:3') +2023-02-05 18:56:48,375 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.55 vs. limit=5.0 +2023-02-05 18:56:51,651 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-05 18:56:55,106 INFO [train.py:901] (3/4) Epoch 1, batch 5250, loss[loss=0.4401, simple_loss=0.46, pruned_loss=0.2101, over 8104.00 frames. ], tot_loss[loss=0.4432, simple_loss=0.4539, pruned_loss=0.2162, over 1622100.54 frames. ], batch size: 23, lr: 4.15e-02, grad_scale: 8.0 +2023-02-05 18:57:11,488 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4946, 1.1599, 2.9247, 1.3697, 1.9509, 3.3406, 2.9002, 2.7075], + device='cuda:3'), covar=tensor([0.1959, 0.2257, 0.0401, 0.2340, 0.1076, 0.0186, 0.0331, 0.0488], + device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0256, 0.0145, 0.0250, 0.0189, 0.0112, 0.0115, 0.0164], + device='cuda:3'), out_proj_covar=tensor([1.9788e-04, 2.0499e-04, 1.3707e-04, 1.9676e-04, 1.7231e-04, 9.8135e-05, + 1.0682e-04, 1.4046e-04], device='cuda:3') +2023-02-05 18:57:28,846 INFO [train.py:901] (3/4) Epoch 1, batch 5300, loss[loss=0.4909, simple_loss=0.4889, pruned_loss=0.2464, over 8351.00 frames. ], tot_loss[loss=0.4416, simple_loss=0.4527, pruned_loss=0.2152, over 1621733.81 frames. ], batch size: 24, lr: 4.14e-02, grad_scale: 8.0 +2023-02-05 18:57:33,641 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.076e+02 4.278e+02 4.955e+02 6.641e+02 1.586e+03, threshold=9.909e+02, percent-clipped=4.0 +2023-02-05 18:57:33,918 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9475, 2.4979, 4.7003, 1.3549, 2.9938, 2.6974, 1.7594, 2.4101], + device='cuda:3'), covar=tensor([0.1071, 0.1336, 0.0199, 0.1342, 0.1118, 0.1573, 0.1158, 0.1252], + device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0241, 0.0200, 0.0264, 0.0300, 0.0308, 0.0260, 0.0286], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 18:57:54,946 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6994, 2.0793, 1.6437, 1.5153, 2.5700, 2.0002, 2.2855, 3.1361], + device='cuda:3'), covar=tensor([0.1815, 0.2418, 0.2443, 0.2680, 0.1635, 0.2195, 0.1829, 0.1162], + device='cuda:3'), in_proj_covar=tensor([0.0267, 0.0282, 0.0277, 0.0279, 0.0277, 0.0260, 0.0264, 0.0257], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:3') +2023-02-05 18:58:04,343 INFO [train.py:901] (3/4) Epoch 1, batch 5350, loss[loss=0.4199, simple_loss=0.4328, pruned_loss=0.2035, over 8285.00 frames. ], tot_loss[loss=0.4397, simple_loss=0.4515, pruned_loss=0.214, over 1622130.21 frames. ], batch size: 23, lr: 4.13e-02, grad_scale: 8.0 +2023-02-05 18:58:15,393 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.07 vs. limit=2.0 +2023-02-05 18:58:39,824 INFO [train.py:901] (3/4) Epoch 1, batch 5400, loss[loss=0.4044, simple_loss=0.4285, pruned_loss=0.1901, over 8235.00 frames. ], tot_loss[loss=0.4375, simple_loss=0.4499, pruned_loss=0.2126, over 1623070.45 frames. ], batch size: 22, lr: 4.12e-02, grad_scale: 8.0 +2023-02-05 18:58:44,298 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.977e+02 4.515e+02 5.788e+02 7.308e+02 1.362e+03, threshold=1.158e+03, percent-clipped=5.0 +2023-02-05 18:59:13,400 INFO [train.py:901] (3/4) Epoch 1, batch 5450, loss[loss=0.4583, simple_loss=0.4698, pruned_loss=0.2234, over 8351.00 frames. ], tot_loss[loss=0.4349, simple_loss=0.4483, pruned_loss=0.2108, over 1623818.58 frames. ], batch size: 25, lr: 4.11e-02, grad_scale: 8.0 +2023-02-05 18:59:41,788 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-05 18:59:49,950 INFO [train.py:901] (3/4) Epoch 1, batch 5500, loss[loss=0.4353, simple_loss=0.4463, pruned_loss=0.2121, over 8131.00 frames. ], tot_loss[loss=0.4375, simple_loss=0.4501, pruned_loss=0.2124, over 1625463.56 frames. ], batch size: 22, lr: 4.10e-02, grad_scale: 8.0 +2023-02-05 18:59:54,516 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.397e+02 4.451e+02 5.295e+02 6.340e+02 1.239e+03, threshold=1.059e+03, percent-clipped=2.0 +2023-02-05 19:00:21,030 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6983, 2.2827, 4.6471, 2.9061, 4.3996, 4.0700, 4.1877, 4.3241], + device='cuda:3'), covar=tensor([0.0124, 0.2226, 0.0140, 0.0776, 0.0201, 0.0173, 0.0167, 0.0159], + device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0306, 0.0149, 0.0184, 0.0169, 0.0165, 0.0149, 0.0168], + device='cuda:3'), out_proj_covar=tensor([7.0860e-05, 1.7314e-04, 9.3464e-05, 1.2142e-04, 9.8619e-05, 1.0081e-04, + 9.2900e-05, 1.0932e-04], device='cuda:3') +2023-02-05 19:00:23,629 INFO [train.py:901] (3/4) Epoch 1, batch 5550, loss[loss=0.3931, simple_loss=0.4077, pruned_loss=0.1893, over 7778.00 frames. ], tot_loss[loss=0.4375, simple_loss=0.45, pruned_loss=0.2126, over 1621533.18 frames. ], batch size: 19, lr: 4.09e-02, grad_scale: 8.0 +2023-02-05 19:00:28,155 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 +2023-02-05 19:01:00,924 INFO [train.py:901] (3/4) Epoch 1, batch 5600, loss[loss=0.4484, simple_loss=0.4586, pruned_loss=0.2191, over 8322.00 frames. ], tot_loss[loss=0.4367, simple_loss=0.4489, pruned_loss=0.2122, over 1617817.30 frames. ], batch size: 25, lr: 4.08e-02, grad_scale: 8.0 +2023-02-05 19:01:05,771 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.857e+02 3.916e+02 5.301e+02 6.582e+02 1.340e+03, threshold=1.060e+03, percent-clipped=3.0 +2023-02-05 19:01:26,253 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 +2023-02-05 19:01:34,543 INFO [train.py:901] (3/4) Epoch 1, batch 5650, loss[loss=0.4044, simple_loss=0.4024, pruned_loss=0.2032, over 7531.00 frames. ], tot_loss[loss=0.4371, simple_loss=0.4491, pruned_loss=0.2125, over 1617585.97 frames. ], batch size: 18, lr: 4.07e-02, grad_scale: 8.0 +2023-02-05 19:01:45,700 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-05 19:01:45,832 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=5668.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 19:02:06,765 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7827, 2.3290, 1.4236, 2.2206, 1.9285, 1.6829, 1.6582, 2.4785], + device='cuda:3'), covar=tensor([0.1616, 0.0736, 0.1712, 0.0889, 0.1460, 0.1319, 0.2307, 0.0763], + device='cuda:3'), in_proj_covar=tensor([0.0280, 0.0197, 0.0316, 0.0229, 0.0286, 0.0240, 0.0305, 0.0246], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 19:02:09,329 INFO [train.py:901] (3/4) Epoch 1, batch 5700, loss[loss=0.4543, simple_loss=0.4666, pruned_loss=0.221, over 8741.00 frames. ], tot_loss[loss=0.439, simple_loss=0.4497, pruned_loss=0.2142, over 1612390.36 frames. ], batch size: 40, lr: 4.06e-02, grad_scale: 8.0 +2023-02-05 19:02:15,264 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.140e+02 4.740e+02 5.744e+02 8.008e+02 1.790e+03, threshold=1.149e+03, percent-clipped=10.0 +2023-02-05 19:02:28,016 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.60 vs. limit=5.0 +2023-02-05 19:02:44,479 INFO [train.py:901] (3/4) Epoch 1, batch 5750, loss[loss=0.4178, simple_loss=0.4464, pruned_loss=0.1946, over 8139.00 frames. ], tot_loss[loss=0.435, simple_loss=0.447, pruned_loss=0.2115, over 1610890.00 frames. ], batch size: 22, lr: 4.05e-02, grad_scale: 8.0 +2023-02-05 19:02:51,405 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-05 19:02:59,846 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=5773.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:03:16,849 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2948, 1.6606, 1.3480, 2.0139, 1.4283, 1.2006, 1.2279, 2.0112], + device='cuda:3'), covar=tensor([0.1255, 0.0680, 0.1484, 0.0586, 0.1394, 0.1291, 0.1643, 0.0738], + device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0207, 0.0322, 0.0236, 0.0290, 0.0244, 0.0307, 0.0253], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 19:03:19,619 INFO [train.py:901] (3/4) Epoch 1, batch 5800, loss[loss=0.4427, simple_loss=0.4582, pruned_loss=0.2136, over 8324.00 frames. ], tot_loss[loss=0.4331, simple_loss=0.4462, pruned_loss=0.21, over 1609196.37 frames. ], batch size: 25, lr: 4.04e-02, grad_scale: 8.0 +2023-02-05 19:03:24,530 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.458e+02 3.671e+02 4.595e+02 5.667e+02 1.405e+03, threshold=9.190e+02, percent-clipped=2.0 +2023-02-05 19:03:48,471 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2023-02-05 19:03:57,248 INFO [train.py:901] (3/4) Epoch 1, batch 5850, loss[loss=0.4272, simple_loss=0.4418, pruned_loss=0.2063, over 8080.00 frames. ], tot_loss[loss=0.4326, simple_loss=0.4461, pruned_loss=0.2095, over 1609670.55 frames. ], batch size: 21, lr: 4.03e-02, grad_scale: 8.0 +2023-02-05 19:04:15,207 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=5876.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:04:21,567 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0921, 1.2567, 2.0160, 0.3793, 1.6034, 1.1244, 0.4392, 1.4242], + device='cuda:3'), covar=tensor([0.0541, 0.0272, 0.0189, 0.0998, 0.0621, 0.0659, 0.0922, 0.0315], + device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0105, 0.0084, 0.0140, 0.0103, 0.0148, 0.0144, 0.0110], + device='cuda:3'), out_proj_covar=tensor([1.0000e-04, 7.5911e-05, 6.4379e-05, 1.1092e-04, 8.4542e-05, 1.1129e-04, + 1.1165e-04, 7.9108e-05], device='cuda:3') +2023-02-05 19:04:32,488 INFO [train.py:901] (3/4) Epoch 1, batch 5900, loss[loss=0.4201, simple_loss=0.4376, pruned_loss=0.2014, over 8292.00 frames. ], tot_loss[loss=0.4302, simple_loss=0.4443, pruned_loss=0.2081, over 1609159.36 frames. ], batch size: 23, lr: 4.02e-02, grad_scale: 8.0 +2023-02-05 19:04:37,230 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.095e+02 4.155e+02 5.559e+02 6.668e+02 2.372e+03, threshold=1.112e+03, percent-clipped=6.0 +2023-02-05 19:05:09,348 INFO [train.py:901] (3/4) Epoch 1, batch 5950, loss[loss=0.4437, simple_loss=0.4347, pruned_loss=0.2263, over 7524.00 frames. ], tot_loss[loss=0.4284, simple_loss=0.4432, pruned_loss=0.2068, over 1612280.16 frames. ], batch size: 18, lr: 4.01e-02, grad_scale: 8.0 +2023-02-05 19:05:13,688 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6391, 1.8462, 2.7939, 2.8603, 2.2292, 1.5125, 2.0386, 1.8560], + device='cuda:3'), covar=tensor([0.1693, 0.0983, 0.0265, 0.0286, 0.0550, 0.0751, 0.0576, 0.1030], + device='cuda:3'), in_proj_covar=tensor([0.0270, 0.0172, 0.0119, 0.0139, 0.0178, 0.0184, 0.0190, 0.0211], + device='cuda:3'), out_proj_covar=tensor([1.6108e-04, 1.0591e-04, 7.2145e-05, 8.2285e-05, 1.0167e-04, 1.0933e-04, + 1.0994e-04, 1.2088e-04], device='cuda:3') +2023-02-05 19:05:44,548 INFO [train.py:901] (3/4) Epoch 1, batch 6000, loss[loss=0.3785, simple_loss=0.3995, pruned_loss=0.1787, over 7428.00 frames. ], tot_loss[loss=0.4287, simple_loss=0.4432, pruned_loss=0.2071, over 1614178.52 frames. ], batch size: 17, lr: 4.00e-02, grad_scale: 16.0 +2023-02-05 19:05:44,548 INFO [train.py:926] (3/4) Computing validation loss +2023-02-05 19:06:02,003 INFO [train.py:935] (3/4) Epoch 1, validation: loss=0.3351, simple_loss=0.4011, pruned_loss=0.1346, over 944034.00 frames. +2023-02-05 19:06:02,004 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6514MB +2023-02-05 19:06:06,793 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.323e+02 3.694e+02 4.999e+02 6.330e+02 1.596e+03, threshold=9.998e+02, percent-clipped=5.0 +2023-02-05 19:06:06,994 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6008.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:06:09,469 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6012.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 19:06:28,692 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 +2023-02-05 19:06:35,737 INFO [train.py:901] (3/4) Epoch 1, batch 6050, loss[loss=0.5255, simple_loss=0.5106, pruned_loss=0.2702, over 8339.00 frames. ], tot_loss[loss=0.4351, simple_loss=0.4475, pruned_loss=0.2113, over 1612275.06 frames. ], batch size: 26, lr: 3.99e-02, grad_scale: 8.0 +2023-02-05 19:06:42,864 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6061.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:07:12,047 INFO [train.py:901] (3/4) Epoch 1, batch 6100, loss[loss=0.4014, simple_loss=0.4319, pruned_loss=0.1855, over 8245.00 frames. ], tot_loss[loss=0.4324, simple_loss=0.4461, pruned_loss=0.2093, over 1613133.52 frames. ], batch size: 24, lr: 3.98e-02, grad_scale: 8.0 +2023-02-05 19:07:17,503 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.508e+02 4.942e+02 6.048e+02 7.564e+02 1.774e+03, threshold=1.210e+03, percent-clipped=15.0 +2023-02-05 19:07:23,143 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6117.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:07:29,001 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-05 19:07:29,786 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6127.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 19:07:45,976 INFO [train.py:901] (3/4) Epoch 1, batch 6150, loss[loss=0.3728, simple_loss=0.4196, pruned_loss=0.163, over 8456.00 frames. ], tot_loss[loss=0.4309, simple_loss=0.4452, pruned_loss=0.2083, over 1618112.80 frames. ], batch size: 27, lr: 3.97e-02, grad_scale: 8.0 +2023-02-05 19:07:47,408 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6153.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:08:12,846 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6188.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:08:14,904 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2117, 1.2041, 1.1965, 1.8544, 0.8377, 0.9913, 1.0658, 1.1170], + device='cuda:3'), covar=tensor([0.1338, 0.1565, 0.1675, 0.0452, 0.2050, 0.2429, 0.1925, 0.1284], + device='cuda:3'), in_proj_covar=tensor([0.0299, 0.0314, 0.0295, 0.0192, 0.0339, 0.0336, 0.0390, 0.0280], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-05 19:08:22,934 INFO [train.py:901] (3/4) Epoch 1, batch 6200, loss[loss=0.4474, simple_loss=0.4584, pruned_loss=0.2182, over 8741.00 frames. ], tot_loss[loss=0.4298, simple_loss=0.4439, pruned_loss=0.2079, over 1611148.22 frames. ], batch size: 49, lr: 3.96e-02, grad_scale: 8.0 +2023-02-05 19:08:28,568 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.743e+02 4.155e+02 5.130e+02 7.106e+02 1.864e+03, threshold=1.026e+03, percent-clipped=2.0 +2023-02-05 19:08:36,408 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6220.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:08:37,190 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6221.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:08:42,653 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6229.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:08:44,652 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6232.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:08:57,426 INFO [train.py:901] (3/4) Epoch 1, batch 6250, loss[loss=0.4511, simple_loss=0.46, pruned_loss=0.2211, over 8476.00 frames. ], tot_loss[loss=0.4277, simple_loss=0.4434, pruned_loss=0.206, over 1612732.00 frames. ], batch size: 29, lr: 3.95e-02, grad_scale: 8.0 +2023-02-05 19:09:10,719 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 +2023-02-05 19:09:15,330 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.31 vs. limit=5.0 +2023-02-05 19:09:20,019 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6284.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:09:32,703 INFO [train.py:901] (3/4) Epoch 1, batch 6300, loss[loss=0.4242, simple_loss=0.4198, pruned_loss=0.2143, over 7545.00 frames. ], tot_loss[loss=0.4258, simple_loss=0.4426, pruned_loss=0.2045, over 1614579.96 frames. ], batch size: 18, lr: 3.94e-02, grad_scale: 8.0 +2023-02-05 19:09:38,778 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.783e+02 4.352e+02 5.159e+02 6.362e+02 1.735e+03, threshold=1.032e+03, percent-clipped=4.0 +2023-02-05 19:09:56,806 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6335.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:10:07,346 INFO [train.py:901] (3/4) Epoch 1, batch 6350, loss[loss=0.4245, simple_loss=0.4358, pruned_loss=0.2066, over 8133.00 frames. ], tot_loss[loss=0.4254, simple_loss=0.442, pruned_loss=0.2044, over 1616424.71 frames. ], batch size: 22, lr: 3.93e-02, grad_scale: 8.0 +2023-02-05 19:10:07,538 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([0.7687, 1.2785, 1.3851, 1.0808, 0.9212, 1.3073, 0.1205, 0.8535], + device='cuda:3'), covar=tensor([0.0693, 0.0564, 0.0294, 0.0424, 0.0586, 0.0403, 0.1352, 0.0669], + device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0114, 0.0101, 0.0132, 0.0122, 0.0085, 0.0176, 0.0147], + device='cuda:3'), out_proj_covar=tensor([1.1777e-04, 1.0304e-04, 8.2847e-05, 1.0529e-04, 1.0752e-04, 7.0384e-05, + 1.4574e-04, 1.2583e-04], device='cuda:3') +2023-02-05 19:10:08,107 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6352.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:10:28,447 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-05 19:10:28,929 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6383.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 19:10:38,776 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.9051, 4.1676, 3.3737, 1.5667, 3.4363, 3.3616, 3.5788, 2.8071], + device='cuda:3'), covar=tensor([0.0914, 0.0427, 0.0769, 0.3304, 0.0453, 0.0512, 0.0851, 0.0559], + device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0179, 0.0207, 0.0268, 0.0164, 0.0128, 0.0192, 0.0120], + device='cuda:3'), out_proj_covar=tensor([1.8633e-04, 1.2836e-04, 1.3568e-04, 1.7520e-04, 1.0601e-04, 9.2311e-05, + 1.3439e-04, 8.7719e-05], device='cuda:3') +2023-02-05 19:10:40,805 INFO [train.py:901] (3/4) Epoch 1, batch 6400, loss[loss=0.4602, simple_loss=0.4748, pruned_loss=0.2228, over 8473.00 frames. ], tot_loss[loss=0.4251, simple_loss=0.4411, pruned_loss=0.2045, over 1612745.88 frames. ], batch size: 25, lr: 3.92e-02, grad_scale: 8.0 +2023-02-05 19:10:43,627 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6405.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:10:45,785 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6408.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 19:10:46,252 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.785e+02 4.017e+02 4.991e+02 6.603e+02 1.156e+03, threshold=9.981e+02, percent-clipped=3.0 +2023-02-05 19:10:59,209 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.24 vs. limit=2.0 +2023-02-05 19:11:16,786 INFO [train.py:901] (3/4) Epoch 1, batch 6450, loss[loss=0.4759, simple_loss=0.4697, pruned_loss=0.241, over 8363.00 frames. ], tot_loss[loss=0.4247, simple_loss=0.4412, pruned_loss=0.2041, over 1614013.89 frames. ], batch size: 24, lr: 3.91e-02, grad_scale: 8.0 +2023-02-05 19:11:27,836 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6467.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:11:33,199 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.5174, 1.3040, 3.5331, 1.6668, 3.0421, 2.8393, 3.0949, 3.0381], + device='cuda:3'), covar=tensor([0.0264, 0.3048, 0.0246, 0.1293, 0.0531, 0.0379, 0.0281, 0.0417], + device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0334, 0.0165, 0.0205, 0.0201, 0.0184, 0.0158, 0.0191], + device='cuda:3'), out_proj_covar=tensor([8.4483e-05, 1.8558e-04, 1.0315e-04, 1.3213e-04, 1.1496e-04, 1.1048e-04, + 9.6604e-05, 1.2205e-04], device='cuda:3') +2023-02-05 19:11:35,915 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7609, 2.0980, 1.0357, 2.0604, 2.0203, 1.2921, 1.4487, 2.4654], + device='cuda:3'), covar=tensor([0.1375, 0.0660, 0.1857, 0.0739, 0.0924, 0.1302, 0.1620, 0.0685], + device='cuda:3'), in_proj_covar=tensor([0.0321, 0.0217, 0.0331, 0.0259, 0.0303, 0.0270, 0.0325, 0.0274], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2023-02-05 19:11:36,486 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6480.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:11:39,741 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6485.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:11:41,926 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6488.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:11:47,728 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6497.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:11:50,293 INFO [train.py:901] (3/4) Epoch 1, batch 6500, loss[loss=0.5012, simple_loss=0.4898, pruned_loss=0.2563, over 8097.00 frames. ], tot_loss[loss=0.4244, simple_loss=0.4414, pruned_loss=0.2037, over 1618832.37 frames. ], batch size: 23, lr: 3.90e-02, grad_scale: 8.0 +2023-02-05 19:11:55,445 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.401e+02 4.204e+02 5.270e+02 6.161e+02 1.286e+03, threshold=1.054e+03, percent-clipped=6.0 +2023-02-05 19:11:58,510 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6513.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:12:03,316 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6520.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:12:11,141 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6532.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:12:24,034 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 +2023-02-05 19:12:25,068 INFO [train.py:901] (3/4) Epoch 1, batch 6550, loss[loss=0.5103, simple_loss=0.4974, pruned_loss=0.2615, over 7106.00 frames. ], tot_loss[loss=0.4248, simple_loss=0.442, pruned_loss=0.2039, over 1617032.27 frames. ], batch size: 72, lr: 3.89e-02, grad_scale: 8.0 +2023-02-05 19:12:29,525 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=5.27 vs. limit=5.0 +2023-02-05 19:12:35,940 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6565.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:12:37,933 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-05 19:12:41,475 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6573.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:12:53,798 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6591.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:12:57,649 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-05 19:13:00,396 INFO [train.py:901] (3/4) Epoch 1, batch 6600, loss[loss=0.454, simple_loss=0.4811, pruned_loss=0.2135, over 8357.00 frames. ], tot_loss[loss=0.4221, simple_loss=0.4408, pruned_loss=0.2018, over 1622166.34 frames. ], batch size: 24, lr: 3.89e-02, grad_scale: 8.0 +2023-02-05 19:13:05,685 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.999e+02 4.035e+02 4.985e+02 6.404e+02 1.328e+03, threshold=9.970e+02, percent-clipped=3.0 +2023-02-05 19:13:07,911 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6612.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:13:10,625 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6616.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:13:18,648 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6628.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:13:31,572 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6647.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:13:31,607 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6647.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:13:34,158 INFO [train.py:901] (3/4) Epoch 1, batch 6650, loss[loss=0.4425, simple_loss=0.4652, pruned_loss=0.2099, over 8700.00 frames. ], tot_loss[loss=0.4192, simple_loss=0.4388, pruned_loss=0.1998, over 1622276.70 frames. ], batch size: 34, lr: 3.88e-02, grad_scale: 8.0 +2023-02-05 19:13:42,280 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7539, 2.6397, 1.2768, 2.7970, 2.2505, 1.8688, 1.7966, 2.8189], + device='cuda:3'), covar=tensor([0.1916, 0.0889, 0.1899, 0.0983, 0.1351, 0.1486, 0.2274, 0.1139], + device='cuda:3'), in_proj_covar=tensor([0.0321, 0.0220, 0.0340, 0.0261, 0.0308, 0.0280, 0.0325, 0.0275], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2023-02-05 19:13:56,261 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6680.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:14:01,307 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6688.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:14:09,932 INFO [train.py:901] (3/4) Epoch 1, batch 6700, loss[loss=0.4447, simple_loss=0.4722, pruned_loss=0.2087, over 8193.00 frames. ], tot_loss[loss=0.4187, simple_loss=0.4379, pruned_loss=0.1997, over 1619187.82 frames. ], batch size: 23, lr: 3.87e-02, grad_scale: 8.0 +2023-02-05 19:14:15,398 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.351e+02 4.140e+02 4.960e+02 6.260e+02 1.494e+03, threshold=9.921e+02, percent-clipped=3.0 +2023-02-05 19:14:25,037 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6723.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:14:26,598 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-05 19:14:38,639 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6743.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:14:42,155 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6748.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:14:44,014 INFO [train.py:901] (3/4) Epoch 1, batch 6750, loss[loss=0.4639, simple_loss=0.4818, pruned_loss=0.223, over 8571.00 frames. ], tot_loss[loss=0.4181, simple_loss=0.4371, pruned_loss=0.1995, over 1616303.21 frames. ], batch size: 31, lr: 3.86e-02, grad_scale: 8.0 +2023-02-05 19:15:00,945 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6776.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:15:14,393 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-05 19:15:19,964 INFO [train.py:901] (3/4) Epoch 1, batch 6800, loss[loss=0.3769, simple_loss=0.4172, pruned_loss=0.1684, over 8101.00 frames. ], tot_loss[loss=0.4158, simple_loss=0.4354, pruned_loss=0.1981, over 1609622.21 frames. ], batch size: 23, lr: 3.85e-02, grad_scale: 8.0 +2023-02-05 19:15:20,167 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6801.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:15:25,324 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.931e+02 4.226e+02 5.434e+02 7.341e+02 1.725e+03, threshold=1.087e+03, percent-clipped=4.0 +2023-02-05 19:15:35,603 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6824.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:15:39,017 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6829.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:15:54,377 INFO [train.py:901] (3/4) Epoch 1, batch 6850, loss[loss=0.4085, simple_loss=0.4348, pruned_loss=0.1911, over 7800.00 frames. ], tot_loss[loss=0.416, simple_loss=0.4358, pruned_loss=0.1981, over 1611633.79 frames. ], batch size: 20, lr: 3.84e-02, grad_scale: 8.0 +2023-02-05 19:15:54,609 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6451, 1.7402, 2.4964, 1.5222, 1.5196, 2.1416, 0.6712, 1.6609], + device='cuda:3'), covar=tensor([0.0938, 0.0607, 0.0514, 0.0525, 0.0863, 0.0823, 0.1633, 0.0877], + device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0106, 0.0098, 0.0129, 0.0117, 0.0083, 0.0162, 0.0132], + device='cuda:3'), out_proj_covar=tensor([1.1180e-04, 9.6890e-05, 8.1399e-05, 1.0407e-04, 1.0467e-04, 6.9712e-05, + 1.3780e-04, 1.1536e-04], device='cuda:3') +2023-02-05 19:16:04,831 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-05 19:16:06,406 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6868.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:16:23,441 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6893.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:16:25,429 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6896.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 19:16:29,306 INFO [train.py:901] (3/4) Epoch 1, batch 6900, loss[loss=0.4047, simple_loss=0.4108, pruned_loss=0.1993, over 7228.00 frames. ], tot_loss[loss=0.4179, simple_loss=0.4371, pruned_loss=0.1993, over 1611732.40 frames. ], batch size: 16, lr: 3.83e-02, grad_scale: 8.0 +2023-02-05 19:16:31,395 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6903.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:16:35,806 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.469e+02 3.796e+02 4.754e+02 6.076e+02 1.448e+03, threshold=9.507e+02, percent-clipped=2.0 +2023-02-05 19:16:48,742 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6927.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:16:49,426 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6928.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:16:54,879 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6936.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:16:56,904 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6939.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:17:00,413 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6944.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:17:00,456 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6944.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:17:05,067 INFO [train.py:901] (3/4) Epoch 1, batch 6950, loss[loss=0.3718, simple_loss=0.4236, pruned_loss=0.16, over 8324.00 frames. ], tot_loss[loss=0.4195, simple_loss=0.4384, pruned_loss=0.2003, over 1606561.03 frames. ], batch size: 25, lr: 3.82e-02, grad_scale: 8.0 +2023-02-05 19:17:11,197 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-05 19:17:12,138 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6961.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:17:17,885 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6969.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:17:32,945 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6991.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:17:38,576 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6999.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:17:39,834 INFO [train.py:901] (3/4) Epoch 1, batch 7000, loss[loss=0.36, simple_loss=0.3982, pruned_loss=0.1609, over 7915.00 frames. ], tot_loss[loss=0.4185, simple_loss=0.4383, pruned_loss=0.1993, over 1609769.37 frames. ], batch size: 20, lr: 3.81e-02, grad_scale: 8.0 +2023-02-05 19:17:45,246 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.380e+02 4.090e+02 4.918e+02 6.048e+02 1.151e+03, threshold=9.836e+02, percent-clipped=6.0 +2023-02-05 19:17:57,728 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7024.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:18:08,481 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 +2023-02-05 19:18:16,024 INFO [train.py:901] (3/4) Epoch 1, batch 7050, loss[loss=0.4197, simple_loss=0.4446, pruned_loss=0.1974, over 8131.00 frames. ], tot_loss[loss=0.4185, simple_loss=0.4381, pruned_loss=0.1994, over 1608104.36 frames. ], batch size: 22, lr: 3.80e-02, grad_scale: 8.0 +2023-02-05 19:18:26,095 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7066.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 19:18:50,234 INFO [train.py:901] (3/4) Epoch 1, batch 7100, loss[loss=0.4291, simple_loss=0.4418, pruned_loss=0.2082, over 8505.00 frames. ], tot_loss[loss=0.4188, simple_loss=0.438, pruned_loss=0.1998, over 1607837.45 frames. ], batch size: 28, lr: 3.79e-02, grad_scale: 8.0 +2023-02-05 19:18:53,868 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7106.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:18:55,758 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.243e+02 3.791e+02 4.613e+02 6.150e+02 1.722e+03, threshold=9.225e+02, percent-clipped=5.0 +2023-02-05 19:19:08,993 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3776, 1.6372, 1.4113, 1.2898, 2.0419, 1.5193, 1.9018, 2.1571], + device='cuda:3'), covar=tensor([0.1413, 0.2172, 0.2658, 0.2352, 0.1256, 0.2175, 0.1383, 0.1041], + device='cuda:3'), in_proj_covar=tensor([0.0273, 0.0292, 0.0297, 0.0280, 0.0273, 0.0261, 0.0258, 0.0252], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2023-02-05 19:19:25,907 INFO [train.py:901] (3/4) Epoch 1, batch 7150, loss[loss=0.4696, simple_loss=0.4532, pruned_loss=0.243, over 7290.00 frames. ], tot_loss[loss=0.4169, simple_loss=0.437, pruned_loss=0.1984, over 1608745.24 frames. ], batch size: 16, lr: 3.78e-02, grad_scale: 8.0 +2023-02-05 19:19:26,118 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([0.8230, 1.2119, 1.8226, 0.9861, 0.9525, 1.5706, 0.1464, 0.8247], + device='cuda:3'), covar=tensor([0.0683, 0.0463, 0.0271, 0.0438, 0.0531, 0.0277, 0.1336, 0.0568], + device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0107, 0.0099, 0.0129, 0.0116, 0.0077, 0.0160, 0.0131], + device='cuda:3'), out_proj_covar=tensor([1.1116e-04, 9.7726e-05, 8.3365e-05, 1.0552e-04, 1.0572e-04, 6.8292e-05, + 1.3667e-04, 1.1411e-04], device='cuda:3') +2023-02-05 19:19:46,604 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7181.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:19:56,059 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7195.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:19:59,481 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7200.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:19:59,951 INFO [train.py:901] (3/4) Epoch 1, batch 7200, loss[loss=0.4302, simple_loss=0.4386, pruned_loss=0.2109, over 7983.00 frames. ], tot_loss[loss=0.4178, simple_loss=0.4381, pruned_loss=0.1988, over 1614154.73 frames. ], batch size: 21, lr: 3.78e-02, grad_scale: 8.0 +2023-02-05 19:20:05,332 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.325e+02 4.231e+02 5.262e+02 7.053e+02 1.685e+03, threshold=1.052e+03, percent-clipped=7.0 +2023-02-05 19:20:13,050 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7220.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:20:16,293 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7225.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:20:22,240 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5776, 2.0608, 3.5017, 0.9407, 2.4054, 1.9043, 1.6564, 2.0271], + device='cuda:3'), covar=tensor([0.1065, 0.1218, 0.0286, 0.1676, 0.1117, 0.1501, 0.0900, 0.1346], + device='cuda:3'), in_proj_covar=tensor([0.0296, 0.0298, 0.0280, 0.0325, 0.0374, 0.0358, 0.0304, 0.0358], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 19:20:25,926 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7240.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 19:20:33,008 INFO [train.py:901] (3/4) Epoch 1, batch 7250, loss[loss=0.4088, simple_loss=0.4413, pruned_loss=0.1881, over 8344.00 frames. ], tot_loss[loss=0.4161, simple_loss=0.4367, pruned_loss=0.1977, over 1617356.65 frames. ], batch size: 26, lr: 3.77e-02, grad_scale: 8.0 +2023-02-05 19:20:49,156 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7271.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:20:58,000 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9377, 2.7550, 1.3173, 2.5357, 2.4511, 1.8569, 2.0377, 2.7744], + device='cuda:3'), covar=tensor([0.1842, 0.0824, 0.1644, 0.1004, 0.1184, 0.1241, 0.1682, 0.1121], + device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0236, 0.0351, 0.0282, 0.0325, 0.0279, 0.0340, 0.0304], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-05 19:21:04,006 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7293.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:21:09,003 INFO [train.py:901] (3/4) Epoch 1, batch 7300, loss[loss=0.4676, simple_loss=0.4748, pruned_loss=0.2302, over 8589.00 frames. ], tot_loss[loss=0.416, simple_loss=0.4366, pruned_loss=0.1977, over 1616224.71 frames. ], batch size: 34, lr: 3.76e-02, grad_scale: 8.0 +2023-02-05 19:21:14,312 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.413e+02 4.263e+02 5.448e+02 6.514e+02 1.215e+03, threshold=1.090e+03, percent-clipped=2.0 +2023-02-05 19:21:29,429 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9787, 2.0392, 1.7780, 2.8474, 1.3104, 1.3151, 2.1633, 2.0606], + device='cuda:3'), covar=tensor([0.1171, 0.1623, 0.1454, 0.0299, 0.2062, 0.2005, 0.1652, 0.1160], + device='cuda:3'), in_proj_covar=tensor([0.0306, 0.0325, 0.0308, 0.0199, 0.0347, 0.0340, 0.0387, 0.0292], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-05 19:21:42,634 INFO [train.py:901] (3/4) Epoch 1, batch 7350, loss[loss=0.414, simple_loss=0.4396, pruned_loss=0.1942, over 8603.00 frames. ], tot_loss[loss=0.4141, simple_loss=0.4353, pruned_loss=0.1965, over 1615605.54 frames. ], batch size: 49, lr: 3.75e-02, grad_scale: 8.0 +2023-02-05 19:21:45,530 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7355.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 19:21:50,108 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7362.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:21:56,032 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-05 19:22:08,451 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7386.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:22:09,131 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7387.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:22:18,191 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-05 19:22:18,967 INFO [train.py:901] (3/4) Epoch 1, batch 7400, loss[loss=0.4148, simple_loss=0.417, pruned_loss=0.2063, over 7793.00 frames. ], tot_loss[loss=0.4128, simple_loss=0.4343, pruned_loss=0.1956, over 1614668.84 frames. ], batch size: 19, lr: 3.74e-02, grad_scale: 8.0 +2023-02-05 19:22:24,409 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.824e+02 4.270e+02 5.603e+02 6.704e+02 2.452e+03, threshold=1.121e+03, percent-clipped=4.0 +2023-02-05 19:22:25,140 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7410.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 19:22:35,066 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7425.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:22:52,555 INFO [train.py:901] (3/4) Epoch 1, batch 7450, loss[loss=0.3993, simple_loss=0.4294, pruned_loss=0.1845, over 8605.00 frames. ], tot_loss[loss=0.4115, simple_loss=0.4335, pruned_loss=0.1948, over 1613957.27 frames. ], batch size: 34, lr: 3.73e-02, grad_scale: 8.0 +2023-02-05 19:22:56,050 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-05 19:23:27,520 INFO [train.py:901] (3/4) Epoch 1, batch 7500, loss[loss=0.3861, simple_loss=0.4179, pruned_loss=0.1772, over 8284.00 frames. ], tot_loss[loss=0.4111, simple_loss=0.4328, pruned_loss=0.1946, over 1613640.33 frames. ], batch size: 23, lr: 3.72e-02, grad_scale: 8.0 +2023-02-05 19:23:34,187 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.398e+02 4.060e+02 5.044e+02 6.934e+02 1.457e+03, threshold=1.009e+03, percent-clipped=3.0 +2023-02-05 19:23:45,047 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7525.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:23:45,156 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7525.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 19:23:51,344 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 +2023-02-05 19:24:02,231 INFO [train.py:901] (3/4) Epoch 1, batch 7550, loss[loss=0.4338, simple_loss=0.46, pruned_loss=0.2038, over 8472.00 frames. ], tot_loss[loss=0.4102, simple_loss=0.4316, pruned_loss=0.1944, over 1610775.79 frames. ], batch size: 25, lr: 3.72e-02, grad_scale: 8.0 +2023-02-05 19:24:36,286 INFO [train.py:901] (3/4) Epoch 1, batch 7600, loss[loss=0.3767, simple_loss=0.4165, pruned_loss=0.1684, over 8509.00 frames. ], tot_loss[loss=0.412, simple_loss=0.4329, pruned_loss=0.1956, over 1613037.76 frames. ], batch size: 26, lr: 3.71e-02, grad_scale: 8.0 +2023-02-05 19:24:41,731 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.765e+02 4.361e+02 5.460e+02 6.853e+02 1.164e+03, threshold=1.092e+03, percent-clipped=2.0 +2023-02-05 19:24:43,944 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7611.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 19:25:03,345 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7636.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 19:25:03,853 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7637.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:25:05,950 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7640.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:25:07,275 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7642.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:25:07,350 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7642.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:25:13,188 INFO [train.py:901] (3/4) Epoch 1, batch 7650, loss[loss=0.3858, simple_loss=0.4045, pruned_loss=0.1836, over 7417.00 frames. ], tot_loss[loss=0.4113, simple_loss=0.4329, pruned_loss=0.1948, over 1613120.06 frames. ], batch size: 17, lr: 3.70e-02, grad_scale: 8.0 +2023-02-05 19:25:23,919 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7667.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:25:30,053 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.30 vs. limit=2.0 +2023-02-05 19:25:46,145 INFO [train.py:901] (3/4) Epoch 1, batch 7700, loss[loss=0.3944, simple_loss=0.4247, pruned_loss=0.182, over 8487.00 frames. ], tot_loss[loss=0.412, simple_loss=0.4327, pruned_loss=0.1957, over 1610710.63 frames. ], batch size: 28, lr: 3.69e-02, grad_scale: 8.0 +2023-02-05 19:25:51,297 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.517e+02 4.083e+02 4.742e+02 6.161e+02 2.101e+03, threshold=9.483e+02, percent-clipped=6.0 +2023-02-05 19:26:07,598 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-05 19:26:21,649 INFO [train.py:901] (3/4) Epoch 1, batch 7750, loss[loss=0.463, simple_loss=0.466, pruned_loss=0.23, over 8728.00 frames. ], tot_loss[loss=0.4112, simple_loss=0.4325, pruned_loss=0.1949, over 1613990.37 frames. ], batch size: 34, lr: 3.68e-02, grad_scale: 8.0 +2023-02-05 19:26:23,165 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7752.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:26:29,122 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7761.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:26:34,419 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7769.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:26:42,672 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7781.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 19:26:56,375 INFO [train.py:901] (3/4) Epoch 1, batch 7800, loss[loss=0.3993, simple_loss=0.4165, pruned_loss=0.1911, over 7974.00 frames. ], tot_loss[loss=0.4094, simple_loss=0.4318, pruned_loss=0.1935, over 1615915.19 frames. ], batch size: 21, lr: 3.67e-02, grad_scale: 8.0 +2023-02-05 19:26:59,822 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7806.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 19:27:01,643 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.208e+02 3.720e+02 4.585e+02 5.523e+02 1.290e+03, threshold=9.170e+02, percent-clipped=3.0 +2023-02-05 19:27:09,239 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7820.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:27:29,703 INFO [train.py:901] (3/4) Epoch 1, batch 7850, loss[loss=0.3691, simple_loss=0.4164, pruned_loss=0.1609, over 8087.00 frames. ], tot_loss[loss=0.409, simple_loss=0.431, pruned_loss=0.1935, over 1611756.27 frames. ], batch size: 21, lr: 3.66e-02, grad_scale: 8.0 +2023-02-05 19:27:44,076 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3173, 2.1037, 3.2772, 3.1386, 2.6664, 1.8484, 1.9183, 2.3612], + device='cuda:3'), covar=tensor([0.1252, 0.0958, 0.0189, 0.0249, 0.0481, 0.0541, 0.0617, 0.0726], + device='cuda:3'), in_proj_covar=tensor([0.0376, 0.0271, 0.0171, 0.0201, 0.0267, 0.0260, 0.0272, 0.0303], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 19:27:52,030 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7884.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:27:59,878 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7896.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:28:03,017 INFO [train.py:901] (3/4) Epoch 1, batch 7900, loss[loss=0.3724, simple_loss=0.4017, pruned_loss=0.1715, over 8472.00 frames. ], tot_loss[loss=0.4063, simple_loss=0.429, pruned_loss=0.1919, over 1614169.27 frames. ], batch size: 25, lr: 3.66e-02, grad_scale: 8.0 +2023-02-05 19:28:08,431 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.445e+02 3.732e+02 4.923e+02 6.190e+02 1.863e+03, threshold=9.845e+02, percent-clipped=5.0 +2023-02-05 19:28:16,376 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7921.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:28:35,804 INFO [train.py:901] (3/4) Epoch 1, batch 7950, loss[loss=0.3425, simple_loss=0.3772, pruned_loss=0.1539, over 7551.00 frames. ], tot_loss[loss=0.4054, simple_loss=0.4286, pruned_loss=0.1911, over 1613412.83 frames. ], batch size: 18, lr: 3.65e-02, grad_scale: 8.0 +2023-02-05 19:28:48,237 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 +2023-02-05 19:28:49,472 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.02 vs. limit=2.0 +2023-02-05 19:28:59,118 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7986.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:29:10,051 INFO [train.py:901] (3/4) Epoch 1, batch 8000, loss[loss=0.5878, simple_loss=0.5425, pruned_loss=0.3165, over 7273.00 frames. ], tot_loss[loss=0.4065, simple_loss=0.4293, pruned_loss=0.1919, over 1613254.04 frames. ], batch size: 71, lr: 3.64e-02, grad_scale: 8.0 +2023-02-05 19:29:15,105 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8008.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:29:15,539 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.650e+02 3.959e+02 4.934e+02 6.403e+02 1.426e+03, threshold=9.868e+02, percent-clipped=4.0 +2023-02-05 19:29:20,382 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4657, 1.3525, 1.9801, 1.6266, 1.3771, 1.7444, 0.7920, 1.4797], + device='cuda:3'), covar=tensor([0.0526, 0.0532, 0.0173, 0.0270, 0.0447, 0.0209, 0.1227, 0.0488], + device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0111, 0.0093, 0.0134, 0.0116, 0.0081, 0.0163, 0.0132], + device='cuda:3'), out_proj_covar=tensor([1.1495e-04, 1.0514e-04, 8.1513e-05, 1.1284e-04, 1.0738e-04, 7.1609e-05, + 1.4203e-04, 1.1730e-04], device='cuda:3') +2023-02-05 19:29:31,438 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8033.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:29:43,013 INFO [train.py:901] (3/4) Epoch 1, batch 8050, loss[loss=0.3641, simple_loss=0.3929, pruned_loss=0.1676, over 7436.00 frames. ], tot_loss[loss=0.4063, simple_loss=0.4287, pruned_loss=0.1919, over 1597148.99 frames. ], batch size: 17, lr: 3.63e-02, grad_scale: 16.0 +2023-02-05 19:30:16,737 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-05 19:30:20,858 INFO [train.py:901] (3/4) Epoch 2, batch 0, loss[loss=0.4977, simple_loss=0.4681, pruned_loss=0.2637, over 7792.00 frames. ], tot_loss[loss=0.4977, simple_loss=0.4681, pruned_loss=0.2637, over 7792.00 frames. ], batch size: 19, lr: 3.56e-02, grad_scale: 8.0 +2023-02-05 19:30:20,858 INFO [train.py:926] (3/4) Computing validation loss +2023-02-05 19:30:32,395 INFO [train.py:935] (3/4) Epoch 2, validation: loss=0.3107, simple_loss=0.3861, pruned_loss=0.1176, over 944034.00 frames. +2023-02-05 19:30:32,396 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6536MB +2023-02-05 19:30:44,061 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8101.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:30:46,614 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-05 19:30:46,680 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8105.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:30:49,933 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.305e+02 3.846e+02 4.676e+02 6.027e+02 1.450e+03, threshold=9.352e+02, percent-clipped=5.0 +2023-02-05 19:31:06,754 INFO [train.py:901] (3/4) Epoch 2, batch 50, loss[loss=0.4538, simple_loss=0.4714, pruned_loss=0.2181, over 8329.00 frames. ], tot_loss[loss=0.4055, simple_loss=0.4291, pruned_loss=0.1909, over 363162.79 frames. ], batch size: 25, lr: 3.55e-02, grad_scale: 8.0 +2023-02-05 19:31:11,120 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8140.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:31:20,781 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-05 19:31:20,999 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0888, 1.3183, 2.2919, 0.7550, 1.6318, 1.4113, 1.1551, 1.5634], + device='cuda:3'), covar=tensor([0.1691, 0.1697, 0.0375, 0.2117, 0.1121, 0.1948, 0.1609, 0.1150], + device='cuda:3'), in_proj_covar=tensor([0.0314, 0.0313, 0.0303, 0.0342, 0.0395, 0.0363, 0.0319, 0.0376], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 19:31:28,340 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8164.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:31:29,178 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8165.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:31:41,594 INFO [train.py:901] (3/4) Epoch 2, batch 100, loss[loss=0.4355, simple_loss=0.4548, pruned_loss=0.208, over 8460.00 frames. ], tot_loss[loss=0.4092, simple_loss=0.4311, pruned_loss=0.1937, over 640367.11 frames. ], batch size: 27, lr: 3.54e-02, grad_scale: 8.0 +2023-02-05 19:31:44,275 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-05 19:31:59,424 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.305e+02 4.246e+02 4.943e+02 6.491e+02 9.375e+02, threshold=9.885e+02, percent-clipped=1.0 +2023-02-05 19:32:06,339 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8220.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:32:07,762 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6156, 1.9599, 3.1216, 0.9820, 2.3549, 1.7417, 1.6091, 1.9689], + device='cuda:3'), covar=tensor([0.1070, 0.1166, 0.0338, 0.1744, 0.0963, 0.1591, 0.0942, 0.1248], + device='cuda:3'), in_proj_covar=tensor([0.0324, 0.0316, 0.0312, 0.0356, 0.0403, 0.0372, 0.0324, 0.0388], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 19:32:15,455 INFO [train.py:901] (3/4) Epoch 2, batch 150, loss[loss=0.4019, simple_loss=0.4296, pruned_loss=0.1871, over 8540.00 frames. ], tot_loss[loss=0.4065, simple_loss=0.4296, pruned_loss=0.1916, over 856394.75 frames. ], batch size: 39, lr: 3.53e-02, grad_scale: 8.0 +2023-02-05 19:32:47,787 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8279.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:32:50,396 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8283.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:32:50,927 INFO [train.py:901] (3/4) Epoch 2, batch 200, loss[loss=0.4125, simple_loss=0.4432, pruned_loss=0.1909, over 8243.00 frames. ], tot_loss[loss=0.404, simple_loss=0.4286, pruned_loss=0.1897, over 1024940.13 frames. ], batch size: 24, lr: 3.52e-02, grad_scale: 8.0 +2023-02-05 19:32:53,836 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 +2023-02-05 19:33:08,598 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.581e+02 3.727e+02 4.975e+02 6.903e+02 1.681e+03, threshold=9.950e+02, percent-clipped=7.0 +2023-02-05 19:33:24,846 INFO [train.py:901] (3/4) Epoch 2, batch 250, loss[loss=0.3923, simple_loss=0.4029, pruned_loss=0.1908, over 7954.00 frames. ], tot_loss[loss=0.4033, simple_loss=0.4282, pruned_loss=0.1892, over 1155873.13 frames. ], batch size: 21, lr: 3.52e-02, grad_scale: 8.0 +2023-02-05 19:33:27,779 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2478, 1.7325, 1.4181, 1.3637, 1.9858, 1.6512, 1.8149, 2.0878], + device='cuda:3'), covar=tensor([0.1212, 0.1728, 0.2441, 0.2086, 0.1046, 0.1778, 0.1343, 0.1017], + device='cuda:3'), in_proj_covar=tensor([0.0259, 0.0281, 0.0299, 0.0272, 0.0260, 0.0252, 0.0257, 0.0246], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2023-02-05 19:33:36,303 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-05 19:33:40,653 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8357.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:33:46,002 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-05 19:33:58,462 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8382.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:33:59,598 INFO [train.py:901] (3/4) Epoch 2, batch 300, loss[loss=0.3645, simple_loss=0.4109, pruned_loss=0.159, over 8450.00 frames. ], tot_loss[loss=0.4017, simple_loss=0.4268, pruned_loss=0.1883, over 1258328.79 frames. ], batch size: 27, lr: 3.51e-02, grad_scale: 8.0 +2023-02-05 19:34:18,661 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.950e+02 4.043e+02 4.737e+02 5.583e+02 9.957e+02, threshold=9.474e+02, percent-clipped=1.0 +2023-02-05 19:34:32,213 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5358, 4.7488, 3.9881, 1.6534, 3.9009, 3.8643, 4.2236, 3.5426], + device='cuda:3'), covar=tensor([0.0793, 0.0314, 0.0732, 0.4047, 0.0437, 0.0661, 0.0793, 0.0394], + device='cuda:3'), in_proj_covar=tensor([0.0291, 0.0186, 0.0236, 0.0308, 0.0195, 0.0147, 0.0213, 0.0136], + device='cuda:3'), out_proj_covar=tensor([2.0865e-04, 1.2919e-04, 1.5256e-04, 1.9551e-04, 1.2491e-04, 1.0519e-04, + 1.4181e-04, 9.6670e-05], device='cuda:3') +2023-02-05 19:34:35,502 INFO [train.py:901] (3/4) Epoch 2, batch 350, loss[loss=0.4392, simple_loss=0.4593, pruned_loss=0.2096, over 8444.00 frames. ], tot_loss[loss=0.4015, simple_loss=0.4264, pruned_loss=0.1883, over 1336092.40 frames. ], batch size: 27, lr: 3.50e-02, grad_scale: 8.0 +2023-02-05 19:34:43,055 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.77 vs. limit=2.0 +2023-02-05 19:35:03,559 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8476.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:35:05,180 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 +2023-02-05 19:35:09,448 INFO [train.py:901] (3/4) Epoch 2, batch 400, loss[loss=0.3832, simple_loss=0.4306, pruned_loss=0.1679, over 8359.00 frames. ], tot_loss[loss=0.4021, simple_loss=0.4269, pruned_loss=0.1886, over 1397605.11 frames. ], batch size: 24, lr: 3.49e-02, grad_scale: 8.0 +2023-02-05 19:35:16,989 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2865, 2.3423, 1.7621, 2.4176, 2.1725, 1.7605, 2.2687, 2.5424], + device='cuda:3'), covar=tensor([0.0997, 0.0526, 0.1000, 0.0660, 0.0958, 0.1112, 0.0986, 0.0719], + device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0226, 0.0347, 0.0297, 0.0340, 0.0308, 0.0351, 0.0295], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:3') +2023-02-05 19:35:20,903 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8501.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:35:27,438 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.659e+02 4.339e+02 4.887e+02 6.099e+02 1.134e+03, threshold=9.773e+02, percent-clipped=6.0 +2023-02-05 19:35:43,486 INFO [train.py:901] (3/4) Epoch 2, batch 450, loss[loss=0.3449, simple_loss=0.3743, pruned_loss=0.1577, over 7799.00 frames. ], tot_loss[loss=0.4024, simple_loss=0.4271, pruned_loss=0.1888, over 1449706.06 frames. ], batch size: 19, lr: 3.49e-02, grad_scale: 8.0 +2023-02-05 19:35:44,343 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8535.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:36:01,172 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1448, 1.8506, 1.2662, 1.9935, 1.6524, 1.1870, 1.2341, 2.1988], + device='cuda:3'), covar=tensor([0.1456, 0.0656, 0.1536, 0.0751, 0.1211, 0.1721, 0.1533, 0.0702], + device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0237, 0.0357, 0.0302, 0.0339, 0.0313, 0.0350, 0.0298], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:3') +2023-02-05 19:36:01,873 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8560.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:36:17,992 INFO [train.py:901] (3/4) Epoch 2, batch 500, loss[loss=0.3989, simple_loss=0.4241, pruned_loss=0.1869, over 8026.00 frames. ], tot_loss[loss=0.4023, simple_loss=0.4276, pruned_loss=0.1885, over 1489056.32 frames. ], batch size: 22, lr: 3.48e-02, grad_scale: 8.0 +2023-02-05 19:36:36,148 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.361e+02 3.910e+02 4.803e+02 5.619e+02 9.699e+02, threshold=9.605e+02, percent-clipped=0.0 +2023-02-05 19:36:39,797 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.80 vs. limit=2.0 +2023-02-05 19:36:46,910 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.0203, 4.2905, 3.5930, 1.5540, 3.4878, 3.5397, 3.7858, 3.1516], + device='cuda:3'), covar=tensor([0.1074, 0.0426, 0.0915, 0.4226, 0.0553, 0.0564, 0.0896, 0.0572], + device='cuda:3'), in_proj_covar=tensor([0.0276, 0.0182, 0.0227, 0.0300, 0.0190, 0.0141, 0.0201, 0.0134], + device='cuda:3'), out_proj_covar=tensor([1.9803e-04, 1.2699e-04, 1.4554e-04, 1.8939e-04, 1.2231e-04, 1.0151e-04, + 1.3450e-04, 9.4873e-05], device='cuda:3') +2023-02-05 19:36:47,558 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8627.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:36:50,876 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.3086, 2.8620, 2.1285, 1.8104, 2.8398, 2.2067, 2.7350, 3.0723], + device='cuda:3'), covar=tensor([0.1226, 0.1696, 0.2199, 0.2035, 0.0937, 0.1856, 0.1196, 0.0896], + device='cuda:3'), in_proj_covar=tensor([0.0263, 0.0284, 0.0299, 0.0278, 0.0261, 0.0261, 0.0259, 0.0246], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2023-02-05 19:36:52,666 INFO [train.py:901] (3/4) Epoch 2, batch 550, loss[loss=0.4291, simple_loss=0.4193, pruned_loss=0.2195, over 7534.00 frames. ], tot_loss[loss=0.4009, simple_loss=0.4266, pruned_loss=0.1876, over 1515111.95 frames. ], batch size: 18, lr: 3.47e-02, grad_scale: 8.0 +2023-02-05 19:37:05,534 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3756, 1.1505, 1.2309, 1.0836, 1.6138, 1.1217, 1.0630, 1.5424], + device='cuda:3'), covar=tensor([0.1460, 0.2447, 0.2913, 0.2514, 0.0986, 0.2575, 0.1567, 0.1164], + device='cuda:3'), in_proj_covar=tensor([0.0260, 0.0279, 0.0291, 0.0273, 0.0257, 0.0256, 0.0256, 0.0247], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2023-02-05 19:37:26,522 INFO [train.py:901] (3/4) Epoch 2, batch 600, loss[loss=0.4362, simple_loss=0.431, pruned_loss=0.2207, over 7961.00 frames. ], tot_loss[loss=0.4016, simple_loss=0.4273, pruned_loss=0.188, over 1538722.47 frames. ], batch size: 21, lr: 3.46e-02, grad_scale: 8.0 +2023-02-05 19:37:43,312 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.752e+02 3.934e+02 5.073e+02 6.758e+02 1.500e+03, threshold=1.015e+03, percent-clipped=5.0 +2023-02-05 19:37:44,744 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-05 19:37:59,726 INFO [train.py:901] (3/4) Epoch 2, batch 650, loss[loss=0.4769, simple_loss=0.4708, pruned_loss=0.2415, over 8128.00 frames. ], tot_loss[loss=0.4001, simple_loss=0.4264, pruned_loss=0.187, over 1560233.24 frames. ], batch size: 22, lr: 3.46e-02, grad_scale: 8.0 +2023-02-05 19:38:05,384 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8742.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:38:31,190 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8778.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:38:35,541 INFO [train.py:901] (3/4) Epoch 2, batch 700, loss[loss=0.4168, simple_loss=0.4439, pruned_loss=0.1949, over 8464.00 frames. ], tot_loss[loss=0.3972, simple_loss=0.4243, pruned_loss=0.1851, over 1571545.87 frames. ], batch size: 27, lr: 3.45e-02, grad_scale: 8.0 +2023-02-05 19:38:53,112 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.421e+02 3.759e+02 4.676e+02 6.060e+02 1.461e+03, threshold=9.352e+02, percent-clipped=1.0 +2023-02-05 19:38:53,988 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.6603, 1.1369, 3.7074, 1.4946, 3.0409, 3.0916, 3.0942, 3.1525], + device='cuda:3'), covar=tensor([0.0328, 0.2907, 0.0249, 0.1329, 0.0769, 0.0336, 0.0350, 0.0426], + device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0350, 0.0195, 0.0231, 0.0265, 0.0215, 0.0192, 0.0228], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2023-02-05 19:38:59,387 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2866, 1.4766, 1.5218, 0.3558, 1.5720, 1.1549, 0.3106, 1.6223], + device='cuda:3'), covar=tensor([0.0177, 0.0143, 0.0254, 0.0389, 0.0150, 0.0466, 0.0517, 0.0137], + device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0129, 0.0108, 0.0173, 0.0120, 0.0200, 0.0180, 0.0145], + device='cuda:3'), out_proj_covar=tensor([1.1934e-04, 9.1277e-05, 8.2637e-05, 1.2607e-04, 9.2538e-05, 1.5435e-04, + 1.3345e-04, 1.0701e-04], device='cuda:3') +2023-02-05 19:39:09,171 INFO [train.py:901] (3/4) Epoch 2, batch 750, loss[loss=0.4067, simple_loss=0.425, pruned_loss=0.1942, over 7808.00 frames. ], tot_loss[loss=0.3986, simple_loss=0.4251, pruned_loss=0.186, over 1581392.38 frames. ], batch size: 20, lr: 3.44e-02, grad_scale: 8.0 +2023-02-05 19:39:26,402 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-05 19:39:35,541 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-05 19:39:44,331 INFO [train.py:901] (3/4) Epoch 2, batch 800, loss[loss=0.3347, simple_loss=0.3706, pruned_loss=0.1494, over 7444.00 frames. ], tot_loss[loss=0.4021, simple_loss=0.4272, pruned_loss=0.1885, over 1590175.84 frames. ], batch size: 17, lr: 3.43e-02, grad_scale: 8.0 +2023-02-05 19:40:02,277 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.406e+02 4.043e+02 5.225e+02 6.708e+02 1.302e+03, threshold=1.045e+03, percent-clipped=9.0 +2023-02-05 19:40:18,487 INFO [train.py:901] (3/4) Epoch 2, batch 850, loss[loss=0.3889, simple_loss=0.4035, pruned_loss=0.1871, over 7930.00 frames. ], tot_loss[loss=0.4014, simple_loss=0.4274, pruned_loss=0.1877, over 1602198.66 frames. ], batch size: 20, lr: 3.43e-02, grad_scale: 8.0 +2023-02-05 19:40:26,055 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8945.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:40:32,045 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7731, 2.3297, 2.4955, 0.3369, 2.3177, 1.5958, 0.8043, 1.3454], + device='cuda:3'), covar=tensor([0.0300, 0.0142, 0.0177, 0.0593, 0.0308, 0.0466, 0.0649, 0.0307], + device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0123, 0.0105, 0.0164, 0.0119, 0.0203, 0.0171, 0.0143], + device='cuda:3'), out_proj_covar=tensor([1.1400e-04, 8.7058e-05, 8.0642e-05, 1.1866e-04, 9.2732e-05, 1.5596e-04, + 1.2641e-04, 1.0579e-04], device='cuda:3') +2023-02-05 19:40:52,644 INFO [train.py:901] (3/4) Epoch 2, batch 900, loss[loss=0.4305, simple_loss=0.4495, pruned_loss=0.2057, over 7806.00 frames. ], tot_loss[loss=0.3996, simple_loss=0.4262, pruned_loss=0.1865, over 1605951.09 frames. ], batch size: 20, lr: 3.42e-02, grad_scale: 8.0 +2023-02-05 19:41:03,942 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8998.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:41:08,133 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9004.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:41:12,012 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.317e+02 3.660e+02 4.402e+02 6.333e+02 1.420e+03, threshold=8.805e+02, percent-clipped=4.0 +2023-02-05 19:41:12,194 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3367, 1.3214, 2.9130, 1.1551, 1.9600, 3.3145, 3.0474, 2.8134], + device='cuda:3'), covar=tensor([0.1703, 0.1984, 0.0402, 0.2214, 0.0838, 0.0233, 0.0247, 0.0454], + device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0263, 0.0174, 0.0254, 0.0188, 0.0137, 0.0134, 0.0196], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-05 19:41:21,858 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9023.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:41:27,244 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9031.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:41:29,113 INFO [train.py:901] (3/4) Epoch 2, batch 950, loss[loss=0.3685, simple_loss=0.3959, pruned_loss=0.1705, over 7817.00 frames. ], tot_loss[loss=0.3992, simple_loss=0.4262, pruned_loss=0.1862, over 1612392.88 frames. ], batch size: 20, lr: 3.41e-02, grad_scale: 8.0 +2023-02-05 19:41:30,788 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.0302, 2.4844, 4.2676, 3.9835, 3.2277, 2.6397, 1.9668, 2.2674], + device='cuda:3'), covar=tensor([0.0871, 0.0994, 0.0144, 0.0241, 0.0460, 0.0386, 0.0580, 0.0965], + device='cuda:3'), in_proj_covar=tensor([0.0412, 0.0319, 0.0221, 0.0253, 0.0327, 0.0294, 0.0319, 0.0365], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 19:41:48,094 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.79 vs. limit=5.0 +2023-02-05 19:41:57,088 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-05 19:42:04,017 INFO [train.py:901] (3/4) Epoch 2, batch 1000, loss[loss=0.3898, simple_loss=0.4111, pruned_loss=0.1843, over 8136.00 frames. ], tot_loss[loss=0.3985, simple_loss=0.426, pruned_loss=0.1855, over 1617163.42 frames. ], batch size: 22, lr: 3.40e-02, grad_scale: 8.0 +2023-02-05 19:42:08,235 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4015, 0.8786, 4.5054, 1.8767, 3.7236, 3.6916, 3.9129, 3.9453], + device='cuda:3'), covar=tensor([0.0367, 0.3976, 0.0221, 0.1576, 0.1039, 0.0338, 0.0296, 0.0364], + device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0353, 0.0196, 0.0237, 0.0278, 0.0214, 0.0195, 0.0227], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2023-02-05 19:42:22,621 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.505e+02 3.676e+02 4.681e+02 5.718e+02 9.745e+02, threshold=9.362e+02, percent-clipped=2.0 +2023-02-05 19:42:30,647 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9122.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:42:31,281 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-05 19:42:39,158 INFO [train.py:901] (3/4) Epoch 2, batch 1050, loss[loss=0.3851, simple_loss=0.4075, pruned_loss=0.1814, over 7203.00 frames. ], tot_loss[loss=0.3981, simple_loss=0.4261, pruned_loss=0.185, over 1619973.11 frames. ], batch size: 16, lr: 3.40e-02, grad_scale: 8.0 +2023-02-05 19:42:43,244 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-05 19:43:12,161 INFO [train.py:901] (3/4) Epoch 2, batch 1100, loss[loss=0.3771, simple_loss=0.4038, pruned_loss=0.1752, over 8289.00 frames. ], tot_loss[loss=0.3951, simple_loss=0.4238, pruned_loss=0.1832, over 1619295.91 frames. ], batch size: 23, lr: 3.39e-02, grad_scale: 8.0 +2023-02-05 19:43:30,063 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.440e+02 3.895e+02 4.986e+02 6.293e+02 1.172e+03, threshold=9.973e+02, percent-clipped=2.0 +2023-02-05 19:43:47,485 INFO [train.py:901] (3/4) Epoch 2, batch 1150, loss[loss=0.3695, simple_loss=0.3793, pruned_loss=0.1799, over 7539.00 frames. ], tot_loss[loss=0.3972, simple_loss=0.4248, pruned_loss=0.1848, over 1619076.10 frames. ], batch size: 18, lr: 3.38e-02, grad_scale: 8.0 +2023-02-05 19:43:49,754 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9237.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:43:50,988 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-05 19:44:18,414 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 +2023-02-05 19:44:22,129 INFO [train.py:901] (3/4) Epoch 2, batch 1200, loss[loss=0.3601, simple_loss=0.3975, pruned_loss=0.1614, over 7804.00 frames. ], tot_loss[loss=0.3978, simple_loss=0.4251, pruned_loss=0.1853, over 1620699.46 frames. ], batch size: 20, lr: 3.38e-02, grad_scale: 8.0 +2023-02-05 19:44:25,542 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9289.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:44:41,024 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.810e+02 4.160e+02 4.885e+02 6.720e+02 4.965e+03, threshold=9.769e+02, percent-clipped=5.0 +2023-02-05 19:44:56,246 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2066, 1.1803, 1.8363, 1.4619, 1.1629, 1.8273, 0.3851, 1.1434], + device='cuda:3'), covar=tensor([0.0881, 0.0652, 0.0313, 0.0434, 0.0602, 0.0311, 0.1618, 0.0771], + device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0103, 0.0094, 0.0130, 0.0114, 0.0085, 0.0158, 0.0126], + device='cuda:3'), out_proj_covar=tensor([1.1616e-04, 1.0030e-04, 8.4678e-05, 1.1451e-04, 1.1153e-04, 7.8152e-05, + 1.3983e-04, 1.1723e-04], device='cuda:3') +2023-02-05 19:44:56,719 INFO [train.py:901] (3/4) Epoch 2, batch 1250, loss[loss=0.3739, simple_loss=0.4112, pruned_loss=0.1683, over 8106.00 frames. ], tot_loss[loss=0.3987, simple_loss=0.4255, pruned_loss=0.1859, over 1618941.58 frames. ], batch size: 23, lr: 3.37e-02, grad_scale: 4.0 +2023-02-05 19:44:59,949 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.81 vs. limit=2.0 +2023-02-05 19:45:07,485 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9348.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:45:25,843 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9375.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:45:31,825 INFO [train.py:901] (3/4) Epoch 2, batch 1300, loss[loss=0.4203, simple_loss=0.4535, pruned_loss=0.1935, over 8321.00 frames. ], tot_loss[loss=0.3979, simple_loss=0.4248, pruned_loss=0.1856, over 1616909.46 frames. ], batch size: 25, lr: 3.36e-02, grad_scale: 4.0 +2023-02-05 19:45:45,741 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9404.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:45:50,304 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.161e+02 4.162e+02 5.656e+02 7.688e+02 2.529e+03, threshold=1.131e+03, percent-clipped=11.0 +2023-02-05 19:46:05,044 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9432.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:46:06,286 INFO [train.py:901] (3/4) Epoch 2, batch 1350, loss[loss=0.4197, simple_loss=0.4391, pruned_loss=0.2002, over 8355.00 frames. ], tot_loss[loss=0.3956, simple_loss=0.4232, pruned_loss=0.184, over 1612943.26 frames. ], batch size: 26, lr: 3.36e-02, grad_scale: 4.0 +2023-02-05 19:46:27,275 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9463.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:46:41,366 INFO [train.py:901] (3/4) Epoch 2, batch 1400, loss[loss=0.4802, simple_loss=0.4769, pruned_loss=0.2418, over 7367.00 frames. ], tot_loss[loss=0.3952, simple_loss=0.4228, pruned_loss=0.1838, over 1610931.37 frames. ], batch size: 71, lr: 3.35e-02, grad_scale: 4.0 +2023-02-05 19:46:45,508 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9490.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:46:47,591 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9493.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:46:59,486 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.192e+02 3.889e+02 4.981e+02 6.326e+02 1.555e+03, threshold=9.962e+02, percent-clipped=1.0 +2023-02-05 19:47:04,257 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9518.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:47:15,070 INFO [train.py:901] (3/4) Epoch 2, batch 1450, loss[loss=0.3825, simple_loss=0.4193, pruned_loss=0.1729, over 8335.00 frames. ], tot_loss[loss=0.3943, simple_loss=0.4221, pruned_loss=0.1832, over 1613719.64 frames. ], batch size: 25, lr: 3.34e-02, grad_scale: 4.0 +2023-02-05 19:47:19,043 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-05 19:47:49,262 INFO [train.py:901] (3/4) Epoch 2, batch 1500, loss[loss=0.4187, simple_loss=0.4373, pruned_loss=0.2, over 7974.00 frames. ], tot_loss[loss=0.3944, simple_loss=0.4224, pruned_loss=0.1832, over 1614673.96 frames. ], batch size: 21, lr: 3.33e-02, grad_scale: 4.0 +2023-02-05 19:48:01,349 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9602.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:48:07,896 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.496e+02 4.006e+02 4.905e+02 6.157e+02 1.300e+03, threshold=9.811e+02, percent-clipped=3.0 +2023-02-05 19:48:23,381 INFO [train.py:901] (3/4) Epoch 2, batch 1550, loss[loss=0.3469, simple_loss=0.3931, pruned_loss=0.1503, over 8358.00 frames. ], tot_loss[loss=0.3936, simple_loss=0.4213, pruned_loss=0.1829, over 1612167.59 frames. ], batch size: 24, lr: 3.33e-02, grad_scale: 4.0 +2023-02-05 19:48:41,694 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9660.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:48:52,343 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9676.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:48:57,573 INFO [train.py:901] (3/4) Epoch 2, batch 1600, loss[loss=0.4571, simple_loss=0.4752, pruned_loss=0.2195, over 8187.00 frames. ], tot_loss[loss=0.3934, simple_loss=0.4211, pruned_loss=0.1828, over 1613421.23 frames. ], batch size: 23, lr: 3.32e-02, grad_scale: 8.0 +2023-02-05 19:48:58,389 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9685.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:49:17,085 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.659e+02 4.192e+02 5.177e+02 6.492e+02 1.266e+03, threshold=1.035e+03, percent-clipped=2.0 +2023-02-05 19:49:22,885 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9719.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:49:33,626 INFO [train.py:901] (3/4) Epoch 2, batch 1650, loss[loss=0.3198, simple_loss=0.3736, pruned_loss=0.133, over 7977.00 frames. ], tot_loss[loss=0.3926, simple_loss=0.4215, pruned_loss=0.1818, over 1615566.32 frames. ], batch size: 21, lr: 3.31e-02, grad_scale: 8.0 +2023-02-05 19:49:40,256 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9744.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:49:41,569 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9746.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:49:51,532 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9761.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:49:58,951 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9771.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:50:02,239 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9776.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:50:07,344 INFO [train.py:901] (3/4) Epoch 2, batch 1700, loss[loss=0.4117, simple_loss=0.4309, pruned_loss=0.1962, over 8466.00 frames. ], tot_loss[loss=0.3929, simple_loss=0.4218, pruned_loss=0.1819, over 1615012.29 frames. ], batch size: 25, lr: 3.31e-02, grad_scale: 8.0 +2023-02-05 19:50:21,970 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2023-02-05 19:50:26,241 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.640e+02 4.068e+02 5.098e+02 6.535e+02 1.207e+03, threshold=1.020e+03, percent-clipped=5.0 +2023-02-05 19:50:42,244 INFO [train.py:901] (3/4) Epoch 2, batch 1750, loss[loss=0.4136, simple_loss=0.4466, pruned_loss=0.1903, over 8641.00 frames. ], tot_loss[loss=0.3925, simple_loss=0.4215, pruned_loss=0.1817, over 1616646.41 frames. ], batch size: 49, lr: 3.30e-02, grad_scale: 8.0 +2023-02-05 19:51:02,102 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6060, 2.0796, 3.7512, 1.1835, 2.3718, 1.9912, 1.7159, 2.0789], + device='cuda:3'), covar=tensor([0.0899, 0.1155, 0.0261, 0.1474, 0.0988, 0.1449, 0.0871, 0.1336], + device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0335, 0.0348, 0.0379, 0.0431, 0.0400, 0.0346, 0.0423], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 19:51:16,319 INFO [train.py:901] (3/4) Epoch 2, batch 1800, loss[loss=0.5017, simple_loss=0.4789, pruned_loss=0.2622, over 6645.00 frames. ], tot_loss[loss=0.3928, simple_loss=0.4214, pruned_loss=0.1821, over 1613178.43 frames. ], batch size: 71, lr: 3.29e-02, grad_scale: 8.0 +2023-02-05 19:51:21,268 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9891.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:51:34,083 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.365e+02 4.111e+02 5.198e+02 6.626e+02 1.120e+03, threshold=1.040e+03, percent-clipped=3.0 +2023-02-05 19:51:49,954 INFO [train.py:901] (3/4) Epoch 2, batch 1850, loss[loss=0.3695, simple_loss=0.4198, pruned_loss=0.1596, over 8686.00 frames. ], tot_loss[loss=0.3929, simple_loss=0.4214, pruned_loss=0.1823, over 1616917.30 frames. ], batch size: 34, lr: 3.29e-02, grad_scale: 8.0 +2023-02-05 19:51:58,652 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9946.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:52:00,331 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.80 vs. limit=2.0 +2023-02-05 19:52:24,350 INFO [train.py:901] (3/4) Epoch 2, batch 1900, loss[loss=0.3827, simple_loss=0.4159, pruned_loss=0.1747, over 8493.00 frames. ], tot_loss[loss=0.3903, simple_loss=0.4197, pruned_loss=0.1804, over 1616940.34 frames. ], batch size: 26, lr: 3.28e-02, grad_scale: 8.0 +2023-02-05 19:52:43,802 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.445e+02 3.513e+02 4.327e+02 5.785e+02 1.080e+03, threshold=8.653e+02, percent-clipped=1.0 +2023-02-05 19:52:49,916 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10020.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:52:54,612 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-05 19:52:59,200 INFO [train.py:901] (3/4) Epoch 2, batch 1950, loss[loss=0.3733, simple_loss=0.4183, pruned_loss=0.1641, over 8579.00 frames. ], tot_loss[loss=0.3897, simple_loss=0.4187, pruned_loss=0.1804, over 1613664.74 frames. ], batch size: 49, lr: 3.27e-02, grad_scale: 8.0 +2023-02-05 19:53:06,849 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-05 19:53:15,906 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10057.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:53:19,396 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10061.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:53:25,610 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-05 19:53:33,046 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10080.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 19:53:33,648 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3109, 2.3711, 1.8895, 3.0941, 1.7425, 1.4844, 1.6103, 2.3719], + device='cuda:3'), covar=tensor([0.0965, 0.1382, 0.1542, 0.0354, 0.1711, 0.2281, 0.2312, 0.1127], + device='cuda:3'), in_proj_covar=tensor([0.0297, 0.0322, 0.0311, 0.0214, 0.0324, 0.0343, 0.0371, 0.0299], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2023-02-05 19:53:35,464 INFO [train.py:901] (3/4) Epoch 2, batch 2000, loss[loss=0.4012, simple_loss=0.4072, pruned_loss=0.1976, over 7691.00 frames. ], tot_loss[loss=0.391, simple_loss=0.4196, pruned_loss=0.1812, over 1613187.33 frames. ], batch size: 18, lr: 3.27e-02, grad_scale: 8.0 +2023-02-05 19:53:42,055 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3059, 1.6110, 1.2316, 1.1550, 1.9531, 1.4918, 1.7090, 1.8804], + device='cuda:3'), covar=tensor([0.1181, 0.1887, 0.2637, 0.2192, 0.0970, 0.1951, 0.1294, 0.1033], + device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0265, 0.0284, 0.0260, 0.0239, 0.0247, 0.0230, 0.0228], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2023-02-05 19:53:46,355 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.0821, 2.6040, 4.2418, 4.1260, 3.1626, 2.3503, 1.7456, 2.1887], + device='cuda:3'), covar=tensor([0.0850, 0.1041, 0.0140, 0.0229, 0.0449, 0.0483, 0.0647, 0.0901], + device='cuda:3'), in_proj_covar=tensor([0.0446, 0.0355, 0.0255, 0.0293, 0.0375, 0.0331, 0.0351, 0.0389], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 19:53:50,417 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10105.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:53:55,727 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.648e+02 4.167e+02 5.413e+02 6.926e+02 6.671e+03, threshold=1.083e+03, percent-clipped=14.0 +2023-02-05 19:54:10,561 INFO [train.py:901] (3/4) Epoch 2, batch 2050, loss[loss=0.3667, simple_loss=0.4154, pruned_loss=0.159, over 8252.00 frames. ], tot_loss[loss=0.3907, simple_loss=0.4198, pruned_loss=0.1808, over 1617414.34 frames. ], batch size: 24, lr: 3.26e-02, grad_scale: 4.0 +2023-02-05 19:54:11,456 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10135.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:54:19,458 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10147.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:54:36,885 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10172.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:54:45,487 INFO [train.py:901] (3/4) Epoch 2, batch 2100, loss[loss=0.3357, simple_loss=0.3746, pruned_loss=0.1484, over 7933.00 frames. ], tot_loss[loss=0.3889, simple_loss=0.4188, pruned_loss=0.1794, over 1620796.65 frames. ], batch size: 20, lr: 3.25e-02, grad_scale: 4.0 +2023-02-05 19:54:55,285 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.30 vs. limit=5.0 +2023-02-05 19:54:59,735 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 +2023-02-05 19:55:06,153 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.637e+02 3.788e+02 4.646e+02 5.840e+02 1.328e+03, threshold=9.292e+02, percent-clipped=3.0 +2023-02-05 19:55:11,276 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10220.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:55:20,261 INFO [train.py:901] (3/4) Epoch 2, batch 2150, loss[loss=0.4237, simple_loss=0.4441, pruned_loss=0.2016, over 8457.00 frames. ], tot_loss[loss=0.3859, simple_loss=0.4167, pruned_loss=0.1776, over 1621620.16 frames. ], batch size: 29, lr: 3.25e-02, grad_scale: 4.0 +2023-02-05 19:55:53,989 INFO [train.py:901] (3/4) Epoch 2, batch 2200, loss[loss=0.3354, simple_loss=0.381, pruned_loss=0.1448, over 8076.00 frames. ], tot_loss[loss=0.385, simple_loss=0.4159, pruned_loss=0.1771, over 1614871.51 frames. ], batch size: 21, lr: 3.24e-02, grad_scale: 4.0 +2023-02-05 19:56:10,342 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 +2023-02-05 19:56:14,592 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.956e+02 3.803e+02 4.971e+02 6.310e+02 1.458e+03, threshold=9.942e+02, percent-clipped=6.0 +2023-02-05 19:56:18,159 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10317.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:56:29,273 INFO [train.py:901] (3/4) Epoch 2, batch 2250, loss[loss=0.4493, simple_loss=0.4646, pruned_loss=0.217, over 8043.00 frames. ], tot_loss[loss=0.3854, simple_loss=0.4164, pruned_loss=0.1772, over 1616101.18 frames. ], batch size: 22, lr: 3.24e-02, grad_scale: 4.0 +2023-02-05 19:56:34,612 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10342.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:57:03,213 INFO [train.py:901] (3/4) Epoch 2, batch 2300, loss[loss=0.3771, simple_loss=0.4042, pruned_loss=0.175, over 7779.00 frames. ], tot_loss[loss=0.3861, simple_loss=0.4168, pruned_loss=0.1777, over 1618346.77 frames. ], batch size: 19, lr: 3.23e-02, grad_scale: 4.0 +2023-02-05 19:57:08,299 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10391.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:57:15,022 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10401.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:57:23,818 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.389e+02 3.989e+02 5.161e+02 7.086e+02 1.471e+03, threshold=1.032e+03, percent-clipped=7.0 +2023-02-05 19:57:25,927 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10416.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:57:31,804 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10424.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 19:57:33,869 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10427.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:57:39,139 INFO [train.py:901] (3/4) Epoch 2, batch 2350, loss[loss=0.3818, simple_loss=0.4131, pruned_loss=0.1752, over 7973.00 frames. ], tot_loss[loss=0.3849, simple_loss=0.4161, pruned_loss=0.1769, over 1618113.98 frames. ], batch size: 21, lr: 3.22e-02, grad_scale: 4.0 +2023-02-05 19:58:05,154 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10472.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:58:07,804 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10476.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:58:12,898 INFO [train.py:901] (3/4) Epoch 2, batch 2400, loss[loss=0.4666, simple_loss=0.475, pruned_loss=0.2291, over 8199.00 frames. ], tot_loss[loss=0.3847, simple_loss=0.4162, pruned_loss=0.1766, over 1620491.40 frames. ], batch size: 23, lr: 3.22e-02, grad_scale: 8.0 +2023-02-05 19:58:24,668 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10501.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:58:32,502 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.364e+02 3.956e+02 5.047e+02 6.263e+02 1.564e+03, threshold=1.009e+03, percent-clipped=2.0 +2023-02-05 19:58:34,745 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10516.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 19:58:47,328 INFO [train.py:901] (3/4) Epoch 2, batch 2450, loss[loss=0.4361, simple_loss=0.4534, pruned_loss=0.2094, over 8249.00 frames. ], tot_loss[loss=0.3843, simple_loss=0.4157, pruned_loss=0.1764, over 1617731.49 frames. ], batch size: 24, lr: 3.21e-02, grad_scale: 8.0 +2023-02-05 19:58:50,875 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10539.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 19:59:07,389 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 +2023-02-05 19:59:21,101 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6711, 2.1304, 2.8326, 1.8435, 1.6755, 2.3837, 0.4785, 1.4969], + device='cuda:3'), covar=tensor([0.0875, 0.0515, 0.0262, 0.0591, 0.0951, 0.0378, 0.1735, 0.0773], + device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0099, 0.0088, 0.0130, 0.0118, 0.0085, 0.0158, 0.0125], + device='cuda:3'), out_proj_covar=tensor([1.0911e-04, 1.0067e-04, 8.2060e-05, 1.1914e-04, 1.1672e-04, 7.9901e-05, + 1.4519e-04, 1.1964e-04], device='cuda:3') +2023-02-05 19:59:22,254 INFO [train.py:901] (3/4) Epoch 2, batch 2500, loss[loss=0.3883, simple_loss=0.4236, pruned_loss=0.1765, over 8473.00 frames. ], tot_loss[loss=0.384, simple_loss=0.4158, pruned_loss=0.1761, over 1619817.75 frames. ], batch size: 29, lr: 3.20e-02, grad_scale: 8.0 +2023-02-05 19:59:23,776 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2422, 2.0058, 1.5349, 1.3117, 1.9781, 1.6594, 1.6736, 1.8671], + device='cuda:3'), covar=tensor([0.1157, 0.1592, 0.2372, 0.1903, 0.0917, 0.1765, 0.1297, 0.0952], + device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0262, 0.0288, 0.0253, 0.0234, 0.0249, 0.0232, 0.0224], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0004], + device='cuda:3') +2023-02-05 19:59:42,163 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.520e+02 3.522e+02 4.438e+02 6.473e+02 1.354e+03, threshold=8.876e+02, percent-clipped=4.0 +2023-02-05 19:59:55,946 INFO [train.py:901] (3/4) Epoch 2, batch 2550, loss[loss=0.3366, simple_loss=0.3973, pruned_loss=0.138, over 8107.00 frames. ], tot_loss[loss=0.3841, simple_loss=0.416, pruned_loss=0.1761, over 1618549.25 frames. ], batch size: 23, lr: 3.20e-02, grad_scale: 8.0 +2023-02-05 20:00:31,346 INFO [train.py:901] (3/4) Epoch 2, batch 2600, loss[loss=0.4578, simple_loss=0.4612, pruned_loss=0.2271, over 7971.00 frames. ], tot_loss[loss=0.3816, simple_loss=0.4137, pruned_loss=0.1748, over 1609586.66 frames. ], batch size: 21, lr: 3.19e-02, grad_scale: 8.0 +2023-02-05 20:00:50,499 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.921e+02 4.188e+02 4.914e+02 6.333e+02 1.432e+03, threshold=9.828e+02, percent-clipped=6.0 +2023-02-05 20:01:05,101 INFO [train.py:901] (3/4) Epoch 2, batch 2650, loss[loss=0.3961, simple_loss=0.4287, pruned_loss=0.1818, over 8097.00 frames. ], tot_loss[loss=0.3832, simple_loss=0.4147, pruned_loss=0.1759, over 1612883.93 frames. ], batch size: 23, lr: 3.19e-02, grad_scale: 8.0 +2023-02-05 20:01:05,522 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-05 20:01:24,206 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10762.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:01:30,880 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10771.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:01:31,706 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10772.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:01:40,305 INFO [train.py:901] (3/4) Epoch 2, batch 2700, loss[loss=0.3817, simple_loss=0.4189, pruned_loss=0.1722, over 8357.00 frames. ], tot_loss[loss=0.3838, simple_loss=0.4151, pruned_loss=0.1763, over 1616298.83 frames. ], batch size: 24, lr: 3.18e-02, grad_scale: 8.0 +2023-02-05 20:01:46,658 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10792.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:01:48,789 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10795.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 20:01:50,097 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10797.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:02:01,035 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.290e+02 4.005e+02 5.458e+02 7.000e+02 2.619e+03, threshold=1.092e+03, percent-clipped=7.0 +2023-02-05 20:02:03,280 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10816.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:02:06,067 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10820.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 20:02:15,181 INFO [train.py:901] (3/4) Epoch 2, batch 2750, loss[loss=0.4953, simple_loss=0.5001, pruned_loss=0.2452, over 8561.00 frames. ], tot_loss[loss=0.3798, simple_loss=0.4121, pruned_loss=0.1737, over 1613885.51 frames. ], batch size: 31, lr: 3.17e-02, grad_scale: 8.0 +2023-02-05 20:02:49,766 INFO [train.py:901] (3/4) Epoch 2, batch 2800, loss[loss=0.3113, simple_loss=0.3576, pruned_loss=0.1325, over 8461.00 frames. ], tot_loss[loss=0.3791, simple_loss=0.412, pruned_loss=0.1731, over 1616421.11 frames. ], batch size: 25, lr: 3.17e-02, grad_scale: 8.0 +2023-02-05 20:02:51,257 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10886.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:03:03,320 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10903.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 20:03:10,621 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.367e+02 3.535e+02 4.531e+02 6.001e+02 1.335e+03, threshold=9.062e+02, percent-clipped=2.0 +2023-02-05 20:03:23,148 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10931.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:03:25,028 INFO [train.py:901] (3/4) Epoch 2, batch 2850, loss[loss=0.4321, simple_loss=0.4556, pruned_loss=0.2043, over 8518.00 frames. ], tot_loss[loss=0.3764, simple_loss=0.4102, pruned_loss=0.1713, over 1620787.53 frames. ], batch size: 26, lr: 3.16e-02, grad_scale: 8.0 +2023-02-05 20:03:59,105 INFO [train.py:901] (3/4) Epoch 2, batch 2900, loss[loss=0.4197, simple_loss=0.4381, pruned_loss=0.2006, over 8129.00 frames. ], tot_loss[loss=0.3797, simple_loss=0.4128, pruned_loss=0.1733, over 1619318.28 frames. ], batch size: 22, lr: 3.16e-02, grad_scale: 8.0 +2023-02-05 20:04:19,446 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.551e+02 4.216e+02 5.196e+02 6.845e+02 2.226e+03, threshold=1.039e+03, percent-clipped=10.0 +2023-02-05 20:04:34,448 INFO [train.py:901] (3/4) Epoch 2, batch 2950, loss[loss=0.4963, simple_loss=0.4841, pruned_loss=0.2542, over 7959.00 frames. ], tot_loss[loss=0.3819, simple_loss=0.4142, pruned_loss=0.1748, over 1616590.36 frames. ], batch size: 21, lr: 3.15e-02, grad_scale: 8.0 +2023-02-05 20:04:39,269 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-05 20:05:08,646 INFO [train.py:901] (3/4) Epoch 2, batch 3000, loss[loss=0.3439, simple_loss=0.3711, pruned_loss=0.1583, over 7694.00 frames. ], tot_loss[loss=0.3798, simple_loss=0.4129, pruned_loss=0.1733, over 1614618.21 frames. ], batch size: 18, lr: 3.14e-02, grad_scale: 8.0 +2023-02-05 20:05:08,647 INFO [train.py:926] (3/4) Computing validation loss +2023-02-05 20:05:24,852 INFO [train.py:935] (3/4) Epoch 2, validation: loss=0.2878, simple_loss=0.369, pruned_loss=0.1033, over 944034.00 frames. +2023-02-05 20:05:24,853 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-05 20:05:40,501 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11106.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:05:45,160 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.542e+02 3.795e+02 4.955e+02 6.193e+02 1.384e+03, threshold=9.910e+02, percent-clipped=4.0 +2023-02-05 20:06:00,078 INFO [train.py:901] (3/4) Epoch 2, batch 3050, loss[loss=0.3526, simple_loss=0.4012, pruned_loss=0.152, over 8195.00 frames. ], tot_loss[loss=0.3814, simple_loss=0.4142, pruned_loss=0.1744, over 1615709.21 frames. ], batch size: 23, lr: 3.14e-02, grad_scale: 8.0 +2023-02-05 20:06:01,579 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11136.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:06:05,905 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11142.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:06:08,869 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.44 vs. limit=5.0 +2023-02-05 20:06:24,332 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11167.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:06:35,398 INFO [train.py:901] (3/4) Epoch 2, batch 3100, loss[loss=0.3583, simple_loss=0.4118, pruned_loss=0.1524, over 8253.00 frames. ], tot_loss[loss=0.3815, simple_loss=0.4143, pruned_loss=0.1743, over 1613124.85 frames. ], batch size: 24, lr: 3.13e-02, grad_scale: 8.0 +2023-02-05 20:06:37,615 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11187.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:06:40,870 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11192.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:06:55,400 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11212.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:06:55,866 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.637e+02 3.930e+02 4.987e+02 6.652e+02 1.229e+03, threshold=9.974e+02, percent-clipped=5.0 +2023-02-05 20:07:01,729 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11221.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:07:10,341 INFO [train.py:901] (3/4) Epoch 2, batch 3150, loss[loss=0.4388, simple_loss=0.4607, pruned_loss=0.2084, over 8470.00 frames. ], tot_loss[loss=0.38, simple_loss=0.4128, pruned_loss=0.1736, over 1612676.79 frames. ], batch size: 25, lr: 3.13e-02, grad_scale: 8.0 +2023-02-05 20:07:20,137 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11247.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 20:07:22,962 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11251.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:07:35,609 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-05 20:07:44,249 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.9556, 4.1919, 3.5387, 1.6815, 3.4598, 3.2433, 3.7016, 2.7286], + device='cuda:3'), covar=tensor([0.1006, 0.0770, 0.1094, 0.4062, 0.0575, 0.0740, 0.1406, 0.0784], + device='cuda:3'), in_proj_covar=tensor([0.0301, 0.0206, 0.0236, 0.0327, 0.0217, 0.0168, 0.0227, 0.0160], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2023-02-05 20:07:46,080 INFO [train.py:901] (3/4) Epoch 2, batch 3200, loss[loss=0.3393, simple_loss=0.3663, pruned_loss=0.1562, over 7535.00 frames. ], tot_loss[loss=0.3805, simple_loss=0.4135, pruned_loss=0.1737, over 1614012.26 frames. ], batch size: 18, lr: 3.12e-02, grad_scale: 8.0 +2023-02-05 20:08:06,205 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.758e+02 3.889e+02 4.508e+02 6.050e+02 1.565e+03, threshold=9.016e+02, percent-clipped=4.0 +2023-02-05 20:08:13,956 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3301, 1.9885, 3.3216, 2.8973, 2.5691, 2.0067, 1.5899, 1.7653], + device='cuda:3'), covar=tensor([0.0795, 0.0784, 0.0140, 0.0215, 0.0338, 0.0345, 0.0475, 0.0684], + device='cuda:3'), in_proj_covar=tensor([0.0468, 0.0385, 0.0271, 0.0307, 0.0403, 0.0346, 0.0361, 0.0407], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 20:08:20,048 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3454, 1.5363, 2.0689, 1.2469, 1.1597, 1.9427, 0.3821, 1.0407], + device='cuda:3'), covar=tensor([0.0940, 0.0737, 0.0468, 0.0660, 0.0861, 0.0324, 0.2173, 0.1198], + device='cuda:3'), in_proj_covar=tensor([0.0118, 0.0098, 0.0087, 0.0136, 0.0123, 0.0082, 0.0163, 0.0131], + device='cuda:3'), out_proj_covar=tensor([1.1292e-04, 1.0073e-04, 8.3813e-05, 1.2701e-04, 1.2407e-04, 7.8640e-05, + 1.5335e-04, 1.2907e-04], device='cuda:3') +2023-02-05 20:08:21,234 INFO [train.py:901] (3/4) Epoch 2, batch 3250, loss[loss=0.2958, simple_loss=0.3406, pruned_loss=0.1254, over 7813.00 frames. ], tot_loss[loss=0.3776, simple_loss=0.411, pruned_loss=0.1721, over 1615047.49 frames. ], batch size: 19, lr: 3.11e-02, grad_scale: 8.0 +2023-02-05 20:08:39,999 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11362.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 20:08:45,874 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0270, 1.1552, 1.1775, 0.0915, 1.0645, 0.8291, 0.1756, 1.0491], + device='cuda:3'), covar=tensor([0.0092, 0.0078, 0.0095, 0.0197, 0.0127, 0.0262, 0.0245, 0.0094], + device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0128, 0.0125, 0.0182, 0.0132, 0.0226, 0.0194, 0.0164], + device='cuda:3'), out_proj_covar=tensor([1.1901e-04, 8.3338e-05, 8.6944e-05, 1.1965e-04, 9.2783e-05, 1.5836e-04, + 1.3218e-04, 1.0966e-04], device='cuda:3') +2023-02-05 20:08:55,031 INFO [train.py:901] (3/4) Epoch 2, batch 3300, loss[loss=0.3703, simple_loss=0.4084, pruned_loss=0.1661, over 7922.00 frames. ], tot_loss[loss=0.3773, simple_loss=0.4109, pruned_loss=0.1719, over 1614631.85 frames. ], batch size: 20, lr: 3.11e-02, grad_scale: 8.0 +2023-02-05 20:09:16,012 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.686e+02 3.650e+02 4.417e+02 5.589e+02 1.513e+03, threshold=8.834e+02, percent-clipped=8.0 +2023-02-05 20:09:20,907 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.5922, 1.2202, 3.7383, 1.3881, 3.1394, 3.0866, 3.2831, 3.3495], + device='cuda:3'), covar=tensor([0.0322, 0.2872, 0.0292, 0.1568, 0.0913, 0.0398, 0.0325, 0.0386], + device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0364, 0.0229, 0.0259, 0.0309, 0.0243, 0.0224, 0.0253], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2023-02-05 20:09:30,177 INFO [train.py:901] (3/4) Epoch 2, batch 3350, loss[loss=0.3063, simple_loss=0.3485, pruned_loss=0.1321, over 7549.00 frames. ], tot_loss[loss=0.3751, simple_loss=0.4093, pruned_loss=0.1705, over 1612808.06 frames. ], batch size: 18, lr: 3.10e-02, grad_scale: 8.0 +2023-02-05 20:10:00,602 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11477.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:10:05,252 INFO [train.py:901] (3/4) Epoch 2, batch 3400, loss[loss=0.2796, simple_loss=0.3301, pruned_loss=0.1146, over 7796.00 frames. ], tot_loss[loss=0.3745, simple_loss=0.4092, pruned_loss=0.1699, over 1616037.93 frames. ], batch size: 19, lr: 3.10e-02, grad_scale: 8.0 +2023-02-05 20:10:17,731 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11502.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:10:21,207 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11507.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:10:25,777 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.120e+02 3.730e+02 4.591e+02 5.662e+02 1.223e+03, threshold=9.181e+02, percent-clipped=5.0 +2023-02-05 20:10:39,499 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11532.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:10:40,655 INFO [train.py:901] (3/4) Epoch 2, batch 3450, loss[loss=0.3682, simple_loss=0.4151, pruned_loss=0.1606, over 8542.00 frames. ], tot_loss[loss=0.375, simple_loss=0.4092, pruned_loss=0.1704, over 1613468.00 frames. ], batch size: 39, lr: 3.09e-02, grad_scale: 8.0 +2023-02-05 20:10:42,065 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11536.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:10:52,326 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11550.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:11:15,646 INFO [train.py:901] (3/4) Epoch 2, batch 3500, loss[loss=0.3654, simple_loss=0.4127, pruned_loss=0.1591, over 8521.00 frames. ], tot_loss[loss=0.3744, simple_loss=0.4087, pruned_loss=0.1701, over 1614982.93 frames. ], batch size: 28, lr: 3.09e-02, grad_scale: 8.0 +2023-02-05 20:11:19,217 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2695, 2.3179, 4.0681, 4.0342, 3.0124, 2.4335, 1.7402, 2.3892], + device='cuda:3'), covar=tensor([0.0631, 0.0949, 0.0133, 0.0194, 0.0394, 0.0369, 0.0528, 0.0725], + device='cuda:3'), in_proj_covar=tensor([0.0471, 0.0387, 0.0282, 0.0311, 0.0413, 0.0355, 0.0373, 0.0408], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 20:11:22,039 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.58 vs. limit=5.0 +2023-02-05 20:11:26,051 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2627, 1.4577, 1.9607, 1.3143, 1.0279, 1.7978, 0.4063, 1.0995], + device='cuda:3'), covar=tensor([0.1066, 0.0908, 0.0451, 0.0720, 0.0914, 0.0411, 0.2369, 0.0976], + device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0101, 0.0086, 0.0135, 0.0120, 0.0078, 0.0156, 0.0123], + device='cuda:3'), out_proj_covar=tensor([1.1200e-04, 1.0293e-04, 8.2897e-05, 1.2719e-04, 1.2137e-04, 7.4997e-05, + 1.4880e-04, 1.2349e-04], device='cuda:3') +2023-02-05 20:11:35,937 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.906e+02 4.071e+02 4.877e+02 6.297e+02 1.257e+03, threshold=9.753e+02, percent-clipped=3.0 +2023-02-05 20:11:39,545 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11618.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 20:11:40,718 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-05 20:11:50,751 INFO [train.py:901] (3/4) Epoch 2, batch 3550, loss[loss=0.2789, simple_loss=0.3275, pruned_loss=0.1152, over 7804.00 frames. ], tot_loss[loss=0.3748, simple_loss=0.4094, pruned_loss=0.1701, over 1620525.07 frames. ], batch size: 19, lr: 3.08e-02, grad_scale: 8.0 +2023-02-05 20:11:57,569 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11643.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 20:12:02,965 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11651.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:12:04,330 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11653.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:12:12,572 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-05 20:12:25,621 INFO [train.py:901] (3/4) Epoch 2, batch 3600, loss[loss=0.385, simple_loss=0.4207, pruned_loss=0.1746, over 8312.00 frames. ], tot_loss[loss=0.3762, simple_loss=0.4103, pruned_loss=0.1711, over 1619843.52 frames. ], batch size: 25, lr: 3.08e-02, grad_scale: 8.0 +2023-02-05 20:12:41,184 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-05 20:12:45,382 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.676e+02 3.688e+02 4.691e+02 6.662e+02 1.491e+03, threshold=9.383e+02, percent-clipped=3.0 +2023-02-05 20:12:48,292 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11717.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:12:59,429 INFO [train.py:901] (3/4) Epoch 2, batch 3650, loss[loss=0.3976, simple_loss=0.4365, pruned_loss=0.1793, over 8509.00 frames. ], tot_loss[loss=0.3779, simple_loss=0.4114, pruned_loss=0.1722, over 1619715.70 frames. ], batch size: 26, lr: 3.07e-02, grad_scale: 8.0 +2023-02-05 20:13:33,833 INFO [train.py:901] (3/4) Epoch 2, batch 3700, loss[loss=0.3926, simple_loss=0.4117, pruned_loss=0.1868, over 8129.00 frames. ], tot_loss[loss=0.3797, simple_loss=0.4121, pruned_loss=0.1736, over 1614256.66 frames. ], batch size: 22, lr: 3.06e-02, grad_scale: 8.0 +2023-02-05 20:13:44,421 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-05 20:13:53,762 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.910e+02 4.224e+02 5.211e+02 6.213e+02 2.304e+03, threshold=1.042e+03, percent-clipped=10.0 +2023-02-05 20:14:08,520 INFO [train.py:901] (3/4) Epoch 2, batch 3750, loss[loss=0.4393, simple_loss=0.4531, pruned_loss=0.2127, over 8186.00 frames. ], tot_loss[loss=0.3783, simple_loss=0.4112, pruned_loss=0.1727, over 1615865.05 frames. ], batch size: 23, lr: 3.06e-02, grad_scale: 8.0 +2023-02-05 20:14:08,633 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11834.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:14:28,588 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11864.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:14:43,031 INFO [train.py:901] (3/4) Epoch 2, batch 3800, loss[loss=0.3687, simple_loss=0.4035, pruned_loss=0.167, over 7973.00 frames. ], tot_loss[loss=0.3756, simple_loss=0.4092, pruned_loss=0.171, over 1611263.09 frames. ], batch size: 21, lr: 3.05e-02, grad_scale: 8.0 +2023-02-05 20:14:49,682 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11894.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:14:58,774 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11907.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:15:02,629 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.750e+02 4.056e+02 4.773e+02 6.198e+02 1.391e+03, threshold=9.546e+02, percent-clipped=3.0 +2023-02-05 20:15:16,344 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11932.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:15:17,493 INFO [train.py:901] (3/4) Epoch 2, batch 3850, loss[loss=0.3509, simple_loss=0.395, pruned_loss=0.1534, over 8485.00 frames. ], tot_loss[loss=0.3753, simple_loss=0.4091, pruned_loss=0.1707, over 1605503.81 frames. ], batch size: 49, lr: 3.05e-02, grad_scale: 8.0 +2023-02-05 20:15:20,309 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11938.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:15:39,314 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2023-02-05 20:15:39,760 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11966.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:15:47,058 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-05 20:15:51,653 INFO [train.py:901] (3/4) Epoch 2, batch 3900, loss[loss=0.3053, simple_loss=0.3578, pruned_loss=0.1264, over 7681.00 frames. ], tot_loss[loss=0.3752, simple_loss=0.4097, pruned_loss=0.1704, over 1606911.93 frames. ], batch size: 18, lr: 3.04e-02, grad_scale: 8.0 +2023-02-05 20:16:01,111 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11997.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:16:06,094 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12002.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:16:10,868 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12009.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:16:13,174 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.113e+02 3.926e+02 4.686e+02 5.678e+02 1.222e+03, threshold=9.373e+02, percent-clipped=4.0 +2023-02-05 20:16:25,605 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5517, 1.6810, 1.4645, 1.9400, 1.6447, 1.1994, 1.2777, 1.8276], + device='cuda:3'), covar=tensor([0.1034, 0.0617, 0.1051, 0.0652, 0.0898, 0.1349, 0.1067, 0.0652], + device='cuda:3'), in_proj_covar=tensor([0.0369, 0.0245, 0.0354, 0.0314, 0.0369, 0.0318, 0.0365, 0.0331], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], + device='cuda:3') +2023-02-05 20:16:26,340 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2764, 1.8600, 3.1400, 2.8219, 2.5075, 1.8215, 1.5588, 1.7555], + device='cuda:3'), covar=tensor([0.0727, 0.0813, 0.0156, 0.0233, 0.0329, 0.0375, 0.0445, 0.0692], + device='cuda:3'), in_proj_covar=tensor([0.0472, 0.0393, 0.0292, 0.0327, 0.0415, 0.0363, 0.0372, 0.0410], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 20:16:27,665 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2466, 1.4541, 1.9224, 1.5565, 0.9738, 1.6423, 0.4757, 0.8717], + device='cuda:3'), covar=tensor([0.0942, 0.0528, 0.0459, 0.0608, 0.0962, 0.0480, 0.1585, 0.0789], + device='cuda:3'), in_proj_covar=tensor([0.0112, 0.0098, 0.0084, 0.0136, 0.0116, 0.0079, 0.0149, 0.0114], + device='cuda:3'), out_proj_covar=tensor([1.1059e-04, 1.0228e-04, 8.2773e-05, 1.2953e-04, 1.1902e-04, 7.7516e-05, + 1.4388e-04, 1.1708e-04], device='cuda:3') +2023-02-05 20:16:28,135 INFO [train.py:901] (3/4) Epoch 2, batch 3950, loss[loss=0.3575, simple_loss=0.4074, pruned_loss=0.1537, over 8482.00 frames. ], tot_loss[loss=0.3747, simple_loss=0.4092, pruned_loss=0.1701, over 1609848.82 frames. ], batch size: 27, lr: 3.04e-02, grad_scale: 8.0 +2023-02-05 20:16:46,989 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12061.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:16:54,547 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5967, 1.9424, 1.6343, 1.3117, 2.3802, 1.8458, 1.9540, 2.1427], + device='cuda:3'), covar=tensor([0.0837, 0.1474, 0.1917, 0.1753, 0.0708, 0.1549, 0.1075, 0.0690], + device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0261, 0.0285, 0.0256, 0.0229, 0.0249, 0.0224, 0.0214], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004], + device='cuda:3') +2023-02-05 20:17:02,480 INFO [train.py:901] (3/4) Epoch 2, batch 4000, loss[loss=0.3706, simple_loss=0.3945, pruned_loss=0.1733, over 7661.00 frames. ], tot_loss[loss=0.3747, simple_loss=0.4092, pruned_loss=0.1701, over 1609718.19 frames. ], batch size: 19, lr: 3.03e-02, grad_scale: 8.0 +2023-02-05 20:17:09,218 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12094.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:17:22,648 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12112.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:17:23,112 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.955e+02 4.453e+02 5.904e+02 7.845e+02 2.502e+03, threshold=1.181e+03, percent-clipped=13.0 +2023-02-05 20:17:36,865 INFO [train.py:901] (3/4) Epoch 2, batch 4050, loss[loss=0.4664, simple_loss=0.4748, pruned_loss=0.229, over 6965.00 frames. ], tot_loss[loss=0.3753, simple_loss=0.4098, pruned_loss=0.1705, over 1611952.28 frames. ], batch size: 71, lr: 3.03e-02, grad_scale: 16.0 +2023-02-05 20:18:06,036 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12176.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:18:07,290 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12178.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:18:11,164 INFO [train.py:901] (3/4) Epoch 2, batch 4100, loss[loss=0.3794, simple_loss=0.4149, pruned_loss=0.1719, over 8301.00 frames. ], tot_loss[loss=0.3757, simple_loss=0.4099, pruned_loss=0.1707, over 1613391.16 frames. ], batch size: 23, lr: 3.02e-02, grad_scale: 16.0 +2023-02-05 20:18:23,677 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4201, 1.7545, 1.4596, 1.2305, 2.3275, 1.5498, 1.8345, 1.8550], + device='cuda:3'), covar=tensor([0.0943, 0.1638, 0.2313, 0.1923, 0.0727, 0.1841, 0.1059, 0.0802], + device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0256, 0.0278, 0.0250, 0.0225, 0.0247, 0.0218, 0.0209], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004], + device='cuda:3') +2023-02-05 20:18:27,597 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12208.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:18:30,904 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.458e+02 3.728e+02 4.672e+02 5.863e+02 2.072e+03, threshold=9.344e+02, percent-clipped=1.0 +2023-02-05 20:18:47,040 INFO [train.py:901] (3/4) Epoch 2, batch 4150, loss[loss=0.3645, simple_loss=0.4105, pruned_loss=0.1593, over 8449.00 frames. ], tot_loss[loss=0.3746, simple_loss=0.4092, pruned_loss=0.17, over 1616681.61 frames. ], batch size: 29, lr: 3.02e-02, grad_scale: 16.0 +2023-02-05 20:19:08,912 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12265.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:19:20,419 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12282.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:19:21,722 INFO [train.py:901] (3/4) Epoch 2, batch 4200, loss[loss=0.3506, simple_loss=0.3803, pruned_loss=0.1604, over 7927.00 frames. ], tot_loss[loss=0.3735, simple_loss=0.4082, pruned_loss=0.1693, over 1610782.26 frames. ], batch size: 20, lr: 3.01e-02, grad_scale: 16.0 +2023-02-05 20:19:24,792 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-05 20:19:25,938 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12290.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:19:28,644 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12293.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:19:40,062 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12310.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:19:42,058 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.324e+02 3.573e+02 4.694e+02 5.833e+02 1.413e+03, threshold=9.388e+02, percent-clipped=6.0 +2023-02-05 20:19:43,521 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-05 20:19:49,250 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12323.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:19:57,056 INFO [train.py:901] (3/4) Epoch 2, batch 4250, loss[loss=0.3567, simple_loss=0.4055, pruned_loss=0.1539, over 8684.00 frames. ], tot_loss[loss=0.3731, simple_loss=0.4082, pruned_loss=0.169, over 1609769.94 frames. ], batch size: 34, lr: 3.01e-02, grad_scale: 16.0 +2023-02-05 20:20:06,020 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12346.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:20:06,630 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-05 20:20:20,963 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12368.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:20:21,630 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0917, 1.8866, 1.8848, 1.3747, 0.9785, 1.5163, 0.3725, 0.8244], + device='cuda:3'), covar=tensor([0.1318, 0.0761, 0.0475, 0.1016, 0.1262, 0.0605, 0.2349, 0.1125], + device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0096, 0.0087, 0.0144, 0.0129, 0.0080, 0.0152, 0.0120], + device='cuda:3'), out_proj_covar=tensor([1.1333e-04, 1.0225e-04, 8.7732e-05, 1.3861e-04, 1.3136e-04, 8.0074e-05, + 1.4814e-04, 1.2437e-04], device='cuda:3') +2023-02-05 20:20:32,219 INFO [train.py:901] (3/4) Epoch 2, batch 4300, loss[loss=0.3601, simple_loss=0.4018, pruned_loss=0.1592, over 7826.00 frames. ], tot_loss[loss=0.3745, simple_loss=0.4093, pruned_loss=0.1698, over 1610601.37 frames. ], batch size: 20, lr: 3.00e-02, grad_scale: 16.0 +2023-02-05 20:20:38,554 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12393.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:20:41,210 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12397.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:20:53,214 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.244e+02 3.864e+02 4.648e+02 5.983e+02 1.525e+03, threshold=9.296e+02, percent-clipped=6.0 +2023-02-05 20:21:00,847 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12425.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:21:05,801 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12432.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:21:06,895 INFO [train.py:901] (3/4) Epoch 2, batch 4350, loss[loss=0.4815, simple_loss=0.4822, pruned_loss=0.2404, over 8623.00 frames. ], tot_loss[loss=0.3733, simple_loss=0.4082, pruned_loss=0.1693, over 1609947.42 frames. ], batch size: 39, lr: 2.99e-02, grad_scale: 8.0 +2023-02-05 20:21:09,737 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12438.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:21:23,517 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12457.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:21:26,047 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12461.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:21:28,058 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12464.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:21:38,136 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-05 20:21:42,135 INFO [train.py:901] (3/4) Epoch 2, batch 4400, loss[loss=0.3246, simple_loss=0.3837, pruned_loss=0.1327, over 8522.00 frames. ], tot_loss[loss=0.3731, simple_loss=0.4082, pruned_loss=0.1691, over 1612539.57 frames. ], batch size: 28, lr: 2.99e-02, grad_scale: 8.0 +2023-02-05 20:21:45,214 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-05 20:22:02,398 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.494e+02 4.041e+02 4.964e+02 6.742e+02 1.213e+03, threshold=9.928e+02, percent-clipped=4.0 +2023-02-05 20:22:16,720 INFO [train.py:901] (3/4) Epoch 2, batch 4450, loss[loss=0.3429, simple_loss=0.375, pruned_loss=0.1554, over 7642.00 frames. ], tot_loss[loss=0.372, simple_loss=0.407, pruned_loss=0.1685, over 1610757.49 frames. ], batch size: 19, lr: 2.98e-02, grad_scale: 8.0 +2023-02-05 20:22:17,428 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-05 20:22:22,526 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5117, 1.9877, 3.0914, 1.0176, 2.3018, 1.8260, 1.4974, 1.9052], + device='cuda:3'), covar=tensor([0.1024, 0.1073, 0.0332, 0.1679, 0.0895, 0.1359, 0.0893, 0.1194], + device='cuda:3'), in_proj_covar=tensor([0.0380, 0.0369, 0.0388, 0.0423, 0.0473, 0.0419, 0.0371, 0.0464], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 20:22:27,287 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12549.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:22:29,914 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12553.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:22:45,076 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12574.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:22:49,154 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12579.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:22:52,404 INFO [train.py:901] (3/4) Epoch 2, batch 4500, loss[loss=0.4511, simple_loss=0.4629, pruned_loss=0.2197, over 8670.00 frames. ], tot_loss[loss=0.3719, simple_loss=0.4066, pruned_loss=0.1685, over 1608616.03 frames. ], batch size: 34, lr: 2.98e-02, grad_scale: 8.0 +2023-02-05 20:22:56,171 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 +2023-02-05 20:23:06,036 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12604.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:23:12,553 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-05 20:23:13,221 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.312e+02 4.309e+02 5.092e+02 6.256e+02 1.421e+03, threshold=1.018e+03, percent-clipped=5.0 +2023-02-05 20:23:27,084 INFO [train.py:901] (3/4) Epoch 2, batch 4550, loss[loss=0.3732, simple_loss=0.4359, pruned_loss=0.1552, over 8476.00 frames. ], tot_loss[loss=0.3727, simple_loss=0.4074, pruned_loss=0.169, over 1611385.03 frames. ], batch size: 25, lr: 2.97e-02, grad_scale: 8.0 +2023-02-05 20:23:40,672 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12653.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:23:57,690 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12678.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:23:59,783 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12681.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:24:01,609 INFO [train.py:901] (3/4) Epoch 2, batch 4600, loss[loss=0.4182, simple_loss=0.4411, pruned_loss=0.1976, over 8288.00 frames. ], tot_loss[loss=0.3728, simple_loss=0.4077, pruned_loss=0.169, over 1610656.88 frames. ], batch size: 23, lr: 2.97e-02, grad_scale: 8.0 +2023-02-05 20:24:01,772 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.9264, 1.1889, 4.0681, 1.6754, 3.4989, 3.4308, 3.6023, 3.5699], + device='cuda:3'), covar=tensor([0.0354, 0.3205, 0.0267, 0.1590, 0.0888, 0.0377, 0.0358, 0.0434], + device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0374, 0.0234, 0.0269, 0.0321, 0.0257, 0.0241, 0.0262], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2023-02-05 20:24:08,005 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2422, 2.3022, 2.8261, 0.6590, 2.8998, 2.0322, 1.3703, 2.1248], + device='cuda:3'), covar=tensor([0.0119, 0.0086, 0.0125, 0.0246, 0.0103, 0.0256, 0.0232, 0.0116], + device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0134, 0.0117, 0.0188, 0.0134, 0.0243, 0.0195, 0.0167], + device='cuda:3'), out_proj_covar=tensor([1.1235e-04, 8.2744e-05, 7.5610e-05, 1.1595e-04, 8.7352e-05, 1.6206e-04, + 1.2512e-04, 1.0585e-04], device='cuda:3') +2023-02-05 20:24:17,929 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12706.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:24:23,145 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.401e+02 3.817e+02 4.647e+02 5.826e+02 1.354e+03, threshold=9.293e+02, percent-clipped=3.0 +2023-02-05 20:24:25,437 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12717.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:24:37,083 INFO [train.py:901] (3/4) Epoch 2, batch 4650, loss[loss=0.3869, simple_loss=0.4204, pruned_loss=0.1767, over 8476.00 frames. ], tot_loss[loss=0.3723, simple_loss=0.407, pruned_loss=0.1688, over 1612865.80 frames. ], batch size: 25, lr: 2.96e-02, grad_scale: 8.0 +2023-02-05 20:24:40,102 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2023-02-05 20:24:42,620 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12742.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:24:53,335 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0978, 2.3006, 2.0403, 2.6328, 1.8470, 1.7897, 2.0960, 2.5205], + device='cuda:3'), covar=tensor([0.0826, 0.1015, 0.1076, 0.0469, 0.1330, 0.1461, 0.1487, 0.0717], + device='cuda:3'), in_proj_covar=tensor([0.0315, 0.0340, 0.0332, 0.0222, 0.0317, 0.0334, 0.0379, 0.0311], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0003, 0.0004, 0.0004, 0.0005, 0.0004], + device='cuda:3') +2023-02-05 20:25:09,738 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12781.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:25:11,642 INFO [train.py:901] (3/4) Epoch 2, batch 4700, loss[loss=0.3524, simple_loss=0.4045, pruned_loss=0.1502, over 8512.00 frames. ], tot_loss[loss=0.3704, simple_loss=0.4055, pruned_loss=0.1676, over 1609609.72 frames. ], batch size: 28, lr: 2.96e-02, grad_scale: 8.0 +2023-02-05 20:25:13,847 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1894, 1.5087, 1.1183, 1.6982, 1.2663, 1.0407, 1.1566, 1.7561], + device='cuda:3'), covar=tensor([0.1018, 0.0702, 0.1438, 0.0688, 0.1260, 0.1449, 0.1195, 0.0551], + device='cuda:3'), in_proj_covar=tensor([0.0367, 0.0249, 0.0357, 0.0309, 0.0362, 0.0322, 0.0359, 0.0321], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], + device='cuda:3') +2023-02-05 20:25:14,122 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 +2023-02-05 20:25:28,731 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12808.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:25:29,557 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12809.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:25:32,792 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.540e+02 4.122e+02 5.358e+02 6.927e+02 1.344e+03, threshold=1.072e+03, percent-clipped=8.0 +2023-02-05 20:25:46,323 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 +2023-02-05 20:25:47,170 INFO [train.py:901] (3/4) Epoch 2, batch 4750, loss[loss=0.2762, simple_loss=0.3381, pruned_loss=0.1072, over 8239.00 frames. ], tot_loss[loss=0.3706, simple_loss=0.4055, pruned_loss=0.1678, over 1613581.02 frames. ], batch size: 22, lr: 2.95e-02, grad_scale: 8.0 +2023-02-05 20:25:47,389 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12834.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:26:18,679 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-05 20:26:20,730 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-05 20:26:22,727 INFO [train.py:901] (3/4) Epoch 2, batch 4800, loss[loss=0.4022, simple_loss=0.4324, pruned_loss=0.186, over 8538.00 frames. ], tot_loss[loss=0.3683, simple_loss=0.4041, pruned_loss=0.1663, over 1610770.13 frames. ], batch size: 49, lr: 2.95e-02, grad_scale: 8.0 +2023-02-05 20:26:37,696 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.72 vs. limit=5.0 +2023-02-05 20:26:41,300 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2023-02-05 20:26:42,214 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6850, 1.2330, 3.3522, 1.3401, 2.3701, 3.8869, 3.5498, 3.2890], + device='cuda:3'), covar=tensor([0.1349, 0.1884, 0.0307, 0.2109, 0.0738, 0.0184, 0.0302, 0.0517], + device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0260, 0.0179, 0.0251, 0.0192, 0.0150, 0.0145, 0.0212], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 20:26:43,405 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.314e+02 3.678e+02 4.471e+02 5.888e+02 1.234e+03, threshold=8.941e+02, percent-clipped=3.0 +2023-02-05 20:26:49,707 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12923.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:26:57,680 INFO [train.py:901] (3/4) Epoch 2, batch 4850, loss[loss=0.3965, simple_loss=0.4381, pruned_loss=0.1775, over 8346.00 frames. ], tot_loss[loss=0.3691, simple_loss=0.4053, pruned_loss=0.1664, over 1618480.21 frames. ], batch size: 26, lr: 2.94e-02, grad_scale: 8.0 +2023-02-05 20:27:12,745 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-05 20:27:32,097 INFO [train.py:901] (3/4) Epoch 2, batch 4900, loss[loss=0.3246, simple_loss=0.3656, pruned_loss=0.1418, over 7921.00 frames. ], tot_loss[loss=0.3672, simple_loss=0.403, pruned_loss=0.1657, over 1614720.11 frames. ], batch size: 20, lr: 2.94e-02, grad_scale: 8.0 +2023-02-05 20:27:53,287 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.918e+02 4.170e+02 5.532e+02 7.452e+02 1.588e+03, threshold=1.106e+03, percent-clipped=9.0 +2023-02-05 20:28:06,714 INFO [train.py:901] (3/4) Epoch 2, batch 4950, loss[loss=0.3877, simple_loss=0.4256, pruned_loss=0.1748, over 8594.00 frames. ], tot_loss[loss=0.3684, simple_loss=0.404, pruned_loss=0.1663, over 1617624.86 frames. ], batch size: 34, lr: 2.93e-02, grad_scale: 8.0 +2023-02-05 20:28:30,712 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9350, 1.7903, 2.4720, 0.9075, 2.4470, 1.8755, 1.2304, 2.1337], + device='cuda:3'), covar=tensor([0.0194, 0.0107, 0.0179, 0.0234, 0.0247, 0.0291, 0.0317, 0.0124], + device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0136, 0.0123, 0.0184, 0.0131, 0.0238, 0.0190, 0.0162], + device='cuda:3'), out_proj_covar=tensor([1.1184e-04, 8.2776e-05, 7.7799e-05, 1.1166e-04, 8.4859e-05, 1.5682e-04, + 1.1915e-04, 1.0001e-04], device='cuda:3') +2023-02-05 20:28:41,848 INFO [train.py:901] (3/4) Epoch 2, batch 5000, loss[loss=0.3924, simple_loss=0.4263, pruned_loss=0.1792, over 8584.00 frames. ], tot_loss[loss=0.3694, simple_loss=0.4047, pruned_loss=0.1671, over 1616370.63 frames. ], batch size: 34, lr: 2.93e-02, grad_scale: 8.0 +2023-02-05 20:29:02,464 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.193e+02 4.113e+02 5.050e+02 6.511e+02 1.788e+03, threshold=1.010e+03, percent-clipped=5.0 +2023-02-05 20:29:09,704 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=13125.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:29:15,847 INFO [train.py:901] (3/4) Epoch 2, batch 5050, loss[loss=0.3523, simple_loss=0.4005, pruned_loss=0.152, over 8464.00 frames. ], tot_loss[loss=0.3692, simple_loss=0.4046, pruned_loss=0.167, over 1615769.25 frames. ], batch size: 49, lr: 2.92e-02, grad_scale: 4.0 +2023-02-05 20:29:47,474 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=13179.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:29:47,972 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-05 20:29:50,616 INFO [train.py:901] (3/4) Epoch 2, batch 5100, loss[loss=0.4316, simple_loss=0.4432, pruned_loss=0.21, over 8534.00 frames. ], tot_loss[loss=0.3688, simple_loss=0.4044, pruned_loss=0.1666, over 1619134.77 frames. ], batch size: 49, lr: 2.92e-02, grad_scale: 4.0 +2023-02-05 20:30:04,642 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=13204.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:30:09,665 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3538, 4.4717, 3.8076, 2.0848, 3.8237, 3.7868, 4.0869, 3.1941], + device='cuda:3'), covar=tensor([0.0600, 0.0348, 0.0720, 0.3310, 0.0472, 0.0604, 0.0757, 0.0533], + device='cuda:3'), in_proj_covar=tensor([0.0326, 0.0220, 0.0261, 0.0346, 0.0240, 0.0188, 0.0249, 0.0165], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2023-02-05 20:30:11,534 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.243e+02 3.930e+02 4.883e+02 5.892e+02 1.355e+03, threshold=9.766e+02, percent-clipped=3.0 +2023-02-05 20:30:24,598 INFO [train.py:901] (3/4) Epoch 2, batch 5150, loss[loss=0.3312, simple_loss=0.3802, pruned_loss=0.1411, over 8185.00 frames. ], tot_loss[loss=0.3696, simple_loss=0.4052, pruned_loss=0.1671, over 1616897.86 frames. ], batch size: 23, lr: 2.91e-02, grad_scale: 4.0 +2023-02-05 20:30:28,706 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=13240.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:30:38,414 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.35 vs. limit=5.0 +2023-02-05 20:30:59,010 INFO [train.py:901] (3/4) Epoch 2, batch 5200, loss[loss=0.3862, simple_loss=0.4139, pruned_loss=0.1792, over 8543.00 frames. ], tot_loss[loss=0.3724, simple_loss=0.4069, pruned_loss=0.169, over 1613027.20 frames. ], batch size: 39, lr: 2.91e-02, grad_scale: 8.0 +2023-02-05 20:31:20,910 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.071e+02 4.339e+02 5.206e+02 6.705e+02 1.063e+03, threshold=1.041e+03, percent-clipped=3.0 +2023-02-05 20:31:33,618 INFO [train.py:901] (3/4) Epoch 2, batch 5250, loss[loss=0.4088, simple_loss=0.4471, pruned_loss=0.1853, over 8507.00 frames. ], tot_loss[loss=0.3708, simple_loss=0.4057, pruned_loss=0.1679, over 1609453.03 frames. ], batch size: 28, lr: 2.91e-02, grad_scale: 8.0 +2023-02-05 20:31:42,977 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-05 20:32:03,333 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 +2023-02-05 20:32:07,567 INFO [train.py:901] (3/4) Epoch 2, batch 5300, loss[loss=0.3745, simple_loss=0.4174, pruned_loss=0.1657, over 8601.00 frames. ], tot_loss[loss=0.3705, simple_loss=0.406, pruned_loss=0.1675, over 1613001.93 frames. ], batch size: 34, lr: 2.90e-02, grad_scale: 8.0 +2023-02-05 20:32:12,473 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3928, 2.0691, 3.2243, 2.9753, 2.6646, 1.9095, 1.5192, 1.9530], + device='cuda:3'), covar=tensor([0.0578, 0.0653, 0.0131, 0.0185, 0.0299, 0.0307, 0.0413, 0.0514], + device='cuda:3'), in_proj_covar=tensor([0.0494, 0.0415, 0.0315, 0.0346, 0.0436, 0.0382, 0.0403, 0.0431], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 20:32:29,092 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.093e+02 3.821e+02 4.884e+02 6.417e+02 1.823e+03, threshold=9.767e+02, percent-clipped=6.0 +2023-02-05 20:32:42,524 INFO [train.py:901] (3/4) Epoch 2, batch 5350, loss[loss=0.3318, simple_loss=0.393, pruned_loss=0.1353, over 8472.00 frames. ], tot_loss[loss=0.3693, simple_loss=0.4049, pruned_loss=0.1668, over 1610544.90 frames. ], batch size: 25, lr: 2.90e-02, grad_scale: 8.0 +2023-02-05 20:33:16,588 INFO [train.py:901] (3/4) Epoch 2, batch 5400, loss[loss=0.3277, simple_loss=0.3692, pruned_loss=0.1431, over 7791.00 frames. ], tot_loss[loss=0.3699, simple_loss=0.4058, pruned_loss=0.167, over 1612424.81 frames. ], batch size: 19, lr: 2.89e-02, grad_scale: 8.0 +2023-02-05 20:33:24,898 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=13496.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:33:38,013 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.355e+02 3.820e+02 4.559e+02 5.766e+02 1.205e+03, threshold=9.119e+02, percent-clipped=6.0 +2023-02-05 20:33:43,049 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=13521.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:33:51,331 INFO [train.py:901] (3/4) Epoch 2, batch 5450, loss[loss=0.3571, simple_loss=0.3954, pruned_loss=0.1594, over 8246.00 frames. ], tot_loss[loss=0.3671, simple_loss=0.4045, pruned_loss=0.1649, over 1614790.52 frames. ], batch size: 22, lr: 2.89e-02, grad_scale: 8.0 +2023-02-05 20:34:25,971 INFO [train.py:901] (3/4) Epoch 2, batch 5500, loss[loss=0.3154, simple_loss=0.3551, pruned_loss=0.1379, over 7809.00 frames. ], tot_loss[loss=0.3687, simple_loss=0.4056, pruned_loss=0.1659, over 1614323.91 frames. ], batch size: 20, lr: 2.88e-02, grad_scale: 8.0 +2023-02-05 20:34:28,053 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-05 20:34:28,412 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=5.45 vs. limit=5.0 +2023-02-05 20:34:46,537 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.267e+02 3.726e+02 4.817e+02 6.308e+02 1.682e+03, threshold=9.635e+02, percent-clipped=6.0 +2023-02-05 20:34:59,986 INFO [train.py:901] (3/4) Epoch 2, batch 5550, loss[loss=0.31, simple_loss=0.3501, pruned_loss=0.1349, over 7819.00 frames. ], tot_loss[loss=0.3684, simple_loss=0.4049, pruned_loss=0.166, over 1612018.77 frames. ], batch size: 20, lr: 2.88e-02, grad_scale: 8.0 +2023-02-05 20:35:35,317 INFO [train.py:901] (3/4) Epoch 2, batch 5600, loss[loss=0.358, simple_loss=0.3812, pruned_loss=0.1673, over 7788.00 frames. ], tot_loss[loss=0.3659, simple_loss=0.4032, pruned_loss=0.1643, over 1613689.42 frames. ], batch size: 19, lr: 2.87e-02, grad_scale: 8.0 +2023-02-05 20:35:55,774 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.674e+02 3.833e+02 4.619e+02 6.071e+02 1.383e+03, threshold=9.238e+02, percent-clipped=5.0 +2023-02-05 20:36:08,578 INFO [train.py:901] (3/4) Epoch 2, batch 5650, loss[loss=0.4043, simple_loss=0.4238, pruned_loss=0.1924, over 8292.00 frames. ], tot_loss[loss=0.3671, simple_loss=0.4037, pruned_loss=0.1653, over 1607681.86 frames. ], batch size: 48, lr: 2.87e-02, grad_scale: 8.0 +2023-02-05 20:36:23,374 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=13755.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:36:34,189 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-05 20:36:35,122 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.13 vs. limit=5.0 +2023-02-05 20:36:43,556 INFO [train.py:901] (3/4) Epoch 2, batch 5700, loss[loss=0.3375, simple_loss=0.3882, pruned_loss=0.1435, over 8029.00 frames. ], tot_loss[loss=0.3659, simple_loss=0.4026, pruned_loss=0.1646, over 1607957.38 frames. ], batch size: 22, lr: 2.86e-02, grad_scale: 8.0 +2023-02-05 20:37:05,589 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.337e+02 4.261e+02 5.123e+02 6.631e+02 2.352e+03, threshold=1.025e+03, percent-clipped=5.0 +2023-02-05 20:37:18,906 INFO [train.py:901] (3/4) Epoch 2, batch 5750, loss[loss=0.3491, simple_loss=0.4019, pruned_loss=0.1482, over 8251.00 frames. ], tot_loss[loss=0.3651, simple_loss=0.402, pruned_loss=0.1641, over 1607389.49 frames. ], batch size: 24, lr: 2.86e-02, grad_scale: 8.0 +2023-02-05 20:37:38,953 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-05 20:37:54,461 INFO [train.py:901] (3/4) Epoch 2, batch 5800, loss[loss=0.4165, simple_loss=0.4425, pruned_loss=0.1953, over 8492.00 frames. ], tot_loss[loss=0.3638, simple_loss=0.4017, pruned_loss=0.163, over 1609899.16 frames. ], batch size: 29, lr: 2.85e-02, grad_scale: 8.0 +2023-02-05 20:38:11,985 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4476, 1.7129, 3.8837, 1.8388, 2.0316, 4.4247, 3.9079, 3.9517], + device='cuda:3'), covar=tensor([0.1175, 0.1659, 0.0315, 0.1894, 0.0972, 0.0320, 0.0354, 0.0552], + device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0262, 0.0186, 0.0254, 0.0194, 0.0157, 0.0148, 0.0229], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 20:38:15,256 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2470, 2.2072, 1.5034, 1.2904, 2.1316, 1.7354, 2.4107, 2.3799], + device='cuda:3'), covar=tensor([0.0922, 0.1502, 0.2152, 0.1943, 0.0881, 0.1798, 0.1036, 0.0696], + device='cuda:3'), in_proj_covar=tensor([0.0221, 0.0250, 0.0279, 0.0248, 0.0220, 0.0243, 0.0210, 0.0205], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004], + device='cuda:3') +2023-02-05 20:38:15,742 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.601e+02 3.784e+02 4.729e+02 6.225e+02 2.390e+03, threshold=9.458e+02, percent-clipped=5.0 +2023-02-05 20:38:29,067 INFO [train.py:901] (3/4) Epoch 2, batch 5850, loss[loss=0.4095, simple_loss=0.4406, pruned_loss=0.1892, over 8539.00 frames. ], tot_loss[loss=0.364, simple_loss=0.4016, pruned_loss=0.1632, over 1613739.62 frames. ], batch size: 31, lr: 2.85e-02, grad_scale: 8.0 +2023-02-05 20:38:58,029 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.3125, 2.4512, 4.1705, 3.9550, 3.0136, 2.5019, 1.8470, 2.2108], + device='cuda:3'), covar=tensor([0.0588, 0.0875, 0.0153, 0.0213, 0.0414, 0.0308, 0.0432, 0.0703], + device='cuda:3'), in_proj_covar=tensor([0.0489, 0.0416, 0.0311, 0.0351, 0.0450, 0.0388, 0.0404, 0.0435], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 20:39:03,904 INFO [train.py:901] (3/4) Epoch 2, batch 5900, loss[loss=0.3359, simple_loss=0.3928, pruned_loss=0.1395, over 8258.00 frames. ], tot_loss[loss=0.3651, simple_loss=0.4023, pruned_loss=0.1639, over 1614372.24 frames. ], batch size: 24, lr: 2.84e-02, grad_scale: 8.0 +2023-02-05 20:39:05,670 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.04 vs. limit=5.0 +2023-02-05 20:39:27,081 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.452e+02 3.946e+02 4.724e+02 6.297e+02 1.551e+03, threshold=9.448e+02, percent-clipped=7.0 +2023-02-05 20:39:30,120 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.60 vs. limit=5.0 +2023-02-05 20:39:40,155 INFO [train.py:901] (3/4) Epoch 2, batch 5950, loss[loss=0.3833, simple_loss=0.4143, pruned_loss=0.1762, over 8296.00 frames. ], tot_loss[loss=0.3642, simple_loss=0.402, pruned_loss=0.1632, over 1614944.72 frames. ], batch size: 23, lr: 2.84e-02, grad_scale: 8.0 +2023-02-05 20:40:14,643 INFO [train.py:901] (3/4) Epoch 2, batch 6000, loss[loss=0.3797, simple_loss=0.4127, pruned_loss=0.1734, over 8496.00 frames. ], tot_loss[loss=0.3631, simple_loss=0.4015, pruned_loss=0.1624, over 1615698.07 frames. ], batch size: 26, lr: 2.84e-02, grad_scale: 8.0 +2023-02-05 20:40:14,643 INFO [train.py:926] (3/4) Computing validation loss +2023-02-05 20:40:24,524 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([0.9187, 1.1742, 1.2710, 0.8340, 0.7641, 1.1568, 0.1463, 0.5454], + device='cuda:3'), covar=tensor([0.1849, 0.1714, 0.0716, 0.1934, 0.2670, 0.0917, 0.3981, 0.2130], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0095, 0.0083, 0.0140, 0.0140, 0.0082, 0.0157, 0.0118], + device='cuda:3'), out_proj_covar=tensor([1.1407e-04, 1.0875e-04, 8.9206e-05, 1.4468e-04, 1.4930e-04, 9.0084e-05, + 1.6301e-04, 1.2994e-04], device='cuda:3') +2023-02-05 20:40:26,576 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0038, 1.5560, 1.4381, 0.3280, 1.4115, 0.9478, 0.2091, 1.4932], + device='cuda:3'), covar=tensor([0.0137, 0.0066, 0.0085, 0.0160, 0.0075, 0.0257, 0.0207, 0.0070], + device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0140, 0.0130, 0.0183, 0.0136, 0.0249, 0.0196, 0.0178], + device='cuda:3'), out_proj_covar=tensor([1.1280e-04, 8.1912e-05, 7.9440e-05, 1.0615e-04, 8.3237e-05, 1.5626e-04, + 1.1841e-04, 1.0724e-04], device='cuda:3') +2023-02-05 20:40:27,828 INFO [train.py:935] (3/4) Epoch 2, validation: loss=0.2758, simple_loss=0.3606, pruned_loss=0.0955, over 944034.00 frames. +2023-02-05 20:40:27,829 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-05 20:40:32,717 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14090.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:40:38,744 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14099.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:40:49,504 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.396e+02 3.733e+02 4.780e+02 6.772e+02 2.203e+03, threshold=9.561e+02, percent-clipped=10.0 +2023-02-05 20:40:53,735 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14121.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:41:02,698 INFO [train.py:901] (3/4) Epoch 2, batch 6050, loss[loss=0.3225, simple_loss=0.373, pruned_loss=0.136, over 8287.00 frames. ], tot_loss[loss=0.3635, simple_loss=0.4013, pruned_loss=0.1629, over 1613946.66 frames. ], batch size: 23, lr: 2.83e-02, grad_scale: 8.0 +2023-02-05 20:41:05,428 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14138.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 20:41:37,178 INFO [train.py:901] (3/4) Epoch 2, batch 6100, loss[loss=0.3072, simple_loss=0.3429, pruned_loss=0.1358, over 7668.00 frames. ], tot_loss[loss=0.3652, simple_loss=0.402, pruned_loss=0.1641, over 1608649.77 frames. ], batch size: 19, lr: 2.83e-02, grad_scale: 8.0 +2023-02-05 20:41:58,433 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14214.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:41:58,921 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.150e+02 3.920e+02 4.920e+02 6.492e+02 2.677e+03, threshold=9.840e+02, percent-clipped=6.0 +2023-02-05 20:42:03,176 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.67 vs. limit=2.0 +2023-02-05 20:42:05,167 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.04 vs. limit=2.0 +2023-02-05 20:42:08,080 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-05 20:42:11,461 INFO [train.py:901] (3/4) Epoch 2, batch 6150, loss[loss=0.3308, simple_loss=0.3934, pruned_loss=0.1341, over 8467.00 frames. ], tot_loss[loss=0.3624, simple_loss=0.4004, pruned_loss=0.1622, over 1611958.19 frames. ], batch size: 25, lr: 2.82e-02, grad_scale: 8.0 +2023-02-05 20:42:46,433 INFO [train.py:901] (3/4) Epoch 2, batch 6200, loss[loss=0.3505, simple_loss=0.3853, pruned_loss=0.1579, over 7816.00 frames. ], tot_loss[loss=0.3644, simple_loss=0.4018, pruned_loss=0.1635, over 1612829.77 frames. ], batch size: 20, lr: 2.82e-02, grad_scale: 8.0 +2023-02-05 20:43:08,135 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.406e+02 3.453e+02 4.846e+02 6.394e+02 2.249e+03, threshold=9.691e+02, percent-clipped=6.0 +2023-02-05 20:43:21,530 INFO [train.py:901] (3/4) Epoch 2, batch 6250, loss[loss=0.3476, simple_loss=0.3869, pruned_loss=0.1541, over 7555.00 frames. ], tot_loss[loss=0.3631, simple_loss=0.4005, pruned_loss=0.1629, over 1609773.32 frames. ], batch size: 18, lr: 2.81e-02, grad_scale: 8.0 +2023-02-05 20:43:55,857 INFO [train.py:901] (3/4) Epoch 2, batch 6300, loss[loss=0.3759, simple_loss=0.4303, pruned_loss=0.1608, over 8256.00 frames. ], tot_loss[loss=0.364, simple_loss=0.4011, pruned_loss=0.1635, over 1609404.18 frames. ], batch size: 24, lr: 2.81e-02, grad_scale: 8.0 +2023-02-05 20:44:17,505 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.643e+02 3.823e+02 4.655e+02 5.877e+02 1.568e+03, threshold=9.309e+02, percent-clipped=4.0 +2023-02-05 20:44:28,394 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14431.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:44:30,257 INFO [train.py:901] (3/4) Epoch 2, batch 6350, loss[loss=0.391, simple_loss=0.4236, pruned_loss=0.1792, over 8649.00 frames. ], tot_loss[loss=0.3615, simple_loss=0.3999, pruned_loss=0.1616, over 1610724.85 frames. ], batch size: 31, lr: 2.81e-02, grad_scale: 8.0 +2023-02-05 20:44:30,321 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14434.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:44:43,820 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2023-02-05 20:44:51,467 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14465.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:44:54,888 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14470.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:45:03,123 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14482.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 20:45:04,274 INFO [train.py:901] (3/4) Epoch 2, batch 6400, loss[loss=0.3954, simple_loss=0.4201, pruned_loss=0.1853, over 8519.00 frames. ], tot_loss[loss=0.3634, simple_loss=0.4014, pruned_loss=0.1627, over 1614130.84 frames. ], batch size: 31, lr: 2.80e-02, grad_scale: 8.0 +2023-02-05 20:45:12,404 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14495.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:45:19,142 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14505.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:45:25,558 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.523e+02 3.974e+02 5.065e+02 7.362e+02 1.328e+03, threshold=1.013e+03, percent-clipped=8.0 +2023-02-05 20:45:38,730 INFO [train.py:901] (3/4) Epoch 2, batch 6450, loss[loss=0.3469, simple_loss=0.3969, pruned_loss=0.1484, over 8198.00 frames. ], tot_loss[loss=0.3639, simple_loss=0.4014, pruned_loss=0.1632, over 1615566.59 frames. ], batch size: 23, lr: 2.80e-02, grad_scale: 8.0 +2023-02-05 20:45:48,962 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14549.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:45:59,284 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.65 vs. limit=2.0 +2023-02-05 20:46:10,576 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14580.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:46:13,166 INFO [train.py:901] (3/4) Epoch 2, batch 6500, loss[loss=0.3102, simple_loss=0.3524, pruned_loss=0.134, over 7692.00 frames. ], tot_loss[loss=0.3605, simple_loss=0.3994, pruned_loss=0.1609, over 1614714.20 frames. ], batch size: 18, lr: 2.79e-02, grad_scale: 8.0 +2023-02-05 20:46:22,646 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14597.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 20:46:35,355 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.440e+02 3.999e+02 5.009e+02 6.288e+02 1.522e+03, threshold=1.002e+03, percent-clipped=8.0 +2023-02-05 20:46:48,426 INFO [train.py:901] (3/4) Epoch 2, batch 6550, loss[loss=0.3455, simple_loss=0.4015, pruned_loss=0.1448, over 8111.00 frames. ], tot_loss[loss=0.3621, simple_loss=0.4003, pruned_loss=0.162, over 1612654.76 frames. ], batch size: 23, lr: 2.79e-02, grad_scale: 8.0 +2023-02-05 20:47:16,653 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-05 20:47:23,549 INFO [train.py:901] (3/4) Epoch 2, batch 6600, loss[loss=0.3661, simple_loss=0.4086, pruned_loss=0.1618, over 8253.00 frames. ], tot_loss[loss=0.3593, simple_loss=0.3982, pruned_loss=0.1602, over 1611346.01 frames. ], batch size: 24, lr: 2.78e-02, grad_scale: 8.0 +2023-02-05 20:47:36,555 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-05 20:47:45,888 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.289e+02 3.681e+02 4.457e+02 5.556e+02 1.208e+03, threshold=8.913e+02, percent-clipped=4.0 +2023-02-05 20:47:58,942 INFO [train.py:901] (3/4) Epoch 2, batch 6650, loss[loss=0.4196, simple_loss=0.4474, pruned_loss=0.1959, over 8445.00 frames. ], tot_loss[loss=0.3601, simple_loss=0.399, pruned_loss=0.1606, over 1615967.39 frames. ], batch size: 29, lr: 2.78e-02, grad_scale: 8.0 +2023-02-05 20:48:14,410 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2066, 1.9257, 3.1520, 2.7113, 2.3454, 1.9594, 1.2968, 1.2772], + device='cuda:3'), covar=tensor([0.0698, 0.0768, 0.0139, 0.0261, 0.0353, 0.0361, 0.0512, 0.0728], + device='cuda:3'), in_proj_covar=tensor([0.0537, 0.0456, 0.0339, 0.0388, 0.0490, 0.0417, 0.0443, 0.0462], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 20:48:16,415 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14758.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:48:28,651 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14775.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:48:34,791 INFO [train.py:901] (3/4) Epoch 2, batch 6700, loss[loss=0.3393, simple_loss=0.3841, pruned_loss=0.1473, over 8034.00 frames. ], tot_loss[loss=0.359, simple_loss=0.3982, pruned_loss=0.1599, over 1613337.03 frames. ], batch size: 22, lr: 2.78e-02, grad_scale: 8.0 +2023-02-05 20:48:50,193 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14805.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:48:56,685 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.056e+02 3.873e+02 4.634e+02 6.203e+02 1.536e+03, threshold=9.268e+02, percent-clipped=6.0 +2023-02-05 20:49:07,131 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14830.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:49:10,358 INFO [train.py:901] (3/4) Epoch 2, batch 6750, loss[loss=0.3279, simple_loss=0.3829, pruned_loss=0.1364, over 8511.00 frames. ], tot_loss[loss=0.3603, simple_loss=0.3995, pruned_loss=0.1606, over 1615895.21 frames. ], batch size: 26, lr: 2.77e-02, grad_scale: 8.0 +2023-02-05 20:49:11,906 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14836.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:49:21,259 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14849.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:49:24,081 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14853.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 20:49:29,654 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14861.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:49:29,663 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([0.8794, 1.0448, 1.0740, 0.9553, 0.7195, 1.0797, 0.0043, 0.6409], + device='cuda:3'), covar=tensor([0.1749, 0.1326, 0.0945, 0.1320, 0.2658, 0.0721, 0.4086, 0.1798], + device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0100, 0.0084, 0.0142, 0.0145, 0.0082, 0.0163, 0.0116], + device='cuda:3'), out_proj_covar=tensor([1.1983e-04, 1.1672e-04, 9.2567e-05, 1.4989e-04, 1.5721e-04, 9.2951e-05, + 1.7304e-04, 1.3071e-04], device='cuda:3') +2023-02-05 20:49:41,490 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14878.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 20:49:45,988 INFO [train.py:901] (3/4) Epoch 2, batch 6800, loss[loss=0.3874, simple_loss=0.4194, pruned_loss=0.1777, over 8608.00 frames. ], tot_loss[loss=0.3597, simple_loss=0.3985, pruned_loss=0.1604, over 1612090.27 frames. ], batch size: 31, lr: 2.77e-02, grad_scale: 8.0 +2023-02-05 20:49:50,348 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14890.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:49:54,306 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-05 20:50:04,552 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.82 vs. limit=5.0 +2023-02-05 20:50:07,690 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.027e+02 3.663e+02 4.715e+02 6.092e+02 1.805e+03, threshold=9.431e+02, percent-clipped=7.0 +2023-02-05 20:50:20,401 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.97 vs. limit=5.0 +2023-02-05 20:50:21,321 INFO [train.py:901] (3/4) Epoch 2, batch 6850, loss[loss=0.2969, simple_loss=0.3601, pruned_loss=0.1169, over 8191.00 frames. ], tot_loss[loss=0.3584, simple_loss=0.3981, pruned_loss=0.1594, over 1616658.73 frames. ], batch size: 23, lr: 2.76e-02, grad_scale: 8.0 +2023-02-05 20:50:42,735 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14964.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:50:45,366 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-05 20:50:57,103 INFO [train.py:901] (3/4) Epoch 2, batch 6900, loss[loss=0.2919, simple_loss=0.3449, pruned_loss=0.1194, over 8033.00 frames. ], tot_loss[loss=0.3577, simple_loss=0.3972, pruned_loss=0.1591, over 1611692.41 frames. ], batch size: 22, lr: 2.76e-02, grad_scale: 8.0 +2023-02-05 20:51:19,285 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.011e+02 4.191e+02 5.097e+02 7.005e+02 1.700e+03, threshold=1.019e+03, percent-clipped=5.0 +2023-02-05 20:51:30,703 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8531, 2.2786, 2.8661, 0.2623, 2.9086, 1.8322, 1.2592, 2.0467], + device='cuda:3'), covar=tensor([0.0145, 0.0066, 0.0087, 0.0225, 0.0118, 0.0205, 0.0225, 0.0102], + device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0133, 0.0123, 0.0186, 0.0132, 0.0252, 0.0200, 0.0171], + device='cuda:3'), out_proj_covar=tensor([1.1061e-04, 7.4798e-05, 7.1916e-05, 1.0345e-04, 7.7725e-05, 1.5465e-04, + 1.1479e-04, 9.8079e-05], device='cuda:3') +2023-02-05 20:51:32,586 INFO [train.py:901] (3/4) Epoch 2, batch 6950, loss[loss=0.3714, simple_loss=0.3869, pruned_loss=0.178, over 7252.00 frames. ], tot_loss[loss=0.3577, simple_loss=0.3966, pruned_loss=0.1594, over 1600959.03 frames. ], batch size: 16, lr: 2.75e-02, grad_scale: 8.0 +2023-02-05 20:51:56,484 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-05 20:52:08,410 INFO [train.py:901] (3/4) Epoch 2, batch 7000, loss[loss=0.3213, simple_loss=0.3728, pruned_loss=0.1349, over 7982.00 frames. ], tot_loss[loss=0.3567, simple_loss=0.3962, pruned_loss=0.1586, over 1609016.27 frames. ], batch size: 21, lr: 2.75e-02, grad_scale: 8.0 +2023-02-05 20:52:21,505 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15102.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:52:30,568 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.233e+02 3.928e+02 4.810e+02 5.818e+02 1.410e+03, threshold=9.621e+02, percent-clipped=1.0 +2023-02-05 20:52:44,338 INFO [train.py:901] (3/4) Epoch 2, batch 7050, loss[loss=0.3643, simple_loss=0.4106, pruned_loss=0.159, over 8500.00 frames. ], tot_loss[loss=0.3571, simple_loss=0.3966, pruned_loss=0.1588, over 1611823.90 frames. ], batch size: 26, lr: 2.75e-02, grad_scale: 16.0 +2023-02-05 20:52:52,907 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15146.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:52:57,901 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-05 20:53:10,279 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15171.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:53:18,778 INFO [train.py:901] (3/4) Epoch 2, batch 7100, loss[loss=0.3975, simple_loss=0.4326, pruned_loss=0.1812, over 8111.00 frames. ], tot_loss[loss=0.3574, simple_loss=0.3967, pruned_loss=0.159, over 1612470.87 frames. ], batch size: 23, lr: 2.74e-02, grad_scale: 16.0 +2023-02-05 20:53:39,783 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15213.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:53:41,005 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.137e+02 3.718e+02 4.413e+02 5.855e+02 1.165e+03, threshold=8.826e+02, percent-clipped=3.0 +2023-02-05 20:53:42,518 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15217.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:53:44,504 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15220.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:53:53,629 INFO [train.py:901] (3/4) Epoch 2, batch 7150, loss[loss=0.3672, simple_loss=0.3966, pruned_loss=0.1689, over 7210.00 frames. ], tot_loss[loss=0.3553, simple_loss=0.3949, pruned_loss=0.1579, over 1608951.16 frames. ], batch size: 16, lr: 2.74e-02, grad_scale: 16.0 +2023-02-05 20:53:55,204 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6575, 2.3648, 3.6865, 1.0852, 2.2068, 1.9036, 1.5932, 1.8825], + device='cuda:3'), covar=tensor([0.1003, 0.0960, 0.0352, 0.1823, 0.1029, 0.1678, 0.1018, 0.1571], + device='cuda:3'), in_proj_covar=tensor([0.0409, 0.0387, 0.0439, 0.0457, 0.0511, 0.0458, 0.0404, 0.0515], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 20:54:02,061 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15245.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:54:17,251 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-05 20:54:29,184 INFO [train.py:901] (3/4) Epoch 2, batch 7200, loss[loss=0.3647, simple_loss=0.4067, pruned_loss=0.1614, over 8324.00 frames. ], tot_loss[loss=0.3543, simple_loss=0.3944, pruned_loss=0.1571, over 1607093.97 frames. ], batch size: 25, lr: 2.73e-02, grad_scale: 16.0 +2023-02-05 20:54:51,169 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.528e+02 3.704e+02 4.905e+02 6.625e+02 1.855e+03, threshold=9.809e+02, percent-clipped=12.0 +2023-02-05 20:54:55,123 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 +2023-02-05 20:55:04,889 INFO [train.py:901] (3/4) Epoch 2, batch 7250, loss[loss=0.3471, simple_loss=0.3939, pruned_loss=0.1502, over 8192.00 frames. ], tot_loss[loss=0.3567, simple_loss=0.396, pruned_loss=0.1586, over 1604175.84 frames. ], batch size: 23, lr: 2.73e-02, grad_scale: 8.0 +2023-02-05 20:55:39,922 INFO [train.py:901] (3/4) Epoch 2, batch 7300, loss[loss=0.2944, simple_loss=0.3591, pruned_loss=0.1148, over 7544.00 frames. ], tot_loss[loss=0.3574, simple_loss=0.397, pruned_loss=0.1589, over 1604917.39 frames. ], batch size: 18, lr: 2.73e-02, grad_scale: 8.0 +2023-02-05 20:56:02,317 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.194e+02 3.434e+02 4.292e+02 5.923e+02 1.449e+03, threshold=8.584e+02, percent-clipped=5.0 +2023-02-05 20:56:14,879 INFO [train.py:901] (3/4) Epoch 2, batch 7350, loss[loss=0.3121, simple_loss=0.3738, pruned_loss=0.1252, over 8041.00 frames. ], tot_loss[loss=0.3552, simple_loss=0.3953, pruned_loss=0.1575, over 1606004.79 frames. ], batch size: 22, lr: 2.72e-02, grad_scale: 8.0 +2023-02-05 20:56:42,793 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15473.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:56:43,923 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-05 20:56:49,843 INFO [train.py:901] (3/4) Epoch 2, batch 7400, loss[loss=0.329, simple_loss=0.394, pruned_loss=0.132, over 8328.00 frames. ], tot_loss[loss=0.356, simple_loss=0.3962, pruned_loss=0.1579, over 1607808.80 frames. ], batch size: 25, lr: 2.72e-02, grad_scale: 8.0 +2023-02-05 20:56:59,513 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15498.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:57:01,976 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-05 20:57:11,815 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.949e+02 3.940e+02 4.956e+02 6.362e+02 1.377e+03, threshold=9.912e+02, percent-clipped=7.0 +2023-02-05 20:57:24,681 INFO [train.py:901] (3/4) Epoch 2, batch 7450, loss[loss=0.3489, simple_loss=0.4047, pruned_loss=0.1466, over 8344.00 frames. ], tot_loss[loss=0.3574, simple_loss=0.3976, pruned_loss=0.1586, over 1609005.28 frames. ], batch size: 26, lr: 2.71e-02, grad_scale: 8.0 +2023-02-05 20:57:40,477 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15557.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:57:41,773 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-05 20:57:59,025 INFO [train.py:901] (3/4) Epoch 2, batch 7500, loss[loss=0.3294, simple_loss=0.3799, pruned_loss=0.1394, over 8030.00 frames. ], tot_loss[loss=0.3557, simple_loss=0.3959, pruned_loss=0.1578, over 1606962.85 frames. ], batch size: 22, lr: 2.71e-02, grad_scale: 8.0 +2023-02-05 20:58:21,359 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.209e+02 3.662e+02 4.519e+02 5.678e+02 1.466e+03, threshold=9.038e+02, percent-clipped=6.0 +2023-02-05 20:58:34,047 INFO [train.py:901] (3/4) Epoch 2, batch 7550, loss[loss=0.3468, simple_loss=0.3966, pruned_loss=0.1485, over 8707.00 frames. ], tot_loss[loss=0.354, simple_loss=0.3943, pruned_loss=0.1568, over 1605818.34 frames. ], batch size: 34, lr: 2.71e-02, grad_scale: 8.0 +2023-02-05 20:58:46,868 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3044, 1.4829, 1.3909, 1.5337, 0.9544, 1.8865, 0.2752, 0.9188], + device='cuda:3'), covar=tensor([0.1736, 0.1087, 0.0698, 0.1306, 0.1834, 0.0546, 0.3648, 0.1422], + device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0091, 0.0077, 0.0134, 0.0135, 0.0076, 0.0144, 0.0108], + device='cuda:3'), out_proj_covar=tensor([1.1914e-04, 1.0971e-04, 8.8119e-05, 1.4635e-04, 1.4930e-04, 9.0133e-05, + 1.5860e-04, 1.2735e-04], device='cuda:3') +2023-02-05 20:59:00,936 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15672.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 20:59:08,582 INFO [train.py:901] (3/4) Epoch 2, batch 7600, loss[loss=0.2765, simple_loss=0.3277, pruned_loss=0.1127, over 7788.00 frames. ], tot_loss[loss=0.3558, simple_loss=0.396, pruned_loss=0.1578, over 1612075.31 frames. ], batch size: 19, lr: 2.70e-02, grad_scale: 8.0 +2023-02-05 20:59:31,058 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.044e+02 3.634e+02 4.473e+02 6.191e+02 1.516e+03, threshold=8.946e+02, percent-clipped=5.0 +2023-02-05 20:59:35,833 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3405, 3.6965, 2.1712, 2.6077, 2.8704, 1.9294, 2.2480, 2.6338], + device='cuda:3'), covar=tensor([0.1474, 0.0436, 0.1043, 0.1000, 0.0863, 0.1320, 0.1521, 0.0932], + device='cuda:3'), in_proj_covar=tensor([0.0388, 0.0244, 0.0350, 0.0324, 0.0359, 0.0332, 0.0363, 0.0323], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-05 20:59:40,160 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-05 20:59:43,079 INFO [train.py:901] (3/4) Epoch 2, batch 7650, loss[loss=0.3897, simple_loss=0.4298, pruned_loss=0.1748, over 8344.00 frames. ], tot_loss[loss=0.3555, simple_loss=0.3957, pruned_loss=0.1576, over 1608661.53 frames. ], batch size: 26, lr: 2.70e-02, grad_scale: 8.0 +2023-02-05 21:00:02,563 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.3405, 2.4696, 4.3645, 3.5512, 3.3120, 2.7750, 1.6375, 1.9670], + device='cuda:3'), covar=tensor([0.0616, 0.0954, 0.0159, 0.0315, 0.0373, 0.0314, 0.0485, 0.0870], + device='cuda:3'), in_proj_covar=tensor([0.0527, 0.0453, 0.0348, 0.0386, 0.0484, 0.0419, 0.0440, 0.0457], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 21:00:19,423 INFO [train.py:901] (3/4) Epoch 2, batch 7700, loss[loss=0.3325, simple_loss=0.3721, pruned_loss=0.1465, over 7824.00 frames. ], tot_loss[loss=0.3543, simple_loss=0.3949, pruned_loss=0.1568, over 1610618.22 frames. ], batch size: 20, lr: 2.69e-02, grad_scale: 8.0 +2023-02-05 21:00:25,146 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.4425, 2.3760, 4.4142, 3.3889, 2.9448, 2.4905, 1.7185, 1.9807], + device='cuda:3'), covar=tensor([0.0607, 0.0948, 0.0176, 0.0314, 0.0471, 0.0366, 0.0512, 0.0842], + device='cuda:3'), in_proj_covar=tensor([0.0537, 0.0466, 0.0358, 0.0396, 0.0496, 0.0429, 0.0451, 0.0468], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 21:00:41,050 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.759e+02 3.880e+02 4.902e+02 6.175e+02 1.322e+03, threshold=9.805e+02, percent-clipped=4.0 +2023-02-05 21:00:43,410 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7733, 1.9069, 3.7748, 1.0649, 2.3229, 1.9947, 1.7170, 2.0684], + device='cuda:3'), covar=tensor([0.0961, 0.1275, 0.0313, 0.2128, 0.1056, 0.1507, 0.0918, 0.1562], + device='cuda:3'), in_proj_covar=tensor([0.0405, 0.0380, 0.0426, 0.0460, 0.0506, 0.0444, 0.0397, 0.0506], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 21:00:51,196 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-05 21:00:53,925 INFO [train.py:901] (3/4) Epoch 2, batch 7750, loss[loss=0.3869, simple_loss=0.4242, pruned_loss=0.1748, over 8507.00 frames. ], tot_loss[loss=0.3541, simple_loss=0.3945, pruned_loss=0.1569, over 1609548.15 frames. ], batch size: 28, lr: 2.69e-02, grad_scale: 8.0 +2023-02-05 21:01:28,170 INFO [train.py:901] (3/4) Epoch 2, batch 7800, loss[loss=0.3127, simple_loss=0.347, pruned_loss=0.1392, over 7689.00 frames. ], tot_loss[loss=0.3531, simple_loss=0.3938, pruned_loss=0.1562, over 1612551.30 frames. ], batch size: 18, lr: 2.69e-02, grad_scale: 8.0 +2023-02-05 21:01:41,032 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15901.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:01:50,927 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.227e+02 3.569e+02 4.742e+02 5.990e+02 9.896e+02, threshold=9.484e+02, percent-clipped=1.0 +2023-02-05 21:01:59,055 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15928.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:02:02,911 INFO [train.py:901] (3/4) Epoch 2, batch 7850, loss[loss=0.343, simple_loss=0.3856, pruned_loss=0.1502, over 7541.00 frames. ], tot_loss[loss=0.3548, simple_loss=0.3954, pruned_loss=0.1571, over 1611802.68 frames. ], batch size: 18, lr: 2.68e-02, grad_scale: 8.0 +2023-02-05 21:02:15,682 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15953.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:02:36,233 INFO [train.py:901] (3/4) Epoch 2, batch 7900, loss[loss=0.3026, simple_loss=0.3485, pruned_loss=0.1283, over 7970.00 frames. ], tot_loss[loss=0.3529, simple_loss=0.3943, pruned_loss=0.1558, over 1609065.57 frames. ], batch size: 21, lr: 2.68e-02, grad_scale: 8.0 +2023-02-05 21:02:58,244 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.267e+02 3.808e+02 4.602e+02 5.936e+02 1.299e+03, threshold=9.205e+02, percent-clipped=9.0 +2023-02-05 21:03:00,388 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16019.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:03:10,220 INFO [train.py:901] (3/4) Epoch 2, batch 7950, loss[loss=0.4178, simple_loss=0.4429, pruned_loss=0.1963, over 8479.00 frames. ], tot_loss[loss=0.3553, simple_loss=0.3961, pruned_loss=0.1572, over 1610354.53 frames. ], batch size: 49, lr: 2.68e-02, grad_scale: 8.0 +2023-02-05 21:03:16,614 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.04 vs. limit=2.0 +2023-02-05 21:03:43,332 INFO [train.py:901] (3/4) Epoch 2, batch 8000, loss[loss=0.3309, simple_loss=0.3904, pruned_loss=0.1357, over 8693.00 frames. ], tot_loss[loss=0.3562, simple_loss=0.3969, pruned_loss=0.1577, over 1612715.58 frames. ], batch size: 34, lr: 2.67e-02, grad_scale: 8.0 +2023-02-05 21:03:55,455 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5003, 1.8401, 3.3949, 1.0492, 2.1326, 1.6158, 1.6440, 1.8365], + device='cuda:3'), covar=tensor([0.1147, 0.1414, 0.0424, 0.2091, 0.1226, 0.1859, 0.0906, 0.1718], + device='cuda:3'), in_proj_covar=tensor([0.0417, 0.0391, 0.0443, 0.0470, 0.0515, 0.0458, 0.0407, 0.0516], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 21:03:56,060 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16103.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:04:04,544 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.336e+02 4.123e+02 4.991e+02 6.647e+02 1.461e+03, threshold=9.983e+02, percent-clipped=10.0 +2023-02-05 21:04:16,511 INFO [train.py:901] (3/4) Epoch 2, batch 8050, loss[loss=0.3267, simple_loss=0.3606, pruned_loss=0.1464, over 7536.00 frames. ], tot_loss[loss=0.3563, simple_loss=0.3959, pruned_loss=0.1583, over 1601070.27 frames. ], batch size: 18, lr: 2.67e-02, grad_scale: 8.0 +2023-02-05 21:04:36,024 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5575, 2.1983, 3.2023, 2.2343, 2.7102, 3.7991, 3.3347, 3.3761], + device='cuda:3'), covar=tensor([0.0878, 0.1156, 0.0646, 0.1446, 0.0650, 0.0270, 0.0373, 0.0500], + device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0255, 0.0189, 0.0255, 0.0194, 0.0159, 0.0152, 0.0231], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 21:04:51,518 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-05 21:04:55,115 INFO [train.py:901] (3/4) Epoch 3, batch 0, loss[loss=0.4623, simple_loss=0.4701, pruned_loss=0.2272, over 8676.00 frames. ], tot_loss[loss=0.4623, simple_loss=0.4701, pruned_loss=0.2272, over 8676.00 frames. ], batch size: 39, lr: 2.53e-02, grad_scale: 8.0 +2023-02-05 21:04:55,115 INFO [train.py:926] (3/4) Computing validation loss +2023-02-05 21:05:06,956 INFO [train.py:935] (3/4) Epoch 3, validation: loss=0.2731, simple_loss=0.3579, pruned_loss=0.09417, over 944034.00 frames. +2023-02-05 21:05:06,957 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-05 21:05:07,097 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16167.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:05:21,209 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.73 vs. limit=5.0 +2023-02-05 21:05:23,569 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-05 21:05:42,762 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.402e+02 4.065e+02 5.070e+02 6.931e+02 1.670e+03, threshold=1.014e+03, percent-clipped=5.0 +2023-02-05 21:05:42,782 INFO [train.py:901] (3/4) Epoch 3, batch 50, loss[loss=0.3827, simple_loss=0.4245, pruned_loss=0.1705, over 8347.00 frames. ], tot_loss[loss=0.3602, simple_loss=0.3995, pruned_loss=0.1605, over 366981.32 frames. ], batch size: 26, lr: 2.53e-02, grad_scale: 4.0 +2023-02-05 21:05:58,800 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-05 21:06:02,997 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16245.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:06:18,208 INFO [train.py:901] (3/4) Epoch 3, batch 100, loss[loss=0.3649, simple_loss=0.4061, pruned_loss=0.1618, over 8495.00 frames. ], tot_loss[loss=0.3584, simple_loss=0.3996, pruned_loss=0.1586, over 649642.13 frames. ], batch size: 26, lr: 2.53e-02, grad_scale: 4.0 +2023-02-05 21:06:18,921 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-05 21:06:20,585 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4310, 1.8402, 3.1540, 0.9621, 2.2257, 1.6481, 1.4264, 1.8010], + device='cuda:3'), covar=tensor([0.1172, 0.1274, 0.0427, 0.2174, 0.1075, 0.1710, 0.1001, 0.1603], + device='cuda:3'), in_proj_covar=tensor([0.0419, 0.0388, 0.0443, 0.0472, 0.0514, 0.0453, 0.0409, 0.0511], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 21:06:53,426 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.291e+02 3.520e+02 4.471e+02 5.811e+02 1.196e+03, threshold=8.942e+02, percent-clipped=3.0 +2023-02-05 21:06:53,448 INFO [train.py:901] (3/4) Epoch 3, batch 150, loss[loss=0.3621, simple_loss=0.3972, pruned_loss=0.1635, over 7914.00 frames. ], tot_loss[loss=0.3523, simple_loss=0.3953, pruned_loss=0.1546, over 867161.29 frames. ], batch size: 20, lr: 2.52e-02, grad_scale: 4.0 +2023-02-05 21:07:22,028 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.80 vs. limit=5.0 +2023-02-05 21:07:23,173 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16360.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:07:24,940 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16363.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:07:27,429 INFO [train.py:901] (3/4) Epoch 3, batch 200, loss[loss=0.3136, simple_loss=0.3832, pruned_loss=0.1219, over 8328.00 frames. ], tot_loss[loss=0.3508, simple_loss=0.394, pruned_loss=0.1538, over 1034717.92 frames. ], batch size: 25, lr: 2.52e-02, grad_scale: 4.0 +2023-02-05 21:07:42,167 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16389.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:08:01,474 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.927e+02 3.609e+02 4.419e+02 5.456e+02 1.161e+03, threshold=8.837e+02, percent-clipped=3.0 +2023-02-05 21:08:01,494 INFO [train.py:901] (3/4) Epoch 3, batch 250, loss[loss=0.2685, simple_loss=0.3282, pruned_loss=0.1044, over 6418.00 frames. ], tot_loss[loss=0.351, simple_loss=0.394, pruned_loss=0.1541, over 1163478.97 frames. ], batch size: 14, lr: 2.51e-02, grad_scale: 4.0 +2023-02-05 21:08:06,810 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6923, 2.2514, 3.0058, 0.8120, 2.9013, 1.8438, 1.1938, 1.6918], + device='cuda:3'), covar=tensor([0.0165, 0.0072, 0.0091, 0.0195, 0.0146, 0.0240, 0.0270, 0.0121], + device='cuda:3'), in_proj_covar=tensor([0.0199, 0.0140, 0.0119, 0.0185, 0.0132, 0.0249, 0.0204, 0.0173], + device='cuda:3'), out_proj_covar=tensor([1.0786e-04, 7.4981e-05, 6.4362e-05, 9.7529e-05, 7.3759e-05, 1.4512e-04, + 1.1234e-04, 9.4053e-05], device='cuda:3') +2023-02-05 21:08:13,913 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-05 21:08:22,558 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-05 21:08:22,620 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16447.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:08:35,585 INFO [train.py:901] (3/4) Epoch 3, batch 300, loss[loss=0.3655, simple_loss=0.3886, pruned_loss=0.1713, over 8091.00 frames. ], tot_loss[loss=0.3494, simple_loss=0.3928, pruned_loss=0.153, over 1264457.38 frames. ], batch size: 21, lr: 2.51e-02, grad_scale: 4.0 +2023-02-05 21:08:43,562 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16478.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:09:05,162 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16511.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:09:09,096 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.982e+02 3.752e+02 4.774e+02 5.919e+02 1.248e+03, threshold=9.549e+02, percent-clipped=6.0 +2023-02-05 21:09:09,117 INFO [train.py:901] (3/4) Epoch 3, batch 350, loss[loss=0.3436, simple_loss=0.3945, pruned_loss=0.1463, over 8101.00 frames. ], tot_loss[loss=0.3507, simple_loss=0.3938, pruned_loss=0.1539, over 1344074.23 frames. ], batch size: 23, lr: 2.51e-02, grad_scale: 4.0 +2023-02-05 21:09:40,933 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16562.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:09:44,011 INFO [train.py:901] (3/4) Epoch 3, batch 400, loss[loss=0.3425, simple_loss=0.3943, pruned_loss=0.1453, over 8097.00 frames. ], tot_loss[loss=0.3494, simple_loss=0.3923, pruned_loss=0.1533, over 1402318.36 frames. ], batch size: 23, lr: 2.50e-02, grad_scale: 8.0 +2023-02-05 21:09:47,772 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2023-02-05 21:10:18,125 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16616.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:10:18,537 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.210e+02 3.588e+02 4.493e+02 6.059e+02 1.047e+03, threshold=8.987e+02, percent-clipped=2.0 +2023-02-05 21:10:18,558 INFO [train.py:901] (3/4) Epoch 3, batch 450, loss[loss=0.4003, simple_loss=0.4177, pruned_loss=0.1915, over 8242.00 frames. ], tot_loss[loss=0.3503, simple_loss=0.3927, pruned_loss=0.154, over 1449026.85 frames. ], batch size: 22, lr: 2.50e-02, grad_scale: 8.0 +2023-02-05 21:10:20,710 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3194, 2.1255, 1.3951, 1.9327, 1.6899, 1.1743, 1.4352, 1.9226], + device='cuda:3'), covar=tensor([0.0998, 0.0379, 0.0960, 0.0508, 0.0697, 0.1206, 0.0880, 0.0705], + device='cuda:3'), in_proj_covar=tensor([0.0367, 0.0244, 0.0339, 0.0307, 0.0346, 0.0312, 0.0350, 0.0319], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-05 21:10:24,822 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16626.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:10:35,607 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16641.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:10:53,050 INFO [train.py:901] (3/4) Epoch 3, batch 500, loss[loss=0.391, simple_loss=0.4296, pruned_loss=0.1762, over 8205.00 frames. ], tot_loss[loss=0.3505, simple_loss=0.3938, pruned_loss=0.1536, over 1491489.18 frames. ], batch size: 23, lr: 2.50e-02, grad_scale: 8.0 +2023-02-05 21:11:27,942 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.737e+02 3.547e+02 4.664e+02 6.145e+02 2.246e+03, threshold=9.327e+02, percent-clipped=7.0 +2023-02-05 21:11:27,962 INFO [train.py:901] (3/4) Epoch 3, batch 550, loss[loss=0.3594, simple_loss=0.3966, pruned_loss=0.1611, over 7962.00 frames. ], tot_loss[loss=0.3506, simple_loss=0.3938, pruned_loss=0.1538, over 1518685.59 frames. ], batch size: 21, lr: 2.49e-02, grad_scale: 8.0 +2023-02-05 21:11:38,639 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16733.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:11:39,423 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16734.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:11:56,557 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16759.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:12:01,683 INFO [train.py:901] (3/4) Epoch 3, batch 600, loss[loss=0.3805, simple_loss=0.4018, pruned_loss=0.1796, over 8608.00 frames. ], tot_loss[loss=0.3513, simple_loss=0.3939, pruned_loss=0.1544, over 1541933.41 frames. ], batch size: 34, lr: 2.49e-02, grad_scale: 8.0 +2023-02-05 21:12:16,339 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-05 21:12:20,848 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2023-02-05 21:12:36,656 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.448e+02 3.715e+02 4.834e+02 5.984e+02 1.404e+03, threshold=9.668e+02, percent-clipped=7.0 +2023-02-05 21:12:36,678 INFO [train.py:901] (3/4) Epoch 3, batch 650, loss[loss=0.3387, simple_loss=0.3657, pruned_loss=0.1559, over 7701.00 frames. ], tot_loss[loss=0.3517, simple_loss=0.3938, pruned_loss=0.1548, over 1555063.75 frames. ], batch size: 18, lr: 2.49e-02, grad_scale: 8.0 +2023-02-05 21:12:37,551 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16818.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:12:54,032 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16843.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:12:57,237 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16848.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:13:10,252 INFO [train.py:901] (3/4) Epoch 3, batch 700, loss[loss=0.3877, simple_loss=0.414, pruned_loss=0.1807, over 8107.00 frames. ], tot_loss[loss=0.3516, simple_loss=0.3936, pruned_loss=0.1548, over 1569845.88 frames. ], batch size: 23, lr: 2.48e-02, grad_scale: 8.0 +2023-02-05 21:13:17,870 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.25 vs. limit=5.0 +2023-02-05 21:13:20,420 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16882.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:13:23,049 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16886.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 21:13:38,452 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16907.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:13:44,822 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.086e+02 3.932e+02 4.613e+02 6.231e+02 2.383e+03, threshold=9.225e+02, percent-clipped=5.0 +2023-02-05 21:13:44,843 INFO [train.py:901] (3/4) Epoch 3, batch 750, loss[loss=0.4409, simple_loss=0.4608, pruned_loss=0.2105, over 8425.00 frames. ], tot_loss[loss=0.3527, simple_loss=0.3943, pruned_loss=0.1555, over 1581545.43 frames. ], batch size: 29, lr: 2.48e-02, grad_scale: 8.0 +2023-02-05 21:13:49,811 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16924.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:13:59,051 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-05 21:14:07,686 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-05 21:14:19,204 INFO [train.py:901] (3/4) Epoch 3, batch 800, loss[loss=0.3375, simple_loss=0.3901, pruned_loss=0.1424, over 8198.00 frames. ], tot_loss[loss=0.3504, simple_loss=0.3928, pruned_loss=0.154, over 1591679.59 frames. ], batch size: 23, lr: 2.48e-02, grad_scale: 8.0 +2023-02-05 21:14:26,118 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4934, 2.0660, 3.4701, 1.0306, 2.3578, 1.6471, 1.4664, 2.0224], + device='cuda:3'), covar=tensor([0.1043, 0.1225, 0.0310, 0.1984, 0.0982, 0.1568, 0.1050, 0.1471], + device='cuda:3'), in_proj_covar=tensor([0.0418, 0.0388, 0.0447, 0.0471, 0.0523, 0.0458, 0.0414, 0.0523], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 21:14:34,633 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4308, 2.2731, 1.3469, 1.7788, 1.8463, 1.1358, 1.7821, 2.0069], + device='cuda:3'), covar=tensor([0.1461, 0.0464, 0.1207, 0.0854, 0.0799, 0.1290, 0.1077, 0.0803], + device='cuda:3'), in_proj_covar=tensor([0.0366, 0.0246, 0.0339, 0.0307, 0.0346, 0.0309, 0.0353, 0.0319], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-05 21:14:53,625 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.211e+02 3.452e+02 4.368e+02 5.287e+02 1.393e+03, threshold=8.735e+02, percent-clipped=4.0 +2023-02-05 21:14:53,646 INFO [train.py:901] (3/4) Epoch 3, batch 850, loss[loss=0.4117, simple_loss=0.4484, pruned_loss=0.1875, over 8517.00 frames. ], tot_loss[loss=0.3485, simple_loss=0.3913, pruned_loss=0.1528, over 1597821.63 frames. ], batch size: 26, lr: 2.47e-02, grad_scale: 8.0 +2023-02-05 21:15:28,356 INFO [train.py:901] (3/4) Epoch 3, batch 900, loss[loss=0.3459, simple_loss=0.3814, pruned_loss=0.1552, over 7919.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.3897, pruned_loss=0.1512, over 1602496.95 frames. ], batch size: 20, lr: 2.47e-02, grad_scale: 8.0 +2023-02-05 21:15:53,799 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17104.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:16:02,283 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.375e+02 3.695e+02 4.540e+02 5.760e+02 9.795e+02, threshold=9.080e+02, percent-clipped=3.0 +2023-02-05 21:16:02,304 INFO [train.py:901] (3/4) Epoch 3, batch 950, loss[loss=0.3276, simple_loss=0.375, pruned_loss=0.1401, over 8578.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.388, pruned_loss=0.1492, over 1604596.47 frames. ], batch size: 34, lr: 2.47e-02, grad_scale: 8.0 +2023-02-05 21:16:10,477 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17129.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:16:25,711 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-05 21:16:36,866 INFO [train.py:901] (3/4) Epoch 3, batch 1000, loss[loss=0.3645, simple_loss=0.4193, pruned_loss=0.1549, over 8489.00 frames. ], tot_loss[loss=0.3477, simple_loss=0.3912, pruned_loss=0.1521, over 1609840.28 frames. ], batch size: 29, lr: 2.46e-02, grad_scale: 8.0 +2023-02-05 21:16:57,953 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-05 21:17:03,574 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17207.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:17:10,141 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.390e+02 4.093e+02 4.952e+02 6.088e+02 1.030e+03, threshold=9.904e+02, percent-clipped=7.0 +2023-02-05 21:17:10,162 INFO [train.py:901] (3/4) Epoch 3, batch 1050, loss[loss=0.3746, simple_loss=0.4036, pruned_loss=0.1728, over 7773.00 frames. ], tot_loss[loss=0.3467, simple_loss=0.3905, pruned_loss=0.1514, over 1610951.49 frames. ], batch size: 19, lr: 2.46e-02, grad_scale: 8.0 +2023-02-05 21:17:10,176 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-05 21:17:19,694 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17230.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 21:17:45,190 INFO [train.py:901] (3/4) Epoch 3, batch 1100, loss[loss=0.3268, simple_loss=0.369, pruned_loss=0.1423, over 8495.00 frames. ], tot_loss[loss=0.3472, simple_loss=0.39, pruned_loss=0.1522, over 1605283.03 frames. ], batch size: 50, lr: 2.46e-02, grad_scale: 8.0 +2023-02-05 21:17:45,923 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17268.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:18:05,709 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.78 vs. limit=2.0 +2023-02-05 21:18:19,098 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.272e+02 3.840e+02 4.434e+02 5.714e+02 1.415e+03, threshold=8.869e+02, percent-clipped=3.0 +2023-02-05 21:18:19,120 INFO [train.py:901] (3/4) Epoch 3, batch 1150, loss[loss=0.4054, simple_loss=0.4399, pruned_loss=0.1854, over 8462.00 frames. ], tot_loss[loss=0.3464, simple_loss=0.39, pruned_loss=0.1514, over 1612789.42 frames. ], batch size: 25, lr: 2.45e-02, grad_scale: 8.0 +2023-02-05 21:18:22,466 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-05 21:18:38,641 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17345.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 21:18:52,840 INFO [train.py:901] (3/4) Epoch 3, batch 1200, loss[loss=0.3198, simple_loss=0.365, pruned_loss=0.1373, over 7537.00 frames. ], tot_loss[loss=0.3465, simple_loss=0.3905, pruned_loss=0.1512, over 1614718.67 frames. ], batch size: 18, lr: 2.45e-02, grad_scale: 8.0 +2023-02-05 21:19:02,176 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17380.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:19:04,946 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17383.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:19:17,680 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17401.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 21:19:28,364 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.147e+02 3.772e+02 4.989e+02 5.905e+02 9.785e+02, threshold=9.978e+02, percent-clipped=4.0 +2023-02-05 21:19:28,385 INFO [train.py:901] (3/4) Epoch 3, batch 1250, loss[loss=0.2313, simple_loss=0.2962, pruned_loss=0.0832, over 7551.00 frames. ], tot_loss[loss=0.3478, simple_loss=0.3916, pruned_loss=0.152, over 1618189.30 frames. ], batch size: 18, lr: 2.45e-02, grad_scale: 8.0 +2023-02-05 21:19:57,608 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.29 vs. limit=5.0 +2023-02-05 21:20:02,604 INFO [train.py:901] (3/4) Epoch 3, batch 1300, loss[loss=0.2586, simple_loss=0.3406, pruned_loss=0.0883, over 8360.00 frames. ], tot_loss[loss=0.3465, simple_loss=0.3905, pruned_loss=0.1512, over 1616398.17 frames. ], batch size: 24, lr: 2.44e-02, grad_scale: 8.0 +2023-02-05 21:20:37,547 INFO [train.py:901] (3/4) Epoch 3, batch 1350, loss[loss=0.3485, simple_loss=0.4087, pruned_loss=0.1441, over 8437.00 frames. ], tot_loss[loss=0.3479, simple_loss=0.3908, pruned_loss=0.1525, over 1615639.97 frames. ], batch size: 29, lr: 2.44e-02, grad_scale: 4.0 +2023-02-05 21:20:38,227 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.198e+02 4.258e+02 5.812e+02 8.345e+02 8.746e+03, threshold=1.162e+03, percent-clipped=16.0 +2023-02-05 21:20:58,310 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1000, 1.2544, 4.3378, 1.8375, 3.7806, 3.5517, 3.7485, 3.8033], + device='cuda:3'), covar=tensor([0.0384, 0.3078, 0.0240, 0.1703, 0.0758, 0.0409, 0.0389, 0.0434], + device='cuda:3'), in_proj_covar=tensor([0.0228, 0.0400, 0.0275, 0.0317, 0.0366, 0.0293, 0.0288, 0.0310], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2023-02-05 21:21:00,246 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17551.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:21:11,045 INFO [train.py:901] (3/4) Epoch 3, batch 1400, loss[loss=0.3144, simple_loss=0.3471, pruned_loss=0.1409, over 7932.00 frames. ], tot_loss[loss=0.3466, simple_loss=0.3899, pruned_loss=0.1517, over 1614067.92 frames. ], batch size: 19, lr: 2.44e-02, grad_scale: 4.0 +2023-02-05 21:21:25,493 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 +2023-02-05 21:21:26,014 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7025, 2.2714, 3.5044, 2.9630, 2.7386, 2.1969, 1.5742, 1.6516], + device='cuda:3'), covar=tensor([0.0698, 0.0888, 0.0167, 0.0318, 0.0403, 0.0437, 0.0584, 0.0852], + device='cuda:3'), in_proj_covar=tensor([0.0556, 0.0479, 0.0377, 0.0422, 0.0528, 0.0450, 0.0469, 0.0479], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 21:21:34,759 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17601.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 21:21:46,906 INFO [train.py:901] (3/4) Epoch 3, batch 1450, loss[loss=0.3423, simple_loss=0.396, pruned_loss=0.1443, over 8465.00 frames. ], tot_loss[loss=0.3454, simple_loss=0.3895, pruned_loss=0.1507, over 1616374.06 frames. ], batch size: 25, lr: 2.43e-02, grad_scale: 4.0 +2023-02-05 21:21:47,591 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.138e+02 3.309e+02 4.161e+02 5.035e+02 1.114e+03, threshold=8.322e+02, percent-clipped=0.0 +2023-02-05 21:21:48,910 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-05 21:21:53,237 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17626.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 21:22:02,546 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17639.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:22:19,039 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17664.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:22:20,430 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17666.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:22:20,937 INFO [train.py:901] (3/4) Epoch 3, batch 1500, loss[loss=0.3356, simple_loss=0.3889, pruned_loss=0.1411, over 8103.00 frames. ], tot_loss[loss=0.3446, simple_loss=0.3884, pruned_loss=0.1504, over 1613098.47 frames. ], batch size: 23, lr: 2.43e-02, grad_scale: 4.0 +2023-02-05 21:22:56,186 INFO [train.py:901] (3/4) Epoch 3, batch 1550, loss[loss=0.3036, simple_loss=0.3768, pruned_loss=0.1152, over 8251.00 frames. ], tot_loss[loss=0.3477, simple_loss=0.3907, pruned_loss=0.1524, over 1616787.58 frames. ], batch size: 24, lr: 2.43e-02, grad_scale: 4.0 +2023-02-05 21:22:56,831 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.415e+02 3.678e+02 4.620e+02 5.892e+02 1.697e+03, threshold=9.239e+02, percent-clipped=9.0 +2023-02-05 21:22:57,030 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17718.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:23:01,086 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17724.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:23:04,116 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2023-02-05 21:23:06,486 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5213, 4.5722, 4.1489, 2.0410, 4.1144, 4.0086, 4.2824, 3.8463], + device='cuda:3'), covar=tensor([0.0835, 0.0447, 0.0781, 0.3967, 0.0473, 0.0475, 0.1003, 0.0518], + device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0233, 0.0282, 0.0370, 0.0254, 0.0207, 0.0266, 0.0195], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2023-02-05 21:23:16,061 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17745.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 21:23:17,958 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17748.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:23:30,933 INFO [train.py:901] (3/4) Epoch 3, batch 1600, loss[loss=0.3447, simple_loss=0.4029, pruned_loss=0.1432, over 8502.00 frames. ], tot_loss[loss=0.3467, simple_loss=0.3899, pruned_loss=0.1517, over 1617436.36 frames. ], batch size: 26, lr: 2.42e-02, grad_scale: 8.0 +2023-02-05 21:24:05,145 INFO [train.py:901] (3/4) Epoch 3, batch 1650, loss[loss=0.2999, simple_loss=0.3453, pruned_loss=0.1272, over 7672.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.3878, pruned_loss=0.15, over 1615013.89 frames. ], batch size: 18, lr: 2.42e-02, grad_scale: 8.0 +2023-02-05 21:24:05,807 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.224e+02 3.309e+02 4.132e+02 5.477e+02 8.650e+02, threshold=8.264e+02, percent-clipped=0.0 +2023-02-05 21:24:10,070 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-05 21:24:20,684 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17839.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:24:35,256 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17860.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 21:24:39,698 INFO [train.py:901] (3/4) Epoch 3, batch 1700, loss[loss=0.3297, simple_loss=0.3689, pruned_loss=0.1452, over 7983.00 frames. ], tot_loss[loss=0.3446, simple_loss=0.3884, pruned_loss=0.1504, over 1618320.68 frames. ], batch size: 21, lr: 2.42e-02, grad_scale: 8.0 +2023-02-05 21:25:05,671 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.72 vs. limit=2.0 +2023-02-05 21:25:06,194 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0535, 1.1034, 1.0685, 1.0503, 0.8188, 1.1423, 0.0481, 0.8444], + device='cuda:3'), covar=tensor([0.2045, 0.1483, 0.1136, 0.1332, 0.3077, 0.0977, 0.3862, 0.1661], + device='cuda:3'), in_proj_covar=tensor([0.0110, 0.0102, 0.0084, 0.0144, 0.0149, 0.0081, 0.0148, 0.0111], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2023-02-05 21:25:13,890 INFO [train.py:901] (3/4) Epoch 3, batch 1750, loss[loss=0.3461, simple_loss=0.3911, pruned_loss=0.1505, over 8148.00 frames. ], tot_loss[loss=0.344, simple_loss=0.388, pruned_loss=0.15, over 1618818.58 frames. ], batch size: 22, lr: 2.42e-02, grad_scale: 8.0 +2023-02-05 21:25:14,593 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.106e+02 3.998e+02 5.161e+02 6.686e+02 1.470e+03, threshold=1.032e+03, percent-clipped=12.0 +2023-02-05 21:25:17,622 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17922.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:25:35,812 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17947.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:25:48,896 INFO [train.py:901] (3/4) Epoch 3, batch 1800, loss[loss=0.3212, simple_loss=0.3739, pruned_loss=0.1342, over 8327.00 frames. ], tot_loss[loss=0.343, simple_loss=0.3874, pruned_loss=0.1493, over 1620415.90 frames. ], batch size: 26, lr: 2.41e-02, grad_scale: 8.0 +2023-02-05 21:25:51,116 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7910, 1.6130, 2.2216, 1.6778, 1.2937, 2.1098, 0.3564, 1.4151], + device='cuda:3'), covar=tensor([0.1745, 0.1936, 0.0861, 0.1804, 0.3240, 0.0721, 0.4211, 0.1610], + device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0099, 0.0082, 0.0142, 0.0148, 0.0078, 0.0142, 0.0109], + device='cuda:3'), out_proj_covar=tensor([1.2939e-04, 1.2438e-04, 1.0166e-04, 1.6370e-04, 1.7154e-04, 9.9048e-05, + 1.6789e-04, 1.3721e-04], device='cuda:3') +2023-02-05 21:25:57,023 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17978.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:26:25,060 INFO [train.py:901] (3/4) Epoch 3, batch 1850, loss[loss=0.2828, simple_loss=0.3382, pruned_loss=0.1137, over 7416.00 frames. ], tot_loss[loss=0.3422, simple_loss=0.3866, pruned_loss=0.1489, over 1617790.57 frames. ], batch size: 17, lr: 2.41e-02, grad_scale: 8.0 +2023-02-05 21:26:25,636 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.340e+02 3.564e+02 4.327e+02 5.819e+02 2.228e+03, threshold=8.654e+02, percent-clipped=8.0 +2023-02-05 21:26:55,949 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18062.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:26:59,907 INFO [train.py:901] (3/4) Epoch 3, batch 1900, loss[loss=0.3361, simple_loss=0.3955, pruned_loss=0.1383, over 8317.00 frames. ], tot_loss[loss=0.3401, simple_loss=0.3852, pruned_loss=0.1475, over 1614463.10 frames. ], batch size: 26, lr: 2.41e-02, grad_scale: 8.0 +2023-02-05 21:27:11,515 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0487, 1.6240, 3.4211, 1.3929, 2.0067, 3.8919, 3.2446, 3.3168], + device='cuda:3'), covar=tensor([0.0986, 0.1266, 0.0316, 0.1818, 0.0790, 0.0187, 0.0402, 0.0553], + device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0263, 0.0200, 0.0260, 0.0198, 0.0165, 0.0166, 0.0241], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 21:27:15,651 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.14 vs. limit=5.0 +2023-02-05 21:27:17,427 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18092.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:27:19,615 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18095.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:27:24,176 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-05 21:27:34,372 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18116.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 21:27:34,801 INFO [train.py:901] (3/4) Epoch 3, batch 1950, loss[loss=0.3344, simple_loss=0.3731, pruned_loss=0.1479, over 7535.00 frames. ], tot_loss[loss=0.3399, simple_loss=0.385, pruned_loss=0.1474, over 1612986.84 frames. ], batch size: 18, lr: 2.40e-02, grad_scale: 8.0 +2023-02-05 21:27:35,481 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.131e+02 3.385e+02 4.094e+02 5.586e+02 1.173e+03, threshold=8.188e+02, percent-clipped=3.0 +2023-02-05 21:27:36,202 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-05 21:27:37,029 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18120.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:27:51,206 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18141.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 21:27:55,034 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-05 21:28:09,114 INFO [train.py:901] (3/4) Epoch 3, batch 2000, loss[loss=0.3793, simple_loss=0.4077, pruned_loss=0.1755, over 7644.00 frames. ], tot_loss[loss=0.3395, simple_loss=0.385, pruned_loss=0.147, over 1614391.47 frames. ], batch size: 19, lr: 2.40e-02, grad_scale: 8.0 +2023-02-05 21:28:17,043 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18177.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:28:27,338 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18192.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:28:28,013 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18193.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:28:33,445 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1998, 1.5622, 1.2181, 1.6683, 1.3932, 1.0200, 1.0715, 1.3611], + device='cuda:3'), covar=tensor([0.0844, 0.0493, 0.0981, 0.0514, 0.0687, 0.1161, 0.0949, 0.0748], + device='cuda:3'), in_proj_covar=tensor([0.0365, 0.0255, 0.0344, 0.0312, 0.0352, 0.0314, 0.0356, 0.0319], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-05 21:28:38,141 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18207.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:28:44,798 INFO [train.py:901] (3/4) Epoch 3, batch 2050, loss[loss=0.3069, simple_loss=0.3587, pruned_loss=0.1276, over 8126.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3856, pruned_loss=0.1475, over 1617937.86 frames. ], batch size: 22, lr: 2.40e-02, grad_scale: 8.0 +2023-02-05 21:28:46,138 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.474e+02 3.817e+02 4.995e+02 6.129e+02 1.664e+03, threshold=9.991e+02, percent-clipped=7.0 +2023-02-05 21:28:51,908 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5284, 1.2229, 3.3130, 1.4883, 2.2430, 3.7804, 3.5099, 3.2934], + device='cuda:3'), covar=tensor([0.1228, 0.1621, 0.0327, 0.1900, 0.0703, 0.0189, 0.0299, 0.0489], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0259, 0.0199, 0.0261, 0.0197, 0.0165, 0.0165, 0.0242], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 21:29:19,882 INFO [train.py:901] (3/4) Epoch 3, batch 2100, loss[loss=0.3814, simple_loss=0.4298, pruned_loss=0.1665, over 8750.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3863, pruned_loss=0.1477, over 1621360.23 frames. ], batch size: 34, lr: 2.39e-02, grad_scale: 8.0 +2023-02-05 21:29:55,182 INFO [train.py:901] (3/4) Epoch 3, batch 2150, loss[loss=0.3659, simple_loss=0.4068, pruned_loss=0.1625, over 8524.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3855, pruned_loss=0.1476, over 1614361.56 frames. ], batch size: 28, lr: 2.39e-02, grad_scale: 8.0 +2023-02-05 21:29:55,874 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.297e+02 3.744e+02 4.718e+02 5.936e+02 1.452e+03, threshold=9.436e+02, percent-clipped=4.0 +2023-02-05 21:29:59,203 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18322.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:30:31,065 INFO [train.py:901] (3/4) Epoch 3, batch 2200, loss[loss=0.3513, simple_loss=0.3783, pruned_loss=0.1622, over 8239.00 frames. ], tot_loss[loss=0.3395, simple_loss=0.3847, pruned_loss=0.1471, over 1612591.54 frames. ], batch size: 22, lr: 2.39e-02, grad_scale: 8.0 +2023-02-05 21:30:35,446 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2023-02-05 21:31:05,526 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5117, 4.7156, 4.0783, 1.7268, 4.0376, 4.0291, 4.3262, 3.4611], + device='cuda:3'), covar=tensor([0.0922, 0.0436, 0.0724, 0.4373, 0.0547, 0.0646, 0.0790, 0.0679], + device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0238, 0.0276, 0.0362, 0.0251, 0.0206, 0.0257, 0.0191], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2023-02-05 21:31:06,830 INFO [train.py:901] (3/4) Epoch 3, batch 2250, loss[loss=0.3125, simple_loss=0.3612, pruned_loss=0.1319, over 7816.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3842, pruned_loss=0.1465, over 1613563.26 frames. ], batch size: 20, lr: 2.38e-02, grad_scale: 8.0 +2023-02-05 21:31:07,499 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.379e+02 3.424e+02 4.222e+02 5.561e+02 1.530e+03, threshold=8.445e+02, percent-clipped=2.0 +2023-02-05 21:31:18,203 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18433.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:31:21,475 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18437.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:31:36,363 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18458.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:31:39,815 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18463.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:31:42,285 INFO [train.py:901] (3/4) Epoch 3, batch 2300, loss[loss=0.331, simple_loss=0.3845, pruned_loss=0.1388, over 8230.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.3837, pruned_loss=0.1461, over 1615214.38 frames. ], batch size: 22, lr: 2.38e-02, grad_scale: 8.0 +2023-02-05 21:31:56,793 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18488.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:32:09,415 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-05 21:32:11,148 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18508.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:32:17,115 INFO [train.py:901] (3/4) Epoch 3, batch 2350, loss[loss=0.281, simple_loss=0.3377, pruned_loss=0.1122, over 7783.00 frames. ], tot_loss[loss=0.3373, simple_loss=0.3827, pruned_loss=0.1459, over 1613202.58 frames. ], batch size: 19, lr: 2.38e-02, grad_scale: 8.0 +2023-02-05 21:32:17,770 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.449e+02 3.759e+02 4.661e+02 5.652e+02 9.227e+02, threshold=9.323e+02, percent-clipped=1.0 +2023-02-05 21:32:23,353 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18526.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 21:32:30,480 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18536.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:32:31,169 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18537.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:32:32,660 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8471, 1.5081, 2.3905, 2.0661, 2.1404, 1.4863, 1.2056, 0.8669], + device='cuda:3'), covar=tensor([0.0785, 0.0841, 0.0200, 0.0301, 0.0301, 0.0420, 0.0555, 0.0725], + device='cuda:3'), in_proj_covar=tensor([0.0563, 0.0491, 0.0393, 0.0443, 0.0544, 0.0458, 0.0476, 0.0486], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 21:32:34,057 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.10 vs. limit=2.0 +2023-02-05 21:32:51,324 INFO [train.py:901] (3/4) Epoch 3, batch 2400, loss[loss=0.3062, simple_loss=0.3384, pruned_loss=0.137, over 7425.00 frames. ], tot_loss[loss=0.3395, simple_loss=0.3842, pruned_loss=0.1474, over 1614078.99 frames. ], batch size: 17, lr: 2.38e-02, grad_scale: 8.0 +2023-02-05 21:32:57,491 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-05 21:33:24,831 INFO [train.py:901] (3/4) Epoch 3, batch 2450, loss[loss=0.3211, simple_loss=0.3806, pruned_loss=0.1308, over 8558.00 frames. ], tot_loss[loss=0.3441, simple_loss=0.3878, pruned_loss=0.1502, over 1615960.89 frames. ], batch size: 31, lr: 2.37e-02, grad_scale: 8.0 +2023-02-05 21:33:25,548 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.037e+02 3.618e+02 4.763e+02 6.456e+02 1.024e+03, threshold=9.527e+02, percent-clipped=2.0 +2023-02-05 21:33:49,161 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18651.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:33:49,834 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18652.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:33:56,583 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4001, 2.1934, 3.1766, 2.8040, 2.4281, 1.8479, 1.2597, 1.3582], + device='cuda:3'), covar=tensor([0.0774, 0.0895, 0.0210, 0.0391, 0.0469, 0.0478, 0.0627, 0.0967], + device='cuda:3'), in_proj_covar=tensor([0.0565, 0.0489, 0.0393, 0.0449, 0.0544, 0.0464, 0.0483, 0.0487], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 21:33:59,429 INFO [train.py:901] (3/4) Epoch 3, batch 2500, loss[loss=0.336, simple_loss=0.4026, pruned_loss=0.1347, over 8361.00 frames. ], tot_loss[loss=0.343, simple_loss=0.3873, pruned_loss=0.1494, over 1615670.57 frames. ], batch size: 24, lr: 2.37e-02, grad_scale: 8.0 +2023-02-05 21:34:17,088 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7126, 2.4299, 1.8229, 2.0889, 1.9439, 1.3637, 1.7485, 2.2055], + device='cuda:3'), covar=tensor([0.1057, 0.0568, 0.0844, 0.0595, 0.0702, 0.1247, 0.0933, 0.0701], + device='cuda:3'), in_proj_covar=tensor([0.0360, 0.0246, 0.0336, 0.0308, 0.0331, 0.0311, 0.0350, 0.0309], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-05 21:34:17,684 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18692.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:34:18,344 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18693.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:34:22,212 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3443, 1.6274, 1.6176, 0.1975, 1.5606, 1.2178, 0.2486, 1.5902], + device='cuda:3'), covar=tensor([0.0110, 0.0067, 0.0080, 0.0156, 0.0090, 0.0249, 0.0205, 0.0057], + device='cuda:3'), in_proj_covar=tensor([0.0213, 0.0151, 0.0130, 0.0196, 0.0146, 0.0266, 0.0214, 0.0175], + device='cuda:3'), out_proj_covar=tensor([1.0783e-04, 7.5567e-05, 6.4230e-05, 9.5794e-05, 7.5737e-05, 1.4382e-04, + 1.1047e-04, 8.9179e-05], device='cuda:3') +2023-02-05 21:34:24,656 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.4677, 5.5791, 4.7266, 1.9710, 4.7662, 4.9826, 5.2199, 4.4546], + device='cuda:3'), covar=tensor([0.0751, 0.0434, 0.0870, 0.4714, 0.0551, 0.0636, 0.1040, 0.0749], + device='cuda:3'), in_proj_covar=tensor([0.0341, 0.0239, 0.0277, 0.0363, 0.0255, 0.0211, 0.0267, 0.0197], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2023-02-05 21:34:33,854 INFO [train.py:901] (3/4) Epoch 3, batch 2550, loss[loss=0.3316, simple_loss=0.3842, pruned_loss=0.1395, over 8566.00 frames. ], tot_loss[loss=0.3427, simple_loss=0.3869, pruned_loss=0.1493, over 1612793.82 frames. ], batch size: 31, lr: 2.37e-02, grad_scale: 8.0 +2023-02-05 21:34:34,506 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.095e+02 3.889e+02 4.529e+02 5.619e+02 1.309e+03, threshold=9.058e+02, percent-clipped=5.0 +2023-02-05 21:34:34,733 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18718.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:35:02,133 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4313, 1.8371, 2.8819, 1.0928, 2.1906, 1.7986, 1.4765, 1.8310], + device='cuda:3'), covar=tensor([0.1240, 0.1245, 0.0445, 0.2266, 0.0983, 0.1753, 0.1153, 0.1546], + device='cuda:3'), in_proj_covar=tensor([0.0440, 0.0410, 0.0486, 0.0494, 0.0552, 0.0495, 0.0441, 0.0546], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 21:35:08,026 INFO [train.py:901] (3/4) Epoch 3, batch 2600, loss[loss=0.3404, simple_loss=0.3957, pruned_loss=0.1426, over 8361.00 frames. ], tot_loss[loss=0.3421, simple_loss=0.3872, pruned_loss=0.1485, over 1613110.39 frames. ], batch size: 24, lr: 2.36e-02, grad_scale: 8.0 +2023-02-05 21:35:25,812 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.1431, 1.2297, 3.2031, 0.9143, 2.7604, 2.6781, 2.8627, 2.8294], + device='cuda:3'), covar=tensor([0.0396, 0.2716, 0.0495, 0.1985, 0.1219, 0.0589, 0.0484, 0.0616], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0419, 0.0279, 0.0322, 0.0388, 0.0303, 0.0295, 0.0325], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2023-02-05 21:35:30,914 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 +2023-02-05 21:35:44,454 INFO [train.py:901] (3/4) Epoch 3, batch 2650, loss[loss=0.2991, simple_loss=0.373, pruned_loss=0.1127, over 8139.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3868, pruned_loss=0.1471, over 1619291.35 frames. ], batch size: 22, lr: 2.36e-02, grad_scale: 8.0 +2023-02-05 21:35:45,136 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.448e+02 3.426e+02 4.272e+02 5.708e+02 1.020e+03, threshold=8.544e+02, percent-clipped=5.0 +2023-02-05 21:35:51,389 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3074, 4.5361, 3.8722, 1.6295, 3.8216, 3.8013, 4.1158, 3.3238], + device='cuda:3'), covar=tensor([0.0982, 0.0523, 0.0927, 0.4668, 0.0617, 0.0694, 0.1156, 0.0632], + device='cuda:3'), in_proj_covar=tensor([0.0326, 0.0231, 0.0267, 0.0351, 0.0249, 0.0204, 0.0256, 0.0188], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2023-02-05 21:36:08,387 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18852.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:36:19,357 INFO [train.py:901] (3/4) Epoch 3, batch 2700, loss[loss=0.301, simple_loss=0.3365, pruned_loss=0.1328, over 7281.00 frames. ], tot_loss[loss=0.3378, simple_loss=0.3845, pruned_loss=0.1455, over 1616754.98 frames. ], batch size: 16, lr: 2.36e-02, grad_scale: 8.0 +2023-02-05 21:36:21,520 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18870.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 21:36:47,165 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18907.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:36:47,855 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18908.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:36:53,563 INFO [train.py:901] (3/4) Epoch 3, batch 2750, loss[loss=0.3582, simple_loss=0.4151, pruned_loss=0.1506, over 8506.00 frames. ], tot_loss[loss=0.3372, simple_loss=0.3843, pruned_loss=0.145, over 1617830.81 frames. ], batch size: 28, lr: 2.36e-02, grad_scale: 8.0 +2023-02-05 21:36:54,225 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.973e+02 3.360e+02 4.052e+02 5.079e+02 9.265e+02, threshold=8.105e+02, percent-clipped=2.0 +2023-02-05 21:37:05,016 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18932.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:37:05,654 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18933.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:37:15,751 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18948.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:37:24,000 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.82 vs. limit=5.0 +2023-02-05 21:37:28,078 INFO [train.py:901] (3/4) Epoch 3, batch 2800, loss[loss=0.302, simple_loss=0.3409, pruned_loss=0.1315, over 7710.00 frames. ], tot_loss[loss=0.3372, simple_loss=0.3841, pruned_loss=0.1451, over 1614938.99 frames. ], batch size: 18, lr: 2.35e-02, grad_scale: 8.0 +2023-02-05 21:37:28,273 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18967.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:37:41,139 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18985.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 21:38:03,343 INFO [train.py:901] (3/4) Epoch 3, batch 2850, loss[loss=0.3411, simple_loss=0.3842, pruned_loss=0.149, over 8246.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3827, pruned_loss=0.1437, over 1612721.80 frames. ], batch size: 22, lr: 2.35e-02, grad_scale: 8.0 +2023-02-05 21:38:03,471 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0042, 1.6056, 3.2387, 1.2498, 2.1634, 3.6930, 3.5175, 2.9726], + device='cuda:3'), covar=tensor([0.1285, 0.1549, 0.0456, 0.2422, 0.0818, 0.0325, 0.0365, 0.0851], + device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0257, 0.0198, 0.0256, 0.0199, 0.0167, 0.0173, 0.0244], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 21:38:03,922 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.788e+02 3.511e+02 4.402e+02 5.555e+02 1.104e+03, threshold=8.804e+02, percent-clipped=5.0 +2023-02-05 21:38:15,999 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19036.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:38:37,625 INFO [train.py:901] (3/4) Epoch 3, batch 2900, loss[loss=0.3005, simple_loss=0.357, pruned_loss=0.122, over 8092.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3843, pruned_loss=0.1453, over 1617949.32 frames. ], batch size: 21, lr: 2.35e-02, grad_scale: 8.0 +2023-02-05 21:38:42,289 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2048, 1.3688, 2.3714, 0.9896, 1.7002, 1.4046, 1.2522, 1.4966], + device='cuda:3'), covar=tensor([0.1320, 0.1441, 0.0495, 0.2230, 0.1038, 0.1965, 0.1241, 0.1332], + device='cuda:3'), in_proj_covar=tensor([0.0436, 0.0406, 0.0477, 0.0486, 0.0534, 0.0481, 0.0427, 0.0536], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 21:39:02,483 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-05 21:39:11,756 INFO [train.py:901] (3/4) Epoch 3, batch 2950, loss[loss=0.3826, simple_loss=0.4126, pruned_loss=0.1763, over 8631.00 frames. ], tot_loss[loss=0.335, simple_loss=0.3826, pruned_loss=0.1436, over 1612074.65 frames. ], batch size: 34, lr: 2.34e-02, grad_scale: 8.0 +2023-02-05 21:39:12,407 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.179e+02 3.613e+02 4.498e+02 5.900e+02 1.326e+03, threshold=8.996e+02, percent-clipped=8.0 +2023-02-05 21:39:13,248 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1690, 1.4905, 1.4554, 1.2611, 0.8967, 1.4860, 0.1423, 0.8329], + device='cuda:3'), covar=tensor([0.2395, 0.1454, 0.1055, 0.1966, 0.3823, 0.0870, 0.4486, 0.2177], + device='cuda:3'), in_proj_covar=tensor([0.0110, 0.0098, 0.0079, 0.0145, 0.0161, 0.0077, 0.0148, 0.0107], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2023-02-05 21:39:32,496 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19147.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 21:39:33,207 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2807, 1.8745, 2.1034, 1.4912, 0.8945, 2.0661, 0.3669, 1.0529], + device='cuda:3'), covar=tensor([0.3705, 0.1570, 0.1198, 0.2839, 0.4776, 0.0818, 0.5863, 0.2288], + device='cuda:3'), in_proj_covar=tensor([0.0112, 0.0098, 0.0078, 0.0146, 0.0162, 0.0077, 0.0148, 0.0106], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2023-02-05 21:39:35,219 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19151.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:39:37,925 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1249, 1.1680, 4.2715, 1.6656, 3.6804, 3.6622, 3.9187, 3.8816], + device='cuda:3'), covar=tensor([0.0445, 0.3319, 0.0310, 0.1723, 0.0941, 0.0398, 0.0387, 0.0440], + device='cuda:3'), in_proj_covar=tensor([0.0225, 0.0409, 0.0276, 0.0316, 0.0377, 0.0301, 0.0294, 0.0323], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2023-02-05 21:39:45,139 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5768, 1.0595, 1.2353, 0.8990, 1.2758, 1.0148, 0.9873, 1.3341], + device='cuda:3'), covar=tensor([0.0867, 0.2075, 0.2786, 0.2131, 0.0820, 0.2421, 0.1099, 0.0820], + device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0220, 0.0259, 0.0222, 0.0187, 0.0221, 0.0181, 0.0186], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0006, 0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0005], + device='cuda:3') +2023-02-05 21:39:45,158 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1304, 1.4825, 1.1851, 1.9375, 0.7594, 0.9406, 1.3027, 1.5039], + device='cuda:3'), covar=tensor([0.1753, 0.1423, 0.2461, 0.0726, 0.2162, 0.3111, 0.1566, 0.1229], + device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0306, 0.0306, 0.0226, 0.0292, 0.0310, 0.0324, 0.0295], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2023-02-05 21:39:46,216 INFO [train.py:901] (3/4) Epoch 3, batch 3000, loss[loss=0.3362, simple_loss=0.3843, pruned_loss=0.144, over 7955.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3848, pruned_loss=0.1457, over 1611619.44 frames. ], batch size: 21, lr: 2.34e-02, grad_scale: 8.0 +2023-02-05 21:39:46,216 INFO [train.py:926] (3/4) Computing validation loss +2023-02-05 21:39:58,665 INFO [train.py:935] (3/4) Epoch 3, validation: loss=0.2584, simple_loss=0.3473, pruned_loss=0.08481, over 944034.00 frames. +2023-02-05 21:39:58,666 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-05 21:40:08,538 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4615, 1.8668, 3.5383, 1.0333, 2.4268, 1.8623, 1.4923, 1.9144], + device='cuda:3'), covar=tensor([0.1316, 0.1401, 0.0444, 0.2279, 0.1180, 0.1777, 0.1092, 0.1899], + device='cuda:3'), in_proj_covar=tensor([0.0433, 0.0406, 0.0466, 0.0482, 0.0526, 0.0475, 0.0424, 0.0530], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 21:40:33,776 INFO [train.py:901] (3/4) Epoch 3, batch 3050, loss[loss=0.3186, simple_loss=0.3695, pruned_loss=0.1338, over 8250.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.3837, pruned_loss=0.1453, over 1613801.15 frames. ], batch size: 24, lr: 2.34e-02, grad_scale: 8.0 +2023-02-05 21:40:34,454 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.016e+02 3.526e+02 4.458e+02 6.217e+02 1.354e+03, threshold=8.917e+02, percent-clipped=3.0 +2023-02-05 21:40:38,080 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19223.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:40:50,807 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19241.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 21:40:55,464 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19248.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:41:03,463 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3027, 1.9036, 3.2761, 2.8129, 2.5362, 1.8598, 1.3500, 1.3937], + device='cuda:3'), covar=tensor([0.0922, 0.1171, 0.0230, 0.0395, 0.0557, 0.0499, 0.0628, 0.1067], + device='cuda:3'), in_proj_covar=tensor([0.0573, 0.0494, 0.0399, 0.0448, 0.0558, 0.0466, 0.0488, 0.0488], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 21:41:07,579 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19266.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 21:41:08,006 INFO [train.py:901] (3/4) Epoch 3, batch 3100, loss[loss=0.3285, simple_loss=0.3598, pruned_loss=0.1487, over 7923.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.3836, pruned_loss=0.1445, over 1614487.06 frames. ], batch size: 20, lr: 2.34e-02, grad_scale: 8.0 +2023-02-05 21:41:26,069 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19292.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:41:43,825 INFO [train.py:901] (3/4) Epoch 3, batch 3150, loss[loss=0.3363, simple_loss=0.3801, pruned_loss=0.1462, over 8291.00 frames. ], tot_loss[loss=0.3354, simple_loss=0.3829, pruned_loss=0.144, over 1615755.77 frames. ], batch size: 23, lr: 2.33e-02, grad_scale: 8.0 +2023-02-05 21:41:44,465 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.556e+02 3.507e+02 4.387e+02 6.193e+02 1.521e+03, threshold=8.773e+02, percent-clipped=4.0 +2023-02-05 21:42:02,650 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6445, 1.5242, 3.2846, 1.1449, 2.2465, 3.4498, 3.3082, 2.9807], + device='cuda:3'), covar=tensor([0.1225, 0.1449, 0.0329, 0.1951, 0.0710, 0.0261, 0.0286, 0.0531], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0264, 0.0204, 0.0267, 0.0206, 0.0172, 0.0175, 0.0250], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 21:42:17,836 INFO [train.py:901] (3/4) Epoch 3, batch 3200, loss[loss=0.4073, simple_loss=0.4403, pruned_loss=0.1871, over 8538.00 frames. ], tot_loss[loss=0.335, simple_loss=0.3823, pruned_loss=0.1439, over 1611758.01 frames. ], batch size: 28, lr: 2.33e-02, grad_scale: 8.0 +2023-02-05 21:42:33,629 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2647, 1.5719, 1.4962, 1.2864, 1.7217, 1.3876, 1.5551, 1.8012], + device='cuda:3'), covar=tensor([0.0732, 0.1450, 0.1885, 0.1679, 0.0756, 0.1673, 0.0937, 0.0656], + device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0221, 0.0259, 0.0223, 0.0189, 0.0224, 0.0185, 0.0188], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0006, 0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0005], + device='cuda:3') +2023-02-05 21:42:45,872 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19407.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:42:45,914 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19407.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:42:49,227 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19412.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:42:53,531 INFO [train.py:901] (3/4) Epoch 3, batch 3250, loss[loss=0.3736, simple_loss=0.3979, pruned_loss=0.1746, over 7680.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3826, pruned_loss=0.1439, over 1609996.60 frames. ], batch size: 18, lr: 2.33e-02, grad_scale: 8.0 +2023-02-05 21:42:54,130 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.836e+02 3.440e+02 4.583e+02 5.736e+02 1.373e+03, threshold=9.167e+02, percent-clipped=8.0 +2023-02-05 21:43:03,731 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19432.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:43:13,590 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 +2023-02-05 21:43:26,831 INFO [train.py:901] (3/4) Epoch 3, batch 3300, loss[loss=0.4007, simple_loss=0.4259, pruned_loss=0.1878, over 7973.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3833, pruned_loss=0.1445, over 1610137.34 frames. ], batch size: 21, lr: 2.32e-02, grad_scale: 8.0 +2023-02-05 21:43:40,000 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 +2023-02-05 21:43:43,381 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19491.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 21:44:01,025 INFO [train.py:901] (3/4) Epoch 3, batch 3350, loss[loss=0.3184, simple_loss=0.3751, pruned_loss=0.1308, over 8038.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.3839, pruned_loss=0.1451, over 1608549.69 frames. ], batch size: 22, lr: 2.32e-02, grad_scale: 16.0 +2023-02-05 21:44:01,692 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.684e+02 3.690e+02 4.650e+02 5.581e+02 1.223e+03, threshold=9.300e+02, percent-clipped=5.0 +2023-02-05 21:44:26,566 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7518, 3.0926, 2.4874, 3.8040, 1.6025, 1.6520, 2.2503, 3.1852], + device='cuda:3'), covar=tensor([0.0968, 0.1511, 0.1644, 0.0472, 0.2555, 0.2696, 0.2413, 0.1116], + device='cuda:3'), in_proj_covar=tensor([0.0298, 0.0313, 0.0308, 0.0235, 0.0300, 0.0314, 0.0335, 0.0292], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2023-02-05 21:44:35,816 INFO [train.py:901] (3/4) Epoch 3, batch 3400, loss[loss=0.3343, simple_loss=0.3798, pruned_loss=0.1443, over 8142.00 frames. ], tot_loss[loss=0.3372, simple_loss=0.3842, pruned_loss=0.1451, over 1609988.01 frames. ], batch size: 22, lr: 2.32e-02, grad_scale: 16.0 +2023-02-05 21:44:55,847 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.1527, 2.2601, 4.0010, 3.5904, 3.0941, 2.1892, 1.4839, 1.8622], + device='cuda:3'), covar=tensor([0.0767, 0.1227, 0.0208, 0.0357, 0.0480, 0.0493, 0.0642, 0.1055], + device='cuda:3'), in_proj_covar=tensor([0.0589, 0.0509, 0.0408, 0.0464, 0.0563, 0.0479, 0.0496, 0.0490], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 21:45:02,612 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19606.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 21:45:09,804 INFO [train.py:901] (3/4) Epoch 3, batch 3450, loss[loss=0.35, simple_loss=0.391, pruned_loss=0.1545, over 8328.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.3844, pruned_loss=0.1457, over 1611291.90 frames. ], batch size: 25, lr: 2.32e-02, grad_scale: 16.0 +2023-02-05 21:45:10,436 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.543e+02 3.801e+02 4.733e+02 6.108e+02 1.526e+03, threshold=9.466e+02, percent-clipped=4.0 +2023-02-05 21:45:23,377 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19636.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:45:42,893 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19663.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:45:45,374 INFO [train.py:901] (3/4) Epoch 3, batch 3500, loss[loss=0.3426, simple_loss=0.3991, pruned_loss=0.143, over 7981.00 frames. ], tot_loss[loss=0.3367, simple_loss=0.3835, pruned_loss=0.145, over 1609257.38 frames. ], batch size: 21, lr: 2.31e-02, grad_scale: 16.0 +2023-02-05 21:45:57,878 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2023-02-05 21:45:58,025 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-05 21:45:59,533 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19688.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:46:19,291 INFO [train.py:901] (3/4) Epoch 3, batch 3550, loss[loss=0.2737, simple_loss=0.3247, pruned_loss=0.1113, over 7533.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3826, pruned_loss=0.1448, over 1605645.15 frames. ], batch size: 18, lr: 2.31e-02, grad_scale: 16.0 +2023-02-05 21:46:19,958 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.944e+02 3.514e+02 4.193e+02 5.166e+02 1.109e+03, threshold=8.387e+02, percent-clipped=2.0 +2023-02-05 21:46:34,071 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2023-02-05 21:46:40,048 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-05 21:46:46,367 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19756.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:46:54,373 INFO [train.py:901] (3/4) Epoch 3, batch 3600, loss[loss=0.339, simple_loss=0.3707, pruned_loss=0.1536, over 7656.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3832, pruned_loss=0.1446, over 1610286.31 frames. ], batch size: 19, lr: 2.31e-02, grad_scale: 16.0 +2023-02-05 21:47:11,110 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6087, 1.7662, 1.4246, 2.4268, 1.2206, 1.0979, 1.5332, 2.1183], + device='cuda:3'), covar=tensor([0.1392, 0.1416, 0.2142, 0.0623, 0.1811, 0.2949, 0.1881, 0.0956], + device='cuda:3'), in_proj_covar=tensor([0.0303, 0.0308, 0.0310, 0.0235, 0.0300, 0.0318, 0.0338, 0.0306], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2023-02-05 21:47:28,233 INFO [train.py:901] (3/4) Epoch 3, batch 3650, loss[loss=0.3587, simple_loss=0.402, pruned_loss=0.1577, over 8648.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3854, pruned_loss=0.1464, over 1619147.84 frames. ], batch size: 39, lr: 2.30e-02, grad_scale: 16.0 +2023-02-05 21:47:28,888 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.850e+02 3.610e+02 4.497e+02 5.952e+02 1.837e+03, threshold=8.994e+02, percent-clipped=7.0 +2023-02-05 21:47:58,248 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1552, 1.5897, 1.3868, 1.0544, 1.6305, 1.3507, 1.2853, 1.7546], + device='cuda:3'), covar=tensor([0.0717, 0.1385, 0.2034, 0.1756, 0.0722, 0.1749, 0.1009, 0.0634], + device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0225, 0.0260, 0.0227, 0.0189, 0.0226, 0.0187, 0.0188], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0005, 0.0005], + device='cuda:3') +2023-02-05 21:47:58,735 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-05 21:47:59,613 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19862.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 21:48:02,765 INFO [train.py:901] (3/4) Epoch 3, batch 3700, loss[loss=0.3392, simple_loss=0.3782, pruned_loss=0.1501, over 8092.00 frames. ], tot_loss[loss=0.3385, simple_loss=0.3855, pruned_loss=0.1457, over 1621368.68 frames. ], batch size: 21, lr: 2.30e-02, grad_scale: 16.0 +2023-02-05 21:48:05,649 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19871.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:48:17,450 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19887.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 21:48:37,848 INFO [train.py:901] (3/4) Epoch 3, batch 3750, loss[loss=0.3529, simple_loss=0.383, pruned_loss=0.1614, over 8236.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3849, pruned_loss=0.145, over 1621600.18 frames. ], batch size: 22, lr: 2.30e-02, grad_scale: 16.0 +2023-02-05 21:48:38,369 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.782e+02 3.342e+02 4.116e+02 5.480e+02 1.463e+03, threshold=8.233e+02, percent-clipped=1.0 +2023-02-05 21:49:08,130 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0015, 1.5396, 3.3139, 1.2175, 2.3725, 3.6059, 3.3480, 3.1478], + device='cuda:3'), covar=tensor([0.1033, 0.1311, 0.0326, 0.1815, 0.0617, 0.0207, 0.0296, 0.0511], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0264, 0.0204, 0.0262, 0.0209, 0.0174, 0.0176, 0.0248], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 21:49:10,146 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.9724, 1.0614, 4.1720, 1.5497, 3.5759, 3.5270, 3.7366, 3.6817], + device='cuda:3'), covar=tensor([0.0317, 0.3374, 0.0269, 0.1697, 0.0914, 0.0396, 0.0381, 0.0414], + device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0432, 0.0295, 0.0332, 0.0394, 0.0314, 0.0311, 0.0334], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2023-02-05 21:49:12,035 INFO [train.py:901] (3/4) Epoch 3, batch 3800, loss[loss=0.3344, simple_loss=0.3741, pruned_loss=0.1474, over 7815.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.3851, pruned_loss=0.1454, over 1614310.90 frames. ], batch size: 20, lr: 2.30e-02, grad_scale: 16.0 +2023-02-05 21:49:16,347 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3626, 1.7845, 2.0224, 1.7388, 0.8942, 2.0043, 0.3546, 1.1878], + device='cuda:3'), covar=tensor([0.3563, 0.1451, 0.1146, 0.1762, 0.5075, 0.0608, 0.5422, 0.2003], + device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0094, 0.0082, 0.0151, 0.0170, 0.0077, 0.0155, 0.0110], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2023-02-05 21:49:20,778 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19980.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:49:48,448 INFO [train.py:901] (3/4) Epoch 3, batch 3850, loss[loss=0.3466, simple_loss=0.39, pruned_loss=0.1516, over 8434.00 frames. ], tot_loss[loss=0.3367, simple_loss=0.384, pruned_loss=0.1447, over 1616934.52 frames. ], batch size: 27, lr: 2.29e-02, grad_scale: 16.0 +2023-02-05 21:49:49,086 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.854e+02 3.536e+02 4.444e+02 5.257e+02 1.055e+03, threshold=8.889e+02, percent-clipped=4.0 +2023-02-05 21:50:01,908 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-05 21:50:02,728 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20038.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:50:09,036 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 +2023-02-05 21:50:22,592 INFO [train.py:901] (3/4) Epoch 3, batch 3900, loss[loss=0.2593, simple_loss=0.3341, pruned_loss=0.09221, over 8320.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.3839, pruned_loss=0.1444, over 1617170.59 frames. ], batch size: 25, lr: 2.29e-02, grad_scale: 16.0 +2023-02-05 21:50:41,541 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20095.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:50:49,334 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 +2023-02-05 21:50:50,426 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20107.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:50:56,837 INFO [train.py:901] (3/4) Epoch 3, batch 3950, loss[loss=0.3293, simple_loss=0.3648, pruned_loss=0.1469, over 7201.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.3825, pruned_loss=0.1436, over 1618040.76 frames. ], batch size: 16, lr: 2.29e-02, grad_scale: 16.0 +2023-02-05 21:50:57,400 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.523e+02 3.492e+02 4.461e+02 6.032e+02 1.371e+03, threshold=8.922e+02, percent-clipped=4.0 +2023-02-05 21:51:05,051 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20127.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:51:06,979 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20130.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:51:21,477 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20152.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:51:31,203 INFO [train.py:901] (3/4) Epoch 3, batch 4000, loss[loss=0.3702, simple_loss=0.4172, pruned_loss=0.1616, over 8364.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3818, pruned_loss=0.1426, over 1619212.31 frames. ], batch size: 24, lr: 2.29e-02, grad_scale: 16.0 +2023-02-05 21:51:45,720 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.21 vs. limit=5.0 +2023-02-05 21:52:05,184 INFO [train.py:901] (3/4) Epoch 3, batch 4050, loss[loss=0.3402, simple_loss=0.384, pruned_loss=0.1482, over 8185.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3813, pruned_loss=0.1423, over 1614110.03 frames. ], batch size: 23, lr: 2.28e-02, grad_scale: 16.0 +2023-02-05 21:52:05,854 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.226e+02 3.505e+02 4.242e+02 5.307e+02 1.364e+03, threshold=8.485e+02, percent-clipped=4.0 +2023-02-05 21:52:09,351 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20222.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:52:40,358 INFO [train.py:901] (3/4) Epoch 3, batch 4100, loss[loss=0.3216, simple_loss=0.3838, pruned_loss=0.1297, over 8449.00 frames. ], tot_loss[loss=0.3347, simple_loss=0.3829, pruned_loss=0.1433, over 1614923.62 frames. ], batch size: 27, lr: 2.28e-02, grad_scale: 8.0 +2023-02-05 21:52:46,100 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0728, 1.7382, 2.8660, 2.3308, 2.4045, 1.8251, 1.2483, 1.0668], + device='cuda:3'), covar=tensor([0.1020, 0.1022, 0.0244, 0.0456, 0.0468, 0.0552, 0.0701, 0.1035], + device='cuda:3'), in_proj_covar=tensor([0.0586, 0.0504, 0.0420, 0.0459, 0.0560, 0.0470, 0.0491, 0.0490], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 21:53:14,414 INFO [train.py:901] (3/4) Epoch 3, batch 4150, loss[loss=0.377, simple_loss=0.4201, pruned_loss=0.1669, over 8496.00 frames. ], tot_loss[loss=0.3366, simple_loss=0.3843, pruned_loss=0.1444, over 1616354.08 frames. ], batch size: 26, lr: 2.28e-02, grad_scale: 8.0 +2023-02-05 21:53:15,797 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.167e+02 3.849e+02 4.660e+02 5.932e+02 1.097e+03, threshold=9.320e+02, percent-clipped=6.0 +2023-02-05 21:53:38,371 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20351.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:53:50,004 INFO [train.py:901] (3/4) Epoch 3, batch 4200, loss[loss=0.3311, simple_loss=0.3727, pruned_loss=0.1447, over 7964.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3839, pruned_loss=0.1444, over 1615240.94 frames. ], batch size: 21, lr: 2.27e-02, grad_scale: 8.0 +2023-02-05 21:53:55,450 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-05 21:53:56,308 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20376.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:54:00,121 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20382.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:54:01,092 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.04 vs. limit=2.0 +2023-02-05 21:54:07,227 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2023-02-05 21:54:16,288 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20406.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:54:16,809 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-05 21:54:24,058 INFO [train.py:901] (3/4) Epoch 3, batch 4250, loss[loss=0.3359, simple_loss=0.3932, pruned_loss=0.1394, over 8631.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.3855, pruned_loss=0.1462, over 1615355.25 frames. ], batch size: 34, lr: 2.27e-02, grad_scale: 8.0 +2023-02-05 21:54:25,368 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.360e+02 3.627e+02 5.036e+02 6.332e+02 1.636e+03, threshold=1.007e+03, percent-clipped=4.0 +2023-02-05 21:54:29,672 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1293, 1.2199, 1.1958, 0.2273, 1.1347, 0.9819, 0.0904, 1.2227], + device='cuda:3'), covar=tensor([0.0088, 0.0068, 0.0050, 0.0138, 0.0074, 0.0212, 0.0164, 0.0064], + device='cuda:3'), in_proj_covar=tensor([0.0227, 0.0164, 0.0133, 0.0212, 0.0162, 0.0279, 0.0216, 0.0189], + device='cuda:3'), out_proj_covar=tensor([1.0911e-04, 7.8531e-05, 6.2246e-05, 9.8894e-05, 7.8683e-05, 1.4315e-04, + 1.0563e-04, 9.0656e-05], device='cuda:3') +2023-02-05 21:54:37,033 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20436.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:54:47,727 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20451.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:54:50,595 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1977, 1.5614, 1.3738, 1.0924, 1.5972, 1.4884, 1.6396, 1.5802], + device='cuda:3'), covar=tensor([0.0704, 0.1372, 0.1985, 0.1774, 0.0751, 0.1587, 0.0920, 0.0661], + device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0221, 0.0257, 0.0223, 0.0187, 0.0222, 0.0182, 0.0184], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0005, 0.0005], + device='cuda:3') +2023-02-05 21:54:54,861 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-05 21:54:59,266 INFO [train.py:901] (3/4) Epoch 3, batch 4300, loss[loss=0.3065, simple_loss=0.3524, pruned_loss=0.1303, over 7812.00 frames. ], tot_loss[loss=0.3385, simple_loss=0.3854, pruned_loss=0.1459, over 1612509.80 frames. ], batch size: 20, lr: 2.27e-02, grad_scale: 8.0 +2023-02-05 21:55:04,855 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20474.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:55:20,175 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20497.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:55:33,534 INFO [train.py:901] (3/4) Epoch 3, batch 4350, loss[loss=0.328, simple_loss=0.3826, pruned_loss=0.1367, over 8500.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3825, pruned_loss=0.144, over 1608454.80 frames. ], batch size: 29, lr: 2.27e-02, grad_scale: 8.0 +2023-02-05 21:55:34,898 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.821e+02 3.452e+02 4.356e+02 5.638e+02 1.577e+03, threshold=8.711e+02, percent-clipped=2.0 +2023-02-05 21:55:46,525 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-05 21:55:50,697 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.91 vs. limit=2.0 +2023-02-05 21:56:06,883 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20566.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:56:06,987 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20566.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:56:07,507 INFO [train.py:901] (3/4) Epoch 3, batch 4400, loss[loss=0.3995, simple_loss=0.4271, pruned_loss=0.1859, over 8582.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3834, pruned_loss=0.1446, over 1611547.37 frames. ], batch size: 39, lr: 2.26e-02, grad_scale: 8.0 +2023-02-05 21:56:13,113 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3494, 1.7675, 1.9496, 1.4296, 1.0375, 1.8409, 0.2512, 1.2657], + device='cuda:3'), covar=tensor([0.3834, 0.1358, 0.1160, 0.2168, 0.4882, 0.0927, 0.5527, 0.2134], + device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0097, 0.0084, 0.0147, 0.0169, 0.0078, 0.0152, 0.0117], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 21:56:22,875 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.83 vs. limit=5.0 +2023-02-05 21:56:24,128 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20589.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:56:27,483 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-05 21:56:36,783 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20606.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:56:44,140 INFO [train.py:901] (3/4) Epoch 3, batch 4450, loss[loss=0.2858, simple_loss=0.3497, pruned_loss=0.1109, over 8130.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3809, pruned_loss=0.1431, over 1607386.81 frames. ], batch size: 22, lr: 2.26e-02, grad_scale: 8.0 +2023-02-05 21:56:45,430 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.923e+02 3.404e+02 4.420e+02 6.069e+02 1.310e+03, threshold=8.839e+02, percent-clipped=8.0 +2023-02-05 21:57:12,227 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-05 21:57:18,492 INFO [train.py:901] (3/4) Epoch 3, batch 4500, loss[loss=0.2909, simple_loss=0.3547, pruned_loss=0.1135, over 8083.00 frames. ], tot_loss[loss=0.3326, simple_loss=0.3809, pruned_loss=0.1421, over 1615496.52 frames. ], batch size: 21, lr: 2.26e-02, grad_scale: 8.0 +2023-02-05 21:57:20,853 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-05 21:57:27,445 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20681.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:57:38,117 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1112, 1.6426, 2.9312, 2.3598, 2.3290, 1.7711, 1.3362, 1.1184], + device='cuda:3'), covar=tensor([0.1087, 0.1254, 0.0229, 0.0487, 0.0475, 0.0604, 0.0707, 0.1075], + device='cuda:3'), in_proj_covar=tensor([0.0601, 0.0514, 0.0433, 0.0470, 0.0585, 0.0476, 0.0497, 0.0503], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 21:57:53,171 INFO [train.py:901] (3/4) Epoch 3, batch 4550, loss[loss=0.32, simple_loss=0.3845, pruned_loss=0.1277, over 8328.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3804, pruned_loss=0.1428, over 1614834.92 frames. ], batch size: 25, lr: 2.26e-02, grad_scale: 8.0 +2023-02-05 21:57:54,484 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.207e+02 3.483e+02 4.570e+02 6.300e+02 1.347e+03, threshold=9.139e+02, percent-clipped=2.0 +2023-02-05 21:58:14,829 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20750.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:58:17,074 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20753.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:58:23,773 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5766, 1.9742, 3.2012, 1.0360, 2.0110, 1.8176, 1.4859, 1.8611], + device='cuda:3'), covar=tensor([0.1217, 0.1233, 0.0426, 0.2435, 0.1184, 0.1828, 0.1147, 0.1639], + device='cuda:3'), in_proj_covar=tensor([0.0439, 0.0410, 0.0481, 0.0501, 0.0551, 0.0482, 0.0427, 0.0544], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 21:58:26,243 INFO [train.py:901] (3/4) Epoch 3, batch 4600, loss[loss=0.3588, simple_loss=0.4041, pruned_loss=0.1568, over 8506.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3796, pruned_loss=0.1426, over 1613379.74 frames. ], batch size: 26, lr: 2.25e-02, grad_scale: 8.0 +2023-02-05 21:58:27,117 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2487, 2.3881, 2.0896, 3.0105, 1.2342, 1.4414, 1.9930, 2.5429], + device='cuda:3'), covar=tensor([0.1019, 0.1387, 0.1439, 0.0551, 0.2105, 0.2507, 0.1900, 0.1066], + device='cuda:3'), in_proj_covar=tensor([0.0303, 0.0305, 0.0305, 0.0225, 0.0296, 0.0316, 0.0329, 0.0299], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2023-02-05 21:58:34,044 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.30 vs. limit=5.0 +2023-02-05 21:58:34,570 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20778.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:58:35,797 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20780.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:59:00,498 INFO [train.py:901] (3/4) Epoch 3, batch 4650, loss[loss=0.3013, simple_loss=0.348, pruned_loss=0.1273, over 7447.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3788, pruned_loss=0.1422, over 1613131.87 frames. ], batch size: 17, lr: 2.25e-02, grad_scale: 8.0 +2023-02-05 21:59:02,524 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.001e+02 3.299e+02 4.239e+02 5.426e+02 9.400e+02, threshold=8.478e+02, percent-clipped=1.0 +2023-02-05 21:59:04,822 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20822.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:59:21,386 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20845.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:59:22,744 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20847.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:59:34,649 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20865.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:59:35,793 INFO [train.py:901] (3/4) Epoch 3, batch 4700, loss[loss=0.3783, simple_loss=0.4179, pruned_loss=0.1694, over 8284.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3792, pruned_loss=0.1424, over 1615003.05 frames. ], batch size: 23, lr: 2.25e-02, grad_scale: 8.0 +2023-02-05 21:59:37,922 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20870.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 21:59:54,909 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20895.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:00:08,984 INFO [train.py:901] (3/4) Epoch 3, batch 4750, loss[loss=0.2904, simple_loss=0.3529, pruned_loss=0.1139, over 8613.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3806, pruned_loss=0.1434, over 1614203.94 frames. ], batch size: 34, lr: 2.25e-02, grad_scale: 8.0 +2023-02-05 22:00:10,302 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.464e+02 3.634e+02 4.432e+02 5.821e+02 1.296e+03, threshold=8.863e+02, percent-clipped=5.0 +2023-02-05 22:00:23,260 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20937.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:00:24,392 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-05 22:00:26,472 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-05 22:00:32,660 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20950.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:00:41,462 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20962.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:00:44,560 INFO [train.py:901] (3/4) Epoch 3, batch 4800, loss[loss=0.3329, simple_loss=0.3929, pruned_loss=0.1364, over 8183.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3797, pruned_loss=0.1419, over 1609418.37 frames. ], batch size: 23, lr: 2.24e-02, grad_scale: 8.0 +2023-02-05 22:01:18,124 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-05 22:01:18,803 INFO [train.py:901] (3/4) Epoch 3, batch 4850, loss[loss=0.3442, simple_loss=0.3877, pruned_loss=0.1503, over 8630.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3796, pruned_loss=0.1419, over 1612706.97 frames. ], batch size: 39, lr: 2.24e-02, grad_scale: 8.0 +2023-02-05 22:01:20,192 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.459e+02 3.687e+02 4.412e+02 5.668e+02 1.155e+03, threshold=8.825e+02, percent-clipped=6.0 +2023-02-05 22:01:52,009 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21065.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:01:53,057 INFO [train.py:901] (3/4) Epoch 3, batch 4900, loss[loss=0.3341, simple_loss=0.3813, pruned_loss=0.1434, over 8085.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.3799, pruned_loss=0.1422, over 1611873.81 frames. ], batch size: 21, lr: 2.24e-02, grad_scale: 8.0 +2023-02-05 22:01:55,912 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21070.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:02:21,872 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.4226, 1.6576, 5.3877, 2.3918, 4.8508, 4.7282, 5.0181, 5.1177], + device='cuda:3'), covar=tensor([0.0212, 0.2969, 0.0170, 0.1427, 0.0654, 0.0266, 0.0266, 0.0220], + device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0426, 0.0294, 0.0331, 0.0396, 0.0314, 0.0312, 0.0339], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2023-02-05 22:02:27,714 INFO [train.py:901] (3/4) Epoch 3, batch 4950, loss[loss=0.356, simple_loss=0.4062, pruned_loss=0.1529, over 8339.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3788, pruned_loss=0.1411, over 1613317.97 frames. ], batch size: 25, lr: 2.24e-02, grad_scale: 8.0 +2023-02-05 22:02:29,094 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.083e+02 3.569e+02 4.502e+02 6.229e+02 1.133e+03, threshold=9.004e+02, percent-clipped=2.0 +2023-02-05 22:02:30,675 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21121.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:02:41,334 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21136.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:02:48,071 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21146.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:02:51,481 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21151.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:03:01,832 INFO [train.py:901] (3/4) Epoch 3, batch 5000, loss[loss=0.3112, simple_loss=0.3653, pruned_loss=0.1285, over 8458.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3786, pruned_loss=0.1407, over 1616405.90 frames. ], batch size: 27, lr: 2.23e-02, grad_scale: 8.0 +2023-02-05 22:03:08,596 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.0793, 1.1665, 3.1872, 0.9591, 2.5629, 2.6709, 2.8366, 2.7933], + device='cuda:3'), covar=tensor([0.0451, 0.2905, 0.0465, 0.1934, 0.1366, 0.0593, 0.0517, 0.0633], + device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0424, 0.0296, 0.0331, 0.0400, 0.0318, 0.0311, 0.0343], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-05 22:03:08,669 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21176.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:03:37,221 INFO [train.py:901] (3/4) Epoch 3, batch 5050, loss[loss=0.3833, simple_loss=0.428, pruned_loss=0.1693, over 8104.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.3795, pruned_loss=0.1403, over 1619849.27 frames. ], batch size: 23, lr: 2.23e-02, grad_scale: 8.0 +2023-02-05 22:03:38,541 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.043e+02 3.325e+02 4.224e+02 5.254e+02 1.187e+03, threshold=8.447e+02, percent-clipped=3.0 +2023-02-05 22:03:57,071 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-05 22:04:11,239 INFO [train.py:901] (3/4) Epoch 3, batch 5100, loss[loss=0.3899, simple_loss=0.4108, pruned_loss=0.1845, over 6918.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3802, pruned_loss=0.1413, over 1618011.82 frames. ], batch size: 71, lr: 2.23e-02, grad_scale: 8.0 +2023-02-05 22:04:34,346 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0866, 1.6439, 1.9365, 1.3663, 1.0441, 1.7681, 0.1640, 0.9704], + device='cuda:3'), covar=tensor([0.2715, 0.2165, 0.0842, 0.1880, 0.4594, 0.0989, 0.5264, 0.2482], + device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0107, 0.0083, 0.0154, 0.0175, 0.0085, 0.0154, 0.0117], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 22:04:43,220 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7812, 1.5828, 1.5993, 1.1835, 1.6169, 1.5801, 2.0385, 1.8556], + device='cuda:3'), covar=tensor([0.0660, 0.1404, 0.2080, 0.1681, 0.0810, 0.1646, 0.0808, 0.0667], + device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0218, 0.0254, 0.0217, 0.0181, 0.0218, 0.0179, 0.0180], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0005, 0.0005], + device='cuda:3') +2023-02-05 22:04:46,363 INFO [train.py:901] (3/4) Epoch 3, batch 5150, loss[loss=0.3338, simple_loss=0.3875, pruned_loss=0.14, over 8131.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3791, pruned_loss=0.1402, over 1618590.84 frames. ], batch size: 22, lr: 2.23e-02, grad_scale: 8.0 +2023-02-05 22:04:47,675 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.142e+02 3.453e+02 4.061e+02 5.332e+02 1.278e+03, threshold=8.122e+02, percent-clipped=4.0 +2023-02-05 22:04:50,017 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21321.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:04:56,024 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21330.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:05:06,501 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21346.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:05:06,558 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21346.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:05:19,997 INFO [train.py:901] (3/4) Epoch 3, batch 5200, loss[loss=0.3587, simple_loss=0.3691, pruned_loss=0.1742, over 7561.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3816, pruned_loss=0.1423, over 1619068.38 frames. ], batch size: 18, lr: 2.22e-02, grad_scale: 8.0 +2023-02-05 22:05:52,850 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21414.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:05:54,862 INFO [train.py:901] (3/4) Epoch 3, batch 5250, loss[loss=0.2852, simple_loss=0.3412, pruned_loss=0.1146, over 7542.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3807, pruned_loss=0.1426, over 1615204.59 frames. ], batch size: 18, lr: 2.22e-02, grad_scale: 8.0 +2023-02-05 22:05:54,879 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-05 22:05:56,260 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.964e+02 3.353e+02 4.281e+02 5.765e+02 2.364e+03, threshold=8.563e+02, percent-clipped=11.0 +2023-02-05 22:06:30,401 INFO [train.py:901] (3/4) Epoch 3, batch 5300, loss[loss=0.2929, simple_loss=0.3495, pruned_loss=0.1182, over 8621.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3808, pruned_loss=0.1418, over 1617001.25 frames. ], batch size: 31, lr: 2.22e-02, grad_scale: 8.0 +2023-02-05 22:06:39,463 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21480.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:07:04,805 INFO [train.py:901] (3/4) Epoch 3, batch 5350, loss[loss=0.2994, simple_loss=0.3614, pruned_loss=0.1187, over 8738.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3805, pruned_loss=0.1413, over 1620322.44 frames. ], batch size: 34, lr: 2.22e-02, grad_scale: 8.0 +2023-02-05 22:07:06,084 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.137e+02 3.338e+02 4.128e+02 5.460e+02 1.129e+03, threshold=8.255e+02, percent-clipped=3.0 +2023-02-05 22:07:13,624 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21529.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:07:14,363 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5629, 2.0484, 3.3649, 0.9935, 2.5503, 1.8804, 1.7074, 1.9737], + device='cuda:3'), covar=tensor([0.1269, 0.1377, 0.0476, 0.2577, 0.1036, 0.2016, 0.1005, 0.1885], + device='cuda:3'), in_proj_covar=tensor([0.0455, 0.0419, 0.0493, 0.0512, 0.0566, 0.0499, 0.0435, 0.0569], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 22:07:40,167 INFO [train.py:901] (3/4) Epoch 3, batch 5400, loss[loss=0.2977, simple_loss=0.3533, pruned_loss=0.1211, over 7544.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3797, pruned_loss=0.1413, over 1616133.75 frames. ], batch size: 18, lr: 2.21e-02, grad_scale: 8.0 +2023-02-05 22:07:48,369 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8249, 2.3855, 1.7217, 2.7914, 1.4229, 1.2967, 1.7274, 2.2366], + device='cuda:3'), covar=tensor([0.1310, 0.1322, 0.1857, 0.0558, 0.1951, 0.2630, 0.2310, 0.1430], + device='cuda:3'), in_proj_covar=tensor([0.0305, 0.0307, 0.0303, 0.0229, 0.0292, 0.0313, 0.0331, 0.0305], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2023-02-05 22:07:59,355 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21595.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:08:12,102 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21613.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:08:14,736 INFO [train.py:901] (3/4) Epoch 3, batch 5450, loss[loss=0.3377, simple_loss=0.3779, pruned_loss=0.1487, over 7807.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3797, pruned_loss=0.1417, over 1615259.70 frames. ], batch size: 20, lr: 2.21e-02, grad_scale: 8.0 +2023-02-05 22:08:16,059 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.864e+02 3.746e+02 4.366e+02 5.874e+02 2.172e+03, threshold=8.732e+02, percent-clipped=6.0 +2023-02-05 22:08:24,289 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21631.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:08:31,561 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 +2023-02-05 22:08:36,096 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 +2023-02-05 22:08:41,803 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-05 22:08:49,803 INFO [train.py:901] (3/4) Epoch 3, batch 5500, loss[loss=0.3351, simple_loss=0.3833, pruned_loss=0.1434, over 8467.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3784, pruned_loss=0.1403, over 1620236.28 frames. ], batch size: 25, lr: 2.21e-02, grad_scale: 8.0 +2023-02-05 22:08:55,224 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21674.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:09:05,908 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21690.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:09:18,076 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2023-02-05 22:09:23,538 INFO [train.py:901] (3/4) Epoch 3, batch 5550, loss[loss=0.2978, simple_loss=0.3505, pruned_loss=0.1225, over 7535.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3783, pruned_loss=0.1404, over 1619966.69 frames. ], batch size: 18, lr: 2.21e-02, grad_scale: 8.0 +2023-02-05 22:09:24,903 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.889e+02 3.296e+02 4.063e+02 5.206e+02 8.291e+02, threshold=8.125e+02, percent-clipped=0.0 +2023-02-05 22:09:57,913 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3821, 1.1863, 4.4692, 1.7751, 3.8010, 3.6230, 3.9342, 3.9883], + device='cuda:3'), covar=tensor([0.0356, 0.3468, 0.0342, 0.1997, 0.1115, 0.0500, 0.0436, 0.0389], + device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0427, 0.0304, 0.0338, 0.0402, 0.0317, 0.0309, 0.0341], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2023-02-05 22:09:58,422 INFO [train.py:901] (3/4) Epoch 3, batch 5600, loss[loss=0.2849, simple_loss=0.355, pruned_loss=0.1074, over 8458.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3775, pruned_loss=0.1397, over 1619825.22 frames. ], batch size: 27, lr: 2.20e-02, grad_scale: 8.0 +2023-02-05 22:10:08,496 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21781.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:10:11,838 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21785.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:10:14,461 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21789.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:10:25,386 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21805.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:10:26,071 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7899, 1.6468, 2.5487, 1.4512, 2.0786, 2.7351, 2.4965, 2.4753], + device='cuda:3'), covar=tensor([0.0915, 0.1189, 0.0844, 0.1533, 0.0971, 0.0390, 0.0409, 0.0599], + device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0260, 0.0198, 0.0252, 0.0207, 0.0180, 0.0176, 0.0252], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 22:10:28,849 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21810.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:10:33,430 INFO [train.py:901] (3/4) Epoch 3, batch 5650, loss[loss=0.3659, simple_loss=0.4046, pruned_loss=0.1636, over 8189.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.378, pruned_loss=0.1401, over 1617597.22 frames. ], batch size: 23, lr: 2.20e-02, grad_scale: 8.0 +2023-02-05 22:10:34,799 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.984e+02 3.614e+02 4.526e+02 5.980e+02 8.654e+02, threshold=9.051e+02, percent-clipped=4.0 +2023-02-05 22:10:45,266 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-05 22:10:56,889 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21851.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:11:07,115 INFO [train.py:901] (3/4) Epoch 3, batch 5700, loss[loss=0.3968, simple_loss=0.421, pruned_loss=0.1863, over 8097.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3792, pruned_loss=0.1406, over 1620210.72 frames. ], batch size: 21, lr: 2.20e-02, grad_scale: 8.0 +2023-02-05 22:11:07,931 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21868.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:11:13,372 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21876.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:11:42,902 INFO [train.py:901] (3/4) Epoch 3, batch 5750, loss[loss=0.35, simple_loss=0.4042, pruned_loss=0.1479, over 8194.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.3792, pruned_loss=0.1405, over 1615049.18 frames. ], batch size: 23, lr: 2.20e-02, grad_scale: 8.0 +2023-02-05 22:11:44,221 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.339e+02 3.657e+02 4.422e+02 5.345e+02 1.248e+03, threshold=8.845e+02, percent-clipped=3.0 +2023-02-05 22:11:49,710 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-05 22:12:10,246 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21957.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:12:16,966 INFO [train.py:901] (3/4) Epoch 3, batch 5800, loss[loss=0.3733, simple_loss=0.4105, pruned_loss=0.168, over 8554.00 frames. ], tot_loss[loss=0.3292, simple_loss=0.3784, pruned_loss=0.14, over 1618816.52 frames. ], batch size: 39, lr: 2.19e-02, grad_scale: 8.0 +2023-02-05 22:12:22,433 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21975.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:12:30,996 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21988.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:12:49,718 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6965, 2.1256, 2.6414, 1.0761, 2.6198, 1.6783, 1.2501, 1.6886], + device='cuda:3'), covar=tensor([0.0195, 0.0076, 0.0134, 0.0205, 0.0156, 0.0252, 0.0258, 0.0135], + device='cuda:3'), in_proj_covar=tensor([0.0227, 0.0156, 0.0135, 0.0210, 0.0157, 0.0283, 0.0224, 0.0192], + device='cuda:3'), out_proj_covar=tensor([1.0566e-04, 7.0679e-05, 6.0602e-05, 9.4311e-05, 7.3969e-05, 1.4039e-04, + 1.0522e-04, 8.7381e-05], device='cuda:3') +2023-02-05 22:12:52,162 INFO [train.py:901] (3/4) Epoch 3, batch 5850, loss[loss=0.348, simple_loss=0.3982, pruned_loss=0.149, over 8190.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3793, pruned_loss=0.1406, over 1623462.18 frames. ], batch size: 23, lr: 2.19e-02, grad_scale: 8.0 +2023-02-05 22:12:53,398 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.077e+02 3.662e+02 4.461e+02 5.594e+02 1.608e+03, threshold=8.923e+02, percent-clipped=8.0 +2023-02-05 22:13:11,817 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22045.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:13:22,199 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22061.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:13:25,941 INFO [train.py:901] (3/4) Epoch 3, batch 5900, loss[loss=0.3449, simple_loss=0.3964, pruned_loss=0.1467, over 8515.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3775, pruned_loss=0.139, over 1620355.49 frames. ], batch size: 28, lr: 2.19e-02, grad_scale: 8.0 +2023-02-05 22:13:28,836 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22070.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:13:30,205 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22072.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:13:39,573 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22086.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:13:42,232 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22090.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:14:00,338 INFO [train.py:901] (3/4) Epoch 3, batch 5950, loss[loss=0.2957, simple_loss=0.3404, pruned_loss=0.1255, over 7716.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3768, pruned_loss=0.1383, over 1614438.04 frames. ], batch size: 18, lr: 2.19e-02, grad_scale: 8.0 +2023-02-05 22:14:02,399 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.045e+02 3.353e+02 4.485e+02 5.691e+02 1.558e+03, threshold=8.970e+02, percent-clipped=6.0 +2023-02-05 22:14:07,181 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22125.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:14:23,958 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22148.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:14:32,281 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7793, 1.3915, 3.0154, 1.2955, 2.2460, 3.3618, 3.1595, 2.9582], + device='cuda:3'), covar=tensor([0.1117, 0.1512, 0.0390, 0.2019, 0.0719, 0.0272, 0.0449, 0.0612], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0260, 0.0196, 0.0254, 0.0204, 0.0176, 0.0174, 0.0252], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 22:14:36,951 INFO [train.py:901] (3/4) Epoch 3, batch 6000, loss[loss=0.2568, simple_loss=0.3174, pruned_loss=0.09808, over 7219.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3767, pruned_loss=0.1386, over 1613496.06 frames. ], batch size: 16, lr: 2.19e-02, grad_scale: 8.0 +2023-02-05 22:14:36,951 INFO [train.py:926] (3/4) Computing validation loss +2023-02-05 22:14:49,934 INFO [train.py:935] (3/4) Epoch 3, validation: loss=0.2472, simple_loss=0.3383, pruned_loss=0.07805, over 944034.00 frames. +2023-02-05 22:14:49,935 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-05 22:15:04,703 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 +2023-02-05 22:15:08,338 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22194.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:15:21,666 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22212.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:15:25,116 INFO [train.py:901] (3/4) Epoch 3, batch 6050, loss[loss=0.2967, simple_loss=0.3508, pruned_loss=0.1212, over 7793.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.378, pruned_loss=0.1396, over 1613748.43 frames. ], batch size: 19, lr: 2.18e-02, grad_scale: 8.0 +2023-02-05 22:15:26,476 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.554e+02 3.417e+02 4.364e+02 5.364e+02 3.571e+03, threshold=8.727e+02, percent-clipped=6.0 +2023-02-05 22:15:33,747 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.95 vs. limit=5.0 +2023-02-05 22:15:36,164 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22233.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:15:40,929 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22240.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:15:51,061 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22255.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:15:52,562 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3826, 1.9239, 2.9561, 2.6930, 2.6557, 1.9472, 1.5792, 1.9016], + device='cuda:3'), covar=tensor([0.0706, 0.1059, 0.0199, 0.0348, 0.0367, 0.0490, 0.0580, 0.0733], + device='cuda:3'), in_proj_covar=tensor([0.0617, 0.0532, 0.0456, 0.0489, 0.0597, 0.0499, 0.0513, 0.0513], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 22:15:59,773 INFO [train.py:901] (3/4) Epoch 3, batch 6100, loss[loss=0.4139, simple_loss=0.4341, pruned_loss=0.1969, over 8612.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3792, pruned_loss=0.1407, over 1612601.15 frames. ], batch size: 39, lr: 2.18e-02, grad_scale: 16.0 +2023-02-05 22:16:18,445 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-05 22:16:35,129 INFO [train.py:901] (3/4) Epoch 3, batch 6150, loss[loss=0.3485, simple_loss=0.3991, pruned_loss=0.149, over 8577.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3792, pruned_loss=0.1406, over 1610050.85 frames. ], batch size: 39, lr: 2.18e-02, grad_scale: 16.0 +2023-02-05 22:16:36,459 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.469e+02 3.615e+02 4.380e+02 5.688e+02 1.525e+03, threshold=8.759e+02, percent-clipped=2.0 +2023-02-05 22:16:41,835 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22327.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:16:42,446 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4726, 1.1307, 1.4229, 0.9992, 1.2808, 1.3134, 1.1562, 1.3772], + device='cuda:3'), covar=tensor([0.0900, 0.1549, 0.2180, 0.1788, 0.0712, 0.1891, 0.0919, 0.0677], + device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0213, 0.0252, 0.0214, 0.0179, 0.0216, 0.0178, 0.0179], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-05 22:16:42,493 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22328.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:16:45,084 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22332.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:16:54,537 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22346.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:16:59,257 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22353.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:17:01,498 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.40 vs. limit=5.0 +2023-02-05 22:17:08,710 INFO [train.py:901] (3/4) Epoch 3, batch 6200, loss[loss=0.4141, simple_loss=0.4375, pruned_loss=0.1953, over 6945.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3789, pruned_loss=0.1404, over 1606131.21 frames. ], batch size: 71, lr: 2.18e-02, grad_scale: 16.0 +2023-02-05 22:17:11,704 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22371.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:17:34,523 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22403.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:17:44,418 INFO [train.py:901] (3/4) Epoch 3, batch 6250, loss[loss=0.3553, simple_loss=0.4062, pruned_loss=0.1522, over 8512.00 frames. ], tot_loss[loss=0.3297, simple_loss=0.3789, pruned_loss=0.1402, over 1609183.42 frames. ], batch size: 39, lr: 2.17e-02, grad_scale: 16.0 +2023-02-05 22:17:45,753 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.208e+02 3.506e+02 4.308e+02 5.585e+02 1.214e+03, threshold=8.617e+02, percent-clipped=6.0 +2023-02-05 22:18:03,163 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.78 vs. limit=2.0 +2023-02-05 22:18:05,783 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22447.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:18:15,774 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-05 22:18:19,135 INFO [train.py:901] (3/4) Epoch 3, batch 6300, loss[loss=0.3402, simple_loss=0.385, pruned_loss=0.1477, over 8032.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3793, pruned_loss=0.1406, over 1610844.01 frames. ], batch size: 22, lr: 2.17e-02, grad_scale: 16.0 +2023-02-05 22:18:36,439 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22492.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:18:39,310 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22496.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:18:46,910 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6252, 2.5453, 3.0787, 0.8136, 2.8378, 1.7662, 1.0794, 1.3603], + device='cuda:3'), covar=tensor([0.0218, 0.0080, 0.0089, 0.0255, 0.0180, 0.0288, 0.0328, 0.0157], + device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0167, 0.0145, 0.0220, 0.0165, 0.0299, 0.0231, 0.0195], + device='cuda:3'), out_proj_covar=tensor([1.0988e-04, 7.5261e-05, 6.5272e-05, 9.8293e-05, 7.7038e-05, 1.4669e-04, + 1.0724e-04, 8.6946e-05], device='cuda:3') +2023-02-05 22:18:54,100 INFO [train.py:901] (3/4) Epoch 3, batch 6350, loss[loss=0.3724, simple_loss=0.4029, pruned_loss=0.171, over 8495.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.3785, pruned_loss=0.1398, over 1615300.20 frames. ], batch size: 26, lr: 2.17e-02, grad_scale: 16.0 +2023-02-05 22:18:55,439 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.104e+02 3.537e+02 4.368e+02 5.315e+02 1.494e+03, threshold=8.736e+02, percent-clipped=5.0 +2023-02-05 22:18:57,022 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22521.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:19:03,400 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-05 22:19:08,840 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22538.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:19:28,747 INFO [train.py:901] (3/4) Epoch 3, batch 6400, loss[loss=0.2794, simple_loss=0.3437, pruned_loss=0.1076, over 8380.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3779, pruned_loss=0.1391, over 1613659.85 frames. ], batch size: 24, lr: 2.17e-02, grad_scale: 16.0 +2023-02-05 22:19:35,384 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22577.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:19:39,446 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22583.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:19:46,085 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4261, 1.5561, 4.2727, 1.9223, 2.2809, 4.9345, 4.4831, 4.3293], + device='cuda:3'), covar=tensor([0.1084, 0.1543, 0.0309, 0.1735, 0.0839, 0.0210, 0.0285, 0.0451], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0263, 0.0200, 0.0260, 0.0209, 0.0179, 0.0176, 0.0249], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 22:19:50,049 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22599.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:19:55,823 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22607.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:19:56,543 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22608.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:20:03,202 INFO [train.py:901] (3/4) Epoch 3, batch 6450, loss[loss=0.3301, simple_loss=0.3911, pruned_loss=0.1346, over 8369.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3779, pruned_loss=0.1397, over 1609692.51 frames. ], batch size: 24, lr: 2.16e-02, grad_scale: 16.0 +2023-02-05 22:20:04,480 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.771e+02 3.557e+02 4.436e+02 5.729e+02 1.082e+03, threshold=8.871e+02, percent-clipped=7.0 +2023-02-05 22:20:28,492 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22653.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:20:31,153 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22657.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:20:37,605 INFO [train.py:901] (3/4) Epoch 3, batch 6500, loss[loss=0.3181, simple_loss=0.3762, pruned_loss=0.13, over 8198.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3773, pruned_loss=0.1398, over 1610362.89 frames. ], batch size: 23, lr: 2.16e-02, grad_scale: 16.0 +2023-02-05 22:20:53,261 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3882, 3.4531, 2.8226, 4.1783, 1.7066, 2.0360, 2.7765, 3.6442], + device='cuda:3'), covar=tensor([0.1223, 0.1364, 0.1344, 0.0287, 0.2141, 0.2079, 0.1768, 0.0786], + device='cuda:3'), in_proj_covar=tensor([0.0305, 0.0305, 0.0304, 0.0225, 0.0287, 0.0310, 0.0319, 0.0293], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0005, 0.0005], + device='cuda:3') +2023-02-05 22:20:55,241 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22692.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:21:01,862 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.6103, 5.6154, 4.6998, 2.1989, 4.7694, 5.3736, 5.1201, 4.6040], + device='cuda:3'), covar=tensor([0.0713, 0.0448, 0.0886, 0.4244, 0.0653, 0.0527, 0.0981, 0.0586], + device='cuda:3'), in_proj_covar=tensor([0.0361, 0.0249, 0.0290, 0.0371, 0.0273, 0.0219, 0.0266, 0.0212], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2023-02-05 22:21:02,667 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22703.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:21:09,922 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22714.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:21:11,759 INFO [train.py:901] (3/4) Epoch 3, batch 6550, loss[loss=0.3553, simple_loss=0.3991, pruned_loss=0.1558, over 8508.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3766, pruned_loss=0.1389, over 1605316.14 frames. ], batch size: 28, lr: 2.16e-02, grad_scale: 16.0 +2023-02-05 22:21:13,165 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.155e+02 3.258e+02 3.883e+02 5.357e+02 1.264e+03, threshold=7.766e+02, percent-clipped=3.0 +2023-02-05 22:21:19,293 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22728.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:21:28,616 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-05 22:21:32,838 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22747.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:21:47,207 INFO [train.py:901] (3/4) Epoch 3, batch 6600, loss[loss=0.3344, simple_loss=0.3718, pruned_loss=0.1485, over 7541.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3751, pruned_loss=0.1382, over 1607993.65 frames. ], batch size: 18, lr: 2.16e-02, grad_scale: 8.0 +2023-02-05 22:21:47,903 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-05 22:22:19,036 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22812.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:22:22,348 INFO [train.py:901] (3/4) Epoch 3, batch 6650, loss[loss=0.2953, simple_loss=0.3534, pruned_loss=0.1186, over 8235.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3747, pruned_loss=0.1379, over 1605018.31 frames. ], batch size: 22, lr: 2.16e-02, grad_scale: 8.0 +2023-02-05 22:22:24,341 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.373e+02 3.456e+02 4.169e+02 5.335e+02 9.931e+02, threshold=8.339e+02, percent-clipped=8.0 +2023-02-05 22:22:40,063 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22843.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:22:53,737 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22862.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:22:54,488 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22863.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:22:56,605 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9015, 2.3683, 4.7020, 1.1962, 2.7848, 2.2725, 1.5661, 2.5061], + device='cuda:3'), covar=tensor([0.1225, 0.1718, 0.0432, 0.2533, 0.1279, 0.1850, 0.1369, 0.1916], + device='cuda:3'), in_proj_covar=tensor([0.0445, 0.0413, 0.0489, 0.0508, 0.0546, 0.0478, 0.0432, 0.0551], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 22:22:57,068 INFO [train.py:901] (3/4) Epoch 3, batch 6700, loss[loss=0.2747, simple_loss=0.3445, pruned_loss=0.1025, over 7974.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.374, pruned_loss=0.1373, over 1602728.43 frames. ], batch size: 21, lr: 2.15e-02, grad_scale: 8.0 +2023-02-05 22:23:12,641 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22888.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:23:26,913 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22909.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:23:32,715 INFO [train.py:901] (3/4) Epoch 3, batch 6750, loss[loss=0.3272, simple_loss=0.3739, pruned_loss=0.1402, over 8619.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.376, pruned_loss=0.1385, over 1606552.00 frames. ], batch size: 39, lr: 2.15e-02, grad_scale: 8.0 +2023-02-05 22:23:34,747 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.971e+02 3.597e+02 4.402e+02 5.483e+02 1.400e+03, threshold=8.804e+02, percent-clipped=7.0 +2023-02-05 22:23:41,011 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.8235, 5.8090, 4.7877, 1.6141, 5.0860, 5.3552, 5.1563, 4.7599], + device='cuda:3'), covar=tensor([0.0538, 0.0513, 0.0980, 0.5298, 0.0589, 0.0511, 0.1389, 0.0500], + device='cuda:3'), in_proj_covar=tensor([0.0356, 0.0244, 0.0283, 0.0369, 0.0267, 0.0223, 0.0263, 0.0211], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2023-02-05 22:23:44,527 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22934.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:23:53,621 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22948.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:24:05,886 INFO [train.py:901] (3/4) Epoch 3, batch 6800, loss[loss=0.2653, simple_loss=0.3312, pruned_loss=0.09975, over 8031.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3764, pruned_loss=0.1386, over 1610385.94 frames. ], batch size: 22, lr: 2.15e-02, grad_scale: 8.0 +2023-02-05 22:24:05,921 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-05 22:24:08,778 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22970.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:24:10,831 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22973.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:24:26,162 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22995.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:24:30,241 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23001.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:24:33,156 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.20 vs. limit=5.0 +2023-02-05 22:24:41,423 INFO [train.py:901] (3/4) Epoch 3, batch 6850, loss[loss=0.3418, simple_loss=0.3694, pruned_loss=0.1571, over 7413.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3758, pruned_loss=0.1377, over 1612655.17 frames. ], batch size: 17, lr: 2.15e-02, grad_scale: 8.0 +2023-02-05 22:24:43,431 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.139e+02 3.425e+02 4.505e+02 5.413e+02 1.323e+03, threshold=9.011e+02, percent-clipped=6.0 +2023-02-05 22:24:54,851 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-05 22:25:15,250 INFO [train.py:901] (3/4) Epoch 3, batch 6900, loss[loss=0.342, simple_loss=0.3973, pruned_loss=0.1433, over 8501.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3767, pruned_loss=0.1388, over 1613106.42 frames. ], batch size: 26, lr: 2.14e-02, grad_scale: 8.0 +2023-02-05 22:25:23,733 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5512, 2.3037, 3.5696, 3.0970, 2.9244, 2.2078, 1.6524, 2.1187], + device='cuda:3'), covar=tensor([0.0895, 0.1243, 0.0248, 0.0421, 0.0547, 0.0573, 0.0658, 0.1094], + device='cuda:3'), in_proj_covar=tensor([0.0621, 0.0541, 0.0454, 0.0503, 0.0617, 0.0511, 0.0515, 0.0527], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 22:25:50,201 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23116.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:25:50,685 INFO [train.py:901] (3/4) Epoch 3, batch 6950, loss[loss=0.327, simple_loss=0.3761, pruned_loss=0.139, over 8346.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3774, pruned_loss=0.1389, over 1613123.16 frames. ], batch size: 24, lr: 2.14e-02, grad_scale: 8.0 +2023-02-05 22:25:51,625 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23118.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:25:52,723 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.940e+02 3.525e+02 4.440e+02 6.025e+02 1.140e+03, threshold=8.880e+02, percent-clipped=3.0 +2023-02-05 22:25:57,668 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23126.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:26:02,129 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-05 22:26:09,874 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23143.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:26:18,630 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23156.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:26:25,707 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5409, 2.1097, 2.1297, 1.0675, 2.0761, 1.5968, 0.9587, 1.8596], + device='cuda:3'), covar=tensor([0.0150, 0.0077, 0.0070, 0.0134, 0.0102, 0.0186, 0.0178, 0.0075], + device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0172, 0.0146, 0.0223, 0.0172, 0.0289, 0.0235, 0.0197], + device='cuda:3'), out_proj_covar=tensor([1.1141e-04, 7.6856e-05, 6.4808e-05, 9.7729e-05, 7.9628e-05, 1.3963e-04, + 1.0637e-04, 8.6732e-05], device='cuda:3') +2023-02-05 22:26:26,189 INFO [train.py:901] (3/4) Epoch 3, batch 7000, loss[loss=0.3439, simple_loss=0.3878, pruned_loss=0.15, over 8286.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3768, pruned_loss=0.1379, over 1618048.23 frames. ], batch size: 49, lr: 2.14e-02, grad_scale: 8.0 +2023-02-05 22:26:30,358 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4059, 1.8626, 3.3869, 2.6512, 2.5961, 1.8413, 1.3622, 1.1545], + device='cuda:3'), covar=tensor([0.1136, 0.1456, 0.0251, 0.0539, 0.0576, 0.0652, 0.0749, 0.1408], + device='cuda:3'), in_proj_covar=tensor([0.0627, 0.0540, 0.0460, 0.0509, 0.0618, 0.0508, 0.0518, 0.0528], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 22:26:39,926 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23187.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:26:49,017 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3901, 4.3162, 3.7907, 1.8798, 3.8204, 3.5648, 3.8822, 3.2926], + device='cuda:3'), covar=tensor([0.0724, 0.0596, 0.0833, 0.4473, 0.0666, 0.0863, 0.1370, 0.0722], + device='cuda:3'), in_proj_covar=tensor([0.0357, 0.0245, 0.0281, 0.0365, 0.0268, 0.0223, 0.0271, 0.0210], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], + device='cuda:3') +2023-02-05 22:27:01,200 INFO [train.py:901] (3/4) Epoch 3, batch 7050, loss[loss=0.4127, simple_loss=0.4329, pruned_loss=0.1962, over 8412.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.378, pruned_loss=0.1387, over 1618011.57 frames. ], batch size: 48, lr: 2.14e-02, grad_scale: 8.0 +2023-02-05 22:27:03,862 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.823e+02 3.682e+02 4.488e+02 5.424e+02 1.788e+03, threshold=8.977e+02, percent-clipped=6.0 +2023-02-05 22:27:14,111 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23235.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:27:22,746 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23247.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:27:36,330 INFO [train.py:901] (3/4) Epoch 3, batch 7100, loss[loss=0.3715, simple_loss=0.417, pruned_loss=0.163, over 8466.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.377, pruned_loss=0.1382, over 1616566.53 frames. ], batch size: 25, lr: 2.14e-02, grad_scale: 8.0 +2023-02-05 22:27:39,097 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23271.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:27:59,147 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23302.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:28:08,841 INFO [train.py:901] (3/4) Epoch 3, batch 7150, loss[loss=0.3016, simple_loss=0.3537, pruned_loss=0.1247, over 7687.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3762, pruned_loss=0.1379, over 1612940.15 frames. ], batch size: 18, lr: 2.13e-02, grad_scale: 8.0 +2023-02-05 22:28:10,863 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.197e+02 3.845e+02 4.572e+02 5.960e+02 1.048e+03, threshold=9.143e+02, percent-clipped=2.0 +2023-02-05 22:28:42,780 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7886, 1.7294, 3.3051, 1.3816, 2.3259, 3.6709, 3.4558, 3.2261], + device='cuda:3'), covar=tensor([0.1181, 0.1356, 0.0417, 0.1926, 0.0709, 0.0241, 0.0428, 0.0602], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0262, 0.0207, 0.0263, 0.0205, 0.0182, 0.0184, 0.0254], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 22:28:43,301 INFO [train.py:901] (3/4) Epoch 3, batch 7200, loss[loss=0.3768, simple_loss=0.4199, pruned_loss=0.1668, over 8621.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3771, pruned_loss=0.1385, over 1615691.95 frames. ], batch size: 39, lr: 2.13e-02, grad_scale: 8.0 +2023-02-05 22:28:47,582 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23372.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:29:04,750 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23397.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:29:17,778 INFO [train.py:901] (3/4) Epoch 3, batch 7250, loss[loss=0.3567, simple_loss=0.3937, pruned_loss=0.1599, over 8182.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3765, pruned_loss=0.1389, over 1611090.33 frames. ], batch size: 23, lr: 2.13e-02, grad_scale: 4.0 +2023-02-05 22:29:20,307 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.518e+02 3.505e+02 4.323e+02 5.847e+02 9.851e+02, threshold=8.646e+02, percent-clipped=2.0 +2023-02-05 22:29:52,881 INFO [train.py:901] (3/4) Epoch 3, batch 7300, loss[loss=0.2713, simple_loss=0.3304, pruned_loss=0.1061, over 7711.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3766, pruned_loss=0.1385, over 1612736.36 frames. ], batch size: 18, lr: 2.13e-02, grad_scale: 4.0 +2023-02-05 22:29:55,076 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23470.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:30:21,341 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23506.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:30:28,665 INFO [train.py:901] (3/4) Epoch 3, batch 7350, loss[loss=0.2854, simple_loss=0.3407, pruned_loss=0.115, over 7937.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3739, pruned_loss=0.1374, over 1601610.26 frames. ], batch size: 20, lr: 2.12e-02, grad_scale: 4.0 +2023-02-05 22:30:31,437 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.204e+02 3.295e+02 4.174e+02 5.897e+02 1.266e+03, threshold=8.348e+02, percent-clipped=6.0 +2023-02-05 22:30:35,724 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23527.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:30:45,653 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-05 22:30:52,328 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23552.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:30:56,357 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23558.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:31:03,074 INFO [train.py:901] (3/4) Epoch 3, batch 7400, loss[loss=0.306, simple_loss=0.3614, pruned_loss=0.1253, over 7819.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3753, pruned_loss=0.1376, over 1607094.73 frames. ], batch size: 20, lr: 2.12e-02, grad_scale: 4.0 +2023-02-05 22:31:05,761 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-05 22:31:11,873 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23579.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:31:14,787 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23583.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:31:16,181 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23585.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:31:20,317 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23591.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:31:38,774 INFO [train.py:901] (3/4) Epoch 3, batch 7450, loss[loss=0.3666, simple_loss=0.4082, pruned_loss=0.1625, over 8461.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3773, pruned_loss=0.1392, over 1606658.81 frames. ], batch size: 27, lr: 2.12e-02, grad_scale: 4.0 +2023-02-05 22:31:41,484 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.151e+02 3.560e+02 4.542e+02 5.434e+02 8.209e+02, threshold=9.083e+02, percent-clipped=0.0 +2023-02-05 22:31:44,194 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-05 22:32:11,864 INFO [train.py:901] (3/4) Epoch 3, batch 7500, loss[loss=0.2788, simple_loss=0.3368, pruned_loss=0.1104, over 8032.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3762, pruned_loss=0.1385, over 1608500.87 frames. ], batch size: 22, lr: 2.12e-02, grad_scale: 4.0 +2023-02-05 22:32:23,285 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0186, 1.0813, 1.0904, 1.0381, 0.7519, 1.1987, 0.0364, 1.0288], + device='cuda:3'), covar=tensor([0.3028, 0.2022, 0.1143, 0.1758, 0.5069, 0.0961, 0.4986, 0.1729], + device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0115, 0.0084, 0.0159, 0.0187, 0.0082, 0.0147, 0.0119], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 22:32:31,249 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23694.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:32:39,266 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23706.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:32:47,010 INFO [train.py:901] (3/4) Epoch 3, batch 7550, loss[loss=0.3611, simple_loss=0.4091, pruned_loss=0.1565, over 8197.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.376, pruned_loss=0.1383, over 1610537.82 frames. ], batch size: 23, lr: 2.12e-02, grad_scale: 4.0 +2023-02-05 22:32:49,788 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.055e+02 3.573e+02 4.120e+02 5.568e+02 9.909e+02, threshold=8.240e+02, percent-clipped=1.0 +2023-02-05 22:33:21,017 INFO [train.py:901] (3/4) Epoch 3, batch 7600, loss[loss=0.3534, simple_loss=0.4114, pruned_loss=0.1477, over 8470.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3764, pruned_loss=0.138, over 1615431.13 frames. ], batch size: 29, lr: 2.11e-02, grad_scale: 8.0 +2023-02-05 22:33:55,888 INFO [train.py:901] (3/4) Epoch 3, batch 7650, loss[loss=0.349, simple_loss=0.378, pruned_loss=0.16, over 7442.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3756, pruned_loss=0.1379, over 1610140.66 frames. ], batch size: 17, lr: 2.11e-02, grad_scale: 8.0 +2023-02-05 22:33:58,392 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.349e+02 3.333e+02 4.379e+02 5.791e+02 1.321e+03, threshold=8.759e+02, percent-clipped=7.0 +2023-02-05 22:34:12,405 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-05 22:34:13,551 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23841.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:34:19,394 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23850.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:34:29,028 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2023-02-05 22:34:30,200 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23866.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:34:30,653 INFO [train.py:901] (3/4) Epoch 3, batch 7700, loss[loss=0.3209, simple_loss=0.3763, pruned_loss=0.1327, over 8501.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3755, pruned_loss=0.1379, over 1613380.31 frames. ], batch size: 28, lr: 2.11e-02, grad_scale: 8.0 +2023-02-05 22:34:50,583 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2023-02-05 22:34:50,862 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-05 22:35:04,780 INFO [train.py:901] (3/4) Epoch 3, batch 7750, loss[loss=0.285, simple_loss=0.3591, pruned_loss=0.1054, over 8352.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3755, pruned_loss=0.1371, over 1614009.70 frames. ], batch size: 24, lr: 2.11e-02, grad_scale: 8.0 +2023-02-05 22:35:08,100 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.077e+02 3.458e+02 4.167e+02 5.729e+02 1.393e+03, threshold=8.335e+02, percent-clipped=8.0 +2023-02-05 22:35:27,673 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23950.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:35:37,130 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23962.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:35:39,048 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23965.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:35:40,188 INFO [train.py:901] (3/4) Epoch 3, batch 7800, loss[loss=0.369, simple_loss=0.4142, pruned_loss=0.1619, over 8342.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3746, pruned_loss=0.1364, over 1613506.93 frames. ], batch size: 26, lr: 2.11e-02, grad_scale: 8.0 +2023-02-05 22:35:45,565 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23975.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:35:53,489 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23987.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:36:14,023 INFO [train.py:901] (3/4) Epoch 3, batch 7850, loss[loss=0.2928, simple_loss=0.3573, pruned_loss=0.1141, over 8535.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3754, pruned_loss=0.1376, over 1610950.92 frames. ], batch size: 28, lr: 2.10e-02, grad_scale: 8.0 +2023-02-05 22:36:16,553 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.230e+02 3.608e+02 4.565e+02 5.801e+02 1.089e+03, threshold=9.129e+02, percent-clipped=5.0 +2023-02-05 22:36:19,400 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.9603, 1.2532, 5.9020, 2.3110, 5.2909, 5.0669, 5.6533, 5.4995], + device='cuda:3'), covar=tensor([0.0241, 0.3166, 0.0159, 0.1496, 0.0668, 0.0300, 0.0216, 0.0264], + device='cuda:3'), in_proj_covar=tensor([0.0261, 0.0432, 0.0320, 0.0343, 0.0410, 0.0338, 0.0317, 0.0358], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2023-02-05 22:36:19,486 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2719, 2.4627, 3.1882, 0.5863, 2.8661, 1.8746, 1.4867, 1.8662], + device='cuda:3'), covar=tensor([0.0185, 0.0082, 0.0071, 0.0254, 0.0157, 0.0258, 0.0251, 0.0125], + device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0176, 0.0141, 0.0227, 0.0170, 0.0308, 0.0241, 0.0202], + device='cuda:3'), out_proj_covar=tensor([1.1150e-04, 7.6177e-05, 6.0818e-05, 9.7618e-05, 7.6524e-05, 1.4657e-04, + 1.0786e-04, 8.6886e-05], device='cuda:3') +2023-02-05 22:36:39,254 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24055.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:36:47,367 INFO [train.py:901] (3/4) Epoch 3, batch 7900, loss[loss=0.3092, simple_loss=0.358, pruned_loss=0.1302, over 8109.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3733, pruned_loss=0.136, over 1608664.80 frames. ], batch size: 23, lr: 2.10e-02, grad_scale: 8.0 +2023-02-05 22:37:20,412 INFO [train.py:901] (3/4) Epoch 3, batch 7950, loss[loss=0.2811, simple_loss=0.3376, pruned_loss=0.1123, over 7557.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3742, pruned_loss=0.1366, over 1612323.13 frames. ], batch size: 18, lr: 2.10e-02, grad_scale: 8.0 +2023-02-05 22:37:23,172 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.080e+02 3.295e+02 4.369e+02 5.897e+02 2.335e+03, threshold=8.738e+02, percent-clipped=5.0 +2023-02-05 22:37:30,776 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.8270, 1.1859, 5.8346, 2.2445, 5.1558, 4.9656, 5.4621, 5.3325], + device='cuda:3'), covar=tensor([0.0245, 0.3479, 0.0199, 0.1573, 0.0772, 0.0396, 0.0275, 0.0302], + device='cuda:3'), in_proj_covar=tensor([0.0273, 0.0443, 0.0332, 0.0357, 0.0429, 0.0351, 0.0337, 0.0375], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-05 22:37:49,164 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9228, 1.5843, 2.4193, 2.0774, 2.1054, 1.6830, 1.3808, 0.6766], + device='cuda:3'), covar=tensor([0.1004, 0.1108, 0.0283, 0.0458, 0.0460, 0.0595, 0.0668, 0.1055], + device='cuda:3'), in_proj_covar=tensor([0.0639, 0.0556, 0.0473, 0.0522, 0.0633, 0.0514, 0.0528, 0.0538], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 22:37:54,056 INFO [train.py:901] (3/4) Epoch 3, batch 8000, loss[loss=0.3002, simple_loss=0.3478, pruned_loss=0.1263, over 7287.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3736, pruned_loss=0.1367, over 1611327.77 frames. ], batch size: 16, lr: 2.10e-02, grad_scale: 8.0 +2023-02-05 22:37:58,422 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2290, 1.6788, 1.3169, 1.6620, 1.4733, 1.1782, 1.2633, 1.5288], + device='cuda:3'), covar=tensor([0.0801, 0.0396, 0.0849, 0.0480, 0.0565, 0.0961, 0.0690, 0.0568], + device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0239, 0.0324, 0.0307, 0.0332, 0.0316, 0.0339, 0.0306], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-05 22:38:27,963 INFO [train.py:901] (3/4) Epoch 3, batch 8050, loss[loss=0.3847, simple_loss=0.4069, pruned_loss=0.1813, over 7230.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3723, pruned_loss=0.137, over 1586903.60 frames. ], batch size: 71, lr: 2.09e-02, grad_scale: 8.0 +2023-02-05 22:38:30,755 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 3.328e+02 4.149e+02 5.404e+02 3.135e+03, threshold=8.298e+02, percent-clipped=6.0 +2023-02-05 22:38:31,005 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24221.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:38:48,122 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24246.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:39:03,930 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-05 22:39:07,722 INFO [train.py:901] (3/4) Epoch 4, batch 0, loss[loss=0.3448, simple_loss=0.3909, pruned_loss=0.1493, over 8583.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.3909, pruned_loss=0.1493, over 8583.00 frames. ], batch size: 31, lr: 1.96e-02, grad_scale: 8.0 +2023-02-05 22:39:07,722 INFO [train.py:926] (3/4) Computing validation loss +2023-02-05 22:39:18,718 INFO [train.py:935] (3/4) Epoch 4, validation: loss=0.2476, simple_loss=0.3384, pruned_loss=0.07836, over 944034.00 frames. +2023-02-05 22:39:18,719 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-05 22:39:34,156 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-05 22:39:52,980 INFO [train.py:901] (3/4) Epoch 4, batch 50, loss[loss=0.2553, simple_loss=0.3309, pruned_loss=0.08989, over 8353.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3674, pruned_loss=0.1322, over 362852.21 frames. ], batch size: 24, lr: 1.96e-02, grad_scale: 8.0 +2023-02-05 22:40:07,592 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.017e+02 3.527e+02 4.250e+02 5.116e+02 9.987e+02, threshold=8.500e+02, percent-clipped=2.0 +2023-02-05 22:40:09,008 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-05 22:40:27,955 INFO [train.py:901] (3/4) Epoch 4, batch 100, loss[loss=0.4026, simple_loss=0.4372, pruned_loss=0.1839, over 8593.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3791, pruned_loss=0.1389, over 645794.14 frames. ], batch size: 49, lr: 1.95e-02, grad_scale: 8.0 +2023-02-05 22:40:31,333 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-05 22:41:01,467 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=24399.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:41:02,085 INFO [train.py:901] (3/4) Epoch 4, batch 150, loss[loss=0.3021, simple_loss=0.3657, pruned_loss=0.1192, over 8281.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3771, pruned_loss=0.1373, over 864773.56 frames. ], batch size: 23, lr: 1.95e-02, grad_scale: 8.0 +2023-02-05 22:41:17,154 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.940e+02 3.490e+02 4.203e+02 5.614e+02 1.653e+03, threshold=8.406e+02, percent-clipped=4.0 +2023-02-05 22:41:37,211 INFO [train.py:901] (3/4) Epoch 4, batch 200, loss[loss=0.3334, simple_loss=0.3873, pruned_loss=0.1398, over 8327.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3751, pruned_loss=0.135, over 1031861.36 frames. ], batch size: 25, lr: 1.95e-02, grad_scale: 8.0 +2023-02-05 22:42:11,040 INFO [train.py:901] (3/4) Epoch 4, batch 250, loss[loss=0.2799, simple_loss=0.3251, pruned_loss=0.1173, over 7690.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.373, pruned_loss=0.1326, over 1164788.93 frames. ], batch size: 18, lr: 1.95e-02, grad_scale: 8.0 +2023-02-05 22:42:20,373 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=24514.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:42:23,570 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-05 22:42:24,844 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.156e+02 3.531e+02 4.434e+02 5.277e+02 1.190e+03, threshold=8.868e+02, percent-clipped=4.0 +2023-02-05 22:42:31,605 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-05 22:42:41,352 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7480, 1.6902, 1.4453, 1.2991, 1.6782, 1.4827, 1.5775, 2.1256], + device='cuda:3'), covar=tensor([0.0545, 0.1258, 0.1951, 0.1485, 0.0684, 0.1605, 0.0927, 0.0542], + device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0209, 0.0252, 0.0207, 0.0168, 0.0211, 0.0176, 0.0174], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0007, 0.0006, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-05 22:42:46,004 INFO [train.py:901] (3/4) Epoch 4, batch 300, loss[loss=0.2569, simple_loss=0.3097, pruned_loss=0.1021, over 7418.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3761, pruned_loss=0.1364, over 1264181.19 frames. ], batch size: 17, lr: 1.95e-02, grad_scale: 8.0 +2023-02-05 22:42:57,005 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24565.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:43:12,332 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24587.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:43:16,435 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3750, 1.8232, 1.7532, 0.4365, 1.6558, 1.2676, 0.3058, 1.5694], + device='cuda:3'), covar=tensor([0.0114, 0.0060, 0.0049, 0.0142, 0.0082, 0.0221, 0.0201, 0.0062], + device='cuda:3'), in_proj_covar=tensor([0.0256, 0.0175, 0.0143, 0.0225, 0.0170, 0.0302, 0.0239, 0.0206], + device='cuda:3'), out_proj_covar=tensor([1.1207e-04, 7.5289e-05, 6.1270e-05, 9.5059e-05, 7.5759e-05, 1.4152e-04, + 1.0535e-04, 8.8042e-05], device='cuda:3') +2023-02-05 22:43:21,554 INFO [train.py:901] (3/4) Epoch 4, batch 350, loss[loss=0.281, simple_loss=0.3394, pruned_loss=0.1113, over 7647.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3746, pruned_loss=0.1358, over 1341676.99 frames. ], batch size: 19, lr: 1.94e-02, grad_scale: 8.0 +2023-02-05 22:43:31,764 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9006, 2.1164, 2.8044, 1.0951, 2.3486, 1.8665, 1.5686, 1.9396], + device='cuda:3'), covar=tensor([0.0203, 0.0083, 0.0058, 0.0180, 0.0149, 0.0178, 0.0200, 0.0111], + device='cuda:3'), in_proj_covar=tensor([0.0259, 0.0178, 0.0146, 0.0229, 0.0172, 0.0306, 0.0242, 0.0210], + device='cuda:3'), out_proj_covar=tensor([1.1364e-04, 7.6424e-05, 6.2785e-05, 9.6814e-05, 7.6290e-05, 1.4319e-04, + 1.0660e-04, 8.9545e-05], device='cuda:3') +2023-02-05 22:43:34,099 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 +2023-02-05 22:43:35,591 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.777e+02 3.300e+02 4.421e+02 5.071e+02 1.044e+03, threshold=8.841e+02, percent-clipped=4.0 +2023-02-05 22:43:52,417 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2023-02-05 22:43:56,478 INFO [train.py:901] (3/4) Epoch 4, batch 400, loss[loss=0.2999, simple_loss=0.3443, pruned_loss=0.1278, over 7176.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.373, pruned_loss=0.1347, over 1403384.03 frames. ], batch size: 16, lr: 1.94e-02, grad_scale: 8.0 +2023-02-05 22:44:04,954 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-05 22:44:30,019 INFO [train.py:901] (3/4) Epoch 4, batch 450, loss[loss=0.311, simple_loss=0.3449, pruned_loss=0.1385, over 7427.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3742, pruned_loss=0.1354, over 1453311.33 frames. ], batch size: 17, lr: 1.94e-02, grad_scale: 8.0 +2023-02-05 22:44:44,808 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.106e+02 3.414e+02 4.548e+02 5.600e+02 1.007e+03, threshold=9.096e+02, percent-clipped=5.0 +2023-02-05 22:45:04,965 INFO [train.py:901] (3/4) Epoch 4, batch 500, loss[loss=0.3341, simple_loss=0.3846, pruned_loss=0.1418, over 8766.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.373, pruned_loss=0.1344, over 1487533.05 frames. ], batch size: 34, lr: 1.94e-02, grad_scale: 8.0 +2023-02-05 22:45:19,921 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24770.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:45:28,220 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24783.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:45:36,835 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24795.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:45:40,054 INFO [train.py:901] (3/4) Epoch 4, batch 550, loss[loss=0.2865, simple_loss=0.3264, pruned_loss=0.1233, over 7414.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3714, pruned_loss=0.1333, over 1516758.02 frames. ], batch size: 17, lr: 1.94e-02, grad_scale: 8.0 +2023-02-05 22:45:53,858 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.817e+02 3.369e+02 4.426e+02 5.591e+02 8.767e+02, threshold=8.852e+02, percent-clipped=0.0 +2023-02-05 22:46:13,959 INFO [train.py:901] (3/4) Epoch 4, batch 600, loss[loss=0.2986, simple_loss=0.3552, pruned_loss=0.121, over 8317.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3726, pruned_loss=0.1334, over 1543921.23 frames. ], batch size: 49, lr: 1.93e-02, grad_scale: 8.0 +2023-02-05 22:46:24,949 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24866.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:46:28,942 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-05 22:46:49,154 INFO [train.py:901] (3/4) Epoch 4, batch 650, loss[loss=0.3868, simple_loss=0.4147, pruned_loss=0.1795, over 7812.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3727, pruned_loss=0.1335, over 1560104.89 frames. ], batch size: 20, lr: 1.93e-02, grad_scale: 8.0 +2023-02-05 22:46:49,645 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 +2023-02-05 22:46:55,189 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=24909.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:47:03,760 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.198e+02 3.310e+02 4.230e+02 5.108e+02 1.167e+03, threshold=8.459e+02, percent-clipped=4.0 +2023-02-05 22:47:10,606 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=24931.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:47:19,517 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6361, 1.3307, 3.3441, 1.4577, 2.3968, 3.8816, 3.6780, 3.3424], + device='cuda:3'), covar=tensor([0.1358, 0.1713, 0.0303, 0.1835, 0.0711, 0.0204, 0.0283, 0.0558], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0262, 0.0207, 0.0264, 0.0212, 0.0187, 0.0187, 0.0265], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 22:47:24,028 INFO [train.py:901] (3/4) Epoch 4, batch 700, loss[loss=0.3196, simple_loss=0.3728, pruned_loss=0.1332, over 8458.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3719, pruned_loss=0.1334, over 1570370.01 frames. ], batch size: 25, lr: 1.93e-02, grad_scale: 8.0 +2023-02-05 22:47:59,150 INFO [train.py:901] (3/4) Epoch 4, batch 750, loss[loss=0.3071, simple_loss=0.3667, pruned_loss=0.1237, over 8025.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3714, pruned_loss=0.133, over 1575538.14 frames. ], batch size: 22, lr: 1.93e-02, grad_scale: 8.0 +2023-02-05 22:48:08,260 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25013.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:48:13,342 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.946e+02 3.175e+02 4.108e+02 5.247e+02 1.235e+03, threshold=8.217e+02, percent-clipped=4.0 +2023-02-05 22:48:14,035 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-05 22:48:15,511 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25024.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:48:22,610 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-05 22:48:30,680 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25046.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:48:33,230 INFO [train.py:901] (3/4) Epoch 4, batch 800, loss[loss=0.2763, simple_loss=0.333, pruned_loss=0.1098, over 7779.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.371, pruned_loss=0.1331, over 1583415.68 frames. ], batch size: 19, lr: 1.93e-02, grad_scale: 8.0 +2023-02-05 22:48:53,299 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3064, 1.9539, 2.1204, 1.3771, 2.1153, 1.4397, 0.5762, 1.7995], + device='cuda:3'), covar=tensor([0.0170, 0.0076, 0.0081, 0.0122, 0.0117, 0.0258, 0.0227, 0.0088], + device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0174, 0.0142, 0.0219, 0.0165, 0.0297, 0.0239, 0.0209], + device='cuda:3'), out_proj_covar=tensor([1.1054e-04, 7.4326e-05, 5.9841e-05, 9.1637e-05, 7.2263e-05, 1.3834e-04, + 1.0356e-04, 8.7703e-05], device='cuda:3') +2023-02-05 22:49:06,963 INFO [train.py:901] (3/4) Epoch 4, batch 850, loss[loss=0.3186, simple_loss=0.364, pruned_loss=0.1366, over 7700.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3697, pruned_loss=0.1329, over 1589333.45 frames. ], batch size: 18, lr: 1.93e-02, grad_scale: 8.0 +2023-02-05 22:49:22,446 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.068e+02 3.301e+02 4.277e+02 5.478e+02 1.022e+03, threshold=8.554e+02, percent-clipped=4.0 +2023-02-05 22:49:26,602 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25127.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:49:42,456 INFO [train.py:901] (3/4) Epoch 4, batch 900, loss[loss=0.3286, simple_loss=0.3855, pruned_loss=0.1359, over 8553.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3696, pruned_loss=0.133, over 1597436.29 frames. ], batch size: 39, lr: 1.92e-02, grad_scale: 8.0 +2023-02-05 22:50:02,302 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5466, 1.1187, 4.5784, 1.8121, 3.8960, 3.7319, 4.0987, 3.9895], + device='cuda:3'), covar=tensor([0.0268, 0.3978, 0.0322, 0.2036, 0.1015, 0.0607, 0.0349, 0.0483], + device='cuda:3'), in_proj_covar=tensor([0.0273, 0.0445, 0.0342, 0.0358, 0.0418, 0.0359, 0.0328, 0.0375], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-05 22:50:16,545 INFO [train.py:901] (3/4) Epoch 4, batch 950, loss[loss=0.3447, simple_loss=0.3824, pruned_loss=0.1536, over 8142.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3694, pruned_loss=0.133, over 1597551.22 frames. ], batch size: 22, lr: 1.92e-02, grad_scale: 8.0 +2023-02-05 22:50:23,514 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25210.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:50:25,724 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3651, 2.3466, 3.0324, 0.5445, 2.8235, 1.7337, 1.2990, 1.4533], + device='cuda:3'), covar=tensor([0.0287, 0.0093, 0.0074, 0.0249, 0.0152, 0.0276, 0.0313, 0.0173], + device='cuda:3'), in_proj_covar=tensor([0.0258, 0.0179, 0.0144, 0.0223, 0.0169, 0.0299, 0.0243, 0.0211], + device='cuda:3'), out_proj_covar=tensor([1.1169e-04, 7.6338e-05, 6.0479e-05, 9.3008e-05, 7.3783e-05, 1.3864e-04, + 1.0534e-04, 8.8758e-05], device='cuda:3') +2023-02-05 22:50:30,882 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.215e+02 3.501e+02 4.488e+02 5.717e+02 1.063e+03, threshold=8.976e+02, percent-clipped=5.0 +2023-02-05 22:50:40,883 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-05 22:50:46,577 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25242.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:50:51,551 INFO [train.py:901] (3/4) Epoch 4, batch 1000, loss[loss=0.34, simple_loss=0.3905, pruned_loss=0.1448, over 8110.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3701, pruned_loss=0.133, over 1602696.60 frames. ], batch size: 23, lr: 1.92e-02, grad_scale: 8.0 +2023-02-05 22:51:12,233 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25280.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:51:13,330 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-05 22:51:21,728 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2023-02-05 22:51:25,838 INFO [train.py:901] (3/4) Epoch 4, batch 1050, loss[loss=0.33, simple_loss=0.3907, pruned_loss=0.1347, over 8251.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3693, pruned_loss=0.132, over 1607056.24 frames. ], batch size: 24, lr: 1.92e-02, grad_scale: 8.0 +2023-02-05 22:51:26,412 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-05 22:51:27,146 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25302.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:51:28,973 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25305.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:51:39,506 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.881e+02 3.519e+02 4.399e+02 5.664e+02 1.146e+03, threshold=8.797e+02, percent-clipped=2.0 +2023-02-05 22:51:42,395 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25325.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:51:43,779 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25327.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:51:58,865 INFO [train.py:901] (3/4) Epoch 4, batch 1100, loss[loss=0.3549, simple_loss=0.3762, pruned_loss=0.1668, over 7710.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3684, pruned_loss=0.1317, over 1605136.73 frames. ], batch size: 18, lr: 1.92e-02, grad_scale: 8.0 +2023-02-05 22:52:03,946 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7897, 3.1328, 3.0308, 4.0621, 1.9240, 1.7543, 2.6248, 3.3313], + device='cuda:3'), covar=tensor([0.0840, 0.1423, 0.1268, 0.0233, 0.1772, 0.2458, 0.1605, 0.1132], + device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0300, 0.0307, 0.0220, 0.0278, 0.0311, 0.0315, 0.0295], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-05 22:52:05,139 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25357.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:52:34,567 INFO [train.py:901] (3/4) Epoch 4, batch 1150, loss[loss=0.3353, simple_loss=0.3864, pruned_loss=0.1421, over 8495.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3675, pruned_loss=0.1307, over 1604546.09 frames. ], batch size: 28, lr: 1.91e-02, grad_scale: 16.0 +2023-02-05 22:52:37,395 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-05 22:52:40,117 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5861, 1.7218, 2.0987, 1.7816, 1.2243, 1.9960, 0.3788, 1.0379], + device='cuda:3'), covar=tensor([0.3607, 0.2366, 0.1041, 0.2514, 0.5734, 0.1481, 0.6260, 0.2444], + device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0112, 0.0082, 0.0158, 0.0190, 0.0081, 0.0154, 0.0115], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 22:52:49,211 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.195e+02 3.278e+02 3.972e+02 4.649e+02 8.065e+02, threshold=7.944e+02, percent-clipped=0.0 +2023-02-05 22:52:55,656 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-05 22:53:06,093 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25446.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 22:53:08,506 INFO [train.py:901] (3/4) Epoch 4, batch 1200, loss[loss=0.2614, simple_loss=0.3186, pruned_loss=0.1021, over 7519.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3677, pruned_loss=0.1306, over 1609749.46 frames. ], batch size: 18, lr: 1.91e-02, grad_scale: 16.0 +2023-02-05 22:53:23,234 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25472.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:53:26,924 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 +2023-02-05 22:53:42,015 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25498.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:53:43,173 INFO [train.py:901] (3/4) Epoch 4, batch 1250, loss[loss=0.3168, simple_loss=0.3747, pruned_loss=0.1294, over 8467.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3671, pruned_loss=0.1301, over 1614121.46 frames. ], batch size: 25, lr: 1.91e-02, grad_scale: 16.0 +2023-02-05 22:53:57,802 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.918e+02 3.538e+02 4.328e+02 6.105e+02 1.271e+03, threshold=8.657e+02, percent-clipped=4.0 +2023-02-05 22:53:59,265 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25523.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:54:18,017 INFO [train.py:901] (3/4) Epoch 4, batch 1300, loss[loss=0.32, simple_loss=0.3924, pruned_loss=0.1238, over 8104.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3678, pruned_loss=0.1299, over 1617575.62 frames. ], batch size: 23, lr: 1.91e-02, grad_scale: 16.0 +2023-02-05 22:54:39,396 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25581.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:54:40,062 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5131, 2.6148, 2.8182, 2.0842, 1.6372, 2.5752, 0.8706, 2.0791], + device='cuda:3'), covar=tensor([0.3150, 0.1368, 0.0835, 0.2238, 0.4711, 0.0806, 0.5690, 0.1971], + device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0110, 0.0086, 0.0159, 0.0190, 0.0082, 0.0152, 0.0118], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 22:54:53,161 INFO [train.py:901] (3/4) Epoch 4, batch 1350, loss[loss=0.2828, simple_loss=0.352, pruned_loss=0.1068, over 8244.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3688, pruned_loss=0.1303, over 1618613.05 frames. ], batch size: 24, lr: 1.91e-02, grad_scale: 16.0 +2023-02-05 22:54:57,489 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25606.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:55:08,861 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.066e+02 3.283e+02 4.098e+02 5.393e+02 1.175e+03, threshold=8.196e+02, percent-clipped=3.0 +2023-02-05 22:55:28,874 INFO [train.py:901] (3/4) Epoch 4, batch 1400, loss[loss=0.3293, simple_loss=0.3828, pruned_loss=0.1379, over 8028.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3701, pruned_loss=0.1318, over 1617913.14 frames. ], batch size: 22, lr: 1.91e-02, grad_scale: 8.0 +2023-02-05 22:56:03,158 INFO [train.py:901] (3/4) Epoch 4, batch 1450, loss[loss=0.327, simple_loss=0.385, pruned_loss=0.1345, over 8528.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3706, pruned_loss=0.1316, over 1620781.71 frames. ], batch size: 28, lr: 1.90e-02, grad_scale: 8.0 +2023-02-05 22:56:05,848 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-05 22:56:14,426 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1827, 1.1493, 2.3185, 1.1609, 1.9388, 2.4694, 2.4259, 2.1364], + device='cuda:3'), covar=tensor([0.1007, 0.1149, 0.0443, 0.1670, 0.0527, 0.0349, 0.0417, 0.0708], + device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0259, 0.0204, 0.0265, 0.0211, 0.0187, 0.0186, 0.0257], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 22:56:18,904 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.916e+02 3.243e+02 3.964e+02 4.847e+02 1.034e+03, threshold=7.929e+02, percent-clipped=2.0 +2023-02-05 22:56:23,194 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25728.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:56:38,603 INFO [train.py:901] (3/4) Epoch 4, batch 1500, loss[loss=0.3592, simple_loss=0.418, pruned_loss=0.1503, over 8281.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3705, pruned_loss=0.1329, over 1616484.27 frames. ], batch size: 23, lr: 1.90e-02, grad_scale: 8.0 +2023-02-05 22:56:40,769 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25753.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 22:57:00,766 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8120, 4.0117, 2.3758, 2.0873, 2.7663, 1.9264, 2.1909, 3.0350], + device='cuda:3'), covar=tensor([0.1400, 0.0237, 0.0712, 0.0738, 0.0610, 0.0949, 0.1069, 0.0668], + device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0239, 0.0325, 0.0301, 0.0337, 0.0318, 0.0346, 0.0312], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-05 22:57:06,013 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25790.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 22:57:12,582 INFO [train.py:901] (3/4) Epoch 4, batch 1550, loss[loss=0.2929, simple_loss=0.3563, pruned_loss=0.1148, over 8685.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.37, pruned_loss=0.1316, over 1616502.39 frames. ], batch size: 39, lr: 1.90e-02, grad_scale: 8.0 +2023-02-05 22:57:20,259 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.62 vs. limit=5.0 +2023-02-05 22:57:23,289 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3272, 2.4258, 1.4374, 1.9210, 1.9396, 1.3871, 1.6687, 2.0340], + device='cuda:3'), covar=tensor([0.1059, 0.0349, 0.0919, 0.0474, 0.0550, 0.1063, 0.0814, 0.0591], + device='cuda:3'), in_proj_covar=tensor([0.0359, 0.0240, 0.0324, 0.0304, 0.0336, 0.0317, 0.0345, 0.0311], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-05 22:57:27,012 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.007e+02 3.100e+02 3.836e+02 5.066e+02 1.009e+03, threshold=7.672e+02, percent-clipped=5.0 +2023-02-05 22:57:46,740 INFO [train.py:901] (3/4) Epoch 4, batch 1600, loss[loss=0.3916, simple_loss=0.4395, pruned_loss=0.1718, over 8467.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3682, pruned_loss=0.1301, over 1614266.24 frames. ], batch size: 48, lr: 1.90e-02, grad_scale: 8.0 +2023-02-05 22:58:03,561 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4074, 1.7014, 1.8550, 1.2631, 1.1402, 1.7672, 0.1675, 1.0354], + device='cuda:3'), covar=tensor([0.2684, 0.1861, 0.1415, 0.2473, 0.6098, 0.1132, 0.5803, 0.2468], + device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0109, 0.0087, 0.0161, 0.0192, 0.0086, 0.0155, 0.0118], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 22:58:04,758 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25876.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 22:58:15,939 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6548, 2.4816, 3.1835, 0.9434, 2.9337, 1.8395, 1.3406, 1.4529], + device='cuda:3'), covar=tensor([0.0256, 0.0111, 0.0075, 0.0206, 0.0115, 0.0265, 0.0269, 0.0157], + device='cuda:3'), in_proj_covar=tensor([0.0258, 0.0184, 0.0146, 0.0226, 0.0169, 0.0305, 0.0248, 0.0208], + device='cuda:3'), out_proj_covar=tensor([1.0990e-04, 7.7610e-05, 6.0591e-05, 9.2636e-05, 7.2225e-05, 1.3973e-04, + 1.0688e-04, 8.6092e-05], device='cuda:3') +2023-02-05 22:58:21,025 INFO [train.py:901] (3/4) Epoch 4, batch 1650, loss[loss=0.3035, simple_loss=0.3731, pruned_loss=0.117, over 8296.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3695, pruned_loss=0.131, over 1613750.25 frames. ], batch size: 23, lr: 1.90e-02, grad_scale: 8.0 +2023-02-05 22:58:24,591 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25905.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 22:58:35,953 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.227e+02 3.823e+02 4.768e+02 5.766e+02 1.707e+03, threshold=9.535e+02, percent-clipped=9.0 +2023-02-05 22:58:36,374 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.28 vs. limit=5.0 +2023-02-05 22:58:56,117 INFO [train.py:901] (3/4) Epoch 4, batch 1700, loss[loss=0.2806, simple_loss=0.3404, pruned_loss=0.1104, over 8253.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.369, pruned_loss=0.1302, over 1615917.76 frames. ], batch size: 22, lr: 1.90e-02, grad_scale: 8.0 +2023-02-05 22:59:28,692 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7636, 2.5178, 3.0825, 0.9869, 2.8377, 1.9021, 1.3425, 1.3695], + device='cuda:3'), covar=tensor([0.0278, 0.0106, 0.0072, 0.0201, 0.0103, 0.0215, 0.0305, 0.0162], + device='cuda:3'), in_proj_covar=tensor([0.0265, 0.0187, 0.0149, 0.0231, 0.0175, 0.0308, 0.0253, 0.0215], + device='cuda:3'), out_proj_covar=tensor([1.1280e-04, 7.8443e-05, 6.1954e-05, 9.5115e-05, 7.5216e-05, 1.4033e-04, + 1.0863e-04, 8.9326e-05], device='cuda:3') +2023-02-05 22:59:31,183 INFO [train.py:901] (3/4) Epoch 4, batch 1750, loss[loss=0.2778, simple_loss=0.3528, pruned_loss=0.1014, over 8527.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3707, pruned_loss=0.1318, over 1614649.63 frames. ], batch size: 31, lr: 1.89e-02, grad_scale: 8.0 +2023-02-05 22:59:46,995 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.961e+02 3.187e+02 3.816e+02 4.801e+02 8.317e+02, threshold=7.632e+02, percent-clipped=0.0 +2023-02-05 23:00:06,090 INFO [train.py:901] (3/4) Epoch 4, batch 1800, loss[loss=0.2984, simple_loss=0.358, pruned_loss=0.1194, over 8030.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3694, pruned_loss=0.1311, over 1612664.70 frames. ], batch size: 22, lr: 1.89e-02, grad_scale: 8.0 +2023-02-05 23:00:40,846 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3978, 1.9215, 3.2873, 2.5949, 2.3664, 1.9528, 1.3626, 1.2037], + device='cuda:3'), covar=tensor([0.1190, 0.1474, 0.0300, 0.0655, 0.0736, 0.0705, 0.0812, 0.1504], + device='cuda:3'), in_proj_covar=tensor([0.0662, 0.0582, 0.0488, 0.0546, 0.0658, 0.0539, 0.0543, 0.0550], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 23:00:41,280 INFO [train.py:901] (3/4) Epoch 4, batch 1850, loss[loss=0.2828, simple_loss=0.343, pruned_loss=0.1113, over 7815.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3679, pruned_loss=0.1303, over 1610111.31 frames. ], batch size: 20, lr: 1.89e-02, grad_scale: 8.0 +2023-02-05 23:00:55,437 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26120.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:00:56,608 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.260e+02 3.379e+02 4.261e+02 5.084e+02 1.608e+03, threshold=8.521e+02, percent-clipped=6.0 +2023-02-05 23:01:05,442 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.4967, 1.3261, 3.6039, 1.4403, 3.0784, 2.9768, 3.1267, 3.0816], + device='cuda:3'), covar=tensor([0.0406, 0.3155, 0.0432, 0.2120, 0.1067, 0.0601, 0.0481, 0.0538], + device='cuda:3'), in_proj_covar=tensor([0.0276, 0.0452, 0.0335, 0.0362, 0.0435, 0.0359, 0.0342, 0.0385], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-05 23:01:15,406 INFO [train.py:901] (3/4) Epoch 4, batch 1900, loss[loss=0.2687, simple_loss=0.3292, pruned_loss=0.104, over 7825.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3685, pruned_loss=0.1307, over 1610715.62 frames. ], batch size: 20, lr: 1.89e-02, grad_scale: 8.0 +2023-02-05 23:01:22,881 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26161.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:01:40,599 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26186.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:01:40,659 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26186.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:01:41,083 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-05 23:01:49,621 INFO [train.py:901] (3/4) Epoch 4, batch 1950, loss[loss=0.3092, simple_loss=0.36, pruned_loss=0.1292, over 7948.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3677, pruned_loss=0.1301, over 1609820.60 frames. ], batch size: 20, lr: 1.89e-02, grad_scale: 8.0 +2023-02-05 23:01:52,432 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-05 23:02:04,052 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26220.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:02:05,149 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.303e+02 3.684e+02 4.572e+02 6.046e+02 1.247e+03, threshold=9.144e+02, percent-clipped=2.0 +2023-02-05 23:02:10,399 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-05 23:02:24,335 INFO [train.py:901] (3/4) Epoch 4, batch 2000, loss[loss=0.2797, simple_loss=0.3425, pruned_loss=0.1084, over 8081.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3673, pruned_loss=0.1291, over 1612482.98 frames. ], batch size: 21, lr: 1.88e-02, grad_scale: 8.0 +2023-02-05 23:02:26,386 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3581, 2.3994, 1.5857, 1.9128, 1.9756, 1.3338, 1.8129, 1.8490], + device='cuda:3'), covar=tensor([0.0989, 0.0311, 0.0826, 0.0461, 0.0545, 0.0994, 0.0720, 0.0589], + device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0233, 0.0324, 0.0308, 0.0329, 0.0314, 0.0339, 0.0307], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-05 23:02:33,520 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-05 23:02:36,635 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26268.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:02:38,771 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8670, 2.1033, 1.8622, 2.8170, 1.2857, 1.4125, 1.6830, 2.2338], + device='cuda:3'), covar=tensor([0.1234, 0.1210, 0.1456, 0.0417, 0.1755, 0.2459, 0.1708, 0.1072], + device='cuda:3'), in_proj_covar=tensor([0.0293, 0.0290, 0.0305, 0.0227, 0.0272, 0.0308, 0.0317, 0.0289], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-05 23:02:46,064 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1277, 1.4193, 2.9521, 0.8845, 2.0776, 1.3354, 1.1606, 1.8090], + device='cuda:3'), covar=tensor([0.2170, 0.2080, 0.0751, 0.3526, 0.1436, 0.2817, 0.1958, 0.2266], + device='cuda:3'), in_proj_covar=tensor([0.0458, 0.0416, 0.0507, 0.0509, 0.0545, 0.0494, 0.0441, 0.0565], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 23:02:59,479 INFO [train.py:901] (3/4) Epoch 4, batch 2050, loss[loss=0.3522, simple_loss=0.4124, pruned_loss=0.146, over 8332.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3681, pruned_loss=0.1295, over 1614708.53 frames. ], batch size: 25, lr: 1.88e-02, grad_scale: 8.0 +2023-02-05 23:03:14,377 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.223e+02 3.433e+02 4.198e+02 5.260e+02 1.263e+03, threshold=8.396e+02, percent-clipped=5.0 +2023-02-05 23:03:24,279 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26335.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:03:34,743 INFO [train.py:901] (3/4) Epoch 4, batch 2100, loss[loss=0.2851, simple_loss=0.3527, pruned_loss=0.1088, over 7971.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3681, pruned_loss=0.1298, over 1617342.62 frames. ], batch size: 21, lr: 1.88e-02, grad_scale: 8.0 +2023-02-05 23:04:08,107 INFO [train.py:901] (3/4) Epoch 4, batch 2150, loss[loss=0.2938, simple_loss=0.3551, pruned_loss=0.1163, over 8234.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3676, pruned_loss=0.129, over 1616270.71 frames. ], batch size: 22, lr: 1.88e-02, grad_scale: 8.0 +2023-02-05 23:04:24,393 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.431e+02 3.407e+02 4.210e+02 5.616e+02 1.521e+03, threshold=8.419e+02, percent-clipped=4.0 +2023-02-05 23:04:31,157 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26432.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:04:43,606 INFO [train.py:901] (3/4) Epoch 4, batch 2200, loss[loss=0.3548, simple_loss=0.3775, pruned_loss=0.1661, over 7971.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3665, pruned_loss=0.1292, over 1610838.77 frames. ], batch size: 21, lr: 1.88e-02, grad_scale: 8.0 +2023-02-05 23:04:53,186 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26464.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:05:18,123 INFO [train.py:901] (3/4) Epoch 4, batch 2250, loss[loss=0.2894, simple_loss=0.3532, pruned_loss=0.1128, over 7218.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3659, pruned_loss=0.1286, over 1614338.61 frames. ], batch size: 16, lr: 1.88e-02, grad_scale: 8.0 +2023-02-05 23:05:33,090 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.934e+02 3.188e+02 3.857e+02 4.748e+02 9.287e+02, threshold=7.714e+02, percent-clipped=1.0 +2023-02-05 23:05:38,938 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26530.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:05:52,817 INFO [train.py:901] (3/4) Epoch 4, batch 2300, loss[loss=0.2829, simple_loss=0.3573, pruned_loss=0.1042, over 8186.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3662, pruned_loss=0.1282, over 1615444.12 frames. ], batch size: 23, lr: 1.87e-02, grad_scale: 8.0 +2023-02-05 23:06:12,949 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26579.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:06:21,880 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26591.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:06:27,554 INFO [train.py:901] (3/4) Epoch 4, batch 2350, loss[loss=0.3627, simple_loss=0.4121, pruned_loss=0.1566, over 8701.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3657, pruned_loss=0.1285, over 1616057.90 frames. ], batch size: 49, lr: 1.87e-02, grad_scale: 8.0 +2023-02-05 23:06:35,960 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26612.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:06:38,764 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26616.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:06:42,405 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.224e+02 3.505e+02 4.841e+02 5.770e+02 1.247e+03, threshold=9.683e+02, percent-clipped=6.0 +2023-02-05 23:06:58,654 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26645.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:07:01,508 INFO [train.py:901] (3/4) Epoch 4, batch 2400, loss[loss=0.3525, simple_loss=0.3928, pruned_loss=0.1561, over 7972.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3673, pruned_loss=0.1299, over 1608781.82 frames. ], batch size: 21, lr: 1.87e-02, grad_scale: 8.0 +2023-02-05 23:07:17,305 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.7252, 1.2497, 3.9312, 1.4563, 3.2634, 3.2546, 3.4683, 3.3882], + device='cuda:3'), covar=tensor([0.0344, 0.2899, 0.0301, 0.1968, 0.0923, 0.0564, 0.0382, 0.0460], + device='cuda:3'), in_proj_covar=tensor([0.0272, 0.0448, 0.0348, 0.0364, 0.0428, 0.0365, 0.0345, 0.0383], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-05 23:07:35,315 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([0.9489, 1.0239, 1.6639, 0.8573, 1.6106, 1.8780, 1.7660, 1.5933], + device='cuda:3'), covar=tensor([0.0822, 0.0906, 0.0529, 0.1511, 0.0504, 0.0313, 0.0436, 0.0595], + device='cuda:3'), in_proj_covar=tensor([0.0225, 0.0259, 0.0205, 0.0262, 0.0209, 0.0187, 0.0195, 0.0260], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 23:07:37,135 INFO [train.py:901] (3/4) Epoch 4, batch 2450, loss[loss=0.2994, simple_loss=0.3443, pruned_loss=0.1273, over 7803.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3684, pruned_loss=0.131, over 1610572.04 frames. ], batch size: 20, lr: 1.87e-02, grad_scale: 8.0 +2023-02-05 23:07:51,856 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.126e+02 3.211e+02 4.300e+02 5.616e+02 1.854e+03, threshold=8.599e+02, percent-clipped=7.0 +2023-02-05 23:07:55,326 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26727.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:08:10,588 INFO [train.py:901] (3/4) Epoch 4, batch 2500, loss[loss=0.3405, simple_loss=0.3808, pruned_loss=0.1501, over 8515.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3689, pruned_loss=0.1312, over 1615582.16 frames. ], batch size: 26, lr: 1.87e-02, grad_scale: 8.0 +2023-02-05 23:08:19,388 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7302, 2.5332, 2.8795, 0.8496, 2.7532, 1.8108, 1.3176, 1.9646], + device='cuda:3'), covar=tensor([0.0234, 0.0081, 0.0093, 0.0207, 0.0121, 0.0230, 0.0297, 0.0115], + device='cuda:3'), in_proj_covar=tensor([0.0258, 0.0185, 0.0150, 0.0222, 0.0170, 0.0303, 0.0248, 0.0207], + device='cuda:3'), out_proj_covar=tensor([1.0812e-04, 7.6519e-05, 6.0870e-05, 8.9652e-05, 7.1888e-05, 1.3598e-04, + 1.0501e-04, 8.4309e-05], device='cuda:3') +2023-02-05 23:08:29,253 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26776.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:08:45,148 INFO [train.py:901] (3/4) Epoch 4, batch 2550, loss[loss=0.2999, simple_loss=0.3747, pruned_loss=0.1126, over 8354.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3681, pruned_loss=0.1303, over 1621133.75 frames. ], batch size: 24, lr: 1.87e-02, grad_scale: 8.0 +2023-02-05 23:09:01,300 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.235e+02 3.301e+02 4.146e+02 5.074e+02 1.055e+03, threshold=8.293e+02, percent-clipped=2.0 +2023-02-05 23:09:10,581 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26835.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:09:11,944 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3248, 1.8967, 1.7569, 0.5661, 1.8551, 1.2774, 0.3774, 1.6454], + device='cuda:3'), covar=tensor([0.0181, 0.0074, 0.0077, 0.0171, 0.0104, 0.0290, 0.0244, 0.0070], + device='cuda:3'), in_proj_covar=tensor([0.0257, 0.0187, 0.0152, 0.0223, 0.0170, 0.0306, 0.0246, 0.0208], + device='cuda:3'), out_proj_covar=tensor([1.0805e-04, 7.7575e-05, 6.1764e-05, 8.9955e-05, 7.1652e-05, 1.3747e-04, + 1.0457e-04, 8.4644e-05], device='cuda:3') +2023-02-05 23:09:20,514 INFO [train.py:901] (3/4) Epoch 4, batch 2600, loss[loss=0.3327, simple_loss=0.3892, pruned_loss=0.1381, over 8501.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3676, pruned_loss=0.1293, over 1624317.99 frames. ], batch size: 31, lr: 1.86e-02, grad_scale: 8.0 +2023-02-05 23:09:27,765 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26860.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:09:50,370 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26891.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:09:56,432 INFO [train.py:901] (3/4) Epoch 4, batch 2650, loss[loss=0.3178, simple_loss=0.369, pruned_loss=0.1332, over 8447.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3664, pruned_loss=0.1281, over 1622562.41 frames. ], batch size: 29, lr: 1.86e-02, grad_scale: 8.0 +2023-02-05 23:09:57,275 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26901.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:10:12,336 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.619e+02 3.245e+02 3.916e+02 5.024e+02 1.006e+03, threshold=7.831e+02, percent-clipped=3.0 +2023-02-05 23:10:12,503 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26922.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:10:15,324 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26926.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:10:21,530 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.03 vs. limit=2.0 +2023-02-05 23:10:32,094 INFO [train.py:901] (3/4) Epoch 4, batch 2700, loss[loss=0.2667, simple_loss=0.336, pruned_loss=0.09867, over 8190.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3648, pruned_loss=0.1274, over 1617077.16 frames. ], batch size: 23, lr: 1.86e-02, grad_scale: 8.0 +2023-02-05 23:10:44,346 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26968.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:10:54,316 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26983.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:11:05,902 INFO [train.py:901] (3/4) Epoch 4, batch 2750, loss[loss=0.2661, simple_loss=0.3269, pruned_loss=0.1027, over 7552.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3645, pruned_loss=0.1274, over 1615706.71 frames. ], batch size: 18, lr: 1.86e-02, grad_scale: 8.0 +2023-02-05 23:11:12,359 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27008.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:11:21,392 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.241e+02 3.589e+02 4.354e+02 5.460e+02 1.197e+03, threshold=8.707e+02, percent-clipped=9.0 +2023-02-05 23:11:40,852 INFO [train.py:901] (3/4) Epoch 4, batch 2800, loss[loss=0.2638, simple_loss=0.3287, pruned_loss=0.09949, over 7522.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3646, pruned_loss=0.1274, over 1613489.43 frames. ], batch size: 18, lr: 1.86e-02, grad_scale: 8.0 +2023-02-05 23:12:11,819 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-05 23:12:14,844 INFO [train.py:901] (3/4) Epoch 4, batch 2850, loss[loss=0.2874, simple_loss=0.3319, pruned_loss=0.1214, over 7935.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3645, pruned_loss=0.1276, over 1615310.23 frames. ], batch size: 20, lr: 1.86e-02, grad_scale: 8.0 +2023-02-05 23:12:16,670 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-05 23:12:25,957 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.88 vs. limit=2.0 +2023-02-05 23:12:30,250 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.884e+02 3.374e+02 4.464e+02 5.831e+02 1.992e+03, threshold=8.927e+02, percent-clipped=6.0 +2023-02-05 23:12:31,895 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-05 23:12:39,737 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5846, 2.7930, 1.7639, 1.9974, 2.1841, 1.4595, 1.8818, 2.1174], + device='cuda:3'), covar=tensor([0.1125, 0.0276, 0.0749, 0.0570, 0.0503, 0.0952, 0.0832, 0.0742], + device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0245, 0.0308, 0.0306, 0.0330, 0.0309, 0.0338, 0.0306], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-05 23:12:47,570 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27147.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:12:49,277 INFO [train.py:901] (3/4) Epoch 4, batch 2900, loss[loss=0.2822, simple_loss=0.3283, pruned_loss=0.1181, over 6819.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3647, pruned_loss=0.1277, over 1613278.71 frames. ], batch size: 15, lr: 1.85e-02, grad_scale: 8.0 +2023-02-05 23:13:00,620 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27166.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:13:05,272 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27172.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:13:11,773 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-05 23:13:15,428 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2632, 1.7930, 2.7746, 1.0243, 1.9977, 1.5256, 1.5889, 1.5434], + device='cuda:3'), covar=tensor([0.1617, 0.1543, 0.0716, 0.3081, 0.1215, 0.2375, 0.1277, 0.1966], + device='cuda:3'), in_proj_covar=tensor([0.0450, 0.0420, 0.0505, 0.0511, 0.0553, 0.0488, 0.0431, 0.0560], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 23:13:24,075 INFO [train.py:901] (3/4) Epoch 4, batch 2950, loss[loss=0.2588, simple_loss=0.316, pruned_loss=0.1009, over 7545.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3655, pruned_loss=0.1279, over 1617810.78 frames. ], batch size: 18, lr: 1.85e-02, grad_scale: 8.0 +2023-02-05 23:13:34,431 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4705, 1.7127, 2.9621, 1.1530, 2.1295, 1.8657, 1.4983, 1.6693], + device='cuda:3'), covar=tensor([0.1262, 0.1482, 0.0451, 0.2594, 0.1055, 0.1858, 0.1247, 0.1680], + device='cuda:3'), in_proj_covar=tensor([0.0447, 0.0419, 0.0502, 0.0510, 0.0553, 0.0489, 0.0433, 0.0560], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 23:13:38,737 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.890e+02 3.092e+02 3.649e+02 5.055e+02 1.216e+03, threshold=7.299e+02, percent-clipped=3.0 +2023-02-05 23:13:58,833 INFO [train.py:901] (3/4) Epoch 4, batch 3000, loss[loss=0.3507, simple_loss=0.4027, pruned_loss=0.1493, over 8325.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3639, pruned_loss=0.1264, over 1618124.07 frames. ], batch size: 25, lr: 1.85e-02, grad_scale: 8.0 +2023-02-05 23:13:58,833 INFO [train.py:926] (3/4) Computing validation loss +2023-02-05 23:14:04,835 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.5123, 1.2262, 3.6821, 1.4935, 3.1495, 3.0924, 3.3008, 3.2446], + device='cuda:3'), covar=tensor([0.0457, 0.3669, 0.0372, 0.2337, 0.1322, 0.0706, 0.0480, 0.0593], + device='cuda:3'), in_proj_covar=tensor([0.0278, 0.0452, 0.0352, 0.0365, 0.0438, 0.0359, 0.0356, 0.0394], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-05 23:14:11,267 INFO [train.py:935] (3/4) Epoch 4, validation: loss=0.2374, simple_loss=0.3304, pruned_loss=0.07225, over 944034.00 frames. +2023-02-05 23:14:11,268 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-05 23:14:23,018 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27266.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:14:24,760 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-05 23:14:45,722 INFO [train.py:901] (3/4) Epoch 4, batch 3050, loss[loss=0.3891, simple_loss=0.4171, pruned_loss=0.1805, over 7316.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3643, pruned_loss=0.1275, over 1609491.73 frames. ], batch size: 71, lr: 1.85e-02, grad_scale: 8.0 +2023-02-05 23:14:54,658 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27312.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:14:54,763 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4046, 1.8809, 1.5723, 1.4205, 1.5841, 1.5170, 1.9059, 1.9482], + device='cuda:3'), covar=tensor([0.0618, 0.1102, 0.1721, 0.1431, 0.0709, 0.1459, 0.0811, 0.0547], + device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0200, 0.0239, 0.0199, 0.0157, 0.0204, 0.0165, 0.0167], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0007, 0.0007, 0.0005, 0.0007, 0.0006, 0.0005], + device='cuda:3') +2023-02-05 23:15:01,945 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.971e+02 3.415e+02 4.317e+02 5.768e+02 1.933e+03, threshold=8.634e+02, percent-clipped=10.0 +2023-02-05 23:15:08,182 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1851, 1.9772, 3.4340, 1.5882, 2.7699, 3.8725, 3.5374, 3.3730], + device='cuda:3'), covar=tensor([0.0879, 0.1195, 0.0325, 0.1774, 0.0504, 0.0239, 0.0346, 0.0485], + device='cuda:3'), in_proj_covar=tensor([0.0228, 0.0265, 0.0216, 0.0267, 0.0211, 0.0191, 0.0198, 0.0265], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 23:15:20,637 INFO [train.py:901] (3/4) Epoch 4, batch 3100, loss[loss=0.3819, simple_loss=0.4069, pruned_loss=0.1784, over 8445.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3648, pruned_loss=0.1281, over 1609788.33 frames. ], batch size: 27, lr: 1.85e-02, grad_scale: 8.0 +2023-02-05 23:15:41,813 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27381.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:15:54,825 INFO [train.py:901] (3/4) Epoch 4, batch 3150, loss[loss=0.2467, simple_loss=0.3169, pruned_loss=0.08828, over 7658.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3662, pruned_loss=0.1289, over 1612064.08 frames. ], batch size: 19, lr: 1.85e-02, grad_scale: 8.0 +2023-02-05 23:16:09,488 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.656e+02 3.237e+02 4.041e+02 5.193e+02 1.210e+03, threshold=8.082e+02, percent-clipped=3.0 +2023-02-05 23:16:13,668 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27427.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:16:27,931 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.07 vs. limit=2.0 +2023-02-05 23:16:29,613 INFO [train.py:901] (3/4) Epoch 4, batch 3200, loss[loss=0.2722, simple_loss=0.3437, pruned_loss=0.1004, over 7973.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3665, pruned_loss=0.1283, over 1613758.47 frames. ], batch size: 21, lr: 1.84e-02, grad_scale: 8.0 +2023-02-05 23:16:31,281 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.02 vs. limit=5.0 +2023-02-05 23:17:03,113 INFO [train.py:901] (3/4) Epoch 4, batch 3250, loss[loss=0.2913, simple_loss=0.3439, pruned_loss=0.1194, over 7454.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3654, pruned_loss=0.1277, over 1611029.66 frames. ], batch size: 17, lr: 1.84e-02, grad_scale: 8.0 +2023-02-05 23:17:08,660 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3839, 1.6157, 1.5326, 1.2820, 1.5268, 1.3455, 1.7307, 1.5238], + device='cuda:3'), covar=tensor([0.0610, 0.1192, 0.1703, 0.1487, 0.0640, 0.1593, 0.0782, 0.0600], + device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0198, 0.0236, 0.0197, 0.0155, 0.0202, 0.0162, 0.0166], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0006, 0.0007, 0.0007, 0.0005, 0.0007, 0.0006, 0.0005], + device='cuda:3') +2023-02-05 23:17:10,576 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27510.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:17:17,970 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3614, 2.4748, 1.5623, 2.2358, 1.9250, 1.4298, 1.6567, 2.0343], + device='cuda:3'), covar=tensor([0.0960, 0.0303, 0.0816, 0.0429, 0.0546, 0.0954, 0.0891, 0.0609], + device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0238, 0.0308, 0.0299, 0.0322, 0.0307, 0.0332, 0.0304], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-05 23:17:18,423 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.911e+02 3.449e+02 4.059e+02 4.930e+02 7.939e+02, threshold=8.117e+02, percent-clipped=0.0 +2023-02-05 23:17:37,484 INFO [train.py:901] (3/4) Epoch 4, batch 3300, loss[loss=0.3255, simple_loss=0.3659, pruned_loss=0.1426, over 7647.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3644, pruned_loss=0.127, over 1613993.24 frames. ], batch size: 19, lr: 1.84e-02, grad_scale: 8.0 +2023-02-05 23:18:12,294 INFO [train.py:901] (3/4) Epoch 4, batch 3350, loss[loss=0.2473, simple_loss=0.3048, pruned_loss=0.09493, over 7427.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3651, pruned_loss=0.1274, over 1617110.79 frames. ], batch size: 17, lr: 1.84e-02, grad_scale: 8.0 +2023-02-05 23:18:28,393 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.390e+02 3.326e+02 4.176e+02 5.439e+02 1.733e+03, threshold=8.353e+02, percent-clipped=9.0 +2023-02-05 23:18:30,513 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27625.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:18:38,359 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27637.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:18:46,891 INFO [train.py:901] (3/4) Epoch 4, batch 3400, loss[loss=0.3365, simple_loss=0.393, pruned_loss=0.14, over 8026.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3661, pruned_loss=0.1279, over 1618136.41 frames. ], batch size: 22, lr: 1.84e-02, grad_scale: 16.0 +2023-02-05 23:18:50,486 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3119, 1.7585, 2.8317, 1.0820, 2.0391, 1.7222, 1.5183, 1.5771], + device='cuda:3'), covar=tensor([0.1457, 0.1507, 0.0536, 0.2717, 0.1098, 0.2029, 0.1245, 0.1612], + device='cuda:3'), in_proj_covar=tensor([0.0462, 0.0430, 0.0517, 0.0521, 0.0569, 0.0501, 0.0443, 0.0573], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 23:18:55,795 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27662.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:19:10,315 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27683.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:19:21,476 INFO [train.py:901] (3/4) Epoch 4, batch 3450, loss[loss=0.336, simple_loss=0.3923, pruned_loss=0.1398, over 8290.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3667, pruned_loss=0.1283, over 1620761.06 frames. ], batch size: 23, lr: 1.84e-02, grad_scale: 16.0 +2023-02-05 23:19:26,949 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27708.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:19:36,065 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.499e+02 3.357e+02 4.072e+02 5.275e+02 9.264e+02, threshold=8.144e+02, percent-clipped=1.0 +2023-02-05 23:19:43,549 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27732.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:19:49,036 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 +2023-02-05 23:19:55,679 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 +2023-02-05 23:19:55,903 INFO [train.py:901] (3/4) Epoch 4, batch 3500, loss[loss=0.3061, simple_loss=0.3644, pruned_loss=0.1238, over 8024.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3676, pruned_loss=0.1287, over 1621711.03 frames. ], batch size: 22, lr: 1.83e-02, grad_scale: 16.0 +2023-02-05 23:19:59,473 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4323, 1.4405, 2.7757, 1.2819, 1.9056, 3.0322, 2.8290, 2.5974], + device='cuda:3'), covar=tensor([0.1046, 0.1281, 0.0418, 0.1816, 0.0749, 0.0259, 0.0455, 0.0596], + device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0267, 0.0215, 0.0262, 0.0216, 0.0189, 0.0198, 0.0265], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 23:20:10,686 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-05 23:20:31,122 INFO [train.py:901] (3/4) Epoch 4, batch 3550, loss[loss=0.3593, simple_loss=0.4049, pruned_loss=0.1568, over 8677.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3656, pruned_loss=0.1273, over 1622594.56 frames. ], batch size: 34, lr: 1.83e-02, grad_scale: 16.0 +2023-02-05 23:20:46,086 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.197e+02 3.262e+02 3.955e+02 5.254e+02 1.114e+03, threshold=7.909e+02, percent-clipped=8.0 +2023-02-05 23:21:05,473 INFO [train.py:901] (3/4) Epoch 4, batch 3600, loss[loss=0.2615, simple_loss=0.3302, pruned_loss=0.09641, over 7968.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3663, pruned_loss=0.1279, over 1624116.93 frames. ], batch size: 21, lr: 1.83e-02, grad_scale: 16.0 +2023-02-05 23:21:27,351 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27881.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:21:39,940 INFO [train.py:901] (3/4) Epoch 4, batch 3650, loss[loss=0.3601, simple_loss=0.3919, pruned_loss=0.1642, over 6784.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3636, pruned_loss=0.1252, over 1614900.18 frames. ], batch size: 71, lr: 1.83e-02, grad_scale: 16.0 +2023-02-05 23:21:44,909 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27906.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:21:56,104 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.998e+02 3.334e+02 3.945e+02 4.811e+02 1.062e+03, threshold=7.891e+02, percent-clipped=4.0 +2023-02-05 23:22:13,487 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-05 23:22:14,787 INFO [train.py:901] (3/4) Epoch 4, batch 3700, loss[loss=0.3761, simple_loss=0.3927, pruned_loss=0.1798, over 8325.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3642, pruned_loss=0.1261, over 1616729.16 frames. ], batch size: 26, lr: 1.83e-02, grad_scale: 16.0 +2023-02-05 23:22:18,647 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 +2023-02-05 23:22:29,744 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 +2023-02-05 23:22:49,606 INFO [train.py:901] (3/4) Epoch 4, batch 3750, loss[loss=0.3147, simple_loss=0.3838, pruned_loss=0.1228, over 8301.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3651, pruned_loss=0.1269, over 1616656.35 frames. ], batch size: 23, lr: 1.83e-02, grad_scale: 8.0 +2023-02-05 23:22:53,491 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.9088, 1.7384, 5.8356, 1.8536, 5.3132, 4.8193, 5.4571, 5.2754], + device='cuda:3'), covar=tensor([0.0320, 0.3328, 0.0221, 0.2069, 0.0769, 0.0474, 0.0309, 0.0357], + device='cuda:3'), in_proj_covar=tensor([0.0281, 0.0457, 0.0352, 0.0368, 0.0440, 0.0365, 0.0357, 0.0395], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-05 23:23:05,800 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.245e+02 3.553e+02 4.442e+02 6.055e+02 1.985e+03, threshold=8.883e+02, percent-clipped=11.0 +2023-02-05 23:23:13,957 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6031, 1.6062, 4.6664, 1.8023, 4.0607, 3.9498, 4.2126, 4.1433], + device='cuda:3'), covar=tensor([0.0357, 0.3006, 0.0274, 0.2087, 0.0955, 0.0552, 0.0404, 0.0433], + device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0457, 0.0354, 0.0370, 0.0443, 0.0367, 0.0357, 0.0394], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-05 23:23:25,246 INFO [train.py:901] (3/4) Epoch 4, batch 3800, loss[loss=0.279, simple_loss=0.3345, pruned_loss=0.1117, over 7780.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.365, pruned_loss=0.1271, over 1616114.42 frames. ], batch size: 19, lr: 1.83e-02, grad_scale: 8.0 +2023-02-05 23:23:42,804 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28076.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:23:47,615 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4940, 2.2826, 4.5064, 1.2254, 2.8924, 2.1307, 1.7221, 2.2263], + device='cuda:3'), covar=tensor([0.1374, 0.1671, 0.0607, 0.2799, 0.1359, 0.2033, 0.1275, 0.2247], + device='cuda:3'), in_proj_covar=tensor([0.0449, 0.0426, 0.0505, 0.0512, 0.0560, 0.0487, 0.0433, 0.0564], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 23:24:00,299 INFO [train.py:901] (3/4) Epoch 4, batch 3850, loss[loss=0.2465, simple_loss=0.3116, pruned_loss=0.09066, over 7943.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3644, pruned_loss=0.1269, over 1614543.36 frames. ], batch size: 20, lr: 1.82e-02, grad_scale: 8.0 +2023-02-05 23:24:15,220 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.006e+02 3.238e+02 4.124e+02 5.182e+02 9.210e+02, threshold=8.247e+02, percent-clipped=1.0 +2023-02-05 23:24:17,309 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-05 23:24:34,697 INFO [train.py:901] (3/4) Epoch 4, batch 3900, loss[loss=0.2995, simple_loss=0.3623, pruned_loss=0.1183, over 8448.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3661, pruned_loss=0.1282, over 1614231.27 frames. ], batch size: 27, lr: 1.82e-02, grad_scale: 8.0 +2023-02-05 23:25:02,735 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28191.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:25:08,590 INFO [train.py:901] (3/4) Epoch 4, batch 3950, loss[loss=0.3702, simple_loss=0.4063, pruned_loss=0.167, over 6962.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3668, pruned_loss=0.1285, over 1614918.25 frames. ], batch size: 72, lr: 1.82e-02, grad_scale: 8.0 +2023-02-05 23:25:19,008 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4621, 1.9601, 3.3713, 1.1843, 2.4904, 1.7252, 1.5619, 1.8092], + device='cuda:3'), covar=tensor([0.1437, 0.1615, 0.0583, 0.2737, 0.1231, 0.2205, 0.1356, 0.2093], + device='cuda:3'), in_proj_covar=tensor([0.0455, 0.0427, 0.0504, 0.0509, 0.0565, 0.0493, 0.0434, 0.0565], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 23:25:24,838 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.029e+02 3.357e+02 4.080e+02 5.453e+02 1.389e+03, threshold=8.161e+02, percent-clipped=8.0 +2023-02-05 23:25:32,133 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-05 23:25:41,198 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28247.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:25:43,099 INFO [train.py:901] (3/4) Epoch 4, batch 4000, loss[loss=0.2556, simple_loss=0.3183, pruned_loss=0.0965, over 7546.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3667, pruned_loss=0.1285, over 1613780.18 frames. ], batch size: 18, lr: 1.82e-02, grad_scale: 8.0 +2023-02-05 23:25:56,829 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.66 vs. limit=5.0 +2023-02-05 23:25:59,940 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28273.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:26:17,616 INFO [train.py:901] (3/4) Epoch 4, batch 4050, loss[loss=0.2874, simple_loss=0.3547, pruned_loss=0.1101, over 8457.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3662, pruned_loss=0.1281, over 1614079.47 frames. ], batch size: 27, lr: 1.82e-02, grad_scale: 8.0 +2023-02-05 23:26:21,162 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28305.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:26:34,449 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.986e+02 3.482e+02 4.201e+02 5.400e+02 1.078e+03, threshold=8.403e+02, percent-clipped=4.0 +2023-02-05 23:26:52,365 INFO [train.py:901] (3/4) Epoch 4, batch 4100, loss[loss=0.3163, simple_loss=0.3464, pruned_loss=0.1431, over 7807.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3669, pruned_loss=0.1286, over 1614984.26 frames. ], batch size: 19, lr: 1.82e-02, grad_scale: 8.0 +2023-02-05 23:27:27,346 INFO [train.py:901] (3/4) Epoch 4, batch 4150, loss[loss=0.3619, simple_loss=0.4091, pruned_loss=0.1573, over 8506.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3652, pruned_loss=0.1273, over 1613242.12 frames. ], batch size: 26, lr: 1.81e-02, grad_scale: 8.0 +2023-02-05 23:27:43,613 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.222e+02 3.372e+02 4.170e+02 5.520e+02 1.384e+03, threshold=8.341e+02, percent-clipped=6.0 +2023-02-05 23:28:00,679 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28447.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:28:02,480 INFO [train.py:901] (3/4) Epoch 4, batch 4200, loss[loss=0.2522, simple_loss=0.3203, pruned_loss=0.09208, over 8090.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3649, pruned_loss=0.1268, over 1613346.80 frames. ], batch size: 21, lr: 1.81e-02, grad_scale: 8.0 +2023-02-05 23:28:07,667 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-05 23:28:17,490 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28472.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:28:18,117 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1389, 1.4037, 1.5151, 1.1972, 1.5005, 1.3942, 1.5695, 1.5431], + device='cuda:3'), covar=tensor([0.0751, 0.1351, 0.1928, 0.1699, 0.0719, 0.1625, 0.0949, 0.0649], + device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0199, 0.0236, 0.0201, 0.0156, 0.0201, 0.0163, 0.0168], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0007, 0.0007, 0.0005, 0.0007, 0.0006, 0.0005], + device='cuda:3') +2023-02-05 23:28:29,077 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-05 23:28:36,444 INFO [train.py:901] (3/4) Epoch 4, batch 4250, loss[loss=0.277, simple_loss=0.3356, pruned_loss=0.1092, over 7657.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3663, pruned_loss=0.1284, over 1618203.68 frames. ], batch size: 19, lr: 1.81e-02, grad_scale: 8.0 +2023-02-05 23:28:39,208 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28504.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:28:43,325 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28510.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:28:51,866 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.723e+02 3.170e+02 4.105e+02 5.662e+02 1.430e+03, threshold=8.210e+02, percent-clipped=9.0 +2023-02-05 23:28:57,105 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-05 23:29:10,383 INFO [train.py:901] (3/4) Epoch 4, batch 4300, loss[loss=0.3491, simple_loss=0.3903, pruned_loss=0.1539, over 7161.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3679, pruned_loss=0.1298, over 1621229.20 frames. ], batch size: 71, lr: 1.81e-02, grad_scale: 8.0 +2023-02-05 23:29:34,048 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 +2023-02-05 23:29:38,464 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28591.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:29:45,232 INFO [train.py:901] (3/4) Epoch 4, batch 4350, loss[loss=0.4299, simple_loss=0.446, pruned_loss=0.2069, over 6992.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3676, pruned_loss=0.1298, over 1617406.85 frames. ], batch size: 71, lr: 1.81e-02, grad_scale: 8.0 +2023-02-05 23:29:57,581 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28617.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:29:58,773 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-05 23:30:01,436 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.096e+02 3.285e+02 3.917e+02 4.771e+02 1.131e+03, threshold=7.833e+02, percent-clipped=1.0 +2023-02-05 23:30:19,076 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28649.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:30:19,682 INFO [train.py:901] (3/4) Epoch 4, batch 4400, loss[loss=0.3425, simple_loss=0.388, pruned_loss=0.1485, over 8481.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3668, pruned_loss=0.1287, over 1615119.75 frames. ], batch size: 27, lr: 1.81e-02, grad_scale: 8.0 +2023-02-05 23:30:37,240 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1810, 2.1122, 1.4548, 2.0548, 1.7418, 1.0987, 1.5348, 1.9627], + device='cuda:3'), covar=tensor([0.1014, 0.0517, 0.1019, 0.0448, 0.0662, 0.1351, 0.0891, 0.0563], + device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0235, 0.0303, 0.0308, 0.0324, 0.0307, 0.0336, 0.0309], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-05 23:30:41,088 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-05 23:30:46,448 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.5786, 5.6508, 5.0249, 2.0606, 4.9308, 5.3062, 5.3127, 4.7235], + device='cuda:3'), covar=tensor([0.0559, 0.0341, 0.0670, 0.4240, 0.0534, 0.0411, 0.0857, 0.0434], + device='cuda:3'), in_proj_covar=tensor([0.0366, 0.0268, 0.0296, 0.0388, 0.0293, 0.0239, 0.0282, 0.0222], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:3') +2023-02-05 23:30:54,262 INFO [train.py:901] (3/4) Epoch 4, batch 4450, loss[loss=0.3361, simple_loss=0.379, pruned_loss=0.1466, over 7809.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3653, pruned_loss=0.128, over 1615726.45 frames. ], batch size: 20, lr: 1.80e-02, grad_scale: 8.0 +2023-02-05 23:30:58,482 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28706.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:31:02,176 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 +2023-02-05 23:31:04,243 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.64 vs. limit=5.0 +2023-02-05 23:31:09,129 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.93 vs. limit=2.0 +2023-02-05 23:31:10,734 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.181e+02 3.229e+02 4.056e+02 4.786e+02 8.259e+02, threshold=8.113e+02, percent-clipped=1.0 +2023-02-05 23:31:17,728 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28732.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:31:30,206 INFO [train.py:901] (3/4) Epoch 4, batch 4500, loss[loss=0.2798, simple_loss=0.3431, pruned_loss=0.1083, over 7803.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3635, pruned_loss=0.1263, over 1615138.66 frames. ], batch size: 20, lr: 1.80e-02, grad_scale: 8.0 +2023-02-05 23:31:36,221 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-05 23:31:39,884 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28764.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:32:05,459 INFO [train.py:901] (3/4) Epoch 4, batch 4550, loss[loss=0.2667, simple_loss=0.3342, pruned_loss=0.09962, over 8140.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3638, pruned_loss=0.1265, over 1615221.32 frames. ], batch size: 22, lr: 1.80e-02, grad_scale: 8.0 +2023-02-05 23:32:21,344 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.937e+02 3.133e+02 4.046e+02 5.517e+02 1.256e+03, threshold=8.093e+02, percent-clipped=3.0 +2023-02-05 23:32:37,064 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4895, 1.7289, 2.7297, 1.1749, 1.9256, 1.6877, 1.5086, 1.7199], + device='cuda:3'), covar=tensor([0.1201, 0.1469, 0.0491, 0.2516, 0.1119, 0.1840, 0.1157, 0.1615], + device='cuda:3'), in_proj_covar=tensor([0.0452, 0.0426, 0.0503, 0.0515, 0.0557, 0.0497, 0.0438, 0.0566], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 23:32:39,629 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28848.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:32:40,907 INFO [train.py:901] (3/4) Epoch 4, batch 4600, loss[loss=0.3479, simple_loss=0.3958, pruned_loss=0.15, over 8251.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3618, pruned_loss=0.1252, over 1616626.26 frames. ], batch size: 24, lr: 1.80e-02, grad_scale: 8.0 +2023-02-05 23:32:43,605 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28854.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:33:00,447 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28879.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:33:05,969 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-05 23:33:14,778 INFO [train.py:901] (3/4) Epoch 4, batch 4650, loss[loss=0.3326, simple_loss=0.3792, pruned_loss=0.143, over 8561.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3616, pruned_loss=0.1257, over 1609052.11 frames. ], batch size: 49, lr: 1.80e-02, grad_scale: 8.0 +2023-02-05 23:33:30,676 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.722e+02 3.425e+02 4.570e+02 5.631e+02 1.457e+03, threshold=9.141e+02, percent-clipped=7.0 +2023-02-05 23:33:49,339 INFO [train.py:901] (3/4) Epoch 4, batch 4700, loss[loss=0.3212, simple_loss=0.372, pruned_loss=0.1352, over 8248.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3628, pruned_loss=0.126, over 1614466.17 frames. ], batch size: 24, lr: 1.80e-02, grad_scale: 8.0 +2023-02-05 23:33:58,483 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28962.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:33:59,157 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28963.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:34:03,891 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28969.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:34:15,893 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28987.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:34:16,547 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28988.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:34:24,384 INFO [train.py:901] (3/4) Epoch 4, batch 4750, loss[loss=0.2987, simple_loss=0.3405, pruned_loss=0.1285, over 7704.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3618, pruned_loss=0.1259, over 1606674.88 frames. ], batch size: 18, lr: 1.80e-02, grad_scale: 8.0 +2023-02-05 23:34:33,284 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29013.0, num_to_drop=1, layers_to_drop={1} +2023-02-05 23:34:38,668 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29020.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:34:38,749 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29020.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:34:40,432 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.078e+02 3.145e+02 3.754e+02 5.040e+02 8.107e+02, threshold=7.508e+02, percent-clipped=0.0 +2023-02-05 23:34:40,462 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-05 23:34:42,472 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-05 23:34:56,219 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29045.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:34:59,308 INFO [train.py:901] (3/4) Epoch 4, batch 4800, loss[loss=0.3806, simple_loss=0.4144, pruned_loss=0.1734, over 8512.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3621, pruned_loss=0.1259, over 1608793.56 frames. ], batch size: 39, lr: 1.79e-02, grad_scale: 8.0 +2023-02-05 23:35:34,008 INFO [train.py:901] (3/4) Epoch 4, batch 4850, loss[loss=0.2929, simple_loss=0.3372, pruned_loss=0.1243, over 7705.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3602, pruned_loss=0.1249, over 1606752.75 frames. ], batch size: 18, lr: 1.79e-02, grad_scale: 8.0 +2023-02-05 23:35:34,022 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-05 23:35:49,592 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.099e+02 3.374e+02 4.405e+02 6.016e+02 1.134e+03, threshold=8.810e+02, percent-clipped=7.0 +2023-02-05 23:35:51,361 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 +2023-02-05 23:36:07,826 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.9817, 1.2306, 4.1137, 1.5108, 3.5079, 3.3324, 3.6963, 3.5860], + device='cuda:3'), covar=tensor([0.0440, 0.3304, 0.0406, 0.2196, 0.1006, 0.0633, 0.0411, 0.0509], + device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0457, 0.0349, 0.0368, 0.0433, 0.0367, 0.0354, 0.0395], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-05 23:36:08,348 INFO [train.py:901] (3/4) Epoch 4, batch 4900, loss[loss=0.3625, simple_loss=0.4015, pruned_loss=0.1618, over 8023.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3598, pruned_loss=0.1244, over 1605444.24 frames. ], batch size: 22, lr: 1.79e-02, grad_scale: 8.0 +2023-02-05 23:36:41,956 INFO [train.py:901] (3/4) Epoch 4, batch 4950, loss[loss=0.3545, simple_loss=0.3999, pruned_loss=0.1546, over 7646.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3622, pruned_loss=0.1263, over 1607548.99 frames. ], batch size: 19, lr: 1.79e-02, grad_scale: 8.0 +2023-02-05 23:36:56,262 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29219.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:36:58,772 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.095e+02 3.208e+02 3.912e+02 5.596e+02 9.849e+02, threshold=7.824e+02, percent-clipped=2.0 +2023-02-05 23:36:58,867 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29223.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:37:00,346 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29225.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:37:12,971 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29244.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:37:16,652 INFO [train.py:901] (3/4) Epoch 4, batch 5000, loss[loss=0.2993, simple_loss=0.3666, pruned_loss=0.116, over 8539.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3612, pruned_loss=0.1253, over 1609444.07 frames. ], batch size: 31, lr: 1.79e-02, grad_scale: 8.0 +2023-02-05 23:37:16,877 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29250.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:37:19,516 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29254.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:37:51,636 INFO [train.py:901] (3/4) Epoch 4, batch 5050, loss[loss=0.3298, simple_loss=0.3862, pruned_loss=0.1367, over 8742.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3619, pruned_loss=0.1258, over 1610130.59 frames. ], batch size: 30, lr: 1.79e-02, grad_scale: 8.0 +2023-02-05 23:38:07,700 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.752e+02 3.436e+02 4.072e+02 5.001e+02 1.022e+03, threshold=8.144e+02, percent-clipped=3.0 +2023-02-05 23:38:14,945 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-05 23:38:18,451 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29338.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:38:20,957 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=5.56 vs. limit=5.0 +2023-02-05 23:38:26,650 INFO [train.py:901] (3/4) Epoch 4, batch 5100, loss[loss=0.3033, simple_loss=0.3684, pruned_loss=0.1192, over 8454.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.362, pruned_loss=0.1262, over 1608301.42 frames. ], batch size: 27, lr: 1.79e-02, grad_scale: 8.0 +2023-02-05 23:38:36,226 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29364.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:38:51,190 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6310, 2.3799, 4.7886, 1.1391, 2.9906, 2.2842, 1.6674, 2.4108], + device='cuda:3'), covar=tensor([0.1372, 0.1539, 0.0458, 0.2871, 0.1324, 0.1908, 0.1272, 0.2361], + device='cuda:3'), in_proj_covar=tensor([0.0454, 0.0434, 0.0516, 0.0520, 0.0561, 0.0497, 0.0440, 0.0575], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 23:39:00,479 INFO [train.py:901] (3/4) Epoch 4, batch 5150, loss[loss=0.3389, simple_loss=0.3917, pruned_loss=0.1431, over 8659.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3619, pruned_loss=0.1257, over 1611154.77 frames. ], batch size: 49, lr: 1.78e-02, grad_scale: 8.0 +2023-02-05 23:39:16,241 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.119e+02 3.145e+02 3.888e+02 4.871e+02 1.199e+03, threshold=7.777e+02, percent-clipped=1.0 +2023-02-05 23:39:35,367 INFO [train.py:901] (3/4) Epoch 4, batch 5200, loss[loss=0.3042, simple_loss=0.3654, pruned_loss=0.1215, over 8508.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3629, pruned_loss=0.1264, over 1609592.47 frames. ], batch size: 26, lr: 1.78e-02, grad_scale: 8.0 +2023-02-05 23:39:37,637 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0423, 2.4737, 2.8724, 1.0271, 2.7296, 1.8254, 1.4894, 1.6565], + device='cuda:3'), covar=tensor([0.0266, 0.0161, 0.0096, 0.0228, 0.0178, 0.0321, 0.0290, 0.0153], + device='cuda:3'), in_proj_covar=tensor([0.0278, 0.0201, 0.0160, 0.0237, 0.0191, 0.0325, 0.0262, 0.0221], + device='cuda:3'), out_proj_covar=tensor([1.1113e-04, 8.0421e-05, 6.0837e-05, 9.1653e-05, 7.6934e-05, 1.3990e-04, + 1.0604e-04, 8.6328e-05], device='cuda:3') +2023-02-05 23:39:45,745 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.97 vs. limit=5.0 +2023-02-05 23:39:54,959 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29479.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:40:09,488 INFO [train.py:901] (3/4) Epoch 4, batch 5250, loss[loss=0.2867, simple_loss=0.3391, pruned_loss=0.1172, over 7797.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.362, pruned_loss=0.1264, over 1606190.57 frames. ], batch size: 19, lr: 1.78e-02, grad_scale: 4.0 +2023-02-05 23:40:12,206 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-05 23:40:25,988 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.346e+02 3.507e+02 4.371e+02 5.555e+02 1.318e+03, threshold=8.742e+02, percent-clipped=11.0 +2023-02-05 23:40:43,419 INFO [train.py:901] (3/4) Epoch 4, batch 5300, loss[loss=0.3814, simple_loss=0.4242, pruned_loss=0.1693, over 8361.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3614, pruned_loss=0.1255, over 1610314.98 frames. ], batch size: 24, lr: 1.78e-02, grad_scale: 4.0 +2023-02-05 23:40:45,984 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 +2023-02-05 23:40:57,663 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29569.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:41:14,745 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29594.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:41:17,292 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29598.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:41:18,613 INFO [train.py:901] (3/4) Epoch 4, batch 5350, loss[loss=0.3874, simple_loss=0.4009, pruned_loss=0.1869, over 6523.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3601, pruned_loss=0.124, over 1609437.86 frames. ], batch size: 71, lr: 1.78e-02, grad_scale: 4.0 +2023-02-05 23:41:32,366 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29619.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:41:35,509 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.071e+02 3.127e+02 4.006e+02 4.952e+02 2.682e+03, threshold=8.012e+02, percent-clipped=7.0 +2023-02-05 23:41:53,618 INFO [train.py:901] (3/4) Epoch 4, batch 5400, loss[loss=0.3584, simple_loss=0.3887, pruned_loss=0.164, over 6859.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3619, pruned_loss=0.1251, over 1611946.68 frames. ], batch size: 72, lr: 1.78e-02, grad_scale: 4.0 +2023-02-05 23:42:14,536 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29680.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:42:28,598 INFO [train.py:901] (3/4) Epoch 4, batch 5450, loss[loss=0.287, simple_loss=0.3589, pruned_loss=0.1075, over 8286.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3614, pruned_loss=0.1246, over 1610687.87 frames. ], batch size: 23, lr: 1.78e-02, grad_scale: 4.0 +2023-02-05 23:42:37,306 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29713.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:42:41,415 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.08 vs. limit=2.0 +2023-02-05 23:42:44,939 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.180e+02 3.089e+02 4.007e+02 5.016e+02 9.074e+02, threshold=8.014e+02, percent-clipped=4.0 +2023-02-05 23:42:52,700 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29735.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:42:57,989 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-05 23:43:02,818 INFO [train.py:901] (3/4) Epoch 4, batch 5500, loss[loss=0.2746, simple_loss=0.3461, pruned_loss=0.1015, over 8023.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3621, pruned_loss=0.1241, over 1612840.77 frames. ], batch size: 22, lr: 1.77e-02, grad_scale: 4.0 +2023-02-05 23:43:10,326 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29760.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:43:38,374 INFO [train.py:901] (3/4) Epoch 4, batch 5550, loss[loss=0.4029, simple_loss=0.4383, pruned_loss=0.1838, over 8661.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3628, pruned_loss=0.1244, over 1618262.58 frames. ], batch size: 39, lr: 1.77e-02, grad_scale: 4.0 +2023-02-05 23:43:51,782 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29820.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:43:53,967 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 +2023-02-05 23:43:54,237 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.923e+02 3.211e+02 3.931e+02 4.808e+02 9.688e+02, threshold=7.861e+02, percent-clipped=2.0 +2023-02-05 23:44:12,161 INFO [train.py:901] (3/4) Epoch 4, batch 5600, loss[loss=0.392, simple_loss=0.4207, pruned_loss=0.1817, over 8646.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3607, pruned_loss=0.1235, over 1615031.22 frames. ], batch size: 34, lr: 1.77e-02, grad_scale: 8.0 +2023-02-05 23:44:17,401 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-05 23:44:46,061 INFO [train.py:901] (3/4) Epoch 4, batch 5650, loss[loss=0.3773, simple_loss=0.3817, pruned_loss=0.1864, over 7801.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3625, pruned_loss=0.1255, over 1613962.77 frames. ], batch size: 19, lr: 1.77e-02, grad_scale: 8.0 +2023-02-05 23:44:55,376 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29913.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:45:03,295 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.126e+02 3.236e+02 4.025e+02 5.119e+02 8.732e+02, threshold=8.050e+02, percent-clipped=2.0 +2023-02-05 23:45:03,329 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-05 23:45:20,778 INFO [train.py:901] (3/4) Epoch 4, batch 5700, loss[loss=0.3125, simple_loss=0.3742, pruned_loss=0.1254, over 8463.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3635, pruned_loss=0.1259, over 1617225.07 frames. ], batch size: 49, lr: 1.77e-02, grad_scale: 8.0 +2023-02-05 23:45:34,487 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29969.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:45:50,023 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.5189, 2.3163, 1.9057, 1.8723, 1.7989, 2.0303, 2.5240, 2.0702], + device='cuda:3'), covar=tensor([0.0562, 0.1146, 0.1761, 0.1351, 0.0666, 0.1418, 0.0722, 0.0600], + device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0193, 0.0229, 0.0194, 0.0151, 0.0198, 0.0159, 0.0161], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0007, 0.0007, 0.0005, 0.0007, 0.0006, 0.0005], + device='cuda:3') +2023-02-05 23:45:52,088 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29994.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:45:55,955 INFO [train.py:901] (3/4) Epoch 4, batch 5750, loss[loss=0.3177, simple_loss=0.374, pruned_loss=0.1307, over 8357.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3614, pruned_loss=0.1245, over 1612424.29 frames. ], batch size: 24, lr: 1.77e-02, grad_scale: 8.0 +2023-02-05 23:46:00,415 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0286, 1.5823, 3.2102, 1.3654, 2.1980, 3.7198, 3.4747, 3.1112], + device='cuda:3'), covar=tensor([0.1082, 0.1532, 0.0420, 0.2125, 0.0847, 0.0245, 0.0376, 0.0646], + device='cuda:3'), in_proj_covar=tensor([0.0227, 0.0266, 0.0215, 0.0266, 0.0220, 0.0193, 0.0200, 0.0264], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 23:46:07,151 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-05 23:46:13,262 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.352e+02 3.278e+02 4.024e+02 4.787e+02 1.009e+03, threshold=8.047e+02, percent-clipped=4.0 +2023-02-05 23:46:13,360 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30024.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:46:17,490 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30028.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:46:32,295 INFO [train.py:901] (3/4) Epoch 4, batch 5800, loss[loss=0.2858, simple_loss=0.3489, pruned_loss=0.1114, over 8109.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3603, pruned_loss=0.1233, over 1613005.15 frames. ], batch size: 23, lr: 1.77e-02, grad_scale: 8.0 +2023-02-05 23:47:06,538 INFO [train.py:901] (3/4) Epoch 4, batch 5850, loss[loss=0.2589, simple_loss=0.3344, pruned_loss=0.09166, over 8025.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3606, pruned_loss=0.1235, over 1610144.79 frames. ], batch size: 22, lr: 1.76e-02, grad_scale: 8.0 +2023-02-05 23:47:23,096 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.818e+02 3.427e+02 4.657e+02 5.932e+02 9.223e+02, threshold=9.314e+02, percent-clipped=4.0 +2023-02-05 23:47:33,289 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30139.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:47:35,878 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30143.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:47:36,576 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2364, 1.7151, 1.6444, 0.5433, 1.6907, 1.2643, 0.1986, 1.5235], + device='cuda:3'), covar=tensor([0.0152, 0.0089, 0.0091, 0.0140, 0.0095, 0.0260, 0.0231, 0.0070], + device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0199, 0.0168, 0.0243, 0.0194, 0.0326, 0.0266, 0.0229], + device='cuda:3'), out_proj_covar=tensor([1.1371e-04, 7.8258e-05, 6.3216e-05, 9.2664e-05, 7.6986e-05, 1.3782e-04, + 1.0688e-04, 8.9229e-05], device='cuda:3') +2023-02-05 23:47:41,684 INFO [train.py:901] (3/4) Epoch 4, batch 5900, loss[loss=0.291, simple_loss=0.3573, pruned_loss=0.1123, over 8367.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3603, pruned_loss=0.1237, over 1610570.56 frames. ], batch size: 24, lr: 1.76e-02, grad_scale: 8.0 +2023-02-05 23:47:43,475 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.83 vs. limit=5.0 +2023-02-05 23:47:51,359 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30164.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:48:15,970 INFO [train.py:901] (3/4) Epoch 4, batch 5950, loss[loss=0.3297, simple_loss=0.3941, pruned_loss=0.1327, over 8107.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3598, pruned_loss=0.1234, over 1612391.34 frames. ], batch size: 23, lr: 1.76e-02, grad_scale: 8.0 +2023-02-05 23:48:32,438 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.869e+02 3.143e+02 3.968e+02 4.977e+02 1.070e+03, threshold=7.937e+02, percent-clipped=1.0 +2023-02-05 23:48:35,349 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30227.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:48:50,185 INFO [train.py:901] (3/4) Epoch 4, batch 6000, loss[loss=0.2978, simple_loss=0.3593, pruned_loss=0.1181, over 8243.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3598, pruned_loss=0.1233, over 1617176.23 frames. ], batch size: 22, lr: 1.76e-02, grad_scale: 8.0 +2023-02-05 23:48:50,185 INFO [train.py:926] (3/4) Computing validation loss +2023-02-05 23:49:02,857 INFO [train.py:935] (3/4) Epoch 4, validation: loss=0.2338, simple_loss=0.3275, pruned_loss=0.07005, over 944034.00 frames. +2023-02-05 23:49:02,858 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-05 23:49:22,554 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30279.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:49:26,599 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30284.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:49:37,708 INFO [train.py:901] (3/4) Epoch 4, batch 6050, loss[loss=0.3202, simple_loss=0.3689, pruned_loss=0.1357, over 8461.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3592, pruned_loss=0.123, over 1613205.13 frames. ], batch size: 29, lr: 1.76e-02, grad_scale: 8.0 +2023-02-05 23:49:44,063 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30309.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:49:53,935 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.902e+02 3.338e+02 3.992e+02 4.649e+02 1.183e+03, threshold=7.984e+02, percent-clipped=3.0 +2023-02-05 23:50:12,466 INFO [train.py:901] (3/4) Epoch 4, batch 6100, loss[loss=0.3141, simple_loss=0.3801, pruned_loss=0.124, over 8252.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3581, pruned_loss=0.1218, over 1615781.08 frames. ], batch size: 24, lr: 1.76e-02, grad_scale: 8.0 +2023-02-05 23:50:32,439 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30378.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:50:37,425 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3282, 1.6184, 2.2982, 1.0406, 1.6365, 1.5790, 1.3505, 1.3384], + device='cuda:3'), covar=tensor([0.1448, 0.1503, 0.0684, 0.2865, 0.1247, 0.2148, 0.1381, 0.1772], + device='cuda:3'), in_proj_covar=tensor([0.0459, 0.0431, 0.0509, 0.0525, 0.0560, 0.0499, 0.0441, 0.0571], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 23:50:39,272 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-05 23:50:44,140 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30395.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:50:47,302 INFO [train.py:901] (3/4) Epoch 4, batch 6150, loss[loss=0.3598, simple_loss=0.4059, pruned_loss=0.1568, over 7967.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3584, pruned_loss=0.1222, over 1615897.17 frames. ], batch size: 21, lr: 1.76e-02, grad_scale: 8.0 +2023-02-05 23:50:47,648 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1988, 1.7190, 1.2616, 1.6874, 1.2944, 1.0627, 1.3212, 1.4274], + device='cuda:3'), covar=tensor([0.0819, 0.0370, 0.0952, 0.0465, 0.0622, 0.1126, 0.0717, 0.0639], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0234, 0.0305, 0.0302, 0.0329, 0.0312, 0.0335, 0.0309], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-05 23:51:02,480 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30420.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:51:05,074 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.693e+02 3.511e+02 4.267e+02 5.249e+02 1.089e+03, threshold=8.535e+02, percent-clipped=6.0 +2023-02-05 23:51:23,139 INFO [train.py:901] (3/4) Epoch 4, batch 6200, loss[loss=0.2523, simple_loss=0.2979, pruned_loss=0.1033, over 7442.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3587, pruned_loss=0.1227, over 1613474.19 frames. ], batch size: 17, lr: 1.75e-02, grad_scale: 8.0 +2023-02-05 23:51:36,654 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2985, 1.9790, 1.9296, 2.0745, 2.1256, 2.0907, 2.6598, 2.1036], + device='cuda:3'), covar=tensor([0.0521, 0.1155, 0.1641, 0.1192, 0.0535, 0.1353, 0.0601, 0.0542], + device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0194, 0.0230, 0.0196, 0.0149, 0.0197, 0.0159, 0.0161], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0007, 0.0007, 0.0005, 0.0007, 0.0006, 0.0005], + device='cuda:3') +2023-02-05 23:51:48,271 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30487.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:51:57,470 INFO [train.py:901] (3/4) Epoch 4, batch 6250, loss[loss=0.3271, simple_loss=0.393, pruned_loss=0.1306, over 8320.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3583, pruned_loss=0.1229, over 1613151.98 frames. ], batch size: 25, lr: 1.75e-02, grad_scale: 8.0 +2023-02-05 23:52:04,992 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.6466, 5.6346, 4.8819, 2.2053, 4.8850, 5.2435, 5.2368, 4.7391], + device='cuda:3'), covar=tensor([0.0555, 0.0359, 0.0687, 0.4160, 0.0590, 0.0499, 0.0877, 0.0435], + device='cuda:3'), in_proj_covar=tensor([0.0367, 0.0267, 0.0299, 0.0375, 0.0295, 0.0246, 0.0279, 0.0222], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:3') +2023-02-05 23:52:14,465 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.635e+02 3.291e+02 3.933e+02 5.014e+02 1.132e+03, threshold=7.866e+02, percent-clipped=4.0 +2023-02-05 23:52:22,914 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30535.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:52:32,704 INFO [train.py:901] (3/4) Epoch 4, batch 6300, loss[loss=0.2284, simple_loss=0.2853, pruned_loss=0.08572, over 7710.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3584, pruned_loss=0.1227, over 1612374.92 frames. ], batch size: 18, lr: 1.75e-02, grad_scale: 8.0 +2023-02-05 23:52:39,498 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30560.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:52:44,917 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 +2023-02-05 23:52:47,317 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30571.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:53:06,768 INFO [train.py:901] (3/4) Epoch 4, batch 6350, loss[loss=0.3048, simple_loss=0.3796, pruned_loss=0.115, over 8466.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3597, pruned_loss=0.1237, over 1613132.14 frames. ], batch size: 29, lr: 1.75e-02, grad_scale: 8.0 +2023-02-05 23:53:08,348 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30602.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:53:23,782 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.916e+02 3.165e+02 3.849e+02 5.077e+02 1.430e+03, threshold=7.697e+02, percent-clipped=4.0 +2023-02-05 23:53:42,470 INFO [train.py:901] (3/4) Epoch 4, batch 6400, loss[loss=0.2799, simple_loss=0.3468, pruned_loss=0.1065, over 8034.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3589, pruned_loss=0.1229, over 1610032.71 frames. ], batch size: 22, lr: 1.75e-02, grad_scale: 8.0 +2023-02-05 23:53:58,109 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30673.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:54:07,701 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30686.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:54:13,776 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2506, 1.8218, 2.9970, 2.3846, 2.4780, 1.8387, 1.3665, 1.2351], + device='cuda:3'), covar=tensor([0.1444, 0.1709, 0.0337, 0.0771, 0.0716, 0.0853, 0.0863, 0.1506], + device='cuda:3'), in_proj_covar=tensor([0.0703, 0.0634, 0.0539, 0.0612, 0.0718, 0.0592, 0.0576, 0.0583], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 23:54:16,870 INFO [train.py:901] (3/4) Epoch 4, batch 6450, loss[loss=0.2583, simple_loss=0.3175, pruned_loss=0.09953, over 7793.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3577, pruned_loss=0.122, over 1609193.41 frames. ], batch size: 19, lr: 1.75e-02, grad_scale: 8.0 +2023-02-05 23:54:31,544 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30722.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:54:32,820 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.794e+02 3.022e+02 3.987e+02 5.645e+02 1.412e+03, threshold=7.975e+02, percent-clipped=10.0 +2023-02-05 23:54:42,853 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-05 23:54:50,947 INFO [train.py:901] (3/4) Epoch 4, batch 6500, loss[loss=0.315, simple_loss=0.3661, pruned_loss=0.132, over 7649.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.358, pruned_loss=0.1223, over 1609657.31 frames. ], batch size: 19, lr: 1.75e-02, grad_scale: 8.0 +2023-02-05 23:55:26,183 INFO [train.py:901] (3/4) Epoch 4, batch 6550, loss[loss=0.2509, simple_loss=0.3157, pruned_loss=0.09303, over 7710.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3595, pruned_loss=0.1231, over 1611568.58 frames. ], batch size: 18, lr: 1.74e-02, grad_scale: 8.0 +2023-02-05 23:55:42,629 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.464e+02 3.539e+02 4.251e+02 5.114e+02 1.135e+03, threshold=8.501e+02, percent-clipped=1.0 +2023-02-05 23:55:50,027 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-05 23:55:51,484 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30837.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:56:00,628 INFO [train.py:901] (3/4) Epoch 4, batch 6600, loss[loss=0.2792, simple_loss=0.3277, pruned_loss=0.1154, over 7653.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3591, pruned_loss=0.1225, over 1615358.13 frames. ], batch size: 19, lr: 1.74e-02, grad_scale: 8.0 +2023-02-05 23:56:06,240 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30858.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:56:08,692 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-05 23:56:24,230 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30883.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:56:35,334 INFO [train.py:901] (3/4) Epoch 4, batch 6650, loss[loss=0.2913, simple_loss=0.3579, pruned_loss=0.1123, over 8106.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3604, pruned_loss=0.1233, over 1611540.10 frames. ], batch size: 23, lr: 1.74e-02, grad_scale: 8.0 +2023-02-05 23:56:50,088 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30921.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:56:51,879 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.853e+02 3.362e+02 4.352e+02 5.461e+02 1.446e+03, threshold=8.703e+02, percent-clipped=3.0 +2023-02-05 23:57:04,055 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30942.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:57:04,679 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30943.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:57:08,262 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.27 vs. limit=5.0 +2023-02-05 23:57:09,173 INFO [train.py:901] (3/4) Epoch 4, batch 6700, loss[loss=0.3203, simple_loss=0.3752, pruned_loss=0.1327, over 8475.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3605, pruned_loss=0.1235, over 1614579.22 frames. ], batch size: 25, lr: 1.74e-02, grad_scale: 8.0 +2023-02-05 23:57:10,717 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30952.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:57:12,267 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 +2023-02-05 23:57:21,648 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30967.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:57:41,914 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6848, 2.8838, 1.9430, 2.1478, 2.2348, 1.4546, 2.0308, 2.2620], + device='cuda:3'), covar=tensor([0.1091, 0.0274, 0.0772, 0.0580, 0.0555, 0.1118, 0.0835, 0.0664], + device='cuda:3'), in_proj_covar=tensor([0.0338, 0.0237, 0.0302, 0.0300, 0.0316, 0.0306, 0.0330, 0.0304], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-05 23:57:45,060 INFO [train.py:901] (3/4) Epoch 4, batch 6750, loss[loss=0.3239, simple_loss=0.3672, pruned_loss=0.1403, over 7807.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3604, pruned_loss=0.1238, over 1612371.68 frames. ], batch size: 20, lr: 1.74e-02, grad_scale: 8.0 +2023-02-05 23:57:56,390 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31017.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:58:00,825 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.225e+02 3.317e+02 4.136e+02 5.252e+02 1.678e+03, threshold=8.272e+02, percent-clipped=4.0 +2023-02-05 23:58:04,868 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31029.0, num_to_drop=1, layers_to_drop={0} +2023-02-05 23:58:14,342 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3177, 1.5717, 1.3541, 1.9154, 0.8258, 1.1864, 1.2941, 1.5260], + device='cuda:3'), covar=tensor([0.1321, 0.1318, 0.1580, 0.0699, 0.1846, 0.2301, 0.1457, 0.1230], + device='cuda:3'), in_proj_covar=tensor([0.0275, 0.0286, 0.0298, 0.0222, 0.0272, 0.0295, 0.0308, 0.0275], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-05 23:58:18,933 INFO [train.py:901] (3/4) Epoch 4, batch 6800, loss[loss=0.3296, simple_loss=0.3947, pruned_loss=0.1323, over 8256.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3619, pruned_loss=0.1245, over 1614717.47 frames. ], batch size: 24, lr: 1.74e-02, grad_scale: 8.0 +2023-02-05 23:58:19,594 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-05 23:58:48,703 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31093.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:58:53,202 INFO [train.py:901] (3/4) Epoch 4, batch 6850, loss[loss=0.3075, simple_loss=0.3638, pruned_loss=0.1256, over 8204.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3611, pruned_loss=0.1243, over 1610012.64 frames. ], batch size: 23, lr: 1.74e-02, grad_scale: 8.0 +2023-02-05 23:59:00,788 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1635, 1.5622, 4.3157, 1.8064, 2.3004, 4.7807, 4.4790, 4.1562], + device='cuda:3'), covar=tensor([0.1195, 0.1597, 0.0316, 0.1970, 0.0858, 0.0232, 0.0376, 0.0529], + device='cuda:3'), in_proj_covar=tensor([0.0227, 0.0255, 0.0211, 0.0257, 0.0209, 0.0192, 0.0203, 0.0259], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-05 23:59:02,901 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1958, 1.3632, 2.3494, 1.0081, 1.9802, 1.4099, 1.2015, 1.7470], + device='cuda:3'), covar=tensor([0.1401, 0.1625, 0.0501, 0.2592, 0.0873, 0.1895, 0.1394, 0.1319], + device='cuda:3'), in_proj_covar=tensor([0.0453, 0.0428, 0.0508, 0.0516, 0.0558, 0.0501, 0.0443, 0.0567], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 23:59:06,907 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31118.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:59:09,997 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-05 23:59:10,560 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.781e+02 3.178e+02 3.797e+02 5.313e+02 1.260e+03, threshold=7.594e+02, percent-clipped=4.0 +2023-02-05 23:59:13,367 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.7219, 3.6616, 3.3306, 1.6829, 3.2968, 3.1761, 3.4009, 2.9295], + device='cuda:3'), covar=tensor([0.1066, 0.0747, 0.0964, 0.4175, 0.0857, 0.0977, 0.1443, 0.0882], + device='cuda:3'), in_proj_covar=tensor([0.0381, 0.0273, 0.0293, 0.0387, 0.0296, 0.0253, 0.0289, 0.0229], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:3') +2023-02-05 23:59:16,202 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31132.0, num_to_drop=0, layers_to_drop=set() +2023-02-05 23:59:16,890 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5968, 3.0853, 2.4874, 4.0918, 1.8034, 1.7453, 2.2539, 3.2726], + device='cuda:3'), covar=tensor([0.1208, 0.1335, 0.1663, 0.0337, 0.2129, 0.2620, 0.2180, 0.1167], + device='cuda:3'), in_proj_covar=tensor([0.0275, 0.0285, 0.0300, 0.0224, 0.0272, 0.0296, 0.0309, 0.0277], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-05 23:59:23,600 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7232, 2.3329, 4.3295, 1.2747, 2.9965, 2.2951, 1.7508, 2.4267], + device='cuda:3'), covar=tensor([0.1203, 0.1589, 0.0506, 0.2611, 0.1195, 0.1835, 0.1194, 0.1923], + device='cuda:3'), in_proj_covar=tensor([0.0455, 0.0432, 0.0510, 0.0519, 0.0559, 0.0497, 0.0445, 0.0567], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-05 23:59:28,757 INFO [train.py:901] (3/4) Epoch 4, batch 6900, loss[loss=0.31, simple_loss=0.3707, pruned_loss=0.1247, over 8573.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3613, pruned_loss=0.124, over 1614501.64 frames. ], batch size: 34, lr: 1.73e-02, grad_scale: 8.0 +2023-02-06 00:00:03,327 INFO [train.py:901] (3/4) Epoch 4, batch 6950, loss[loss=0.3318, simple_loss=0.3903, pruned_loss=0.1366, over 8297.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3603, pruned_loss=0.1236, over 1612394.78 frames. ], batch size: 23, lr: 1.73e-02, grad_scale: 8.0 +2023-02-06 00:00:18,144 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 00:00:20,076 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.679e+02 3.425e+02 4.122e+02 5.302e+02 9.579e+02, threshold=8.244e+02, percent-clipped=6.0 +2023-02-06 00:00:25,014 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2839, 1.5656, 1.2291, 1.8890, 0.9368, 1.1082, 1.3925, 1.5584], + device='cuda:3'), covar=tensor([0.1489, 0.1481, 0.1824, 0.0821, 0.1873, 0.2629, 0.1365, 0.1179], + device='cuda:3'), in_proj_covar=tensor([0.0270, 0.0281, 0.0295, 0.0218, 0.0264, 0.0295, 0.0299, 0.0268], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 00:00:38,134 INFO [train.py:901] (3/4) Epoch 4, batch 7000, loss[loss=0.2452, simple_loss=0.308, pruned_loss=0.09124, over 7425.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3593, pruned_loss=0.1231, over 1609918.69 frames. ], batch size: 17, lr: 1.73e-02, grad_scale: 8.0 +2023-02-06 00:00:48,784 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31265.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:01:01,632 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2023-02-06 00:01:03,285 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31287.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:01:05,299 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3639, 4.4414, 3.9420, 1.9480, 3.9226, 4.0834, 4.0061, 3.5009], + device='cuda:3'), covar=tensor([0.0886, 0.0502, 0.0924, 0.4285, 0.0590, 0.0607, 0.1247, 0.0608], + device='cuda:3'), in_proj_covar=tensor([0.0378, 0.0269, 0.0296, 0.0391, 0.0295, 0.0250, 0.0288, 0.0228], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:3') +2023-02-06 00:01:09,191 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31296.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:01:11,657 INFO [train.py:901] (3/4) Epoch 4, batch 7050, loss[loss=0.3403, simple_loss=0.4006, pruned_loss=0.14, over 8623.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3594, pruned_loss=0.1235, over 1606961.62 frames. ], batch size: 31, lr: 1.73e-02, grad_scale: 8.0 +2023-02-06 00:01:21,134 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8167, 1.1529, 3.9182, 1.6038, 2.9106, 3.0555, 3.5272, 3.5438], + device='cuda:3'), covar=tensor([0.0662, 0.5148, 0.0927, 0.2926, 0.2498, 0.1412, 0.0816, 0.0838], + device='cuda:3'), in_proj_covar=tensor([0.0306, 0.0471, 0.0377, 0.0397, 0.0463, 0.0386, 0.0382, 0.0424], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 00:01:28,375 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.698e+02 3.221e+02 3.873e+02 5.326e+02 1.178e+03, threshold=7.746e+02, percent-clipped=8.0 +2023-02-06 00:01:47,448 INFO [train.py:901] (3/4) Epoch 4, batch 7100, loss[loss=0.2898, simple_loss=0.3439, pruned_loss=0.1178, over 7657.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3593, pruned_loss=0.1229, over 1606203.40 frames. ], batch size: 19, lr: 1.73e-02, grad_scale: 8.0 +2023-02-06 00:02:00,269 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 +2023-02-06 00:02:02,610 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31373.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 00:02:08,043 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31380.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:02:13,379 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31388.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:02:14,690 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8383, 2.5010, 3.1567, 0.8464, 3.0945, 2.2058, 1.2432, 1.7843], + device='cuda:3'), covar=tensor([0.0241, 0.0086, 0.0084, 0.0224, 0.0113, 0.0221, 0.0334, 0.0142], + device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0203, 0.0164, 0.0244, 0.0196, 0.0332, 0.0271, 0.0232], + device='cuda:3'), out_proj_covar=tensor([1.0952e-04, 7.8012e-05, 6.1065e-05, 9.1123e-05, 7.6527e-05, 1.3799e-04, + 1.0631e-04, 8.8663e-05], device='cuda:3') +2023-02-06 00:02:21,054 INFO [train.py:901] (3/4) Epoch 4, batch 7150, loss[loss=0.2609, simple_loss=0.328, pruned_loss=0.09694, over 7976.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3605, pruned_loss=0.1233, over 1608852.37 frames. ], batch size: 21, lr: 1.73e-02, grad_scale: 8.0 +2023-02-06 00:02:22,730 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31402.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:02:28,577 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31411.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:02:29,926 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31413.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:02:37,019 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.027e+02 3.187e+02 3.955e+02 5.000e+02 8.847e+02, threshold=7.910e+02, percent-clipped=2.0 +2023-02-06 00:02:39,208 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31427.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:02:56,011 INFO [train.py:901] (3/4) Epoch 4, batch 7200, loss[loss=0.3339, simple_loss=0.3693, pruned_loss=0.1492, over 8406.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3588, pruned_loss=0.1224, over 1604867.02 frames. ], batch size: 49, lr: 1.73e-02, grad_scale: 8.0 +2023-02-06 00:03:22,243 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31488.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 00:03:30,969 INFO [train.py:901] (3/4) Epoch 4, batch 7250, loss[loss=0.2942, simple_loss=0.3642, pruned_loss=0.1121, over 8466.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.358, pruned_loss=0.1218, over 1607261.32 frames. ], batch size: 25, lr: 1.73e-02, grad_scale: 16.0 +2023-02-06 00:03:37,363 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31509.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 00:03:47,339 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.140e+02 3.150e+02 3.858e+02 4.938e+02 9.845e+02, threshold=7.715e+02, percent-clipped=4.0 +2023-02-06 00:04:05,058 INFO [train.py:901] (3/4) Epoch 4, batch 7300, loss[loss=0.3163, simple_loss=0.3768, pruned_loss=0.1279, over 8335.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3578, pruned_loss=0.1216, over 1606929.34 frames. ], batch size: 25, lr: 1.72e-02, grad_scale: 16.0 +2023-02-06 00:04:40,426 INFO [train.py:901] (3/4) Epoch 4, batch 7350, loss[loss=0.2458, simple_loss=0.3119, pruned_loss=0.08989, over 8243.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3575, pruned_loss=0.1213, over 1610833.16 frames. ], batch size: 22, lr: 1.72e-02, grad_scale: 16.0 +2023-02-06 00:04:57,285 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.182e+02 2.774e+02 3.613e+02 4.483e+02 1.102e+03, threshold=7.227e+02, percent-clipped=2.0 +2023-02-06 00:04:59,980 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 00:05:00,991 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2023-02-06 00:05:05,421 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31636.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:05:14,569 INFO [train.py:901] (3/4) Epoch 4, batch 7400, loss[loss=0.275, simple_loss=0.3369, pruned_loss=0.1066, over 5983.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.359, pruned_loss=0.1216, over 1614778.03 frames. ], batch size: 13, lr: 1.72e-02, grad_scale: 16.0 +2023-02-06 00:05:19,274 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 00:05:20,113 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31658.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:05:21,990 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31661.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:05:26,693 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31667.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:05:37,009 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31683.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:05:43,513 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31692.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:05:48,578 INFO [train.py:901] (3/4) Epoch 4, batch 7450, loss[loss=0.2709, simple_loss=0.341, pruned_loss=0.1004, over 8360.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3613, pruned_loss=0.1232, over 1619363.46 frames. ], batch size: 24, lr: 1.72e-02, grad_scale: 16.0 +2023-02-06 00:05:50,973 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2023-02-06 00:05:58,006 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 00:06:04,224 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31722.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:06:05,436 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.933e+02 3.216e+02 3.933e+02 5.503e+02 1.387e+03, threshold=7.866e+02, percent-clipped=9.0 +2023-02-06 00:06:19,848 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31744.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 00:06:23,714 INFO [train.py:901] (3/4) Epoch 4, batch 7500, loss[loss=0.3812, simple_loss=0.411, pruned_loss=0.1757, over 8491.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3612, pruned_loss=0.1232, over 1620205.92 frames. ], batch size: 28, lr: 1.72e-02, grad_scale: 16.0 +2023-02-06 00:06:36,763 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31769.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 00:06:37,958 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31771.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:06:56,290 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6374, 2.9063, 2.0032, 2.1708, 2.2981, 1.6045, 2.1469, 2.3114], + device='cuda:3'), covar=tensor([0.1165, 0.0304, 0.0806, 0.0547, 0.0545, 0.1037, 0.0770, 0.0704], + device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0235, 0.0309, 0.0300, 0.0321, 0.0310, 0.0333, 0.0310], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 00:06:58,136 INFO [train.py:901] (3/4) Epoch 4, batch 7550, loss[loss=0.3991, simple_loss=0.423, pruned_loss=0.1876, over 8510.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.361, pruned_loss=0.1237, over 1618453.88 frames. ], batch size: 49, lr: 1.72e-02, grad_scale: 8.0 +2023-02-06 00:07:02,979 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31806.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:07:16,191 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.990e+02 2.842e+02 3.963e+02 5.244e+02 1.193e+03, threshold=7.926e+02, percent-clipped=8.0 +2023-02-06 00:07:21,433 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.25 vs. limit=5.0 +2023-02-06 00:07:27,230 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31841.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:07:34,061 INFO [train.py:901] (3/4) Epoch 4, batch 7600, loss[loss=0.3238, simple_loss=0.3748, pruned_loss=0.1365, over 8614.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3603, pruned_loss=0.1232, over 1621450.85 frames. ], batch size: 49, lr: 1.72e-02, grad_scale: 8.0 +2023-02-06 00:07:36,150 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31853.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 00:07:58,613 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31886.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:08:05,432 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31896.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:08:06,813 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9636, 3.2593, 2.5725, 4.2564, 1.7723, 2.0528, 2.0867, 3.5067], + device='cuda:3'), covar=tensor([0.0939, 0.1328, 0.1655, 0.0313, 0.1982, 0.2254, 0.2112, 0.1154], + device='cuda:3'), in_proj_covar=tensor([0.0274, 0.0276, 0.0299, 0.0225, 0.0263, 0.0287, 0.0298, 0.0270], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 00:08:07,996 INFO [train.py:901] (3/4) Epoch 4, batch 7650, loss[loss=0.3421, simple_loss=0.3889, pruned_loss=0.1476, over 8767.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3587, pruned_loss=0.1223, over 1617572.21 frames. ], batch size: 30, lr: 1.71e-02, grad_scale: 8.0 +2023-02-06 00:08:25,718 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6038, 2.5480, 2.9592, 2.2239, 1.4070, 2.7092, 0.5307, 1.8377], + device='cuda:3'), covar=tensor([0.2435, 0.2731, 0.1216, 0.2146, 0.5933, 0.0606, 0.7771, 0.1966], + device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0114, 0.0082, 0.0164, 0.0204, 0.0083, 0.0151, 0.0122], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 00:08:26,156 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.122e+02 3.181e+02 3.860e+02 4.828e+02 9.649e+02, threshold=7.720e+02, percent-clipped=2.0 +2023-02-06 00:08:31,519 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31933.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:08:31,633 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0421, 1.7266, 2.7436, 2.2151, 2.2720, 1.8609, 1.3943, 0.9220], + device='cuda:3'), covar=tensor([0.1705, 0.1690, 0.0390, 0.0820, 0.0813, 0.0873, 0.0974, 0.1823], + device='cuda:3'), in_proj_covar=tensor([0.0718, 0.0638, 0.0534, 0.0612, 0.0732, 0.0593, 0.0578, 0.0594], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 00:08:43,631 INFO [train.py:901] (3/4) Epoch 4, batch 7700, loss[loss=0.2843, simple_loss=0.3584, pruned_loss=0.1051, over 8470.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3584, pruned_loss=0.1214, over 1616340.90 frames. ], batch size: 27, lr: 1.71e-02, grad_scale: 8.0 +2023-02-06 00:08:55,989 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31968.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:08:56,026 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31968.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 00:09:06,796 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 00:09:17,470 INFO [train.py:901] (3/4) Epoch 4, batch 7750, loss[loss=0.317, simple_loss=0.3627, pruned_loss=0.1357, over 7689.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3587, pruned_loss=0.1217, over 1617235.04 frames. ], batch size: 18, lr: 1.71e-02, grad_scale: 8.0 +2023-02-06 00:09:35,965 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.165e+02 3.163e+02 3.927e+02 5.355e+02 1.239e+03, threshold=7.853e+02, percent-clipped=4.0 +2023-02-06 00:09:53,605 INFO [train.py:901] (3/4) Epoch 4, batch 7800, loss[loss=0.3631, simple_loss=0.4059, pruned_loss=0.1601, over 8565.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3576, pruned_loss=0.1214, over 1613679.61 frames. ], batch size: 39, lr: 1.71e-02, grad_scale: 8.0 +2023-02-06 00:10:05,169 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32066.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:10:11,338 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8664, 1.5174, 2.3028, 2.0028, 2.0439, 1.7375, 1.3416, 0.7002], + device='cuda:3'), covar=tensor([0.1619, 0.1689, 0.0441, 0.0753, 0.0656, 0.0862, 0.0877, 0.1592], + device='cuda:3'), in_proj_covar=tensor([0.0710, 0.0637, 0.0537, 0.0604, 0.0717, 0.0589, 0.0572, 0.0587], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 00:10:27,388 INFO [train.py:901] (3/4) Epoch 4, batch 7850, loss[loss=0.2653, simple_loss=0.3366, pruned_loss=0.09698, over 8362.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.359, pruned_loss=0.1222, over 1614473.32 frames. ], batch size: 24, lr: 1.71e-02, grad_scale: 8.0 +2023-02-06 00:10:43,938 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.460e+02 3.521e+02 4.480e+02 6.179e+02 1.308e+03, threshold=8.960e+02, percent-clipped=13.0 +2023-02-06 00:10:55,619 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32142.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:11:00,787 INFO [train.py:901] (3/4) Epoch 4, batch 7900, loss[loss=0.2848, simple_loss=0.3365, pruned_loss=0.1165, over 7528.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3591, pruned_loss=0.1224, over 1613291.03 frames. ], batch size: 18, lr: 1.71e-02, grad_scale: 8.0 +2023-02-06 00:11:00,855 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32150.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:11:12,545 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32167.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:11:21,573 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32181.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:11:24,163 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32185.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:11:34,311 INFO [train.py:901] (3/4) Epoch 4, batch 7950, loss[loss=0.3281, simple_loss=0.3856, pruned_loss=0.1353, over 8553.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3583, pruned_loss=0.1213, over 1615743.19 frames. ], batch size: 49, lr: 1.71e-02, grad_scale: 8.0 +2023-02-06 00:11:51,232 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32224.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 00:11:51,622 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.248e+02 3.536e+02 4.226e+02 5.315e+02 1.259e+03, threshold=8.452e+02, percent-clipped=4.0 +2023-02-06 00:12:01,775 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32240.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:12:08,279 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32249.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 00:12:08,740 INFO [train.py:901] (3/4) Epoch 4, batch 8000, loss[loss=0.2984, simple_loss=0.3507, pruned_loss=0.123, over 7937.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3594, pruned_loss=0.1219, over 1620155.18 frames. ], batch size: 20, lr: 1.71e-02, grad_scale: 8.0 +2023-02-06 00:12:19,090 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32265.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:12:26,801 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32277.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:12:42,460 INFO [train.py:901] (3/4) Epoch 4, batch 8050, loss[loss=0.2561, simple_loss=0.3109, pruned_loss=0.1007, over 7238.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3596, pruned_loss=0.1236, over 1602038.87 frames. ], batch size: 16, lr: 1.70e-02, grad_scale: 8.0 +2023-02-06 00:12:42,654 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32300.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:12:50,708 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32312.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:12:51,458 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6578, 1.7672, 1.8898, 1.5222, 0.9740, 1.7117, 0.2435, 1.1944], + device='cuda:3'), covar=tensor([0.4053, 0.3222, 0.1563, 0.2428, 0.7927, 0.1511, 0.6101, 0.2880], + device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0109, 0.0079, 0.0162, 0.0201, 0.0081, 0.0145, 0.0119], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 00:12:58,911 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.697e+02 3.496e+02 4.220e+02 5.135e+02 1.064e+03, threshold=8.441e+02, percent-clipped=2.0 +2023-02-06 00:13:15,885 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 00:13:19,665 INFO [train.py:901] (3/4) Epoch 5, batch 0, loss[loss=0.3183, simple_loss=0.3579, pruned_loss=0.1393, over 7934.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3579, pruned_loss=0.1393, over 7934.00 frames. ], batch size: 20, lr: 1.59e-02, grad_scale: 8.0 +2023-02-06 00:13:19,666 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 00:13:31,615 INFO [train.py:935] (3/4) Epoch 5, validation: loss=0.2309, simple_loss=0.3254, pruned_loss=0.06822, over 944034.00 frames. +2023-02-06 00:13:31,616 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 00:13:46,435 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 00:13:46,612 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32355.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:14:06,989 INFO [train.py:901] (3/4) Epoch 5, batch 50, loss[loss=0.3401, simple_loss=0.3817, pruned_loss=0.1492, over 8235.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3628, pruned_loss=0.1236, over 369010.72 frames. ], batch size: 22, lr: 1.58e-02, grad_scale: 8.0 +2023-02-06 00:14:14,084 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32392.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:14:22,029 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2023-02-06 00:14:22,894 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 00:14:36,524 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.029e+02 3.148e+02 3.721e+02 4.839e+02 1.477e+03, threshold=7.442e+02, percent-clipped=1.0 +2023-02-06 00:14:38,056 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32427.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:14:41,779 INFO [train.py:901] (3/4) Epoch 5, batch 100, loss[loss=0.2827, simple_loss=0.3227, pruned_loss=0.1214, over 7242.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3596, pruned_loss=0.1219, over 647849.36 frames. ], batch size: 16, lr: 1.58e-02, grad_scale: 8.0 +2023-02-06 00:14:44,572 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32437.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:14:45,046 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 00:15:02,115 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32462.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:15:15,799 INFO [train.py:901] (3/4) Epoch 5, batch 150, loss[loss=0.3056, simple_loss=0.363, pruned_loss=0.1241, over 7791.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3575, pruned_loss=0.1209, over 862620.50 frames. ], batch size: 20, lr: 1.58e-02, grad_scale: 8.0 +2023-02-06 00:15:43,039 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32521.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:15:45,465 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.082e+02 3.007e+02 3.818e+02 4.644e+02 8.323e+02, threshold=7.636e+02, percent-clipped=1.0 +2023-02-06 00:15:50,801 INFO [train.py:901] (3/4) Epoch 5, batch 200, loss[loss=0.2857, simple_loss=0.3529, pruned_loss=0.1093, over 8318.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3564, pruned_loss=0.1193, over 1024260.86 frames. ], batch size: 25, lr: 1.58e-02, grad_scale: 8.0 +2023-02-06 00:15:59,814 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32546.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:16:06,406 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32556.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:16:23,700 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32581.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:16:24,867 INFO [train.py:901] (3/4) Epoch 5, batch 250, loss[loss=0.258, simple_loss=0.3162, pruned_loss=0.0999, over 7661.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3581, pruned_loss=0.1201, over 1159264.08 frames. ], batch size: 19, lr: 1.58e-02, grad_scale: 8.0 +2023-02-06 00:16:36,174 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 00:16:45,158 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32611.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:16:46,301 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 00:16:54,493 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.598e+02 3.241e+02 4.131e+02 4.869e+02 1.219e+03, threshold=8.263e+02, percent-clipped=9.0 +2023-02-06 00:17:00,684 INFO [train.py:901] (3/4) Epoch 5, batch 300, loss[loss=0.3294, simple_loss=0.3994, pruned_loss=0.1297, over 8510.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3589, pruned_loss=0.1199, over 1263755.28 frames. ], batch size: 26, lr: 1.58e-02, grad_scale: 8.0 +2023-02-06 00:17:03,012 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32636.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:17:10,889 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32648.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:17:27,574 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32673.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:17:34,190 INFO [train.py:901] (3/4) Epoch 5, batch 350, loss[loss=0.3739, simple_loss=0.4009, pruned_loss=0.1735, over 6965.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3602, pruned_loss=0.1213, over 1346348.77 frames. ], batch size: 73, lr: 1.58e-02, grad_scale: 8.0 +2023-02-06 00:17:34,410 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32683.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:17:48,606 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.30 vs. limit=5.0 +2023-02-06 00:17:51,716 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32708.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:18:04,025 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.794e+02 3.189e+02 4.031e+02 4.810e+02 8.158e+02, threshold=8.062e+02, percent-clipped=0.0 +2023-02-06 00:18:09,323 INFO [train.py:901] (3/4) Epoch 5, batch 400, loss[loss=0.3389, simple_loss=0.3965, pruned_loss=0.1407, over 8504.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3601, pruned_loss=0.1216, over 1408663.59 frames. ], batch size: 49, lr: 1.58e-02, grad_scale: 8.0 +2023-02-06 00:18:29,055 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8423, 1.4838, 2.3307, 1.9730, 2.1543, 1.6504, 1.2599, 0.6391], + device='cuda:3'), covar=tensor([0.1646, 0.1774, 0.0417, 0.0703, 0.0582, 0.0832, 0.0931, 0.1590], + device='cuda:3'), in_proj_covar=tensor([0.0715, 0.0646, 0.0553, 0.0615, 0.0725, 0.0593, 0.0586, 0.0596], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 00:18:43,765 INFO [train.py:901] (3/4) Epoch 5, batch 450, loss[loss=0.2943, simple_loss=0.3551, pruned_loss=0.1167, over 8098.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.358, pruned_loss=0.1204, over 1454544.50 frames. ], batch size: 23, lr: 1.58e-02, grad_scale: 8.0 +2023-02-06 00:19:12,446 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.263e+02 3.122e+02 4.068e+02 4.898e+02 9.897e+02, threshold=8.137e+02, percent-clipped=5.0 +2023-02-06 00:19:17,687 INFO [train.py:901] (3/4) Epoch 5, batch 500, loss[loss=0.3005, simple_loss=0.3587, pruned_loss=0.1211, over 7919.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3582, pruned_loss=0.1207, over 1488718.72 frames. ], batch size: 20, lr: 1.57e-02, grad_scale: 8.0 +2023-02-06 00:19:39,191 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8814, 2.4658, 4.8093, 1.4121, 2.9237, 2.5247, 1.7966, 2.6416], + device='cuda:3'), covar=tensor([0.1289, 0.1606, 0.0551, 0.2660, 0.1267, 0.1827, 0.1234, 0.1975], + device='cuda:3'), in_proj_covar=tensor([0.0468, 0.0440, 0.0522, 0.0526, 0.0577, 0.0502, 0.0448, 0.0587], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 00:19:52,905 INFO [train.py:901] (3/4) Epoch 5, batch 550, loss[loss=0.3362, simple_loss=0.3979, pruned_loss=0.1373, over 8533.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3586, pruned_loss=0.121, over 1520031.19 frames. ], batch size: 31, lr: 1.57e-02, grad_scale: 8.0 +2023-02-06 00:20:02,444 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.15 vs. limit=5.0 +2023-02-06 00:20:21,235 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.184e+02 3.133e+02 3.697e+02 5.126e+02 1.321e+03, threshold=7.393e+02, percent-clipped=4.0 +2023-02-06 00:20:26,724 INFO [train.py:901] (3/4) Epoch 5, batch 600, loss[loss=0.3022, simple_loss=0.3352, pruned_loss=0.1346, over 7217.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3571, pruned_loss=0.1198, over 1543312.80 frames. ], batch size: 16, lr: 1.57e-02, grad_scale: 8.0 +2023-02-06 00:20:50,758 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 00:21:02,150 INFO [train.py:901] (3/4) Epoch 5, batch 650, loss[loss=0.3495, simple_loss=0.3879, pruned_loss=0.1555, over 8026.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3561, pruned_loss=0.1191, over 1557452.52 frames. ], batch size: 22, lr: 1.57e-02, grad_scale: 8.0 +2023-02-06 00:21:02,979 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32984.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:21:04,609 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-06 00:21:30,782 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.418e+02 3.090e+02 3.854e+02 5.024e+02 8.355e+02, threshold=7.708e+02, percent-clipped=4.0 +2023-02-06 00:21:36,138 INFO [train.py:901] (3/4) Epoch 5, batch 700, loss[loss=0.2569, simple_loss=0.3363, pruned_loss=0.08877, over 8459.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3559, pruned_loss=0.119, over 1574072.52 frames. ], batch size: 25, lr: 1.57e-02, grad_scale: 8.0 +2023-02-06 00:22:11,097 INFO [train.py:901] (3/4) Epoch 5, batch 750, loss[loss=0.3373, simple_loss=0.3801, pruned_loss=0.1472, over 8497.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3557, pruned_loss=0.1189, over 1581999.45 frames. ], batch size: 26, lr: 1.57e-02, grad_scale: 8.0 +2023-02-06 00:22:12,595 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5001, 1.5151, 1.4094, 1.2714, 1.5219, 1.3201, 1.9440, 1.9594], + device='cuda:3'), covar=tensor([0.0501, 0.1380, 0.2043, 0.1498, 0.0667, 0.1858, 0.0730, 0.0521], + device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0187, 0.0226, 0.0187, 0.0141, 0.0195, 0.0152, 0.0156], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 00:22:14,422 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33087.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:22:16,106 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 +2023-02-06 00:22:36,865 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 00:22:40,967 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.060e+02 3.306e+02 4.079e+02 5.042e+02 1.499e+03, threshold=8.159e+02, percent-clipped=7.0 +2023-02-06 00:22:45,524 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 00:22:46,150 INFO [train.py:901] (3/4) Epoch 5, batch 800, loss[loss=0.3014, simple_loss=0.3682, pruned_loss=0.1173, over 8442.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3568, pruned_loss=0.1193, over 1593105.23 frames. ], batch size: 49, lr: 1.57e-02, grad_scale: 8.0 +2023-02-06 00:23:13,726 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33173.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:23:19,950 INFO [train.py:901] (3/4) Epoch 5, batch 850, loss[loss=0.2532, simple_loss=0.3278, pruned_loss=0.08927, over 8471.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3544, pruned_loss=0.1177, over 1597267.66 frames. ], batch size: 25, lr: 1.57e-02, grad_scale: 8.0 +2023-02-06 00:23:49,970 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.524e+02 2.888e+02 3.855e+02 5.468e+02 1.103e+03, threshold=7.709e+02, percent-clipped=2.0 +2023-02-06 00:23:56,030 INFO [train.py:901] (3/4) Epoch 5, batch 900, loss[loss=0.2949, simple_loss=0.3519, pruned_loss=0.1189, over 8138.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3557, pruned_loss=0.1185, over 1603195.76 frames. ], batch size: 22, lr: 1.57e-02, grad_scale: 8.0 +2023-02-06 00:24:29,752 INFO [train.py:901] (3/4) Epoch 5, batch 950, loss[loss=0.2742, simple_loss=0.3357, pruned_loss=0.1063, over 7649.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3547, pruned_loss=0.1177, over 1605272.78 frames. ], batch size: 19, lr: 1.56e-02, grad_scale: 8.0 +2023-02-06 00:25:01,024 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.012e+02 3.004e+02 3.759e+02 4.642e+02 8.675e+02, threshold=7.519e+02, percent-clipped=2.0 +2023-02-06 00:25:03,069 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33328.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:25:05,120 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 00:25:06,447 INFO [train.py:901] (3/4) Epoch 5, batch 1000, loss[loss=0.3677, simple_loss=0.4047, pruned_loss=0.1653, over 7157.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3529, pruned_loss=0.1165, over 1601554.84 frames. ], batch size: 71, lr: 1.56e-02, grad_scale: 8.0 +2023-02-06 00:25:06,675 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5591, 1.7770, 2.0718, 1.7293, 1.0138, 2.1708, 0.3163, 1.1265], + device='cuda:3'), covar=tensor([0.3832, 0.2422, 0.0988, 0.2918, 0.6964, 0.0880, 0.7528, 0.3879], + device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0117, 0.0080, 0.0165, 0.0210, 0.0081, 0.0146, 0.0123], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 00:25:16,753 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1973, 1.5488, 1.5105, 1.2461, 1.4761, 1.4201, 1.6522, 1.7293], + device='cuda:3'), covar=tensor([0.0597, 0.1207, 0.1949, 0.1452, 0.0631, 0.1594, 0.0763, 0.0564], + device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0184, 0.0224, 0.0183, 0.0140, 0.0193, 0.0153, 0.0156], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 00:25:40,440 INFO [train.py:901] (3/4) Epoch 5, batch 1050, loss[loss=0.2447, simple_loss=0.3253, pruned_loss=0.08208, over 8255.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3529, pruned_loss=0.1168, over 1600672.64 frames. ], batch size: 24, lr: 1.56e-02, grad_scale: 8.0 +2023-02-06 00:25:40,444 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 00:25:47,767 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 00:25:52,111 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 00:26:08,784 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.885e+02 3.252e+02 3.786e+02 4.850e+02 9.380e+02, threshold=7.572e+02, percent-clipped=3.0 +2023-02-06 00:26:13,599 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33431.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:26:14,849 INFO [train.py:901] (3/4) Epoch 5, batch 1100, loss[loss=0.3121, simple_loss=0.3614, pruned_loss=0.1314, over 7536.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3539, pruned_loss=0.117, over 1610131.76 frames. ], batch size: 71, lr: 1.56e-02, grad_scale: 8.0 +2023-02-06 00:26:23,006 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33443.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:26:50,041 INFO [train.py:901] (3/4) Epoch 5, batch 1150, loss[loss=0.2298, simple_loss=0.2989, pruned_loss=0.08035, over 7444.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3545, pruned_loss=0.1176, over 1611431.12 frames. ], batch size: 17, lr: 1.56e-02, grad_scale: 8.0 +2023-02-06 00:26:54,935 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33490.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:27:02,050 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 00:27:13,083 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33517.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:27:18,265 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.110e+02 3.101e+02 4.052e+02 5.357e+02 1.331e+03, threshold=8.105e+02, percent-clipped=11.0 +2023-02-06 00:27:23,603 INFO [train.py:901] (3/4) Epoch 5, batch 1200, loss[loss=0.376, simple_loss=0.4166, pruned_loss=0.1677, over 8567.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3535, pruned_loss=0.1173, over 1608516.80 frames. ], batch size: 31, lr: 1.56e-02, grad_scale: 8.0 +2023-02-06 00:27:32,555 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33546.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:28:00,131 INFO [train.py:901] (3/4) Epoch 5, batch 1250, loss[loss=0.2893, simple_loss=0.3433, pruned_loss=0.1177, over 7973.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3538, pruned_loss=0.1172, over 1613931.18 frames. ], batch size: 21, lr: 1.56e-02, grad_scale: 8.0 +2023-02-06 00:28:29,029 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.930e+02 3.057e+02 3.737e+02 5.343e+02 1.068e+03, threshold=7.474e+02, percent-clipped=1.0 +2023-02-06 00:28:34,054 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33632.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:28:34,561 INFO [train.py:901] (3/4) Epoch 5, batch 1300, loss[loss=0.2465, simple_loss=0.3213, pruned_loss=0.08583, over 7532.00 frames. ], tot_loss[loss=0.2905, simple_loss=0.3508, pruned_loss=0.1151, over 1611837.21 frames. ], batch size: 18, lr: 1.56e-02, grad_scale: 8.0 +2023-02-06 00:28:38,322 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.96 vs. limit=5.0 +2023-02-06 00:29:10,162 INFO [train.py:901] (3/4) Epoch 5, batch 1350, loss[loss=0.2794, simple_loss=0.3463, pruned_loss=0.1063, over 8136.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3505, pruned_loss=0.1145, over 1613072.86 frames. ], batch size: 22, lr: 1.55e-02, grad_scale: 4.0 +2023-02-06 00:29:18,826 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33695.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:29:21,585 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33699.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:29:23,806 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 00:29:38,302 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33724.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:29:39,358 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.845e+02 3.141e+02 3.942e+02 4.566e+02 9.800e+02, threshold=7.885e+02, percent-clipped=1.0 +2023-02-06 00:29:44,199 INFO [train.py:901] (3/4) Epoch 5, batch 1400, loss[loss=0.2699, simple_loss=0.3364, pruned_loss=0.1017, over 8107.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3516, pruned_loss=0.1149, over 1615203.35 frames. ], batch size: 23, lr: 1.55e-02, grad_scale: 4.0 +2023-02-06 00:30:16,757 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.25 vs. limit=5.0 +2023-02-06 00:30:18,172 INFO [train.py:901] (3/4) Epoch 5, batch 1450, loss[loss=0.2491, simple_loss=0.3089, pruned_loss=0.09463, over 7660.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3512, pruned_loss=0.1152, over 1614804.48 frames. ], batch size: 19, lr: 1.55e-02, grad_scale: 4.0 +2023-02-06 00:30:32,277 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 00:30:32,504 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33802.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:30:49,113 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.073e+02 3.068e+02 3.705e+02 5.190e+02 1.303e+03, threshold=7.410e+02, percent-clipped=4.0 +2023-02-06 00:30:50,020 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33827.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:30:54,017 INFO [train.py:901] (3/4) Epoch 5, batch 1500, loss[loss=0.3321, simple_loss=0.3832, pruned_loss=0.1405, over 8595.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3517, pruned_loss=0.1152, over 1615609.83 frames. ], batch size: 31, lr: 1.55e-02, grad_scale: 4.0 +2023-02-06 00:30:54,781 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33834.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:31:16,100 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 +2023-02-06 00:31:27,604 INFO [train.py:901] (3/4) Epoch 5, batch 1550, loss[loss=0.288, simple_loss=0.3479, pruned_loss=0.114, over 8247.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.351, pruned_loss=0.1158, over 1608066.68 frames. ], batch size: 24, lr: 1.55e-02, grad_scale: 4.0 +2023-02-06 00:31:31,095 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33888.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:31:48,390 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2586, 1.4148, 2.1584, 0.9978, 1.9796, 2.1810, 2.3791, 1.7047], + device='cuda:3'), covar=tensor([0.1224, 0.1208, 0.0657, 0.2400, 0.0763, 0.0646, 0.0649, 0.1280], + device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0261, 0.0215, 0.0255, 0.0219, 0.0193, 0.0218, 0.0267], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 00:31:48,419 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33913.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:31:58,126 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.067e+02 3.419e+02 4.027e+02 4.998e+02 8.696e+02, threshold=8.054e+02, percent-clipped=2.0 +2023-02-06 00:32:03,155 INFO [train.py:901] (3/4) Epoch 5, batch 1600, loss[loss=0.3517, simple_loss=0.4054, pruned_loss=0.149, over 8235.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3527, pruned_loss=0.1165, over 1610455.63 frames. ], batch size: 24, lr: 1.55e-02, grad_scale: 8.0 +2023-02-06 00:32:14,803 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33949.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:32:17,009 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 +2023-02-06 00:32:20,121 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33957.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:32:31,556 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.91 vs. limit=5.0 +2023-02-06 00:32:37,050 INFO [train.py:901] (3/4) Epoch 5, batch 1650, loss[loss=0.2753, simple_loss=0.344, pruned_loss=0.1033, over 8105.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3518, pruned_loss=0.1154, over 1612801.88 frames. ], batch size: 23, lr: 1.55e-02, grad_scale: 8.0 +2023-02-06 00:33:07,304 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.007e+02 2.926e+02 3.722e+02 4.611e+02 9.053e+02, threshold=7.444e+02, percent-clipped=4.0 +2023-02-06 00:33:11,834 INFO [train.py:901] (3/4) Epoch 5, batch 1700, loss[loss=0.3617, simple_loss=0.3954, pruned_loss=0.1639, over 8335.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3536, pruned_loss=0.1173, over 1613172.75 frames. ], batch size: 25, lr: 1.55e-02, grad_scale: 8.0 +2023-02-06 00:33:17,289 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34039.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:33:33,627 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34062.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:33:47,837 INFO [train.py:901] (3/4) Epoch 5, batch 1750, loss[loss=0.2887, simple_loss=0.3596, pruned_loss=0.1089, over 8505.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3537, pruned_loss=0.1171, over 1615706.77 frames. ], batch size: 28, lr: 1.55e-02, grad_scale: 8.0 +2023-02-06 00:34:16,515 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.177e+02 3.118e+02 3.687e+02 4.787e+02 9.448e+02, threshold=7.373e+02, percent-clipped=7.0 +2023-02-06 00:34:21,854 INFO [train.py:901] (3/4) Epoch 5, batch 1800, loss[loss=0.3319, simple_loss=0.3855, pruned_loss=0.1391, over 8201.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3543, pruned_loss=0.1176, over 1615387.92 frames. ], batch size: 23, lr: 1.54e-02, grad_scale: 8.0 +2023-02-06 00:34:36,714 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34154.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:34:38,709 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4579, 1.7014, 1.8367, 1.6493, 0.9896, 1.9092, 0.3481, 1.1225], + device='cuda:3'), covar=tensor([0.3592, 0.2131, 0.1255, 0.1774, 0.7175, 0.0948, 0.5914, 0.2882], + device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0120, 0.0081, 0.0167, 0.0216, 0.0084, 0.0148, 0.0127], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 00:34:43,247 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34163.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:34:57,300 INFO [train.py:901] (3/4) Epoch 5, batch 1850, loss[loss=0.3522, simple_loss=0.403, pruned_loss=0.1507, over 8254.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3539, pruned_loss=0.117, over 1617754.99 frames. ], batch size: 24, lr: 1.54e-02, grad_scale: 8.0 +2023-02-06 00:35:12,369 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34205.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:35:26,215 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.821e+02 3.489e+02 4.150e+02 5.670e+02 1.027e+03, threshold=8.299e+02, percent-clipped=7.0 +2023-02-06 00:35:29,066 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34230.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:35:30,897 INFO [train.py:901] (3/4) Epoch 5, batch 1900, loss[loss=0.3266, simple_loss=0.3874, pruned_loss=0.1329, over 8105.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3519, pruned_loss=0.1159, over 1610516.22 frames. ], batch size: 23, lr: 1.54e-02, grad_scale: 8.0 +2023-02-06 00:36:05,945 INFO [train.py:901] (3/4) Epoch 5, batch 1950, loss[loss=0.2881, simple_loss=0.3428, pruned_loss=0.1167, over 8336.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3514, pruned_loss=0.116, over 1608438.49 frames. ], batch size: 26, lr: 1.54e-02, grad_scale: 8.0 +2023-02-06 00:36:09,873 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 00:36:18,719 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34301.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:36:23,388 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 00:36:35,388 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.917e+02 2.967e+02 3.945e+02 4.927e+02 1.257e+03, threshold=7.890e+02, percent-clipped=2.0 +2023-02-06 00:36:40,163 INFO [train.py:901] (3/4) Epoch 5, batch 2000, loss[loss=0.3383, simple_loss=0.3842, pruned_loss=0.1462, over 8024.00 frames. ], tot_loss[loss=0.2905, simple_loss=0.3504, pruned_loss=0.1153, over 1608245.62 frames. ], batch size: 22, lr: 1.54e-02, grad_scale: 8.0 +2023-02-06 00:36:42,282 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 00:37:14,373 INFO [train.py:901] (3/4) Epoch 5, batch 2050, loss[loss=0.2248, simple_loss=0.2933, pruned_loss=0.07817, over 7713.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3511, pruned_loss=0.1158, over 1607017.51 frames. ], batch size: 18, lr: 1.54e-02, grad_scale: 8.0 +2023-02-06 00:37:29,113 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9554, 3.3200, 2.3326, 4.1214, 1.6912, 2.2496, 2.1186, 3.2436], + device='cuda:3'), covar=tensor([0.0752, 0.1095, 0.1297, 0.0363, 0.1661, 0.1846, 0.1959, 0.0966], + device='cuda:3'), in_proj_covar=tensor([0.0269, 0.0268, 0.0286, 0.0221, 0.0261, 0.0287, 0.0289, 0.0263], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 00:37:31,143 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34406.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:37:33,921 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34410.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:37:38,677 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34416.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:37:45,186 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.678e+02 3.245e+02 4.066e+02 4.898e+02 1.293e+03, threshold=8.132e+02, percent-clipped=4.0 +2023-02-06 00:37:49,840 INFO [train.py:901] (3/4) Epoch 5, batch 2100, loss[loss=0.2554, simple_loss=0.3237, pruned_loss=0.09352, over 8034.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3522, pruned_loss=0.1167, over 1609788.02 frames. ], batch size: 22, lr: 1.54e-02, grad_scale: 8.0 +2023-02-06 00:37:51,415 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34435.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:38:23,166 INFO [train.py:901] (3/4) Epoch 5, batch 2150, loss[loss=0.3279, simple_loss=0.3866, pruned_loss=0.1346, over 8099.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3523, pruned_loss=0.1165, over 1616040.72 frames. ], batch size: 23, lr: 1.54e-02, grad_scale: 8.0 +2023-02-06 00:38:39,927 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34507.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:38:50,619 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34521.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:38:53,806 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.931e+02 2.929e+02 3.753e+02 4.663e+02 1.529e+03, threshold=7.506e+02, percent-clipped=2.0 +2023-02-06 00:38:59,141 INFO [train.py:901] (3/4) Epoch 5, batch 2200, loss[loss=0.3375, simple_loss=0.3899, pruned_loss=0.1425, over 8511.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3524, pruned_loss=0.1167, over 1611597.27 frames. ], batch size: 26, lr: 1.54e-02, grad_scale: 8.0 +2023-02-06 00:39:32,427 INFO [train.py:901] (3/4) Epoch 5, batch 2250, loss[loss=0.3164, simple_loss=0.3773, pruned_loss=0.1277, over 8470.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3518, pruned_loss=0.1164, over 1611205.45 frames. ], batch size: 25, lr: 1.53e-02, grad_scale: 8.0 +2023-02-06 00:39:52,222 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34611.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:39:59,502 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34622.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:40:02,640 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.996e+02 3.688e+02 4.883e+02 6.349e+02 4.437e+03, threshold=9.766e+02, percent-clipped=16.0 +2023-02-06 00:40:07,919 INFO [train.py:901] (3/4) Epoch 5, batch 2300, loss[loss=0.3134, simple_loss=0.3429, pruned_loss=0.142, over 7248.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3513, pruned_loss=0.116, over 1613512.96 frames. ], batch size: 16, lr: 1.53e-02, grad_scale: 8.0 +2023-02-06 00:40:12,726 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34640.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:40:16,085 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34644.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:40:34,982 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34672.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:40:42,256 INFO [train.py:901] (3/4) Epoch 5, batch 2350, loss[loss=0.2637, simple_loss=0.328, pruned_loss=0.09967, over 8147.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3508, pruned_loss=0.1158, over 1610665.93 frames. ], batch size: 22, lr: 1.53e-02, grad_scale: 8.0 +2023-02-06 00:40:51,695 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34697.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:41:11,439 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.800e+02 3.189e+02 4.018e+02 4.942e+02 1.178e+03, threshold=8.036e+02, percent-clipped=1.0 +2023-02-06 00:41:12,937 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34728.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 00:41:14,290 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3127, 1.9513, 1.5227, 1.4690, 1.4869, 1.3700, 1.7783, 1.7855], + device='cuda:3'), covar=tensor([0.0617, 0.1208, 0.1884, 0.1426, 0.0628, 0.1661, 0.0757, 0.0558], + device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0185, 0.0226, 0.0190, 0.0139, 0.0195, 0.0150, 0.0159], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 00:41:16,112 INFO [train.py:901] (3/4) Epoch 5, batch 2400, loss[loss=0.2476, simple_loss=0.3107, pruned_loss=0.09223, over 7927.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.351, pruned_loss=0.1158, over 1614657.69 frames. ], batch size: 20, lr: 1.53e-02, grad_scale: 8.0 +2023-02-06 00:41:47,413 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34777.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:41:49,403 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5047, 1.9420, 2.0886, 0.8355, 2.1660, 1.5103, 0.5345, 1.7573], + device='cuda:3'), covar=tensor([0.0194, 0.0099, 0.0088, 0.0175, 0.0102, 0.0299, 0.0264, 0.0084], + device='cuda:3'), in_proj_covar=tensor([0.0303, 0.0217, 0.0179, 0.0260, 0.0208, 0.0351, 0.0273, 0.0250], + device='cuda:3'), out_proj_covar=tensor([1.1232e-04, 7.9085e-05, 6.3760e-05, 9.3411e-05, 7.7770e-05, 1.3878e-04, + 1.0153e-04, 9.1232e-05], device='cuda:3') +2023-02-06 00:41:51,224 INFO [train.py:901] (3/4) Epoch 5, batch 2450, loss[loss=0.3574, simple_loss=0.4054, pruned_loss=0.1547, over 8551.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3508, pruned_loss=0.1156, over 1613003.04 frames. ], batch size: 28, lr: 1.53e-02, grad_scale: 8.0 +2023-02-06 00:42:04,203 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34802.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:42:09,448 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7543, 2.8086, 3.0652, 1.7337, 1.2644, 2.8573, 0.4969, 1.8108], + device='cuda:3'), covar=tensor([0.2867, 0.1774, 0.1625, 0.4132, 0.7857, 0.1088, 0.7006, 0.2304], + device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0121, 0.0080, 0.0167, 0.0210, 0.0082, 0.0147, 0.0120], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 00:42:18,442 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3655, 2.8324, 1.6851, 2.0988, 2.4130, 1.2591, 1.8706, 2.1175], + device='cuda:3'), covar=tensor([0.1338, 0.0340, 0.0946, 0.0648, 0.0540, 0.1194, 0.1034, 0.0835], + device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0243, 0.0315, 0.0308, 0.0322, 0.0310, 0.0340, 0.0313], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 00:42:19,463 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.849e+02 2.991e+02 3.791e+02 4.954e+02 1.109e+03, threshold=7.583e+02, percent-clipped=3.0 +2023-02-06 00:42:24,147 INFO [train.py:901] (3/4) Epoch 5, batch 2500, loss[loss=0.3123, simple_loss=0.3793, pruned_loss=0.1226, over 8326.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3525, pruned_loss=0.1165, over 1619421.16 frames. ], batch size: 25, lr: 1.53e-02, grad_scale: 8.0 +2023-02-06 00:42:55,971 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34878.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:42:59,763 INFO [train.py:901] (3/4) Epoch 5, batch 2550, loss[loss=0.2998, simple_loss=0.3536, pruned_loss=0.123, over 8669.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3519, pruned_loss=0.1154, over 1616670.22 frames. ], batch size: 39, lr: 1.53e-02, grad_scale: 8.0 +2023-02-06 00:43:13,491 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34903.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:43:29,133 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.077e+02 2.947e+02 3.618e+02 4.736e+02 1.253e+03, threshold=7.237e+02, percent-clipped=4.0 +2023-02-06 00:43:33,905 INFO [train.py:901] (3/4) Epoch 5, batch 2600, loss[loss=0.3364, simple_loss=0.4068, pruned_loss=0.1331, over 8318.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.352, pruned_loss=0.1158, over 1614475.69 frames. ], batch size: 25, lr: 1.53e-02, grad_scale: 8.0 +2023-02-06 00:43:49,009 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34955.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:44:09,524 INFO [train.py:901] (3/4) Epoch 5, batch 2650, loss[loss=0.3782, simple_loss=0.4268, pruned_loss=0.1647, over 8243.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3517, pruned_loss=0.1162, over 1610220.03 frames. ], batch size: 22, lr: 1.53e-02, grad_scale: 8.0 +2023-02-06 00:44:10,292 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34984.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:44:13,023 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34988.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:44:39,308 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.917e+02 2.922e+02 3.827e+02 4.980e+02 8.274e+02, threshold=7.654e+02, percent-clipped=5.0 +2023-02-06 00:44:43,764 INFO [train.py:901] (3/4) Epoch 5, batch 2700, loss[loss=0.2612, simple_loss=0.3302, pruned_loss=0.09611, over 7648.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3519, pruned_loss=0.1163, over 1610200.07 frames. ], batch size: 19, lr: 1.53e-02, grad_scale: 8.0 +2023-02-06 00:45:09,483 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35070.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:45:10,724 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35072.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 00:45:12,465 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 00:45:18,323 INFO [train.py:901] (3/4) Epoch 5, batch 2750, loss[loss=0.333, simple_loss=0.385, pruned_loss=0.1405, over 8351.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3513, pruned_loss=0.1157, over 1613780.79 frames. ], batch size: 26, lr: 1.52e-02, grad_scale: 8.0 +2023-02-06 00:45:29,757 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35099.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:45:32,522 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35103.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:45:34,566 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3574, 1.8116, 1.5703, 1.4474, 1.5322, 1.4870, 1.9220, 1.8479], + device='cuda:3'), covar=tensor([0.0616, 0.1135, 0.1755, 0.1457, 0.0643, 0.1577, 0.0758, 0.0570], + device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0185, 0.0226, 0.0189, 0.0139, 0.0195, 0.0149, 0.0158], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 00:45:37,947 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35110.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:45:48,687 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.004e+02 3.039e+02 3.659e+02 5.251e+02 1.248e+03, threshold=7.317e+02, percent-clipped=8.0 +2023-02-06 00:45:53,668 INFO [train.py:901] (3/4) Epoch 5, batch 2800, loss[loss=0.2878, simple_loss=0.355, pruned_loss=0.1103, over 8348.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3499, pruned_loss=0.1146, over 1617163.71 frames. ], batch size: 24, lr: 1.52e-02, grad_scale: 8.0 +2023-02-06 00:46:27,514 INFO [train.py:901] (3/4) Epoch 5, batch 2850, loss[loss=0.3224, simple_loss=0.382, pruned_loss=0.1314, over 8621.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3497, pruned_loss=0.1149, over 1614526.24 frames. ], batch size: 31, lr: 1.52e-02, grad_scale: 8.0 +2023-02-06 00:46:30,525 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35187.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 00:46:58,369 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.147e+02 2.990e+02 3.598e+02 4.675e+02 1.498e+03, threshold=7.197e+02, percent-clipped=4.0 +2023-02-06 00:47:03,665 INFO [train.py:901] (3/4) Epoch 5, batch 2900, loss[loss=0.3242, simple_loss=0.3814, pruned_loss=0.1335, over 8105.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.35, pruned_loss=0.1157, over 1610876.55 frames. ], batch size: 23, lr: 1.52e-02, grad_scale: 8.0 +2023-02-06 00:47:36,539 INFO [train.py:901] (3/4) Epoch 5, batch 2950, loss[loss=0.3144, simple_loss=0.387, pruned_loss=0.1209, over 8290.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3512, pruned_loss=0.1158, over 1606501.76 frames. ], batch size: 23, lr: 1.52e-02, grad_scale: 8.0 +2023-02-06 00:47:41,847 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 00:47:58,889 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7150, 2.2722, 4.7899, 1.2583, 3.3221, 2.2782, 1.7358, 2.6701], + device='cuda:3'), covar=tensor([0.1821, 0.2070, 0.0477, 0.3679, 0.1252, 0.2419, 0.1824, 0.2250], + device='cuda:3'), in_proj_covar=tensor([0.0463, 0.0436, 0.0520, 0.0526, 0.0573, 0.0511, 0.0447, 0.0582], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 00:48:06,832 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.668e+02 3.185e+02 3.825e+02 4.988e+02 1.295e+03, threshold=7.649e+02, percent-clipped=4.0 +2023-02-06 00:48:07,072 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35326.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:48:12,109 INFO [train.py:901] (3/4) Epoch 5, batch 3000, loss[loss=0.3041, simple_loss=0.3645, pruned_loss=0.1218, over 8294.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3523, pruned_loss=0.1163, over 1608681.40 frames. ], batch size: 23, lr: 1.52e-02, grad_scale: 8.0 +2023-02-06 00:48:12,109 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 00:48:22,590 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2702, 2.1946, 1.4480, 1.9687, 1.8109, 1.2048, 1.5636, 1.7552], + device='cuda:3'), covar=tensor([0.1125, 0.0327, 0.0985, 0.0472, 0.0584, 0.1200, 0.0891, 0.0701], + device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0236, 0.0310, 0.0304, 0.0320, 0.0302, 0.0337, 0.0309], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 00:48:25,508 INFO [train.py:935] (3/4) Epoch 5, validation: loss=0.2228, simple_loss=0.319, pruned_loss=0.0633, over 944034.00 frames. +2023-02-06 00:48:25,509 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 00:48:39,272 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35351.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:48:41,954 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3379, 2.1646, 3.4884, 2.1462, 2.8392, 3.9242, 3.7428, 3.5038], + device='cuda:3'), covar=tensor([0.0837, 0.1051, 0.0599, 0.1457, 0.0835, 0.0218, 0.0287, 0.0454], + device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0267, 0.0226, 0.0269, 0.0221, 0.0199, 0.0230, 0.0278], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 00:48:42,026 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35355.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:48:44,709 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35359.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:48:47,351 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35363.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 00:48:59,256 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35380.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:49:01,099 INFO [train.py:901] (3/4) Epoch 5, batch 3050, loss[loss=0.3202, simple_loss=0.3826, pruned_loss=0.1289, over 8412.00 frames. ], tot_loss[loss=0.293, simple_loss=0.353, pruned_loss=0.1165, over 1609433.20 frames. ], batch size: 49, lr: 1.52e-02, grad_scale: 8.0 +2023-02-06 00:49:01,973 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35384.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:49:07,211 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35392.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:49:29,672 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.238e+02 3.010e+02 3.735e+02 4.816e+02 9.592e+02, threshold=7.471e+02, percent-clipped=3.0 +2023-02-06 00:49:33,914 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-06 00:49:34,200 INFO [train.py:901] (3/4) Epoch 5, batch 3100, loss[loss=0.3144, simple_loss=0.3777, pruned_loss=0.1256, over 8316.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.355, pruned_loss=0.1177, over 1614386.45 frames. ], batch size: 25, lr: 1.52e-02, grad_scale: 8.0 +2023-02-06 00:49:41,040 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35443.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 00:49:48,741 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35454.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:49:53,600 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2657, 1.7129, 1.3496, 1.6764, 1.3870, 1.1459, 1.3201, 1.5743], + device='cuda:3'), covar=tensor([0.0832, 0.0386, 0.0878, 0.0441, 0.0588, 0.1071, 0.0707, 0.0584], + device='cuda:3'), in_proj_covar=tensor([0.0360, 0.0242, 0.0312, 0.0309, 0.0323, 0.0310, 0.0346, 0.0318], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 00:49:59,648 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35468.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 00:50:09,986 INFO [train.py:901] (3/4) Epoch 5, batch 3150, loss[loss=0.3077, simple_loss=0.3643, pruned_loss=0.1255, over 8140.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3541, pruned_loss=0.1166, over 1612955.24 frames. ], batch size: 22, lr: 1.52e-02, grad_scale: 8.0 +2023-02-06 00:50:29,958 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4469, 1.9361, 3.3937, 1.0511, 2.2067, 1.8402, 1.5146, 1.7078], + device='cuda:3'), covar=tensor([0.1498, 0.1580, 0.0602, 0.3075, 0.1384, 0.2239, 0.1414, 0.2310], + device='cuda:3'), in_proj_covar=tensor([0.0472, 0.0445, 0.0530, 0.0538, 0.0579, 0.0518, 0.0457, 0.0591], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 00:50:39,622 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.108e+02 3.249e+02 4.087e+02 5.030e+02 9.472e+02, threshold=8.174e+02, percent-clipped=3.0 +2023-02-06 00:50:44,420 INFO [train.py:901] (3/4) Epoch 5, batch 3200, loss[loss=0.2471, simple_loss=0.3314, pruned_loss=0.08141, over 8511.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3546, pruned_loss=0.1171, over 1610101.89 frames. ], batch size: 28, lr: 1.51e-02, grad_scale: 8.0 +2023-02-06 00:50:54,835 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35548.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:51:09,467 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35569.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:51:20,160 INFO [train.py:901] (3/4) Epoch 5, batch 3250, loss[loss=0.2839, simple_loss=0.3576, pruned_loss=0.1051, over 8252.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3552, pruned_loss=0.1175, over 1611942.33 frames. ], batch size: 24, lr: 1.51e-02, grad_scale: 8.0 +2023-02-06 00:51:44,439 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6331, 3.9058, 2.2638, 2.1184, 2.7253, 1.7797, 2.4928, 2.8063], + device='cuda:3'), covar=tensor([0.1630, 0.0217, 0.0905, 0.0901, 0.0699, 0.1174, 0.1111, 0.1086], + device='cuda:3'), in_proj_covar=tensor([0.0364, 0.0240, 0.0317, 0.0306, 0.0320, 0.0313, 0.0349, 0.0324], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 00:51:50,492 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.693e+02 3.378e+02 4.149e+02 5.121e+02 1.146e+03, threshold=8.298e+02, percent-clipped=3.0 +2023-02-06 00:51:51,044 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-06 00:51:55,288 INFO [train.py:901] (3/4) Epoch 5, batch 3300, loss[loss=0.2995, simple_loss=0.3571, pruned_loss=0.1209, over 8634.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3543, pruned_loss=0.1171, over 1610055.71 frames. ], batch size: 34, lr: 1.51e-02, grad_scale: 8.0 +2023-02-06 00:52:30,143 INFO [train.py:901] (3/4) Epoch 5, batch 3350, loss[loss=0.2827, simple_loss=0.3452, pruned_loss=0.1101, over 8109.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3527, pruned_loss=0.1159, over 1609880.56 frames. ], batch size: 23, lr: 1.51e-02, grad_scale: 16.0 +2023-02-06 00:52:47,837 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35707.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 00:53:01,469 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.966e+02 3.555e+02 4.125e+02 4.946e+02 1.065e+03, threshold=8.250e+02, percent-clipped=5.0 +2023-02-06 00:53:06,231 INFO [train.py:901] (3/4) Epoch 5, batch 3400, loss[loss=0.3018, simple_loss=0.3606, pruned_loss=0.1215, over 8453.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.354, pruned_loss=0.1168, over 1608614.73 frames. ], batch size: 48, lr: 1.51e-02, grad_scale: 16.0 +2023-02-06 00:53:08,429 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35736.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:53:25,785 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-06 00:53:26,934 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-06 00:53:34,781 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 00:53:39,726 INFO [train.py:901] (3/4) Epoch 5, batch 3450, loss[loss=0.2759, simple_loss=0.3442, pruned_loss=0.1038, over 8039.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3529, pruned_loss=0.1163, over 1607795.67 frames. ], batch size: 22, lr: 1.51e-02, grad_scale: 16.0 +2023-02-06 00:53:43,882 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35789.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:54:06,639 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6493, 4.3769, 2.4979, 2.1047, 2.7121, 2.1547, 2.3732, 3.0976], + device='cuda:3'), covar=tensor([0.1414, 0.0120, 0.0679, 0.0786, 0.0555, 0.0853, 0.0945, 0.0738], + device='cuda:3'), in_proj_covar=tensor([0.0359, 0.0237, 0.0312, 0.0302, 0.0321, 0.0311, 0.0346, 0.0321], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 00:54:07,873 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35822.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 00:54:09,940 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35825.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:54:10,372 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.123e+02 3.051e+02 3.738e+02 4.571e+02 6.690e+02, threshold=7.475e+02, percent-clipped=0.0 +2023-02-06 00:54:12,583 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35829.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:54:15,094 INFO [train.py:901] (3/4) Epoch 5, batch 3500, loss[loss=0.299, simple_loss=0.3626, pruned_loss=0.1177, over 8368.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.353, pruned_loss=0.116, over 1609066.13 frames. ], batch size: 24, lr: 1.51e-02, grad_scale: 16.0 +2023-02-06 00:54:15,201 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5282, 4.4885, 4.0655, 2.0085, 3.9030, 3.8691, 4.0499, 3.5800], + device='cuda:3'), covar=tensor([0.0716, 0.0570, 0.0888, 0.4116, 0.0794, 0.0858, 0.1365, 0.0796], + device='cuda:3'), in_proj_covar=tensor([0.0383, 0.0277, 0.0309, 0.0393, 0.0306, 0.0266, 0.0293, 0.0238], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:3') +2023-02-06 00:54:27,026 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35850.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:54:27,677 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35851.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:54:40,697 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 00:54:48,820 INFO [train.py:901] (3/4) Epoch 5, batch 3550, loss[loss=0.393, simple_loss=0.4372, pruned_loss=0.1744, over 8330.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3534, pruned_loss=0.1163, over 1612294.11 frames. ], batch size: 26, lr: 1.51e-02, grad_scale: 16.0 +2023-02-06 00:54:54,913 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35892.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:55:19,538 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.232e+02 3.318e+02 3.882e+02 4.908e+02 1.221e+03, threshold=7.763e+02, percent-clipped=6.0 +2023-02-06 00:55:23,998 INFO [train.py:901] (3/4) Epoch 5, batch 3600, loss[loss=0.3247, simple_loss=0.3817, pruned_loss=0.1338, over 8475.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3535, pruned_loss=0.1156, over 1619145.27 frames. ], batch size: 25, lr: 1.51e-02, grad_scale: 16.0 +2023-02-06 00:55:37,602 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0433, 1.2981, 1.1783, 0.2158, 1.1263, 0.9182, 0.1323, 1.1173], + device='cuda:3'), covar=tensor([0.0133, 0.0111, 0.0089, 0.0196, 0.0111, 0.0356, 0.0256, 0.0095], + device='cuda:3'), in_proj_covar=tensor([0.0302, 0.0223, 0.0178, 0.0268, 0.0206, 0.0352, 0.0276, 0.0248], + device='cuda:3'), out_proj_covar=tensor([1.1013e-04, 7.9926e-05, 6.2670e-05, 9.5372e-05, 7.5519e-05, 1.3767e-04, + 1.0143e-04, 8.8963e-05], device='cuda:3') +2023-02-06 00:55:57,697 INFO [train.py:901] (3/4) Epoch 5, batch 3650, loss[loss=0.2595, simple_loss=0.3172, pruned_loss=0.1009, over 7219.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.354, pruned_loss=0.1157, over 1621348.43 frames. ], batch size: 16, lr: 1.51e-02, grad_scale: 16.0 +2023-02-06 00:56:01,318 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4518, 2.1422, 3.1633, 2.7680, 2.7847, 1.9984, 1.4184, 1.5330], + device='cuda:3'), covar=tensor([0.1464, 0.1807, 0.0414, 0.0901, 0.0753, 0.0943, 0.0986, 0.1786], + device='cuda:3'), in_proj_covar=tensor([0.0732, 0.0670, 0.0568, 0.0653, 0.0762, 0.0627, 0.0603, 0.0624], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 00:56:15,066 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36007.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:56:27,680 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.237e+02 3.345e+02 4.197e+02 5.280e+02 9.599e+02, threshold=8.394e+02, percent-clipped=10.0 +2023-02-06 00:56:32,339 INFO [train.py:901] (3/4) Epoch 5, batch 3700, loss[loss=0.2901, simple_loss=0.3521, pruned_loss=0.114, over 8461.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3546, pruned_loss=0.1166, over 1618094.17 frames. ], batch size: 25, lr: 1.50e-02, grad_scale: 16.0 +2023-02-06 00:56:40,796 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 00:56:44,923 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7265, 1.4498, 2.8798, 1.1868, 2.2243, 3.1384, 3.0026, 2.6711], + device='cuda:3'), covar=tensor([0.0963, 0.1262, 0.0366, 0.1944, 0.0584, 0.0257, 0.0414, 0.0671], + device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0262, 0.0218, 0.0258, 0.0217, 0.0194, 0.0224, 0.0267], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 00:57:04,801 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36078.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 00:57:07,929 INFO [train.py:901] (3/4) Epoch 5, batch 3750, loss[loss=0.3157, simple_loss=0.373, pruned_loss=0.1292, over 8239.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3536, pruned_loss=0.1163, over 1613673.69 frames. ], batch size: 22, lr: 1.50e-02, grad_scale: 16.0 +2023-02-06 00:57:09,711 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2023-02-06 00:57:21,376 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36103.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 00:57:24,139 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36107.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:57:37,192 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.699e+02 3.033e+02 3.704e+02 4.599e+02 1.470e+03, threshold=7.408e+02, percent-clipped=9.0 +2023-02-06 00:57:40,796 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36132.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:57:41,247 INFO [train.py:901] (3/4) Epoch 5, batch 3800, loss[loss=0.3073, simple_loss=0.3679, pruned_loss=0.1233, over 8109.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3527, pruned_loss=0.1153, over 1616536.60 frames. ], batch size: 23, lr: 1.50e-02, grad_scale: 8.0 +2023-02-06 00:57:41,320 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36133.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:57:48,141 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5813, 1.9475, 3.5466, 1.2065, 2.3912, 1.7675, 1.5892, 2.1776], + device='cuda:3'), covar=tensor([0.1326, 0.1719, 0.0484, 0.2836, 0.1291, 0.2275, 0.1400, 0.1861], + device='cuda:3'), in_proj_covar=tensor([0.0457, 0.0438, 0.0513, 0.0527, 0.0565, 0.0512, 0.0447, 0.0575], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 00:58:09,345 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36173.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:58:15,714 INFO [train.py:901] (3/4) Epoch 5, batch 3850, loss[loss=0.2898, simple_loss=0.3294, pruned_loss=0.1251, over 8142.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3516, pruned_loss=0.1152, over 1614316.52 frames. ], batch size: 22, lr: 1.50e-02, grad_scale: 8.0 +2023-02-06 00:58:27,240 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36199.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:58:41,077 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 00:58:45,749 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.620e+02 3.284e+02 4.097e+02 5.243e+02 1.380e+03, threshold=8.194e+02, percent-clipped=10.0 +2023-02-06 00:58:49,716 INFO [train.py:901] (3/4) Epoch 5, batch 3900, loss[loss=0.3211, simple_loss=0.3743, pruned_loss=0.134, over 8656.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3514, pruned_loss=0.1152, over 1616301.04 frames. ], batch size: 39, lr: 1.50e-02, grad_scale: 8.0 +2023-02-06 00:58:59,594 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36248.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:59:09,642 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36263.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:59:22,980 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5065, 1.4805, 4.4878, 1.8157, 2.1614, 5.1313, 4.8008, 4.4682], + device='cuda:3'), covar=tensor([0.1068, 0.1527, 0.0220, 0.1845, 0.1084, 0.0198, 0.0434, 0.0536], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0267, 0.0224, 0.0261, 0.0225, 0.0199, 0.0231, 0.0273], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 00:59:24,828 INFO [train.py:901] (3/4) Epoch 5, batch 3950, loss[loss=0.29, simple_loss=0.3585, pruned_loss=0.1108, over 8498.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.35, pruned_loss=0.1136, over 1616726.98 frames. ], batch size: 28, lr: 1.50e-02, grad_scale: 8.0 +2023-02-06 00:59:27,795 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4019, 2.0465, 3.1854, 2.6336, 2.6162, 2.1026, 1.5020, 1.2338], + device='cuda:3'), covar=tensor([0.1614, 0.1777, 0.0388, 0.0816, 0.0831, 0.0913, 0.0993, 0.2047], + device='cuda:3'), in_proj_covar=tensor([0.0739, 0.0667, 0.0567, 0.0659, 0.0759, 0.0623, 0.0597, 0.0613], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 00:59:28,405 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36288.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:59:28,429 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36288.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 00:59:46,710 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5371, 1.8173, 3.3432, 0.9891, 2.3634, 1.8332, 1.3398, 1.9743], + device='cuda:3'), covar=tensor([0.1429, 0.1800, 0.0618, 0.3306, 0.1340, 0.2317, 0.1599, 0.2174], + device='cuda:3'), in_proj_covar=tensor([0.0468, 0.0445, 0.0528, 0.0537, 0.0573, 0.0521, 0.0453, 0.0589], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 00:59:48,302 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.83 vs. limit=5.0 +2023-02-06 00:59:53,672 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2023-02-06 00:59:54,113 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5129, 1.4349, 2.8475, 1.0402, 1.9596, 3.0667, 2.9242, 2.5816], + device='cuda:3'), covar=tensor([0.1037, 0.1254, 0.0395, 0.1911, 0.0703, 0.0255, 0.0402, 0.0667], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0265, 0.0223, 0.0260, 0.0223, 0.0197, 0.0228, 0.0272], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 00:59:54,577 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.894e+02 2.959e+02 3.540e+02 4.519e+02 1.633e+03, threshold=7.079e+02, percent-clipped=6.0 +2023-02-06 00:59:58,379 INFO [train.py:901] (3/4) Epoch 5, batch 4000, loss[loss=0.2439, simple_loss=0.3135, pruned_loss=0.08711, over 8081.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3508, pruned_loss=0.1143, over 1613394.37 frames. ], batch size: 21, lr: 1.50e-02, grad_scale: 8.0 +2023-02-06 01:00:31,747 INFO [train.py:901] (3/4) Epoch 5, batch 4050, loss[loss=0.2819, simple_loss=0.3571, pruned_loss=0.1033, over 8013.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3505, pruned_loss=0.1146, over 1605728.60 frames. ], batch size: 22, lr: 1.50e-02, grad_scale: 8.0 +2023-02-06 01:01:03,218 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.718e+02 3.036e+02 3.577e+02 4.439e+02 7.437e+02, threshold=7.154e+02, percent-clipped=1.0 +2023-02-06 01:01:07,959 INFO [train.py:901] (3/4) Epoch 5, batch 4100, loss[loss=0.2671, simple_loss=0.3367, pruned_loss=0.09872, over 8083.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3504, pruned_loss=0.1142, over 1610608.60 frames. ], batch size: 21, lr: 1.50e-02, grad_scale: 8.0 +2023-02-06 01:01:41,918 INFO [train.py:901] (3/4) Epoch 5, batch 4150, loss[loss=0.2589, simple_loss=0.321, pruned_loss=0.09847, over 8650.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3486, pruned_loss=0.1133, over 1608929.99 frames. ], batch size: 34, lr: 1.50e-02, grad_scale: 8.0 +2023-02-06 01:01:44,924 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9147, 1.3855, 1.4698, 1.1647, 1.0969, 1.3461, 1.5384, 1.4272], + device='cuda:3'), covar=tensor([0.0608, 0.1255, 0.1948, 0.1598, 0.0666, 0.1615, 0.0774, 0.0602], + device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0181, 0.0221, 0.0184, 0.0133, 0.0192, 0.0146, 0.0155], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 01:01:56,717 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36504.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:02:13,608 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.587e+02 2.751e+02 3.740e+02 4.679e+02 9.033e+02, threshold=7.480e+02, percent-clipped=3.0 +2023-02-06 01:02:15,162 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36529.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:02:17,734 INFO [train.py:901] (3/4) Epoch 5, batch 4200, loss[loss=0.2973, simple_loss=0.3642, pruned_loss=0.1152, over 8482.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3483, pruned_loss=0.1129, over 1609107.08 frames. ], batch size: 27, lr: 1.49e-02, grad_scale: 8.0 +2023-02-06 01:02:24,878 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36543.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:02:25,684 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36544.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:02:37,294 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7441, 2.2274, 1.6161, 2.8289, 1.1427, 1.4520, 1.6555, 2.3581], + device='cuda:3'), covar=tensor([0.1281, 0.1131, 0.1662, 0.0518, 0.1829, 0.2200, 0.1701, 0.1040], + device='cuda:3'), in_proj_covar=tensor([0.0274, 0.0265, 0.0291, 0.0231, 0.0258, 0.0289, 0.0293, 0.0266], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 01:02:43,334 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 01:02:43,528 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36569.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:02:52,965 INFO [train.py:901] (3/4) Epoch 5, batch 4250, loss[loss=0.284, simple_loss=0.3435, pruned_loss=0.1123, over 7307.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.35, pruned_loss=0.1138, over 1610606.47 frames. ], batch size: 16, lr: 1.49e-02, grad_scale: 8.0 +2023-02-06 01:03:05,690 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 01:03:13,555 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4184, 1.4004, 4.5639, 1.7024, 3.9116, 3.7573, 4.0971, 4.0116], + device='cuda:3'), covar=tensor([0.0442, 0.3415, 0.0339, 0.2532, 0.1102, 0.0691, 0.0453, 0.0509], + device='cuda:3'), in_proj_covar=tensor([0.0324, 0.0481, 0.0404, 0.0416, 0.0476, 0.0400, 0.0387, 0.0442], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 01:03:20,905 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36623.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:03:24,279 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36626.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:03:24,817 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.959e+02 3.120e+02 3.802e+02 4.654e+02 9.583e+02, threshold=7.605e+02, percent-clipped=3.0 +2023-02-06 01:03:27,087 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1018, 4.1519, 3.6706, 1.7306, 3.6941, 3.6712, 3.8377, 3.3217], + device='cuda:3'), covar=tensor([0.0988, 0.0615, 0.1130, 0.5046, 0.0852, 0.1004, 0.1351, 0.0941], + device='cuda:3'), in_proj_covar=tensor([0.0387, 0.0279, 0.0318, 0.0399, 0.0312, 0.0270, 0.0298, 0.0241], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:3') +2023-02-06 01:03:27,131 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36630.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:03:28,982 INFO [train.py:901] (3/4) Epoch 5, batch 4300, loss[loss=0.3344, simple_loss=0.3732, pruned_loss=0.1478, over 8032.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3511, pruned_loss=0.1142, over 1614649.61 frames. ], batch size: 22, lr: 1.49e-02, grad_scale: 8.0 +2023-02-06 01:03:47,187 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36658.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:03:57,097 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-06 01:04:05,001 INFO [train.py:901] (3/4) Epoch 5, batch 4350, loss[loss=0.2247, simple_loss=0.2904, pruned_loss=0.07952, over 7653.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.351, pruned_loss=0.1137, over 1611445.24 frames. ], batch size: 19, lr: 1.49e-02, grad_scale: 8.0 +2023-02-06 01:04:28,883 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36718.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 01:04:34,836 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.000e+02 3.265e+02 4.032e+02 4.973e+02 1.053e+03, threshold=8.064e+02, percent-clipped=5.0 +2023-02-06 01:04:36,262 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 01:04:38,955 INFO [train.py:901] (3/4) Epoch 5, batch 4400, loss[loss=0.2149, simple_loss=0.2853, pruned_loss=0.07225, over 7786.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3509, pruned_loss=0.1145, over 1611858.91 frames. ], batch size: 19, lr: 1.49e-02, grad_scale: 8.0 +2023-02-06 01:05:15,297 INFO [train.py:901] (3/4) Epoch 5, batch 4450, loss[loss=0.281, simple_loss=0.3535, pruned_loss=0.1042, over 8084.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3512, pruned_loss=0.1145, over 1613249.05 frames. ], batch size: 21, lr: 1.49e-02, grad_scale: 8.0 +2023-02-06 01:05:18,070 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 01:05:43,198 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36823.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 01:05:44,599 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7026, 2.1665, 2.1843, 1.2141, 2.1326, 1.4424, 0.6403, 1.7260], + device='cuda:3'), covar=tensor([0.0179, 0.0084, 0.0068, 0.0166, 0.0149, 0.0336, 0.0284, 0.0104], + device='cuda:3'), in_proj_covar=tensor([0.0310, 0.0228, 0.0180, 0.0266, 0.0213, 0.0361, 0.0286, 0.0259], + device='cuda:3'), out_proj_covar=tensor([1.1198e-04, 8.0690e-05, 6.2385e-05, 9.3859e-05, 7.7463e-05, 1.3974e-04, + 1.0432e-04, 9.2529e-05], device='cuda:3') +2023-02-06 01:05:45,713 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.705e+02 3.029e+02 3.648e+02 4.687e+02 9.435e+02, threshold=7.296e+02, percent-clipped=4.0 +2023-02-06 01:05:49,701 INFO [train.py:901] (3/4) Epoch 5, batch 4500, loss[loss=0.2978, simple_loss=0.3575, pruned_loss=0.119, over 8187.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3512, pruned_loss=0.1147, over 1618150.12 frames. ], batch size: 23, lr: 1.49e-02, grad_scale: 8.0 +2023-02-06 01:06:14,965 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 01:06:26,112 INFO [train.py:901] (3/4) Epoch 5, batch 4550, loss[loss=0.3301, simple_loss=0.3702, pruned_loss=0.145, over 7971.00 frames. ], tot_loss[loss=0.2905, simple_loss=0.3515, pruned_loss=0.1148, over 1618349.35 frames. ], batch size: 21, lr: 1.49e-02, grad_scale: 8.0 +2023-02-06 01:06:47,961 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36914.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:06:56,551 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.694e+02 3.174e+02 3.779e+02 4.790e+02 8.988e+02, threshold=7.559e+02, percent-clipped=4.0 +2023-02-06 01:06:58,095 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36929.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:07:00,566 INFO [train.py:901] (3/4) Epoch 5, batch 4600, loss[loss=0.2655, simple_loss=0.3386, pruned_loss=0.09626, over 8323.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3514, pruned_loss=0.115, over 1616596.13 frames. ], batch size: 25, lr: 1.49e-02, grad_scale: 8.0 +2023-02-06 01:07:04,198 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.61 vs. limit=5.0 +2023-02-06 01:07:04,718 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36939.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:07:14,074 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3908, 1.9600, 1.7957, 0.8022, 1.9469, 1.5000, 0.3394, 1.6986], + device='cuda:3'), covar=tensor([0.0215, 0.0099, 0.0094, 0.0188, 0.0115, 0.0331, 0.0312, 0.0113], + device='cuda:3'), in_proj_covar=tensor([0.0309, 0.0225, 0.0179, 0.0265, 0.0212, 0.0359, 0.0282, 0.0258], + device='cuda:3'), out_proj_covar=tensor([1.1165e-04, 7.9618e-05, 6.1792e-05, 9.3386e-05, 7.6775e-05, 1.3866e-04, + 1.0261e-04, 9.2074e-05], device='cuda:3') +2023-02-06 01:07:14,112 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3327, 1.8987, 3.0776, 2.3980, 2.3852, 1.8372, 1.4388, 0.9369], + device='cuda:3'), covar=tensor([0.1699, 0.1938, 0.0418, 0.1026, 0.0921, 0.1027, 0.0961, 0.2148], + device='cuda:3'), in_proj_covar=tensor([0.0751, 0.0680, 0.0570, 0.0655, 0.0757, 0.0628, 0.0601, 0.0621], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 01:07:18,360 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-06 01:07:23,549 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36967.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:07:25,627 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36970.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:07:26,415 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.9352, 1.2136, 5.8581, 2.1816, 5.3779, 5.0141, 5.6045, 5.4614], + device='cuda:3'), covar=tensor([0.0358, 0.3462, 0.0187, 0.2074, 0.0658, 0.0414, 0.0256, 0.0321], + device='cuda:3'), in_proj_covar=tensor([0.0323, 0.0477, 0.0404, 0.0404, 0.0468, 0.0394, 0.0389, 0.0441], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 01:07:29,031 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36974.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:07:30,330 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36976.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:07:35,634 INFO [train.py:901] (3/4) Epoch 5, batch 4650, loss[loss=0.2731, simple_loss=0.3349, pruned_loss=0.1056, over 7659.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3497, pruned_loss=0.1137, over 1616819.03 frames. ], batch size: 19, lr: 1.49e-02, grad_scale: 8.0 +2023-02-06 01:08:02,860 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37022.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:08:06,067 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.888e+02 3.207e+02 3.974e+02 5.163e+02 9.904e+02, threshold=7.949e+02, percent-clipped=4.0 +2023-02-06 01:08:10,736 INFO [train.py:901] (3/4) Epoch 5, batch 4700, loss[loss=0.3006, simple_loss=0.3679, pruned_loss=0.1166, over 8326.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3498, pruned_loss=0.1133, over 1618203.26 frames. ], batch size: 25, lr: 1.48e-02, grad_scale: 8.0 +2023-02-06 01:08:29,886 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37062.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 01:08:43,597 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37082.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:08:44,113 INFO [train.py:901] (3/4) Epoch 5, batch 4750, loss[loss=0.2829, simple_loss=0.3463, pruned_loss=0.1098, over 8140.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3502, pruned_loss=0.1143, over 1616509.59 frames. ], batch size: 22, lr: 1.48e-02, grad_scale: 8.0 +2023-02-06 01:08:45,721 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37085.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:08:49,146 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37089.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:09:15,976 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.555e+02 3.010e+02 3.846e+02 4.879e+02 1.523e+03, threshold=7.692e+02, percent-clipped=5.0 +2023-02-06 01:09:17,401 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 01:09:20,175 INFO [train.py:901] (3/4) Epoch 5, batch 4800, loss[loss=0.3805, simple_loss=0.3987, pruned_loss=0.1811, over 7067.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3515, pruned_loss=0.1152, over 1615758.57 frames. ], batch size: 71, lr: 1.48e-02, grad_scale: 8.0 +2023-02-06 01:09:20,181 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 01:09:44,352 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37167.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 01:09:51,284 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37177.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 01:09:55,226 INFO [train.py:901] (3/4) Epoch 5, batch 4850, loss[loss=0.333, simple_loss=0.3773, pruned_loss=0.1444, over 6887.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.353, pruned_loss=0.1172, over 1614988.43 frames. ], batch size: 71, lr: 1.48e-02, grad_scale: 8.0 +2023-02-06 01:09:59,031 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.60 vs. limit=5.0 +2023-02-06 01:10:10,652 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 01:10:27,460 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.840e+02 2.956e+02 3.581e+02 4.871e+02 1.087e+03, threshold=7.163e+02, percent-clipped=6.0 +2023-02-06 01:10:31,482 INFO [train.py:901] (3/4) Epoch 5, batch 4900, loss[loss=0.2689, simple_loss=0.3436, pruned_loss=0.09711, over 8252.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3533, pruned_loss=0.1173, over 1616277.04 frames. ], batch size: 24, lr: 1.48e-02, grad_scale: 8.0 +2023-02-06 01:10:36,645 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-06 01:10:59,367 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37273.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:11:05,619 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37282.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 01:11:06,080 INFO [train.py:901] (3/4) Epoch 5, batch 4950, loss[loss=0.2818, simple_loss=0.36, pruned_loss=0.1018, over 8199.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3517, pruned_loss=0.1155, over 1614171.68 frames. ], batch size: 23, lr: 1.48e-02, grad_scale: 8.0 +2023-02-06 01:11:08,201 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37286.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:11:31,042 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37320.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:11:35,592 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.941e+02 3.052e+02 3.616e+02 4.696e+02 1.143e+03, threshold=7.231e+02, percent-clipped=5.0 +2023-02-06 01:11:40,306 INFO [train.py:901] (3/4) Epoch 5, batch 5000, loss[loss=0.2756, simple_loss=0.3323, pruned_loss=0.1094, over 8096.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3509, pruned_loss=0.1151, over 1615802.00 frames. ], batch size: 21, lr: 1.48e-02, grad_scale: 8.0 +2023-02-06 01:11:43,902 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37338.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:11:45,952 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37341.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:11:49,267 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37345.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:12:01,337 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37363.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:12:03,194 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37366.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:12:03,302 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37366.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:12:05,948 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37370.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:12:14,482 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37382.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:12:15,020 INFO [train.py:901] (3/4) Epoch 5, batch 5050, loss[loss=0.2664, simple_loss=0.318, pruned_loss=0.1074, over 7790.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3512, pruned_loss=0.1151, over 1613871.68 frames. ], batch size: 19, lr: 1.48e-02, grad_scale: 8.0 +2023-02-06 01:12:18,417 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37388.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:12:18,584 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-06 01:12:39,796 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 +2023-02-06 01:12:44,492 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.841e+02 3.496e+02 4.122e+02 5.072e+02 9.522e+02, threshold=8.245e+02, percent-clipped=6.0 +2023-02-06 01:12:48,504 INFO [train.py:901] (3/4) Epoch 5, batch 5100, loss[loss=0.2318, simple_loss=0.3004, pruned_loss=0.08158, over 7664.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3516, pruned_loss=0.1155, over 1615926.81 frames. ], batch size: 19, lr: 1.48e-02, grad_scale: 8.0 +2023-02-06 01:12:48,513 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 01:12:48,713 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37433.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 01:12:49,923 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37435.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:13:06,537 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37458.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 01:13:22,248 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37481.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:13:23,387 INFO [train.py:901] (3/4) Epoch 5, batch 5150, loss[loss=0.3387, simple_loss=0.3873, pruned_loss=0.145, over 8317.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3515, pruned_loss=0.1151, over 1613266.39 frames. ], batch size: 25, lr: 1.48e-02, grad_scale: 8.0 +2023-02-06 01:13:46,994 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.0353, 1.0653, 3.2323, 0.9406, 2.6972, 2.7461, 2.9077, 2.8804], + device='cuda:3'), covar=tensor([0.0703, 0.3473, 0.0648, 0.2548, 0.1557, 0.0759, 0.0709, 0.0747], + device='cuda:3'), in_proj_covar=tensor([0.0331, 0.0494, 0.0410, 0.0413, 0.0485, 0.0403, 0.0400, 0.0445], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 01:13:53,504 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.119e+02 3.138e+02 3.879e+02 5.454e+02 1.167e+03, threshold=7.757e+02, percent-clipped=4.0 +2023-02-06 01:13:57,569 INFO [train.py:901] (3/4) Epoch 5, batch 5200, loss[loss=0.2878, simple_loss=0.3543, pruned_loss=0.1107, over 8108.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3503, pruned_loss=0.1143, over 1608838.94 frames. ], batch size: 23, lr: 1.47e-02, grad_scale: 8.0 +2023-02-06 01:14:01,148 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37538.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 01:14:19,392 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37563.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 01:14:21,941 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2001, 1.2523, 2.0114, 0.9625, 1.8820, 2.2227, 2.2053, 1.8959], + device='cuda:3'), covar=tensor([0.1060, 0.1236, 0.0639, 0.1989, 0.0650, 0.0427, 0.0548, 0.0864], + device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0274, 0.0224, 0.0263, 0.0225, 0.0199, 0.0230, 0.0278], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 01:14:33,074 INFO [train.py:901] (3/4) Epoch 5, batch 5250, loss[loss=0.274, simple_loss=0.3498, pruned_loss=0.0991, over 8517.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3518, pruned_loss=0.1147, over 1616114.76 frames. ], batch size: 28, lr: 1.47e-02, grad_scale: 8.0 +2023-02-06 01:14:44,071 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6111, 1.7881, 1.8442, 1.5233, 1.0739, 1.8873, 0.2790, 1.0122], + device='cuda:3'), covar=tensor([0.3106, 0.1774, 0.1337, 0.2538, 0.5873, 0.0950, 0.5637, 0.2993], + device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0123, 0.0080, 0.0172, 0.0209, 0.0083, 0.0149, 0.0127], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 01:14:45,226 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 01:15:03,494 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.897e+02 3.205e+02 3.781e+02 5.298e+02 9.083e+02, threshold=7.562e+02, percent-clipped=4.0 +2023-02-06 01:15:05,555 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37630.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:15:07,536 INFO [train.py:901] (3/4) Epoch 5, batch 5300, loss[loss=0.2743, simple_loss=0.3523, pruned_loss=0.09814, over 8467.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3502, pruned_loss=0.1137, over 1612715.40 frames. ], batch size: 27, lr: 1.47e-02, grad_scale: 8.0 +2023-02-06 01:15:15,188 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37644.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:15:20,672 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4852, 1.4260, 2.7681, 1.1833, 1.8984, 2.9690, 2.9028, 2.5024], + device='cuda:3'), covar=tensor([0.1068, 0.1334, 0.0455, 0.1866, 0.0766, 0.0293, 0.0478, 0.0695], + device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0276, 0.0227, 0.0264, 0.0226, 0.0201, 0.0232, 0.0281], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 01:15:32,590 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37669.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:15:36,180 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 +2023-02-06 01:15:42,454 INFO [train.py:901] (3/4) Epoch 5, batch 5350, loss[loss=0.2617, simple_loss=0.3315, pruned_loss=0.09594, over 7981.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3493, pruned_loss=0.1127, over 1613747.84 frames. ], batch size: 21, lr: 1.47e-02, grad_scale: 8.0 +2023-02-06 01:15:47,822 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37691.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:16:05,033 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37716.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:16:11,802 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37726.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:16:12,341 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.725e+02 3.472e+02 4.206e+02 5.536e+02 1.524e+03, threshold=8.412e+02, percent-clipped=7.0 +2023-02-06 01:16:17,022 INFO [train.py:901] (3/4) Epoch 5, batch 5400, loss[loss=0.2357, simple_loss=0.2931, pruned_loss=0.08913, over 7452.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3493, pruned_loss=0.1131, over 1613649.07 frames. ], batch size: 17, lr: 1.47e-02, grad_scale: 8.0 +2023-02-06 01:16:19,918 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37737.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:16:25,229 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37745.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:16:36,760 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37762.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:16:40,898 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-06 01:16:50,838 INFO [train.py:901] (3/4) Epoch 5, batch 5450, loss[loss=0.3041, simple_loss=0.3513, pruned_loss=0.1284, over 7658.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3486, pruned_loss=0.1135, over 1609206.09 frames. ], batch size: 19, lr: 1.47e-02, grad_scale: 8.0 +2023-02-06 01:17:21,245 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4631, 1.4236, 1.5613, 1.3481, 1.2966, 1.3763, 1.7631, 1.6923], + device='cuda:3'), covar=tensor([0.0495, 0.1253, 0.1624, 0.1312, 0.0624, 0.1534, 0.0737, 0.0543], + device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0179, 0.0218, 0.0181, 0.0130, 0.0188, 0.0146, 0.0153], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 01:17:22,464 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.949e+02 2.936e+02 3.882e+02 5.021e+02 1.156e+03, threshold=7.764e+02, percent-clipped=3.0 +2023-02-06 01:17:23,312 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7552, 2.2143, 3.7059, 3.0836, 2.8719, 2.1009, 1.5762, 1.7260], + device='cuda:3'), covar=tensor([0.1751, 0.2344, 0.0501, 0.1057, 0.1170, 0.1012, 0.1032, 0.2285], + device='cuda:3'), in_proj_covar=tensor([0.0755, 0.0678, 0.0571, 0.0665, 0.0778, 0.0628, 0.0607, 0.0631], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 01:17:26,495 INFO [train.py:901] (3/4) Epoch 5, batch 5500, loss[loss=0.3377, simple_loss=0.3819, pruned_loss=0.1468, over 6904.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3492, pruned_loss=0.1139, over 1610578.90 frames. ], batch size: 71, lr: 1.47e-02, grad_scale: 8.0 +2023-02-06 01:17:30,474 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 01:17:32,020 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37841.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:17:40,095 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3818, 2.5132, 1.6105, 2.0479, 1.9548, 1.3564, 1.7624, 1.8064], + device='cuda:3'), covar=tensor([0.1150, 0.0258, 0.0867, 0.0479, 0.0541, 0.1074, 0.0848, 0.0724], + device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0242, 0.0312, 0.0302, 0.0318, 0.0308, 0.0343, 0.0314], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 01:18:00,739 INFO [train.py:901] (3/4) Epoch 5, batch 5550, loss[loss=0.334, simple_loss=0.3816, pruned_loss=0.1432, over 8242.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3489, pruned_loss=0.1139, over 1610305.67 frames. ], batch size: 24, lr: 1.47e-02, grad_scale: 8.0 +2023-02-06 01:18:32,131 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.468e+02 3.124e+02 3.941e+02 5.093e+02 9.977e+02, threshold=7.882e+02, percent-clipped=4.0 +2023-02-06 01:18:36,202 INFO [train.py:901] (3/4) Epoch 5, batch 5600, loss[loss=0.2767, simple_loss=0.3356, pruned_loss=0.1089, over 8036.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3481, pruned_loss=0.1132, over 1610355.56 frames. ], batch size: 22, lr: 1.47e-02, grad_scale: 8.0 +2023-02-06 01:18:40,423 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.6999, 1.1886, 3.8857, 1.4093, 3.3262, 3.2173, 3.4937, 3.3517], + device='cuda:3'), covar=tensor([0.0436, 0.3689, 0.0478, 0.2628, 0.1155, 0.0768, 0.0474, 0.0570], + device='cuda:3'), in_proj_covar=tensor([0.0329, 0.0491, 0.0412, 0.0421, 0.0480, 0.0407, 0.0397, 0.0447], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 01:18:46,649 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-06 01:19:10,749 INFO [train.py:901] (3/4) Epoch 5, batch 5650, loss[loss=0.285, simple_loss=0.3542, pruned_loss=0.1079, over 8245.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3489, pruned_loss=0.1138, over 1609890.06 frames. ], batch size: 24, lr: 1.47e-02, grad_scale: 4.0 +2023-02-06 01:19:11,865 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.60 vs. limit=5.0 +2023-02-06 01:19:20,217 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37997.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:19:23,903 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38001.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:19:33,263 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7828, 2.1720, 4.6317, 1.1560, 3.1263, 2.4956, 1.6413, 2.4429], + device='cuda:3'), covar=tensor([0.1402, 0.1972, 0.0655, 0.3339, 0.1281, 0.2140, 0.1519, 0.2363], + device='cuda:3'), in_proj_covar=tensor([0.0463, 0.0446, 0.0530, 0.0532, 0.0577, 0.0516, 0.0446, 0.0588], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 01:19:34,403 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 01:19:41,416 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38026.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:19:42,548 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.040e+02 3.005e+02 3.717e+02 4.758e+02 1.120e+03, threshold=7.434e+02, percent-clipped=3.0 +2023-02-06 01:19:45,907 INFO [train.py:901] (3/4) Epoch 5, batch 5700, loss[loss=0.2865, simple_loss=0.3566, pruned_loss=0.1082, over 8105.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3519, pruned_loss=0.1163, over 1607095.02 frames. ], batch size: 23, lr: 1.46e-02, grad_scale: 4.0 +2023-02-06 01:19:58,761 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5292, 4.5192, 4.0787, 1.9780, 4.1100, 4.0978, 4.1922, 3.8457], + device='cuda:3'), covar=tensor([0.0598, 0.0428, 0.0768, 0.3931, 0.0601, 0.0565, 0.0865, 0.0564], + device='cuda:3'), in_proj_covar=tensor([0.0399, 0.0287, 0.0319, 0.0405, 0.0321, 0.0269, 0.0298, 0.0250], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 01:20:20,737 INFO [train.py:901] (3/4) Epoch 5, batch 5750, loss[loss=0.2719, simple_loss=0.3422, pruned_loss=0.1008, over 8351.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3504, pruned_loss=0.1149, over 1607608.02 frames. ], batch size: 26, lr: 1.46e-02, grad_scale: 4.0 +2023-02-06 01:20:30,236 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38097.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:20:37,196 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 01:20:46,792 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38122.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:20:50,580 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.930e+02 3.053e+02 3.876e+02 4.925e+02 1.023e+03, threshold=7.752e+02, percent-clipped=4.0 +2023-02-06 01:20:54,528 INFO [train.py:901] (3/4) Epoch 5, batch 5800, loss[loss=0.2835, simple_loss=0.358, pruned_loss=0.1045, over 8456.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.351, pruned_loss=0.1146, over 1613683.07 frames. ], batch size: 29, lr: 1.46e-02, grad_scale: 4.0 +2023-02-06 01:20:58,012 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9486, 2.3104, 2.9667, 1.1043, 2.9592, 1.6703, 1.4085, 1.5788], + device='cuda:3'), covar=tensor([0.0289, 0.0129, 0.0086, 0.0278, 0.0176, 0.0377, 0.0373, 0.0207], + device='cuda:3'), in_proj_covar=tensor([0.0315, 0.0224, 0.0183, 0.0269, 0.0214, 0.0359, 0.0282, 0.0261], + device='cuda:3'), out_proj_covar=tensor([1.1298e-04, 7.7778e-05, 6.2967e-05, 9.3688e-05, 7.6369e-05, 1.3633e-04, + 1.0106e-04, 9.1467e-05], device='cuda:3') +2023-02-06 01:21:14,944 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4257, 1.6280, 1.5521, 1.3900, 1.4000, 1.5112, 1.8202, 1.8796], + device='cuda:3'), covar=tensor([0.0551, 0.1291, 0.1806, 0.1435, 0.0647, 0.1653, 0.0810, 0.0531], + device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0180, 0.0219, 0.0183, 0.0131, 0.0191, 0.0145, 0.0153], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 01:21:29,415 INFO [train.py:901] (3/4) Epoch 5, batch 5850, loss[loss=0.2818, simple_loss=0.3555, pruned_loss=0.104, over 8510.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3506, pruned_loss=0.114, over 1615078.82 frames. ], batch size: 28, lr: 1.46e-02, grad_scale: 4.0 +2023-02-06 01:21:31,398 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 +2023-02-06 01:21:57,290 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.5737, 3.5160, 3.0653, 2.0035, 3.0645, 3.1877, 3.3079, 2.8165], + device='cuda:3'), covar=tensor([0.0954, 0.0749, 0.1061, 0.3855, 0.0833, 0.0895, 0.1090, 0.1019], + device='cuda:3'), in_proj_covar=tensor([0.0404, 0.0289, 0.0323, 0.0409, 0.0318, 0.0271, 0.0304, 0.0255], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 01:22:01,294 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.005e+02 3.016e+02 3.759e+02 4.889e+02 1.185e+03, threshold=7.518e+02, percent-clipped=2.0 +2023-02-06 01:22:04,818 INFO [train.py:901] (3/4) Epoch 5, batch 5900, loss[loss=0.2573, simple_loss=0.3109, pruned_loss=0.1018, over 7790.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3478, pruned_loss=0.1125, over 1610963.55 frames. ], batch size: 19, lr: 1.46e-02, grad_scale: 4.0 +2023-02-06 01:22:40,895 INFO [train.py:901] (3/4) Epoch 5, batch 5950, loss[loss=0.3692, simple_loss=0.3935, pruned_loss=0.1724, over 6938.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3483, pruned_loss=0.1127, over 1613891.67 frames. ], batch size: 71, lr: 1.46e-02, grad_scale: 4.0 +2023-02-06 01:23:12,028 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.179e+02 3.191e+02 3.790e+02 5.332e+02 1.075e+03, threshold=7.580e+02, percent-clipped=7.0 +2023-02-06 01:23:15,479 INFO [train.py:901] (3/4) Epoch 5, batch 6000, loss[loss=0.2447, simple_loss=0.3277, pruned_loss=0.08082, over 8249.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.347, pruned_loss=0.1122, over 1610555.36 frames. ], batch size: 24, lr: 1.46e-02, grad_scale: 8.0 +2023-02-06 01:23:15,479 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 01:23:23,744 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4417, 1.7206, 2.7938, 1.0934, 2.0820, 1.6816, 1.5074, 1.7903], + device='cuda:3'), covar=tensor([0.1518, 0.2101, 0.0717, 0.3574, 0.1479, 0.2532, 0.1562, 0.1993], + device='cuda:3'), in_proj_covar=tensor([0.0464, 0.0446, 0.0522, 0.0537, 0.0577, 0.0520, 0.0443, 0.0593], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 01:23:28,272 INFO [train.py:935] (3/4) Epoch 5, validation: loss=0.2196, simple_loss=0.3162, pruned_loss=0.06146, over 944034.00 frames. +2023-02-06 01:23:28,273 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 01:23:33,799 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=38341.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:23:56,506 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3283, 1.5947, 3.0940, 1.0346, 2.1760, 1.7881, 1.4375, 1.7477], + device='cuda:3'), covar=tensor([0.1966, 0.2261, 0.0777, 0.3935, 0.1575, 0.2634, 0.1845, 0.2429], + device='cuda:3'), in_proj_covar=tensor([0.0465, 0.0446, 0.0524, 0.0537, 0.0579, 0.0519, 0.0444, 0.0595], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 01:23:58,484 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38378.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:24:01,710 INFO [train.py:901] (3/4) Epoch 5, batch 6050, loss[loss=0.2926, simple_loss=0.3397, pruned_loss=0.1228, over 7653.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3472, pruned_loss=0.1126, over 1613142.81 frames. ], batch size: 19, lr: 1.46e-02, grad_scale: 8.0 +2023-02-06 01:24:33,770 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.931e+02 3.108e+02 3.868e+02 4.827e+02 8.119e+02, threshold=7.737e+02, percent-clipped=1.0 +2023-02-06 01:24:37,062 INFO [train.py:901] (3/4) Epoch 5, batch 6100, loss[loss=0.2257, simple_loss=0.2875, pruned_loss=0.08197, over 7544.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3486, pruned_loss=0.1137, over 1613120.73 frames. ], batch size: 18, lr: 1.46e-02, grad_scale: 8.0 +2023-02-06 01:24:53,399 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38456.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:25:09,655 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 01:25:10,930 INFO [train.py:901] (3/4) Epoch 5, batch 6150, loss[loss=0.3141, simple_loss=0.3796, pruned_loss=0.1243, over 8443.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3488, pruned_loss=0.1137, over 1613830.54 frames. ], batch size: 27, lr: 1.46e-02, grad_scale: 8.0 +2023-02-06 01:25:29,207 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5191, 1.8980, 2.0044, 1.4637, 0.8855, 2.0611, 0.3239, 1.0301], + device='cuda:3'), covar=tensor([0.3517, 0.2308, 0.1255, 0.3057, 0.6500, 0.0627, 0.5857, 0.3184], + device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0127, 0.0083, 0.0173, 0.0216, 0.0083, 0.0145, 0.0131], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 01:25:42,289 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.784e+02 3.089e+02 4.008e+02 5.119e+02 1.011e+03, threshold=8.016e+02, percent-clipped=7.0 +2023-02-06 01:25:44,511 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1723, 1.3136, 2.3221, 1.0299, 2.0900, 2.4779, 2.4610, 2.1430], + device='cuda:3'), covar=tensor([0.1029, 0.1091, 0.0487, 0.1870, 0.0554, 0.0362, 0.0502, 0.0786], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0267, 0.0220, 0.0260, 0.0222, 0.0198, 0.0228, 0.0271], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 01:25:45,625 INFO [train.py:901] (3/4) Epoch 5, batch 6200, loss[loss=0.3147, simple_loss=0.3631, pruned_loss=0.1332, over 7973.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3482, pruned_loss=0.1132, over 1609855.47 frames. ], batch size: 21, lr: 1.46e-02, grad_scale: 8.0 +2023-02-06 01:26:20,118 INFO [train.py:901] (3/4) Epoch 5, batch 6250, loss[loss=0.271, simple_loss=0.3216, pruned_loss=0.1102, over 7269.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3478, pruned_loss=0.1129, over 1609466.27 frames. ], batch size: 16, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:26:29,033 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2764, 1.9495, 3.1794, 2.3558, 2.6026, 1.9443, 1.5115, 1.2266], + device='cuda:3'), covar=tensor([0.1893, 0.2035, 0.0445, 0.1112, 0.0971, 0.1056, 0.1020, 0.2146], + device='cuda:3'), in_proj_covar=tensor([0.0758, 0.0685, 0.0580, 0.0666, 0.0783, 0.0641, 0.0612, 0.0634], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 01:26:36,084 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.93 vs. limit=2.0 +2023-02-06 01:26:51,177 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.140e+02 3.239e+02 3.994e+02 4.997e+02 1.061e+03, threshold=7.988e+02, percent-clipped=3.0 +2023-02-06 01:26:54,600 INFO [train.py:901] (3/4) Epoch 5, batch 6300, loss[loss=0.2862, simple_loss=0.3676, pruned_loss=0.1024, over 8459.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.348, pruned_loss=0.1132, over 1608628.98 frames. ], batch size: 25, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:27:29,613 INFO [train.py:901] (3/4) Epoch 5, batch 6350, loss[loss=0.3036, simple_loss=0.3487, pruned_loss=0.1292, over 7934.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3486, pruned_loss=0.1132, over 1608957.11 frames. ], batch size: 20, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:27:49,936 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38712.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:27:56,425 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=38722.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:28:00,326 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.824e+02 2.862e+02 3.826e+02 4.732e+02 1.596e+03, threshold=7.652e+02, percent-clipped=5.0 +2023-02-06 01:28:03,612 INFO [train.py:901] (3/4) Epoch 5, batch 6400, loss[loss=0.3239, simple_loss=0.3763, pruned_loss=0.1357, over 7813.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3483, pruned_loss=0.1126, over 1613052.56 frames. ], batch size: 20, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:28:06,534 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38737.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:28:38,892 INFO [train.py:901] (3/4) Epoch 5, batch 6450, loss[loss=0.2446, simple_loss=0.3054, pruned_loss=0.09191, over 7935.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3493, pruned_loss=0.1133, over 1614084.19 frames. ], batch size: 20, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:28:59,421 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4852, 2.4794, 2.7475, 2.0229, 1.6245, 2.5916, 0.9042, 1.8929], + device='cuda:3'), covar=tensor([0.2434, 0.1775, 0.0608, 0.2491, 0.4383, 0.0525, 0.5502, 0.2064], + device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0123, 0.0078, 0.0166, 0.0207, 0.0077, 0.0140, 0.0124], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 01:29:09,817 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.470e+02 3.536e+02 4.141e+02 5.010e+02 9.096e+02, threshold=8.281e+02, percent-clipped=4.0 +2023-02-06 01:29:13,135 INFO [train.py:901] (3/4) Epoch 5, batch 6500, loss[loss=0.3023, simple_loss=0.3699, pruned_loss=0.1173, over 8569.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3509, pruned_loss=0.1145, over 1617220.84 frames. ], batch size: 31, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:29:16,034 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38837.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:29:18,659 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38841.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:29:18,852 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0 +2023-02-06 01:29:38,053 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4857, 1.5164, 1.4316, 1.1133, 1.2119, 1.3947, 1.7086, 1.7266], + device='cuda:3'), covar=tensor([0.0546, 0.1411, 0.1922, 0.1618, 0.0691, 0.1758, 0.0825, 0.0597], + device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0179, 0.0221, 0.0186, 0.0131, 0.0189, 0.0144, 0.0154], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 01:29:48,037 INFO [train.py:901] (3/4) Epoch 5, batch 6550, loss[loss=0.2774, simple_loss=0.3458, pruned_loss=0.1045, over 8357.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3486, pruned_loss=0.1129, over 1616066.91 frames. ], batch size: 24, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:30:09,430 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0 +2023-02-06 01:30:19,268 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.070e+02 3.239e+02 3.737e+02 4.952e+02 1.438e+03, threshold=7.474e+02, percent-clipped=4.0 +2023-02-06 01:30:22,018 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 01:30:22,217 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4550, 1.8045, 1.9463, 1.4495, 1.0673, 2.1216, 0.2911, 1.0274], + device='cuda:3'), covar=tensor([0.3221, 0.2443, 0.1049, 0.2996, 0.6551, 0.0676, 0.5764, 0.2919], + device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0127, 0.0083, 0.0172, 0.0215, 0.0079, 0.0142, 0.0129], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 01:30:22,676 INFO [train.py:901] (3/4) Epoch 5, batch 6600, loss[loss=0.2524, simple_loss=0.3098, pruned_loss=0.09749, over 7657.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3485, pruned_loss=0.1133, over 1613708.93 frames. ], batch size: 19, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:30:39,837 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 01:30:58,095 INFO [train.py:901] (3/4) Epoch 5, batch 6650, loss[loss=0.3017, simple_loss=0.3443, pruned_loss=0.1295, over 7223.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3476, pruned_loss=0.113, over 1603961.34 frames. ], batch size: 16, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:31:29,853 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.971e+02 3.068e+02 3.660e+02 4.252e+02 1.265e+03, threshold=7.321e+02, percent-clipped=3.0 +2023-02-06 01:31:33,319 INFO [train.py:901] (3/4) Epoch 5, batch 6700, loss[loss=0.2453, simple_loss=0.3024, pruned_loss=0.09411, over 7694.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3473, pruned_loss=0.1119, over 1608023.51 frames. ], batch size: 18, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:32:02,064 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.1064, 1.0730, 3.1944, 1.0054, 2.7102, 2.7034, 2.8790, 2.7746], + device='cuda:3'), covar=tensor([0.0618, 0.3877, 0.0623, 0.2841, 0.1527, 0.0817, 0.0694, 0.0795], + device='cuda:3'), in_proj_covar=tensor([0.0339, 0.0497, 0.0425, 0.0426, 0.0494, 0.0408, 0.0406, 0.0451], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 01:32:08,097 INFO [train.py:901] (3/4) Epoch 5, batch 6750, loss[loss=0.2317, simple_loss=0.3114, pruned_loss=0.07598, over 7923.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3466, pruned_loss=0.1116, over 1608722.08 frames. ], batch size: 20, lr: 1.45e-02, grad_scale: 8.0 +2023-02-06 01:32:15,830 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39093.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:32:33,105 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39118.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:32:40,328 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.849e+02 2.802e+02 3.437e+02 4.300e+02 8.945e+02, threshold=6.874e+02, percent-clipped=2.0 +2023-02-06 01:32:43,702 INFO [train.py:901] (3/4) Epoch 5, batch 6800, loss[loss=0.296, simple_loss=0.3681, pruned_loss=0.1119, over 8292.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3464, pruned_loss=0.1115, over 1608619.60 frames. ], batch size: 23, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:32:54,741 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 01:32:59,187 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-06 01:33:17,611 INFO [train.py:901] (3/4) Epoch 5, batch 6850, loss[loss=0.2461, simple_loss=0.2981, pruned_loss=0.09707, over 7428.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3481, pruned_loss=0.1129, over 1609073.43 frames. ], batch size: 17, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:33:19,091 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39185.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:33:31,848 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39203.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:33:43,504 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 01:33:48,721 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.815e+02 3.284e+02 3.960e+02 5.468e+02 1.321e+03, threshold=7.919e+02, percent-clipped=11.0 +2023-02-06 01:33:52,156 INFO [train.py:901] (3/4) Epoch 5, batch 6900, loss[loss=0.3041, simple_loss=0.363, pruned_loss=0.1226, over 7935.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3476, pruned_loss=0.1125, over 1610001.66 frames. ], batch size: 20, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:34:18,979 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 01:34:26,657 INFO [train.py:901] (3/4) Epoch 5, batch 6950, loss[loss=0.3655, simple_loss=0.4003, pruned_loss=0.1654, over 7255.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3486, pruned_loss=0.113, over 1611331.01 frames. ], batch size: 71, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:34:38,156 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39300.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:34:39,404 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39302.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:34:49,520 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 01:34:58,304 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.949e+02 3.231e+02 3.801e+02 5.196e+02 1.038e+03, threshold=7.603e+02, percent-clipped=4.0 +2023-02-06 01:35:01,666 INFO [train.py:901] (3/4) Epoch 5, batch 7000, loss[loss=0.2827, simple_loss=0.359, pruned_loss=0.1032, over 8463.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3468, pruned_loss=0.1121, over 1604681.36 frames. ], batch size: 29, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:35:34,098 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6833, 2.9478, 1.9520, 2.2241, 2.5011, 1.6198, 2.3144, 2.2636], + device='cuda:3'), covar=tensor([0.1118, 0.0225, 0.0809, 0.0558, 0.0460, 0.1110, 0.0677, 0.0751], + device='cuda:3'), in_proj_covar=tensor([0.0342, 0.0241, 0.0307, 0.0300, 0.0312, 0.0313, 0.0333, 0.0305], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 01:35:35,822 INFO [train.py:901] (3/4) Epoch 5, batch 7050, loss[loss=0.2411, simple_loss=0.3097, pruned_loss=0.08627, over 6802.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3455, pruned_loss=0.1114, over 1600199.61 frames. ], batch size: 15, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:36:06,897 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.068e+02 2.867e+02 3.538e+02 4.706e+02 1.662e+03, threshold=7.075e+02, percent-clipped=2.0 +2023-02-06 01:36:09,166 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8122, 2.2798, 3.8188, 3.1079, 3.0668, 2.2964, 1.5007, 1.8185], + device='cuda:3'), covar=tensor([0.1829, 0.2459, 0.0508, 0.1065, 0.1097, 0.1056, 0.1119, 0.2319], + device='cuda:3'), in_proj_covar=tensor([0.0756, 0.0686, 0.0581, 0.0673, 0.0777, 0.0645, 0.0607, 0.0633], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 01:36:10,286 INFO [train.py:901] (3/4) Epoch 5, batch 7100, loss[loss=0.3854, simple_loss=0.4105, pruned_loss=0.1801, over 7125.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3456, pruned_loss=0.1114, over 1597881.84 frames. ], batch size: 71, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:36:38,200 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2640, 2.5069, 1.7368, 2.1774, 2.0611, 1.4860, 1.8029, 1.9480], + device='cuda:3'), covar=tensor([0.1215, 0.0255, 0.0789, 0.0456, 0.0613, 0.1025, 0.0895, 0.0757], + device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0240, 0.0310, 0.0303, 0.0321, 0.0317, 0.0341, 0.0312], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 01:36:44,804 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39481.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:36:46,000 INFO [train.py:901] (3/4) Epoch 5, batch 7150, loss[loss=0.2512, simple_loss=0.3342, pruned_loss=0.08407, over 8192.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3457, pruned_loss=0.1118, over 1598110.90 frames. ], batch size: 23, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:37:17,185 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.977e+02 2.912e+02 3.907e+02 4.774e+02 1.202e+03, threshold=7.813e+02, percent-clipped=7.0 +2023-02-06 01:37:20,761 INFO [train.py:901] (3/4) Epoch 5, batch 7200, loss[loss=0.2728, simple_loss=0.3458, pruned_loss=0.09995, over 8469.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3465, pruned_loss=0.1124, over 1600275.53 frames. ], batch size: 25, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:37:22,385 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5287, 1.9792, 3.6013, 1.1286, 2.5100, 1.8896, 1.5459, 2.0404], + device='cuda:3'), covar=tensor([0.1435, 0.1829, 0.0476, 0.3045, 0.1233, 0.2235, 0.1442, 0.2069], + device='cuda:3'), in_proj_covar=tensor([0.0470, 0.0453, 0.0520, 0.0529, 0.0570, 0.0514, 0.0448, 0.0592], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 01:37:30,595 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39547.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:37:37,491 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39556.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:37:55,145 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39581.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:37:56,307 INFO [train.py:901] (3/4) Epoch 5, batch 7250, loss[loss=0.3072, simple_loss=0.3647, pruned_loss=0.1249, over 8234.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3473, pruned_loss=0.1129, over 1603876.23 frames. ], batch size: 24, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:38:09,671 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4366, 2.1875, 1.3646, 1.9695, 1.8799, 1.0797, 1.4096, 1.8110], + device='cuda:3'), covar=tensor([0.1083, 0.0348, 0.1163, 0.0509, 0.0655, 0.1479, 0.1041, 0.0736], + device='cuda:3'), in_proj_covar=tensor([0.0344, 0.0242, 0.0309, 0.0303, 0.0320, 0.0315, 0.0340, 0.0311], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 01:38:27,209 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.040e+02 2.862e+02 3.679e+02 5.056e+02 1.142e+03, threshold=7.358e+02, percent-clipped=8.0 +2023-02-06 01:38:30,502 INFO [train.py:901] (3/4) Epoch 5, batch 7300, loss[loss=0.3378, simple_loss=0.3971, pruned_loss=0.1393, over 8480.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3473, pruned_loss=0.1122, over 1610156.13 frames. ], batch size: 28, lr: 1.44e-02, grad_scale: 8.0 +2023-02-06 01:38:33,401 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5567, 1.3184, 1.3948, 1.2057, 1.0935, 1.2489, 1.2531, 1.4494], + device='cuda:3'), covar=tensor([0.0610, 0.1283, 0.1794, 0.1407, 0.0589, 0.1622, 0.0732, 0.0541], + device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0178, 0.0219, 0.0183, 0.0131, 0.0189, 0.0144, 0.0151], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 01:38:39,558 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39646.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:38:47,121 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39657.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 01:38:50,516 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39662.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:39:05,559 INFO [train.py:901] (3/4) Epoch 5, batch 7350, loss[loss=0.3419, simple_loss=0.3964, pruned_loss=0.1437, over 8292.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3489, pruned_loss=0.1135, over 1611228.72 frames. ], batch size: 23, lr: 1.43e-02, grad_scale: 8.0 +2023-02-06 01:39:10,023 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-06 01:39:12,259 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.5730, 3.3719, 3.1463, 1.9870, 3.0535, 3.1655, 3.2269, 2.8538], + device='cuda:3'), covar=tensor([0.0871, 0.0777, 0.1021, 0.3750, 0.0810, 0.0940, 0.1221, 0.0882], + device='cuda:3'), in_proj_covar=tensor([0.0392, 0.0282, 0.0310, 0.0399, 0.0302, 0.0270, 0.0292, 0.0243], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:3') +2023-02-06 01:39:33,447 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 01:39:36,162 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.150e+02 2.846e+02 3.982e+02 4.999e+02 1.878e+03, threshold=7.964e+02, percent-clipped=11.0 +2023-02-06 01:39:39,633 INFO [train.py:901] (3/4) Epoch 5, batch 7400, loss[loss=0.3359, simple_loss=0.3877, pruned_loss=0.1421, over 8471.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3479, pruned_loss=0.113, over 1607873.39 frames. ], batch size: 25, lr: 1.43e-02, grad_scale: 8.0 +2023-02-06 01:39:45,832 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7301, 3.0561, 3.2394, 1.9168, 1.3217, 3.1775, 0.5364, 2.0484], + device='cuda:3'), covar=tensor([0.4860, 0.1621, 0.0883, 0.3746, 0.6774, 0.1011, 0.6890, 0.2335], + device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0131, 0.0083, 0.0179, 0.0221, 0.0084, 0.0146, 0.0133], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 01:39:52,995 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 01:39:59,175 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39761.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:40:13,746 INFO [train.py:901] (3/4) Epoch 5, batch 7450, loss[loss=0.2421, simple_loss=0.3196, pruned_loss=0.08226, over 8468.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3473, pruned_loss=0.1121, over 1610483.18 frames. ], batch size: 27, lr: 1.43e-02, grad_scale: 8.0 +2023-02-06 01:40:15,954 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39785.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:40:32,190 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 01:40:43,642 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39825.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:40:45,624 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.975e+02 3.116e+02 3.779e+02 4.440e+02 1.107e+03, threshold=7.558e+02, percent-clipped=3.0 +2023-02-06 01:40:49,054 INFO [train.py:901] (3/4) Epoch 5, batch 7500, loss[loss=0.2471, simple_loss=0.3254, pruned_loss=0.08438, over 8106.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3469, pruned_loss=0.1117, over 1611241.41 frames. ], batch size: 23, lr: 1.43e-02, grad_scale: 8.0 +2023-02-06 01:40:49,201 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39833.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:41:07,198 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1655, 1.4580, 1.4358, 1.2061, 1.2842, 1.2901, 1.6234, 1.5576], + device='cuda:3'), covar=tensor([0.0590, 0.1212, 0.1768, 0.1441, 0.0585, 0.1598, 0.0745, 0.0585], + device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0178, 0.0218, 0.0181, 0.0128, 0.0188, 0.0143, 0.0150], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 01:41:22,994 INFO [train.py:901] (3/4) Epoch 5, batch 7550, loss[loss=0.2858, simple_loss=0.3331, pruned_loss=0.1193, over 7567.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.347, pruned_loss=0.1126, over 1609389.22 frames. ], batch size: 18, lr: 1.43e-02, grad_scale: 8.0 +2023-02-06 01:41:41,173 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3068, 1.0386, 4.4795, 1.8089, 3.7663, 3.7617, 3.9999, 3.8644], + device='cuda:3'), covar=tensor([0.0402, 0.3709, 0.0365, 0.2258, 0.1086, 0.0653, 0.0402, 0.0528], + device='cuda:3'), in_proj_covar=tensor([0.0337, 0.0486, 0.0429, 0.0425, 0.0493, 0.0411, 0.0399, 0.0451], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 01:41:48,007 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39918.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:41:54,538 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.961e+02 3.144e+02 3.978e+02 5.379e+02 1.554e+03, threshold=7.957e+02, percent-clipped=6.0 +2023-02-06 01:41:57,840 INFO [train.py:901] (3/4) Epoch 5, batch 7600, loss[loss=0.2583, simple_loss=0.3357, pruned_loss=0.09048, over 7811.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3465, pruned_loss=0.1123, over 1606718.00 frames. ], batch size: 20, lr: 1.43e-02, grad_scale: 8.0 +2023-02-06 01:42:02,473 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39940.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:42:04,534 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39943.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:42:32,747 INFO [train.py:901] (3/4) Epoch 5, batch 7650, loss[loss=0.2532, simple_loss=0.3309, pruned_loss=0.08772, over 8293.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3469, pruned_loss=0.1121, over 1609683.15 frames. ], batch size: 23, lr: 1.43e-02, grad_scale: 16.0 +2023-02-06 01:42:45,885 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40001.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 01:42:56,999 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40017.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:43:05,594 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.281e+02 3.028e+02 3.689e+02 4.703e+02 1.290e+03, threshold=7.379e+02, percent-clipped=1.0 +2023-02-06 01:43:08,878 INFO [train.py:901] (3/4) Epoch 5, batch 7700, loss[loss=0.2731, simple_loss=0.3404, pruned_loss=0.1028, over 8517.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3481, pruned_loss=0.1128, over 1613266.47 frames. ], batch size: 28, lr: 1.43e-02, grad_scale: 16.0 +2023-02-06 01:43:15,433 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40042.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:43:15,443 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3105, 1.5562, 1.7139, 1.3153, 0.7626, 1.6853, 0.0724, 0.9333], + device='cuda:3'), covar=tensor([0.3783, 0.2102, 0.0875, 0.2133, 0.6117, 0.0712, 0.4735, 0.2601], + device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0126, 0.0078, 0.0173, 0.0212, 0.0083, 0.0141, 0.0130], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 01:43:21,219 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-06 01:43:44,274 INFO [train.py:901] (3/4) Epoch 5, batch 7750, loss[loss=0.2642, simple_loss=0.3176, pruned_loss=0.1054, over 7542.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3472, pruned_loss=0.1125, over 1610945.67 frames. ], batch size: 18, lr: 1.43e-02, grad_scale: 16.0 +2023-02-06 01:43:44,292 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 01:44:07,435 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40116.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 01:44:12,724 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5271, 4.5322, 4.0434, 1.7042, 4.0962, 3.9157, 4.1860, 3.8121], + device='cuda:3'), covar=tensor([0.0696, 0.0553, 0.0868, 0.5123, 0.0636, 0.0739, 0.1249, 0.0559], + device='cuda:3'), in_proj_covar=tensor([0.0390, 0.0277, 0.0313, 0.0402, 0.0303, 0.0271, 0.0292, 0.0247], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001], + device='cuda:3') +2023-02-06 01:44:15,426 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.050e+02 3.016e+02 3.638e+02 4.428e+02 8.911e+02, threshold=7.276e+02, percent-clipped=8.0 +2023-02-06 01:44:16,238 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40129.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:44:18,934 INFO [train.py:901] (3/4) Epoch 5, batch 7800, loss[loss=0.3182, simple_loss=0.3852, pruned_loss=0.1256, over 8502.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3474, pruned_loss=0.1124, over 1614259.00 frames. ], batch size: 26, lr: 1.43e-02, grad_scale: 16.0 +2023-02-06 01:44:50,152 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40177.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:44:54,256 INFO [train.py:901] (3/4) Epoch 5, batch 7850, loss[loss=0.2829, simple_loss=0.3545, pruned_loss=0.1057, over 8460.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3485, pruned_loss=0.1127, over 1619158.16 frames. ], batch size: 29, lr: 1.43e-02, grad_scale: 16.0 +2023-02-06 01:45:03,200 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40196.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:45:14,682 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40213.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:45:20,124 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40221.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:45:24,748 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.315e+02 3.285e+02 3.978e+02 4.753e+02 1.108e+03, threshold=7.955e+02, percent-clipped=4.0 +2023-02-06 01:45:28,277 INFO [train.py:901] (3/4) Epoch 5, batch 7900, loss[loss=0.2928, simple_loss=0.3601, pruned_loss=0.1127, over 8251.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3478, pruned_loss=0.1118, over 1618338.19 frames. ], batch size: 24, lr: 1.42e-02, grad_scale: 16.0 +2023-02-06 01:45:30,497 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40236.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:45:35,930 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40244.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:46:02,027 INFO [train.py:901] (3/4) Epoch 5, batch 7950, loss[loss=0.2541, simple_loss=0.3083, pruned_loss=0.09988, over 7701.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3473, pruned_loss=0.1116, over 1611991.08 frames. ], batch size: 18, lr: 1.42e-02, grad_scale: 16.0 +2023-02-06 01:46:07,555 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40290.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:46:09,020 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40292.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:46:24,696 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-06 01:46:33,072 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.942e+02 3.003e+02 3.931e+02 4.743e+02 9.937e+02, threshold=7.862e+02, percent-clipped=4.0 +2023-02-06 01:46:36,398 INFO [train.py:901] (3/4) Epoch 5, batch 8000, loss[loss=0.2985, simple_loss=0.3594, pruned_loss=0.1188, over 8496.00 frames. ], tot_loss[loss=0.286, simple_loss=0.348, pruned_loss=0.112, over 1616572.33 frames. ], batch size: 26, lr: 1.42e-02, grad_scale: 16.0 +2023-02-06 01:46:46,582 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40348.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:47:03,142 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40372.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 01:47:10,475 INFO [train.py:901] (3/4) Epoch 5, batch 8050, loss[loss=0.2848, simple_loss=0.3509, pruned_loss=0.1093, over 8496.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3475, pruned_loss=0.1121, over 1608253.52 frames. ], batch size: 49, lr: 1.42e-02, grad_scale: 8.0 +2023-02-06 01:47:20,250 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40397.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 01:47:43,720 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 01:47:48,100 INFO [train.py:901] (3/4) Epoch 6, batch 0, loss[loss=0.3374, simple_loss=0.3912, pruned_loss=0.1418, over 8463.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3912, pruned_loss=0.1418, over 8463.00 frames. ], batch size: 25, lr: 1.33e-02, grad_scale: 8.0 +2023-02-06 01:47:48,100 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 01:47:59,053 INFO [train.py:935] (3/4) Epoch 6, validation: loss=0.2203, simple_loss=0.3165, pruned_loss=0.06206, over 944034.00 frames. +2023-02-06 01:47:59,054 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 01:48:07,799 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 3.052e+02 3.992e+02 5.098e+02 1.227e+03, threshold=7.983e+02, percent-clipped=7.0 +2023-02-06 01:48:13,427 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 01:48:24,586 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.27 vs. limit=5.0 +2023-02-06 01:48:34,110 INFO [train.py:901] (3/4) Epoch 6, batch 50, loss[loss=0.2479, simple_loss=0.3229, pruned_loss=0.08647, over 7524.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3532, pruned_loss=0.1154, over 366970.96 frames. ], batch size: 18, lr: 1.33e-02, grad_scale: 8.0 +2023-02-06 01:48:48,505 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 01:48:57,360 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40500.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:49:04,765 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40510.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:49:08,810 INFO [train.py:901] (3/4) Epoch 6, batch 100, loss[loss=0.2488, simple_loss=0.3091, pruned_loss=0.09421, over 7706.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3461, pruned_loss=0.1098, over 644149.60 frames. ], batch size: 18, lr: 1.33e-02, grad_scale: 8.0 +2023-02-06 01:49:13,092 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 01:49:15,331 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40525.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:49:17,934 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.747e+02 2.877e+02 3.627e+02 4.294e+02 7.601e+02, threshold=7.253e+02, percent-clipped=0.0 +2023-02-06 01:49:31,506 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40548.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:49:37,660 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40557.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:49:44,253 INFO [train.py:901] (3/4) Epoch 6, batch 150, loss[loss=0.2639, simple_loss=0.3414, pruned_loss=0.09323, over 8290.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3421, pruned_loss=0.1074, over 854353.21 frames. ], batch size: 23, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:49:49,722 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40573.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:49:54,327 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40580.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:50:19,240 INFO [train.py:901] (3/4) Epoch 6, batch 200, loss[loss=0.2844, simple_loss=0.3532, pruned_loss=0.1078, over 8539.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3441, pruned_loss=0.1087, over 1024941.49 frames. ], batch size: 49, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:50:25,200 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 +2023-02-06 01:50:28,752 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.837e+02 3.079e+02 3.898e+02 5.213e+02 9.157e+02, threshold=7.795e+02, percent-clipped=3.0 +2023-02-06 01:50:32,269 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40634.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:50:54,051 INFO [train.py:901] (3/4) Epoch 6, batch 250, loss[loss=0.2671, simple_loss=0.3459, pruned_loss=0.09414, over 8462.00 frames. ], tot_loss[loss=0.282, simple_loss=0.346, pruned_loss=0.109, over 1159819.49 frames. ], batch size: 25, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:50:56,283 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40669.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:50:58,988 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40672.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:51:04,073 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 01:51:12,261 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 01:51:13,018 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40692.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:51:15,105 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40695.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:51:20,705 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 +2023-02-06 01:51:29,244 INFO [train.py:901] (3/4) Epoch 6, batch 300, loss[loss=0.2671, simple_loss=0.3395, pruned_loss=0.09731, over 8334.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.348, pruned_loss=0.1107, over 1262796.66 frames. ], batch size: 25, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:51:38,581 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.762e+02 3.025e+02 3.729e+02 4.724e+02 9.863e+02, threshold=7.458e+02, percent-clipped=3.0 +2023-02-06 01:51:52,721 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40749.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:52:03,705 INFO [train.py:901] (3/4) Epoch 6, batch 350, loss[loss=0.2871, simple_loss=0.357, pruned_loss=0.1087, over 8253.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3471, pruned_loss=0.1107, over 1341579.31 frames. ], batch size: 24, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:52:32,427 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40807.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:52:35,329 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2023-02-06 01:52:38,231 INFO [train.py:901] (3/4) Epoch 6, batch 400, loss[loss=0.3403, simple_loss=0.391, pruned_loss=0.1448, over 8564.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3453, pruned_loss=0.1098, over 1396007.10 frames. ], batch size: 31, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:52:39,046 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3364, 4.3212, 3.8801, 1.7887, 3.7254, 3.8870, 4.1173, 3.4826], + device='cuda:3'), covar=tensor([0.0850, 0.0603, 0.1030, 0.4976, 0.0759, 0.0766, 0.1156, 0.0763], + device='cuda:3'), in_proj_covar=tensor([0.0400, 0.0285, 0.0321, 0.0410, 0.0311, 0.0275, 0.0298, 0.0251], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 01:52:46,391 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.2902, 1.4434, 4.3819, 1.6563, 3.8219, 3.6854, 3.8713, 3.8150], + device='cuda:3'), covar=tensor([0.0411, 0.3593, 0.0399, 0.2613, 0.0984, 0.0649, 0.0510, 0.0522], + device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0498, 0.0439, 0.0432, 0.0499, 0.0420, 0.0405, 0.0461], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 01:52:46,905 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.939e+02 3.080e+02 3.801e+02 5.022e+02 1.220e+03, threshold=7.601e+02, percent-clipped=4.0 +2023-02-06 01:53:04,906 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40854.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:53:04,945 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.7218, 5.7771, 4.9990, 2.1547, 5.1067, 5.4710, 5.5655, 5.0770], + device='cuda:3'), covar=tensor([0.0741, 0.0366, 0.0848, 0.4736, 0.0690, 0.0550, 0.0900, 0.0501], + device='cuda:3'), in_proj_covar=tensor([0.0411, 0.0289, 0.0325, 0.0414, 0.0315, 0.0280, 0.0306, 0.0255], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 01:53:12,737 INFO [train.py:901] (3/4) Epoch 6, batch 450, loss[loss=0.2418, simple_loss=0.3104, pruned_loss=0.08667, over 7801.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3474, pruned_loss=0.1108, over 1448983.93 frames. ], batch size: 20, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:53:24,454 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40883.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:53:47,834 INFO [train.py:901] (3/4) Epoch 6, batch 500, loss[loss=0.3719, simple_loss=0.408, pruned_loss=0.1678, over 7224.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3471, pruned_loss=0.1104, over 1489524.59 frames. ], batch size: 73, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:53:49,263 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40918.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:53:56,788 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40928.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:53:57,231 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.956e+02 3.058e+02 3.738e+02 5.288e+02 8.550e+02, threshold=7.476e+02, percent-clipped=3.0 +2023-02-06 01:54:12,302 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40951.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:54:13,628 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40953.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:54:22,939 INFO [train.py:901] (3/4) Epoch 6, batch 550, loss[loss=0.282, simple_loss=0.3505, pruned_loss=0.1067, over 8485.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3456, pruned_loss=0.1092, over 1519847.18 frames. ], batch size: 34, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:54:25,225 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40969.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:54:30,599 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40976.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:54:49,829 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41005.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:54:55,069 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41013.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:54:56,493 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41015.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:54:57,033 INFO [train.py:901] (3/4) Epoch 6, batch 600, loss[loss=0.3101, simple_loss=0.3716, pruned_loss=0.1242, over 8538.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3464, pruned_loss=0.1096, over 1543040.66 frames. ], batch size: 49, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:55:06,096 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.689e+02 2.839e+02 3.515e+02 4.292e+02 8.268e+02, threshold=7.031e+02, percent-clipped=4.0 +2023-02-06 01:55:06,998 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41030.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:55:09,476 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 01:55:16,705 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.6852, 5.8268, 4.9880, 2.3394, 5.0101, 5.2752, 5.3896, 4.8589], + device='cuda:3'), covar=tensor([0.0561, 0.0328, 0.0730, 0.4124, 0.0553, 0.0637, 0.0877, 0.0568], + device='cuda:3'), in_proj_covar=tensor([0.0398, 0.0284, 0.0318, 0.0410, 0.0310, 0.0278, 0.0300, 0.0253], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 01:55:29,301 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41063.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:55:29,380 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41063.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:55:31,088 INFO [train.py:901] (3/4) Epoch 6, batch 650, loss[loss=0.2021, simple_loss=0.2703, pruned_loss=0.06698, over 7301.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3461, pruned_loss=0.1094, over 1555462.88 frames. ], batch size: 16, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:55:32,749 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.84 vs. limit=5.0 +2023-02-06 01:55:46,229 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41088.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:56:05,817 INFO [train.py:901] (3/4) Epoch 6, batch 700, loss[loss=0.2118, simple_loss=0.2962, pruned_loss=0.06368, over 7989.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3445, pruned_loss=0.108, over 1569716.25 frames. ], batch size: 21, lr: 1.32e-02, grad_scale: 8.0 +2023-02-06 01:56:08,625 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41120.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:56:13,565 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9813, 2.6300, 3.1803, 1.0500, 3.2303, 1.9539, 1.4868, 1.8786], + device='cuda:3'), covar=tensor([0.0278, 0.0116, 0.0124, 0.0273, 0.0119, 0.0306, 0.0323, 0.0183], + device='cuda:3'), in_proj_covar=tensor([0.0324, 0.0229, 0.0198, 0.0285, 0.0227, 0.0372, 0.0297, 0.0275], + device='cuda:3'), out_proj_covar=tensor([1.1200e-04, 7.6334e-05, 6.6504e-05, 9.7055e-05, 7.8084e-05, 1.3704e-04, + 1.0279e-04, 9.3551e-05], device='cuda:3') +2023-02-06 01:56:14,229 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41128.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:56:14,737 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.921e+02 2.998e+02 3.776e+02 4.654e+02 1.221e+03, threshold=7.553e+02, percent-clipped=4.0 +2023-02-06 01:56:40,063 INFO [train.py:901] (3/4) Epoch 6, batch 750, loss[loss=0.3099, simple_loss=0.3542, pruned_loss=0.1328, over 6887.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3433, pruned_loss=0.1078, over 1578985.05 frames. ], batch size: 71, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 01:56:52,315 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1375, 1.4209, 1.6752, 1.2675, 1.2097, 1.4297, 1.7092, 1.4222], + device='cuda:3'), covar=tensor([0.0575, 0.1272, 0.1696, 0.1411, 0.0603, 0.1529, 0.0681, 0.0599], + device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0176, 0.0216, 0.0180, 0.0129, 0.0186, 0.0140, 0.0150], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 01:56:52,816 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 01:57:00,951 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 01:57:15,163 INFO [train.py:901] (3/4) Epoch 6, batch 800, loss[loss=0.3141, simple_loss=0.3632, pruned_loss=0.1325, over 8478.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3441, pruned_loss=0.1083, over 1585659.57 frames. ], batch size: 29, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 01:57:21,520 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41225.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:57:22,747 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41227.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:57:24,005 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.548e+02 2.937e+02 3.578e+02 4.897e+02 8.076e+02, threshold=7.157e+02, percent-clipped=3.0 +2023-02-06 01:57:38,352 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41250.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:57:46,852 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41262.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:57:49,532 INFO [train.py:901] (3/4) Epoch 6, batch 850, loss[loss=0.296, simple_loss=0.3599, pruned_loss=0.1161, over 8347.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3439, pruned_loss=0.1089, over 1588790.30 frames. ], batch size: 24, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 01:57:54,388 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8580, 2.3379, 4.8175, 1.3443, 3.3496, 2.4265, 1.9128, 2.9725], + device='cuda:3'), covar=tensor([0.1305, 0.1770, 0.0507, 0.3048, 0.1155, 0.1991, 0.1332, 0.1935], + device='cuda:3'), in_proj_covar=tensor([0.0462, 0.0450, 0.0520, 0.0533, 0.0585, 0.0511, 0.0443, 0.0588], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 01:58:23,894 INFO [train.py:901] (3/4) Epoch 6, batch 900, loss[loss=0.2711, simple_loss=0.3479, pruned_loss=0.0971, over 8488.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3433, pruned_loss=0.1084, over 1593757.27 frames. ], batch size: 29, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 01:58:33,480 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.830e+02 2.871e+02 3.405e+02 4.321e+02 1.147e+03, threshold=6.810e+02, percent-clipped=2.0 +2023-02-06 01:58:34,642 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-06 01:58:42,537 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41342.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:58:52,794 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.68 vs. limit=5.0 +2023-02-06 01:58:53,861 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41359.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:58:58,556 INFO [train.py:901] (3/4) Epoch 6, batch 950, loss[loss=0.2547, simple_loss=0.3204, pruned_loss=0.09452, over 8082.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3429, pruned_loss=0.1081, over 1598469.55 frames. ], batch size: 21, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 01:59:06,361 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41377.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:59:11,063 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41384.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:59:21,435 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 01:59:26,995 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41407.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:59:28,450 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41409.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 01:59:32,977 INFO [train.py:901] (3/4) Epoch 6, batch 1000, loss[loss=0.2403, simple_loss=0.3041, pruned_loss=0.08822, over 7640.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3415, pruned_loss=0.1067, over 1597873.76 frames. ], batch size: 19, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 01:59:41,359 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.042e+02 3.293e+02 3.921e+02 5.074e+02 1.211e+03, threshold=7.843e+02, percent-clipped=6.0 +2023-02-06 01:59:55,291 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 02:00:06,251 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41464.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:00:07,520 INFO [train.py:901] (3/4) Epoch 6, batch 1050, loss[loss=0.3016, simple_loss=0.3518, pruned_loss=0.1257, over 6821.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3443, pruned_loss=0.1091, over 1602906.41 frames. ], batch size: 15, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 02:00:08,234 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 02:00:13,141 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41474.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:00:19,147 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7366, 1.7698, 3.1983, 1.3118, 2.2316, 3.6654, 3.6142, 3.0478], + device='cuda:3'), covar=tensor([0.1195, 0.1320, 0.0443, 0.2159, 0.0864, 0.0272, 0.0429, 0.0670], + device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0266, 0.0229, 0.0261, 0.0229, 0.0207, 0.0243, 0.0279], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 02:00:42,276 INFO [train.py:901] (3/4) Epoch 6, batch 1100, loss[loss=0.2252, simple_loss=0.3127, pruned_loss=0.06886, over 8084.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3453, pruned_loss=0.1098, over 1604333.46 frames. ], batch size: 21, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 02:00:46,700 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41522.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:00:51,108 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.799e+02 2.937e+02 3.488e+02 4.376e+02 9.981e+02, threshold=6.976e+02, percent-clipped=3.0 +2023-02-06 02:01:16,051 INFO [train.py:901] (3/4) Epoch 6, batch 1150, loss[loss=0.2639, simple_loss=0.3306, pruned_loss=0.09865, over 8358.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3449, pruned_loss=0.1096, over 1606192.58 frames. ], batch size: 24, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 02:01:18,801 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 02:01:25,454 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41579.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:01:27,462 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4860, 1.6518, 1.8335, 1.5230, 0.9564, 1.9086, 0.2600, 1.0418], + device='cuda:3'), covar=tensor([0.2375, 0.1902, 0.0680, 0.1979, 0.5824, 0.0847, 0.5086, 0.2652], + device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0131, 0.0082, 0.0175, 0.0213, 0.0083, 0.0147, 0.0136], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 02:01:38,067 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41598.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:01:50,380 INFO [train.py:901] (3/4) Epoch 6, batch 1200, loss[loss=0.2414, simple_loss=0.3128, pruned_loss=0.08499, over 7788.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3445, pruned_loss=0.1089, over 1611368.99 frames. ], batch size: 19, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 02:01:53,225 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2405, 2.0025, 2.9638, 2.3264, 2.4175, 1.9126, 1.5208, 1.1939], + device='cuda:3'), covar=tensor([0.1927, 0.1957, 0.0464, 0.1135, 0.0947, 0.1035, 0.1056, 0.1982], + device='cuda:3'), in_proj_covar=tensor([0.0770, 0.0711, 0.0604, 0.0703, 0.0794, 0.0656, 0.0626, 0.0654], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 02:01:55,788 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41623.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:02:00,257 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.965e+02 3.060e+02 3.864e+02 4.910e+02 1.275e+03, threshold=7.729e+02, percent-clipped=9.0 +2023-02-06 02:02:03,225 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41633.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:02:19,548 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41658.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:02:24,698 INFO [train.py:901] (3/4) Epoch 6, batch 1250, loss[loss=0.348, simple_loss=0.3723, pruned_loss=0.1618, over 7538.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3427, pruned_loss=0.1076, over 1608256.28 frames. ], batch size: 18, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 02:02:29,689 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41672.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:02:34,701 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 +2023-02-06 02:02:59,807 INFO [train.py:901] (3/4) Epoch 6, batch 1300, loss[loss=0.2703, simple_loss=0.3157, pruned_loss=0.1124, over 7791.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3417, pruned_loss=0.1074, over 1606246.80 frames. ], batch size: 19, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 02:03:08,602 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.781e+02 3.137e+02 4.028e+02 4.813e+02 9.668e+02, threshold=8.056e+02, percent-clipped=5.0 +2023-02-06 02:03:09,498 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41730.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:03:27,440 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41755.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:03:34,647 INFO [train.py:901] (3/4) Epoch 6, batch 1350, loss[loss=0.252, simple_loss=0.31, pruned_loss=0.09697, over 7428.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3413, pruned_loss=0.1068, over 1610632.39 frames. ], batch size: 17, lr: 1.31e-02, grad_scale: 8.0 +2023-02-06 02:03:42,972 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41778.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:04:00,426 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41803.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:04:06,499 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3161, 1.5540, 2.2972, 1.1347, 1.6662, 1.5582, 1.3539, 1.5309], + device='cuda:3'), covar=tensor([0.1445, 0.1612, 0.0629, 0.2863, 0.1235, 0.2305, 0.1454, 0.1510], + device='cuda:3'), in_proj_covar=tensor([0.0471, 0.0459, 0.0528, 0.0542, 0.0592, 0.0520, 0.0445, 0.0587], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 02:04:09,399 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 +2023-02-06 02:04:09,619 INFO [train.py:901] (3/4) Epoch 6, batch 1400, loss[loss=0.2045, simple_loss=0.2879, pruned_loss=0.06055, over 8111.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3397, pruned_loss=0.1059, over 1609798.37 frames. ], batch size: 23, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:04:18,118 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.090e+02 3.079e+02 3.704e+02 4.589e+02 8.838e+02, threshold=7.407e+02, percent-clipped=2.0 +2023-02-06 02:04:22,381 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41835.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:04:39,963 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41860.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:04:44,132 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2433, 2.1320, 1.5300, 2.0200, 1.8312, 1.2064, 1.5134, 1.7432], + device='cuda:3'), covar=tensor([0.0972, 0.0334, 0.0937, 0.0458, 0.0578, 0.1268, 0.0879, 0.0685], + device='cuda:3'), in_proj_covar=tensor([0.0341, 0.0237, 0.0308, 0.0297, 0.0312, 0.0310, 0.0336, 0.0307], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 02:04:44,592 INFO [train.py:901] (3/4) Epoch 6, batch 1450, loss[loss=0.2703, simple_loss=0.3436, pruned_loss=0.09851, over 8248.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3409, pruned_loss=0.1067, over 1605956.49 frames. ], batch size: 24, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:04:47,872 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 02:05:18,595 INFO [train.py:901] (3/4) Epoch 6, batch 1500, loss[loss=0.3083, simple_loss=0.3737, pruned_loss=0.1214, over 8202.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3429, pruned_loss=0.1078, over 1609640.53 frames. ], batch size: 23, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:05:24,647 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41924.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:05:27,835 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.902e+02 2.922e+02 3.542e+02 4.432e+02 1.007e+03, threshold=7.084e+02, percent-clipped=2.0 +2023-02-06 02:05:53,225 INFO [train.py:901] (3/4) Epoch 6, batch 1550, loss[loss=0.244, simple_loss=0.3214, pruned_loss=0.08329, over 8728.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.343, pruned_loss=0.1082, over 1606594.23 frames. ], batch size: 34, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:05:53,367 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41966.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 02:06:28,439 INFO [train.py:901] (3/4) Epoch 6, batch 1600, loss[loss=0.3281, simple_loss=0.392, pruned_loss=0.1321, over 8607.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.343, pruned_loss=0.1077, over 1610121.83 frames. ], batch size: 39, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:06:28,517 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42016.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:06:33,628 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 +2023-02-06 02:06:37,873 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.135e+02 3.132e+02 3.836e+02 5.392e+02 3.005e+03, threshold=7.672e+02, percent-clipped=11.0 +2023-02-06 02:07:03,787 INFO [train.py:901] (3/4) Epoch 6, batch 1650, loss[loss=0.2749, simple_loss=0.3567, pruned_loss=0.09652, over 8364.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3405, pruned_loss=0.1057, over 1612212.79 frames. ], batch size: 24, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:07:39,091 INFO [train.py:901] (3/4) Epoch 6, batch 1700, loss[loss=0.2691, simple_loss=0.3358, pruned_loss=0.1012, over 7246.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3424, pruned_loss=0.107, over 1613883.37 frames. ], batch size: 72, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:07:47,894 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.821e+02 2.826e+02 3.670e+02 4.452e+02 1.049e+03, threshold=7.339e+02, percent-clipped=2.0 +2023-02-06 02:07:49,334 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42131.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:08:00,568 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 +2023-02-06 02:08:14,465 INFO [train.py:901] (3/4) Epoch 6, batch 1750, loss[loss=0.3281, simple_loss=0.3776, pruned_loss=0.1393, over 8463.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3408, pruned_loss=0.107, over 1607229.71 frames. ], batch size: 27, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:08:49,384 INFO [train.py:901] (3/4) Epoch 6, batch 1800, loss[loss=0.2615, simple_loss=0.3179, pruned_loss=0.1025, over 7276.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.34, pruned_loss=0.1066, over 1603239.88 frames. ], batch size: 16, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:08:59,176 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.886e+02 3.032e+02 3.540e+02 4.353e+02 2.015e+03, threshold=7.080e+02, percent-clipped=5.0 +2023-02-06 02:09:24,922 INFO [train.py:901] (3/4) Epoch 6, batch 1850, loss[loss=0.2779, simple_loss=0.3413, pruned_loss=0.1072, over 8138.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3416, pruned_loss=0.1075, over 1612118.47 frames. ], batch size: 22, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:09:26,417 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42268.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:09:55,345 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42310.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 02:09:59,279 INFO [train.py:901] (3/4) Epoch 6, batch 1900, loss[loss=0.2657, simple_loss=0.3113, pruned_loss=0.1101, over 7536.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3423, pruned_loss=0.1076, over 1615564.20 frames. ], batch size: 18, lr: 1.30e-02, grad_scale: 8.0 +2023-02-06 02:10:08,776 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.992e+02 2.715e+02 3.297e+02 4.142e+02 7.213e+02, threshold=6.594e+02, percent-clipped=2.0 +2023-02-06 02:10:23,952 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 02:10:34,058 INFO [train.py:901] (3/4) Epoch 6, batch 1950, loss[loss=0.2276, simple_loss=0.3051, pruned_loss=0.07502, over 8223.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3426, pruned_loss=0.107, over 1617926.25 frames. ], batch size: 22, lr: 1.30e-02, grad_scale: 16.0 +2023-02-06 02:10:36,640 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 02:10:46,221 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42383.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:10:48,981 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42387.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:10:56,190 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 02:10:56,319 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42397.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:11:06,409 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42412.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:11:08,983 INFO [train.py:901] (3/4) Epoch 6, batch 2000, loss[loss=0.2598, simple_loss=0.3351, pruned_loss=0.09224, over 8326.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3426, pruned_loss=0.1071, over 1617873.52 frames. ], batch size: 25, lr: 1.30e-02, grad_scale: 16.0 +2023-02-06 02:11:15,172 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42425.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 02:11:18,299 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.899e+02 2.766e+02 3.581e+02 4.303e+02 8.011e+02, threshold=7.162e+02, percent-clipped=3.0 +2023-02-06 02:11:43,879 INFO [train.py:901] (3/4) Epoch 6, batch 2050, loss[loss=0.2785, simple_loss=0.3496, pruned_loss=0.1037, over 8342.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3416, pruned_loss=0.1058, over 1617275.04 frames. ], batch size: 26, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:12:17,698 INFO [train.py:901] (3/4) Epoch 6, batch 2100, loss[loss=0.3897, simple_loss=0.4246, pruned_loss=0.1774, over 8569.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3428, pruned_loss=0.1067, over 1622100.09 frames. ], batch size: 49, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:12:23,886 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42524.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:12:27,687 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.694e+02 2.916e+02 3.481e+02 4.572e+02 1.310e+03, threshold=6.962e+02, percent-clipped=2.0 +2023-02-06 02:12:31,136 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.6039, 5.5796, 4.9617, 1.9874, 5.0171, 5.3622, 5.3176, 5.0100], + device='cuda:3'), covar=tensor([0.0661, 0.0444, 0.0814, 0.4557, 0.0565, 0.0723, 0.0969, 0.0618], + device='cuda:3'), in_proj_covar=tensor([0.0393, 0.0293, 0.0323, 0.0401, 0.0314, 0.0283, 0.0306, 0.0258], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 02:12:52,387 INFO [train.py:901] (3/4) Epoch 6, batch 2150, loss[loss=0.2818, simple_loss=0.3493, pruned_loss=0.1071, over 8485.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3427, pruned_loss=0.1071, over 1616862.55 frames. ], batch size: 28, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:13:25,966 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8830, 1.5489, 2.3097, 1.9224, 1.9373, 1.6882, 1.2952, 0.6276], + device='cuda:3'), covar=tensor([0.2194, 0.2190, 0.0596, 0.1152, 0.0949, 0.1299, 0.1375, 0.2005], + device='cuda:3'), in_proj_covar=tensor([0.0775, 0.0710, 0.0618, 0.0712, 0.0797, 0.0657, 0.0622, 0.0651], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 02:13:27,037 INFO [train.py:901] (3/4) Epoch 6, batch 2200, loss[loss=0.2941, simple_loss=0.3542, pruned_loss=0.117, over 8489.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.342, pruned_loss=0.1071, over 1608615.35 frames. ], batch size: 25, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:13:36,153 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.877e+02 3.020e+02 3.729e+02 5.072e+02 1.122e+03, threshold=7.459e+02, percent-clipped=5.0 +2023-02-06 02:13:39,964 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-06 02:13:41,693 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.6135, 5.7012, 4.9592, 1.8116, 5.0295, 5.1806, 5.4093, 4.7886], + device='cuda:3'), covar=tensor([0.0744, 0.0331, 0.0785, 0.5004, 0.0635, 0.0647, 0.0818, 0.0654], + device='cuda:3'), in_proj_covar=tensor([0.0407, 0.0298, 0.0332, 0.0415, 0.0321, 0.0290, 0.0309, 0.0263], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 02:13:43,115 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42639.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:13:51,252 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.43 vs. limit=5.0 +2023-02-06 02:13:59,706 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42664.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:14:00,835 INFO [train.py:901] (3/4) Epoch 6, batch 2250, loss[loss=0.2992, simple_loss=0.3636, pruned_loss=0.1174, over 8498.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3421, pruned_loss=0.107, over 1614249.66 frames. ], batch size: 28, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:14:01,636 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42667.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:14:11,906 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42681.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 02:14:14,915 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.04 vs. limit=5.0 +2023-02-06 02:14:29,260 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42706.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 02:14:35,823 INFO [train.py:901] (3/4) Epoch 6, batch 2300, loss[loss=0.2988, simple_loss=0.357, pruned_loss=0.1203, over 8543.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3418, pruned_loss=0.1072, over 1610389.10 frames. ], batch size: 39, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:14:37,439 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6013, 1.9404, 2.2259, 1.2968, 2.3735, 1.4909, 0.7703, 1.8157], + device='cuda:3'), covar=tensor([0.0241, 0.0105, 0.0086, 0.0203, 0.0088, 0.0349, 0.0317, 0.0112], + device='cuda:3'), in_proj_covar=tensor([0.0331, 0.0238, 0.0211, 0.0296, 0.0233, 0.0389, 0.0307, 0.0276], + device='cuda:3'), out_proj_covar=tensor([1.1251e-04, 7.8386e-05, 6.9930e-05, 9.9007e-05, 7.9130e-05, 1.4115e-04, + 1.0483e-04, 9.2462e-05], device='cuda:3') +2023-02-06 02:14:45,237 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.705e+02 2.977e+02 3.532e+02 4.435e+02 7.362e+02, threshold=7.063e+02, percent-clipped=0.0 +2023-02-06 02:14:53,236 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42741.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:15:11,242 INFO [train.py:901] (3/4) Epoch 6, batch 2350, loss[loss=0.3137, simple_loss=0.3755, pruned_loss=0.126, over 8345.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3422, pruned_loss=0.1069, over 1614591.28 frames. ], batch size: 26, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:15:18,563 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3155, 1.9887, 3.1727, 2.6992, 2.7280, 2.0030, 1.5909, 1.4335], + device='cuda:3'), covar=tensor([0.2057, 0.2414, 0.0513, 0.1147, 0.1028, 0.1207, 0.1062, 0.2326], + device='cuda:3'), in_proj_covar=tensor([0.0772, 0.0705, 0.0608, 0.0705, 0.0794, 0.0649, 0.0617, 0.0646], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 02:15:43,015 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.9898, 1.4592, 6.0652, 2.3068, 5.4130, 5.0754, 5.6618, 5.5218], + device='cuda:3'), covar=tensor([0.0333, 0.3853, 0.0229, 0.2297, 0.0795, 0.0454, 0.0279, 0.0326], + device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0495, 0.0432, 0.0427, 0.0499, 0.0411, 0.0405, 0.0461], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 02:15:46,902 INFO [train.py:901] (3/4) Epoch 6, batch 2400, loss[loss=0.2653, simple_loss=0.3397, pruned_loss=0.09543, over 8185.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3426, pruned_loss=0.1074, over 1615097.49 frames. ], batch size: 23, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:15:56,301 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.044e+02 3.099e+02 3.712e+02 4.452e+02 1.076e+03, threshold=7.425e+02, percent-clipped=4.0 +2023-02-06 02:16:14,319 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42856.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:16:20,856 INFO [train.py:901] (3/4) Epoch 6, batch 2450, loss[loss=0.1919, simple_loss=0.276, pruned_loss=0.05389, over 7235.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3414, pruned_loss=0.1069, over 1613405.36 frames. ], batch size: 16, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:16:22,281 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42868.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:16:29,155 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42877.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:16:31,746 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3255, 1.5982, 1.3198, 1.9572, 0.8512, 1.1881, 1.2235, 1.5695], + device='cuda:3'), covar=tensor([0.1045, 0.0892, 0.1466, 0.0538, 0.1383, 0.1756, 0.1021, 0.0804], + device='cuda:3'), in_proj_covar=tensor([0.0267, 0.0250, 0.0284, 0.0225, 0.0246, 0.0278, 0.0282, 0.0252], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 02:16:32,685 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.23 vs. limit=5.0 +2023-02-06 02:16:42,221 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42897.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:16:54,610 INFO [train.py:901] (3/4) Epoch 6, batch 2500, loss[loss=0.2675, simple_loss=0.3395, pruned_loss=0.09775, over 8468.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3424, pruned_loss=0.1075, over 1613361.54 frames. ], batch size: 27, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:17:05,200 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.826e+02 3.094e+02 4.004e+02 4.995e+02 1.056e+03, threshold=8.009e+02, percent-clipped=4.0 +2023-02-06 02:17:17,684 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-06 02:17:29,432 INFO [train.py:901] (3/4) Epoch 6, batch 2550, loss[loss=0.2246, simple_loss=0.2856, pruned_loss=0.08179, over 7437.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3407, pruned_loss=0.1068, over 1610753.93 frames. ], batch size: 17, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:17:39,755 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.3935, 2.2037, 1.9761, 1.8482, 1.5478, 1.9114, 2.5767, 1.8873], + device='cuda:3'), covar=tensor([0.0470, 0.1145, 0.1675, 0.1319, 0.0608, 0.1451, 0.0591, 0.0601], + device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0173, 0.0216, 0.0180, 0.0125, 0.0183, 0.0139, 0.0150], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 02:17:41,648 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42983.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:17:42,283 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0356, 2.3556, 1.9640, 3.0838, 1.4272, 1.5130, 1.7943, 2.3127], + device='cuda:3'), covar=tensor([0.0895, 0.0955, 0.1260, 0.0421, 0.1403, 0.1895, 0.1399, 0.0933], + device='cuda:3'), in_proj_covar=tensor([0.0266, 0.0250, 0.0281, 0.0223, 0.0243, 0.0278, 0.0280, 0.0252], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 02:18:01,220 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43011.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:18:04,436 INFO [train.py:901] (3/4) Epoch 6, batch 2600, loss[loss=0.2061, simple_loss=0.2709, pruned_loss=0.07066, over 7711.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.34, pruned_loss=0.1063, over 1606174.63 frames. ], batch size: 18, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:18:10,796 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3238, 1.7636, 2.6906, 1.0944, 1.7988, 1.6289, 1.4059, 1.6292], + device='cuda:3'), covar=tensor([0.1856, 0.1946, 0.0759, 0.3695, 0.1778, 0.2807, 0.1748, 0.2298], + device='cuda:3'), in_proj_covar=tensor([0.0468, 0.0453, 0.0521, 0.0536, 0.0583, 0.0520, 0.0445, 0.0587], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 02:18:13,979 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.066e+02 3.052e+02 3.779e+02 5.019e+02 1.784e+03, threshold=7.558e+02, percent-clipped=4.0 +2023-02-06 02:18:33,097 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.7861, 3.7506, 3.3712, 1.6025, 3.3500, 3.2637, 3.4449, 2.8803], + device='cuda:3'), covar=tensor([0.0922, 0.0616, 0.0945, 0.4719, 0.0813, 0.1101, 0.1089, 0.1067], + device='cuda:3'), in_proj_covar=tensor([0.0394, 0.0285, 0.0322, 0.0399, 0.0313, 0.0280, 0.0306, 0.0253], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 02:18:39,577 INFO [train.py:901] (3/4) Epoch 6, batch 2650, loss[loss=0.2827, simple_loss=0.3535, pruned_loss=0.106, over 8468.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3405, pruned_loss=0.106, over 1606874.73 frames. ], batch size: 27, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:19:11,157 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43112.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:19:14,337 INFO [train.py:901] (3/4) Epoch 6, batch 2700, loss[loss=0.3167, simple_loss=0.3802, pruned_loss=0.1266, over 8334.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3404, pruned_loss=0.1057, over 1605067.45 frames. ], batch size: 26, lr: 1.29e-02, grad_scale: 8.0 +2023-02-06 02:19:20,973 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43126.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:19:23,436 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.791e+02 2.935e+02 3.532e+02 4.548e+02 1.003e+03, threshold=7.064e+02, percent-clipped=2.0 +2023-02-06 02:19:28,343 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43137.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:19:47,960 INFO [train.py:901] (3/4) Epoch 6, batch 2750, loss[loss=0.3381, simple_loss=0.392, pruned_loss=0.1421, over 8617.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3394, pruned_loss=0.1048, over 1609960.85 frames. ], batch size: 34, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:19:51,155 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 +2023-02-06 02:20:22,662 INFO [train.py:901] (3/4) Epoch 6, batch 2800, loss[loss=0.2671, simple_loss=0.3236, pruned_loss=0.1053, over 7647.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3394, pruned_loss=0.1047, over 1611212.97 frames. ], batch size: 19, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:20:26,193 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43221.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:20:32,052 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.494e+02 2.696e+02 3.315e+02 4.271e+02 8.534e+02, threshold=6.630e+02, percent-clipped=4.0 +2023-02-06 02:20:39,133 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43239.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:20:40,382 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43241.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:20:55,849 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43264.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:20:56,917 INFO [train.py:901] (3/4) Epoch 6, batch 2850, loss[loss=0.2358, simple_loss=0.3137, pruned_loss=0.07894, over 8297.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3405, pruned_loss=0.1054, over 1614073.64 frames. ], batch size: 23, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:21:06,153 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.87 vs. limit=5.0 +2023-02-06 02:21:31,687 INFO [train.py:901] (3/4) Epoch 6, batch 2900, loss[loss=0.2639, simple_loss=0.3413, pruned_loss=0.09322, over 8245.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.34, pruned_loss=0.1049, over 1614473.86 frames. ], batch size: 24, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:21:41,577 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.891e+02 2.996e+02 3.885e+02 4.976e+02 9.964e+02, threshold=7.771e+02, percent-clipped=9.0 +2023-02-06 02:21:42,516 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9306, 3.8617, 2.4094, 2.5528, 3.0011, 1.8617, 2.8186, 3.1600], + device='cuda:3'), covar=tensor([0.1409, 0.0263, 0.0794, 0.0824, 0.0636, 0.1121, 0.0907, 0.0794], + device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0240, 0.0314, 0.0308, 0.0322, 0.0316, 0.0344, 0.0311], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 02:21:46,023 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43336.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:22:00,459 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43356.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:22:01,689 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 02:22:07,136 INFO [train.py:901] (3/4) Epoch 6, batch 2950, loss[loss=0.3368, simple_loss=0.3816, pruned_loss=0.1459, over 8613.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3401, pruned_loss=0.1051, over 1610041.66 frames. ], batch size: 34, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:22:17,905 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43382.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:22:33,423 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6309, 2.7018, 1.8091, 2.1780, 2.1171, 1.5466, 2.0659, 2.2822], + device='cuda:3'), covar=tensor([0.1237, 0.0381, 0.0867, 0.0631, 0.0614, 0.1232, 0.0883, 0.0701], + device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0236, 0.0309, 0.0303, 0.0317, 0.0317, 0.0342, 0.0308], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 02:22:36,001 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43407.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:22:41,772 INFO [train.py:901] (3/4) Epoch 6, batch 3000, loss[loss=0.2918, simple_loss=0.3636, pruned_loss=0.11, over 8534.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3414, pruned_loss=0.1059, over 1614425.30 frames. ], batch size: 28, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:22:41,772 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 02:22:53,877 INFO [train.py:935] (3/4) Epoch 6, validation: loss=0.2158, simple_loss=0.3124, pruned_loss=0.05962, over 944034.00 frames. +2023-02-06 02:22:53,878 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 02:23:03,880 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.066e+02 3.288e+02 4.080e+02 5.339e+02 1.082e+03, threshold=8.161e+02, percent-clipped=5.0 +2023-02-06 02:23:28,757 INFO [train.py:901] (3/4) Epoch 6, batch 3050, loss[loss=0.2954, simple_loss=0.3633, pruned_loss=0.1137, over 8746.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3412, pruned_loss=0.1061, over 1607324.48 frames. ], batch size: 30, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:23:45,125 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.0986, 4.0667, 3.6800, 1.7698, 3.6136, 3.6202, 3.7741, 3.1519], + device='cuda:3'), covar=tensor([0.0894, 0.0777, 0.1013, 0.4544, 0.0783, 0.0744, 0.1609, 0.0910], + device='cuda:3'), in_proj_covar=tensor([0.0406, 0.0302, 0.0333, 0.0417, 0.0331, 0.0292, 0.0320, 0.0264], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 02:24:03,331 INFO [train.py:901] (3/4) Epoch 6, batch 3100, loss[loss=0.3103, simple_loss=0.3695, pruned_loss=0.1256, over 8183.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3423, pruned_loss=0.107, over 1609135.32 frames. ], batch size: 23, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:24:12,759 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.408e+02 2.871e+02 3.509e+02 4.582e+02 1.148e+03, threshold=7.017e+02, percent-clipped=4.0 +2023-02-06 02:24:14,880 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43532.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:24:25,050 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0785, 1.3959, 1.5642, 1.2592, 1.0823, 1.3839, 1.6645, 1.5634], + device='cuda:3'), covar=tensor([0.0581, 0.1247, 0.1775, 0.1473, 0.0651, 0.1586, 0.0727, 0.0588], + device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0174, 0.0219, 0.0180, 0.0124, 0.0186, 0.0141, 0.0149], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 02:24:38,267 INFO [train.py:901] (3/4) Epoch 6, batch 3150, loss[loss=0.2518, simple_loss=0.3294, pruned_loss=0.08709, over 8335.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3427, pruned_loss=0.107, over 1614930.16 frames. ], batch size: 26, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:24:38,611 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.93 vs. limit=5.0 +2023-02-06 02:24:50,306 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-06 02:24:57,059 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43592.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:25:07,804 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.66 vs. limit=5.0 +2023-02-06 02:25:10,978 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43612.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:25:13,479 INFO [train.py:901] (3/4) Epoch 6, batch 3200, loss[loss=0.255, simple_loss=0.307, pruned_loss=0.1015, over 7293.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3429, pruned_loss=0.1076, over 1617889.31 frames. ], batch size: 16, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:25:14,386 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43617.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:25:23,589 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.828e+02 3.409e+02 4.222e+02 1.719e+03, threshold=6.818e+02, percent-clipped=4.0 +2023-02-06 02:25:28,589 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43637.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:25:34,656 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3274, 1.1559, 4.4556, 1.6553, 3.7349, 3.7066, 3.9918, 3.8182], + device='cuda:3'), covar=tensor([0.0412, 0.3974, 0.0370, 0.2703, 0.1224, 0.0695, 0.0508, 0.0593], + device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0504, 0.0449, 0.0439, 0.0508, 0.0423, 0.0421, 0.0476], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 02:25:49,151 INFO [train.py:901] (3/4) Epoch 6, batch 3250, loss[loss=0.2721, simple_loss=0.3522, pruned_loss=0.09598, over 8480.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3416, pruned_loss=0.1065, over 1614600.05 frames. ], batch size: 25, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:26:14,704 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6739, 2.3746, 4.3930, 1.1865, 3.0282, 2.3068, 1.7014, 2.6088], + device='cuda:3'), covar=tensor([0.1339, 0.1715, 0.0552, 0.3080, 0.1174, 0.2065, 0.1460, 0.2100], + device='cuda:3'), in_proj_covar=tensor([0.0465, 0.0456, 0.0517, 0.0532, 0.0586, 0.0515, 0.0445, 0.0583], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 02:26:23,265 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 +2023-02-06 02:26:23,408 INFO [train.py:901] (3/4) Epoch 6, batch 3300, loss[loss=0.2674, simple_loss=0.3297, pruned_loss=0.1025, over 8287.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3411, pruned_loss=0.1061, over 1615714.02 frames. ], batch size: 23, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:26:33,006 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.765e+02 2.968e+02 3.670e+02 5.054e+02 9.057e+02, threshold=7.341e+02, percent-clipped=6.0 +2023-02-06 02:26:40,676 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.3844, 1.4770, 1.5744, 1.4407, 1.3940, 1.5022, 2.6842, 2.3707], + device='cuda:3'), covar=tensor([0.0502, 0.1854, 0.2574, 0.1798, 0.0745, 0.2159, 0.0701, 0.0592], + device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0172, 0.0214, 0.0178, 0.0123, 0.0183, 0.0139, 0.0148], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 02:26:58,036 INFO [train.py:901] (3/4) Epoch 6, batch 3350, loss[loss=0.2671, simple_loss=0.3412, pruned_loss=0.0965, over 8295.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3412, pruned_loss=0.1056, over 1619458.36 frames. ], batch size: 23, lr: 1.28e-02, grad_scale: 8.0 +2023-02-06 02:27:25,496 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43805.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:27:33,331 INFO [train.py:901] (3/4) Epoch 6, batch 3400, loss[loss=0.3541, simple_loss=0.4054, pruned_loss=0.1514, over 8289.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3416, pruned_loss=0.1057, over 1620495.14 frames. ], batch size: 23, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:27:42,441 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.884e+02 2.693e+02 3.397e+02 4.441e+02 9.371e+02, threshold=6.793e+02, percent-clipped=2.0 +2023-02-06 02:28:07,549 INFO [train.py:901] (3/4) Epoch 6, batch 3450, loss[loss=0.2341, simple_loss=0.3064, pruned_loss=0.08091, over 8084.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3408, pruned_loss=0.1055, over 1616985.89 frames. ], batch size: 21, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:28:09,784 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0385, 1.7064, 1.4336, 1.6546, 1.3864, 1.1287, 1.2617, 1.4779], + device='cuda:3'), covar=tensor([0.0793, 0.0335, 0.0827, 0.0411, 0.0555, 0.1081, 0.0663, 0.0596], + device='cuda:3'), in_proj_covar=tensor([0.0341, 0.0233, 0.0307, 0.0301, 0.0311, 0.0309, 0.0332, 0.0303], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 02:28:14,344 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43876.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:28:42,236 INFO [train.py:901] (3/4) Epoch 6, batch 3500, loss[loss=0.2853, simple_loss=0.3652, pruned_loss=0.1027, over 8506.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3408, pruned_loss=0.1052, over 1616050.67 frames. ], batch size: 31, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:28:50,533 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43927.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:28:52,404 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.590e+02 3.111e+02 3.775e+02 4.956e+02 7.195e+02, threshold=7.550e+02, percent-clipped=1.0 +2023-02-06 02:28:54,757 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.53 vs. limit=5.0 +2023-02-06 02:28:59,194 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 02:29:16,404 INFO [train.py:901] (3/4) Epoch 6, batch 3550, loss[loss=0.3717, simple_loss=0.4073, pruned_loss=0.168, over 6862.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.342, pruned_loss=0.1063, over 1612960.34 frames. ], batch size: 73, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:29:24,737 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43977.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:29:24,979 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.02 vs. limit=2.0 +2023-02-06 02:29:34,252 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43991.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:29:52,626 INFO [train.py:901] (3/4) Epoch 6, batch 3600, loss[loss=0.3038, simple_loss=0.364, pruned_loss=0.1219, over 8356.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3409, pruned_loss=0.1058, over 1610139.11 frames. ], batch size: 24, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:30:02,264 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.897e+02 2.983e+02 3.632e+02 4.470e+02 1.452e+03, threshold=7.265e+02, percent-clipped=1.0 +2023-02-06 02:30:27,001 INFO [train.py:901] (3/4) Epoch 6, batch 3650, loss[loss=0.2379, simple_loss=0.2985, pruned_loss=0.08865, over 7301.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3409, pruned_loss=0.1059, over 1606869.95 frames. ], batch size: 16, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:31:00,588 INFO [train.py:901] (3/4) Epoch 6, batch 3700, loss[loss=0.2376, simple_loss=0.3184, pruned_loss=0.07843, over 8469.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3404, pruned_loss=0.1057, over 1604808.43 frames. ], batch size: 25, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:31:01,285 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 02:31:11,218 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.832e+02 3.019e+02 3.651e+02 4.413e+02 8.839e+02, threshold=7.303e+02, percent-clipped=3.0 +2023-02-06 02:31:23,984 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44149.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:31:27,458 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.92 vs. limit=5.0 +2023-02-06 02:31:35,821 INFO [train.py:901] (3/4) Epoch 6, batch 3750, loss[loss=0.2739, simple_loss=0.337, pruned_loss=0.1054, over 7976.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3413, pruned_loss=0.1064, over 1604476.03 frames. ], batch size: 21, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:32:06,703 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.7660, 3.7532, 3.3925, 1.7283, 3.3727, 3.4092, 3.4413, 3.0287], + device='cuda:3'), covar=tensor([0.1091, 0.0776, 0.1214, 0.4684, 0.0931, 0.1080, 0.1432, 0.1072], + device='cuda:3'), in_proj_covar=tensor([0.0405, 0.0302, 0.0335, 0.0417, 0.0323, 0.0293, 0.0319, 0.0265], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 02:32:09,322 INFO [train.py:901] (3/4) Epoch 6, batch 3800, loss[loss=0.2933, simple_loss=0.3459, pruned_loss=0.1204, over 7804.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3408, pruned_loss=0.1065, over 1597072.45 frames. ], batch size: 19, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:32:19,578 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.066e+02 3.008e+02 3.761e+02 4.930e+02 1.044e+03, threshold=7.521e+02, percent-clipped=7.0 +2023-02-06 02:32:32,439 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44247.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:32:34,456 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5032, 1.4263, 4.6864, 1.8005, 4.1716, 3.9148, 4.2582, 4.1683], + device='cuda:3'), covar=tensor([0.0419, 0.3419, 0.0401, 0.2343, 0.0999, 0.0721, 0.0375, 0.0461], + device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0499, 0.0445, 0.0436, 0.0503, 0.0413, 0.0419, 0.0474], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 02:32:44,252 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44264.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:32:45,446 INFO [train.py:901] (3/4) Epoch 6, batch 3850, loss[loss=0.3331, simple_loss=0.3953, pruned_loss=0.1355, over 8296.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3426, pruned_loss=0.1077, over 1601854.30 frames. ], batch size: 23, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:32:48,966 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44271.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:32:49,779 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44272.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:33:02,839 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 02:33:20,475 INFO [train.py:901] (3/4) Epoch 6, batch 3900, loss[loss=0.2374, simple_loss=0.3168, pruned_loss=0.07901, over 8018.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3427, pruned_loss=0.1073, over 1610224.43 frames. ], batch size: 22, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:33:23,541 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-06 02:33:23,957 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44321.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:33:30,569 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.224e+02 2.909e+02 3.535e+02 4.398e+02 8.405e+02, threshold=7.069e+02, percent-clipped=2.0 +2023-02-06 02:33:56,226 INFO [train.py:901] (3/4) Epoch 6, batch 3950, loss[loss=0.281, simple_loss=0.3481, pruned_loss=0.107, over 8607.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3427, pruned_loss=0.1073, over 1610138.14 frames. ], batch size: 34, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:34:09,754 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44386.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:34:30,792 INFO [train.py:901] (3/4) Epoch 6, batch 4000, loss[loss=0.3142, simple_loss=0.3846, pruned_loss=0.1219, over 8026.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3429, pruned_loss=0.1068, over 1615538.40 frames. ], batch size: 22, lr: 1.27e-02, grad_scale: 8.0 +2023-02-06 02:34:39,800 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4736, 1.5025, 2.7828, 1.2218, 1.9913, 3.0171, 3.0105, 2.5541], + device='cuda:3'), covar=tensor([0.1047, 0.1244, 0.0399, 0.1959, 0.0711, 0.0308, 0.0446, 0.0660], + device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0273, 0.0227, 0.0267, 0.0235, 0.0212, 0.0247, 0.0284], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 02:34:40,314 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.014e+02 2.805e+02 3.702e+02 4.857e+02 8.487e+02, threshold=7.405e+02, percent-clipped=7.0 +2023-02-06 02:34:44,606 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44436.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:35:05,734 INFO [train.py:901] (3/4) Epoch 6, batch 4050, loss[loss=0.2681, simple_loss=0.3177, pruned_loss=0.1092, over 6815.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3426, pruned_loss=0.1066, over 1614562.96 frames. ], batch size: 15, lr: 1.27e-02, grad_scale: 16.0 +2023-02-06 02:35:15,521 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5420, 2.8698, 1.8753, 2.2644, 2.4115, 1.4866, 1.9673, 2.2058], + device='cuda:3'), covar=tensor([0.1145, 0.0249, 0.0790, 0.0517, 0.0476, 0.1069, 0.0703, 0.0682], + device='cuda:3'), in_proj_covar=tensor([0.0344, 0.0232, 0.0307, 0.0301, 0.0316, 0.0311, 0.0333, 0.0304], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 02:35:41,182 INFO [train.py:901] (3/4) Epoch 6, batch 4100, loss[loss=0.2658, simple_loss=0.3313, pruned_loss=0.1002, over 6389.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3418, pruned_loss=0.1061, over 1610724.59 frames. ], batch size: 14, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:35:44,020 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44520.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:35:50,484 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.946e+02 3.131e+02 3.987e+02 5.314e+02 1.327e+03, threshold=7.973e+02, percent-clipped=4.0 +2023-02-06 02:35:53,973 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0014, 4.0814, 2.7437, 2.7210, 3.1167, 2.0484, 2.5495, 3.0083], + device='cuda:3'), covar=tensor([0.1397, 0.0231, 0.0730, 0.0690, 0.0583, 0.1130, 0.0952, 0.0826], + device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0235, 0.0310, 0.0306, 0.0323, 0.0316, 0.0337, 0.0310], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 02:36:00,537 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44545.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:36:00,594 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44545.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:36:14,407 INFO [train.py:901] (3/4) Epoch 6, batch 4150, loss[loss=0.2543, simple_loss=0.3162, pruned_loss=0.09616, over 7973.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.34, pruned_loss=0.1053, over 1610328.55 frames. ], batch size: 21, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:36:15,790 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44568.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:36:49,802 INFO [train.py:901] (3/4) Epoch 6, batch 4200, loss[loss=0.2539, simple_loss=0.3071, pruned_loss=0.1003, over 7642.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3393, pruned_loss=0.1054, over 1612118.73 frames. ], batch size: 19, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:36:58,995 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.001e+02 2.806e+02 3.559e+02 4.787e+02 1.284e+03, threshold=7.119e+02, percent-clipped=4.0 +2023-02-06 02:37:05,643 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 02:37:07,966 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44642.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:37:23,774 INFO [train.py:901] (3/4) Epoch 6, batch 4250, loss[loss=0.3034, simple_loss=0.3603, pruned_loss=0.1233, over 7693.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3385, pruned_loss=0.1051, over 1610297.36 frames. ], batch size: 18, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:37:24,680 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44667.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:37:29,243 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 02:37:41,576 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44692.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:37:47,606 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4844, 2.0033, 3.2431, 2.6161, 2.7037, 2.1163, 1.5207, 1.3694], + device='cuda:3'), covar=tensor([0.1949, 0.2318, 0.0551, 0.1173, 0.1023, 0.1182, 0.1224, 0.2580], + device='cuda:3'), in_proj_covar=tensor([0.0783, 0.0721, 0.0623, 0.0719, 0.0804, 0.0661, 0.0629, 0.0661], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 02:37:51,714 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.68 vs. limit=5.0 +2023-02-06 02:37:58,490 INFO [train.py:901] (3/4) Epoch 6, batch 4300, loss[loss=0.2662, simple_loss=0.3203, pruned_loss=0.106, over 7800.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3381, pruned_loss=0.1045, over 1611799.72 frames. ], batch size: 19, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:38:00,027 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44717.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:38:08,669 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.808e+02 2.771e+02 3.321e+02 4.102e+02 9.930e+02, threshold=6.641e+02, percent-clipped=2.0 +2023-02-06 02:38:33,191 INFO [train.py:901] (3/4) Epoch 6, batch 4350, loss[loss=0.254, simple_loss=0.3202, pruned_loss=0.09388, over 7789.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3394, pruned_loss=0.1052, over 1614057.90 frames. ], batch size: 19, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:38:38,300 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6743, 2.4786, 4.4220, 1.2230, 2.8418, 2.1352, 1.9713, 2.3630], + device='cuda:3'), covar=tensor([0.1679, 0.1969, 0.0863, 0.3968, 0.1633, 0.2639, 0.1582, 0.2828], + device='cuda:3'), in_proj_covar=tensor([0.0473, 0.0462, 0.0538, 0.0545, 0.0589, 0.0531, 0.0445, 0.0599], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 02:39:00,033 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 02:39:06,571 INFO [train.py:901] (3/4) Epoch 6, batch 4400, loss[loss=0.2707, simple_loss=0.3371, pruned_loss=0.1022, over 8233.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3386, pruned_loss=0.1044, over 1614375.55 frames. ], batch size: 22, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:39:12,866 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3497, 2.1508, 1.1967, 2.9935, 1.2830, 1.1094, 1.8661, 2.3296], + device='cuda:3'), covar=tensor([0.2300, 0.1629, 0.3305, 0.0476, 0.2061, 0.3121, 0.1825, 0.1245], + device='cuda:3'), in_proj_covar=tensor([0.0267, 0.0245, 0.0277, 0.0220, 0.0241, 0.0273, 0.0284, 0.0252], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 02:39:17,275 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.185e+02 3.434e+02 4.206e+02 5.183e+02 1.151e+03, threshold=8.413e+02, percent-clipped=11.0 +2023-02-06 02:39:31,753 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0908, 2.1776, 1.4742, 1.9169, 1.9228, 1.2255, 1.5720, 1.7166], + device='cuda:3'), covar=tensor([0.1094, 0.0296, 0.1007, 0.0436, 0.0541, 0.1204, 0.0744, 0.0737], + device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0234, 0.0307, 0.0302, 0.0315, 0.0312, 0.0335, 0.0309], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 02:39:40,250 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 02:39:42,256 INFO [train.py:901] (3/4) Epoch 6, batch 4450, loss[loss=0.2621, simple_loss=0.3448, pruned_loss=0.08975, over 8333.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3388, pruned_loss=0.1043, over 1619383.86 frames. ], batch size: 25, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:39:58,546 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44889.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:40:13,884 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44912.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:40:16,360 INFO [train.py:901] (3/4) Epoch 6, batch 4500, loss[loss=0.3065, simple_loss=0.3564, pruned_loss=0.1283, over 7934.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3393, pruned_loss=0.1047, over 1616612.16 frames. ], batch size: 20, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:40:26,435 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.070e+02 3.100e+02 3.740e+02 5.266e+02 1.703e+03, threshold=7.479e+02, percent-clipped=4.0 +2023-02-06 02:40:31,315 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2879, 2.3861, 1.5787, 2.2144, 2.0114, 1.2261, 1.7589, 2.0423], + device='cuda:3'), covar=tensor([0.1198, 0.0304, 0.1048, 0.0434, 0.0610, 0.1398, 0.0899, 0.0673], + device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0234, 0.0306, 0.0302, 0.0313, 0.0311, 0.0334, 0.0303], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 02:40:31,751 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 02:40:36,019 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7157, 2.9788, 2.0285, 2.4425, 2.4170, 1.7705, 2.0784, 2.4888], + device='cuda:3'), covar=tensor([0.1202, 0.0275, 0.0748, 0.0519, 0.0519, 0.1057, 0.0805, 0.0643], + device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0234, 0.0307, 0.0302, 0.0313, 0.0312, 0.0333, 0.0303], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 02:40:44,171 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5382, 4.5135, 4.0717, 1.5558, 4.1361, 4.0704, 4.2632, 3.7086], + device='cuda:3'), covar=tensor([0.0719, 0.0536, 0.0995, 0.4436, 0.0654, 0.0824, 0.1099, 0.0758], + device='cuda:3'), in_proj_covar=tensor([0.0411, 0.0305, 0.0335, 0.0412, 0.0329, 0.0293, 0.0319, 0.0265], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 02:40:52,041 INFO [train.py:901] (3/4) Epoch 6, batch 4550, loss[loss=0.3191, simple_loss=0.3769, pruned_loss=0.1307, over 8622.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3387, pruned_loss=0.1038, over 1614190.69 frames. ], batch size: 39, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:41:10,193 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.0256, 3.9638, 3.6629, 1.8046, 3.5504, 3.6177, 3.7409, 3.3060], + device='cuda:3'), covar=tensor([0.0959, 0.0644, 0.0909, 0.4764, 0.0852, 0.0922, 0.1121, 0.0947], + device='cuda:3'), in_proj_covar=tensor([0.0408, 0.0304, 0.0333, 0.0413, 0.0328, 0.0293, 0.0316, 0.0265], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 02:41:18,803 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45004.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:41:27,054 INFO [train.py:901] (3/4) Epoch 6, batch 4600, loss[loss=0.2504, simple_loss=0.3133, pruned_loss=0.0937, over 7964.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3392, pruned_loss=0.1042, over 1615865.66 frames. ], batch size: 21, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:41:28,727 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6947, 2.1392, 3.8124, 2.9252, 3.1943, 2.1812, 1.5443, 1.8099], + device='cuda:3'), covar=tensor([0.2293, 0.2962, 0.0630, 0.1368, 0.1256, 0.1207, 0.1207, 0.2681], + device='cuda:3'), in_proj_covar=tensor([0.0785, 0.0731, 0.0621, 0.0726, 0.0817, 0.0665, 0.0632, 0.0663], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 02:41:34,828 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45027.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:41:36,658 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.840e+02 2.977e+02 3.732e+02 4.434e+02 1.135e+03, threshold=7.465e+02, percent-clipped=1.0 +2023-02-06 02:42:02,755 INFO [train.py:901] (3/4) Epoch 6, batch 4650, loss[loss=0.2927, simple_loss=0.3427, pruned_loss=0.1214, over 7916.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3384, pruned_loss=0.1038, over 1615978.20 frames. ], batch size: 20, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:42:08,413 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45074.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:42:25,262 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45099.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:42:36,980 INFO [train.py:901] (3/4) Epoch 6, batch 4700, loss[loss=0.2588, simple_loss=0.326, pruned_loss=0.09577, over 7192.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3397, pruned_loss=0.1052, over 1610147.36 frames. ], batch size: 16, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:42:46,396 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.041e+02 3.187e+02 3.833e+02 4.569e+02 1.251e+03, threshold=7.667e+02, percent-clipped=2.0 +2023-02-06 02:43:11,110 INFO [train.py:901] (3/4) Epoch 6, batch 4750, loss[loss=0.2897, simple_loss=0.3569, pruned_loss=0.1113, over 8318.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3396, pruned_loss=0.1046, over 1613138.44 frames. ], batch size: 25, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:43:30,408 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 02:43:31,805 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 02:43:46,231 INFO [train.py:901] (3/4) Epoch 6, batch 4800, loss[loss=0.3384, simple_loss=0.3793, pruned_loss=0.1487, over 8734.00 frames. ], tot_loss[loss=0.276, simple_loss=0.341, pruned_loss=0.1055, over 1612148.88 frames. ], batch size: 34, lr: 1.26e-02, grad_scale: 16.0 +2023-02-06 02:43:55,770 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.654e+02 3.021e+02 3.501e+02 4.623e+02 8.497e+02, threshold=7.001e+02, percent-clipped=1.0 +2023-02-06 02:44:16,259 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45260.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:44:20,164 INFO [train.py:901] (3/4) Epoch 6, batch 4850, loss[loss=0.3338, simple_loss=0.3883, pruned_loss=0.1397, over 8339.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3394, pruned_loss=0.1047, over 1610357.86 frames. ], batch size: 26, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:44:20,856 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 02:44:33,164 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45283.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:44:35,180 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45285.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:44:50,595 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45308.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:44:51,247 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0420, 1.3261, 1.1773, 0.2997, 1.2664, 0.9418, 0.0883, 1.1409], + device='cuda:3'), covar=tensor([0.0166, 0.0130, 0.0117, 0.0212, 0.0124, 0.0433, 0.0294, 0.0120], + device='cuda:3'), in_proj_covar=tensor([0.0324, 0.0236, 0.0204, 0.0292, 0.0231, 0.0377, 0.0298, 0.0277], + device='cuda:3'), out_proj_covar=tensor([1.0791e-04, 7.6319e-05, 6.6182e-05, 9.6015e-05, 7.6680e-05, 1.3412e-04, + 9.9780e-05, 9.1389e-05], device='cuda:3') +2023-02-06 02:44:56,003 INFO [train.py:901] (3/4) Epoch 6, batch 4900, loss[loss=0.2487, simple_loss=0.3239, pruned_loss=0.08669, over 8533.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.339, pruned_loss=0.1046, over 1608655.49 frames. ], batch size: 39, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:45:02,904 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6605, 1.6524, 3.1864, 1.2569, 2.2269, 3.6101, 3.5361, 3.1779], + device='cuda:3'), covar=tensor([0.1025, 0.1242, 0.0371, 0.1876, 0.0735, 0.0233, 0.0336, 0.0532], + device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0263, 0.0224, 0.0265, 0.0231, 0.0211, 0.0246, 0.0275], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 02:45:05,437 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.764e+02 2.896e+02 3.521e+02 4.501e+02 9.960e+02, threshold=7.042e+02, percent-clipped=7.0 +2023-02-06 02:45:30,236 INFO [train.py:901] (3/4) Epoch 6, batch 4950, loss[loss=0.2561, simple_loss=0.3399, pruned_loss=0.08611, over 8296.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3399, pruned_loss=0.1048, over 1615302.65 frames. ], batch size: 23, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:45:30,384 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45366.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 02:46:05,707 INFO [train.py:901] (3/4) Epoch 6, batch 5000, loss[loss=0.252, simple_loss=0.3192, pruned_loss=0.09238, over 8245.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3393, pruned_loss=0.1051, over 1614517.68 frames. ], batch size: 22, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:46:07,201 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45418.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:46:07,295 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1170, 1.4282, 2.3083, 1.1545, 2.1221, 2.5238, 2.5104, 2.1510], + device='cuda:3'), covar=tensor([0.1067, 0.1039, 0.0460, 0.1845, 0.0540, 0.0366, 0.0524, 0.0747], + device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0261, 0.0222, 0.0264, 0.0230, 0.0210, 0.0246, 0.0275], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 02:46:15,105 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.151e+02 3.255e+02 4.005e+02 4.887e+02 1.315e+03, threshold=8.009e+02, percent-clipped=7.0 +2023-02-06 02:46:24,039 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45443.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:46:40,014 INFO [train.py:901] (3/4) Epoch 6, batch 5050, loss[loss=0.2581, simple_loss=0.3244, pruned_loss=0.09594, over 7942.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3381, pruned_loss=0.1039, over 1615226.04 frames. ], batch size: 20, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:46:51,616 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.0102, 4.0886, 3.6295, 1.9027, 3.5535, 3.6299, 3.7868, 3.3547], + device='cuda:3'), covar=tensor([0.1094, 0.0697, 0.1058, 0.4290, 0.0902, 0.0823, 0.1376, 0.0739], + device='cuda:3'), in_proj_covar=tensor([0.0400, 0.0302, 0.0330, 0.0411, 0.0322, 0.0292, 0.0310, 0.0261], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 02:46:58,892 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 02:47:09,014 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4486, 1.7370, 1.8788, 1.3639, 0.9463, 2.0148, 0.1628, 1.2288], + device='cuda:3'), covar=tensor([0.4345, 0.2483, 0.1133, 0.3288, 0.6862, 0.0660, 0.5417, 0.2265], + device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0135, 0.0087, 0.0183, 0.0228, 0.0083, 0.0146, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 02:47:13,239 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.94 vs. limit=5.0 +2023-02-06 02:47:14,057 INFO [train.py:901] (3/4) Epoch 6, batch 5100, loss[loss=0.2538, simple_loss=0.3034, pruned_loss=0.1021, over 7532.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3375, pruned_loss=0.1033, over 1610746.13 frames. ], batch size: 18, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:47:24,713 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.922e+02 2.842e+02 3.419e+02 4.219e+02 7.828e+02, threshold=6.837e+02, percent-clipped=0.0 +2023-02-06 02:47:26,970 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45533.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:47:40,375 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1195, 1.7411, 3.0210, 2.4664, 2.4061, 1.7281, 1.4461, 1.1331], + device='cuda:3'), covar=tensor([0.2867, 0.2979, 0.0546, 0.1332, 0.1305, 0.1992, 0.1875, 0.2497], + device='cuda:3'), in_proj_covar=tensor([0.0800, 0.0736, 0.0628, 0.0727, 0.0837, 0.0678, 0.0643, 0.0678], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 02:47:43,443 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45558.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:47:47,199 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-06 02:47:49,316 INFO [train.py:901] (3/4) Epoch 6, batch 5150, loss[loss=0.3533, simple_loss=0.3948, pruned_loss=0.1559, over 8561.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3406, pruned_loss=0.1051, over 1617458.98 frames. ], batch size: 31, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:47:50,927 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8890, 1.5473, 2.2947, 1.9743, 2.0177, 1.6315, 1.3182, 1.1536], + device='cuda:3'), covar=tensor([0.1526, 0.1864, 0.0455, 0.0853, 0.0818, 0.0973, 0.0971, 0.1671], + device='cuda:3'), in_proj_covar=tensor([0.0791, 0.0729, 0.0621, 0.0719, 0.0828, 0.0672, 0.0637, 0.0672], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 02:48:23,749 INFO [train.py:901] (3/4) Epoch 6, batch 5200, loss[loss=0.2722, simple_loss=0.3444, pruned_loss=0.09998, over 8689.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3411, pruned_loss=0.1053, over 1612973.00 frames. ], batch size: 34, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:48:34,007 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.012e+02 3.204e+02 4.015e+02 4.654e+02 8.708e+02, threshold=8.029e+02, percent-clipped=4.0 +2023-02-06 02:48:57,684 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 02:48:59,760 INFO [train.py:901] (3/4) Epoch 6, batch 5250, loss[loss=0.2826, simple_loss=0.3588, pruned_loss=0.1032, over 8322.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3415, pruned_loss=0.1058, over 1613745.22 frames. ], batch size: 25, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:49:30,225 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45710.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 02:49:34,108 INFO [train.py:901] (3/4) Epoch 6, batch 5300, loss[loss=0.2761, simple_loss=0.3412, pruned_loss=0.1055, over 8245.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3405, pruned_loss=0.1055, over 1612482.25 frames. ], batch size: 24, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:49:36,957 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45720.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:49:43,554 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.559e+02 2.935e+02 3.437e+02 4.667e+02 1.283e+03, threshold=6.874e+02, percent-clipped=3.0 +2023-02-06 02:50:09,986 INFO [train.py:901] (3/4) Epoch 6, batch 5350, loss[loss=0.2733, simple_loss=0.3374, pruned_loss=0.1046, over 8108.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3402, pruned_loss=0.1055, over 1612979.96 frames. ], batch size: 23, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:50:20,660 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4014, 1.8539, 3.4386, 1.0567, 2.4118, 1.8732, 1.3599, 2.0971], + device='cuda:3'), covar=tensor([0.1584, 0.1935, 0.0664, 0.3316, 0.1394, 0.2369, 0.1541, 0.2269], + device='cuda:3'), in_proj_covar=tensor([0.0462, 0.0458, 0.0525, 0.0540, 0.0579, 0.0526, 0.0439, 0.0581], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 02:50:25,441 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45789.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:50:27,348 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45792.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:50:38,025 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5901, 1.2828, 4.8333, 1.7709, 4.0701, 3.9810, 4.3687, 4.2316], + device='cuda:3'), covar=tensor([0.0498, 0.3685, 0.0301, 0.2649, 0.1044, 0.0601, 0.0417, 0.0497], + device='cuda:3'), in_proj_covar=tensor([0.0366, 0.0518, 0.0463, 0.0455, 0.0517, 0.0425, 0.0439, 0.0482], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 02:50:42,760 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45814.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:50:42,778 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45814.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:50:43,428 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3947, 1.2631, 1.2402, 1.1240, 0.8952, 1.1119, 1.2939, 1.1398], + device='cuda:3'), covar=tensor([0.0657, 0.1314, 0.1826, 0.1439, 0.0603, 0.1530, 0.0704, 0.0632], + device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0170, 0.0209, 0.0173, 0.0120, 0.0178, 0.0133, 0.0146], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 02:50:43,905 INFO [train.py:901] (3/4) Epoch 6, batch 5400, loss[loss=0.3158, simple_loss=0.3704, pruned_loss=0.1306, over 8589.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3415, pruned_loss=0.1063, over 1614299.58 frames. ], batch size: 31, lr: 1.25e-02, grad_scale: 16.0 +2023-02-06 02:50:49,968 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45825.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 02:50:53,703 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.737e+02 2.842e+02 3.609e+02 4.644e+02 1.367e+03, threshold=7.218e+02, percent-clipped=2.0 +2023-02-06 02:50:59,048 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45839.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:51:17,301 INFO [train.py:901] (3/4) Epoch 6, batch 5450, loss[loss=0.2603, simple_loss=0.3386, pruned_loss=0.09101, over 7921.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3401, pruned_loss=0.1051, over 1610459.93 frames. ], batch size: 20, lr: 1.25e-02, grad_scale: 8.0 +2023-02-06 02:51:22,257 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9701, 3.4796, 2.5621, 3.9590, 1.9355, 2.1627, 2.4196, 3.3103], + device='cuda:3'), covar=tensor([0.0743, 0.0818, 0.1130, 0.0271, 0.1322, 0.1749, 0.1514, 0.0757], + device='cuda:3'), in_proj_covar=tensor([0.0267, 0.0249, 0.0280, 0.0227, 0.0244, 0.0278, 0.0287, 0.0248], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 02:51:26,215 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45877.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:51:47,555 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 02:51:51,930 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9683, 1.5899, 2.2581, 1.9378, 1.9797, 1.7282, 1.4595, 0.6276], + device='cuda:3'), covar=tensor([0.2339, 0.2240, 0.0647, 0.1174, 0.1046, 0.1340, 0.1160, 0.2270], + device='cuda:3'), in_proj_covar=tensor([0.0798, 0.0734, 0.0629, 0.0722, 0.0836, 0.0679, 0.0639, 0.0675], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 02:51:52,372 INFO [train.py:901] (3/4) Epoch 6, batch 5500, loss[loss=0.256, simple_loss=0.3154, pruned_loss=0.09827, over 7537.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3393, pruned_loss=0.1044, over 1610529.50 frames. ], batch size: 18, lr: 1.25e-02, grad_scale: 8.0 +2023-02-06 02:52:03,086 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.639e+02 2.821e+02 3.418e+02 4.385e+02 9.516e+02, threshold=6.836e+02, percent-clipped=4.0 +2023-02-06 02:52:27,018 INFO [train.py:901] (3/4) Epoch 6, batch 5550, loss[loss=0.2836, simple_loss=0.3322, pruned_loss=0.1176, over 7706.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3391, pruned_loss=0.1042, over 1611767.52 frames. ], batch size: 18, lr: 1.25e-02, grad_scale: 8.0 +2023-02-06 02:52:39,460 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 +2023-02-06 02:52:56,205 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4362, 1.5384, 1.5700, 1.3718, 1.1801, 1.3635, 1.7510, 1.8000], + device='cuda:3'), covar=tensor([0.0544, 0.1278, 0.1802, 0.1365, 0.0658, 0.1581, 0.0749, 0.0505], + device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0170, 0.0211, 0.0175, 0.0121, 0.0180, 0.0135, 0.0147], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 02:53:03,502 INFO [train.py:901] (3/4) Epoch 6, batch 5600, loss[loss=0.2463, simple_loss=0.3038, pruned_loss=0.09441, over 7973.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.34, pruned_loss=0.1052, over 1610232.58 frames. ], batch size: 21, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:53:13,363 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.150e+02 2.809e+02 3.495e+02 4.670e+02 1.291e+03, threshold=6.989e+02, percent-clipped=6.0 +2023-02-06 02:53:33,464 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0781, 1.7250, 2.7724, 2.0732, 2.4032, 1.8959, 1.4221, 0.9501], + device='cuda:3'), covar=tensor([0.2341, 0.2388, 0.0522, 0.1372, 0.1036, 0.1275, 0.1150, 0.2516], + device='cuda:3'), in_proj_covar=tensor([0.0813, 0.0744, 0.0637, 0.0729, 0.0842, 0.0687, 0.0647, 0.0683], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 02:53:36,014 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46064.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:53:37,308 INFO [train.py:901] (3/4) Epoch 6, batch 5650, loss[loss=0.2802, simple_loss=0.343, pruned_loss=0.1087, over 8608.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3402, pruned_loss=0.1055, over 1611356.72 frames. ], batch size: 34, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:53:47,430 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46081.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 02:53:51,172 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 02:53:59,577 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-06 02:54:04,756 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46106.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 02:54:12,525 INFO [train.py:901] (3/4) Epoch 6, batch 5700, loss[loss=0.2644, simple_loss=0.335, pruned_loss=0.09687, over 8536.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3397, pruned_loss=0.1046, over 1610988.81 frames. ], batch size: 28, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:54:22,543 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.760e+02 2.829e+02 3.489e+02 4.392e+02 1.030e+03, threshold=6.978e+02, percent-clipped=3.0 +2023-02-06 02:54:25,975 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46136.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:54:46,332 INFO [train.py:901] (3/4) Epoch 6, batch 5750, loss[loss=0.2661, simple_loss=0.3381, pruned_loss=0.09701, over 8513.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3401, pruned_loss=0.1048, over 1613122.79 frames. ], batch size: 28, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:54:53,641 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 02:54:55,248 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46179.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:55:01,704 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-06 02:55:21,182 INFO [train.py:901] (3/4) Epoch 6, batch 5800, loss[loss=0.288, simple_loss=0.3608, pruned_loss=0.1076, over 8500.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3408, pruned_loss=0.1044, over 1619352.73 frames. ], batch size: 28, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:55:24,803 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46221.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:55:32,627 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.746e+02 2.949e+02 3.358e+02 4.338e+02 9.471e+02, threshold=6.717e+02, percent-clipped=1.0 +2023-02-06 02:55:45,837 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46251.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:55:55,787 INFO [train.py:901] (3/4) Epoch 6, batch 5850, loss[loss=0.2243, simple_loss=0.2968, pruned_loss=0.07586, over 7792.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3403, pruned_loss=0.1042, over 1620781.16 frames. ], batch size: 19, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:56:14,992 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46294.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:56:29,476 INFO [train.py:901] (3/4) Epoch 6, batch 5900, loss[loss=0.2917, simple_loss=0.3436, pruned_loss=0.1198, over 8445.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3416, pruned_loss=0.1056, over 1622486.98 frames. ], batch size: 27, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:56:39,482 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.969e+02 3.022e+02 3.849e+02 5.141e+02 8.536e+02, threshold=7.697e+02, percent-clipped=7.0 +2023-02-06 02:56:43,702 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46336.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:57:04,658 INFO [train.py:901] (3/4) Epoch 6, batch 5950, loss[loss=0.2424, simple_loss=0.3122, pruned_loss=0.08627, over 7973.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3404, pruned_loss=0.1052, over 1617017.14 frames. ], batch size: 21, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:57:38,431 INFO [train.py:901] (3/4) Epoch 6, batch 6000, loss[loss=0.3541, simple_loss=0.3964, pruned_loss=0.1559, over 7065.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3404, pruned_loss=0.1049, over 1617804.51 frames. ], batch size: 71, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:57:38,432 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 02:57:50,760 INFO [train.py:935] (3/4) Epoch 6, validation: loss=0.2127, simple_loss=0.3094, pruned_loss=0.05799, over 944034.00 frames. +2023-02-06 02:57:50,761 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 02:58:01,259 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.784e+02 2.683e+02 3.226e+02 4.100e+02 1.140e+03, threshold=6.453e+02, percent-clipped=1.0 +2023-02-06 02:58:04,347 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46435.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:58:06,388 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.7160, 3.6963, 3.3656, 1.7014, 3.2668, 3.3486, 3.3761, 2.9576], + device='cuda:3'), covar=tensor([0.1079, 0.0782, 0.1134, 0.5397, 0.1000, 0.1172, 0.1719, 0.1105], + device='cuda:3'), in_proj_covar=tensor([0.0413, 0.0312, 0.0340, 0.0429, 0.0330, 0.0304, 0.0326, 0.0273], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 02:58:11,325 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2007, 1.6193, 3.8774, 1.7973, 2.5530, 4.4070, 4.1931, 3.8580], + device='cuda:3'), covar=tensor([0.1025, 0.1476, 0.0367, 0.1905, 0.0883, 0.0177, 0.0320, 0.0496], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0258, 0.0224, 0.0263, 0.0230, 0.0209, 0.0248, 0.0270], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2023-02-06 02:58:21,760 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46460.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:58:25,654 INFO [train.py:901] (3/4) Epoch 6, batch 6050, loss[loss=0.2752, simple_loss=0.3427, pruned_loss=0.1038, over 8487.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3411, pruned_loss=0.1049, over 1617480.71 frames. ], batch size: 49, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:58:56,115 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46507.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:59:02,121 INFO [train.py:901] (3/4) Epoch 6, batch 6100, loss[loss=0.2009, simple_loss=0.2756, pruned_loss=0.06315, over 7938.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3418, pruned_loss=0.1051, over 1619636.83 frames. ], batch size: 20, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:59:12,642 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.922e+02 3.046e+02 3.657e+02 4.398e+02 9.620e+02, threshold=7.315e+02, percent-clipped=4.0 +2023-02-06 02:59:13,539 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46532.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 02:59:20,475 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2159, 1.5721, 3.3991, 1.4389, 2.2413, 3.7877, 3.6996, 3.2450], + device='cuda:3'), covar=tensor([0.0957, 0.1418, 0.0412, 0.2049, 0.0860, 0.0246, 0.0398, 0.0609], + device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0266, 0.0230, 0.0268, 0.0237, 0.0213, 0.0256, 0.0278], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 02:59:24,583 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 02:59:37,764 INFO [train.py:901] (3/4) Epoch 6, batch 6150, loss[loss=0.3023, simple_loss=0.36, pruned_loss=0.1223, over 7933.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3415, pruned_loss=0.1052, over 1617819.22 frames. ], batch size: 20, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 02:59:56,266 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46592.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:00:11,873 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46613.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:00:13,815 INFO [train.py:901] (3/4) Epoch 6, batch 6200, loss[loss=0.2804, simple_loss=0.3568, pruned_loss=0.102, over 8734.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3425, pruned_loss=0.1064, over 1617095.86 frames. ], batch size: 49, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 03:00:14,154 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 +2023-02-06 03:00:14,732 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46617.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:00:16,763 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46620.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:00:24,151 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.053e+02 3.051e+02 3.861e+02 4.926e+02 1.016e+03, threshold=7.722e+02, percent-clipped=3.0 +2023-02-06 03:00:28,968 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46638.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:00:48,696 INFO [train.py:901] (3/4) Epoch 6, batch 6250, loss[loss=0.2945, simple_loss=0.3606, pruned_loss=0.1142, over 8200.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3407, pruned_loss=0.1053, over 1616080.05 frames. ], batch size: 23, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 03:01:22,866 INFO [train.py:901] (3/4) Epoch 6, batch 6300, loss[loss=0.3172, simple_loss=0.3731, pruned_loss=0.1306, over 8245.00 frames. ], tot_loss[loss=0.274, simple_loss=0.339, pruned_loss=0.1045, over 1613158.16 frames. ], batch size: 24, lr: 1.24e-02, grad_scale: 8.0 +2023-02-06 03:01:24,468 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2859, 1.3231, 2.3280, 1.1568, 2.1760, 2.4972, 2.5171, 2.1399], + device='cuda:3'), covar=tensor([0.0973, 0.1094, 0.0430, 0.1917, 0.0511, 0.0370, 0.0491, 0.0707], + device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0267, 0.0228, 0.0270, 0.0234, 0.0213, 0.0253, 0.0276], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 03:01:34,435 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.898e+02 2.734e+02 3.399e+02 4.377e+02 1.449e+03, threshold=6.797e+02, percent-clipped=4.0 +2023-02-06 03:01:49,381 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46753.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:01:57,778 INFO [train.py:901] (3/4) Epoch 6, batch 6350, loss[loss=0.2546, simple_loss=0.3236, pruned_loss=0.09276, over 7650.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3386, pruned_loss=0.1043, over 1614559.87 frames. ], batch size: 19, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:02:32,250 INFO [train.py:901] (3/4) Epoch 6, batch 6400, loss[loss=0.2491, simple_loss=0.3003, pruned_loss=0.09892, over 7543.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3381, pruned_loss=0.1042, over 1611710.12 frames. ], batch size: 18, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:02:41,198 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46828.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:02:43,104 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.753e+02 2.744e+02 3.578e+02 4.396e+02 9.504e+02, threshold=7.157e+02, percent-clipped=5.0 +2023-02-06 03:03:07,338 INFO [train.py:901] (3/4) Epoch 6, batch 6450, loss[loss=0.3358, simple_loss=0.3914, pruned_loss=0.1402, over 8462.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3374, pruned_loss=0.1037, over 1610496.07 frames. ], batch size: 29, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:03:15,539 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8642, 1.3223, 1.4064, 1.2091, 1.0665, 1.3194, 1.5280, 1.2917], + device='cuda:3'), covar=tensor([0.0609, 0.1338, 0.1909, 0.1489, 0.0630, 0.1626, 0.0763, 0.0668], + device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0169, 0.0209, 0.0172, 0.0119, 0.0178, 0.0132, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 03:03:41,560 INFO [train.py:901] (3/4) Epoch 6, batch 6500, loss[loss=0.2621, simple_loss=0.3398, pruned_loss=0.09221, over 8536.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3379, pruned_loss=0.1041, over 1610923.27 frames. ], batch size: 34, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:03:51,595 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.994e+02 3.001e+02 3.759e+02 4.377e+02 1.086e+03, threshold=7.517e+02, percent-clipped=1.0 +2023-02-06 03:04:09,448 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46957.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:04:14,549 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46964.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:04:15,758 INFO [train.py:901] (3/4) Epoch 6, batch 6550, loss[loss=0.3234, simple_loss=0.3926, pruned_loss=0.1271, over 8107.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3387, pruned_loss=0.1046, over 1613653.54 frames. ], batch size: 23, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:04:37,517 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 03:04:45,803 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47009.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:04:50,982 INFO [train.py:901] (3/4) Epoch 6, batch 6600, loss[loss=0.2459, simple_loss=0.3309, pruned_loss=0.08043, over 8475.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.34, pruned_loss=0.1051, over 1615223.11 frames. ], batch size: 25, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:04:56,435 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 03:05:01,160 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.072e+02 2.927e+02 3.687e+02 4.772e+02 1.123e+03, threshold=7.374e+02, percent-clipped=4.0 +2023-02-06 03:05:03,302 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47034.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:05:04,658 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3509, 2.8763, 1.9748, 2.0925, 2.2395, 1.5613, 1.8004, 2.2788], + device='cuda:3'), covar=tensor([0.1285, 0.0300, 0.0755, 0.0561, 0.0580, 0.1242, 0.0973, 0.0780], + device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0233, 0.0309, 0.0301, 0.0312, 0.0315, 0.0340, 0.0313], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 03:05:25,535 INFO [train.py:901] (3/4) Epoch 6, batch 6650, loss[loss=0.3755, simple_loss=0.4097, pruned_loss=0.1707, over 6680.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.34, pruned_loss=0.1055, over 1611313.55 frames. ], batch size: 72, lr: 1.23e-02, grad_scale: 4.0 +2023-02-06 03:05:30,595 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47072.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:05:35,423 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47079.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:06:00,860 INFO [train.py:901] (3/4) Epoch 6, batch 6700, loss[loss=0.2716, simple_loss=0.3531, pruned_loss=0.09509, over 8466.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3396, pruned_loss=0.1046, over 1614142.08 frames. ], batch size: 25, lr: 1.23e-02, grad_scale: 4.0 +2023-02-06 03:06:07,896 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4998, 2.8412, 1.9031, 2.2487, 2.3155, 1.5590, 2.1773, 2.2303], + device='cuda:3'), covar=tensor([0.1174, 0.0257, 0.0773, 0.0511, 0.0574, 0.1186, 0.0749, 0.0714], + device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0231, 0.0309, 0.0299, 0.0309, 0.0314, 0.0337, 0.0310], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 03:06:12,485 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.787e+02 2.777e+02 3.640e+02 4.922e+02 1.093e+03, threshold=7.281e+02, percent-clipped=6.0 +2023-02-06 03:06:33,731 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2456, 1.8905, 1.9061, 1.7861, 1.5654, 1.7740, 2.4555, 2.1151], + device='cuda:3'), covar=tensor([0.0474, 0.1140, 0.1791, 0.1232, 0.0529, 0.1541, 0.0590, 0.0597], + device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0168, 0.0208, 0.0171, 0.0118, 0.0177, 0.0132, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 03:06:34,837 INFO [train.py:901] (3/4) Epoch 6, batch 6750, loss[loss=0.2691, simple_loss=0.342, pruned_loss=0.09805, over 8250.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3382, pruned_loss=0.1043, over 1611372.55 frames. ], batch size: 22, lr: 1.23e-02, grad_scale: 4.0 +2023-02-06 03:06:38,969 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47172.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:07:10,710 INFO [train.py:901] (3/4) Epoch 6, batch 6800, loss[loss=0.3078, simple_loss=0.3711, pruned_loss=0.1222, over 8497.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3374, pruned_loss=0.1033, over 1612382.17 frames. ], batch size: 26, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:07:12,675 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 03:07:21,208 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.621e+02 2.942e+02 3.591e+02 4.804e+02 1.528e+03, threshold=7.182e+02, percent-clipped=7.0 +2023-02-06 03:07:35,889 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9453, 1.3154, 1.3736, 1.1479, 1.0569, 1.2741, 1.5308, 1.5881], + device='cuda:3'), covar=tensor([0.0539, 0.1385, 0.1798, 0.1453, 0.0674, 0.1732, 0.0739, 0.0559], + device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0167, 0.0207, 0.0170, 0.0119, 0.0176, 0.0131, 0.0144], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0007, 0.0008, 0.0007, 0.0005, 0.0007, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 03:07:40,015 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47259.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:07:41,479 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8439, 2.9926, 2.5264, 4.2525, 1.7531, 1.7546, 2.4332, 3.2016], + device='cuda:3'), covar=tensor([0.0874, 0.1178, 0.1322, 0.0217, 0.1685, 0.2087, 0.1644, 0.0998], + device='cuda:3'), in_proj_covar=tensor([0.0267, 0.0252, 0.0279, 0.0225, 0.0249, 0.0274, 0.0280, 0.0249], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 03:07:44,575 INFO [train.py:901] (3/4) Epoch 6, batch 6850, loss[loss=0.3601, simple_loss=0.4043, pruned_loss=0.158, over 8631.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3371, pruned_loss=0.1029, over 1609869.33 frames. ], batch size: 34, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:07:50,959 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7437, 1.5732, 3.4189, 1.3819, 2.2753, 3.7814, 3.6851, 3.3287], + device='cuda:3'), covar=tensor([0.0993, 0.1335, 0.0288, 0.1792, 0.0727, 0.0218, 0.0394, 0.0523], + device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0263, 0.0221, 0.0263, 0.0230, 0.0207, 0.0252, 0.0269], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2023-02-06 03:07:58,973 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47287.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:08:00,897 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 03:08:19,248 INFO [train.py:901] (3/4) Epoch 6, batch 6900, loss[loss=0.2924, simple_loss=0.3676, pruned_loss=0.1086, over 8562.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3378, pruned_loss=0.1032, over 1613946.47 frames. ], batch size: 34, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:08:28,208 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47328.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:08:30,616 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.066e+02 2.873e+02 3.537e+02 4.379e+02 9.664e+02, threshold=7.075e+02, percent-clipped=2.0 +2023-02-06 03:08:32,865 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47335.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:08:44,861 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47353.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:08:49,798 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47360.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:08:54,247 INFO [train.py:901] (3/4) Epoch 6, batch 6950, loss[loss=0.2471, simple_loss=0.3321, pruned_loss=0.08103, over 8507.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3371, pruned_loss=0.1024, over 1616118.56 frames. ], batch size: 48, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:09:09,780 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 03:09:15,148 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47397.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:09:15,352 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 03:09:28,483 INFO [train.py:901] (3/4) Epoch 6, batch 7000, loss[loss=0.2259, simple_loss=0.2981, pruned_loss=0.07689, over 7703.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3371, pruned_loss=0.1023, over 1615185.91 frames. ], batch size: 18, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:09:39,921 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.860e+02 2.784e+02 3.553e+02 4.437e+02 1.281e+03, threshold=7.106e+02, percent-clipped=4.0 +2023-02-06 03:10:03,577 INFO [train.py:901] (3/4) Epoch 6, batch 7050, loss[loss=0.2419, simple_loss=0.3017, pruned_loss=0.09103, over 7795.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.338, pruned_loss=0.1036, over 1614650.39 frames. ], batch size: 19, lr: 1.23e-02, grad_scale: 8.0 +2023-02-06 03:10:18,702 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1045, 3.8740, 2.2842, 2.5931, 2.8769, 2.0066, 2.6883, 3.0033], + device='cuda:3'), covar=tensor([0.1295, 0.0273, 0.0847, 0.0659, 0.0623, 0.1077, 0.0864, 0.0805], + device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0232, 0.0307, 0.0299, 0.0314, 0.0312, 0.0342, 0.0309], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 03:10:19,958 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4893, 1.2330, 4.6477, 1.7736, 3.9009, 3.9334, 4.1240, 4.0310], + device='cuda:3'), covar=tensor([0.0488, 0.4128, 0.0344, 0.2765, 0.1157, 0.0684, 0.0491, 0.0598], + device='cuda:3'), in_proj_covar=tensor([0.0369, 0.0508, 0.0450, 0.0445, 0.0508, 0.0424, 0.0432, 0.0475], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2023-02-06 03:10:37,641 INFO [train.py:901] (3/4) Epoch 6, batch 7100, loss[loss=0.2565, simple_loss=0.3223, pruned_loss=0.0954, over 7974.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3387, pruned_loss=0.1039, over 1617187.33 frames. ], batch size: 21, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:10:48,827 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.154e+02 3.207e+02 3.842e+02 5.073e+02 1.424e+03, threshold=7.684e+02, percent-clipped=2.0 +2023-02-06 03:10:56,280 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47543.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:11:00,495 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.29 vs. limit=5.0 +2023-02-06 03:11:12,602 INFO [train.py:901] (3/4) Epoch 6, batch 7150, loss[loss=0.2912, simple_loss=0.3555, pruned_loss=0.1134, over 8601.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3379, pruned_loss=0.1038, over 1615312.58 frames. ], batch size: 39, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:11:14,091 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47568.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:11:37,946 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47603.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:11:46,759 INFO [train.py:901] (3/4) Epoch 6, batch 7200, loss[loss=0.3385, simple_loss=0.3896, pruned_loss=0.1437, over 7197.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3368, pruned_loss=0.104, over 1606679.35 frames. ], batch size: 72, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:11:57,791 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.907e+02 2.983e+02 3.737e+02 4.630e+02 8.445e+02, threshold=7.473e+02, percent-clipped=4.0 +2023-02-06 03:12:02,846 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5730, 1.3389, 3.0811, 1.4019, 2.1791, 3.3646, 3.3357, 2.9315], + device='cuda:3'), covar=tensor([0.1096, 0.1313, 0.0355, 0.1844, 0.0720, 0.0253, 0.0411, 0.0589], + device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0266, 0.0225, 0.0268, 0.0232, 0.0209, 0.0256, 0.0275], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 03:12:22,028 INFO [train.py:901] (3/4) Epoch 6, batch 7250, loss[loss=0.2819, simple_loss=0.3457, pruned_loss=0.1091, over 8285.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3367, pruned_loss=0.1036, over 1608685.50 frames. ], batch size: 23, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:12:42,080 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 +2023-02-06 03:12:45,389 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2093, 1.7759, 2.7498, 2.1189, 2.2293, 1.9386, 1.5175, 0.9999], + device='cuda:3'), covar=tensor([0.2169, 0.2390, 0.0551, 0.1431, 0.1176, 0.1337, 0.1278, 0.2443], + device='cuda:3'), in_proj_covar=tensor([0.0826, 0.0755, 0.0655, 0.0745, 0.0850, 0.0694, 0.0654, 0.0693], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 03:12:56,481 INFO [train.py:901] (3/4) Epoch 6, batch 7300, loss[loss=0.2848, simple_loss=0.3558, pruned_loss=0.1068, over 8630.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3373, pruned_loss=0.1038, over 1605631.29 frames. ], batch size: 34, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:12:57,903 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47718.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:13:07,209 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.066e+02 3.071e+02 3.696e+02 4.839e+02 1.031e+03, threshold=7.393e+02, percent-clipped=2.0 +2023-02-06 03:13:13,285 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47741.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:13:23,407 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47756.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:13:30,060 INFO [train.py:901] (3/4) Epoch 6, batch 7350, loss[loss=0.3171, simple_loss=0.3758, pruned_loss=0.1292, over 8473.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3368, pruned_loss=0.1031, over 1604439.58 frames. ], batch size: 25, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:13:45,831 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 03:13:48,798 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 03:13:59,324 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47806.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:14:05,871 INFO [train.py:901] (3/4) Epoch 6, batch 7400, loss[loss=0.2034, simple_loss=0.2806, pruned_loss=0.0631, over 7447.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3355, pruned_loss=0.1031, over 1600293.26 frames. ], batch size: 17, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:14:08,036 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 03:14:13,527 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47827.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:14:17,406 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.816e+02 3.124e+02 3.904e+02 4.877e+02 9.892e+02, threshold=7.808e+02, percent-clipped=5.0 +2023-02-06 03:14:33,733 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47856.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:14:40,242 INFO [train.py:901] (3/4) Epoch 6, batch 7450, loss[loss=0.2723, simple_loss=0.329, pruned_loss=0.1078, over 7531.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.338, pruned_loss=0.1047, over 1605447.82 frames. ], batch size: 18, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:14:46,244 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 03:15:15,311 INFO [train.py:901] (3/4) Epoch 6, batch 7500, loss[loss=0.2166, simple_loss=0.2986, pruned_loss=0.06732, over 8365.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3383, pruned_loss=0.1046, over 1607524.26 frames. ], batch size: 24, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:15:25,969 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.098e+02 3.102e+02 3.706e+02 4.699e+02 1.511e+03, threshold=7.412e+02, percent-clipped=9.0 +2023-02-06 03:15:49,279 INFO [train.py:901] (3/4) Epoch 6, batch 7550, loss[loss=0.2508, simple_loss=0.3317, pruned_loss=0.08498, over 8333.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3376, pruned_loss=0.1039, over 1607707.82 frames. ], batch size: 25, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:15:54,927 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47974.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:16:04,325 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5343, 2.9590, 3.0524, 1.8997, 1.4498, 3.0062, 0.5399, 2.0888], + device='cuda:3'), covar=tensor([0.3244, 0.1844, 0.1197, 0.3983, 0.6298, 0.0784, 0.5406, 0.2046], + device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0132, 0.0081, 0.0182, 0.0226, 0.0082, 0.0142, 0.0136], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 03:16:11,854 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47999.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:16:24,750 INFO [train.py:901] (3/4) Epoch 6, batch 7600, loss[loss=0.2947, simple_loss=0.3457, pruned_loss=0.1218, over 8072.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3374, pruned_loss=0.1035, over 1612573.38 frames. ], batch size: 21, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:16:27,980 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 +2023-02-06 03:16:34,272 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-06 03:16:37,192 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.899e+02 2.897e+02 3.536e+02 4.611e+02 2.294e+03, threshold=7.072e+02, percent-clipped=5.0 +2023-02-06 03:17:01,525 INFO [train.py:901] (3/4) Epoch 6, batch 7650, loss[loss=0.213, simple_loss=0.295, pruned_loss=0.06553, over 7668.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3369, pruned_loss=0.103, over 1611054.98 frames. ], batch size: 19, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:17:17,558 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48090.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:17:24,181 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48100.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:17:32,462 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48112.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:17:34,935 INFO [train.py:901] (3/4) Epoch 6, batch 7700, loss[loss=0.2675, simple_loss=0.3221, pruned_loss=0.1064, over 7268.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3371, pruned_loss=0.103, over 1613580.86 frames. ], batch size: 16, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:17:46,038 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.945e+02 2.821e+02 3.617e+02 4.667e+02 9.808e+02, threshold=7.234e+02, percent-clipped=3.0 +2023-02-06 03:17:50,861 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48137.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:17:57,309 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 03:17:59,320 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48150.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:18:03,433 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.1845, 3.1641, 2.9098, 1.4904, 2.8168, 2.7131, 2.9209, 2.5215], + device='cuda:3'), covar=tensor([0.1476, 0.0959, 0.1472, 0.4738, 0.1115, 0.1531, 0.1886, 0.1224], + device='cuda:3'), in_proj_covar=tensor([0.0403, 0.0310, 0.0332, 0.0412, 0.0324, 0.0301, 0.0311, 0.0268], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 03:18:10,106 INFO [train.py:901] (3/4) Epoch 6, batch 7750, loss[loss=0.292, simple_loss=0.3577, pruned_loss=0.1131, over 8296.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3382, pruned_loss=0.1035, over 1616209.38 frames. ], batch size: 23, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:18:13,414 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48171.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:18:43,311 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48215.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:18:43,803 INFO [train.py:901] (3/4) Epoch 6, batch 7800, loss[loss=0.2842, simple_loss=0.3496, pruned_loss=0.1094, over 8500.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3392, pruned_loss=0.1039, over 1618146.53 frames. ], batch size: 26, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:18:47,460 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3021, 2.2533, 1.4668, 1.9861, 1.8503, 1.2660, 1.6289, 1.7393], + device='cuda:3'), covar=tensor([0.1009, 0.0290, 0.0949, 0.0426, 0.0573, 0.1127, 0.0725, 0.0648], + device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0230, 0.0316, 0.0302, 0.0313, 0.0316, 0.0341, 0.0314], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 03:18:53,437 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48230.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:18:54,591 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.145e+02 3.053e+02 3.731e+02 4.789e+02 1.133e+03, threshold=7.462e+02, percent-clipped=3.0 +2023-02-06 03:19:16,584 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48265.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:19:17,072 INFO [train.py:901] (3/4) Epoch 6, batch 7850, loss[loss=0.2397, simple_loss=0.3084, pruned_loss=0.08547, over 7927.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3395, pruned_loss=0.1037, over 1622738.12 frames. ], batch size: 20, lr: 1.22e-02, grad_scale: 8.0 +2023-02-06 03:19:30,593 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48286.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:19:51,000 INFO [train.py:901] (3/4) Epoch 6, batch 7900, loss[loss=0.2562, simple_loss=0.3351, pruned_loss=0.08867, over 8314.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3391, pruned_loss=0.1034, over 1620618.08 frames. ], batch size: 25, lr: 1.21e-02, grad_scale: 8.0 +2023-02-06 03:19:57,278 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.2455, 4.2239, 3.7717, 1.9487, 3.7670, 3.7406, 3.9525, 3.3625], + device='cuda:3'), covar=tensor([0.0860, 0.0638, 0.0967, 0.4837, 0.0870, 0.0980, 0.1194, 0.1195], + device='cuda:3'), in_proj_covar=tensor([0.0402, 0.0311, 0.0333, 0.0413, 0.0322, 0.0302, 0.0310, 0.0267], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 03:20:01,872 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.890e+02 2.927e+02 3.494e+02 4.326e+02 7.205e+02, threshold=6.988e+02, percent-clipped=0.0 +2023-02-06 03:20:09,338 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48342.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:20:25,105 INFO [train.py:901] (3/4) Epoch 6, batch 7950, loss[loss=0.2609, simple_loss=0.3209, pruned_loss=0.1005, over 7645.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3378, pruned_loss=0.1029, over 1622202.44 frames. ], batch size: 19, lr: 1.21e-02, grad_scale: 8.0 +2023-02-06 03:20:58,689 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48415.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:20:59,237 INFO [train.py:901] (3/4) Epoch 6, batch 8000, loss[loss=0.2668, simple_loss=0.3348, pruned_loss=0.09936, over 8474.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3378, pruned_loss=0.1027, over 1621555.24 frames. ], batch size: 28, lr: 1.21e-02, grad_scale: 8.0 +2023-02-06 03:21:10,358 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.174e+02 2.873e+02 3.488e+02 4.217e+02 8.104e+02, threshold=6.977e+02, percent-clipped=2.0 +2023-02-06 03:21:11,764 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48434.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:21:16,779 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48441.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:21:33,917 INFO [train.py:901] (3/4) Epoch 6, batch 8050, loss[loss=0.216, simple_loss=0.2816, pruned_loss=0.07518, over 7554.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3361, pruned_loss=0.103, over 1596911.05 frames. ], batch size: 18, lr: 1.21e-02, grad_scale: 8.0 +2023-02-06 03:21:36,916 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3939, 2.2907, 3.1511, 2.0686, 2.6774, 3.4821, 3.2304, 3.1690], + device='cuda:3'), covar=tensor([0.0709, 0.0920, 0.0593, 0.1362, 0.0783, 0.0221, 0.0438, 0.0456], + device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0273, 0.0225, 0.0268, 0.0234, 0.0211, 0.0260, 0.0279], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 03:21:37,664 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48471.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:21:54,567 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48496.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:22:07,093 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 03:22:10,949 INFO [train.py:901] (3/4) Epoch 7, batch 0, loss[loss=0.2996, simple_loss=0.3591, pruned_loss=0.1201, over 8558.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3591, pruned_loss=0.1201, over 8558.00 frames. ], batch size: 39, lr: 1.14e-02, grad_scale: 8.0 +2023-02-06 03:22:10,950 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 03:22:22,759 INFO [train.py:935] (3/4) Epoch 7, validation: loss=0.2113, simple_loss=0.3091, pruned_loss=0.05678, over 944034.00 frames. +2023-02-06 03:22:22,760 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 03:22:28,414 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8674, 2.1743, 1.7800, 2.6929, 1.1407, 1.5530, 1.7914, 2.1690], + device='cuda:3'), covar=tensor([0.0964, 0.0941, 0.1279, 0.0465, 0.1399, 0.1713, 0.1222, 0.0895], + device='cuda:3'), in_proj_covar=tensor([0.0263, 0.0239, 0.0276, 0.0221, 0.0240, 0.0267, 0.0273, 0.0244], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 03:22:37,613 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48521.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:22:38,081 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 03:22:39,699 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6350, 1.8537, 2.1963, 1.0464, 2.3379, 1.3406, 0.7125, 1.6980], + device='cuda:3'), covar=tensor([0.0287, 0.0186, 0.0110, 0.0281, 0.0140, 0.0412, 0.0418, 0.0153], + device='cuda:3'), in_proj_covar=tensor([0.0340, 0.0251, 0.0203, 0.0302, 0.0238, 0.0385, 0.0310, 0.0287], + device='cuda:3'), out_proj_covar=tensor([1.1150e-04, 8.0551e-05, 6.3816e-05, 9.6298e-05, 7.6935e-05, 1.3405e-04, + 1.0118e-04, 9.2305e-05], device='cuda:3') +2023-02-06 03:22:41,674 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6137, 2.8933, 2.3544, 4.1629, 1.5997, 1.8397, 2.2062, 2.7918], + device='cuda:3'), covar=tensor([0.0926, 0.1224, 0.1455, 0.0298, 0.1785, 0.2031, 0.1817, 0.1315], + device='cuda:3'), in_proj_covar=tensor([0.0266, 0.0242, 0.0278, 0.0224, 0.0244, 0.0271, 0.0277, 0.0247], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 03:22:45,429 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.948e+02 2.727e+02 3.570e+02 4.321e+02 1.428e+03, threshold=7.140e+02, percent-clipped=5.0 +2023-02-06 03:22:53,284 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48542.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:22:55,826 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48546.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:22:56,687 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 03:22:57,631 INFO [train.py:901] (3/4) Epoch 7, batch 50, loss[loss=0.2501, simple_loss=0.3189, pruned_loss=0.09067, over 8093.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3328, pruned_loss=0.1008, over 362824.91 frames. ], batch size: 21, lr: 1.14e-02, grad_scale: 8.0 +2023-02-06 03:22:57,789 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48549.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:23:09,791 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48567.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:23:12,926 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 03:23:14,280 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48574.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:23:31,378 INFO [train.py:901] (3/4) Epoch 7, batch 100, loss[loss=0.3118, simple_loss=0.3703, pruned_loss=0.1267, over 8129.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3345, pruned_loss=0.1017, over 642318.41 frames. ], batch size: 22, lr: 1.14e-02, grad_scale: 8.0 +2023-02-06 03:23:34,987 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 03:23:44,206 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5141, 2.2897, 4.4781, 1.3061, 3.0203, 2.1785, 1.7053, 2.5153], + device='cuda:3'), covar=tensor([0.1565, 0.1809, 0.0589, 0.3158, 0.1315, 0.2290, 0.1509, 0.2225], + device='cuda:3'), in_proj_covar=tensor([0.0474, 0.0473, 0.0526, 0.0547, 0.0594, 0.0534, 0.0451, 0.0592], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 03:23:54,579 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.847e+02 2.936e+02 3.434e+02 4.642e+02 8.961e+02, threshold=6.868e+02, percent-clipped=3.0 +2023-02-06 03:24:06,735 INFO [train.py:901] (3/4) Epoch 7, batch 150, loss[loss=0.241, simple_loss=0.3215, pruned_loss=0.08025, over 8135.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.336, pruned_loss=0.1022, over 860941.54 frames. ], batch size: 22, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:24:28,699 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2373, 1.4659, 2.2279, 1.0884, 1.6126, 1.4903, 1.3236, 1.5135], + device='cuda:3'), covar=tensor([0.1527, 0.1835, 0.0701, 0.3083, 0.1448, 0.2593, 0.1561, 0.1652], + device='cuda:3'), in_proj_covar=tensor([0.0470, 0.0470, 0.0523, 0.0541, 0.0591, 0.0527, 0.0448, 0.0587], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 03:24:31,975 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48686.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:24:34,069 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48689.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:24:40,631 INFO [train.py:901] (3/4) Epoch 7, batch 200, loss[loss=0.2974, simple_loss=0.3595, pruned_loss=0.1176, over 8633.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3358, pruned_loss=0.1024, over 1026603.26 frames. ], batch size: 39, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:25:03,499 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.914e+02 2.629e+02 3.306e+02 4.274e+02 1.004e+03, threshold=6.612e+02, percent-clipped=3.0 +2023-02-06 03:25:15,498 INFO [train.py:901] (3/4) Epoch 7, batch 250, loss[loss=0.2793, simple_loss=0.3485, pruned_loss=0.1051, over 8684.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3356, pruned_loss=0.1012, over 1161054.12 frames. ], batch size: 39, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:25:22,711 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48759.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:25:26,770 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 03:25:35,605 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 03:25:41,109 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48785.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:25:44,865 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-06 03:25:50,539 INFO [train.py:901] (3/4) Epoch 7, batch 300, loss[loss=0.2632, simple_loss=0.3373, pruned_loss=0.09461, over 8556.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3372, pruned_loss=0.1018, over 1266871.54 frames. ], batch size: 31, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:25:52,230 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48801.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:25:54,978 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48805.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:26:12,387 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48830.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:26:13,528 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.768e+02 2.973e+02 3.476e+02 4.340e+02 1.124e+03, threshold=6.953e+02, percent-clipped=5.0 +2023-02-06 03:26:18,391 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48839.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:26:18,505 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2751, 1.8020, 2.9008, 2.2864, 2.5057, 2.0384, 1.5099, 1.1888], + device='cuda:3'), covar=tensor([0.2445, 0.2681, 0.0640, 0.1437, 0.1183, 0.1323, 0.1293, 0.2577], + device='cuda:3'), in_proj_covar=tensor([0.0800, 0.0742, 0.0640, 0.0731, 0.0829, 0.0683, 0.0638, 0.0672], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 03:26:25,098 INFO [train.py:901] (3/4) Epoch 7, batch 350, loss[loss=0.2757, simple_loss=0.3463, pruned_loss=0.1026, over 8477.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3366, pruned_loss=0.1009, over 1349370.12 frames. ], batch size: 25, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:26:30,605 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48856.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:26:34,339 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 +2023-02-06 03:26:43,222 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48874.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:26:49,431 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.39 vs. limit=5.0 +2023-02-06 03:27:00,371 INFO [train.py:901] (3/4) Epoch 7, batch 400, loss[loss=0.2129, simple_loss=0.2799, pruned_loss=0.07296, over 7425.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3345, pruned_loss=0.1002, over 1406979.28 frames. ], batch size: 17, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:27:01,279 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48900.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:27:06,078 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 +2023-02-06 03:27:09,953 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4688, 1.9569, 3.3733, 1.2238, 2.3679, 1.9533, 1.6369, 2.1742], + device='cuda:3'), covar=tensor([0.1488, 0.1627, 0.0536, 0.3009, 0.1206, 0.2134, 0.1380, 0.1743], + device='cuda:3'), in_proj_covar=tensor([0.0472, 0.0467, 0.0524, 0.0542, 0.0585, 0.0524, 0.0451, 0.0584], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 03:27:22,460 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.855e+02 2.734e+02 3.619e+02 4.506e+02 1.679e+03, threshold=7.237e+02, percent-clipped=8.0 +2023-02-06 03:27:29,397 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4113, 4.3078, 3.9923, 1.9108, 3.8631, 3.9463, 3.9767, 3.5207], + device='cuda:3'), covar=tensor([0.0675, 0.0597, 0.0911, 0.4606, 0.0873, 0.0828, 0.1458, 0.0793], + device='cuda:3'), in_proj_covar=tensor([0.0413, 0.0321, 0.0344, 0.0427, 0.0333, 0.0315, 0.0320, 0.0271], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 03:27:32,148 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48945.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:27:34,591 INFO [train.py:901] (3/4) Epoch 7, batch 450, loss[loss=0.2272, simple_loss=0.2955, pruned_loss=0.0794, over 7919.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.336, pruned_loss=0.1006, over 1454575.77 frames. ], batch size: 20, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:27:48,253 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.9534, 6.0602, 5.1113, 2.3833, 5.2257, 5.5897, 5.4876, 5.1515], + device='cuda:3'), covar=tensor([0.0668, 0.0469, 0.1040, 0.5015, 0.0711, 0.0622, 0.1432, 0.0598], + device='cuda:3'), in_proj_covar=tensor([0.0416, 0.0321, 0.0344, 0.0428, 0.0334, 0.0316, 0.0320, 0.0272], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 03:27:49,656 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48970.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:28:10,030 INFO [train.py:901] (3/4) Epoch 7, batch 500, loss[loss=0.2997, simple_loss=0.3616, pruned_loss=0.1189, over 8452.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3341, pruned_loss=0.09991, over 1487477.70 frames. ], batch size: 27, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:28:32,337 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.604e+02 2.572e+02 3.184e+02 4.227e+02 8.649e+02, threshold=6.369e+02, percent-clipped=1.0 +2023-02-06 03:28:43,905 INFO [train.py:901] (3/4) Epoch 7, batch 550, loss[loss=0.309, simple_loss=0.3723, pruned_loss=0.1229, over 8492.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3347, pruned_loss=0.1004, over 1515404.97 frames. ], batch size: 26, lr: 1.13e-02, grad_scale: 16.0 +2023-02-06 03:28:50,297 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49057.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:29:05,360 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 +2023-02-06 03:29:06,570 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1256, 2.6689, 3.2338, 1.1123, 3.1161, 1.9946, 1.5581, 1.8359], + device='cuda:3'), covar=tensor([0.0376, 0.0152, 0.0111, 0.0373, 0.0258, 0.0438, 0.0403, 0.0236], + device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0256, 0.0212, 0.0315, 0.0245, 0.0403, 0.0322, 0.0298], + device='cuda:3'), out_proj_covar=tensor([1.1610e-04, 8.1755e-05, 6.6466e-05, 1.0013e-04, 7.8668e-05, 1.3977e-04, + 1.0509e-04, 9.5720e-05], device='cuda:3') +2023-02-06 03:29:07,260 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49082.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:29:19,500 INFO [train.py:901] (3/4) Epoch 7, batch 600, loss[loss=0.2754, simple_loss=0.3429, pruned_loss=0.104, over 8671.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3361, pruned_loss=0.1013, over 1543369.09 frames. ], batch size: 39, lr: 1.13e-02, grad_scale: 16.0 +2023-02-06 03:29:24,352 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3676, 1.3903, 4.4064, 2.0435, 2.3361, 5.0660, 4.9207, 4.4468], + device='cuda:3'), covar=tensor([0.0934, 0.1500, 0.0231, 0.1725, 0.0975, 0.0190, 0.0312, 0.0556], + device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0275, 0.0228, 0.0272, 0.0237, 0.0214, 0.0267, 0.0281], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 03:29:31,447 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 03:29:41,663 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49130.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:29:42,821 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.770e+02 2.845e+02 3.510e+02 4.694e+02 1.227e+03, threshold=7.020e+02, percent-clipped=5.0 +2023-02-06 03:29:54,465 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 +2023-02-06 03:29:54,594 INFO [train.py:901] (3/4) Epoch 7, batch 650, loss[loss=0.2442, simple_loss=0.3225, pruned_loss=0.08296, over 7919.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3346, pruned_loss=0.1006, over 1554542.59 frames. ], batch size: 20, lr: 1.13e-02, grad_scale: 16.0 +2023-02-06 03:29:58,972 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49155.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:29:59,709 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49156.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:30:13,069 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.68 vs. limit=5.0 +2023-02-06 03:30:17,667 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49181.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:30:18,988 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49183.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:30:29,822 INFO [train.py:901] (3/4) Epoch 7, batch 700, loss[loss=0.294, simple_loss=0.355, pruned_loss=0.1165, over 8657.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.335, pruned_loss=0.1006, over 1569490.78 frames. ], batch size: 39, lr: 1.13e-02, grad_scale: 16.0 +2023-02-06 03:30:31,258 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49200.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:30:43,158 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3414, 1.6549, 3.6677, 1.1202, 2.5342, 1.9617, 1.4702, 2.4777], + device='cuda:3'), covar=tensor([0.2069, 0.2807, 0.0664, 0.4271, 0.1380, 0.2778, 0.2025, 0.2069], + device='cuda:3'), in_proj_covar=tensor([0.0476, 0.0474, 0.0529, 0.0550, 0.0589, 0.0528, 0.0453, 0.0594], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 03:30:54,555 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.106e+02 2.876e+02 3.436e+02 4.276e+02 6.994e+02, threshold=6.873e+02, percent-clipped=0.0 +2023-02-06 03:31:06,146 INFO [train.py:901] (3/4) Epoch 7, batch 750, loss[loss=0.2294, simple_loss=0.289, pruned_loss=0.08486, over 7781.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3345, pruned_loss=0.1004, over 1583006.02 frames. ], batch size: 19, lr: 1.13e-02, grad_scale: 16.0 +2023-02-06 03:31:09,779 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49254.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:31:18,085 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 03:31:26,429 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 03:31:28,182 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-06 03:31:40,968 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49298.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:31:41,430 INFO [train.py:901] (3/4) Epoch 7, batch 800, loss[loss=0.2281, simple_loss=0.298, pruned_loss=0.07912, over 8078.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3347, pruned_loss=0.1006, over 1591806.50 frames. ], batch size: 21, lr: 1.13e-02, grad_scale: 16.0 +2023-02-06 03:31:53,367 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49315.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:32:05,543 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.006e+02 2.832e+02 3.318e+02 4.162e+02 1.224e+03, threshold=6.636e+02, percent-clipped=6.0 +2023-02-06 03:32:17,934 INFO [train.py:901] (3/4) Epoch 7, batch 850, loss[loss=0.253, simple_loss=0.3099, pruned_loss=0.09806, over 7712.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3346, pruned_loss=0.1008, over 1597249.64 frames. ], batch size: 18, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:32:22,243 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5338, 2.7230, 1.7834, 2.2654, 2.2788, 1.3102, 1.9205, 2.1005], + device='cuda:3'), covar=tensor([0.1120, 0.0266, 0.0831, 0.0452, 0.0579, 0.1268, 0.0881, 0.0737], + device='cuda:3'), in_proj_covar=tensor([0.0343, 0.0229, 0.0311, 0.0296, 0.0308, 0.0313, 0.0338, 0.0313], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 03:32:31,305 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7675, 2.0829, 2.3722, 1.7748, 1.0822, 2.4012, 0.4249, 1.3541], + device='cuda:3'), covar=tensor([0.3193, 0.2400, 0.0657, 0.2592, 0.6466, 0.0442, 0.5227, 0.2640], + device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0131, 0.0082, 0.0183, 0.0227, 0.0082, 0.0144, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 03:32:52,545 INFO [train.py:901] (3/4) Epoch 7, batch 900, loss[loss=0.2661, simple_loss=0.3447, pruned_loss=0.09369, over 8483.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3354, pruned_loss=0.1011, over 1605561.51 frames. ], batch size: 25, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:33:17,133 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.626e+02 2.812e+02 3.278e+02 4.578e+02 1.649e+03, threshold=6.556e+02, percent-clipped=8.0 +2023-02-06 03:33:22,001 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49440.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:33:28,079 INFO [train.py:901] (3/4) Epoch 7, batch 950, loss[loss=0.2787, simple_loss=0.3482, pruned_loss=0.1046, over 8462.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3348, pruned_loss=0.1004, over 1608240.38 frames. ], batch size: 29, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:33:28,880 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49450.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:33:35,919 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-06 03:33:50,518 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 03:34:03,658 INFO [train.py:901] (3/4) Epoch 7, batch 1000, loss[loss=0.2921, simple_loss=0.3574, pruned_loss=0.1134, over 8605.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3341, pruned_loss=0.09986, over 1608798.76 frames. ], batch size: 34, lr: 1.13e-02, grad_scale: 8.0 +2023-02-06 03:34:24,199 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 03:34:27,690 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.937e+02 3.091e+02 3.599e+02 4.515e+02 1.445e+03, threshold=7.198e+02, percent-clipped=7.0 +2023-02-06 03:34:35,923 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 03:34:38,730 INFO [train.py:901] (3/4) Epoch 7, batch 1050, loss[loss=0.2238, simple_loss=0.3, pruned_loss=0.07377, over 7931.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3345, pruned_loss=0.1004, over 1609268.77 frames. ], batch size: 20, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:34:43,053 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49554.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:34:54,544 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49571.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:35:00,751 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49579.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:35:13,244 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49596.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:35:14,558 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49598.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:35:14,686 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6081, 1.9065, 2.0998, 1.5999, 0.9636, 2.1710, 0.3778, 1.0729], + device='cuda:3'), covar=tensor([0.4010, 0.2179, 0.1200, 0.3153, 0.7070, 0.0704, 0.5242, 0.3463], + device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0132, 0.0083, 0.0183, 0.0227, 0.0084, 0.0143, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 03:35:15,138 INFO [train.py:901] (3/4) Epoch 7, batch 1100, loss[loss=0.286, simple_loss=0.35, pruned_loss=0.111, over 8460.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3338, pruned_loss=0.09942, over 1614287.47 frames. ], batch size: 25, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:35:38,342 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.978e+02 2.766e+02 3.386e+02 4.310e+02 6.415e+02, threshold=6.771e+02, percent-clipped=0.0 +2023-02-06 03:35:45,264 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.30 vs. limit=5.0 +2023-02-06 03:35:46,076 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 03:35:49,188 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 +2023-02-06 03:35:49,445 INFO [train.py:901] (3/4) Epoch 7, batch 1150, loss[loss=0.2264, simple_loss=0.2941, pruned_loss=0.07932, over 8234.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3347, pruned_loss=0.09971, over 1616386.90 frames. ], batch size: 22, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:36:23,365 INFO [train.py:901] (3/4) Epoch 7, batch 1200, loss[loss=0.2475, simple_loss=0.3256, pruned_loss=0.08473, over 8137.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3351, pruned_loss=0.09995, over 1612538.35 frames. ], batch size: 22, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:36:33,695 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49713.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:36:47,167 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.006e+02 2.915e+02 3.820e+02 5.048e+02 1.193e+03, threshold=7.640e+02, percent-clipped=11.0 +2023-02-06 03:36:57,980 INFO [train.py:901] (3/4) Epoch 7, batch 1250, loss[loss=0.2535, simple_loss=0.331, pruned_loss=0.08803, over 7976.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3337, pruned_loss=0.09862, over 1616947.00 frames. ], batch size: 21, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:37:22,190 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49784.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:37:22,260 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49784.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:37:29,679 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49794.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:37:32,811 INFO [train.py:901] (3/4) Epoch 7, batch 1300, loss[loss=0.4713, simple_loss=0.4731, pruned_loss=0.2348, over 7041.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3349, pruned_loss=0.09934, over 1615180.59 frames. ], batch size: 72, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:37:57,663 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.968e+02 2.650e+02 3.390e+02 4.402e+02 9.600e+02, threshold=6.781e+02, percent-clipped=3.0 +2023-02-06 03:38:08,116 INFO [train.py:901] (3/4) Epoch 7, batch 1350, loss[loss=0.2521, simple_loss=0.3271, pruned_loss=0.08851, over 7965.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3359, pruned_loss=0.1006, over 1612444.35 frames. ], batch size: 21, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:38:38,485 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0881, 1.2966, 1.1400, 0.4128, 1.2205, 0.9146, 0.0819, 1.1563], + device='cuda:3'), covar=tensor([0.0181, 0.0141, 0.0130, 0.0231, 0.0153, 0.0443, 0.0340, 0.0135], + device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0263, 0.0216, 0.0309, 0.0251, 0.0404, 0.0319, 0.0297], + device='cuda:3'), out_proj_covar=tensor([1.1481e-04, 8.3460e-05, 6.7401e-05, 9.7767e-05, 8.0563e-05, 1.3962e-04, + 1.0383e-04, 9.4980e-05], device='cuda:3') +2023-02-06 03:38:42,146 INFO [train.py:901] (3/4) Epoch 7, batch 1400, loss[loss=0.2311, simple_loss=0.2916, pruned_loss=0.08533, over 7802.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3357, pruned_loss=0.1005, over 1616769.77 frames. ], batch size: 20, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:38:43,022 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49899.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:38:49,798 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49909.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:39:07,158 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.950e+02 2.921e+02 3.790e+02 4.996e+02 8.997e+02, threshold=7.579e+02, percent-clipped=6.0 +2023-02-06 03:39:11,351 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 03:39:18,097 INFO [train.py:901] (3/4) Epoch 7, batch 1450, loss[loss=0.2857, simple_loss=0.3575, pruned_loss=0.107, over 8460.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3364, pruned_loss=0.1009, over 1618318.35 frames. ], batch size: 25, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:39:31,697 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49969.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:39:49,094 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49994.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:39:52,397 INFO [train.py:901] (3/4) Epoch 7, batch 1500, loss[loss=0.2611, simple_loss=0.3285, pruned_loss=0.09679, over 7979.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3348, pruned_loss=0.1, over 1613851.70 frames. ], batch size: 21, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:40:16,582 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.509e+02 2.823e+02 3.555e+02 4.038e+02 9.229e+02, threshold=7.110e+02, percent-clipped=3.0 +2023-02-06 03:40:27,896 INFO [train.py:901] (3/4) Epoch 7, batch 1550, loss[loss=0.3849, simple_loss=0.4369, pruned_loss=0.1664, over 8507.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3374, pruned_loss=0.1013, over 1620684.62 frames. ], batch size: 26, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:41:02,320 INFO [train.py:901] (3/4) Epoch 7, batch 1600, loss[loss=0.2978, simple_loss=0.3612, pruned_loss=0.1172, over 8028.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3379, pruned_loss=0.1014, over 1623134.60 frames. ], batch size: 22, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:41:12,237 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.07 vs. limit=5.0 +2023-02-06 03:41:22,667 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50128.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:41:25,920 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.888e+02 2.849e+02 3.464e+02 4.418e+02 7.019e+02, threshold=6.928e+02, percent-clipped=0.0 +2023-02-06 03:41:35,383 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50146.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:41:37,167 INFO [train.py:901] (3/4) Epoch 7, batch 1650, loss[loss=0.2542, simple_loss=0.3203, pruned_loss=0.09399, over 7916.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3364, pruned_loss=0.1008, over 1619565.44 frames. ], batch size: 20, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:41:41,456 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50155.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:41:48,659 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50165.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:41:59,250 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50180.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:42:05,843 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50190.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:42:11,959 INFO [train.py:901] (3/4) Epoch 7, batch 1700, loss[loss=0.3836, simple_loss=0.4104, pruned_loss=0.1784, over 6483.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3365, pruned_loss=0.1004, over 1620548.37 frames. ], batch size: 71, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:42:22,989 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50215.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:42:35,242 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.948e+02 2.886e+02 3.481e+02 4.608e+02 1.233e+03, threshold=6.962e+02, percent-clipped=3.0 +2023-02-06 03:42:42,122 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50243.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:42:46,680 INFO [train.py:901] (3/4) Epoch 7, batch 1750, loss[loss=0.3067, simple_loss=0.3725, pruned_loss=0.1204, over 8440.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3359, pruned_loss=0.1, over 1625935.84 frames. ], batch size: 49, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:43:21,543 INFO [train.py:901] (3/4) Epoch 7, batch 1800, loss[loss=0.2887, simple_loss=0.3457, pruned_loss=0.1158, over 7816.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3359, pruned_loss=0.1002, over 1624765.64 frames. ], batch size: 20, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:43:44,800 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.963e+02 2.922e+02 3.562e+02 4.379e+02 1.030e+03, threshold=7.125e+02, percent-clipped=4.0 +2023-02-06 03:43:56,170 INFO [train.py:901] (3/4) Epoch 7, batch 1850, loss[loss=0.239, simple_loss=0.2976, pruned_loss=0.09015, over 6285.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3353, pruned_loss=0.09917, over 1629049.37 frames. ], batch size: 14, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:44:30,587 INFO [train.py:901] (3/4) Epoch 7, batch 1900, loss[loss=0.2807, simple_loss=0.3432, pruned_loss=0.1091, over 8240.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3357, pruned_loss=0.09993, over 1624823.03 frames. ], batch size: 22, lr: 1.12e-02, grad_scale: 8.0 +2023-02-06 03:44:42,162 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.0714, 1.1565, 4.2180, 1.5536, 3.6019, 3.3865, 3.7043, 3.5472], + device='cuda:3'), covar=tensor([0.0465, 0.4225, 0.0483, 0.2895, 0.1210, 0.0807, 0.0526, 0.0684], + device='cuda:3'), in_proj_covar=tensor([0.0371, 0.0512, 0.0464, 0.0455, 0.0516, 0.0428, 0.0432, 0.0490], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 03:44:43,978 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 03:44:53,922 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.927e+02 2.786e+02 3.637e+02 4.614e+02 8.948e+02, threshold=7.273e+02, percent-clipped=3.0 +2023-02-06 03:44:56,014 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 03:45:04,598 INFO [train.py:901] (3/4) Epoch 7, batch 1950, loss[loss=0.2402, simple_loss=0.3258, pruned_loss=0.07729, over 8470.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3352, pruned_loss=0.1001, over 1619348.68 frames. ], batch size: 29, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:45:05,201 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 +2023-02-06 03:45:15,300 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 03:45:19,139 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.31 vs. limit=5.0 +2023-02-06 03:45:33,289 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50490.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:45:39,205 INFO [train.py:901] (3/4) Epoch 7, batch 2000, loss[loss=0.2472, simple_loss=0.3096, pruned_loss=0.09243, over 7691.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3353, pruned_loss=0.1003, over 1618870.49 frames. ], batch size: 18, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:45:39,435 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50499.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:45:40,222 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 +2023-02-06 03:45:57,458 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50524.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:45:57,476 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1114, 2.6066, 3.1934, 1.2052, 3.3005, 2.0956, 1.4738, 1.8942], + device='cuda:3'), covar=tensor([0.0359, 0.0150, 0.0093, 0.0327, 0.0131, 0.0361, 0.0446, 0.0228], + device='cuda:3'), in_proj_covar=tensor([0.0344, 0.0254, 0.0212, 0.0307, 0.0248, 0.0395, 0.0314, 0.0286], + device='cuda:3'), out_proj_covar=tensor([1.1192e-04, 7.9927e-05, 6.5928e-05, 9.6853e-05, 7.8975e-05, 1.3617e-04, + 1.0205e-04, 9.0992e-05], device='cuda:3') +2023-02-06 03:46:03,227 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.870e+02 2.800e+02 3.583e+02 4.591e+02 1.075e+03, threshold=7.166e+02, percent-clipped=7.0 +2023-02-06 03:46:13,949 INFO [train.py:901] (3/4) Epoch 7, batch 2050, loss[loss=0.2574, simple_loss=0.3264, pruned_loss=0.0942, over 8291.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3358, pruned_loss=0.1008, over 1618417.60 frames. ], batch size: 23, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:46:20,540 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50559.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:46:23,784 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50564.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:46:47,297 INFO [train.py:901] (3/4) Epoch 7, batch 2100, loss[loss=0.276, simple_loss=0.3482, pruned_loss=0.102, over 8568.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3352, pruned_loss=0.1001, over 1617216.75 frames. ], batch size: 31, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:46:52,372 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50605.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:47:11,026 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.849e+02 3.066e+02 3.697e+02 4.610e+02 1.063e+03, threshold=7.394e+02, percent-clipped=3.0 +2023-02-06 03:47:22,370 INFO [train.py:901] (3/4) Epoch 7, batch 2150, loss[loss=0.2313, simple_loss=0.2874, pruned_loss=0.0876, over 7802.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3351, pruned_loss=0.1006, over 1616359.28 frames. ], batch size: 19, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:47:40,258 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50674.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:47:52,177 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2618, 1.2564, 1.3705, 1.1128, 1.2359, 1.1632, 1.7553, 1.7095], + device='cuda:3'), covar=tensor([0.0607, 0.1819, 0.2731, 0.1881, 0.0740, 0.2290, 0.0884, 0.0702], + device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0167, 0.0208, 0.0170, 0.0117, 0.0176, 0.0129, 0.0146], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 03:47:56,609 INFO [train.py:901] (3/4) Epoch 7, batch 2200, loss[loss=0.2254, simple_loss=0.3013, pruned_loss=0.07475, over 7655.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3353, pruned_loss=0.1006, over 1616578.18 frames. ], batch size: 19, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:48:20,832 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.790e+02 2.939e+02 3.492e+02 4.230e+02 8.261e+02, threshold=6.983e+02, percent-clipped=2.0 +2023-02-06 03:48:31,240 INFO [train.py:901] (3/4) Epoch 7, batch 2250, loss[loss=0.2691, simple_loss=0.3454, pruned_loss=0.0964, over 8658.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3349, pruned_loss=0.09982, over 1622420.23 frames. ], batch size: 34, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:49:05,387 INFO [train.py:901] (3/4) Epoch 7, batch 2300, loss[loss=0.2377, simple_loss=0.3057, pruned_loss=0.08488, over 7809.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3363, pruned_loss=0.1015, over 1621418.98 frames. ], batch size: 20, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:49:06,290 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50800.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:49:23,415 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50826.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:49:28,610 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.693e+02 3.140e+02 4.073e+02 5.620e+02 1.608e+03, threshold=8.146e+02, percent-clipped=16.0 +2023-02-06 03:49:39,820 INFO [train.py:901] (3/4) Epoch 7, batch 2350, loss[loss=0.2824, simple_loss=0.3582, pruned_loss=0.1033, over 8581.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3361, pruned_loss=0.1013, over 1617515.52 frames. ], batch size: 49, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:49:47,971 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50861.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:50:05,013 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50886.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:50:14,212 INFO [train.py:901] (3/4) Epoch 7, batch 2400, loss[loss=0.2658, simple_loss=0.3329, pruned_loss=0.0994, over 7807.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3358, pruned_loss=0.101, over 1616746.71 frames. ], batch size: 20, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:50:20,197 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50908.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:50:29,050 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6709, 1.7533, 1.9537, 1.7434, 1.0833, 1.9627, 0.3400, 1.3956], + device='cuda:3'), covar=tensor([0.3757, 0.2063, 0.0806, 0.1812, 0.5525, 0.0692, 0.4257, 0.2090], + device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0136, 0.0083, 0.0186, 0.0230, 0.0085, 0.0145, 0.0137], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 03:50:35,194 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50930.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:50:36,985 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.859e+02 2.900e+02 3.414e+02 4.196e+02 7.276e+02, threshold=6.828e+02, percent-clipped=0.0 +2023-02-06 03:50:47,544 INFO [train.py:901] (3/4) Epoch 7, batch 2450, loss[loss=0.2452, simple_loss=0.3128, pruned_loss=0.08879, over 8079.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3342, pruned_loss=0.1002, over 1617265.49 frames. ], batch size: 21, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:50:48,394 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.7328, 1.3783, 3.9092, 1.2849, 3.3845, 3.2792, 3.5562, 3.3886], + device='cuda:3'), covar=tensor([0.0524, 0.3706, 0.0496, 0.3005, 0.1196, 0.0779, 0.0554, 0.0711], + device='cuda:3'), in_proj_covar=tensor([0.0375, 0.0523, 0.0475, 0.0455, 0.0523, 0.0434, 0.0437, 0.0497], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 03:50:52,348 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50955.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:50:54,522 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 +2023-02-06 03:50:59,079 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50964.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:51:22,801 INFO [train.py:901] (3/4) Epoch 7, batch 2500, loss[loss=0.2694, simple_loss=0.3463, pruned_loss=0.09622, over 8465.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3351, pruned_loss=0.1005, over 1622062.83 frames. ], batch size: 27, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:51:39,950 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51023.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:51:46,386 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.883e+02 2.634e+02 3.421e+02 4.023e+02 8.503e+02, threshold=6.842e+02, percent-clipped=1.0 +2023-02-06 03:51:56,890 INFO [train.py:901] (3/4) Epoch 7, batch 2550, loss[loss=0.2615, simple_loss=0.3316, pruned_loss=0.09568, over 8468.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3337, pruned_loss=0.09984, over 1619882.37 frames. ], batch size: 25, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:52:18,142 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51080.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:52:31,081 INFO [train.py:901] (3/4) Epoch 7, batch 2600, loss[loss=0.295, simple_loss=0.3701, pruned_loss=0.11, over 8445.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3337, pruned_loss=0.1003, over 1616335.06 frames. ], batch size: 27, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:52:54,869 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.187e+02 3.140e+02 3.874e+02 4.757e+02 8.436e+02, threshold=7.747e+02, percent-clipped=5.0 +2023-02-06 03:53:02,028 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51144.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:53:05,173 INFO [train.py:901] (3/4) Epoch 7, batch 2650, loss[loss=0.3399, simple_loss=0.3837, pruned_loss=0.1481, over 8576.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3334, pruned_loss=0.1005, over 1614220.22 frames. ], batch size: 39, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:53:19,253 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51170.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:53:32,643 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51190.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 03:53:39,221 INFO [train.py:901] (3/4) Epoch 7, batch 2700, loss[loss=0.2476, simple_loss=0.3223, pruned_loss=0.08647, over 8500.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3322, pruned_loss=0.09952, over 1616749.14 frames. ], batch size: 26, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:53:46,473 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 +2023-02-06 03:54:02,460 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.830e+02 3.046e+02 3.584e+02 4.560e+02 9.753e+02, threshold=7.169e+02, percent-clipped=4.0 +2023-02-06 03:54:14,236 INFO [train.py:901] (3/4) Epoch 7, batch 2750, loss[loss=0.2887, simple_loss=0.368, pruned_loss=0.1047, over 8321.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3331, pruned_loss=0.09946, over 1619229.08 frames. ], batch size: 25, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:54:14,458 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7294, 2.9969, 1.9526, 2.4843, 2.4678, 1.6230, 2.1668, 2.4132], + device='cuda:3'), covar=tensor([0.1061, 0.0229, 0.0822, 0.0458, 0.0527, 0.1002, 0.0700, 0.0714], + device='cuda:3'), in_proj_covar=tensor([0.0341, 0.0229, 0.0310, 0.0294, 0.0305, 0.0310, 0.0335, 0.0313], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 03:54:21,079 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51259.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:54:34,381 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51279.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:54:38,286 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51285.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:54:40,393 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7964, 2.2104, 3.9188, 2.8540, 3.1897, 2.2539, 1.7048, 1.8630], + device='cuda:3'), covar=tensor([0.2250, 0.2881, 0.0576, 0.1645, 0.1313, 0.1365, 0.1281, 0.2962], + device='cuda:3'), in_proj_covar=tensor([0.0826, 0.0769, 0.0667, 0.0765, 0.0857, 0.0708, 0.0666, 0.0703], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 03:54:47,540 INFO [train.py:901] (3/4) Epoch 7, batch 2800, loss[loss=0.2118, simple_loss=0.2891, pruned_loss=0.06724, over 7220.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.333, pruned_loss=0.09909, over 1623062.70 frames. ], batch size: 16, lr: 1.11e-02, grad_scale: 8.0 +2023-02-06 03:54:51,153 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51304.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:54:53,741 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51308.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:55:01,962 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9754, 1.5961, 1.5521, 1.4750, 1.1488, 1.4098, 1.6768, 1.4743], + device='cuda:3'), covar=tensor([0.0611, 0.1206, 0.1757, 0.1312, 0.0575, 0.1452, 0.0668, 0.0586], + device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0167, 0.0208, 0.0171, 0.0117, 0.0176, 0.0129, 0.0148], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 03:55:11,783 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.791e+02 2.811e+02 3.563e+02 4.674e+02 6.809e+02, threshold=7.126e+02, percent-clipped=0.0 +2023-02-06 03:55:22,692 INFO [train.py:901] (3/4) Epoch 7, batch 2850, loss[loss=0.2689, simple_loss=0.3399, pruned_loss=0.09895, over 8132.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3351, pruned_loss=0.09998, over 1625330.45 frames. ], batch size: 22, lr: 1.10e-02, grad_scale: 16.0 +2023-02-06 03:55:57,259 INFO [train.py:901] (3/4) Epoch 7, batch 2900, loss[loss=0.25, simple_loss=0.3223, pruned_loss=0.08887, over 8335.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3348, pruned_loss=0.09995, over 1620775.11 frames. ], batch size: 26, lr: 1.10e-02, grad_scale: 16.0 +2023-02-06 03:55:57,412 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.8666, 1.4199, 5.8675, 1.9331, 5.2576, 4.9742, 5.5580, 5.4512], + device='cuda:3'), covar=tensor([0.0341, 0.3773, 0.0233, 0.2713, 0.0892, 0.0597, 0.0314, 0.0358], + device='cuda:3'), in_proj_covar=tensor([0.0377, 0.0521, 0.0468, 0.0462, 0.0524, 0.0436, 0.0430, 0.0489], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 03:55:58,200 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51400.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:55:58,867 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.5625, 5.5910, 4.8678, 2.2238, 4.8699, 5.0992, 5.1965, 4.7966], + device='cuda:3'), covar=tensor([0.0672, 0.0371, 0.0765, 0.4383, 0.0646, 0.0588, 0.0933, 0.0579], + device='cuda:3'), in_proj_covar=tensor([0.0412, 0.0310, 0.0336, 0.0419, 0.0328, 0.0306, 0.0315, 0.0272], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 03:56:04,944 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51410.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:56:13,557 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51423.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:56:14,147 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51424.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:56:19,630 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 03:56:20,272 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.021e+02 2.833e+02 3.577e+02 4.732e+02 1.075e+03, threshold=7.153e+02, percent-clipped=9.0 +2023-02-06 03:56:28,586 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-06 03:56:32,325 INFO [train.py:901] (3/4) Epoch 7, batch 2950, loss[loss=0.2448, simple_loss=0.3107, pruned_loss=0.08946, over 6449.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.334, pruned_loss=0.09873, over 1622551.41 frames. ], batch size: 14, lr: 1.10e-02, grad_scale: 16.0 +2023-02-06 03:57:06,412 INFO [train.py:901] (3/4) Epoch 7, batch 3000, loss[loss=0.2269, simple_loss=0.2863, pruned_loss=0.0837, over 7199.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3343, pruned_loss=0.09928, over 1620999.45 frames. ], batch size: 16, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 03:57:06,412 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 03:57:21,703 INFO [train.py:935] (3/4) Epoch 7, validation: loss=0.2071, simple_loss=0.305, pruned_loss=0.05459, over 944034.00 frames. +2023-02-06 03:57:21,703 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 03:57:31,201 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51513.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:57:32,578 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51515.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:57:45,150 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.886e+02 2.842e+02 3.422e+02 4.197e+02 1.269e+03, threshold=6.844e+02, percent-clipped=2.0 +2023-02-06 03:57:45,245 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51534.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:57:48,659 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51539.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:57:49,365 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51540.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:57:50,090 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51541.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:57:55,339 INFO [train.py:901] (3/4) Epoch 7, batch 3050, loss[loss=0.3549, simple_loss=0.3944, pruned_loss=0.1578, over 8455.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3358, pruned_loss=0.1005, over 1621650.50 frames. ], batch size: 27, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 03:57:59,273 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-06 03:58:06,936 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51566.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:58:25,337 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.0105, 3.8732, 3.5901, 1.9517, 3.4255, 3.4461, 3.6935, 3.1891], + device='cuda:3'), covar=tensor([0.0873, 0.0742, 0.0919, 0.4266, 0.0900, 0.1081, 0.1276, 0.0993], + device='cuda:3'), in_proj_covar=tensor([0.0411, 0.0315, 0.0342, 0.0424, 0.0332, 0.0307, 0.0322, 0.0275], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 03:58:29,888 INFO [train.py:901] (3/4) Epoch 7, batch 3100, loss[loss=0.2728, simple_loss=0.3591, pruned_loss=0.09326, over 8203.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3369, pruned_loss=0.1014, over 1619188.01 frames. ], batch size: 23, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 03:58:38,418 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-06 03:58:54,843 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.934e+02 3.035e+02 3.902e+02 5.145e+02 1.067e+03, threshold=7.804e+02, percent-clipped=7.0 +2023-02-06 03:59:05,326 INFO [train.py:901] (3/4) Epoch 7, batch 3150, loss[loss=0.254, simple_loss=0.3167, pruned_loss=0.09565, over 7434.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3363, pruned_loss=0.1014, over 1616057.09 frames. ], batch size: 17, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 03:59:05,504 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51649.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 03:59:10,954 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5940, 1.7633, 1.9355, 1.5648, 1.1860, 2.0222, 0.2530, 1.2232], + device='cuda:3'), covar=tensor([0.3444, 0.2018, 0.0836, 0.2322, 0.5255, 0.0765, 0.4309, 0.2389], + device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0140, 0.0082, 0.0186, 0.0227, 0.0088, 0.0142, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 03:59:17,824 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5761, 4.5651, 4.0095, 1.7447, 3.9765, 4.1667, 4.2018, 3.5526], + device='cuda:3'), covar=tensor([0.0767, 0.0529, 0.0889, 0.5055, 0.0782, 0.0789, 0.1146, 0.0987], + device='cuda:3'), in_proj_covar=tensor([0.0419, 0.0320, 0.0346, 0.0428, 0.0333, 0.0312, 0.0324, 0.0281], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 03:59:26,403 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51679.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 03:59:40,111 INFO [train.py:901] (3/4) Epoch 7, batch 3200, loss[loss=0.2172, simple_loss=0.2766, pruned_loss=0.07888, over 7710.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3366, pruned_loss=0.1015, over 1618105.70 frames. ], batch size: 18, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 03:59:43,623 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51704.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:00:05,272 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.836e+02 2.946e+02 3.588e+02 4.680e+02 7.788e+02, threshold=7.176e+02, percent-clipped=0.0 +2023-02-06 04:00:12,049 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51744.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:00:16,054 INFO [train.py:901] (3/4) Epoch 7, batch 3250, loss[loss=0.2291, simple_loss=0.306, pruned_loss=0.07606, over 7812.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3357, pruned_loss=0.101, over 1614336.68 frames. ], batch size: 20, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:00:19,466 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51754.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:00:23,712 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-06 04:00:26,786 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51765.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:00:46,821 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51795.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:00:49,340 INFO [train.py:901] (3/4) Epoch 7, batch 3300, loss[loss=0.2206, simple_loss=0.2971, pruned_loss=0.07207, over 8099.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3351, pruned_loss=0.1007, over 1613665.39 frames. ], batch size: 21, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:01:03,668 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51820.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:01:14,384 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.688e+02 2.988e+02 3.662e+02 4.246e+02 9.313e+02, threshold=7.324e+02, percent-clipped=2.0 +2023-02-06 04:01:24,626 INFO [train.py:901] (3/4) Epoch 7, batch 3350, loss[loss=0.2297, simple_loss=0.3119, pruned_loss=0.07379, over 8029.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3354, pruned_loss=0.1003, over 1616770.05 frames. ], batch size: 22, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:01:30,919 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51857.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:01:32,354 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51859.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:01:39,546 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51869.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:01:59,346 INFO [train.py:901] (3/4) Epoch 7, batch 3400, loss[loss=0.2923, simple_loss=0.3602, pruned_loss=0.1121, over 8479.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3342, pruned_loss=0.09946, over 1618834.26 frames. ], batch size: 29, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:02:03,619 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51905.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 04:02:20,935 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51930.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 04:02:23,261 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.857e+02 2.723e+02 3.470e+02 4.144e+02 7.359e+02, threshold=6.940e+02, percent-clipped=1.0 +2023-02-06 04:02:24,191 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4857, 2.2351, 4.2690, 1.1705, 2.7690, 1.9734, 1.6491, 2.4628], + device='cuda:3'), covar=tensor([0.1876, 0.2208, 0.0686, 0.3931, 0.1686, 0.2781, 0.1810, 0.2421], + device='cuda:3'), in_proj_covar=tensor([0.0475, 0.0470, 0.0534, 0.0547, 0.0588, 0.0524, 0.0452, 0.0590], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 04:02:34,623 INFO [train.py:901] (3/4) Epoch 7, batch 3450, loss[loss=0.2046, simple_loss=0.2913, pruned_loss=0.05895, over 8185.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3335, pruned_loss=0.09897, over 1620251.56 frames. ], batch size: 23, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:02:50,954 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51972.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:03:08,878 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0981, 1.5255, 1.4833, 1.4065, 1.0826, 1.3198, 1.5845, 1.5309], + device='cuda:3'), covar=tensor([0.0499, 0.1172, 0.1809, 0.1338, 0.0585, 0.1497, 0.0720, 0.0584], + device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0165, 0.0208, 0.0171, 0.0116, 0.0174, 0.0127, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0007, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 04:03:09,379 INFO [train.py:901] (3/4) Epoch 7, batch 3500, loss[loss=0.2392, simple_loss=0.3218, pruned_loss=0.07824, over 8120.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3327, pruned_loss=0.09889, over 1616159.65 frames. ], batch size: 22, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:03:22,431 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 04:03:33,496 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.660e+02 2.822e+02 3.302e+02 4.435e+02 1.594e+03, threshold=6.604e+02, percent-clipped=5.0 +2023-02-06 04:03:43,727 INFO [train.py:901] (3/4) Epoch 7, batch 3550, loss[loss=0.2891, simple_loss=0.3502, pruned_loss=0.114, over 8238.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3329, pruned_loss=0.09878, over 1615476.43 frames. ], batch size: 24, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:03:50,008 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52058.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:03:52,702 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1671, 1.8406, 2.5990, 2.1150, 2.1294, 1.9422, 1.4799, 0.9394], + device='cuda:3'), covar=tensor([0.2704, 0.2559, 0.0694, 0.1450, 0.1283, 0.1472, 0.1520, 0.2657], + device='cuda:3'), in_proj_covar=tensor([0.0827, 0.0763, 0.0660, 0.0758, 0.0857, 0.0710, 0.0656, 0.0694], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 04:04:10,082 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3973, 1.3135, 4.6567, 1.6473, 3.9147, 3.9032, 4.1844, 4.0254], + device='cuda:3'), covar=tensor([0.0540, 0.3891, 0.0456, 0.2912, 0.1323, 0.0708, 0.0454, 0.0567], + device='cuda:3'), in_proj_covar=tensor([0.0375, 0.0517, 0.0467, 0.0459, 0.0519, 0.0429, 0.0426, 0.0484], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], + device='cuda:3') +2023-02-06 04:04:19,844 INFO [train.py:901] (3/4) Epoch 7, batch 3600, loss[loss=0.25, simple_loss=0.3192, pruned_loss=0.09041, over 8130.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3329, pruned_loss=0.09872, over 1615728.97 frames. ], batch size: 22, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:04:26,703 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52109.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:04:30,881 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52115.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:04:37,710 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52125.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:04:43,463 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.793e+02 2.818e+02 3.176e+02 4.094e+02 8.086e+02, threshold=6.353e+02, percent-clipped=5.0 +2023-02-06 04:04:47,728 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52140.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:04:53,394 INFO [train.py:901] (3/4) Epoch 7, batch 3650, loss[loss=0.2273, simple_loss=0.2964, pruned_loss=0.07907, over 7452.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3331, pruned_loss=0.09874, over 1615416.67 frames. ], batch size: 17, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:04:54,178 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52150.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:05:23,190 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 04:05:28,583 INFO [train.py:901] (3/4) Epoch 7, batch 3700, loss[loss=0.2405, simple_loss=0.3237, pruned_loss=0.07867, over 8362.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3319, pruned_loss=0.09839, over 1610948.81 frames. ], batch size: 24, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:05:36,877 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52211.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:05:47,074 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52224.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:05:49,855 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52228.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:05:53,719 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.627e+02 2.602e+02 3.554e+02 4.404e+02 9.700e+02, threshold=7.108e+02, percent-clipped=5.0 +2023-02-06 04:06:04,132 INFO [train.py:901] (3/4) Epoch 7, batch 3750, loss[loss=0.2711, simple_loss=0.3516, pruned_loss=0.09534, over 8357.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3304, pruned_loss=0.09713, over 1611304.74 frames. ], batch size: 24, lr: 1.10e-02, grad_scale: 8.0 +2023-02-06 04:06:07,177 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52253.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:06:38,744 INFO [train.py:901] (3/4) Epoch 7, batch 3800, loss[loss=0.2518, simple_loss=0.3276, pruned_loss=0.08802, over 8249.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3316, pruned_loss=0.09765, over 1615276.40 frames. ], batch size: 22, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:07:01,891 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52330.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:07:04,421 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.417e+02 2.780e+02 3.361e+02 4.228e+02 6.516e+02, threshold=6.722e+02, percent-clipped=0.0 +2023-02-06 04:07:15,844 INFO [train.py:901] (3/4) Epoch 7, batch 3850, loss[loss=0.2609, simple_loss=0.3458, pruned_loss=0.08796, over 8328.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3315, pruned_loss=0.09758, over 1610652.97 frames. ], batch size: 25, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:07:18,009 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.67 vs. limit=5.0 +2023-02-06 04:07:30,481 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 04:07:49,736 INFO [train.py:901] (3/4) Epoch 7, batch 3900, loss[loss=0.2498, simple_loss=0.3104, pruned_loss=0.09462, over 7266.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3324, pruned_loss=0.09786, over 1611338.37 frames. ], batch size: 16, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:07:51,995 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52402.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:08:15,069 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.442e+02 2.697e+02 3.207e+02 4.225e+02 1.297e+03, threshold=6.414e+02, percent-clipped=5.0 +2023-02-06 04:08:25,215 INFO [train.py:901] (3/4) Epoch 7, batch 3950, loss[loss=0.2234, simple_loss=0.2972, pruned_loss=0.07479, over 7800.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3336, pruned_loss=0.09855, over 1617513.04 frames. ], batch size: 20, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:08:48,002 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52480.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:09:00,797 INFO [train.py:901] (3/4) Epoch 7, batch 4000, loss[loss=0.3037, simple_loss=0.3441, pruned_loss=0.1316, over 7692.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3328, pruned_loss=0.09815, over 1614858.26 frames. ], batch size: 18, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:09:05,166 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52505.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:09:13,183 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52517.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:09:24,186 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.802e+02 2.935e+02 3.629e+02 4.693e+02 1.248e+03, threshold=7.258e+02, percent-clipped=9.0 +2023-02-06 04:09:24,393 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2807, 2.1754, 1.5036, 1.9823, 1.7704, 1.1895, 1.5678, 1.8437], + device='cuda:3'), covar=tensor([0.1125, 0.0343, 0.0992, 0.0460, 0.0612, 0.1251, 0.0781, 0.0669], + device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0239, 0.0313, 0.0304, 0.0315, 0.0321, 0.0340, 0.0323], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 04:09:35,730 INFO [train.py:901] (3/4) Epoch 7, batch 4050, loss[loss=0.2751, simple_loss=0.353, pruned_loss=0.09861, over 8197.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3341, pruned_loss=0.09904, over 1620422.07 frames. ], batch size: 23, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:09:39,788 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52555.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:09:53,155 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52573.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 04:10:10,473 INFO [train.py:901] (3/4) Epoch 7, batch 4100, loss[loss=0.2361, simple_loss=0.3044, pruned_loss=0.08385, over 7932.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3337, pruned_loss=0.09915, over 1616538.80 frames. ], batch size: 20, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:10:33,850 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.429e+02 2.709e+02 3.346e+02 4.687e+02 1.096e+03, threshold=6.691e+02, percent-clipped=5.0 +2023-02-06 04:10:44,015 INFO [train.py:901] (3/4) Epoch 7, batch 4150, loss[loss=0.2532, simple_loss=0.3314, pruned_loss=0.08752, over 8450.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3323, pruned_loss=0.09864, over 1615827.44 frames. ], batch size: 27, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:10:59,828 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52670.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:11:02,381 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52674.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:11:20,449 INFO [train.py:901] (3/4) Epoch 7, batch 4200, loss[loss=0.4247, simple_loss=0.4566, pruned_loss=0.1964, over 8561.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3321, pruned_loss=0.09845, over 1614584.70 frames. ], batch size: 39, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:11:30,521 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 04:11:43,884 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.677e+02 2.675e+02 3.334e+02 4.108e+02 1.082e+03, threshold=6.669e+02, percent-clipped=4.0 +2023-02-06 04:11:53,178 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 04:11:53,853 INFO [train.py:901] (3/4) Epoch 7, batch 4250, loss[loss=0.2985, simple_loss=0.3575, pruned_loss=0.1197, over 8571.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3319, pruned_loss=0.09881, over 1611350.86 frames. ], batch size: 34, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:12:08,141 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52770.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:12:10,297 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52773.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:12:22,429 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52789.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:12:28,468 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52798.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:12:28,944 INFO [train.py:901] (3/4) Epoch 7, batch 4300, loss[loss=0.2696, simple_loss=0.3397, pruned_loss=0.09975, over 8332.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3303, pruned_loss=0.09783, over 1609911.40 frames. ], batch size: 25, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:12:53,669 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.536e+02 2.879e+02 3.462e+02 4.347e+02 1.112e+03, threshold=6.924e+02, percent-clipped=5.0 +2023-02-06 04:13:03,882 INFO [train.py:901] (3/4) Epoch 7, batch 4350, loss[loss=0.2407, simple_loss=0.3287, pruned_loss=0.07636, over 8640.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3294, pruned_loss=0.09737, over 1612855.92 frames. ], batch size: 49, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:13:24,437 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 04:13:38,504 INFO [train.py:901] (3/4) Epoch 7, batch 4400, loss[loss=0.264, simple_loss=0.3258, pruned_loss=0.101, over 7967.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3311, pruned_loss=0.09898, over 1608066.06 frames. ], batch size: 21, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:13:50,776 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52917.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 04:13:56,765 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52926.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:14:02,503 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.837e+02 2.715e+02 3.689e+02 4.508e+02 8.331e+02, threshold=7.379e+02, percent-clipped=6.0 +2023-02-06 04:14:06,689 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 04:14:13,200 INFO [train.py:901] (3/4) Epoch 7, batch 4450, loss[loss=0.3552, simple_loss=0.4053, pruned_loss=0.1525, over 8502.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3314, pruned_loss=0.09878, over 1610538.15 frames. ], batch size: 26, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:14:14,731 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52951.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:14:46,593 INFO [train.py:901] (3/4) Epoch 7, batch 4500, loss[loss=0.2552, simple_loss=0.3072, pruned_loss=0.1016, over 7658.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3325, pruned_loss=0.09969, over 1612166.61 frames. ], batch size: 19, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:14:59,361 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 04:15:10,353 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53032.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 04:15:11,488 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.236e+02 2.890e+02 3.405e+02 4.030e+02 1.067e+03, threshold=6.809e+02, percent-clipped=4.0 +2023-02-06 04:15:18,996 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53045.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:15:22,052 INFO [train.py:901] (3/4) Epoch 7, batch 4550, loss[loss=0.2597, simple_loss=0.3286, pruned_loss=0.09537, over 7636.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3312, pruned_loss=0.09904, over 1612434.23 frames. ], batch size: 19, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:15:27,688 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5520, 2.0254, 2.0398, 1.0920, 2.2947, 1.3753, 0.6862, 1.6940], + device='cuda:3'), covar=tensor([0.0301, 0.0152, 0.0130, 0.0273, 0.0162, 0.0445, 0.0409, 0.0151], + device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0263, 0.0221, 0.0322, 0.0261, 0.0408, 0.0318, 0.0298], + device='cuda:3'), out_proj_covar=tensor([1.1291e-04, 8.1943e-05, 6.8022e-05, 1.0050e-04, 8.2528e-05, 1.3840e-04, + 1.0194e-04, 9.3595e-05], device='cuda:3') +2023-02-06 04:15:29,785 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 +2023-02-06 04:15:36,992 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53070.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:15:39,630 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53074.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:15:56,227 INFO [train.py:901] (3/4) Epoch 7, batch 4600, loss[loss=0.2746, simple_loss=0.3462, pruned_loss=0.1015, over 8328.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3318, pruned_loss=0.09954, over 1612391.90 frames. ], batch size: 26, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:16:06,498 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53114.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:16:20,975 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.818e+02 2.951e+02 3.579e+02 4.375e+02 1.013e+03, threshold=7.158e+02, percent-clipped=5.0 +2023-02-06 04:16:31,866 INFO [train.py:901] (3/4) Epoch 7, batch 4650, loss[loss=0.3678, simple_loss=0.4019, pruned_loss=0.1669, over 7087.00 frames. ], tot_loss[loss=0.265, simple_loss=0.332, pruned_loss=0.09899, over 1613794.95 frames. ], batch size: 71, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:17:05,162 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1349, 1.5239, 1.6615, 1.3606, 1.0444, 1.4582, 1.6462, 1.6406], + device='cuda:3'), covar=tensor([0.0546, 0.1300, 0.1771, 0.1421, 0.0608, 0.1565, 0.0719, 0.0593], + device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0167, 0.0207, 0.0169, 0.0115, 0.0174, 0.0127, 0.0146], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 04:17:06,942 INFO [train.py:901] (3/4) Epoch 7, batch 4700, loss[loss=0.28, simple_loss=0.3421, pruned_loss=0.109, over 8038.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3315, pruned_loss=0.09848, over 1612435.16 frames. ], batch size: 22, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:17:27,464 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53229.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:17:30,531 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.642e+02 2.793e+02 3.469e+02 4.275e+02 9.300e+02, threshold=6.939e+02, percent-clipped=3.0 +2023-02-06 04:17:41,203 INFO [train.py:901] (3/4) Epoch 7, batch 4750, loss[loss=0.3161, simple_loss=0.377, pruned_loss=0.1276, over 8361.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3327, pruned_loss=0.09871, over 1616816.27 frames. ], batch size: 24, lr: 1.09e-02, grad_scale: 8.0 +2023-02-06 04:17:42,009 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4744, 2.1280, 3.4675, 2.1462, 2.7317, 3.9642, 3.7183, 3.5579], + device='cuda:3'), covar=tensor([0.0738, 0.1204, 0.0560, 0.1434, 0.0912, 0.0193, 0.0482, 0.0449], + device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0276, 0.0230, 0.0270, 0.0242, 0.0217, 0.0274, 0.0281], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 04:17:46,729 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2583, 2.6006, 1.7978, 2.1396, 1.9797, 1.4524, 1.8349, 1.9514], + device='cuda:3'), covar=tensor([0.1345, 0.0347, 0.0943, 0.0547, 0.0652, 0.1226, 0.0942, 0.0824], + device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0236, 0.0310, 0.0298, 0.0311, 0.0314, 0.0337, 0.0317], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 04:17:48,716 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53259.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:17:56,082 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.8817, 1.7601, 5.8329, 1.9718, 5.2669, 4.8464, 5.5108, 5.3786], + device='cuda:3'), covar=tensor([0.0308, 0.3646, 0.0286, 0.2974, 0.0804, 0.0567, 0.0345, 0.0378], + device='cuda:3'), in_proj_covar=tensor([0.0390, 0.0531, 0.0479, 0.0469, 0.0529, 0.0448, 0.0435, 0.0497], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 04:17:56,639 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 04:17:59,361 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 04:18:01,605 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5050, 1.7750, 4.6322, 2.1619, 4.1310, 3.9176, 4.2844, 4.1804], + device='cuda:3'), covar=tensor([0.0391, 0.3381, 0.0360, 0.2495, 0.0871, 0.0603, 0.0405, 0.0458], + device='cuda:3'), in_proj_covar=tensor([0.0390, 0.0531, 0.0479, 0.0469, 0.0530, 0.0448, 0.0435, 0.0497], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 04:18:04,735 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 +2023-02-06 04:18:09,725 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53288.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 04:18:16,601 INFO [train.py:901] (3/4) Epoch 7, batch 4800, loss[loss=0.2414, simple_loss=0.3177, pruned_loss=0.08255, over 7946.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3315, pruned_loss=0.09807, over 1617266.95 frames. ], batch size: 20, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:18:26,454 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53313.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 04:18:32,413 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53322.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:18:39,903 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53333.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:18:41,111 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.599e+02 2.630e+02 3.191e+02 3.984e+02 9.617e+02, threshold=6.381e+02, percent-clipped=3.0 +2023-02-06 04:18:50,455 INFO [train.py:901] (3/4) Epoch 7, batch 4850, loss[loss=0.2536, simple_loss=0.3305, pruned_loss=0.0883, over 7649.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3307, pruned_loss=0.0978, over 1616357.20 frames. ], batch size: 19, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:18:51,164 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 04:19:16,485 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53385.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:19:20,663 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9133, 2.1982, 3.8259, 2.9821, 3.2090, 2.4728, 1.6558, 1.7854], + device='cuda:3'), covar=tensor([0.2312, 0.3261, 0.0652, 0.1332, 0.1230, 0.1192, 0.1231, 0.2942], + device='cuda:3'), in_proj_covar=tensor([0.0818, 0.0767, 0.0659, 0.0755, 0.0847, 0.0698, 0.0650, 0.0693], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 04:19:26,612 INFO [train.py:901] (3/4) Epoch 7, batch 4900, loss[loss=0.2489, simple_loss=0.3295, pruned_loss=0.08415, over 8340.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3306, pruned_loss=0.09723, over 1619208.19 frames. ], batch size: 26, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:19:40,217 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53418.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:19:51,455 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.021e+02 2.811e+02 3.269e+02 4.328e+02 9.769e+02, threshold=6.539e+02, percent-clipped=6.0 +2023-02-06 04:20:00,859 INFO [train.py:901] (3/4) Epoch 7, batch 4950, loss[loss=0.2976, simple_loss=0.3684, pruned_loss=0.1134, over 8349.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3315, pruned_loss=0.09786, over 1622024.15 frames. ], batch size: 24, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:20:27,191 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53485.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:20:36,822 INFO [train.py:901] (3/4) Epoch 7, batch 5000, loss[loss=0.2638, simple_loss=0.342, pruned_loss=0.09279, over 8506.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.331, pruned_loss=0.09734, over 1620636.19 frames. ], batch size: 28, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:20:37,041 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6517, 2.0553, 2.0524, 1.1644, 2.3603, 1.4524, 0.6475, 1.6820], + device='cuda:3'), covar=tensor([0.0325, 0.0154, 0.0132, 0.0258, 0.0137, 0.0451, 0.0405, 0.0153], + device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0259, 0.0217, 0.0316, 0.0256, 0.0402, 0.0311, 0.0288], + device='cuda:3'), out_proj_covar=tensor([1.1241e-04, 8.0720e-05, 6.6808e-05, 9.7905e-05, 8.0513e-05, 1.3622e-04, + 9.9228e-05, 9.0097e-05], device='cuda:3') +2023-02-06 04:20:45,261 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53510.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:21:01,965 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53533.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:21:03,199 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 2.612e+02 3.156e+02 4.000e+02 8.821e+02, threshold=6.312e+02, percent-clipped=7.0 +2023-02-06 04:21:12,888 INFO [train.py:901] (3/4) Epoch 7, batch 5050, loss[loss=0.2367, simple_loss=0.3127, pruned_loss=0.08033, over 7802.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3302, pruned_loss=0.09687, over 1615788.27 frames. ], batch size: 20, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:21:31,995 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 04:21:46,780 INFO [train.py:901] (3/4) Epoch 7, batch 5100, loss[loss=0.2525, simple_loss=0.3189, pruned_loss=0.09306, over 8295.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3305, pruned_loss=0.09715, over 1615681.99 frames. ], batch size: 23, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:21:51,219 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53603.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:22:13,232 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.895e+02 2.889e+02 3.391e+02 4.238e+02 9.606e+02, threshold=6.783e+02, percent-clipped=10.0 +2023-02-06 04:22:23,450 INFO [train.py:901] (3/4) Epoch 7, batch 5150, loss[loss=0.2621, simple_loss=0.3289, pruned_loss=0.09769, over 8524.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.332, pruned_loss=0.09819, over 1617774.24 frames. ], batch size: 39, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:22:23,630 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3626, 1.6090, 1.7146, 1.2918, 1.1076, 1.5862, 1.7518, 1.7378], + device='cuda:3'), covar=tensor([0.0515, 0.1185, 0.1663, 0.1432, 0.0605, 0.1531, 0.0660, 0.0552], + device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0166, 0.0205, 0.0171, 0.0115, 0.0175, 0.0127, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 04:22:34,970 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53666.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:22:42,267 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53677.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:22:56,582 INFO [train.py:901] (3/4) Epoch 7, batch 5200, loss[loss=0.2462, simple_loss=0.3147, pruned_loss=0.08885, over 8612.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.332, pruned_loss=0.09856, over 1617554.60 frames. ], batch size: 34, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:23:10,744 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53718.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:23:17,973 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53729.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:23:22,034 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.983e+02 3.078e+02 4.028e+02 5.378e+02 1.177e+03, threshold=8.056e+02, percent-clipped=8.0 +2023-02-06 04:23:28,937 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 04:23:32,187 INFO [train.py:901] (3/4) Epoch 7, batch 5250, loss[loss=0.3331, simple_loss=0.3794, pruned_loss=0.1434, over 8579.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3316, pruned_loss=0.09821, over 1616692.80 frames. ], batch size: 34, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:23:54,699 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53781.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:23:58,805 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.6903, 5.7099, 5.0719, 1.7851, 5.1279, 5.4607, 5.5080, 4.9399], + device='cuda:3'), covar=tensor([0.0587, 0.0455, 0.0924, 0.5416, 0.0636, 0.0658, 0.1080, 0.0644], + device='cuda:3'), in_proj_covar=tensor([0.0425, 0.0324, 0.0357, 0.0443, 0.0343, 0.0322, 0.0334, 0.0282], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 04:24:00,280 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53789.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:24:02,339 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53792.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:24:06,925 INFO [train.py:901] (3/4) Epoch 7, batch 5300, loss[loss=0.2051, simple_loss=0.2807, pruned_loss=0.06478, over 7249.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3332, pruned_loss=0.09884, over 1613641.82 frames. ], batch size: 16, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:24:17,563 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53814.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:24:31,354 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.704e+02 2.647e+02 3.169e+02 3.870e+02 1.211e+03, threshold=6.339e+02, percent-clipped=2.0 +2023-02-06 04:24:36,505 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 04:24:39,058 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53844.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:24:42,222 INFO [train.py:901] (3/4) Epoch 7, batch 5350, loss[loss=0.2459, simple_loss=0.33, pruned_loss=0.0809, over 8489.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.333, pruned_loss=0.09837, over 1614509.83 frames. ], batch size: 28, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:25:17,625 INFO [train.py:901] (3/4) Epoch 7, batch 5400, loss[loss=0.2377, simple_loss=0.3123, pruned_loss=0.08158, over 8304.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3318, pruned_loss=0.09709, over 1612753.91 frames. ], batch size: 23, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:25:41,932 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.925e+02 2.728e+02 3.458e+02 4.119e+02 1.009e+03, threshold=6.915e+02, percent-clipped=3.0 +2023-02-06 04:25:43,934 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.1808, 1.3578, 3.3196, 0.9409, 2.8833, 2.8423, 3.0187, 2.9207], + device='cuda:3'), covar=tensor([0.0603, 0.3027, 0.0604, 0.2905, 0.1226, 0.0733, 0.0632, 0.0695], + device='cuda:3'), in_proj_covar=tensor([0.0391, 0.0526, 0.0478, 0.0461, 0.0523, 0.0441, 0.0440, 0.0494], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 04:25:51,250 INFO [train.py:901] (3/4) Epoch 7, batch 5450, loss[loss=0.286, simple_loss=0.3564, pruned_loss=0.1078, over 8741.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3317, pruned_loss=0.097, over 1619731.04 frames. ], batch size: 39, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:26:09,515 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53974.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:26:11,714 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 +2023-02-06 04:26:18,665 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 04:26:26,958 INFO [train.py:901] (3/4) Epoch 7, batch 5500, loss[loss=0.2387, simple_loss=0.307, pruned_loss=0.08521, over 7914.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3299, pruned_loss=0.09667, over 1610587.25 frames. ], batch size: 20, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:26:27,120 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53999.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:26:52,078 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.929e+02 2.794e+02 3.496e+02 4.646e+02 1.157e+03, threshold=6.993e+02, percent-clipped=7.0 +2023-02-06 04:26:53,573 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54037.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:27:01,157 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54048.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:27:01,616 INFO [train.py:901] (3/4) Epoch 7, batch 5550, loss[loss=0.2637, simple_loss=0.3397, pruned_loss=0.09385, over 8610.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3303, pruned_loss=0.09673, over 1615613.14 frames. ], batch size: 39, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:27:10,522 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54062.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:27:18,062 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54073.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:27:37,055 INFO [train.py:901] (3/4) Epoch 7, batch 5600, loss[loss=0.3206, simple_loss=0.3757, pruned_loss=0.1328, over 8337.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3311, pruned_loss=0.09685, over 1616516.37 frames. ], batch size: 26, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:27:37,987 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54100.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:27:55,818 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54125.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:28:02,458 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.954e+02 2.795e+02 3.455e+02 4.516e+02 9.788e+02, threshold=6.911e+02, percent-clipped=3.0 +2023-02-06 04:28:04,939 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-06 04:28:12,203 INFO [train.py:901] (3/4) Epoch 7, batch 5650, loss[loss=0.2442, simple_loss=0.3227, pruned_loss=0.0828, over 8366.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3319, pruned_loss=0.09762, over 1618663.91 frames. ], batch size: 24, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:28:25,370 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 04:28:42,171 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.69 vs. limit=5.0 +2023-02-06 04:28:47,273 INFO [train.py:901] (3/4) Epoch 7, batch 5700, loss[loss=0.2411, simple_loss=0.3135, pruned_loss=0.08429, over 5167.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3305, pruned_loss=0.09639, over 1610391.95 frames. ], batch size: 11, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:29:12,994 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.604e+02 2.786e+02 3.155e+02 4.023e+02 8.991e+02, threshold=6.311e+02, percent-clipped=4.0 +2023-02-06 04:29:22,470 INFO [train.py:901] (3/4) Epoch 7, batch 5750, loss[loss=0.3023, simple_loss=0.3604, pruned_loss=0.1221, over 8574.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.329, pruned_loss=0.09572, over 1608444.18 frames. ], batch size: 31, lr: 1.08e-02, grad_scale: 8.0 +2023-02-06 04:29:31,469 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 04:29:39,674 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5943, 1.3485, 2.8853, 1.2615, 1.9088, 3.0927, 3.0217, 2.6518], + device='cuda:3'), covar=tensor([0.0966, 0.1304, 0.0405, 0.1945, 0.0769, 0.0286, 0.0510, 0.0629], + device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0272, 0.0231, 0.0270, 0.0237, 0.0217, 0.0275, 0.0279], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 04:29:48,772 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 +2023-02-06 04:29:56,274 INFO [train.py:901] (3/4) Epoch 7, batch 5800, loss[loss=0.1835, simple_loss=0.2616, pruned_loss=0.05274, over 6858.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.329, pruned_loss=0.09619, over 1607311.72 frames. ], batch size: 15, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:30:03,822 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3959, 1.5583, 1.6617, 1.4467, 0.9158, 1.7543, 0.0672, 1.0895], + device='cuda:3'), covar=tensor([0.3097, 0.1964, 0.0744, 0.1906, 0.5615, 0.0650, 0.4407, 0.2208], + device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0144, 0.0086, 0.0193, 0.0230, 0.0089, 0.0152, 0.0141], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 04:30:22,036 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.127e+02 2.882e+02 3.737e+02 4.385e+02 9.194e+02, threshold=7.474e+02, percent-clipped=5.0 +2023-02-06 04:30:25,045 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.9773, 1.2268, 5.9069, 2.0522, 5.3455, 5.0678, 5.6019, 5.4203], + device='cuda:3'), covar=tensor([0.0281, 0.4463, 0.0218, 0.2750, 0.0754, 0.0566, 0.0295, 0.0371], + device='cuda:3'), in_proj_covar=tensor([0.0395, 0.0532, 0.0474, 0.0462, 0.0541, 0.0445, 0.0439, 0.0498], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 04:30:32,282 INFO [train.py:901] (3/4) Epoch 7, batch 5850, loss[loss=0.2711, simple_loss=0.3455, pruned_loss=0.09833, over 8366.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3302, pruned_loss=0.09709, over 1607254.27 frames. ], batch size: 24, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:31:06,116 INFO [train.py:901] (3/4) Epoch 7, batch 5900, loss[loss=0.222, simple_loss=0.2949, pruned_loss=0.07453, over 7439.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.33, pruned_loss=0.09735, over 1606727.36 frames. ], batch size: 17, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:31:30,650 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.792e+02 2.634e+02 3.151e+02 3.851e+02 7.879e+02, threshold=6.301e+02, percent-clipped=2.0 +2023-02-06 04:31:40,254 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5327, 1.7839, 2.8793, 1.2526, 2.0679, 1.7332, 1.6737, 1.7079], + device='cuda:3'), covar=tensor([0.1422, 0.1808, 0.0553, 0.3151, 0.1244, 0.2416, 0.1489, 0.1933], + device='cuda:3'), in_proj_covar=tensor([0.0477, 0.0477, 0.0529, 0.0552, 0.0597, 0.0530, 0.0456, 0.0585], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 04:31:40,698 INFO [train.py:901] (3/4) Epoch 7, batch 5950, loss[loss=0.2698, simple_loss=0.3376, pruned_loss=0.101, over 8621.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3307, pruned_loss=0.09712, over 1612398.25 frames. ], batch size: 39, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:31:55,859 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9927, 2.3128, 1.8839, 2.8204, 1.2809, 1.6182, 1.7473, 2.4276], + device='cuda:3'), covar=tensor([0.0859, 0.0969, 0.1245, 0.0470, 0.1536, 0.1802, 0.1432, 0.0862], + device='cuda:3'), in_proj_covar=tensor([0.0254, 0.0240, 0.0277, 0.0228, 0.0240, 0.0269, 0.0280, 0.0242], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 04:32:14,309 INFO [train.py:901] (3/4) Epoch 7, batch 6000, loss[loss=0.25, simple_loss=0.3222, pruned_loss=0.08894, over 8254.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3302, pruned_loss=0.09669, over 1612045.40 frames. ], batch size: 24, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:32:14,309 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 04:32:26,542 INFO [train.py:935] (3/4) Epoch 7, validation: loss=0.2048, simple_loss=0.3036, pruned_loss=0.05298, over 944034.00 frames. +2023-02-06 04:32:26,543 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 04:32:50,868 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.791e+02 2.687e+02 3.524e+02 4.445e+02 8.914e+02, threshold=7.048e+02, percent-clipped=8.0 +2023-02-06 04:33:00,119 INFO [train.py:901] (3/4) Epoch 7, batch 6050, loss[loss=0.2904, simple_loss=0.3595, pruned_loss=0.1106, over 8449.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3318, pruned_loss=0.09805, over 1614112.61 frames. ], batch size: 27, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:33:22,591 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.7323, 1.1917, 3.9320, 1.3483, 3.4725, 3.4117, 3.5646, 3.4279], + device='cuda:3'), covar=tensor([0.0574, 0.3835, 0.0577, 0.2958, 0.1303, 0.0852, 0.0588, 0.0685], + device='cuda:3'), in_proj_covar=tensor([0.0390, 0.0532, 0.0473, 0.0466, 0.0540, 0.0446, 0.0445, 0.0502], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 04:33:36,263 INFO [train.py:901] (3/4) Epoch 7, batch 6100, loss[loss=0.2359, simple_loss=0.324, pruned_loss=0.07389, over 8508.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3319, pruned_loss=0.09795, over 1613940.69 frames. ], batch size: 26, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:33:52,423 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6498, 2.0760, 3.3878, 1.3801, 2.4120, 2.1057, 1.7385, 2.1632], + device='cuda:3'), covar=tensor([0.1440, 0.1807, 0.0608, 0.3286, 0.1269, 0.2252, 0.1489, 0.1902], + device='cuda:3'), in_proj_covar=tensor([0.0478, 0.0472, 0.0525, 0.0557, 0.0599, 0.0530, 0.0453, 0.0586], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 04:33:57,248 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9283, 3.1184, 2.7178, 4.0679, 1.9521, 2.3058, 2.3716, 3.3493], + device='cuda:3'), covar=tensor([0.0689, 0.1012, 0.0992, 0.0290, 0.1354, 0.1456, 0.1439, 0.0855], + device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0237, 0.0275, 0.0225, 0.0236, 0.0268, 0.0276, 0.0239], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 04:34:00,442 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 04:34:01,815 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.948e+02 2.824e+02 3.447e+02 4.351e+02 1.012e+03, threshold=6.894e+02, percent-clipped=2.0 +2023-02-06 04:34:11,162 INFO [train.py:901] (3/4) Epoch 7, batch 6150, loss[loss=0.2711, simple_loss=0.327, pruned_loss=0.1076, over 7421.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3326, pruned_loss=0.09898, over 1612022.82 frames. ], batch size: 17, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:34:34,025 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=54682.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 04:34:46,676 INFO [train.py:901] (3/4) Epoch 7, batch 6200, loss[loss=0.3221, simple_loss=0.3661, pruned_loss=0.1391, over 8588.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.334, pruned_loss=0.1, over 1614801.83 frames. ], batch size: 50, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:35:12,146 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.915e+02 2.928e+02 3.624e+02 4.953e+02 9.267e+02, threshold=7.248e+02, percent-clipped=4.0 +2023-02-06 04:35:21,775 INFO [train.py:901] (3/4) Epoch 7, batch 6250, loss[loss=0.3075, simple_loss=0.3621, pruned_loss=0.1264, over 7198.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3324, pruned_loss=0.09935, over 1610305.21 frames. ], batch size: 71, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:35:21,999 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2770, 1.5485, 1.3753, 1.8839, 0.8620, 1.1847, 1.2752, 1.4696], + device='cuda:3'), covar=tensor([0.1159, 0.0998, 0.1368, 0.0685, 0.1490, 0.1905, 0.1037, 0.0970], + device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0236, 0.0272, 0.0223, 0.0236, 0.0264, 0.0272, 0.0238], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 04:35:44,427 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.66 vs. limit=5.0 +2023-02-06 04:35:55,499 INFO [train.py:901] (3/4) Epoch 7, batch 6300, loss[loss=0.3027, simple_loss=0.3554, pruned_loss=0.125, over 8197.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3327, pruned_loss=0.09918, over 1612677.95 frames. ], batch size: 23, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:36:22,281 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.742e+02 3.001e+02 3.662e+02 4.451e+02 9.002e+02, threshold=7.325e+02, percent-clipped=3.0 +2023-02-06 04:36:23,839 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5661, 1.7411, 2.8542, 1.2812, 2.0336, 1.9027, 1.5158, 1.6756], + device='cuda:3'), covar=tensor([0.1519, 0.1986, 0.0640, 0.3477, 0.1347, 0.2329, 0.1720, 0.2016], + device='cuda:3'), in_proj_covar=tensor([0.0481, 0.0475, 0.0534, 0.0564, 0.0602, 0.0529, 0.0453, 0.0590], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 04:36:24,160 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 +2023-02-06 04:36:32,317 INFO [train.py:901] (3/4) Epoch 7, batch 6350, loss[loss=0.2498, simple_loss=0.3149, pruned_loss=0.09234, over 7705.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3339, pruned_loss=0.09991, over 1612547.49 frames. ], batch size: 18, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:36:53,639 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=54880.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:37:06,007 INFO [train.py:901] (3/4) Epoch 7, batch 6400, loss[loss=0.2112, simple_loss=0.2904, pruned_loss=0.06602, over 8286.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3338, pruned_loss=0.09983, over 1616446.39 frames. ], batch size: 23, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:37:26,014 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9565, 4.0505, 2.6136, 2.7124, 2.7542, 2.2343, 2.4258, 2.9970], + device='cuda:3'), covar=tensor([0.1505, 0.0248, 0.0747, 0.0728, 0.0673, 0.1047, 0.0985, 0.0938], + device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0233, 0.0313, 0.0298, 0.0314, 0.0316, 0.0339, 0.0318], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 04:37:31,195 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.863e+02 2.661e+02 3.281e+02 3.949e+02 1.010e+03, threshold=6.562e+02, percent-clipped=2.0 +2023-02-06 04:37:40,714 INFO [train.py:901] (3/4) Epoch 7, batch 6450, loss[loss=0.2549, simple_loss=0.322, pruned_loss=0.0939, over 8240.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3327, pruned_loss=0.09908, over 1614063.16 frames. ], batch size: 22, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:37:56,506 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5831, 1.9318, 3.3136, 1.2875, 2.1726, 1.9415, 1.6413, 1.9964], + device='cuda:3'), covar=tensor([0.1471, 0.1820, 0.0555, 0.3342, 0.1384, 0.2339, 0.1538, 0.1960], + device='cuda:3'), in_proj_covar=tensor([0.0472, 0.0468, 0.0524, 0.0550, 0.0589, 0.0521, 0.0445, 0.0582], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 04:38:15,589 INFO [train.py:901] (3/4) Epoch 7, batch 6500, loss[loss=0.2835, simple_loss=0.3513, pruned_loss=0.1078, over 8261.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3323, pruned_loss=0.09886, over 1612278.48 frames. ], batch size: 24, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:38:33,834 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55026.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 04:38:39,660 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.787e+02 2.546e+02 3.271e+02 4.197e+02 5.859e+02, threshold=6.542e+02, percent-clipped=0.0 +2023-02-06 04:38:49,579 INFO [train.py:901] (3/4) Epoch 7, batch 6550, loss[loss=0.1928, simple_loss=0.2838, pruned_loss=0.05091, over 7944.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3336, pruned_loss=0.09952, over 1613721.08 frames. ], batch size: 20, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:39:08,940 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0510, 4.1434, 2.6323, 2.7910, 3.1469, 2.3690, 2.8230, 3.0007], + device='cuda:3'), covar=tensor([0.1484, 0.0180, 0.0868, 0.0713, 0.0639, 0.1129, 0.0975, 0.1192], + device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0231, 0.0311, 0.0296, 0.0306, 0.0313, 0.0337, 0.0318], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 04:39:12,182 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 04:39:17,120 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7525, 2.1375, 1.6721, 2.5964, 1.1693, 1.2997, 1.7093, 2.1538], + device='cuda:3'), covar=tensor([0.0990, 0.0864, 0.1333, 0.0497, 0.1424, 0.1946, 0.1240, 0.0940], + device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0239, 0.0279, 0.0225, 0.0235, 0.0270, 0.0278, 0.0242], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 04:39:25,813 INFO [train.py:901] (3/4) Epoch 7, batch 6600, loss[loss=0.207, simple_loss=0.2815, pruned_loss=0.06627, over 7436.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3328, pruned_loss=0.09825, over 1616544.13 frames. ], batch size: 17, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:39:32,333 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 04:39:49,419 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.976e+02 2.699e+02 3.503e+02 4.413e+02 7.218e+02, threshold=7.007e+02, percent-clipped=4.0 +2023-02-06 04:39:53,585 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55141.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 04:39:58,770 INFO [train.py:901] (3/4) Epoch 7, batch 6650, loss[loss=0.266, simple_loss=0.3354, pruned_loss=0.09834, over 8550.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3332, pruned_loss=0.09832, over 1617085.04 frames. ], batch size: 39, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:40:10,776 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55166.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:40:29,609 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5299, 2.8069, 1.8006, 2.1666, 2.2163, 1.5404, 2.1257, 2.2516], + device='cuda:3'), covar=tensor([0.1483, 0.0314, 0.0962, 0.0711, 0.0669, 0.1267, 0.0936, 0.0947], + device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0232, 0.0309, 0.0297, 0.0311, 0.0311, 0.0337, 0.0316], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 04:40:34,123 INFO [train.py:901] (3/4) Epoch 7, batch 6700, loss[loss=0.2087, simple_loss=0.286, pruned_loss=0.06568, over 7817.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3333, pruned_loss=0.09849, over 1618889.79 frames. ], batch size: 20, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:40:42,563 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-06 04:40:51,595 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55224.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:40:55,055 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55229.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:40:58,789 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.861e+02 3.031e+02 3.759e+02 4.673e+02 1.170e+03, threshold=7.519e+02, percent-clipped=9.0 +2023-02-06 04:41:07,993 INFO [train.py:901] (3/4) Epoch 7, batch 6750, loss[loss=0.2488, simple_loss=0.3229, pruned_loss=0.08734, over 8190.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3317, pruned_loss=0.09752, over 1615269.32 frames. ], batch size: 23, lr: 1.07e-02, grad_scale: 8.0 +2023-02-06 04:41:17,500 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7281, 1.9926, 2.1029, 1.3397, 2.2049, 1.4758, 1.0145, 1.7612], + device='cuda:3'), covar=tensor([0.0237, 0.0146, 0.0102, 0.0215, 0.0176, 0.0384, 0.0315, 0.0145], + device='cuda:3'), in_proj_covar=tensor([0.0344, 0.0263, 0.0218, 0.0318, 0.0258, 0.0407, 0.0317, 0.0297], + device='cuda:3'), out_proj_covar=tensor([1.0850e-04, 8.1515e-05, 6.6539e-05, 9.7277e-05, 8.0360e-05, 1.3681e-04, + 9.9991e-05, 9.2540e-05], device='cuda:3') +2023-02-06 04:41:28,742 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4097, 2.8297, 1.6554, 2.0642, 1.9410, 1.3750, 1.7238, 2.1448], + device='cuda:3'), covar=tensor([0.1459, 0.0284, 0.1036, 0.0707, 0.0778, 0.1435, 0.1143, 0.0937], + device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0231, 0.0310, 0.0299, 0.0311, 0.0315, 0.0339, 0.0318], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 04:41:38,038 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2658, 2.2105, 1.4131, 1.9402, 1.7558, 1.3154, 1.6063, 1.7253], + device='cuda:3'), covar=tensor([0.1254, 0.0342, 0.1056, 0.0494, 0.0590, 0.1289, 0.0802, 0.0807], + device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0232, 0.0313, 0.0301, 0.0313, 0.0317, 0.0340, 0.0320], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 04:41:42,537 INFO [train.py:901] (3/4) Epoch 7, batch 6800, loss[loss=0.1895, simple_loss=0.2637, pruned_loss=0.05769, over 7230.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3325, pruned_loss=0.09815, over 1616328.25 frames. ], batch size: 16, lr: 1.07e-02, grad_scale: 16.0 +2023-02-06 04:41:47,230 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 04:42:08,120 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5016, 2.8012, 1.5628, 2.0815, 2.1207, 1.4222, 1.9196, 2.0864], + device='cuda:3'), covar=tensor([0.1343, 0.0310, 0.1036, 0.0687, 0.0632, 0.1232, 0.0913, 0.0894], + device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0231, 0.0310, 0.0299, 0.0311, 0.0314, 0.0337, 0.0318], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 04:42:08,554 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.679e+02 2.859e+02 3.364e+02 4.161e+02 9.626e+02, threshold=6.728e+02, percent-clipped=3.0 +2023-02-06 04:42:11,524 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55339.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:42:18,203 INFO [train.py:901] (3/4) Epoch 7, batch 6850, loss[loss=0.281, simple_loss=0.343, pruned_loss=0.1095, over 8282.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.333, pruned_loss=0.09875, over 1617854.67 frames. ], batch size: 23, lr: 1.06e-02, grad_scale: 16.0 +2023-02-06 04:42:34,166 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 04:42:50,632 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55397.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 04:42:52,432 INFO [train.py:901] (3/4) Epoch 7, batch 6900, loss[loss=0.2131, simple_loss=0.2879, pruned_loss=0.06916, over 7924.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3332, pruned_loss=0.09831, over 1620766.38 frames. ], batch size: 20, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:43:09,632 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55422.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 04:43:19,277 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.721e+02 2.767e+02 3.318e+02 4.413e+02 7.718e+02, threshold=6.635e+02, percent-clipped=1.0 +2023-02-06 04:43:28,911 INFO [train.py:901] (3/4) Epoch 7, batch 6950, loss[loss=0.3456, simple_loss=0.3842, pruned_loss=0.1535, over 8645.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3318, pruned_loss=0.09772, over 1616163.74 frames. ], batch size: 49, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:43:42,924 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 04:43:46,612 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 04:43:48,201 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2023-02-06 04:44:02,247 INFO [train.py:901] (3/4) Epoch 7, batch 7000, loss[loss=0.2143, simple_loss=0.2747, pruned_loss=0.07693, over 6792.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3342, pruned_loss=0.09895, over 1619791.16 frames. ], batch size: 15, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:44:04,388 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1648, 3.1517, 2.2362, 2.4313, 2.5189, 2.1208, 2.4195, 2.7790], + device='cuda:3'), covar=tensor([0.0956, 0.0237, 0.0650, 0.0534, 0.0469, 0.0796, 0.0656, 0.0658], + device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0232, 0.0308, 0.0297, 0.0308, 0.0317, 0.0335, 0.0316], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 04:44:09,458 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55510.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:44:28,243 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.786e+02 2.942e+02 3.699e+02 4.542e+02 1.220e+03, threshold=7.399e+02, percent-clipped=11.0 +2023-02-06 04:44:37,059 INFO [train.py:901] (3/4) Epoch 7, batch 7050, loss[loss=0.2339, simple_loss=0.2963, pruned_loss=0.08572, over 8235.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.333, pruned_loss=0.0981, over 1613962.19 frames. ], batch size: 22, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:44:43,035 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 04:44:54,228 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55573.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:45:09,299 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55595.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:45:11,742 INFO [train.py:901] (3/4) Epoch 7, batch 7100, loss[loss=0.2193, simple_loss=0.2923, pruned_loss=0.0732, over 7793.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3309, pruned_loss=0.09703, over 1613322.55 frames. ], batch size: 19, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:45:26,280 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55620.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:45:29,498 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55625.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:45:36,698 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.658e+02 2.720e+02 3.297e+02 4.008e+02 7.250e+02, threshold=6.594e+02, percent-clipped=0.0 +2023-02-06 04:45:38,761 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3214, 4.2420, 3.8524, 1.8955, 3.8426, 3.6911, 3.8946, 3.3983], + device='cuda:3'), covar=tensor([0.0758, 0.0568, 0.1017, 0.4676, 0.0785, 0.0932, 0.1217, 0.0919], + device='cuda:3'), in_proj_covar=tensor([0.0413, 0.0324, 0.0358, 0.0439, 0.0339, 0.0317, 0.0328, 0.0281], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 04:45:45,953 INFO [train.py:901] (3/4) Epoch 7, batch 7150, loss[loss=0.2685, simple_loss=0.3443, pruned_loss=0.09633, over 8246.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3301, pruned_loss=0.09634, over 1610605.96 frames. ], batch size: 24, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:46:14,224 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55688.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:46:21,659 INFO [train.py:901] (3/4) Epoch 7, batch 7200, loss[loss=0.2699, simple_loss=0.337, pruned_loss=0.1014, over 8250.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3314, pruned_loss=0.09673, over 1617312.56 frames. ], batch size: 24, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:46:30,269 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.96 vs. limit=5.0 +2023-02-06 04:46:38,149 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-06 04:46:47,138 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.695e+02 2.857e+02 3.487e+02 4.455e+02 1.230e+03, threshold=6.974e+02, percent-clipped=5.0 +2023-02-06 04:46:55,811 INFO [train.py:901] (3/4) Epoch 7, batch 7250, loss[loss=0.2551, simple_loss=0.3136, pruned_loss=0.09825, over 7228.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3297, pruned_loss=0.09615, over 1611565.07 frames. ], batch size: 16, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:47:07,788 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55766.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:47:15,884 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6302, 1.3584, 2.8285, 1.1625, 2.0065, 3.0632, 3.0694, 2.5909], + device='cuda:3'), covar=tensor([0.1005, 0.1288, 0.0352, 0.1928, 0.0740, 0.0283, 0.0429, 0.0640], + device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0271, 0.0229, 0.0269, 0.0239, 0.0212, 0.0277, 0.0276], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2023-02-06 04:47:31,018 INFO [train.py:901] (3/4) Epoch 7, batch 7300, loss[loss=0.2421, simple_loss=0.3088, pruned_loss=0.08766, over 8691.00 frames. ], tot_loss[loss=0.261, simple_loss=0.33, pruned_loss=0.09599, over 1617361.25 frames. ], batch size: 34, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:47:55,727 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.870e+02 2.887e+02 3.402e+02 4.424e+02 1.529e+03, threshold=6.804e+02, percent-clipped=7.0 +2023-02-06 04:48:04,227 INFO [train.py:901] (3/4) Epoch 7, batch 7350, loss[loss=0.2086, simple_loss=0.2785, pruned_loss=0.06931, over 7550.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3302, pruned_loss=0.09693, over 1613848.46 frames. ], batch size: 18, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:48:09,146 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9014, 2.5721, 4.6961, 1.4280, 3.2558, 2.4411, 2.0390, 2.9725], + device='cuda:3'), covar=tensor([0.1441, 0.1805, 0.0546, 0.3275, 0.1357, 0.2249, 0.1405, 0.1958], + device='cuda:3'), in_proj_covar=tensor([0.0474, 0.0474, 0.0531, 0.0554, 0.0596, 0.0531, 0.0452, 0.0592], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 04:48:15,839 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55866.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:48:16,838 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.06 vs. limit=5.0 +2023-02-06 04:48:26,613 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55881.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:48:27,826 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 04:48:40,229 INFO [train.py:901] (3/4) Epoch 7, batch 7400, loss[loss=0.2825, simple_loss=0.3223, pruned_loss=0.1213, over 7438.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3298, pruned_loss=0.09667, over 1615845.61 frames. ], batch size: 17, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:48:45,324 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55906.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:48:50,024 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 04:49:05,903 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.566e+02 2.646e+02 3.471e+02 4.467e+02 1.348e+03, threshold=6.942e+02, percent-clipped=5.0 +2023-02-06 04:49:11,674 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55944.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:49:14,924 INFO [train.py:901] (3/4) Epoch 7, batch 7450, loss[loss=0.3126, simple_loss=0.3675, pruned_loss=0.1289, over 8500.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3312, pruned_loss=0.09783, over 1612093.69 frames. ], batch size: 39, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:49:25,913 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 04:49:28,833 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55969.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:49:38,425 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8988, 2.0171, 2.3370, 1.6483, 1.2104, 2.4651, 0.3960, 1.3989], + device='cuda:3'), covar=tensor([0.3531, 0.1562, 0.0633, 0.2501, 0.5515, 0.0763, 0.4522, 0.2341], + device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0144, 0.0085, 0.0191, 0.0229, 0.0088, 0.0145, 0.0141], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 04:49:50,667 INFO [train.py:901] (3/4) Epoch 7, batch 7500, loss[loss=0.3191, simple_loss=0.3779, pruned_loss=0.1301, over 8621.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3317, pruned_loss=0.09832, over 1614617.98 frames. ], batch size: 39, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:50:17,129 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.595e+02 2.816e+02 3.537e+02 4.737e+02 9.745e+02, threshold=7.074e+02, percent-clipped=6.0 +2023-02-06 04:50:23,244 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5436, 4.4501, 4.0722, 1.8495, 3.9849, 4.0387, 4.0376, 3.4973], + device='cuda:3'), covar=tensor([0.0758, 0.0619, 0.0974, 0.4989, 0.0739, 0.0743, 0.1306, 0.1026], + device='cuda:3'), in_proj_covar=tensor([0.0425, 0.0330, 0.0361, 0.0447, 0.0345, 0.0320, 0.0335, 0.0287], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 04:50:25,625 INFO [train.py:901] (3/4) Epoch 7, batch 7550, loss[loss=0.2469, simple_loss=0.306, pruned_loss=0.09391, over 8240.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3302, pruned_loss=0.0973, over 1613203.70 frames. ], batch size: 22, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:50:58,643 INFO [train.py:901] (3/4) Epoch 7, batch 7600, loss[loss=0.3025, simple_loss=0.362, pruned_loss=0.1215, over 8433.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3305, pruned_loss=0.098, over 1611752.87 frames. ], batch size: 27, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:51:06,785 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56110.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:51:25,294 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.963e+02 2.750e+02 3.495e+02 4.537e+02 9.121e+02, threshold=6.990e+02, percent-clipped=3.0 +2023-02-06 04:51:34,927 INFO [train.py:901] (3/4) Epoch 7, batch 7650, loss[loss=0.2753, simple_loss=0.3398, pruned_loss=0.1054, over 7014.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3312, pruned_loss=0.09809, over 1610843.30 frames. ], batch size: 71, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:51:44,429 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56163.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:51:47,899 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5813, 2.5773, 4.6124, 1.3053, 3.0081, 1.9982, 1.8645, 2.3313], + device='cuda:3'), covar=tensor([0.2007, 0.1938, 0.0678, 0.4183, 0.1723, 0.2901, 0.1837, 0.2800], + device='cuda:3'), in_proj_covar=tensor([0.0479, 0.0478, 0.0533, 0.0562, 0.0599, 0.0529, 0.0457, 0.0594], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 04:51:50,263 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 +2023-02-06 04:52:04,033 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.8117, 4.7221, 4.3069, 2.2408, 4.2453, 4.2235, 4.3173, 3.8334], + device='cuda:3'), covar=tensor([0.0544, 0.0444, 0.0738, 0.3665, 0.0683, 0.0667, 0.0920, 0.0796], + device='cuda:3'), in_proj_covar=tensor([0.0417, 0.0327, 0.0358, 0.0442, 0.0342, 0.0316, 0.0329, 0.0283], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 04:52:08,629 INFO [train.py:901] (3/4) Epoch 7, batch 7700, loss[loss=0.256, simple_loss=0.3278, pruned_loss=0.09209, over 8029.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3314, pruned_loss=0.09819, over 1613114.92 frames. ], batch size: 22, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:52:16,096 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56210.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:52:22,688 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1844, 1.4619, 3.4707, 1.3610, 2.3594, 3.9442, 3.9169, 3.3687], + device='cuda:3'), covar=tensor([0.0931, 0.1392, 0.0323, 0.1994, 0.0781, 0.0231, 0.0377, 0.0625], + device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0269, 0.0231, 0.0268, 0.0238, 0.0210, 0.0274, 0.0272], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2023-02-06 04:52:22,936 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-06 04:52:26,763 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56225.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:52:34,688 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.593e+02 2.791e+02 3.394e+02 3.978e+02 9.035e+02, threshold=6.788e+02, percent-clipped=3.0 +2023-02-06 04:52:34,717 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 04:52:44,020 INFO [train.py:901] (3/4) Epoch 7, batch 7750, loss[loss=0.2772, simple_loss=0.3405, pruned_loss=0.1069, over 7792.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3302, pruned_loss=0.09754, over 1609058.53 frames. ], batch size: 19, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:52:49,380 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 +2023-02-06 04:53:18,265 INFO [train.py:901] (3/4) Epoch 7, batch 7800, loss[loss=0.2132, simple_loss=0.2825, pruned_loss=0.07194, over 7558.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3276, pruned_loss=0.0957, over 1608499.67 frames. ], batch size: 18, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:53:35,684 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56325.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:53:42,620 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.660e+02 2.702e+02 3.307e+02 4.383e+02 8.490e+02, threshold=6.613e+02, percent-clipped=4.0 +2023-02-06 04:53:51,366 INFO [train.py:901] (3/4) Epoch 7, batch 7850, loss[loss=0.2519, simple_loss=0.332, pruned_loss=0.0859, over 8251.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3275, pruned_loss=0.09537, over 1610195.82 frames. ], batch size: 24, lr: 1.06e-02, grad_scale: 8.0 +2023-02-06 04:54:24,864 INFO [train.py:901] (3/4) Epoch 7, batch 7900, loss[loss=0.2882, simple_loss=0.3589, pruned_loss=0.1087, over 8366.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3287, pruned_loss=0.09571, over 1611820.68 frames. ], batch size: 24, lr: 1.05e-02, grad_scale: 8.0 +2023-02-06 04:54:49,429 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 2.758e+02 3.520e+02 4.424e+02 1.197e+03, threshold=7.039e+02, percent-clipped=9.0 +2023-02-06 04:54:58,058 INFO [train.py:901] (3/4) Epoch 7, batch 7950, loss[loss=0.2446, simple_loss=0.3264, pruned_loss=0.08134, over 8448.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3292, pruned_loss=0.09579, over 1613798.87 frames. ], batch size: 27, lr: 1.05e-02, grad_scale: 8.0 +2023-02-06 04:55:19,809 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56481.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:55:31,649 INFO [train.py:901] (3/4) Epoch 7, batch 8000, loss[loss=0.224, simple_loss=0.2952, pruned_loss=0.07634, over 8069.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3286, pruned_loss=0.0953, over 1612047.12 frames. ], batch size: 21, lr: 1.05e-02, grad_scale: 8.0 +2023-02-06 04:55:36,308 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56506.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:55:36,851 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56507.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:55:56,107 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.751e+02 2.809e+02 3.378e+02 4.457e+02 7.052e+02, threshold=6.755e+02, percent-clipped=1.0 +2023-02-06 04:55:56,474 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-06 04:55:59,646 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56541.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 04:56:04,838 INFO [train.py:901] (3/4) Epoch 7, batch 8050, loss[loss=0.3255, simple_loss=0.3737, pruned_loss=0.1386, over 6622.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3272, pruned_loss=0.0954, over 1599754.43 frames. ], batch size: 72, lr: 1.05e-02, grad_scale: 8.0 +2023-02-06 04:56:37,957 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 04:56:42,617 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56581.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:56:43,089 INFO [train.py:901] (3/4) Epoch 8, batch 0, loss[loss=0.256, simple_loss=0.3244, pruned_loss=0.0938, over 7980.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3244, pruned_loss=0.0938, over 7980.00 frames. ], batch size: 21, lr: 9.92e-03, grad_scale: 8.0 +2023-02-06 04:56:43,090 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 04:56:54,073 INFO [train.py:935] (3/4) Epoch 8, validation: loss=0.205, simple_loss=0.3028, pruned_loss=0.05355, over 944034.00 frames. +2023-02-06 04:56:54,074 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 04:56:54,903 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9354, 2.8337, 2.6396, 1.4244, 2.5576, 2.6299, 2.6366, 2.4658], + device='cuda:3'), covar=tensor([0.1512, 0.1167, 0.1597, 0.5062, 0.1234, 0.1570, 0.1903, 0.1333], + device='cuda:3'), in_proj_covar=tensor([0.0419, 0.0327, 0.0354, 0.0451, 0.0343, 0.0322, 0.0334, 0.0287], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 04:57:08,612 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 04:57:08,792 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4152, 1.8152, 1.3152, 2.3719, 1.0417, 1.1504, 1.5754, 1.8441], + device='cuda:3'), covar=tensor([0.1164, 0.1059, 0.1551, 0.0598, 0.1508, 0.2085, 0.1338, 0.0947], + device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0240, 0.0279, 0.0223, 0.0236, 0.0269, 0.0274, 0.0236], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 04:57:10,750 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56606.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:57:22,333 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56622.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:57:28,989 INFO [train.py:901] (3/4) Epoch 8, batch 50, loss[loss=0.2792, simple_loss=0.3465, pruned_loss=0.1059, over 8356.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3336, pruned_loss=0.09748, over 370774.27 frames. ], batch size: 24, lr: 9.92e-03, grad_scale: 8.0 +2023-02-06 04:57:31,765 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.799e+02 2.831e+02 3.488e+02 4.265e+02 1.069e+03, threshold=6.975e+02, percent-clipped=2.0 +2023-02-06 04:57:43,125 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 04:58:03,657 INFO [train.py:901] (3/4) Epoch 8, batch 100, loss[loss=0.2086, simple_loss=0.2696, pruned_loss=0.07377, over 7447.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3331, pruned_loss=0.09862, over 649301.08 frames. ], batch size: 17, lr: 9.91e-03, grad_scale: 8.0 +2023-02-06 04:58:05,751 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 04:58:08,609 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.5915, 5.6142, 4.9442, 2.2569, 5.0410, 5.4643, 5.1610, 4.9841], + device='cuda:3'), covar=tensor([0.0578, 0.0396, 0.0681, 0.4189, 0.0545, 0.0462, 0.0896, 0.0556], + device='cuda:3'), in_proj_covar=tensor([0.0423, 0.0326, 0.0358, 0.0450, 0.0345, 0.0325, 0.0335, 0.0290], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 04:58:37,162 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3667, 1.8830, 2.9235, 2.3950, 2.5679, 2.0689, 1.7144, 1.7987], + device='cuda:3'), covar=tensor([0.2024, 0.2783, 0.0734, 0.1465, 0.1189, 0.1431, 0.1170, 0.2247], + device='cuda:3'), in_proj_covar=tensor([0.0827, 0.0780, 0.0671, 0.0780, 0.0868, 0.0720, 0.0671, 0.0708], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 04:58:38,288 INFO [train.py:901] (3/4) Epoch 8, batch 150, loss[loss=0.225, simple_loss=0.2985, pruned_loss=0.07571, over 8229.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3307, pruned_loss=0.09586, over 863582.53 frames. ], batch size: 22, lr: 9.91e-03, grad_scale: 8.0 +2023-02-06 04:58:39,112 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1896, 1.6800, 4.2711, 1.3678, 2.4808, 4.8742, 5.0284, 3.8404], + device='cuda:3'), covar=tensor([0.1355, 0.1666, 0.0323, 0.2509, 0.0990, 0.0358, 0.0401, 0.1017], + device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0277, 0.0236, 0.0273, 0.0243, 0.0214, 0.0281, 0.0281], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 04:58:40,996 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.667e+02 2.710e+02 3.372e+02 4.105e+02 8.611e+02, threshold=6.744e+02, percent-clipped=2.0 +2023-02-06 04:59:12,801 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-06 04:59:13,823 INFO [train.py:901] (3/4) Epoch 8, batch 200, loss[loss=0.3055, simple_loss=0.3663, pruned_loss=0.1224, over 8734.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3296, pruned_loss=0.09471, over 1032745.35 frames. ], batch size: 34, lr: 9.90e-03, grad_scale: 8.0 +2023-02-06 04:59:26,215 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 +2023-02-06 04:59:41,302 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56821.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 04:59:48,622 INFO [train.py:901] (3/4) Epoch 8, batch 250, loss[loss=0.2922, simple_loss=0.3654, pruned_loss=0.1095, over 8333.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3316, pruned_loss=0.09583, over 1161185.85 frames. ], batch size: 25, lr: 9.90e-03, grad_scale: 8.0 +2023-02-06 04:59:51,336 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.820e+02 2.703e+02 3.318e+02 4.204e+02 1.022e+03, threshold=6.636e+02, percent-clipped=1.0 +2023-02-06 04:59:56,902 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 05:00:06,241 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 05:00:21,346 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0889, 2.3360, 1.9900, 2.7577, 1.4618, 1.5663, 2.0104, 2.3649], + device='cuda:3'), covar=tensor([0.0826, 0.0929, 0.1199, 0.0544, 0.1296, 0.1752, 0.1033, 0.0815], + device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0235, 0.0275, 0.0223, 0.0234, 0.0266, 0.0271, 0.0237], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 05:00:21,381 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56878.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:00:23,889 INFO [train.py:901] (3/4) Epoch 8, batch 300, loss[loss=0.2413, simple_loss=0.3122, pruned_loss=0.08524, over 7909.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3314, pruned_loss=0.09622, over 1262964.70 frames. ], batch size: 20, lr: 9.89e-03, grad_scale: 8.0 +2023-02-06 05:00:25,999 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56885.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 05:00:38,049 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56903.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:00:54,963 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56926.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:00:58,776 INFO [train.py:901] (3/4) Epoch 8, batch 350, loss[loss=0.3267, simple_loss=0.3802, pruned_loss=0.1366, over 8460.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3303, pruned_loss=0.09528, over 1343807.09 frames. ], batch size: 27, lr: 9.89e-03, grad_scale: 8.0 +2023-02-06 05:01:01,452 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.824e+02 2.612e+02 3.168e+02 3.951e+02 1.059e+03, threshold=6.336e+02, percent-clipped=3.0 +2023-02-06 05:01:33,167 INFO [train.py:901] (3/4) Epoch 8, batch 400, loss[loss=0.2329, simple_loss=0.3216, pruned_loss=0.07213, over 8517.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.331, pruned_loss=0.09564, over 1407290.10 frames. ], batch size: 28, lr: 9.89e-03, grad_scale: 8.0 +2023-02-06 05:01:45,435 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57000.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 05:01:50,619 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.0091, 3.0737, 3.3354, 2.0775, 1.8389, 3.3923, 0.6973, 1.8772], + device='cuda:3'), covar=tensor([0.2904, 0.1249, 0.0411, 0.3317, 0.5484, 0.0504, 0.4045, 0.2542], + device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0145, 0.0088, 0.0197, 0.0238, 0.0092, 0.0146, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:01:52,588 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57011.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:01:53,825 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57013.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:02:04,583 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5986, 1.9933, 3.4173, 1.2825, 2.4464, 2.0813, 1.5257, 2.1951], + device='cuda:3'), covar=tensor([0.1593, 0.1957, 0.0701, 0.3504, 0.1476, 0.2510, 0.1732, 0.2195], + device='cuda:3'), in_proj_covar=tensor([0.0483, 0.0478, 0.0543, 0.0559, 0.0603, 0.0537, 0.0458, 0.0602], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 05:02:07,049 INFO [train.py:901] (3/4) Epoch 8, batch 450, loss[loss=0.24, simple_loss=0.305, pruned_loss=0.08745, over 7797.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3311, pruned_loss=0.09583, over 1454891.31 frames. ], batch size: 19, lr: 9.88e-03, grad_scale: 8.0 +2023-02-06 05:02:10,305 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.037e+02 2.769e+02 3.532e+02 4.551e+02 9.004e+02, threshold=7.064e+02, percent-clipped=7.0 +2023-02-06 05:02:41,859 INFO [train.py:901] (3/4) Epoch 8, batch 500, loss[loss=0.2485, simple_loss=0.331, pruned_loss=0.08302, over 8340.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3292, pruned_loss=0.09456, over 1487056.85 frames. ], batch size: 26, lr: 9.88e-03, grad_scale: 8.0 +2023-02-06 05:03:15,882 INFO [train.py:901] (3/4) Epoch 8, batch 550, loss[loss=0.2513, simple_loss=0.3289, pruned_loss=0.0869, over 8505.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3295, pruned_loss=0.09513, over 1513316.64 frames. ], batch size: 26, lr: 9.87e-03, grad_scale: 8.0 +2023-02-06 05:03:18,519 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.804e+02 2.761e+02 3.532e+02 4.192e+02 1.400e+03, threshold=7.064e+02, percent-clipped=6.0 +2023-02-06 05:03:39,524 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57165.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:03:50,883 INFO [train.py:901] (3/4) Epoch 8, batch 600, loss[loss=0.2768, simple_loss=0.3336, pruned_loss=0.11, over 8497.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3294, pruned_loss=0.09462, over 1540145.51 frames. ], batch size: 26, lr: 9.87e-03, grad_scale: 8.0 +2023-02-06 05:04:02,995 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 05:04:06,349 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57204.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:04:13,913 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57215.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:04:25,746 INFO [train.py:901] (3/4) Epoch 8, batch 650, loss[loss=0.278, simple_loss=0.3423, pruned_loss=0.1069, over 8372.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3285, pruned_loss=0.09462, over 1552149.95 frames. ], batch size: 24, lr: 9.86e-03, grad_scale: 8.0 +2023-02-06 05:04:28,376 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.822e+02 2.474e+02 3.242e+02 4.284e+02 1.059e+03, threshold=6.484e+02, percent-clipped=6.0 +2023-02-06 05:04:29,950 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1888, 2.8103, 2.9020, 1.4111, 3.0759, 1.7997, 1.6761, 1.6525], + device='cuda:3'), covar=tensor([0.0386, 0.0181, 0.0170, 0.0331, 0.0301, 0.0450, 0.0467, 0.0290], + device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0271, 0.0224, 0.0330, 0.0266, 0.0418, 0.0325, 0.0302], + device='cuda:3'), out_proj_covar=tensor([1.1024e-04, 8.2685e-05, 6.7404e-05, 1.0004e-04, 8.2607e-05, 1.3904e-04, + 1.0134e-04, 9.3016e-05], device='cuda:3') +2023-02-06 05:04:41,274 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4650, 2.0513, 1.9830, 1.0523, 2.2235, 1.3786, 0.5700, 1.5823], + device='cuda:3'), covar=tensor([0.0281, 0.0147, 0.0121, 0.0255, 0.0150, 0.0433, 0.0392, 0.0149], + device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0272, 0.0224, 0.0333, 0.0268, 0.0420, 0.0328, 0.0303], + device='cuda:3'), out_proj_covar=tensor([1.1101e-04, 8.3030e-05, 6.7443e-05, 1.0093e-04, 8.3005e-05, 1.3962e-04, + 1.0202e-04, 9.3388e-05], device='cuda:3') +2023-02-06 05:04:41,994 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57256.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 05:04:46,656 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57262.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:04:51,987 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57270.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:04:59,512 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57280.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:05:00,219 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57281.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 05:05:00,683 INFO [train.py:901] (3/4) Epoch 8, batch 700, loss[loss=0.2364, simple_loss=0.3104, pruned_loss=0.08124, over 8238.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3277, pruned_loss=0.09459, over 1563253.53 frames. ], batch size: 22, lr: 9.86e-03, grad_scale: 8.0 +2023-02-06 05:05:34,915 INFO [train.py:901] (3/4) Epoch 8, batch 750, loss[loss=0.2985, simple_loss=0.3651, pruned_loss=0.1159, over 8331.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3267, pruned_loss=0.09417, over 1568443.46 frames. ], batch size: 26, lr: 9.86e-03, grad_scale: 8.0 +2023-02-06 05:05:35,407 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 05:05:38,344 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.790e+02 2.846e+02 3.371e+02 4.091e+02 7.333e+02, threshold=6.742e+02, percent-clipped=1.0 +2023-02-06 05:05:49,697 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 05:05:51,159 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57355.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:05:52,580 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57357.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:05:57,732 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 05:05:58,592 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9435, 2.1306, 1.7442, 2.6735, 1.1434, 1.5192, 1.8171, 2.1920], + device='cuda:3'), covar=tensor([0.0829, 0.0927, 0.1261, 0.0405, 0.1269, 0.1577, 0.1047, 0.0727], + device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0235, 0.0275, 0.0217, 0.0232, 0.0267, 0.0271, 0.0237], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 05:06:09,705 INFO [train.py:901] (3/4) Epoch 8, batch 800, loss[loss=0.2867, simple_loss=0.3478, pruned_loss=0.1128, over 8357.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3302, pruned_loss=0.09636, over 1585182.81 frames. ], batch size: 24, lr: 9.85e-03, grad_scale: 16.0 +2023-02-06 05:06:10,112 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2023-02-06 05:06:11,947 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57385.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:06:33,606 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57416.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:06:38,323 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7774, 1.5168, 3.2313, 1.1613, 2.1069, 3.5750, 3.6298, 2.8127], + device='cuda:3'), covar=tensor([0.1263, 0.1638, 0.0482, 0.2324, 0.1065, 0.0362, 0.0441, 0.0849], + device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0276, 0.0234, 0.0271, 0.0244, 0.0217, 0.0281, 0.0281], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 05:06:44,177 INFO [train.py:901] (3/4) Epoch 8, batch 850, loss[loss=0.2601, simple_loss=0.3257, pruned_loss=0.09725, over 8344.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3308, pruned_loss=0.09667, over 1593197.86 frames. ], batch size: 26, lr: 9.85e-03, grad_scale: 16.0 +2023-02-06 05:06:46,893 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.644e+02 2.664e+02 3.287e+02 4.255e+02 8.769e+02, threshold=6.575e+02, percent-clipped=4.0 +2023-02-06 05:07:11,117 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57470.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:07:12,470 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57472.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:07:13,429 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-06 05:07:17,839 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57480.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:07:19,065 INFO [train.py:901] (3/4) Epoch 8, batch 900, loss[loss=0.2268, simple_loss=0.306, pruned_loss=0.07385, over 8029.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3304, pruned_loss=0.0965, over 1597290.98 frames. ], batch size: 22, lr: 9.84e-03, grad_scale: 16.0 +2023-02-06 05:07:40,917 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57513.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:07:53,661 INFO [train.py:901] (3/4) Epoch 8, batch 950, loss[loss=0.2417, simple_loss=0.3182, pruned_loss=0.08254, over 8459.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3299, pruned_loss=0.09647, over 1599244.96 frames. ], batch size: 25, lr: 9.84e-03, grad_scale: 16.0 +2023-02-06 05:07:56,417 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.722e+02 2.713e+02 3.197e+02 4.416e+02 7.629e+02, threshold=6.394e+02, percent-clipped=6.0 +2023-02-06 05:07:56,683 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57536.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:07:57,304 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5354, 1.4376, 2.7471, 1.2093, 2.1028, 3.0142, 3.0018, 2.5543], + device='cuda:3'), covar=tensor([0.1097, 0.1396, 0.0484, 0.2017, 0.0780, 0.0320, 0.0577, 0.0678], + device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0281, 0.0240, 0.0275, 0.0248, 0.0219, 0.0288, 0.0286], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 05:08:04,593 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57548.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:08:13,027 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57559.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:08:14,312 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 05:08:14,512 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57561.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:08:29,126 INFO [train.py:901] (3/4) Epoch 8, batch 1000, loss[loss=0.2822, simple_loss=0.3525, pruned_loss=0.106, over 8446.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3294, pruned_loss=0.09549, over 1608556.26 frames. ], batch size: 27, lr: 9.83e-03, grad_scale: 16.0 +2023-02-06 05:08:45,909 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57606.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:08:47,895 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 05:09:00,595 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 05:09:03,903 INFO [train.py:901] (3/4) Epoch 8, batch 1050, loss[loss=0.2587, simple_loss=0.3268, pruned_loss=0.09528, over 7923.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3271, pruned_loss=0.0943, over 1603595.25 frames. ], batch size: 20, lr: 9.83e-03, grad_scale: 16.0 +2023-02-06 05:09:06,632 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.778e+02 2.733e+02 3.382e+02 4.210e+02 1.523e+03, threshold=6.765e+02, percent-clipped=11.0 +2023-02-06 05:09:10,055 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57641.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:09:24,435 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57663.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:09:26,530 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57666.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:09:31,692 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57674.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:09:37,589 INFO [train.py:901] (3/4) Epoch 8, batch 1100, loss[loss=0.2938, simple_loss=0.3539, pruned_loss=0.1168, over 6875.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3269, pruned_loss=0.09448, over 1607719.46 frames. ], batch size: 71, lr: 9.83e-03, grad_scale: 16.0 +2023-02-06 05:10:05,404 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57721.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:10:08,891 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57726.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:10:10,258 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57728.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:10:10,700 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 05:10:12,767 INFO [train.py:901] (3/4) Epoch 8, batch 1150, loss[loss=0.2948, simple_loss=0.3687, pruned_loss=0.1105, over 8202.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.326, pruned_loss=0.09375, over 1611657.61 frames. ], batch size: 23, lr: 9.82e-03, grad_scale: 16.0 +2023-02-06 05:10:15,541 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.995e+02 2.752e+02 3.349e+02 4.211e+02 1.172e+03, threshold=6.698e+02, percent-clipped=4.0 +2023-02-06 05:10:26,720 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57751.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:10:26,782 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57751.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:10:28,142 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57753.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:10:32,806 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57760.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:10:48,142 INFO [train.py:901] (3/4) Epoch 8, batch 1200, loss[loss=0.2747, simple_loss=0.3494, pruned_loss=0.09998, over 8771.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3265, pruned_loss=0.09404, over 1616815.21 frames. ], batch size: 30, lr: 9.82e-03, grad_scale: 16.0 +2023-02-06 05:11:17,883 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57824.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:11:23,253 INFO [train.py:901] (3/4) Epoch 8, batch 1250, loss[loss=0.2783, simple_loss=0.3454, pruned_loss=0.1056, over 7125.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3274, pruned_loss=0.09512, over 1613865.81 frames. ], batch size: 71, lr: 9.81e-03, grad_scale: 16.0 +2023-02-06 05:11:25,902 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.793e+02 2.817e+02 3.577e+02 4.191e+02 8.690e+02, threshold=7.155e+02, percent-clipped=5.0 +2023-02-06 05:11:41,411 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57857.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:11:53,671 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57875.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:11:58,369 INFO [train.py:901] (3/4) Epoch 8, batch 1300, loss[loss=0.2773, simple_loss=0.3505, pruned_loss=0.102, over 8360.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3274, pruned_loss=0.09518, over 1610194.23 frames. ], batch size: 24, lr: 9.81e-03, grad_scale: 16.0 +2023-02-06 05:12:24,237 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57919.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:12:31,156 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.22 vs. limit=5.0 +2023-02-06 05:12:32,241 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57930.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:12:33,387 INFO [train.py:901] (3/4) Epoch 8, batch 1350, loss[loss=0.2234, simple_loss=0.2859, pruned_loss=0.08041, over 7270.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3261, pruned_loss=0.09414, over 1609947.92 frames. ], batch size: 16, lr: 9.80e-03, grad_scale: 16.0 +2023-02-06 05:12:36,124 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.963e+02 2.787e+02 3.281e+02 4.089e+02 1.129e+03, threshold=6.562e+02, percent-clipped=4.0 +2023-02-06 05:12:38,227 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57939.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:12:41,656 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57944.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:12:49,721 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57955.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:12:51,658 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6436, 1.8434, 2.2436, 1.7020, 0.9716, 2.2987, 0.3267, 1.2902], + device='cuda:3'), covar=tensor([0.2737, 0.1668, 0.0603, 0.2177, 0.5280, 0.0504, 0.3611, 0.2154], + device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0143, 0.0083, 0.0188, 0.0229, 0.0089, 0.0142, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:13:01,725 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57972.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:13:05,186 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57977.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:13:08,423 INFO [train.py:901] (3/4) Epoch 8, batch 1400, loss[loss=0.2761, simple_loss=0.3391, pruned_loss=0.1066, over 8337.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3275, pruned_loss=0.09488, over 1613880.53 frames. ], batch size: 48, lr: 9.80e-03, grad_scale: 16.0 +2023-02-06 05:13:23,091 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58002.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:13:41,319 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 05:13:42,234 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8657, 2.7611, 3.2255, 1.4563, 3.3525, 1.9064, 1.5039, 1.8338], + device='cuda:3'), covar=tensor([0.0478, 0.0197, 0.0113, 0.0388, 0.0181, 0.0505, 0.0519, 0.0317], + device='cuda:3'), in_proj_covar=tensor([0.0357, 0.0274, 0.0224, 0.0335, 0.0269, 0.0422, 0.0328, 0.0309], + device='cuda:3'), out_proj_covar=tensor([1.1115e-04, 8.3089e-05, 6.7083e-05, 1.0102e-04, 8.3117e-05, 1.3975e-04, + 1.0160e-04, 9.4776e-05], device='cuda:3') +2023-02-06 05:13:43,360 INFO [train.py:901] (3/4) Epoch 8, batch 1450, loss[loss=0.2557, simple_loss=0.3273, pruned_loss=0.09202, over 8605.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3269, pruned_loss=0.09383, over 1617045.21 frames. ], batch size: 39, lr: 9.80e-03, grad_scale: 16.0 +2023-02-06 05:13:46,045 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.540e+02 2.668e+02 3.298e+02 4.223e+02 1.032e+03, threshold=6.596e+02, percent-clipped=5.0 +2023-02-06 05:14:18,670 INFO [train.py:901] (3/4) Epoch 8, batch 1500, loss[loss=0.2005, simple_loss=0.2791, pruned_loss=0.0609, over 7817.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3273, pruned_loss=0.09416, over 1617224.26 frames. ], batch size: 20, lr: 9.79e-03, grad_scale: 16.0 +2023-02-06 05:14:28,081 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58095.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:14:43,585 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-06 05:14:52,818 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58131.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:14:53,269 INFO [train.py:901] (3/4) Epoch 8, batch 1550, loss[loss=0.2631, simple_loss=0.3392, pruned_loss=0.0935, over 8698.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3256, pruned_loss=0.09284, over 1615023.37 frames. ], batch size: 34, lr: 9.79e-03, grad_scale: 16.0 +2023-02-06 05:14:56,007 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.671e+02 2.601e+02 3.218e+02 3.979e+02 6.246e+02, threshold=6.435e+02, percent-clipped=0.0 +2023-02-06 05:15:10,021 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58156.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:15:12,046 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58159.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:15:14,755 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58162.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:15:27,769 INFO [train.py:901] (3/4) Epoch 8, batch 1600, loss[loss=0.2182, simple_loss=0.2896, pruned_loss=0.07336, over 8073.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3249, pruned_loss=0.09239, over 1615700.18 frames. ], batch size: 21, lr: 9.78e-03, grad_scale: 16.0 +2023-02-06 05:15:37,318 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58195.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:15:42,313 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-06 05:15:42,380 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-06 05:15:47,385 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58210.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:15:54,734 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58220.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:16:00,085 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58228.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:16:02,613 INFO [train.py:901] (3/4) Epoch 8, batch 1650, loss[loss=0.1862, simple_loss=0.2667, pruned_loss=0.05282, over 7788.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.326, pruned_loss=0.09307, over 1612620.11 frames. ], batch size: 19, lr: 9.78e-03, grad_scale: 16.0 +2023-02-06 05:16:05,270 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.744e+02 2.785e+02 3.241e+02 4.331e+02 1.468e+03, threshold=6.482e+02, percent-clipped=4.0 +2023-02-06 05:16:16,844 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58253.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:16:37,569 INFO [train.py:901] (3/4) Epoch 8, batch 1700, loss[loss=0.309, simple_loss=0.3703, pruned_loss=0.1238, over 8463.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.324, pruned_loss=0.09258, over 1608447.92 frames. ], batch size: 25, lr: 9.78e-03, grad_scale: 16.0 +2023-02-06 05:16:57,472 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-06 05:17:11,722 INFO [train.py:901] (3/4) Epoch 8, batch 1750, loss[loss=0.2246, simple_loss=0.2972, pruned_loss=0.07599, over 8087.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3247, pruned_loss=0.09301, over 1609898.00 frames. ], batch size: 21, lr: 9.77e-03, grad_scale: 16.0 +2023-02-06 05:17:15,044 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.673e+02 2.752e+02 3.204e+02 3.949e+02 8.384e+02, threshold=6.409e+02, percent-clipped=4.0 +2023-02-06 05:17:45,739 INFO [train.py:901] (3/4) Epoch 8, batch 1800, loss[loss=0.2492, simple_loss=0.3278, pruned_loss=0.08531, over 8324.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3267, pruned_loss=0.09373, over 1616180.56 frames. ], batch size: 25, lr: 9.77e-03, grad_scale: 16.0 +2023-02-06 05:17:51,960 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4752, 1.8437, 3.3985, 1.3039, 2.3996, 2.0170, 1.6165, 2.1090], + device='cuda:3'), covar=tensor([0.1872, 0.2089, 0.0610, 0.3584, 0.1285, 0.2405, 0.1858, 0.2020], + device='cuda:3'), in_proj_covar=tensor([0.0481, 0.0482, 0.0530, 0.0559, 0.0609, 0.0534, 0.0457, 0.0595], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:18:16,053 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4481, 4.3909, 3.8620, 1.8036, 3.8963, 3.9741, 4.1652, 3.5599], + device='cuda:3'), covar=tensor([0.0898, 0.0648, 0.1127, 0.5773, 0.0899, 0.1051, 0.1198, 0.0952], + device='cuda:3'), in_proj_covar=tensor([0.0423, 0.0329, 0.0359, 0.0451, 0.0352, 0.0329, 0.0338, 0.0290], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:18:21,313 INFO [train.py:901] (3/4) Epoch 8, batch 1850, loss[loss=0.2387, simple_loss=0.3085, pruned_loss=0.0844, over 7539.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3264, pruned_loss=0.09339, over 1612446.04 frames. ], batch size: 18, lr: 9.76e-03, grad_scale: 16.0 +2023-02-06 05:18:24,003 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.582e+02 2.956e+02 3.603e+02 4.636e+02 8.044e+02, threshold=7.207e+02, percent-clipped=5.0 +2023-02-06 05:18:45,037 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58466.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:18:55,853 INFO [train.py:901] (3/4) Epoch 8, batch 1900, loss[loss=0.2716, simple_loss=0.3284, pruned_loss=0.1073, over 7542.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3261, pruned_loss=0.09328, over 1610496.70 frames. ], batch size: 18, lr: 9.76e-03, grad_scale: 16.0 +2023-02-06 05:19:02,004 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58491.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:19:10,024 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58503.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:19:12,718 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58506.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:19:19,281 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 05:19:24,512 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4994, 3.2020, 2.3017, 4.0840, 1.8728, 2.2006, 2.1433, 3.2246], + device='cuda:3'), covar=tensor([0.0772, 0.0706, 0.1096, 0.0262, 0.1296, 0.1489, 0.1312, 0.0748], + device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0229, 0.0265, 0.0214, 0.0234, 0.0265, 0.0265, 0.0235], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 05:19:30,166 INFO [train.py:901] (3/4) Epoch 8, batch 1950, loss[loss=0.233, simple_loss=0.3157, pruned_loss=0.07512, over 8350.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3272, pruned_loss=0.09389, over 1613154.62 frames. ], batch size: 24, lr: 9.75e-03, grad_scale: 16.0 +2023-02-06 05:19:30,815 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 05:19:32,812 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.635e+02 2.699e+02 3.417e+02 4.103e+02 8.210e+02, threshold=6.834e+02, percent-clipped=5.0 +2023-02-06 05:19:41,935 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 +2023-02-06 05:19:50,869 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 05:19:59,410 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 +2023-02-06 05:20:04,893 INFO [train.py:901] (3/4) Epoch 8, batch 2000, loss[loss=0.2119, simple_loss=0.2852, pruned_loss=0.06925, over 7248.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3258, pruned_loss=0.09356, over 1610496.42 frames. ], batch size: 16, lr: 9.75e-03, grad_scale: 8.0 +2023-02-06 05:20:25,673 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.22 vs. limit=5.0 +2023-02-06 05:20:29,402 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58618.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:20:31,400 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58621.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:20:39,655 INFO [train.py:901] (3/4) Epoch 8, batch 2050, loss[loss=0.3171, simple_loss=0.3757, pruned_loss=0.1293, over 8506.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3265, pruned_loss=0.09352, over 1616679.48 frames. ], batch size: 26, lr: 9.75e-03, grad_scale: 8.0 +2023-02-06 05:20:42,933 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.764e+02 2.785e+02 3.396e+02 4.687e+02 1.585e+03, threshold=6.792e+02, percent-clipped=4.0 +2023-02-06 05:21:13,656 INFO [train.py:901] (3/4) Epoch 8, batch 2100, loss[loss=0.2402, simple_loss=0.3055, pruned_loss=0.08748, over 7696.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3274, pruned_loss=0.09405, over 1616528.79 frames. ], batch size: 18, lr: 9.74e-03, grad_scale: 8.0 +2023-02-06 05:21:25,817 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58699.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:21:34,548 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.6540, 1.2020, 3.8472, 1.4429, 3.2155, 3.1889, 3.4360, 3.3268], + device='cuda:3'), covar=tensor([0.0589, 0.4448, 0.0560, 0.3152, 0.1274, 0.0887, 0.0593, 0.0701], + device='cuda:3'), in_proj_covar=tensor([0.0406, 0.0532, 0.0506, 0.0476, 0.0543, 0.0459, 0.0457, 0.0513], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 05:21:47,831 INFO [train.py:901] (3/4) Epoch 8, batch 2150, loss[loss=0.2604, simple_loss=0.3358, pruned_loss=0.09243, over 8501.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3279, pruned_loss=0.09393, over 1621613.51 frames. ], batch size: 28, lr: 9.74e-03, grad_scale: 8.0 +2023-02-06 05:21:51,089 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.839e+02 2.818e+02 3.372e+02 4.104e+02 8.704e+02, threshold=6.743e+02, percent-clipped=2.0 +2023-02-06 05:21:53,305 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3402, 1.5031, 1.4365, 1.8461, 0.8410, 1.2159, 1.3553, 1.4543], + device='cuda:3'), covar=tensor([0.1074, 0.0941, 0.1203, 0.0608, 0.1322, 0.1787, 0.0990, 0.0866], + device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0227, 0.0265, 0.0214, 0.0233, 0.0264, 0.0263, 0.0234], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 05:22:23,671 INFO [train.py:901] (3/4) Epoch 8, batch 2200, loss[loss=0.2857, simple_loss=0.3558, pruned_loss=0.1078, over 8636.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3275, pruned_loss=0.09408, over 1618824.43 frames. ], batch size: 49, lr: 9.73e-03, grad_scale: 8.0 +2023-02-06 05:22:31,534 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58793.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:22:58,800 INFO [train.py:901] (3/4) Epoch 8, batch 2250, loss[loss=0.206, simple_loss=0.2834, pruned_loss=0.06426, over 7711.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3264, pruned_loss=0.09319, over 1620137.95 frames. ], batch size: 18, lr: 9.73e-03, grad_scale: 8.0 +2023-02-06 05:23:02,318 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.890e+02 2.600e+02 3.138e+02 4.259e+02 8.800e+02, threshold=6.276e+02, percent-clipped=5.0 +2023-02-06 05:23:04,491 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.0908, 1.3294, 3.2043, 0.9988, 2.7671, 2.6668, 2.8593, 2.7899], + device='cuda:3'), covar=tensor([0.0686, 0.3577, 0.0757, 0.3286, 0.1520, 0.1059, 0.0726, 0.0819], + device='cuda:3'), in_proj_covar=tensor([0.0400, 0.0524, 0.0500, 0.0470, 0.0540, 0.0454, 0.0455, 0.0506], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 05:23:25,723 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4600, 3.0731, 2.1620, 3.9302, 1.9505, 1.8695, 2.0991, 3.1780], + device='cuda:3'), covar=tensor([0.0955, 0.0952, 0.1236, 0.0342, 0.1402, 0.1847, 0.1541, 0.1046], + device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0228, 0.0267, 0.0215, 0.0233, 0.0266, 0.0267, 0.0235], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 05:23:29,264 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58874.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:23:31,267 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58877.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:23:34,473 INFO [train.py:901] (3/4) Epoch 8, batch 2300, loss[loss=0.255, simple_loss=0.3256, pruned_loss=0.09216, over 8358.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3272, pruned_loss=0.09316, over 1621484.42 frames. ], batch size: 26, lr: 9.73e-03, grad_scale: 8.0 +2023-02-06 05:23:36,059 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6211, 1.7838, 1.8964, 1.5619, 0.9780, 1.9972, 0.1859, 1.2016], + device='cuda:3'), covar=tensor([0.2725, 0.1688, 0.0718, 0.2193, 0.5277, 0.0721, 0.4028, 0.2171], + device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0142, 0.0084, 0.0194, 0.0231, 0.0088, 0.0149, 0.0143], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:23:46,516 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58899.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:23:48,587 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58902.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:23:50,664 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9630, 1.8531, 1.7634, 1.9283, 1.1833, 1.6423, 2.2000, 2.1988], + device='cuda:3'), covar=tensor([0.0460, 0.1035, 0.1630, 0.1187, 0.0542, 0.1307, 0.0590, 0.0497], + device='cuda:3'), in_proj_covar=tensor([0.0111, 0.0161, 0.0199, 0.0166, 0.0110, 0.0169, 0.0123, 0.0142], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 05:24:00,183 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3633, 1.4314, 1.2953, 1.8567, 0.8179, 1.1433, 1.3314, 1.4779], + device='cuda:3'), covar=tensor([0.1031, 0.0935, 0.1434, 0.0578, 0.1251, 0.1757, 0.0908, 0.0860], + device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0229, 0.0267, 0.0215, 0.0233, 0.0267, 0.0266, 0.0236], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 05:24:03,549 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58924.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:24:09,485 INFO [train.py:901] (3/4) Epoch 8, batch 2350, loss[loss=0.2093, simple_loss=0.2955, pruned_loss=0.06152, over 8031.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.326, pruned_loss=0.09182, over 1618938.85 frames. ], batch size: 22, lr: 9.72e-03, grad_scale: 8.0 +2023-02-06 05:24:12,940 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.625e+02 2.611e+02 3.221e+02 3.780e+02 8.999e+02, threshold=6.441e+02, percent-clipped=2.0 +2023-02-06 05:24:44,048 INFO [train.py:901] (3/4) Epoch 8, batch 2400, loss[loss=0.3105, simple_loss=0.3709, pruned_loss=0.125, over 8366.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3252, pruned_loss=0.09144, over 1618716.54 frames. ], batch size: 24, lr: 9.72e-03, grad_scale: 8.0 +2023-02-06 05:25:18,654 INFO [train.py:901] (3/4) Epoch 8, batch 2450, loss[loss=0.2505, simple_loss=0.332, pruned_loss=0.08447, over 8288.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3246, pruned_loss=0.0911, over 1617996.64 frames. ], batch size: 23, lr: 9.71e-03, grad_scale: 8.0 +2023-02-06 05:25:21,882 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.833e+02 3.001e+02 3.706e+02 4.542e+02 9.599e+02, threshold=7.413e+02, percent-clipped=3.0 +2023-02-06 05:25:26,108 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59043.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:25:29,906 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 +2023-02-06 05:25:52,506 INFO [train.py:901] (3/4) Epoch 8, batch 2500, loss[loss=0.2497, simple_loss=0.3126, pruned_loss=0.09341, over 8083.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3256, pruned_loss=0.09191, over 1619819.30 frames. ], batch size: 21, lr: 9.71e-03, grad_scale: 8.0 +2023-02-06 05:26:27,547 INFO [train.py:901] (3/4) Epoch 8, batch 2550, loss[loss=0.2795, simple_loss=0.3587, pruned_loss=0.1001, over 8324.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3257, pruned_loss=0.09195, over 1622319.78 frames. ], batch size: 25, lr: 9.71e-03, grad_scale: 8.0 +2023-02-06 05:26:30,875 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.667e+02 2.676e+02 3.180e+02 4.175e+02 9.807e+02, threshold=6.360e+02, percent-clipped=4.0 +2023-02-06 05:26:30,965 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59137.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:26:45,976 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59158.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:27:03,035 INFO [train.py:901] (3/4) Epoch 8, batch 2600, loss[loss=0.2211, simple_loss=0.2922, pruned_loss=0.07494, over 7931.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3244, pruned_loss=0.09144, over 1618122.63 frames. ], batch size: 20, lr: 9.70e-03, grad_scale: 8.0 +2023-02-06 05:27:03,171 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59182.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:27:19,766 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2023-02-06 05:27:38,230 INFO [train.py:901] (3/4) Epoch 8, batch 2650, loss[loss=0.3689, simple_loss=0.4047, pruned_loss=0.1665, over 8549.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3252, pruned_loss=0.09246, over 1612203.35 frames. ], batch size: 39, lr: 9.70e-03, grad_scale: 8.0 +2023-02-06 05:27:41,652 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.517e+02 2.757e+02 3.213e+02 4.207e+02 1.360e+03, threshold=6.426e+02, percent-clipped=6.0 +2023-02-06 05:27:51,997 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59252.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:28:03,088 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59268.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:28:05,173 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2759, 1.4049, 4.2824, 1.9319, 2.4633, 4.9700, 4.8406, 4.3013], + device='cuda:3'), covar=tensor([0.1093, 0.1755, 0.0298, 0.1956, 0.1017, 0.0201, 0.0330, 0.0582], + device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0284, 0.0242, 0.0277, 0.0249, 0.0223, 0.0293, 0.0286], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 05:28:12,220 INFO [train.py:901] (3/4) Epoch 8, batch 2700, loss[loss=0.2334, simple_loss=0.2997, pruned_loss=0.08358, over 7513.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3242, pruned_loss=0.09165, over 1613451.89 frames. ], batch size: 18, lr: 9.69e-03, grad_scale: 8.0 +2023-02-06 05:28:46,767 INFO [train.py:901] (3/4) Epoch 8, batch 2750, loss[loss=0.2468, simple_loss=0.3138, pruned_loss=0.08991, over 8133.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3241, pruned_loss=0.09193, over 1613097.32 frames. ], batch size: 22, lr: 9.69e-03, grad_scale: 8.0 +2023-02-06 05:28:50,092 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.941e+02 2.846e+02 3.367e+02 4.274e+02 9.837e+02, threshold=6.735e+02, percent-clipped=6.0 +2023-02-06 05:28:51,051 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4070, 1.9600, 3.1790, 2.5189, 2.7088, 2.1241, 1.5944, 1.3997], + device='cuda:3'), covar=tensor([0.2891, 0.3421, 0.0799, 0.1934, 0.1577, 0.1691, 0.1482, 0.3473], + device='cuda:3'), in_proj_covar=tensor([0.0843, 0.0796, 0.0686, 0.0790, 0.0887, 0.0736, 0.0672, 0.0721], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:29:14,220 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.99 vs. limit=5.0 +2023-02-06 05:29:22,654 INFO [train.py:901] (3/4) Epoch 8, batch 2800, loss[loss=0.2722, simple_loss=0.3235, pruned_loss=0.1105, over 7547.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.325, pruned_loss=0.09176, over 1617171.74 frames. ], batch size: 18, lr: 9.69e-03, grad_scale: 8.0 +2023-02-06 05:29:23,474 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59383.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:29:36,338 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5467, 2.3872, 2.7482, 2.1087, 1.6953, 2.7259, 0.9734, 2.1664], + device='cuda:3'), covar=tensor([0.2217, 0.1555, 0.0570, 0.2520, 0.4500, 0.0544, 0.4033, 0.1829], + device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0146, 0.0083, 0.0196, 0.0232, 0.0089, 0.0152, 0.0144], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:29:44,998 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59414.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:29:56,659 INFO [train.py:901] (3/4) Epoch 8, batch 2850, loss[loss=0.3103, simple_loss=0.3679, pruned_loss=0.1263, over 8277.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3251, pruned_loss=0.09191, over 1620936.22 frames. ], batch size: 23, lr: 9.68e-03, grad_scale: 8.0 +2023-02-06 05:30:00,145 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.577e+02 2.974e+02 3.773e+02 5.956e+02, threshold=5.948e+02, percent-clipped=0.0 +2023-02-06 05:30:01,810 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59439.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:30:32,530 INFO [train.py:901] (3/4) Epoch 8, batch 2900, loss[loss=0.2479, simple_loss=0.3284, pruned_loss=0.08375, over 8132.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3255, pruned_loss=0.09212, over 1616634.77 frames. ], batch size: 22, lr: 9.68e-03, grad_scale: 8.0 +2023-02-06 05:30:51,217 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59508.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:30:59,183 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59520.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:31:03,080 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59526.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:31:04,363 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 05:31:07,552 INFO [train.py:901] (3/4) Epoch 8, batch 2950, loss[loss=0.3744, simple_loss=0.4182, pruned_loss=0.1653, over 8362.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3253, pruned_loss=0.09223, over 1614687.04 frames. ], batch size: 24, lr: 9.67e-03, grad_scale: 8.0 +2023-02-06 05:31:08,346 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59533.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:31:10,780 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.890e+02 2.847e+02 3.468e+02 5.057e+02 9.591e+02, threshold=6.936e+02, percent-clipped=13.0 +2023-02-06 05:31:42,181 INFO [train.py:901] (3/4) Epoch 8, batch 3000, loss[loss=0.2188, simple_loss=0.2978, pruned_loss=0.06986, over 7972.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3256, pruned_loss=0.09284, over 1611533.13 frames. ], batch size: 21, lr: 9.67e-03, grad_scale: 8.0 +2023-02-06 05:31:42,181 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 05:31:54,427 INFO [train.py:935] (3/4) Epoch 8, validation: loss=0.2021, simple_loss=0.3001, pruned_loss=0.05199, over 944034.00 frames. +2023-02-06 05:31:54,428 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 05:32:30,902 INFO [train.py:901] (3/4) Epoch 8, batch 3050, loss[loss=0.2855, simple_loss=0.3528, pruned_loss=0.1092, over 8356.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3247, pruned_loss=0.09284, over 1602419.57 frames. ], batch size: 24, lr: 9.67e-03, grad_scale: 8.0 +2023-02-06 05:32:34,247 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.666e+02 2.623e+02 3.324e+02 4.059e+02 7.396e+02, threshold=6.648e+02, percent-clipped=1.0 +2023-02-06 05:32:35,805 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59639.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:32:37,138 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59641.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:32:52,396 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59664.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:33:05,113 INFO [train.py:901] (3/4) Epoch 8, batch 3100, loss[loss=0.2794, simple_loss=0.347, pruned_loss=0.1059, over 8632.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3268, pruned_loss=0.09401, over 1607313.79 frames. ], batch size: 34, lr: 9.66e-03, grad_scale: 8.0 +2023-02-06 05:33:40,029 INFO [train.py:901] (3/4) Epoch 8, batch 3150, loss[loss=0.1932, simple_loss=0.2664, pruned_loss=0.05993, over 5986.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.326, pruned_loss=0.0936, over 1608147.79 frames. ], batch size: 13, lr: 9.66e-03, grad_scale: 8.0 +2023-02-06 05:33:43,224 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.052e+02 2.894e+02 3.427e+02 4.526e+02 8.691e+02, threshold=6.853e+02, percent-clipped=4.0 +2023-02-06 05:34:14,629 INFO [train.py:901] (3/4) Epoch 8, batch 3200, loss[loss=0.2724, simple_loss=0.341, pruned_loss=0.102, over 8455.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3254, pruned_loss=0.09289, over 1611716.74 frames. ], batch size: 27, lr: 9.65e-03, grad_scale: 8.0 +2023-02-06 05:34:50,982 INFO [train.py:901] (3/4) Epoch 8, batch 3250, loss[loss=0.2188, simple_loss=0.3058, pruned_loss=0.06587, over 8136.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3252, pruned_loss=0.09261, over 1611669.75 frames. ], batch size: 22, lr: 9.65e-03, grad_scale: 8.0 +2023-02-06 05:34:53,137 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.0366, 2.9880, 3.0345, 2.3398, 1.9428, 3.3361, 0.6831, 2.2998], + device='cuda:3'), covar=tensor([0.2269, 0.1559, 0.1091, 0.2783, 0.4723, 0.0781, 0.4785, 0.2439], + device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0144, 0.0084, 0.0191, 0.0231, 0.0090, 0.0151, 0.0143], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:34:54,318 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.704e+02 2.545e+02 3.201e+02 4.295e+02 9.179e+02, threshold=6.402e+02, percent-clipped=6.0 +2023-02-06 05:35:13,114 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59864.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:35:14,548 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59866.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:35:23,456 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4569, 2.0054, 3.2366, 2.3981, 2.7436, 2.1005, 1.7111, 1.4060], + device='cuda:3'), covar=tensor([0.2983, 0.3379, 0.0737, 0.2004, 0.1615, 0.1927, 0.1643, 0.3367], + device='cuda:3'), in_proj_covar=tensor([0.0850, 0.0811, 0.0689, 0.0795, 0.0895, 0.0746, 0.0684, 0.0729], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:35:25,230 INFO [train.py:901] (3/4) Epoch 8, batch 3300, loss[loss=0.1652, simple_loss=0.2448, pruned_loss=0.04283, over 6841.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3253, pruned_loss=0.09262, over 1614328.73 frames. ], batch size: 15, lr: 9.65e-03, grad_scale: 8.0 +2023-02-06 05:35:31,615 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 +2023-02-06 05:35:35,293 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59897.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:35:41,747 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59907.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:35:48,082 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 05:35:52,455 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59922.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:35:59,528 INFO [train.py:901] (3/4) Epoch 8, batch 3350, loss[loss=0.2388, simple_loss=0.3023, pruned_loss=0.08762, over 7921.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3258, pruned_loss=0.093, over 1620084.47 frames. ], batch size: 20, lr: 9.64e-03, grad_scale: 8.0 +2023-02-06 05:36:02,898 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.805e+02 2.795e+02 3.400e+02 4.166e+02 8.824e+02, threshold=6.801e+02, percent-clipped=5.0 +2023-02-06 05:36:31,812 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59979.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:36:33,699 INFO [train.py:901] (3/4) Epoch 8, batch 3400, loss[loss=0.2526, simple_loss=0.3233, pruned_loss=0.09098, over 7522.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3248, pruned_loss=0.09244, over 1616734.63 frames. ], batch size: 18, lr: 9.64e-03, grad_scale: 8.0 +2023-02-06 05:36:48,498 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7181, 2.2693, 3.4338, 2.8055, 3.0488, 2.3229, 2.0568, 2.1647], + device='cuda:3'), covar=tensor([0.2113, 0.3012, 0.0795, 0.1681, 0.1247, 0.1602, 0.1172, 0.2566], + device='cuda:3'), in_proj_covar=tensor([0.0848, 0.0815, 0.0691, 0.0793, 0.0897, 0.0744, 0.0682, 0.0727], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:37:09,558 INFO [train.py:901] (3/4) Epoch 8, batch 3450, loss[loss=0.2497, simple_loss=0.323, pruned_loss=0.08819, over 8471.00 frames. ], tot_loss[loss=0.255, simple_loss=0.325, pruned_loss=0.09252, over 1612588.78 frames. ], batch size: 25, lr: 9.63e-03, grad_scale: 8.0 +2023-02-06 05:37:12,878 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.731e+02 2.668e+02 3.106e+02 3.891e+02 9.201e+02, threshold=6.211e+02, percent-clipped=2.0 +2023-02-06 05:37:27,906 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60058.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:37:35,081 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0338, 1.4727, 1.5324, 1.3011, 1.0593, 1.3121, 1.5498, 1.6214], + device='cuda:3'), covar=tensor([0.0582, 0.1287, 0.1781, 0.1408, 0.0600, 0.1527, 0.0703, 0.0596], + device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0162, 0.0198, 0.0164, 0.0111, 0.0169, 0.0123, 0.0142], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 05:37:36,417 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60070.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:37:44,252 INFO [train.py:901] (3/4) Epoch 8, batch 3500, loss[loss=0.25, simple_loss=0.3257, pruned_loss=0.08712, over 8239.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3246, pruned_loss=0.09257, over 1611983.20 frames. ], batch size: 24, lr: 9.63e-03, grad_scale: 8.0 +2023-02-06 05:37:45,090 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60083.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:37:53,903 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60096.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:37:58,640 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 +2023-02-06 05:38:02,911 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 05:38:18,645 INFO [train.py:901] (3/4) Epoch 8, batch 3550, loss[loss=0.2861, simple_loss=0.3502, pruned_loss=0.111, over 8338.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3255, pruned_loss=0.09339, over 1612532.12 frames. ], batch size: 26, lr: 9.63e-03, grad_scale: 8.0 +2023-02-06 05:38:22,095 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.667e+02 2.844e+02 3.449e+02 4.512e+02 7.529e+02, threshold=6.898e+02, percent-clipped=5.0 +2023-02-06 05:38:41,391 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.0811, 1.2057, 4.2515, 1.5297, 3.7316, 3.5509, 3.7980, 3.6673], + device='cuda:3'), covar=tensor([0.0477, 0.3798, 0.0448, 0.2985, 0.1051, 0.0747, 0.0517, 0.0628], + device='cuda:3'), in_proj_covar=tensor([0.0415, 0.0535, 0.0514, 0.0478, 0.0552, 0.0465, 0.0460, 0.0514], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 05:38:54,374 INFO [train.py:901] (3/4) Epoch 8, batch 3600, loss[loss=0.2444, simple_loss=0.323, pruned_loss=0.08295, over 8514.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3254, pruned_loss=0.09281, over 1613259.85 frames. ], batch size: 26, lr: 9.62e-03, grad_scale: 8.0 +2023-02-06 05:38:59,446 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8870, 1.8598, 1.8958, 1.7589, 1.1018, 1.8269, 2.4061, 2.7433], + device='cuda:3'), covar=tensor([0.0470, 0.1127, 0.1621, 0.1239, 0.0577, 0.1435, 0.0573, 0.0434], + device='cuda:3'), in_proj_covar=tensor([0.0110, 0.0162, 0.0199, 0.0164, 0.0110, 0.0169, 0.0123, 0.0142], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 05:39:03,692 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.59 vs. limit=5.0 +2023-02-06 05:39:14,697 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60210.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:39:18,773 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60216.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:39:29,812 INFO [train.py:901] (3/4) Epoch 8, batch 3650, loss[loss=0.3384, simple_loss=0.393, pruned_loss=0.1419, over 8601.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3268, pruned_loss=0.09426, over 1610246.93 frames. ], batch size: 49, lr: 9.62e-03, grad_scale: 8.0 +2023-02-06 05:39:32,082 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60235.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:39:33,128 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.875e+02 2.702e+02 3.457e+02 4.155e+02 9.631e+02, threshold=6.915e+02, percent-clipped=4.0 +2023-02-06 05:39:42,753 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60251.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:39:48,812 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60260.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:40:02,764 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 05:40:04,774 INFO [train.py:901] (3/4) Epoch 8, batch 3700, loss[loss=0.2319, simple_loss=0.3108, pruned_loss=0.07654, over 8332.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3257, pruned_loss=0.09368, over 1610655.21 frames. ], batch size: 26, lr: 9.61e-03, grad_scale: 8.0 +2023-02-06 05:40:34,564 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60325.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:40:39,103 INFO [train.py:901] (3/4) Epoch 8, batch 3750, loss[loss=0.2338, simple_loss=0.315, pruned_loss=0.07629, over 8467.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3258, pruned_loss=0.09319, over 1613213.52 frames. ], batch size: 25, lr: 9.61e-03, grad_scale: 8.0 +2023-02-06 05:40:43,053 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.515e+02 2.685e+02 3.295e+02 3.882e+02 8.274e+02, threshold=6.589e+02, percent-clipped=2.0 +2023-02-06 05:41:02,727 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60366.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:41:13,769 INFO [train.py:901] (3/4) Epoch 8, batch 3800, loss[loss=0.2626, simple_loss=0.3245, pruned_loss=0.1004, over 7817.00 frames. ], tot_loss[loss=0.257, simple_loss=0.327, pruned_loss=0.09346, over 1611287.73 frames. ], batch size: 20, lr: 9.61e-03, grad_scale: 8.0 +2023-02-06 05:41:28,152 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60402.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:41:36,465 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60414.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:41:45,879 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60427.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:41:49,069 INFO [train.py:901] (3/4) Epoch 8, batch 3850, loss[loss=0.3296, simple_loss=0.3944, pruned_loss=0.1324, over 8568.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3257, pruned_loss=0.09243, over 1609661.97 frames. ], batch size: 34, lr: 9.60e-03, grad_scale: 8.0 +2023-02-06 05:41:52,287 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.593e+02 2.691e+02 3.271e+02 4.212e+02 1.032e+03, threshold=6.541e+02, percent-clipped=5.0 +2023-02-06 05:41:54,201 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60440.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:42:08,499 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 05:42:12,035 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6785, 4.7127, 4.0644, 1.8184, 3.9997, 4.0807, 4.2531, 3.9225], + device='cuda:3'), covar=tensor([0.0650, 0.0469, 0.0918, 0.5439, 0.0819, 0.0993, 0.1208, 0.0773], + device='cuda:3'), in_proj_covar=tensor([0.0417, 0.0326, 0.0351, 0.0449, 0.0351, 0.0329, 0.0336, 0.0284], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:42:22,376 INFO [train.py:901] (3/4) Epoch 8, batch 3900, loss[loss=0.2658, simple_loss=0.335, pruned_loss=0.09836, over 8357.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3255, pruned_loss=0.09258, over 1611514.90 frames. ], batch size: 24, lr: 9.60e-03, grad_scale: 8.0 +2023-02-06 05:42:41,578 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 +2023-02-06 05:42:46,591 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60517.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:42:55,303 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60529.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:42:57,131 INFO [train.py:901] (3/4) Epoch 8, batch 3950, loss[loss=0.2813, simple_loss=0.3532, pruned_loss=0.1047, over 8252.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3248, pruned_loss=0.09234, over 1610496.21 frames. ], batch size: 24, lr: 9.59e-03, grad_scale: 8.0 +2023-02-06 05:43:00,413 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.833e+02 2.562e+02 3.362e+02 4.082e+02 8.516e+02, threshold=6.724e+02, percent-clipped=2.0 +2023-02-06 05:43:03,872 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60542.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:43:13,331 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60555.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:43:16,391 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60560.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:43:31,328 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60581.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:43:31,760 INFO [train.py:901] (3/4) Epoch 8, batch 4000, loss[loss=0.2315, simple_loss=0.3077, pruned_loss=0.07771, over 7937.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3263, pruned_loss=0.09289, over 1613560.13 frames. ], batch size: 20, lr: 9.59e-03, grad_scale: 16.0 +2023-02-06 05:43:48,277 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60606.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:43:50,230 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8784, 2.4296, 4.6566, 1.4012, 3.2781, 2.2835, 1.9321, 2.8317], + device='cuda:3'), covar=tensor([0.1329, 0.1749, 0.0607, 0.3294, 0.1160, 0.2264, 0.1438, 0.1994], + device='cuda:3'), in_proj_covar=tensor([0.0469, 0.0479, 0.0525, 0.0552, 0.0595, 0.0533, 0.0452, 0.0587], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:43:58,368 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7874, 2.1289, 4.6609, 1.2505, 3.1942, 2.1804, 1.7559, 2.8431], + device='cuda:3'), covar=tensor([0.1569, 0.2311, 0.0601, 0.3676, 0.1470, 0.2731, 0.1807, 0.2247], + device='cuda:3'), in_proj_covar=tensor([0.0470, 0.0481, 0.0526, 0.0554, 0.0598, 0.0536, 0.0454, 0.0589], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:43:59,795 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60622.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:44:06,269 INFO [train.py:901] (3/4) Epoch 8, batch 4050, loss[loss=0.2678, simple_loss=0.3555, pruned_loss=0.08999, over 8358.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3271, pruned_loss=0.09326, over 1616301.96 frames. ], batch size: 24, lr: 9.59e-03, grad_scale: 16.0 +2023-02-06 05:44:09,650 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.751e+02 2.835e+02 3.722e+02 4.462e+02 8.493e+02, threshold=7.445e+02, percent-clipped=1.0 +2023-02-06 05:44:17,204 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60647.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:44:36,991 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60675.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:44:39,700 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2948, 1.5410, 1.5559, 0.7855, 1.7239, 1.2243, 0.2574, 1.5196], + device='cuda:3'), covar=tensor([0.0254, 0.0164, 0.0124, 0.0228, 0.0170, 0.0423, 0.0393, 0.0132], + device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0276, 0.0226, 0.0336, 0.0267, 0.0427, 0.0329, 0.0304], + device='cuda:3'), out_proj_covar=tensor([1.0820e-04, 8.2460e-05, 6.7271e-05, 1.0027e-04, 8.1688e-05, 1.4057e-04, + 1.0064e-04, 9.2114e-05], device='cuda:3') +2023-02-06 05:44:41,543 INFO [train.py:901] (3/4) Epoch 8, batch 4100, loss[loss=0.2393, simple_loss=0.3094, pruned_loss=0.08462, over 7691.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3261, pruned_loss=0.09246, over 1615172.42 frames. ], batch size: 18, lr: 9.58e-03, grad_scale: 16.0 +2023-02-06 05:44:50,520 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2178, 1.1503, 1.2268, 1.1712, 0.8052, 1.2973, 0.0329, 1.0175], + device='cuda:3'), covar=tensor([0.2559, 0.2341, 0.0858, 0.1651, 0.4985, 0.0759, 0.4031, 0.2113], + device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0146, 0.0086, 0.0196, 0.0235, 0.0091, 0.0153, 0.0148], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:45:13,057 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4410, 1.8462, 1.4093, 2.3492, 1.1552, 1.1542, 1.6321, 1.9410], + device='cuda:3'), covar=tensor([0.1249, 0.1085, 0.1673, 0.0649, 0.1433, 0.2217, 0.1300, 0.0900], + device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0232, 0.0267, 0.0219, 0.0228, 0.0264, 0.0267, 0.0233], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 05:45:16,315 INFO [train.py:901] (3/4) Epoch 8, batch 4150, loss[loss=0.3012, simple_loss=0.3593, pruned_loss=0.1216, over 8362.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3257, pruned_loss=0.09228, over 1614893.58 frames. ], batch size: 24, lr: 9.58e-03, grad_scale: 16.0 +2023-02-06 05:45:19,086 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60736.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:45:19,619 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.991e+02 2.904e+02 3.574e+02 4.093e+02 8.234e+02, threshold=7.147e+02, percent-clipped=2.0 +2023-02-06 05:45:45,319 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60773.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:45:51,060 INFO [train.py:901] (3/4) Epoch 8, batch 4200, loss[loss=0.2426, simple_loss=0.3176, pruned_loss=0.08382, over 8342.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3263, pruned_loss=0.0927, over 1615433.61 frames. ], batch size: 26, lr: 9.57e-03, grad_scale: 8.0 +2023-02-06 05:45:54,023 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60785.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:46:02,594 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60798.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:46:02,614 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60798.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:46:07,985 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 05:46:10,882 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60810.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:46:11,541 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60811.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:46:20,182 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60823.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:46:25,972 INFO [train.py:901] (3/4) Epoch 8, batch 4250, loss[loss=0.244, simple_loss=0.3252, pruned_loss=0.08141, over 8091.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3259, pruned_loss=0.09229, over 1616817.72 frames. ], batch size: 21, lr: 9.57e-03, grad_scale: 8.0 +2023-02-06 05:46:28,909 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60836.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:46:30,099 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.898e+02 2.807e+02 3.546e+02 4.515e+02 1.213e+03, threshold=7.092e+02, percent-clipped=3.0 +2023-02-06 05:46:30,813 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 05:46:51,904 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8116, 3.7886, 3.3493, 1.7609, 3.3390, 3.3600, 3.4664, 2.9853], + device='cuda:3'), covar=tensor([0.0979, 0.0658, 0.1245, 0.4642, 0.0911, 0.0941, 0.1393, 0.1081], + device='cuda:3'), in_proj_covar=tensor([0.0422, 0.0330, 0.0353, 0.0454, 0.0349, 0.0328, 0.0338, 0.0290], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:47:01,075 INFO [train.py:901] (3/4) Epoch 8, batch 4300, loss[loss=0.2316, simple_loss=0.3088, pruned_loss=0.07716, over 8290.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3272, pruned_loss=0.09318, over 1614380.55 frames. ], batch size: 23, lr: 9.57e-03, grad_scale: 8.0 +2023-02-06 05:47:03,344 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8329, 1.8725, 2.1516, 1.6101, 1.2926, 2.1670, 0.2675, 1.3464], + device='cuda:3'), covar=tensor([0.3078, 0.1868, 0.0562, 0.2493, 0.5133, 0.0612, 0.4151, 0.2335], + device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0149, 0.0085, 0.0197, 0.0235, 0.0091, 0.0153, 0.0149], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:47:35,225 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60931.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:47:35,242 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0693, 2.3699, 3.8619, 1.7517, 2.9088, 2.5051, 2.0283, 2.7598], + device='cuda:3'), covar=tensor([0.1118, 0.1683, 0.0454, 0.2620, 0.1081, 0.1759, 0.1310, 0.1689], + device='cuda:3'), in_proj_covar=tensor([0.0471, 0.0484, 0.0525, 0.0559, 0.0599, 0.0542, 0.0455, 0.0594], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:47:35,684 INFO [train.py:901] (3/4) Epoch 8, batch 4350, loss[loss=0.2831, simple_loss=0.3385, pruned_loss=0.1139, over 7136.00 frames. ], tot_loss[loss=0.259, simple_loss=0.329, pruned_loss=0.0945, over 1617482.02 frames. ], batch size: 71, lr: 9.56e-03, grad_scale: 8.0 +2023-02-06 05:47:39,642 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.403e+02 2.751e+02 3.442e+02 4.335e+02 7.709e+02, threshold=6.884e+02, percent-clipped=1.0 +2023-02-06 05:47:51,980 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60956.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:47:59,291 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 05:48:10,331 INFO [train.py:901] (3/4) Epoch 8, batch 4400, loss[loss=0.252, simple_loss=0.328, pruned_loss=0.08799, over 8242.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3278, pruned_loss=0.09408, over 1614290.06 frames. ], batch size: 24, lr: 9.56e-03, grad_scale: 8.0 +2023-02-06 05:48:33,377 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6428, 2.0255, 4.6328, 1.3220, 3.2011, 2.1608, 1.7228, 2.7506], + device='cuda:3'), covar=tensor([0.1815, 0.2278, 0.0567, 0.3873, 0.1483, 0.2814, 0.1874, 0.2202], + device='cuda:3'), in_proj_covar=tensor([0.0471, 0.0487, 0.0522, 0.0559, 0.0598, 0.0542, 0.0455, 0.0593], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:48:42,540 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 05:48:44,685 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4605, 1.7660, 1.8559, 0.9078, 1.9740, 1.4151, 0.4044, 1.6899], + device='cuda:3'), covar=tensor([0.0264, 0.0166, 0.0120, 0.0293, 0.0162, 0.0444, 0.0450, 0.0120], + device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0280, 0.0227, 0.0339, 0.0268, 0.0430, 0.0333, 0.0306], + device='cuda:3'), out_proj_covar=tensor([1.0817e-04, 8.3407e-05, 6.7626e-05, 1.0127e-04, 8.2175e-05, 1.4119e-04, + 1.0195e-04, 9.2573e-05], device='cuda:3') +2023-02-06 05:48:45,132 INFO [train.py:901] (3/4) Epoch 8, batch 4450, loss[loss=0.2499, simple_loss=0.3167, pruned_loss=0.09151, over 7815.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3271, pruned_loss=0.09394, over 1614408.79 frames. ], batch size: 20, lr: 9.55e-03, grad_scale: 8.0 +2023-02-06 05:48:49,127 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.733e+02 2.403e+02 3.130e+02 3.948e+02 8.767e+02, threshold=6.260e+02, percent-clipped=3.0 +2023-02-06 05:49:13,356 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5094, 1.8201, 2.0384, 1.1622, 2.2323, 1.3778, 0.6821, 1.5787], + device='cuda:3'), covar=tensor([0.0296, 0.0177, 0.0128, 0.0284, 0.0166, 0.0480, 0.0416, 0.0172], + device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0280, 0.0228, 0.0339, 0.0268, 0.0430, 0.0332, 0.0308], + device='cuda:3'), out_proj_covar=tensor([1.0793e-04, 8.3361e-05, 6.7925e-05, 1.0123e-04, 8.2006e-05, 1.4153e-04, + 1.0163e-04, 9.2944e-05], device='cuda:3') +2023-02-06 05:49:17,368 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61079.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:49:17,967 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=61080.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:49:19,304 INFO [train.py:901] (3/4) Epoch 8, batch 4500, loss[loss=0.2251, simple_loss=0.2936, pruned_loss=0.07831, over 7707.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3275, pruned_loss=0.09411, over 1616451.83 frames. ], batch size: 18, lr: 9.55e-03, grad_scale: 8.0 +2023-02-06 05:49:24,899 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-06 05:49:36,534 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 05:49:38,012 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6261, 1.9081, 1.4932, 2.3225, 1.0256, 1.1613, 1.4607, 1.9425], + device='cuda:3'), covar=tensor([0.0925, 0.0924, 0.1212, 0.0556, 0.1308, 0.1859, 0.1182, 0.0870], + device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0231, 0.0271, 0.0218, 0.0229, 0.0266, 0.0268, 0.0237], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 05:49:44,119 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 +2023-02-06 05:49:50,317 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8186, 2.1375, 4.8461, 1.4310, 3.3181, 2.2060, 1.7016, 3.0008], + device='cuda:3'), covar=tensor([0.1511, 0.2239, 0.0480, 0.3574, 0.1336, 0.2547, 0.1664, 0.2015], + device='cuda:3'), in_proj_covar=tensor([0.0475, 0.0488, 0.0522, 0.0563, 0.0602, 0.0541, 0.0456, 0.0595], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:49:53,205 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2023-02-06 05:49:53,478 INFO [train.py:901] (3/4) Epoch 8, batch 4550, loss[loss=0.2585, simple_loss=0.3415, pruned_loss=0.08775, over 8678.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3274, pruned_loss=0.09462, over 1615350.73 frames. ], batch size: 34, lr: 9.55e-03, grad_scale: 8.0 +2023-02-06 05:49:58,147 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.532e+02 2.768e+02 3.493e+02 4.645e+02 1.007e+03, threshold=6.986e+02, percent-clipped=6.0 +2023-02-06 05:50:07,866 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4664, 1.9787, 3.1851, 2.4402, 2.6614, 2.2110, 1.6724, 1.2472], + device='cuda:3'), covar=tensor([0.3083, 0.3596, 0.0916, 0.2211, 0.1788, 0.1827, 0.1544, 0.3879], + device='cuda:3'), in_proj_covar=tensor([0.0835, 0.0798, 0.0680, 0.0782, 0.0881, 0.0730, 0.0668, 0.0717], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:50:29,340 INFO [train.py:901] (3/4) Epoch 8, batch 4600, loss[loss=0.2523, simple_loss=0.3047, pruned_loss=0.09992, over 7923.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3267, pruned_loss=0.09448, over 1614790.13 frames. ], batch size: 20, lr: 9.54e-03, grad_scale: 8.0 +2023-02-06 05:50:31,565 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6026, 2.0464, 3.2714, 2.4446, 2.7489, 2.3151, 1.7959, 1.4117], + device='cuda:3'), covar=tensor([0.3048, 0.3532, 0.0887, 0.2351, 0.1725, 0.1816, 0.1623, 0.3662], + device='cuda:3'), in_proj_covar=tensor([0.0833, 0.0796, 0.0681, 0.0782, 0.0880, 0.0728, 0.0668, 0.0715], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:50:38,249 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61195.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:50:38,324 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4312, 1.9370, 2.9304, 2.2896, 2.4850, 2.1641, 1.6725, 1.1854], + device='cuda:3'), covar=tensor([0.2843, 0.3256, 0.0946, 0.2024, 0.1616, 0.1715, 0.1432, 0.3207], + device='cuda:3'), in_proj_covar=tensor([0.0837, 0.0800, 0.0685, 0.0787, 0.0886, 0.0733, 0.0671, 0.0719], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:50:39,770 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-06 05:50:53,689 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3163, 1.7069, 1.6532, 1.6411, 1.2107, 1.5015, 1.8961, 1.6296], + device='cuda:3'), covar=tensor([0.0469, 0.1143, 0.1705, 0.1255, 0.0573, 0.1454, 0.0635, 0.0568], + device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0161, 0.0199, 0.0163, 0.0110, 0.0169, 0.0122, 0.0141], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 05:51:04,231 INFO [train.py:901] (3/4) Epoch 8, batch 4650, loss[loss=0.2796, simple_loss=0.3455, pruned_loss=0.1069, over 8350.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3267, pruned_loss=0.0941, over 1614314.58 frames. ], batch size: 24, lr: 9.54e-03, grad_scale: 8.0 +2023-02-06 05:51:08,278 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.975e+02 2.693e+02 3.115e+02 3.876e+02 8.832e+02, threshold=6.229e+02, percent-clipped=3.0 +2023-02-06 05:51:11,645 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3182, 4.2881, 3.8871, 1.9843, 3.8485, 3.9633, 3.9105, 3.5734], + device='cuda:3'), covar=tensor([0.1071, 0.0693, 0.1191, 0.4521, 0.0924, 0.0942, 0.1319, 0.0983], + device='cuda:3'), in_proj_covar=tensor([0.0416, 0.0333, 0.0347, 0.0442, 0.0348, 0.0324, 0.0337, 0.0286], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:51:19,732 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.1259, 1.7837, 1.8123, 1.6441, 1.2687, 1.6524, 2.3050, 2.2700], + device='cuda:3'), covar=tensor([0.0441, 0.1205, 0.1766, 0.1390, 0.0599, 0.1481, 0.0605, 0.0468], + device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0161, 0.0198, 0.0163, 0.0111, 0.0168, 0.0122, 0.0141], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 05:51:38,690 INFO [train.py:901] (3/4) Epoch 8, batch 4700, loss[loss=0.2856, simple_loss=0.3587, pruned_loss=0.1063, over 8583.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3279, pruned_loss=0.09453, over 1614541.60 frames. ], batch size: 31, lr: 9.54e-03, grad_scale: 8.0 +2023-02-06 05:52:09,393 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5308, 1.8481, 1.9547, 1.1338, 2.0424, 1.3509, 0.6016, 1.5713], + device='cuda:3'), covar=tensor([0.0270, 0.0163, 0.0119, 0.0269, 0.0178, 0.0451, 0.0417, 0.0153], + device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0278, 0.0228, 0.0332, 0.0265, 0.0425, 0.0328, 0.0305], + device='cuda:3'), out_proj_covar=tensor([1.0611e-04, 8.2890e-05, 6.7800e-05, 9.8778e-05, 8.0798e-05, 1.3965e-04, + 1.0029e-04, 9.2068e-05], device='cuda:3') +2023-02-06 05:52:13,893 INFO [train.py:901] (3/4) Epoch 8, batch 4750, loss[loss=0.2657, simple_loss=0.3333, pruned_loss=0.09905, over 8596.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.328, pruned_loss=0.09485, over 1612480.74 frames. ], batch size: 31, lr: 9.53e-03, grad_scale: 8.0 +2023-02-06 05:52:17,857 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.549e+02 2.870e+02 3.425e+02 4.672e+02 9.837e+02, threshold=6.850e+02, percent-clipped=8.0 +2023-02-06 05:52:22,852 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-06 05:52:34,193 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2915, 2.4999, 1.8573, 2.9995, 1.4313, 1.6378, 1.8839, 2.4499], + device='cuda:3'), covar=tensor([0.0763, 0.0843, 0.1124, 0.0386, 0.1267, 0.1563, 0.1206, 0.0853], + device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0231, 0.0272, 0.0218, 0.0230, 0.0264, 0.0270, 0.0236], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 05:52:36,774 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 05:52:39,421 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 05:52:48,205 INFO [train.py:901] (3/4) Epoch 8, batch 4800, loss[loss=0.2797, simple_loss=0.3578, pruned_loss=0.1008, over 8331.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3266, pruned_loss=0.09408, over 1603733.36 frames. ], batch size: 26, lr: 9.53e-03, grad_scale: 8.0 +2023-02-06 05:53:12,939 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3384, 1.5842, 2.9560, 1.2157, 2.0369, 1.7899, 1.5145, 1.8126], + device='cuda:3'), covar=tensor([0.2017, 0.2461, 0.0641, 0.4027, 0.1534, 0.2870, 0.2065, 0.2145], + device='cuda:3'), in_proj_covar=tensor([0.0474, 0.0487, 0.0525, 0.0563, 0.0603, 0.0539, 0.0455, 0.0596], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:53:16,794 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=61423.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:53:22,686 INFO [train.py:901] (3/4) Epoch 8, batch 4850, loss[loss=0.2437, simple_loss=0.3124, pruned_loss=0.08752, over 8134.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3268, pruned_loss=0.09345, over 1611567.01 frames. ], batch size: 22, lr: 9.52e-03, grad_scale: 8.0 +2023-02-06 05:53:26,653 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.608e+02 2.780e+02 3.448e+02 4.323e+02 7.771e+02, threshold=6.895e+02, percent-clipped=1.0 +2023-02-06 05:53:28,671 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 05:53:36,192 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61451.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:53:53,266 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61476.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:53:54,744 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-06 05:53:56,860 INFO [train.py:901] (3/4) Epoch 8, batch 4900, loss[loss=0.2887, simple_loss=0.3534, pruned_loss=0.112, over 8489.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3268, pruned_loss=0.09387, over 1612502.83 frames. ], batch size: 28, lr: 9.52e-03, grad_scale: 8.0 +2023-02-06 05:53:59,790 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5670, 1.8888, 3.2293, 1.2241, 2.2634, 1.9951, 1.6042, 1.8915], + device='cuda:3'), covar=tensor([0.1485, 0.2014, 0.0639, 0.3518, 0.1408, 0.2399, 0.1550, 0.2020], + device='cuda:3'), in_proj_covar=tensor([0.0479, 0.0492, 0.0535, 0.0568, 0.0610, 0.0545, 0.0460, 0.0606], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 05:54:07,101 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3074, 1.7617, 2.7409, 2.1227, 2.3321, 2.0530, 1.6669, 1.0070], + device='cuda:3'), covar=tensor([0.2768, 0.3182, 0.0791, 0.1749, 0.1339, 0.1762, 0.1447, 0.3140], + device='cuda:3'), in_proj_covar=tensor([0.0852, 0.0811, 0.0689, 0.0793, 0.0889, 0.0744, 0.0678, 0.0731], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:54:31,292 INFO [train.py:901] (3/4) Epoch 8, batch 4950, loss[loss=0.2359, simple_loss=0.3125, pruned_loss=0.07967, over 8193.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3271, pruned_loss=0.09385, over 1611918.96 frames. ], batch size: 23, lr: 9.52e-03, grad_scale: 8.0 +2023-02-06 05:54:35,325 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.817e+02 2.701e+02 3.325e+02 4.582e+02 7.633e+02, threshold=6.649e+02, percent-clipped=1.0 +2023-02-06 05:54:35,521 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61538.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:54:47,371 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9071, 2.4012, 3.9974, 2.8293, 3.3055, 2.4137, 1.9516, 1.7485], + device='cuda:3'), covar=tensor([0.2875, 0.3618, 0.0763, 0.2314, 0.1730, 0.1711, 0.1483, 0.3939], + device='cuda:3'), in_proj_covar=tensor([0.0846, 0.0805, 0.0680, 0.0787, 0.0880, 0.0737, 0.0674, 0.0725], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:55:07,153 INFO [train.py:901] (3/4) Epoch 8, batch 5000, loss[loss=0.228, simple_loss=0.2979, pruned_loss=0.0791, over 7797.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3254, pruned_loss=0.09268, over 1609936.55 frames. ], batch size: 19, lr: 9.51e-03, grad_scale: 8.0 +2023-02-06 05:55:17,611 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.7139, 1.5617, 5.7841, 2.2465, 5.1510, 4.8661, 5.3825, 5.2318], + device='cuda:3'), covar=tensor([0.0455, 0.4575, 0.0290, 0.3086, 0.0910, 0.0763, 0.0478, 0.0466], + device='cuda:3'), in_proj_covar=tensor([0.0432, 0.0549, 0.0522, 0.0497, 0.0564, 0.0474, 0.0474, 0.0522], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 05:55:42,165 INFO [train.py:901] (3/4) Epoch 8, batch 5050, loss[loss=0.266, simple_loss=0.335, pruned_loss=0.0985, over 7971.00 frames. ], tot_loss[loss=0.255, simple_loss=0.325, pruned_loss=0.0925, over 1612171.43 frames. ], batch size: 21, lr: 9.51e-03, grad_scale: 8.0 +2023-02-06 05:55:46,809 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.732e+02 2.771e+02 3.459e+02 4.924e+02 1.310e+03, threshold=6.919e+02, percent-clipped=9.0 +2023-02-06 05:55:55,071 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2005, 1.5630, 1.5463, 1.4523, 1.0561, 1.3762, 1.6582, 1.7223], + device='cuda:3'), covar=tensor([0.0497, 0.1172, 0.1863, 0.1325, 0.0615, 0.1520, 0.0728, 0.0559], + device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0160, 0.0200, 0.0163, 0.0111, 0.0168, 0.0121, 0.0141], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 05:56:07,702 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 05:56:17,393 INFO [train.py:901] (3/4) Epoch 8, batch 5100, loss[loss=0.2335, simple_loss=0.2972, pruned_loss=0.08485, over 7434.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3233, pruned_loss=0.09167, over 1610220.17 frames. ], batch size: 17, lr: 9.50e-03, grad_scale: 8.0 +2023-02-06 05:56:52,941 INFO [train.py:901] (3/4) Epoch 8, batch 5150, loss[loss=0.2203, simple_loss=0.2963, pruned_loss=0.07221, over 7795.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3235, pruned_loss=0.09149, over 1613773.64 frames. ], batch size: 20, lr: 9.50e-03, grad_scale: 8.0 +2023-02-06 05:56:57,120 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.735e+02 2.584e+02 3.190e+02 4.018e+02 8.337e+02, threshold=6.381e+02, percent-clipped=2.0 +2023-02-06 05:57:06,179 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61750.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:57:07,616 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1760, 1.0934, 1.1380, 1.0423, 0.8026, 1.2680, 0.0530, 0.9251], + device='cuda:3'), covar=tensor([0.3397, 0.2710, 0.0909, 0.1861, 0.5325, 0.0865, 0.4352, 0.2145], + device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0151, 0.0087, 0.0200, 0.0238, 0.0092, 0.0160, 0.0153], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0003, 0.0002], + device='cuda:3') +2023-02-06 05:57:08,917 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.7194, 1.8592, 5.7919, 2.1277, 5.1541, 4.9236, 5.3962, 5.2445], + device='cuda:3'), covar=tensor([0.0392, 0.3677, 0.0300, 0.3073, 0.0936, 0.0721, 0.0434, 0.0408], + device='cuda:3'), in_proj_covar=tensor([0.0428, 0.0541, 0.0515, 0.0495, 0.0559, 0.0468, 0.0469, 0.0518], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 05:57:13,701 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61761.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:57:27,554 INFO [train.py:901] (3/4) Epoch 8, batch 5200, loss[loss=0.2518, simple_loss=0.3157, pruned_loss=0.09399, over 7936.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3248, pruned_loss=0.09251, over 1612659.17 frames. ], batch size: 20, lr: 9.50e-03, grad_scale: 8.0 +2023-02-06 05:57:35,720 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61794.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:57:53,740 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61819.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:58:02,409 INFO [train.py:901] (3/4) Epoch 8, batch 5250, loss[loss=0.2449, simple_loss=0.3161, pruned_loss=0.08687, over 8199.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3246, pruned_loss=0.092, over 1616405.27 frames. ], batch size: 23, lr: 9.49e-03, grad_scale: 8.0 +2023-02-06 05:58:06,468 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.578e+02 2.711e+02 3.309e+02 4.013e+02 1.150e+03, threshold=6.618e+02, percent-clipped=3.0 +2023-02-06 05:58:07,305 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5062, 2.4866, 1.6263, 2.0662, 1.9224, 1.3288, 1.9460, 1.9023], + device='cuda:3'), covar=tensor([0.1143, 0.0327, 0.1128, 0.0493, 0.0631, 0.1399, 0.0790, 0.0811], + device='cuda:3'), in_proj_covar=tensor([0.0341, 0.0231, 0.0308, 0.0293, 0.0302, 0.0308, 0.0334, 0.0308], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 05:58:07,802 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 05:58:10,096 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3303, 1.9427, 2.9267, 2.3355, 2.5485, 2.0686, 1.6414, 1.2360], + device='cuda:3'), covar=tensor([0.2754, 0.2987, 0.0817, 0.1762, 0.1366, 0.1618, 0.1400, 0.3119], + device='cuda:3'), in_proj_covar=tensor([0.0851, 0.0807, 0.0681, 0.0793, 0.0892, 0.0741, 0.0675, 0.0730], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 05:58:37,170 INFO [train.py:901] (3/4) Epoch 8, batch 5300, loss[loss=0.2705, simple_loss=0.3442, pruned_loss=0.09844, over 8474.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.324, pruned_loss=0.09221, over 1612544.47 frames. ], batch size: 25, lr: 9.49e-03, grad_scale: 8.0 +2023-02-06 05:59:12,098 INFO [train.py:901] (3/4) Epoch 8, batch 5350, loss[loss=0.2136, simple_loss=0.2943, pruned_loss=0.06649, over 8243.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3253, pruned_loss=0.09308, over 1610824.23 frames. ], batch size: 22, lr: 9.49e-03, grad_scale: 8.0 +2023-02-06 05:59:12,261 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61932.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 05:59:16,956 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.785e+02 2.596e+02 3.249e+02 3.983e+02 1.109e+03, threshold=6.498e+02, percent-clipped=6.0 +2023-02-06 05:59:47,805 INFO [train.py:901] (3/4) Epoch 8, batch 5400, loss[loss=0.2441, simple_loss=0.3296, pruned_loss=0.07929, over 8470.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3248, pruned_loss=0.0922, over 1612366.75 frames. ], batch size: 25, lr: 9.48e-03, grad_scale: 8.0 +2023-02-06 06:00:23,965 INFO [train.py:901] (3/4) Epoch 8, batch 5450, loss[loss=0.2657, simple_loss=0.3313, pruned_loss=0.1, over 8254.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3249, pruned_loss=0.09205, over 1614876.74 frames. ], batch size: 22, lr: 9.48e-03, grad_scale: 8.0 +2023-02-06 06:00:28,689 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.755e+02 2.625e+02 3.240e+02 4.068e+02 8.471e+02, threshold=6.479e+02, percent-clipped=5.0 +2023-02-06 06:00:33,864 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3568, 1.8995, 2.9865, 2.3989, 2.5542, 2.0953, 1.6032, 1.3354], + device='cuda:3'), covar=tensor([0.2835, 0.3284, 0.0836, 0.1914, 0.1458, 0.1717, 0.1358, 0.3362], + device='cuda:3'), in_proj_covar=tensor([0.0847, 0.0805, 0.0679, 0.0789, 0.0887, 0.0739, 0.0669, 0.0728], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 06:00:49,039 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5469, 1.5049, 1.6887, 1.3919, 0.9975, 1.8146, 0.1006, 1.2248], + device='cuda:3'), covar=tensor([0.2569, 0.2125, 0.0629, 0.1676, 0.4964, 0.0565, 0.3640, 0.1830], + device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0152, 0.0087, 0.0200, 0.0242, 0.0092, 0.0160, 0.0151], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0003, 0.0002], + device='cuda:3') +2023-02-06 06:01:00,412 INFO [train.py:901] (3/4) Epoch 8, batch 5500, loss[loss=0.2072, simple_loss=0.2809, pruned_loss=0.06677, over 7527.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3235, pruned_loss=0.09139, over 1613861.71 frames. ], batch size: 18, lr: 9.47e-03, grad_scale: 8.0 +2023-02-06 06:01:01,792 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 06:01:08,468 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62094.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:01:16,547 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62105.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:01:34,829 INFO [train.py:901] (3/4) Epoch 8, batch 5550, loss[loss=0.2264, simple_loss=0.3073, pruned_loss=0.07275, over 8193.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.325, pruned_loss=0.09138, over 1622760.53 frames. ], batch size: 23, lr: 9.47e-03, grad_scale: 8.0 +2023-02-06 06:01:38,603 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.598e+02 2.724e+02 3.277e+02 4.222e+02 9.983e+02, threshold=6.553e+02, percent-clipped=5.0 +2023-02-06 06:01:40,112 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62140.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:02:09,239 INFO [train.py:901] (3/4) Epoch 8, batch 5600, loss[loss=0.2189, simple_loss=0.285, pruned_loss=0.07636, over 7442.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3258, pruned_loss=0.09236, over 1622120.10 frames. ], batch size: 17, lr: 9.47e-03, grad_scale: 8.0 +2023-02-06 06:02:19,446 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2023-02-06 06:02:27,947 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62209.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:02:35,378 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62220.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:02:44,114 INFO [train.py:901] (3/4) Epoch 8, batch 5650, loss[loss=0.2632, simple_loss=0.3429, pruned_loss=0.09171, over 8521.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3251, pruned_loss=0.09192, over 1621999.64 frames. ], batch size: 39, lr: 9.46e-03, grad_scale: 8.0 +2023-02-06 06:02:48,201 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.899e+02 2.893e+02 3.442e+02 4.058e+02 7.819e+02, threshold=6.884e+02, percent-clipped=2.0 +2023-02-06 06:02:58,226 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 06:03:03,307 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 06:03:13,678 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3528, 2.7789, 1.8105, 2.1386, 2.2337, 1.4025, 1.9431, 1.9903], + device='cuda:3'), covar=tensor([0.1161, 0.0264, 0.0905, 0.0522, 0.0541, 0.1284, 0.0837, 0.0786], + device='cuda:3'), in_proj_covar=tensor([0.0342, 0.0231, 0.0309, 0.0295, 0.0305, 0.0316, 0.0339, 0.0308], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 06:03:14,928 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62276.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:03:19,746 INFO [train.py:901] (3/4) Epoch 8, batch 5700, loss[loss=0.2317, simple_loss=0.3019, pruned_loss=0.08069, over 7984.00 frames. ], tot_loss[loss=0.255, simple_loss=0.326, pruned_loss=0.09196, over 1625932.30 frames. ], batch size: 21, lr: 9.46e-03, grad_scale: 8.0 +2023-02-06 06:03:26,021 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62291.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:03:32,321 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2023-02-06 06:03:53,772 INFO [train.py:901] (3/4) Epoch 8, batch 5750, loss[loss=0.2785, simple_loss=0.3435, pruned_loss=0.1067, over 8495.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3249, pruned_loss=0.0919, over 1620682.85 frames. ], batch size: 26, lr: 9.45e-03, grad_scale: 8.0 +2023-02-06 06:03:58,446 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.993e+02 2.702e+02 3.342e+02 4.214e+02 1.406e+03, threshold=6.684e+02, percent-clipped=3.0 +2023-02-06 06:04:07,821 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 06:04:28,273 INFO [train.py:901] (3/4) Epoch 8, batch 5800, loss[loss=0.251, simple_loss=0.3033, pruned_loss=0.09933, over 7426.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3249, pruned_loss=0.09175, over 1621268.67 frames. ], batch size: 17, lr: 9.45e-03, grad_scale: 8.0 +2023-02-06 06:04:35,223 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62391.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:04:48,168 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62409.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:05:04,097 INFO [train.py:901] (3/4) Epoch 8, batch 5850, loss[loss=0.2212, simple_loss=0.3006, pruned_loss=0.07091, over 8138.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3239, pruned_loss=0.09093, over 1620208.75 frames. ], batch size: 22, lr: 9.45e-03, grad_scale: 8.0 +2023-02-06 06:05:06,527 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 +2023-02-06 06:05:08,203 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.970e+02 2.690e+02 3.286e+02 4.000e+02 6.740e+02, threshold=6.571e+02, percent-clipped=1.0 +2023-02-06 06:05:20,411 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0821, 1.2760, 1.1992, 0.3455, 1.2323, 1.0137, 0.1273, 1.1083], + device='cuda:3'), covar=tensor([0.0187, 0.0174, 0.0148, 0.0297, 0.0187, 0.0530, 0.0383, 0.0152], + device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0279, 0.0226, 0.0336, 0.0267, 0.0425, 0.0328, 0.0305], + device='cuda:3'), out_proj_covar=tensor([1.0555e-04, 8.2629e-05, 6.6702e-05, 1.0024e-04, 8.0871e-05, 1.3890e-04, + 1.0028e-04, 9.1450e-05], device='cuda:3') +2023-02-06 06:05:27,195 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62465.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:05:34,614 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62476.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:05:38,457 INFO [train.py:901] (3/4) Epoch 8, batch 5900, loss[loss=0.2686, simple_loss=0.3371, pruned_loss=0.1, over 8322.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3236, pruned_loss=0.09104, over 1616161.44 frames. ], batch size: 26, lr: 9.44e-03, grad_scale: 8.0 +2023-02-06 06:05:39,871 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62484.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:05:44,020 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62490.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:05:52,029 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62501.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:06:13,414 INFO [train.py:901] (3/4) Epoch 8, batch 5950, loss[loss=0.1905, simple_loss=0.2701, pruned_loss=0.05547, over 7799.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3232, pruned_loss=0.09018, over 1619122.64 frames. ], batch size: 19, lr: 9.44e-03, grad_scale: 8.0 +2023-02-06 06:06:17,413 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.881e+02 2.652e+02 3.220e+02 3.904e+02 8.315e+02, threshold=6.439e+02, percent-clipped=2.0 +2023-02-06 06:06:34,802 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62563.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:06:47,821 INFO [train.py:901] (3/4) Epoch 8, batch 6000, loss[loss=0.2655, simple_loss=0.3305, pruned_loss=0.1002, over 8237.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3237, pruned_loss=0.09091, over 1618992.11 frames. ], batch size: 22, lr: 9.44e-03, grad_scale: 8.0 +2023-02-06 06:06:47,822 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 06:06:56,083 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2890, 2.1585, 1.5099, 1.8462, 1.8148, 1.2620, 1.5891, 1.6110], + device='cuda:3'), covar=tensor([0.1056, 0.0302, 0.1002, 0.0468, 0.0633, 0.1234, 0.0867, 0.0796], + device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0233, 0.0309, 0.0293, 0.0307, 0.0315, 0.0338, 0.0313], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 06:07:00,012 INFO [train.py:935] (3/4) Epoch 8, validation: loss=0.1996, simple_loss=0.2985, pruned_loss=0.05037, over 944034.00 frames. +2023-02-06 06:07:00,013 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 06:07:12,285 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62599.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:07:14,967 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62603.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:07:33,819 INFO [train.py:901] (3/4) Epoch 8, batch 6050, loss[loss=0.2044, simple_loss=0.2819, pruned_loss=0.06343, over 7814.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3234, pruned_loss=0.09073, over 1617194.28 frames. ], batch size: 20, lr: 9.43e-03, grad_scale: 8.0 +2023-02-06 06:07:35,913 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62635.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:07:37,858 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.666e+02 2.575e+02 3.268e+02 4.071e+02 9.720e+02, threshold=6.536e+02, percent-clipped=3.0 +2023-02-06 06:07:44,180 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62647.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:08:02,426 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62672.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:08:08,753 INFO [train.py:901] (3/4) Epoch 8, batch 6100, loss[loss=0.2032, simple_loss=0.2759, pruned_loss=0.06522, over 7550.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3239, pruned_loss=0.09083, over 1619395.82 frames. ], batch size: 18, lr: 9.43e-03, grad_scale: 8.0 +2023-02-06 06:08:10,161 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62684.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:08:37,242 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 06:08:43,081 INFO [train.py:901] (3/4) Epoch 8, batch 6150, loss[loss=0.2624, simple_loss=0.3386, pruned_loss=0.09311, over 8736.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3231, pruned_loss=0.09088, over 1617839.34 frames. ], batch size: 30, lr: 9.42e-03, grad_scale: 8.0 +2023-02-06 06:08:47,077 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.762e+02 2.667e+02 3.544e+02 4.037e+02 8.376e+02, threshold=7.087e+02, percent-clipped=5.0 +2023-02-06 06:08:55,088 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62750.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:08:57,067 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62753.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:09:17,711 INFO [train.py:901] (3/4) Epoch 8, batch 6200, loss[loss=0.2289, simple_loss=0.3144, pruned_loss=0.07165, over 8342.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3229, pruned_loss=0.09115, over 1613531.81 frames. ], batch size: 25, lr: 9.42e-03, grad_scale: 16.0 +2023-02-06 06:09:23,752 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62791.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:09:52,742 INFO [train.py:901] (3/4) Epoch 8, batch 6250, loss[loss=0.217, simple_loss=0.288, pruned_loss=0.07302, over 7524.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3235, pruned_loss=0.09183, over 1610789.18 frames. ], batch size: 18, lr: 9.42e-03, grad_scale: 16.0 +2023-02-06 06:09:56,741 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.905e+02 2.717e+02 3.222e+02 4.596e+02 9.217e+02, threshold=6.445e+02, percent-clipped=3.0 +2023-02-06 06:10:08,419 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62855.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:10:16,912 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62868.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:10:23,670 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9903, 3.9937, 2.4128, 2.6466, 2.9990, 2.2838, 2.8289, 2.8293], + device='cuda:3'), covar=tensor([0.1341, 0.0226, 0.0794, 0.0700, 0.0584, 0.0970, 0.0857, 0.0907], + device='cuda:3'), in_proj_covar=tensor([0.0339, 0.0233, 0.0305, 0.0288, 0.0300, 0.0311, 0.0332, 0.0308], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 06:10:24,988 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62880.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:10:26,044 INFO [train.py:901] (3/4) Epoch 8, batch 6300, loss[loss=0.2809, simple_loss=0.3477, pruned_loss=0.107, over 8022.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3238, pruned_loss=0.09187, over 1613486.45 frames. ], batch size: 22, lr: 9.41e-03, grad_scale: 16.0 +2023-02-06 06:10:31,573 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9440, 2.0363, 1.7510, 2.6474, 1.2090, 1.5268, 1.7662, 2.1788], + device='cuda:3'), covar=tensor([0.0874, 0.1004, 0.1255, 0.0419, 0.1235, 0.1574, 0.1028, 0.0782], + device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0226, 0.0267, 0.0215, 0.0227, 0.0263, 0.0264, 0.0232], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 06:10:44,100 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62907.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:11:01,294 INFO [train.py:901] (3/4) Epoch 8, batch 6350, loss[loss=0.2881, simple_loss=0.3595, pruned_loss=0.1084, over 8665.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3235, pruned_loss=0.09167, over 1613406.56 frames. ], batch size: 34, lr: 9.41e-03, grad_scale: 16.0 +2023-02-06 06:11:05,349 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.489e+02 2.824e+02 3.504e+02 4.161e+02 7.437e+02, threshold=7.007e+02, percent-clipped=2.0 +2023-02-06 06:11:12,102 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62947.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:11:14,900 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9173, 1.9576, 2.2917, 1.8128, 1.2520, 2.3115, 0.3716, 1.3972], + device='cuda:3'), covar=tensor([0.2439, 0.1656, 0.0471, 0.2083, 0.5054, 0.0497, 0.3903, 0.2146], + device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0150, 0.0088, 0.0197, 0.0236, 0.0092, 0.0156, 0.0154], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 06:11:17,753 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 06:11:35,238 INFO [train.py:901] (3/4) Epoch 8, batch 6400, loss[loss=0.2274, simple_loss=0.3124, pruned_loss=0.07123, over 8298.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3232, pruned_loss=0.09124, over 1615277.81 frames. ], batch size: 23, lr: 9.41e-03, grad_scale: 16.0 +2023-02-06 06:11:52,116 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63006.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:12:03,499 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63022.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:12:07,442 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63028.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:12:09,585 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63031.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:12:10,068 INFO [train.py:901] (3/4) Epoch 8, batch 6450, loss[loss=0.2733, simple_loss=0.3393, pruned_loss=0.1036, over 8143.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3226, pruned_loss=0.09107, over 1611362.69 frames. ], batch size: 22, lr: 9.40e-03, grad_scale: 16.0 +2023-02-06 06:12:13,710 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6325, 2.2470, 4.3771, 1.1687, 3.1012, 2.1718, 1.6320, 2.4787], + device='cuda:3'), covar=tensor([0.1571, 0.1892, 0.0568, 0.3537, 0.1220, 0.2369, 0.1660, 0.2342], + device='cuda:3'), in_proj_covar=tensor([0.0482, 0.0490, 0.0538, 0.0569, 0.0604, 0.0537, 0.0459, 0.0606], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003], + device='cuda:3') +2023-02-06 06:12:14,125 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.060e+02 2.983e+02 3.820e+02 5.218e+02 9.633e+02, threshold=7.640e+02, percent-clipped=4.0 +2023-02-06 06:12:31,361 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63062.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:12:45,220 INFO [train.py:901] (3/4) Epoch 8, batch 6500, loss[loss=0.2825, simple_loss=0.3487, pruned_loss=0.1082, over 8696.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3228, pruned_loss=0.09102, over 1607875.36 frames. ], batch size: 34, lr: 9.40e-03, grad_scale: 16.0 +2023-02-06 06:13:03,586 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.16 vs. limit=5.0 +2023-02-06 06:13:14,197 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63124.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:13:20,107 INFO [train.py:901] (3/4) Epoch 8, batch 6550, loss[loss=0.2168, simple_loss=0.2954, pruned_loss=0.06913, over 8086.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3239, pruned_loss=0.09181, over 1607636.51 frames. ], batch size: 21, lr: 9.40e-03, grad_scale: 16.0 +2023-02-06 06:13:22,295 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63135.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:13:24,225 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.694e+02 2.645e+02 3.116e+02 3.905e+02 9.747e+02, threshold=6.232e+02, percent-clipped=3.0 +2023-02-06 06:13:27,751 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63143.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:13:27,810 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4177, 1.6857, 2.8296, 1.2011, 2.1661, 1.8492, 1.4181, 1.7722], + device='cuda:3'), covar=tensor([0.1458, 0.1889, 0.0622, 0.3326, 0.1149, 0.2309, 0.1543, 0.1831], + device='cuda:3'), in_proj_covar=tensor([0.0479, 0.0487, 0.0533, 0.0566, 0.0600, 0.0535, 0.0458, 0.0603], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 06:13:29,203 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3077, 2.5260, 1.8124, 2.0118, 2.0665, 1.4483, 1.7070, 2.0064], + device='cuda:3'), covar=tensor([0.1234, 0.0310, 0.0874, 0.0488, 0.0631, 0.1235, 0.0941, 0.0721], + device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0235, 0.0308, 0.0294, 0.0304, 0.0315, 0.0337, 0.0309], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 06:13:31,960 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63149.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:13:46,259 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2241, 2.8718, 2.9126, 1.7698, 1.6546, 3.2133, 0.6347, 2.0266], + device='cuda:3'), covar=tensor([0.1586, 0.1462, 0.1191, 0.2979, 0.5694, 0.0472, 0.4994, 0.2324], + device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0155, 0.0091, 0.0204, 0.0243, 0.0095, 0.0159, 0.0158], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 06:13:49,427 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 06:13:54,668 INFO [train.py:901] (3/4) Epoch 8, batch 6600, loss[loss=0.2413, simple_loss=0.3125, pruned_loss=0.08501, over 8194.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3241, pruned_loss=0.09177, over 1610895.11 frames. ], batch size: 23, lr: 9.39e-03, grad_scale: 16.0 +2023-02-06 06:13:56,259 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2879, 1.8750, 2.8918, 2.2990, 2.5113, 2.0758, 1.6481, 1.1637], + device='cuda:3'), covar=tensor([0.3139, 0.3362, 0.0859, 0.2025, 0.1732, 0.1684, 0.1448, 0.3544], + device='cuda:3'), in_proj_covar=tensor([0.0845, 0.0806, 0.0680, 0.0792, 0.0892, 0.0741, 0.0673, 0.0727], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 06:14:07,399 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 06:14:25,322 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.51 vs. limit=5.0 +2023-02-06 06:14:29,753 INFO [train.py:901] (3/4) Epoch 8, batch 6650, loss[loss=0.2332, simple_loss=0.3078, pruned_loss=0.07936, over 8106.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.325, pruned_loss=0.09216, over 1616507.80 frames. ], batch size: 23, lr: 9.39e-03, grad_scale: 16.0 +2023-02-06 06:14:33,644 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.260e+02 2.943e+02 3.528e+02 4.449e+02 1.178e+03, threshold=7.055e+02, percent-clipped=8.0 +2023-02-06 06:14:34,518 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63239.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:14:42,442 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63250.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:15:01,179 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63278.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:15:03,732 INFO [train.py:901] (3/4) Epoch 8, batch 6700, loss[loss=0.2, simple_loss=0.2862, pruned_loss=0.0569, over 8278.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3258, pruned_loss=0.09268, over 1618229.89 frames. ], batch size: 23, lr: 9.38e-03, grad_scale: 16.0 +2023-02-06 06:15:19,256 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63303.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:15:29,352 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63318.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:15:34,397 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 +2023-02-06 06:15:38,596 INFO [train.py:901] (3/4) Epoch 8, batch 6750, loss[loss=0.2137, simple_loss=0.2831, pruned_loss=0.07217, over 8283.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3244, pruned_loss=0.09169, over 1619122.79 frames. ], batch size: 23, lr: 9.38e-03, grad_scale: 8.0 +2023-02-06 06:15:43,298 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.362e+02 2.603e+02 3.068e+02 3.707e+02 1.416e+03, threshold=6.136e+02, percent-clipped=3.0 +2023-02-06 06:15:46,216 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63343.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:16:13,022 INFO [train.py:901] (3/4) Epoch 8, batch 6800, loss[loss=0.2444, simple_loss=0.3192, pruned_loss=0.08478, over 8129.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3238, pruned_loss=0.09134, over 1618178.57 frames. ], batch size: 22, lr: 9.38e-03, grad_scale: 8.0 +2023-02-06 06:16:19,067 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1471, 4.0858, 3.7041, 1.8488, 3.6517, 3.6429, 3.7028, 3.1850], + device='cuda:3'), covar=tensor([0.0815, 0.0600, 0.0987, 0.4686, 0.0943, 0.0988, 0.1276, 0.1059], + device='cuda:3'), in_proj_covar=tensor([0.0436, 0.0343, 0.0360, 0.0451, 0.0358, 0.0335, 0.0349, 0.0299], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 06:16:20,992 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 06:16:24,486 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63399.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:16:42,724 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63424.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:16:47,925 INFO [train.py:901] (3/4) Epoch 8, batch 6850, loss[loss=0.228, simple_loss=0.3082, pruned_loss=0.07391, over 8240.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.323, pruned_loss=0.09085, over 1616905.79 frames. ], batch size: 22, lr: 9.37e-03, grad_scale: 8.0 +2023-02-06 06:16:52,299 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63438.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:16:52,777 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.904e+02 2.669e+02 3.418e+02 4.059e+02 7.847e+02, threshold=6.836e+02, percent-clipped=4.0 +2023-02-06 06:16:53,667 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63440.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:16:57,207 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0255, 1.7217, 1.2747, 1.6328, 1.3901, 1.0959, 1.3073, 1.3577], + device='cuda:3'), covar=tensor([0.0923, 0.0403, 0.1026, 0.0420, 0.0568, 0.1198, 0.0721, 0.0668], + device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0237, 0.0309, 0.0295, 0.0306, 0.0315, 0.0336, 0.0312], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 06:17:11,111 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 06:17:22,530 INFO [train.py:901] (3/4) Epoch 8, batch 6900, loss[loss=0.2297, simple_loss=0.2988, pruned_loss=0.08028, over 7701.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3239, pruned_loss=0.09178, over 1615601.32 frames. ], batch size: 18, lr: 9.37e-03, grad_scale: 8.0 +2023-02-06 06:17:39,554 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63506.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:17:45,254 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.71 vs. limit=5.0 +2023-02-06 06:17:58,279 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63531.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:17:58,757 INFO [train.py:901] (3/4) Epoch 8, batch 6950, loss[loss=0.2182, simple_loss=0.3008, pruned_loss=0.06779, over 8022.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3221, pruned_loss=0.09063, over 1612422.01 frames. ], batch size: 22, lr: 9.37e-03, grad_scale: 8.0 +2023-02-06 06:18:03,563 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.884e+02 2.671e+02 3.369e+02 4.495e+02 9.890e+02, threshold=6.738e+02, percent-clipped=4.0 +2023-02-06 06:18:18,590 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 06:18:28,136 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4572, 1.9269, 3.2494, 1.1397, 2.3135, 1.7297, 1.5227, 1.8470], + device='cuda:3'), covar=tensor([0.1537, 0.1782, 0.0664, 0.3429, 0.1292, 0.2617, 0.1598, 0.2334], + device='cuda:3'), in_proj_covar=tensor([0.0478, 0.0483, 0.0530, 0.0563, 0.0597, 0.0538, 0.0457, 0.0596], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 06:18:32,009 INFO [train.py:901] (3/4) Epoch 8, batch 7000, loss[loss=0.2372, simple_loss=0.3251, pruned_loss=0.07468, over 8038.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3229, pruned_loss=0.09084, over 1614295.00 frames. ], batch size: 22, lr: 9.36e-03, grad_scale: 8.0 +2023-02-06 06:18:32,810 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63583.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:18:36,325 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63587.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:18:48,868 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63604.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:18:55,138 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4463, 1.4031, 3.0734, 1.3329, 2.0989, 3.2775, 3.3285, 2.7847], + device='cuda:3'), covar=tensor([0.1334, 0.1687, 0.0402, 0.2058, 0.1012, 0.0320, 0.0499, 0.0781], + device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0279, 0.0241, 0.0268, 0.0253, 0.0224, 0.0292, 0.0280], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2023-02-06 06:19:08,076 INFO [train.py:901] (3/4) Epoch 8, batch 7050, loss[loss=0.2816, simple_loss=0.3407, pruned_loss=0.1112, over 7222.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3234, pruned_loss=0.0907, over 1615370.55 frames. ], batch size: 71, lr: 9.36e-03, grad_scale: 8.0 +2023-02-06 06:19:12,575 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.671e+02 2.720e+02 3.242e+02 3.930e+02 7.648e+02, threshold=6.484e+02, percent-clipped=3.0 +2023-02-06 06:19:42,443 INFO [train.py:901] (3/4) Epoch 8, batch 7100, loss[loss=0.2635, simple_loss=0.3373, pruned_loss=0.09484, over 8503.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3231, pruned_loss=0.09018, over 1618419.30 frames. ], batch size: 28, lr: 9.35e-03, grad_scale: 8.0 +2023-02-06 06:19:52,879 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5223, 2.0363, 3.4057, 1.2442, 2.4173, 1.9357, 1.6213, 2.1732], + device='cuda:3'), covar=tensor([0.1590, 0.1927, 0.0685, 0.3676, 0.1405, 0.2626, 0.1640, 0.2191], + device='cuda:3'), in_proj_covar=tensor([0.0481, 0.0485, 0.0529, 0.0565, 0.0601, 0.0541, 0.0461, 0.0600], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 06:19:53,533 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63698.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:20:17,191 INFO [train.py:901] (3/4) Epoch 8, batch 7150, loss[loss=0.2466, simple_loss=0.3125, pruned_loss=0.09032, over 8285.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3229, pruned_loss=0.09024, over 1617403.56 frames. ], batch size: 23, lr: 9.35e-03, grad_scale: 8.0 +2023-02-06 06:20:21,748 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.819e+02 2.778e+02 3.549e+02 4.516e+02 1.097e+03, threshold=7.098e+02, percent-clipped=7.0 +2023-02-06 06:20:51,749 INFO [train.py:901] (3/4) Epoch 8, batch 7200, loss[loss=0.2841, simple_loss=0.3496, pruned_loss=0.1093, over 6847.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3236, pruned_loss=0.09087, over 1613804.75 frames. ], batch size: 71, lr: 9.35e-03, grad_scale: 8.0 +2023-02-06 06:20:51,835 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63782.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:20:53,157 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63784.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:21:15,508 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3404, 1.1172, 1.3092, 1.0767, 0.7999, 1.1504, 1.1165, 1.1206], + device='cuda:3'), covar=tensor([0.0557, 0.1446, 0.1887, 0.1539, 0.0628, 0.1724, 0.0767, 0.0686], + device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0161, 0.0198, 0.0164, 0.0111, 0.0168, 0.0122, 0.0141], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 06:21:23,973 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63828.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:21:26,385 INFO [train.py:901] (3/4) Epoch 8, batch 7250, loss[loss=0.3124, simple_loss=0.372, pruned_loss=0.1264, over 8558.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3238, pruned_loss=0.09173, over 1609842.02 frames. ], batch size: 31, lr: 9.34e-03, grad_scale: 8.0 +2023-02-06 06:21:30,976 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.883e+02 2.694e+02 3.202e+02 4.148e+02 8.009e+02, threshold=6.403e+02, percent-clipped=2.0 +2023-02-06 06:21:35,245 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 06:21:47,993 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63863.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:22:00,780 INFO [train.py:901] (3/4) Epoch 8, batch 7300, loss[loss=0.2862, simple_loss=0.3517, pruned_loss=0.1103, over 8526.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3233, pruned_loss=0.09114, over 1609094.21 frames. ], batch size: 28, lr: 9.34e-03, grad_scale: 8.0 +2023-02-06 06:22:12,942 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63897.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:22:14,342 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63899.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:22:36,711 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63931.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:22:37,304 INFO [train.py:901] (3/4) Epoch 8, batch 7350, loss[loss=0.2881, simple_loss=0.3651, pruned_loss=0.1056, over 8325.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3229, pruned_loss=0.09082, over 1607226.36 frames. ], batch size: 25, lr: 9.34e-03, grad_scale: 8.0 +2023-02-06 06:22:38,888 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63934.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:22:42,245 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.863e+02 2.551e+02 3.183e+02 3.767e+02 5.416e+02, threshold=6.365e+02, percent-clipped=0.0 +2023-02-06 06:22:48,931 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63948.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:22:50,009 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.43 vs. limit=5.0 +2023-02-06 06:22:53,096 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3765, 1.9466, 4.5318, 2.0177, 3.9247, 3.8488, 4.0604, 4.0024], + device='cuda:3'), covar=tensor([0.0450, 0.3236, 0.0439, 0.2833, 0.1011, 0.0688, 0.0469, 0.0545], + device='cuda:3'), in_proj_covar=tensor([0.0424, 0.0539, 0.0527, 0.0490, 0.0553, 0.0467, 0.0464, 0.0523], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 06:22:53,935 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63954.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:23:05,953 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 06:23:10,858 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63979.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:23:12,554 INFO [train.py:901] (3/4) Epoch 8, batch 7400, loss[loss=0.2912, simple_loss=0.3489, pruned_loss=0.1167, over 8031.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3231, pruned_loss=0.09082, over 1607200.51 frames. ], batch size: 22, lr: 9.33e-03, grad_scale: 8.0 +2023-02-06 06:23:24,827 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 06:23:48,430 INFO [train.py:901] (3/4) Epoch 8, batch 7450, loss[loss=0.2662, simple_loss=0.3469, pruned_loss=0.09277, over 8448.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3235, pruned_loss=0.09117, over 1608756.90 frames. ], batch size: 27, lr: 9.33e-03, grad_scale: 8.0 +2023-02-06 06:23:53,158 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.642e+02 2.827e+02 3.358e+02 3.935e+02 9.777e+02, threshold=6.715e+02, percent-clipped=5.0 +2023-02-06 06:23:54,693 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64041.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:23:58,118 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64046.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:24:05,490 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 06:24:10,191 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64063.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:24:23,166 INFO [train.py:901] (3/4) Epoch 8, batch 7500, loss[loss=0.2254, simple_loss=0.3009, pruned_loss=0.07498, over 8079.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3229, pruned_loss=0.0905, over 1612404.33 frames. ], batch size: 21, lr: 9.33e-03, grad_scale: 8.0 +2023-02-06 06:24:33,030 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([6.0630, 1.7135, 6.0718, 2.1528, 5.4736, 5.1015, 5.5836, 5.5371], + device='cuda:3'), covar=tensor([0.0301, 0.3998, 0.0240, 0.2859, 0.0857, 0.0633, 0.0333, 0.0390], + device='cuda:3'), in_proj_covar=tensor([0.0420, 0.0538, 0.0528, 0.0489, 0.0552, 0.0467, 0.0466, 0.0524], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 06:24:57,458 INFO [train.py:901] (3/4) Epoch 8, batch 7550, loss[loss=0.2975, simple_loss=0.3626, pruned_loss=0.1162, over 8501.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3239, pruned_loss=0.09162, over 1610502.30 frames. ], batch size: 49, lr: 9.32e-03, grad_scale: 8.0 +2023-02-06 06:25:02,136 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.846e+02 3.017e+02 3.905e+02 4.969e+02 7.546e+02, threshold=7.810e+02, percent-clipped=1.0 +2023-02-06 06:25:06,402 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4480, 1.9908, 3.0928, 2.4523, 2.5987, 2.0983, 1.6878, 1.1800], + device='cuda:3'), covar=tensor([0.2636, 0.3120, 0.0808, 0.1844, 0.1569, 0.1587, 0.1377, 0.3421], + device='cuda:3'), in_proj_covar=tensor([0.0846, 0.0806, 0.0688, 0.0801, 0.0892, 0.0746, 0.0677, 0.0724], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 06:25:11,679 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64153.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:25:13,083 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64155.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:25:24,861 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64172.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:25:28,964 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64178.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:25:30,997 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64180.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:25:32,166 INFO [train.py:901] (3/4) Epoch 8, batch 7600, loss[loss=0.2234, simple_loss=0.3091, pruned_loss=0.06886, over 8106.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3241, pruned_loss=0.09154, over 1612765.85 frames. ], batch size: 23, lr: 9.32e-03, grad_scale: 8.0 +2023-02-06 06:25:40,512 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8654, 1.9953, 1.7514, 2.5621, 1.2376, 1.4125, 1.7679, 2.0644], + device='cuda:3'), covar=tensor([0.0881, 0.1108, 0.1191, 0.0485, 0.1310, 0.1745, 0.1103, 0.0884], + device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0228, 0.0270, 0.0215, 0.0229, 0.0264, 0.0268, 0.0232], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 06:25:49,066 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64207.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:25:58,615 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 +2023-02-06 06:26:06,999 INFO [train.py:901] (3/4) Epoch 8, batch 7650, loss[loss=0.2959, simple_loss=0.3515, pruned_loss=0.1202, over 8646.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3242, pruned_loss=0.09217, over 1611294.18 frames. ], batch size: 39, lr: 9.31e-03, grad_scale: 8.0 +2023-02-06 06:26:11,829 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.722e+02 2.778e+02 3.467e+02 5.154e+02 1.113e+03, threshold=6.933e+02, percent-clipped=3.0 +2023-02-06 06:26:19,377 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64250.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:26:38,168 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64278.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:26:40,887 INFO [train.py:901] (3/4) Epoch 8, batch 7700, loss[loss=0.233, simple_loss=0.3234, pruned_loss=0.07125, over 8193.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3242, pruned_loss=0.09177, over 1614681.38 frames. ], batch size: 23, lr: 9.31e-03, grad_scale: 8.0 +2023-02-06 06:26:45,250 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64287.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:26:56,492 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64302.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:27:08,125 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64319.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:27:10,000 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64322.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:27:10,463 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 06:27:13,360 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64327.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:27:16,536 INFO [train.py:901] (3/4) Epoch 8, batch 7750, loss[loss=0.2055, simple_loss=0.2916, pruned_loss=0.05974, over 8511.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.324, pruned_loss=0.09136, over 1614594.49 frames. ], batch size: 26, lr: 9.31e-03, grad_scale: 8.0 +2023-02-06 06:27:21,028 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.379e+02 2.516e+02 3.070e+02 3.996e+02 6.859e+02, threshold=6.139e+02, percent-clipped=0.0 +2023-02-06 06:27:25,294 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64344.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:27:38,606 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64363.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:27:43,970 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4772, 2.0065, 2.0161, 1.1197, 2.1012, 1.3464, 0.5046, 1.7139], + device='cuda:3'), covar=tensor([0.0299, 0.0132, 0.0137, 0.0300, 0.0179, 0.0527, 0.0453, 0.0133], + device='cuda:3'), in_proj_covar=tensor([0.0359, 0.0284, 0.0236, 0.0346, 0.0273, 0.0441, 0.0336, 0.0317], + device='cuda:3'), out_proj_covar=tensor([1.0808e-04, 8.3007e-05, 6.9521e-05, 1.0235e-04, 8.1936e-05, 1.4384e-04, + 1.0169e-04, 9.4832e-05], device='cuda:3') +2023-02-06 06:27:51,131 INFO [train.py:901] (3/4) Epoch 8, batch 7800, loss[loss=0.2239, simple_loss=0.2926, pruned_loss=0.07761, over 7420.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3237, pruned_loss=0.0918, over 1609694.19 frames. ], batch size: 17, lr: 9.30e-03, grad_scale: 8.0 +2023-02-06 06:27:51,953 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8696, 1.0690, 3.0126, 0.9910, 2.5700, 2.5016, 2.7036, 2.6378], + device='cuda:3'), covar=tensor([0.0726, 0.3590, 0.0788, 0.3355, 0.1427, 0.0913, 0.0701, 0.0828], + device='cuda:3'), in_proj_covar=tensor([0.0428, 0.0544, 0.0536, 0.0496, 0.0560, 0.0473, 0.0471, 0.0528], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 06:27:53,244 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64385.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:27:54,024 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3582, 2.8430, 1.8507, 2.2508, 2.1300, 1.4485, 1.9477, 2.1323], + device='cuda:3'), covar=tensor([0.1434, 0.0344, 0.0988, 0.0638, 0.0596, 0.1353, 0.0873, 0.0837], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0233, 0.0311, 0.0297, 0.0306, 0.0317, 0.0339, 0.0313], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 06:27:58,800 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64393.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:27:59,137 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.29 vs. limit=5.0 +2023-02-06 06:28:25,439 INFO [train.py:901] (3/4) Epoch 8, batch 7850, loss[loss=0.2869, simple_loss=0.3362, pruned_loss=0.1188, over 8260.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3238, pruned_loss=0.09165, over 1611907.19 frames. ], batch size: 24, lr: 9.30e-03, grad_scale: 8.0 +2023-02-06 06:28:30,089 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.075e+02 2.873e+02 3.519e+02 4.505e+02 1.254e+03, threshold=7.037e+02, percent-clipped=6.0 +2023-02-06 06:28:58,104 INFO [train.py:901] (3/4) Epoch 8, batch 7900, loss[loss=0.2678, simple_loss=0.342, pruned_loss=0.09677, over 8253.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3248, pruned_loss=0.09254, over 1611405.43 frames. ], batch size: 24, lr: 9.30e-03, grad_scale: 8.0 +2023-02-06 06:28:58,914 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64483.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:29:10,210 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64500.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:29:12,952 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4811, 2.7861, 1.9309, 2.2294, 2.0907, 1.5134, 1.9139, 2.1036], + device='cuda:3'), covar=tensor([0.1212, 0.0305, 0.0910, 0.0524, 0.0561, 0.1221, 0.0849, 0.0708], + device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0236, 0.0310, 0.0295, 0.0305, 0.0318, 0.0339, 0.0313], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 06:29:17,969 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2023-02-06 06:29:32,351 INFO [train.py:901] (3/4) Epoch 8, batch 7950, loss[loss=0.2657, simple_loss=0.3391, pruned_loss=0.09609, over 8361.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3241, pruned_loss=0.09176, over 1609296.55 frames. ], batch size: 24, lr: 9.29e-03, grad_scale: 8.0 +2023-02-06 06:29:37,062 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.747e+02 2.689e+02 3.383e+02 4.341e+02 8.251e+02, threshold=6.766e+02, percent-clipped=4.0 +2023-02-06 06:29:38,020 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0940, 1.7119, 1.3844, 1.6824, 1.4094, 1.1781, 1.3655, 1.4111], + device='cuda:3'), covar=tensor([0.0947, 0.0462, 0.1080, 0.0467, 0.0622, 0.1265, 0.0772, 0.0709], + device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0235, 0.0309, 0.0293, 0.0303, 0.0318, 0.0337, 0.0311], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 06:29:40,088 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64543.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:29:56,722 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64568.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:30:03,508 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64578.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:30:06,038 INFO [train.py:901] (3/4) Epoch 8, batch 8000, loss[loss=0.1976, simple_loss=0.2745, pruned_loss=0.06037, over 7655.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3231, pruned_loss=0.09177, over 1601001.47 frames. ], batch size: 19, lr: 9.29e-03, grad_scale: 8.0 +2023-02-06 06:30:14,258 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64594.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:30:20,522 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64603.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:30:23,072 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64607.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:30:39,924 INFO [train.py:901] (3/4) Epoch 8, batch 8050, loss[loss=0.2172, simple_loss=0.2807, pruned_loss=0.07689, over 7557.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3227, pruned_loss=0.09272, over 1585011.03 frames. ], batch size: 18, lr: 9.29e-03, grad_scale: 8.0 +2023-02-06 06:30:44,636 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.375e+02 2.434e+02 2.955e+02 3.616e+02 6.730e+02, threshold=5.909e+02, percent-clipped=0.0 +2023-02-06 06:30:51,580 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64649.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:31:13,049 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 06:31:17,613 INFO [train.py:901] (3/4) Epoch 9, batch 0, loss[loss=0.2478, simple_loss=0.3181, pruned_loss=0.08873, over 8622.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3181, pruned_loss=0.08873, over 8622.00 frames. ], batch size: 31, lr: 8.79e-03, grad_scale: 8.0 +2023-02-06 06:31:17,613 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 06:31:28,850 INFO [train.py:935] (3/4) Epoch 9, validation: loss=0.1983, simple_loss=0.2974, pruned_loss=0.04961, over 944034.00 frames. +2023-02-06 06:31:28,851 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 06:31:29,662 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64666.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:31:35,200 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64674.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:31:35,991 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.59 vs. limit=2.0 +2023-02-06 06:31:43,422 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 06:31:56,883 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64707.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:31:58,425 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64709.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:32:02,443 INFO [train.py:901] (3/4) Epoch 9, batch 50, loss[loss=0.2464, simple_loss=0.3218, pruned_loss=0.08553, over 8452.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3225, pruned_loss=0.09218, over 365814.18 frames. ], batch size: 25, lr: 8.79e-03, grad_scale: 8.0 +2023-02-06 06:32:06,633 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64721.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:32:07,956 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3696, 1.9522, 3.2760, 1.1221, 2.4500, 1.7898, 1.6189, 2.0197], + device='cuda:3'), covar=tensor([0.2002, 0.2216, 0.0705, 0.4354, 0.1555, 0.3054, 0.1932, 0.2480], + device='cuda:3'), in_proj_covar=tensor([0.0475, 0.0482, 0.0522, 0.0562, 0.0595, 0.0532, 0.0457, 0.0595], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 06:32:16,334 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 06:32:18,980 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.993e+02 2.818e+02 3.347e+02 4.122e+02 1.189e+03, threshold=6.695e+02, percent-clipped=9.0 +2023-02-06 06:32:30,212 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2657, 1.5650, 1.6071, 0.8990, 1.7128, 1.1993, 0.2336, 1.4324], + device='cuda:3'), covar=tensor([0.0266, 0.0179, 0.0169, 0.0264, 0.0200, 0.0580, 0.0448, 0.0144], + device='cuda:3'), in_proj_covar=tensor([0.0372, 0.0288, 0.0239, 0.0353, 0.0281, 0.0447, 0.0340, 0.0324], + device='cuda:3'), out_proj_covar=tensor([1.1223e-04, 8.4071e-05, 7.0473e-05, 1.0408e-04, 8.4099e-05, 1.4525e-04, + 1.0258e-04, 9.6760e-05], device='cuda:3') +2023-02-06 06:32:31,575 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64756.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:32:36,077 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64763.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:32:37,301 INFO [train.py:901] (3/4) Epoch 9, batch 100, loss[loss=0.2189, simple_loss=0.2966, pruned_loss=0.07055, over 8183.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3237, pruned_loss=0.09064, over 649705.35 frames. ], batch size: 23, lr: 8.78e-03, grad_scale: 8.0 +2023-02-06 06:32:41,538 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64770.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:32:42,074 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 06:32:49,745 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64781.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:33:11,715 INFO [train.py:901] (3/4) Epoch 9, batch 150, loss[loss=0.3216, simple_loss=0.3763, pruned_loss=0.1334, over 8590.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3248, pruned_loss=0.09168, over 864523.48 frames. ], batch size: 31, lr: 8.78e-03, grad_scale: 8.0 +2023-02-06 06:33:16,748 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64822.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:33:20,059 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64827.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:33:27,774 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.896e+02 2.577e+02 3.213e+02 3.848e+02 9.281e+02, threshold=6.425e+02, percent-clipped=3.0 +2023-02-06 06:33:45,633 INFO [train.py:901] (3/4) Epoch 9, batch 200, loss[loss=0.2953, simple_loss=0.3402, pruned_loss=0.1252, over 8080.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3266, pruned_loss=0.09359, over 1035685.55 frames. ], batch size: 21, lr: 8.78e-03, grad_scale: 8.0 +2023-02-06 06:34:21,132 INFO [train.py:901] (3/4) Epoch 9, batch 250, loss[loss=0.2725, simple_loss=0.3388, pruned_loss=0.1031, over 8490.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3254, pruned_loss=0.09227, over 1166272.92 frames. ], batch size: 29, lr: 8.77e-03, grad_scale: 8.0 +2023-02-06 06:34:21,950 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7627, 1.3561, 2.7413, 1.1887, 2.0645, 2.9474, 2.9751, 2.5060], + device='cuda:3'), covar=tensor([0.1087, 0.1481, 0.0445, 0.2188, 0.0857, 0.0350, 0.0546, 0.0769], + device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0282, 0.0241, 0.0273, 0.0254, 0.0222, 0.0295, 0.0282], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 06:34:34,312 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 06:34:36,811 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.096e+02 2.841e+02 3.295e+02 4.179e+02 1.029e+03, threshold=6.590e+02, percent-clipped=5.0 +2023-02-06 06:34:39,027 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64942.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:34:42,836 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 06:34:44,920 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64951.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:34:54,272 INFO [train.py:901] (3/4) Epoch 9, batch 300, loss[loss=0.215, simple_loss=0.2879, pruned_loss=0.07102, over 8092.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3254, pruned_loss=0.09212, over 1272605.52 frames. ], batch size: 21, lr: 8.77e-03, grad_scale: 8.0 +2023-02-06 06:34:54,483 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64965.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:35:09,995 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-06 06:35:11,915 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64990.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:35:15,045 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64994.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:35:26,442 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65010.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:35:30,329 INFO [train.py:901] (3/4) Epoch 9, batch 350, loss[loss=0.2324, simple_loss=0.2952, pruned_loss=0.08478, over 7551.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3244, pruned_loss=0.09168, over 1345956.69 frames. ], batch size: 18, lr: 8.77e-03, grad_scale: 8.0 +2023-02-06 06:35:41,464 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-06 06:35:46,488 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.892e+02 2.570e+02 3.183e+02 3.796e+02 1.000e+03, threshold=6.367e+02, percent-clipped=4.0 +2023-02-06 06:36:03,866 INFO [train.py:901] (3/4) Epoch 9, batch 400, loss[loss=0.2679, simple_loss=0.3452, pruned_loss=0.09531, over 8343.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.325, pruned_loss=0.09139, over 1407080.80 frames. ], batch size: 26, lr: 8.76e-03, grad_scale: 8.0 +2023-02-06 06:36:03,966 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65065.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:36:04,759 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65066.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:36:08,000 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6867, 1.3182, 4.7887, 1.8665, 4.1914, 3.9596, 4.3137, 4.1567], + device='cuda:3'), covar=tensor([0.0397, 0.4281, 0.0398, 0.3051, 0.1094, 0.0785, 0.0499, 0.0570], + device='cuda:3'), in_proj_covar=tensor([0.0429, 0.0547, 0.0535, 0.0500, 0.0565, 0.0474, 0.0476, 0.0532], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 06:36:09,409 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65073.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:36:12,868 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65078.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:36:22,093 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-06 06:36:30,518 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65103.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:36:33,064 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65107.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:36:36,574 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6400, 2.3915, 4.7081, 1.2588, 3.2158, 2.2472, 1.8008, 2.7486], + device='cuda:3'), covar=tensor([0.1627, 0.2054, 0.0591, 0.3768, 0.1297, 0.2547, 0.1613, 0.2324], + device='cuda:3'), in_proj_covar=tensor([0.0474, 0.0486, 0.0523, 0.0561, 0.0599, 0.0533, 0.0456, 0.0594], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 06:36:37,776 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65114.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:36:38,389 INFO [train.py:901] (3/4) Epoch 9, batch 450, loss[loss=0.2459, simple_loss=0.3246, pruned_loss=0.08361, over 8130.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3245, pruned_loss=0.09058, over 1453494.59 frames. ], batch size: 22, lr: 8.76e-03, grad_scale: 8.0 +2023-02-06 06:36:46,192 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65125.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:36:56,315 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.677e+02 2.705e+02 3.323e+02 3.920e+02 9.407e+02, threshold=6.647e+02, percent-clipped=6.0 +2023-02-06 06:37:13,476 INFO [train.py:901] (3/4) Epoch 9, batch 500, loss[loss=0.2407, simple_loss=0.3218, pruned_loss=0.07982, over 8252.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3229, pruned_loss=0.08932, over 1491357.65 frames. ], batch size: 24, lr: 8.76e-03, grad_scale: 8.0 +2023-02-06 06:37:23,270 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65180.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:37:35,049 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65198.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:37:45,913 INFO [train.py:901] (3/4) Epoch 9, batch 550, loss[loss=0.3019, simple_loss=0.3674, pruned_loss=0.1182, over 8344.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3248, pruned_loss=0.09083, over 1524150.17 frames. ], batch size: 25, lr: 8.75e-03, grad_scale: 8.0 +2023-02-06 06:37:51,951 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65222.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:37:52,675 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65223.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:37:56,719 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65229.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:38:03,161 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.816e+02 2.829e+02 3.496e+02 4.355e+02 8.306e+02, threshold=6.991e+02, percent-clipped=2.0 +2023-02-06 06:38:21,250 INFO [train.py:901] (3/4) Epoch 9, batch 600, loss[loss=0.257, simple_loss=0.3404, pruned_loss=0.08677, over 8341.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3242, pruned_loss=0.09033, over 1544764.25 frames. ], batch size: 26, lr: 8.75e-03, grad_scale: 8.0 +2023-02-06 06:38:37,438 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 +2023-02-06 06:38:38,355 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 06:38:39,248 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1805, 2.2152, 1.6857, 1.9827, 1.7429, 1.2807, 1.6532, 1.6307], + device='cuda:3'), covar=tensor([0.1097, 0.0283, 0.0814, 0.0447, 0.0540, 0.1198, 0.0756, 0.0766], + device='cuda:3'), in_proj_covar=tensor([0.0341, 0.0234, 0.0307, 0.0295, 0.0300, 0.0316, 0.0336, 0.0312], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 06:38:43,796 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3793, 1.7046, 4.4210, 1.9204, 2.2497, 4.9459, 4.8692, 4.2097], + device='cuda:3'), covar=tensor([0.0964, 0.1536, 0.0249, 0.1869, 0.1135, 0.0172, 0.0346, 0.0554], + device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0284, 0.0241, 0.0274, 0.0255, 0.0225, 0.0298, 0.0284], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 06:38:46,525 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65303.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:38:54,293 INFO [train.py:901] (3/4) Epoch 9, batch 650, loss[loss=0.2673, simple_loss=0.3317, pruned_loss=0.1014, over 8327.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3253, pruned_loss=0.09149, over 1563064.97 frames. ], batch size: 25, lr: 8.75e-03, grad_scale: 16.0 +2023-02-06 06:38:59,115 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65322.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:39:08,415 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7227, 1.8219, 2.1874, 1.7069, 1.1347, 2.2227, 0.3621, 1.2883], + device='cuda:3'), covar=tensor([0.2738, 0.1429, 0.0480, 0.1745, 0.4973, 0.0531, 0.3653, 0.1962], + device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0153, 0.0092, 0.0202, 0.0242, 0.0095, 0.0156, 0.0156], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0001, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 06:39:10,302 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65338.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:39:10,872 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.745e+02 2.656e+02 3.252e+02 4.080e+02 6.220e+02, threshold=6.503e+02, percent-clipped=0.0 +2023-02-06 06:39:17,065 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65347.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:39:29,074 INFO [train.py:901] (3/4) Epoch 9, batch 700, loss[loss=0.2158, simple_loss=0.3125, pruned_loss=0.05957, over 8192.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3238, pruned_loss=0.08977, over 1580772.13 frames. ], batch size: 23, lr: 8.74e-03, grad_scale: 16.0 +2023-02-06 06:39:41,010 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65381.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:39:57,320 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2514, 1.7121, 1.6428, 0.7102, 1.7310, 1.2183, 0.2694, 1.4908], + device='cuda:3'), covar=tensor([0.0271, 0.0163, 0.0182, 0.0302, 0.0230, 0.0547, 0.0471, 0.0151], + device='cuda:3'), in_proj_covar=tensor([0.0364, 0.0288, 0.0239, 0.0350, 0.0281, 0.0446, 0.0334, 0.0317], + device='cuda:3'), out_proj_covar=tensor([1.0970e-04, 8.4167e-05, 7.0472e-05, 1.0279e-04, 8.4140e-05, 1.4444e-04, + 1.0045e-04, 9.4514e-05], device='cuda:3') +2023-02-06 06:39:57,985 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65406.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:40:03,865 INFO [train.py:901] (3/4) Epoch 9, batch 750, loss[loss=0.2224, simple_loss=0.2996, pruned_loss=0.07254, over 8037.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3227, pruned_loss=0.08896, over 1593322.33 frames. ], batch size: 22, lr: 8.74e-03, grad_scale: 16.0 +2023-02-06 06:40:05,346 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65417.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:40:18,167 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65436.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:40:19,959 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.493e+02 2.803e+02 3.527e+02 4.474e+02 1.505e+03, threshold=7.053e+02, percent-clipped=7.0 +2023-02-06 06:40:21,305 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 06:40:23,501 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5613, 2.7724, 1.8921, 2.2660, 2.2666, 1.4667, 2.0891, 2.0802], + device='cuda:3'), covar=tensor([0.1355, 0.0352, 0.0967, 0.0625, 0.0631, 0.1390, 0.0992, 0.0974], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0239, 0.0313, 0.0301, 0.0306, 0.0322, 0.0345, 0.0317], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 06:40:30,030 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 06:40:30,195 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65453.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:40:36,163 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65461.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:40:38,584 INFO [train.py:901] (3/4) Epoch 9, batch 800, loss[loss=0.2469, simple_loss=0.32, pruned_loss=0.08695, over 8572.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3235, pruned_loss=0.08951, over 1599719.68 frames. ], batch size: 34, lr: 8.74e-03, grad_scale: 16.0 +2023-02-06 06:40:47,463 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65478.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:40:53,641 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65485.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:41:05,678 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65503.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:41:10,500 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65510.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:41:13,593 INFO [train.py:901] (3/4) Epoch 9, batch 850, loss[loss=0.2818, simple_loss=0.3532, pruned_loss=0.1051, over 8577.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3231, pruned_loss=0.08961, over 1605394.64 frames. ], batch size: 49, lr: 8.73e-03, grad_scale: 16.0 +2023-02-06 06:41:13,771 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9010, 3.8967, 2.1664, 2.6930, 2.8707, 1.8328, 2.3309, 2.7223], + device='cuda:3'), covar=tensor([0.1364, 0.0277, 0.0888, 0.0610, 0.0523, 0.1175, 0.0895, 0.0875], + device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0234, 0.0308, 0.0296, 0.0300, 0.0316, 0.0339, 0.0312], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 06:41:25,166 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65532.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:41:29,557 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.696e+02 2.736e+02 3.271e+02 4.209e+02 1.110e+03, threshold=6.542e+02, percent-clipped=5.0 +2023-02-06 06:41:37,091 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65550.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 06:41:47,587 INFO [train.py:901] (3/4) Epoch 9, batch 900, loss[loss=0.237, simple_loss=0.3064, pruned_loss=0.0838, over 8338.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3251, pruned_loss=0.09073, over 1610607.58 frames. ], batch size: 26, lr: 8.73e-03, grad_scale: 16.0 +2023-02-06 06:41:59,861 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4925, 2.9452, 2.3726, 3.7190, 1.6896, 1.8938, 2.2773, 3.0549], + device='cuda:3'), covar=tensor([0.0796, 0.0852, 0.1008, 0.0327, 0.1311, 0.1484, 0.1172, 0.0734], + device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0225, 0.0269, 0.0219, 0.0227, 0.0262, 0.0265, 0.0229], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 06:42:23,181 INFO [train.py:901] (3/4) Epoch 9, batch 950, loss[loss=0.2149, simple_loss=0.3073, pruned_loss=0.06122, over 8284.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3248, pruned_loss=0.09056, over 1616671.68 frames. ], batch size: 23, lr: 8.73e-03, grad_scale: 16.0 +2023-02-06 06:42:39,187 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.656e+02 2.498e+02 3.047e+02 4.041e+02 6.463e+02, threshold=6.094e+02, percent-clipped=0.0 +2023-02-06 06:42:44,697 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65647.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:42:50,357 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 06:42:56,309 INFO [train.py:901] (3/4) Epoch 9, batch 1000, loss[loss=0.2628, simple_loss=0.3448, pruned_loss=0.09033, over 8256.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3235, pruned_loss=0.08997, over 1613195.78 frames. ], batch size: 24, lr: 8.72e-03, grad_scale: 16.0 +2023-02-06 06:43:05,055 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0366, 4.1526, 2.2730, 2.7002, 2.8597, 1.9280, 2.7294, 2.7535], + device='cuda:3'), covar=tensor([0.1469, 0.0243, 0.0932, 0.0698, 0.0592, 0.1203, 0.0962, 0.1003], + device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0234, 0.0312, 0.0299, 0.0304, 0.0318, 0.0341, 0.0315], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 06:43:23,106 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 06:43:27,428 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65709.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:43:31,857 INFO [train.py:901] (3/4) Epoch 9, batch 1050, loss[loss=0.2271, simple_loss=0.3081, pruned_loss=0.07302, over 8501.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3237, pruned_loss=0.08978, over 1616516.99 frames. ], batch size: 28, lr: 8.72e-03, grad_scale: 16.0 +2023-02-06 06:43:32,060 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4985, 3.0664, 2.5648, 3.9670, 1.6770, 1.9412, 2.4619, 3.3211], + device='cuda:3'), covar=tensor([0.0830, 0.0817, 0.0918, 0.0240, 0.1219, 0.1548, 0.1042, 0.0731], + device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0223, 0.0269, 0.0218, 0.0226, 0.0263, 0.0265, 0.0229], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 06:43:35,701 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 06:43:44,978 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65734.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:43:47,995 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.070e+02 2.877e+02 3.398e+02 4.338e+02 8.070e+02, threshold=6.796e+02, percent-clipped=6.0 +2023-02-06 06:43:48,507 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2023-02-06 06:44:03,490 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65762.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:44:05,257 INFO [train.py:901] (3/4) Epoch 9, batch 1100, loss[loss=0.2936, simple_loss=0.3723, pruned_loss=0.1074, over 8556.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3231, pruned_loss=0.08972, over 1617168.99 frames. ], batch size: 31, lr: 8.72e-03, grad_scale: 16.0 +2023-02-06 06:44:20,645 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65788.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:44:38,688 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65813.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:44:39,894 INFO [train.py:901] (3/4) Epoch 9, batch 1150, loss[loss=0.231, simple_loss=0.3129, pruned_loss=0.07456, over 8521.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3228, pruned_loss=0.08999, over 1620874.90 frames. ], batch size: 28, lr: 8.71e-03, grad_scale: 16.0 +2023-02-06 06:44:44,630 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 06:44:56,767 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.732e+02 2.552e+02 3.121e+02 3.966e+02 8.304e+02, threshold=6.242e+02, percent-clipped=2.0 +2023-02-06 06:45:09,008 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65856.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 06:45:14,590 INFO [train.py:901] (3/4) Epoch 9, batch 1200, loss[loss=0.2381, simple_loss=0.304, pruned_loss=0.08614, over 7794.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3222, pruned_loss=0.09004, over 1609627.34 frames. ], batch size: 19, lr: 8.71e-03, grad_scale: 16.0 +2023-02-06 06:45:33,745 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65894.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 06:45:47,641 INFO [train.py:901] (3/4) Epoch 9, batch 1250, loss[loss=0.2465, simple_loss=0.3248, pruned_loss=0.08409, over 8245.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3221, pruned_loss=0.08975, over 1610280.83 frames. ], batch size: 24, lr: 8.71e-03, grad_scale: 16.0 +2023-02-06 06:46:05,032 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.752e+02 2.844e+02 3.477e+02 4.312e+02 8.167e+02, threshold=6.953e+02, percent-clipped=5.0 +2023-02-06 06:46:23,825 INFO [train.py:901] (3/4) Epoch 9, batch 1300, loss[loss=0.2365, simple_loss=0.3084, pruned_loss=0.08227, over 8502.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3227, pruned_loss=0.09052, over 1612133.30 frames. ], batch size: 26, lr: 8.70e-03, grad_scale: 16.0 +2023-02-06 06:46:46,259 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-06 06:46:46,668 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8950, 3.7923, 3.5257, 1.6928, 3.4657, 3.3611, 3.6007, 3.1063], + device='cuda:3'), covar=tensor([0.0872, 0.0615, 0.0917, 0.4453, 0.0827, 0.0999, 0.1079, 0.0950], + device='cuda:3'), in_proj_covar=tensor([0.0438, 0.0348, 0.0362, 0.0457, 0.0353, 0.0332, 0.0355, 0.0302], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 06:46:55,297 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66009.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 06:46:59,214 INFO [train.py:901] (3/4) Epoch 9, batch 1350, loss[loss=0.2478, simple_loss=0.3244, pruned_loss=0.08556, over 8614.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3219, pruned_loss=0.09009, over 1610092.67 frames. ], batch size: 34, lr: 8.70e-03, grad_scale: 8.0 +2023-02-06 06:47:00,007 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3021, 2.9569, 2.2227, 3.8428, 1.7603, 1.7070, 2.0520, 3.1736], + device='cuda:3'), covar=tensor([0.0876, 0.0871, 0.1104, 0.0344, 0.1266, 0.1670, 0.1297, 0.0737], + device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0224, 0.0268, 0.0217, 0.0224, 0.0261, 0.0264, 0.0229], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 06:47:01,385 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66018.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:47:16,348 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6083, 2.5764, 2.0198, 2.2420, 2.2152, 1.5657, 2.1487, 2.1445], + device='cuda:3'), covar=tensor([0.1194, 0.0301, 0.0777, 0.0457, 0.0511, 0.1178, 0.0736, 0.0749], + device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0234, 0.0311, 0.0297, 0.0307, 0.0319, 0.0341, 0.0311], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 06:47:17,555 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.923e+02 2.557e+02 3.336e+02 4.233e+02 1.201e+03, threshold=6.672e+02, percent-clipped=8.0 +2023-02-06 06:47:19,778 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66043.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:47:35,101 INFO [train.py:901] (3/4) Epoch 9, batch 1400, loss[loss=0.2484, simple_loss=0.3272, pruned_loss=0.08478, over 8507.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3241, pruned_loss=0.09138, over 1613515.51 frames. ], batch size: 28, lr: 8.70e-03, grad_scale: 8.0 +2023-02-06 06:48:09,443 INFO [train.py:901] (3/4) Epoch 9, batch 1450, loss[loss=0.3189, simple_loss=0.3717, pruned_loss=0.1331, over 6960.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3236, pruned_loss=0.09183, over 1608848.47 frames. ], batch size: 71, lr: 8.69e-03, grad_scale: 8.0 +2023-02-06 06:48:12,195 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 06:48:26,154 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.917e+02 2.633e+02 3.463e+02 4.686e+02 9.003e+02, threshold=6.925e+02, percent-clipped=5.0 +2023-02-06 06:48:42,248 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66162.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:48:44,178 INFO [train.py:901] (3/4) Epoch 9, batch 1500, loss[loss=0.2245, simple_loss=0.3119, pruned_loss=0.06854, over 8289.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3242, pruned_loss=0.09137, over 1613685.71 frames. ], batch size: 23, lr: 8.69e-03, grad_scale: 8.0 +2023-02-06 06:48:52,878 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66178.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 06:49:05,569 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3072, 2.3348, 4.3895, 2.6966, 4.0290, 3.7868, 4.1309, 4.0499], + device='cuda:3'), covar=tensor([0.0424, 0.2885, 0.0541, 0.2449, 0.0797, 0.0661, 0.0401, 0.0455], + device='cuda:3'), in_proj_covar=tensor([0.0427, 0.0538, 0.0518, 0.0494, 0.0558, 0.0473, 0.0463, 0.0528], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 06:49:07,243 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2023-02-06 06:49:08,956 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66200.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 06:49:18,748 INFO [train.py:901] (3/4) Epoch 9, batch 1550, loss[loss=0.2275, simple_loss=0.3192, pruned_loss=0.06785, over 8498.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3233, pruned_loss=0.09058, over 1612372.40 frames. ], batch size: 28, lr: 8.69e-03, grad_scale: 8.0 +2023-02-06 06:49:28,444 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3187, 1.2385, 1.3656, 1.1771, 0.7833, 1.2264, 1.1945, 0.8971], + device='cuda:3'), covar=tensor([0.0619, 0.1249, 0.1817, 0.1445, 0.0596, 0.1601, 0.0682, 0.0716], + device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0158, 0.0198, 0.0163, 0.0109, 0.0167, 0.0122, 0.0142], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 06:49:35,629 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.465e+02 2.559e+02 2.942e+02 3.565e+02 7.942e+02, threshold=5.885e+02, percent-clipped=2.0 +2023-02-06 06:49:53,203 INFO [train.py:901] (3/4) Epoch 9, batch 1600, loss[loss=0.1983, simple_loss=0.2718, pruned_loss=0.06236, over 7925.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.322, pruned_loss=0.08982, over 1611007.26 frames. ], batch size: 20, lr: 8.68e-03, grad_scale: 8.0 +2023-02-06 06:49:53,436 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66265.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 06:50:02,299 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66276.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:50:11,802 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66290.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 06:50:20,671 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-06 06:50:29,758 INFO [train.py:901] (3/4) Epoch 9, batch 1650, loss[loss=0.2127, simple_loss=0.282, pruned_loss=0.07166, over 7559.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3226, pruned_loss=0.08998, over 1614452.87 frames. ], batch size: 18, lr: 8.68e-03, grad_scale: 8.0 +2023-02-06 06:50:29,933 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66315.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 06:50:46,552 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.749e+02 2.536e+02 3.360e+02 4.258e+02 7.701e+02, threshold=6.719e+02, percent-clipped=5.0 +2023-02-06 06:51:03,467 INFO [train.py:901] (3/4) Epoch 9, batch 1700, loss[loss=0.2584, simple_loss=0.336, pruned_loss=0.09038, over 8296.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3219, pruned_loss=0.08936, over 1616369.62 frames. ], batch size: 23, lr: 8.68e-03, grad_scale: 8.0 +2023-02-06 06:51:39,908 INFO [train.py:901] (3/4) Epoch 9, batch 1750, loss[loss=0.262, simple_loss=0.3416, pruned_loss=0.09119, over 8770.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3204, pruned_loss=0.08818, over 1617884.81 frames. ], batch size: 30, lr: 8.67e-03, grad_scale: 8.0 +2023-02-06 06:51:57,455 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.817e+02 2.942e+02 3.542e+02 4.261e+02 7.419e+02, threshold=7.084e+02, percent-clipped=2.0 +2023-02-06 06:52:04,253 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.2177, 5.2407, 4.6484, 2.3743, 4.7494, 4.8146, 4.9675, 4.2785], + device='cuda:3'), covar=tensor([0.0646, 0.0416, 0.0927, 0.4321, 0.0671, 0.0654, 0.0963, 0.0695], + device='cuda:3'), in_proj_covar=tensor([0.0444, 0.0345, 0.0358, 0.0451, 0.0356, 0.0336, 0.0358, 0.0302], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 06:52:13,907 INFO [train.py:901] (3/4) Epoch 9, batch 1800, loss[loss=0.2405, simple_loss=0.3135, pruned_loss=0.08378, over 8583.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3216, pruned_loss=0.08874, over 1621727.22 frames. ], batch size: 39, lr: 8.67e-03, grad_scale: 8.0 +2023-02-06 06:52:42,873 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66506.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:52:48,997 INFO [train.py:901] (3/4) Epoch 9, batch 1850, loss[loss=0.2359, simple_loss=0.3097, pruned_loss=0.08109, over 8108.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3208, pruned_loss=0.08824, over 1621980.26 frames. ], batch size: 23, lr: 8.67e-03, grad_scale: 8.0 +2023-02-06 06:52:53,698 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66522.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 06:53:05,497 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66539.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 06:53:05,997 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.478e+02 2.848e+02 3.228e+02 4.154e+02 1.120e+03, threshold=6.457e+02, percent-clipped=1.0 +2023-02-06 06:53:13,960 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6898, 1.2813, 1.5297, 1.2344, 1.0598, 1.3027, 1.5406, 1.4289], + device='cuda:3'), covar=tensor([0.0573, 0.1273, 0.1675, 0.1352, 0.0566, 0.1495, 0.0677, 0.0595], + device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0158, 0.0199, 0.0162, 0.0110, 0.0167, 0.0122, 0.0141], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 06:53:22,983 INFO [train.py:901] (3/4) Epoch 9, batch 1900, loss[loss=0.2525, simple_loss=0.3278, pruned_loss=0.0886, over 8455.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3216, pruned_loss=0.08949, over 1616646.04 frames. ], batch size: 27, lr: 8.66e-03, grad_scale: 8.0 +2023-02-06 06:53:27,132 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66571.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 06:53:43,851 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66596.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 06:53:47,139 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 06:53:48,636 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6863, 4.6033, 4.2062, 2.0260, 4.1375, 4.1202, 4.3522, 3.7197], + device='cuda:3'), covar=tensor([0.0727, 0.0494, 0.0939, 0.4418, 0.0746, 0.0804, 0.0972, 0.0771], + device='cuda:3'), in_proj_covar=tensor([0.0443, 0.0346, 0.0361, 0.0455, 0.0361, 0.0337, 0.0359, 0.0302], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 06:53:56,562 INFO [train.py:901] (3/4) Epoch 9, batch 1950, loss[loss=0.2216, simple_loss=0.2849, pruned_loss=0.07912, over 7198.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.322, pruned_loss=0.08944, over 1615652.52 frames. ], batch size: 16, lr: 8.66e-03, grad_scale: 8.0 +2023-02-06 06:53:58,607 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 06:54:00,037 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66620.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:54:00,840 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66621.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:54:12,900 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66637.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 06:54:14,621 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.940e+02 2.852e+02 3.410e+02 4.369e+02 9.021e+02, threshold=6.820e+02, percent-clipped=7.0 +2023-02-06 06:54:20,054 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 06:54:32,103 INFO [train.py:901] (3/4) Epoch 9, batch 2000, loss[loss=0.2334, simple_loss=0.3144, pruned_loss=0.07621, over 8360.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3206, pruned_loss=0.08803, over 1615020.32 frames. ], batch size: 26, lr: 8.66e-03, grad_scale: 8.0 +2023-02-06 06:55:06,982 INFO [train.py:901] (3/4) Epoch 9, batch 2050, loss[loss=0.2042, simple_loss=0.294, pruned_loss=0.05718, over 8034.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3218, pruned_loss=0.08847, over 1620354.05 frames. ], batch size: 22, lr: 8.65e-03, grad_scale: 8.0 +2023-02-06 06:55:12,325 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1559, 1.3513, 4.3001, 1.7281, 3.8359, 3.5741, 3.8501, 3.7648], + device='cuda:3'), covar=tensor([0.0466, 0.3890, 0.0473, 0.2736, 0.1047, 0.0780, 0.0517, 0.0610], + device='cuda:3'), in_proj_covar=tensor([0.0435, 0.0551, 0.0538, 0.0503, 0.0572, 0.0486, 0.0478, 0.0543], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 06:55:20,407 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66735.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:55:23,653 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.024e+02 2.763e+02 3.349e+02 4.333e+02 1.017e+03, threshold=6.698e+02, percent-clipped=4.0 +2023-02-06 06:55:29,223 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1196, 1.3834, 3.1566, 1.3983, 2.1329, 3.5103, 3.5426, 2.9691], + device='cuda:3'), covar=tensor([0.0937, 0.1685, 0.0367, 0.2061, 0.1033, 0.0284, 0.0505, 0.0704], + device='cuda:3'), in_proj_covar=tensor([0.0256, 0.0288, 0.0246, 0.0279, 0.0263, 0.0230, 0.0305, 0.0291], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 06:55:31,185 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66749.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 06:55:42,432 INFO [train.py:901] (3/4) Epoch 9, batch 2100, loss[loss=0.2594, simple_loss=0.3354, pruned_loss=0.09168, over 7053.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3223, pruned_loss=0.08911, over 1619357.03 frames. ], batch size: 71, lr: 8.65e-03, grad_scale: 8.0 +2023-02-06 06:56:17,429 INFO [train.py:901] (3/4) Epoch 9, batch 2150, loss[loss=0.2384, simple_loss=0.3095, pruned_loss=0.08361, over 7660.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3236, pruned_loss=0.08975, over 1620953.95 frames. ], batch size: 19, lr: 8.65e-03, grad_scale: 8.0 +2023-02-06 06:56:34,546 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 2.810e+02 3.362e+02 4.511e+02 1.000e+03, threshold=6.724e+02, percent-clipped=7.0 +2023-02-06 06:56:53,076 INFO [train.py:901] (3/4) Epoch 9, batch 2200, loss[loss=0.2868, simple_loss=0.3512, pruned_loss=0.1112, over 8623.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3224, pruned_loss=0.08952, over 1617156.39 frames. ], batch size: 34, lr: 8.64e-03, grad_scale: 8.0 +2023-02-06 06:57:01,059 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66877.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:57:05,629 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66883.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 06:57:09,744 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2322, 1.8948, 3.0731, 2.3176, 2.5120, 2.0589, 1.6328, 1.2626], + device='cuda:3'), covar=tensor([0.3447, 0.3761, 0.0950, 0.2532, 0.1954, 0.1921, 0.1641, 0.3919], + device='cuda:3'), in_proj_covar=tensor([0.0842, 0.0809, 0.0694, 0.0805, 0.0896, 0.0756, 0.0686, 0.0731], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 06:57:12,276 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66893.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 06:57:18,130 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66902.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:57:26,998 INFO [train.py:901] (3/4) Epoch 9, batch 2250, loss[loss=0.2265, simple_loss=0.2967, pruned_loss=0.07814, over 7922.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3218, pruned_loss=0.08925, over 1613393.97 frames. ], batch size: 20, lr: 8.64e-03, grad_scale: 8.0 +2023-02-06 06:57:29,218 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66918.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 06:57:43,689 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.550e+02 2.716e+02 3.375e+02 4.203e+02 7.579e+02, threshold=6.750e+02, percent-clipped=1.0 +2023-02-06 06:58:00,318 INFO [train.py:901] (3/4) Epoch 9, batch 2300, loss[loss=0.2269, simple_loss=0.3189, pruned_loss=0.06746, over 8200.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3206, pruned_loss=0.08878, over 1610820.60 frames. ], batch size: 23, lr: 8.64e-03, grad_scale: 8.0 +2023-02-06 06:58:07,513 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66975.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:58:19,742 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66991.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:58:21,141 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4783, 1.7745, 2.8479, 1.2732, 2.1570, 1.9049, 1.4476, 1.8160], + device='cuda:3'), covar=tensor([0.1535, 0.1875, 0.0642, 0.3463, 0.1250, 0.2465, 0.1679, 0.1849], + device='cuda:3'), in_proj_covar=tensor([0.0477, 0.0488, 0.0534, 0.0570, 0.0603, 0.0535, 0.0463, 0.0601], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 06:58:24,432 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66998.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 06:58:36,102 INFO [train.py:901] (3/4) Epoch 9, batch 2350, loss[loss=0.2336, simple_loss=0.3256, pruned_loss=0.07085, over 8234.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3204, pruned_loss=0.08876, over 1612455.41 frames. ], batch size: 22, lr: 8.63e-03, grad_scale: 8.0 +2023-02-06 06:58:37,002 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67016.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:58:53,554 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.636e+02 2.755e+02 3.236e+02 4.430e+02 1.005e+03, threshold=6.472e+02, percent-clipped=3.0 +2023-02-06 06:59:10,050 INFO [train.py:901] (3/4) Epoch 9, batch 2400, loss[loss=0.243, simple_loss=0.3179, pruned_loss=0.08411, over 8466.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3208, pruned_loss=0.08924, over 1608778.20 frames. ], batch size: 25, lr: 8.63e-03, grad_scale: 8.0 +2023-02-06 06:59:10,297 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3333, 1.6793, 1.6592, 0.8849, 1.7219, 1.2633, 0.3000, 1.4803], + device='cuda:3'), covar=tensor([0.0293, 0.0198, 0.0153, 0.0275, 0.0178, 0.0554, 0.0477, 0.0148], + device='cuda:3'), in_proj_covar=tensor([0.0367, 0.0292, 0.0243, 0.0355, 0.0284, 0.0441, 0.0336, 0.0321], + device='cuda:3'), out_proj_covar=tensor([1.0949e-04, 8.5292e-05, 7.1180e-05, 1.0401e-04, 8.4612e-05, 1.4138e-04, + 1.0074e-04, 9.4993e-05], device='cuda:3') +2023-02-06 06:59:28,534 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67093.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 06:59:35,235 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67101.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 06:59:44,439 INFO [train.py:901] (3/4) Epoch 9, batch 2450, loss[loss=0.2594, simple_loss=0.3296, pruned_loss=0.09461, over 7977.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3212, pruned_loss=0.08936, over 1611804.76 frames. ], batch size: 21, lr: 8.63e-03, grad_scale: 8.0 +2023-02-06 07:00:02,010 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.801e+02 2.787e+02 3.467e+02 4.148e+02 8.119e+02, threshold=6.934e+02, percent-clipped=3.0 +2023-02-06 07:00:17,987 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8587, 2.0183, 1.7213, 2.5580, 1.3885, 1.3622, 1.7281, 2.0502], + device='cuda:3'), covar=tensor([0.0850, 0.0990, 0.1231, 0.0480, 0.1231, 0.1829, 0.1105, 0.0923], + device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0226, 0.0265, 0.0218, 0.0224, 0.0264, 0.0267, 0.0231], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 07:00:18,493 INFO [train.py:901] (3/4) Epoch 9, batch 2500, loss[loss=0.2595, simple_loss=0.3371, pruned_loss=0.09094, over 8466.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3229, pruned_loss=0.0905, over 1614758.86 frames. ], batch size: 25, lr: 8.62e-03, grad_scale: 8.0 +2023-02-06 07:00:48,181 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67208.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 07:00:52,861 INFO [train.py:901] (3/4) Epoch 9, batch 2550, loss[loss=0.253, simple_loss=0.3297, pruned_loss=0.08814, over 8331.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3226, pruned_loss=0.09047, over 1611660.83 frames. ], batch size: 25, lr: 8.62e-03, grad_scale: 8.0 +2023-02-06 07:01:00,613 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7598, 2.2114, 3.4077, 2.7451, 3.0258, 2.4161, 2.0650, 2.0979], + device='cuda:3'), covar=tensor([0.2425, 0.3513, 0.0949, 0.2292, 0.1588, 0.1702, 0.1296, 0.3287], + device='cuda:3'), in_proj_covar=tensor([0.0844, 0.0811, 0.0693, 0.0807, 0.0894, 0.0752, 0.0684, 0.0733], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 07:01:12,346 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.684e+02 2.679e+02 3.405e+02 4.203e+02 8.726e+02, threshold=6.810e+02, percent-clipped=2.0 +2023-02-06 07:01:19,266 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4224, 1.7037, 2.8722, 1.2096, 2.1253, 1.8052, 1.5717, 2.0116], + device='cuda:3'), covar=tensor([0.1520, 0.1920, 0.0507, 0.3460, 0.1310, 0.2502, 0.1562, 0.1752], + device='cuda:3'), in_proj_covar=tensor([0.0477, 0.0489, 0.0531, 0.0566, 0.0602, 0.0534, 0.0463, 0.0606], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 07:01:19,289 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1630, 1.7425, 2.6363, 2.1645, 2.3578, 1.9731, 1.6274, 1.1139], + device='cuda:3'), covar=tensor([0.3053, 0.3405, 0.0886, 0.1962, 0.1423, 0.1950, 0.1477, 0.3165], + device='cuda:3'), in_proj_covar=tensor([0.0850, 0.0814, 0.0696, 0.0809, 0.0896, 0.0756, 0.0687, 0.0734], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 07:01:21,215 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3264, 1.6603, 2.8248, 1.1150, 1.9811, 1.7285, 1.4660, 1.8632], + device='cuda:3'), covar=tensor([0.1749, 0.2005, 0.0759, 0.3661, 0.1494, 0.2649, 0.1655, 0.2021], + device='cuda:3'), in_proj_covar=tensor([0.0476, 0.0489, 0.0530, 0.0565, 0.0601, 0.0533, 0.0462, 0.0605], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 07:01:21,878 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67254.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 07:01:28,483 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5858, 1.8760, 3.3787, 1.3082, 2.2885, 1.9750, 1.6702, 2.0693], + device='cuda:3'), covar=tensor([0.1512, 0.2008, 0.0597, 0.3526, 0.1333, 0.2502, 0.1517, 0.2151], + device='cuda:3'), in_proj_covar=tensor([0.0476, 0.0489, 0.0528, 0.0565, 0.0600, 0.0533, 0.0461, 0.0604], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 07:01:29,607 INFO [train.py:901] (3/4) Epoch 9, batch 2600, loss[loss=0.2687, simple_loss=0.3318, pruned_loss=0.1028, over 8024.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.325, pruned_loss=0.09185, over 1615802.26 frames. ], batch size: 22, lr: 8.62e-03, grad_scale: 8.0 +2023-02-06 07:01:32,416 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4723, 1.5143, 2.3495, 1.1851, 2.1602, 2.5145, 2.6073, 2.1753], + device='cuda:3'), covar=tensor([0.0935, 0.1063, 0.0472, 0.1840, 0.0619, 0.0377, 0.0546, 0.0755], + device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0282, 0.0244, 0.0274, 0.0256, 0.0227, 0.0302, 0.0286], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 07:01:39,250 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67279.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 07:01:46,306 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 +2023-02-06 07:02:03,433 INFO [train.py:901] (3/4) Epoch 9, batch 2650, loss[loss=0.2536, simple_loss=0.3288, pruned_loss=0.08921, over 8740.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3244, pruned_loss=0.09179, over 1612207.07 frames. ], batch size: 30, lr: 8.62e-03, grad_scale: 8.0 +2023-02-06 07:02:06,282 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67319.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:02:18,506 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8428, 1.5453, 1.7592, 1.4095, 1.1596, 1.4189, 1.6983, 1.5018], + device='cuda:3'), covar=tensor([0.0527, 0.1143, 0.1575, 0.1267, 0.0533, 0.1406, 0.0621, 0.0580], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0158, 0.0195, 0.0161, 0.0108, 0.0166, 0.0119, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008, 0.0005, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 07:02:20,594 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6601, 1.5540, 4.4236, 1.8276, 2.4131, 5.1081, 5.0245, 4.4314], + device='cuda:3'), covar=tensor([0.1034, 0.1619, 0.0253, 0.2050, 0.1023, 0.0195, 0.0310, 0.0510], + device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0280, 0.0242, 0.0274, 0.0255, 0.0225, 0.0302, 0.0285], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 07:02:21,820 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.856e+02 2.739e+02 3.376e+02 4.238e+02 9.756e+02, threshold=6.752e+02, percent-clipped=4.0 +2023-02-06 07:02:39,314 INFO [train.py:901] (3/4) Epoch 9, batch 2700, loss[loss=0.32, simple_loss=0.3827, pruned_loss=0.1287, over 8262.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3234, pruned_loss=0.09112, over 1612545.98 frames. ], batch size: 24, lr: 8.61e-03, grad_scale: 8.0 +2023-02-06 07:02:55,842 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9399, 2.4005, 4.5780, 1.4313, 3.2922, 2.3789, 1.9610, 2.9404], + device='cuda:3'), covar=tensor([0.1514, 0.2019, 0.0677, 0.3586, 0.1302, 0.2411, 0.1509, 0.2176], + device='cuda:3'), in_proj_covar=tensor([0.0481, 0.0493, 0.0532, 0.0571, 0.0605, 0.0540, 0.0465, 0.0607], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 07:03:13,103 INFO [train.py:901] (3/4) Epoch 9, batch 2750, loss[loss=0.3373, simple_loss=0.385, pruned_loss=0.1448, over 7419.00 frames. ], tot_loss[loss=0.253, simple_loss=0.324, pruned_loss=0.09106, over 1612166.35 frames. ], batch size: 72, lr: 8.61e-03, grad_scale: 8.0 +2023-02-06 07:03:13,294 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7582, 3.3151, 2.6513, 4.0098, 2.1832, 2.5298, 2.4659, 3.5445], + device='cuda:3'), covar=tensor([0.0684, 0.0746, 0.0892, 0.0249, 0.1025, 0.1153, 0.1155, 0.0557], + device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0224, 0.0265, 0.0217, 0.0224, 0.0262, 0.0266, 0.0228], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 07:03:26,078 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67434.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:03:29,963 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.803e+02 2.867e+02 3.446e+02 4.196e+02 9.783e+02, threshold=6.892e+02, percent-clipped=3.0 +2023-02-06 07:03:30,377 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 +2023-02-06 07:03:34,268 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67445.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:03:48,012 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67464.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 07:03:48,381 INFO [train.py:901] (3/4) Epoch 9, batch 2800, loss[loss=0.2879, simple_loss=0.355, pruned_loss=0.1104, over 8027.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3231, pruned_loss=0.09082, over 1613090.43 frames. ], batch size: 22, lr: 8.61e-03, grad_scale: 8.0 +2023-02-06 07:04:05,334 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67489.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 07:04:23,680 INFO [train.py:901] (3/4) Epoch 9, batch 2850, loss[loss=0.2538, simple_loss=0.325, pruned_loss=0.09131, over 8660.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3228, pruned_loss=0.09044, over 1612429.36 frames. ], batch size: 39, lr: 8.60e-03, grad_scale: 8.0 +2023-02-06 07:04:34,549 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67531.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:04:40,475 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.340e+02 2.765e+02 3.269e+02 4.105e+02 6.649e+02, threshold=6.538e+02, percent-clipped=0.0 +2023-02-06 07:04:55,092 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67560.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:04:58,190 INFO [train.py:901] (3/4) Epoch 9, batch 2900, loss[loss=0.2024, simple_loss=0.2749, pruned_loss=0.06495, over 7803.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3229, pruned_loss=0.09105, over 1608013.95 frames. ], batch size: 20, lr: 8.60e-03, grad_scale: 8.0 +2023-02-06 07:05:23,402 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2023-02-06 07:05:24,273 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 07:05:33,887 INFO [train.py:901] (3/4) Epoch 9, batch 2950, loss[loss=0.2324, simple_loss=0.2852, pruned_loss=0.08978, over 7423.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3217, pruned_loss=0.09003, over 1610732.32 frames. ], batch size: 17, lr: 8.60e-03, grad_scale: 8.0 +2023-02-06 07:05:41,791 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9906, 1.3699, 3.2416, 1.4438, 2.2515, 3.5142, 3.5421, 3.0377], + device='cuda:3'), covar=tensor([0.1007, 0.1636, 0.0421, 0.2064, 0.0945, 0.0275, 0.0527, 0.0617], + device='cuda:3'), in_proj_covar=tensor([0.0254, 0.0285, 0.0248, 0.0279, 0.0261, 0.0228, 0.0309, 0.0293], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 07:05:51,260 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.115e+02 2.827e+02 3.390e+02 4.435e+02 7.404e+02, threshold=6.780e+02, percent-clipped=4.0 +2023-02-06 07:06:07,060 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0290, 3.1119, 2.3177, 2.3647, 2.4832, 2.0047, 2.3321, 2.7436], + device='cuda:3'), covar=tensor([0.1296, 0.0267, 0.0766, 0.0703, 0.0543, 0.1060, 0.0961, 0.0983], + device='cuda:3'), in_proj_covar=tensor([0.0341, 0.0230, 0.0307, 0.0294, 0.0301, 0.0315, 0.0337, 0.0308], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 07:06:08,213 INFO [train.py:901] (3/4) Epoch 9, batch 3000, loss[loss=0.2016, simple_loss=0.2637, pruned_loss=0.06979, over 7708.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3208, pruned_loss=0.08931, over 1608863.94 frames. ], batch size: 18, lr: 8.59e-03, grad_scale: 8.0 +2023-02-06 07:06:08,213 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 07:06:18,918 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3498, 1.7696, 2.7001, 1.1344, 2.0433, 1.5689, 1.5463, 1.8880], + device='cuda:3'), covar=tensor([0.1643, 0.2093, 0.0732, 0.3788, 0.1429, 0.2770, 0.1716, 0.1995], + device='cuda:3'), in_proj_covar=tensor([0.0475, 0.0494, 0.0527, 0.0568, 0.0601, 0.0537, 0.0458, 0.0602], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 07:06:20,340 INFO [train.py:935] (3/4) Epoch 9, validation: loss=0.1965, simple_loss=0.2957, pruned_loss=0.04864, over 944034.00 frames. +2023-02-06 07:06:20,341 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 07:06:37,452 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67690.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:06:37,688 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-06 07:06:43,356 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67698.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:06:52,181 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67710.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:06:55,496 INFO [train.py:901] (3/4) Epoch 9, batch 3050, loss[loss=0.2116, simple_loss=0.2729, pruned_loss=0.07518, over 7220.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3203, pruned_loss=0.0892, over 1607301.71 frames. ], batch size: 16, lr: 8.59e-03, grad_scale: 8.0 +2023-02-06 07:06:55,710 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67715.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:07:13,229 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.763e+02 2.629e+02 3.194e+02 3.976e+02 7.575e+02, threshold=6.387e+02, percent-clipped=1.0 +2023-02-06 07:07:29,768 INFO [train.py:901] (3/4) Epoch 9, batch 3100, loss[loss=0.2605, simple_loss=0.329, pruned_loss=0.09602, over 8304.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3199, pruned_loss=0.08894, over 1605443.91 frames. ], batch size: 23, lr: 8.59e-03, grad_scale: 8.0 +2023-02-06 07:07:42,788 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8062, 4.0244, 2.5397, 2.7268, 2.8673, 1.8413, 2.7479, 3.0910], + device='cuda:3'), covar=tensor([0.1482, 0.0222, 0.0843, 0.0723, 0.0592, 0.1292, 0.0926, 0.0828], + device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0231, 0.0309, 0.0296, 0.0304, 0.0318, 0.0339, 0.0309], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 07:08:04,168 INFO [train.py:901] (3/4) Epoch 9, batch 3150, loss[loss=0.2681, simple_loss=0.3407, pruned_loss=0.09773, over 8363.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3209, pruned_loss=0.08907, over 1608881.25 frames. ], batch size: 24, lr: 8.58e-03, grad_scale: 8.0 +2023-02-06 07:08:05,060 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67816.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:08:21,137 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.202e+02 2.768e+02 3.401e+02 4.235e+02 8.418e+02, threshold=6.801e+02, percent-clipped=5.0 +2023-02-06 07:08:21,987 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67841.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:08:37,978 INFO [train.py:901] (3/4) Epoch 9, batch 3200, loss[loss=0.2833, simple_loss=0.3549, pruned_loss=0.1059, over 8450.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3225, pruned_loss=0.08973, over 1613188.17 frames. ], batch size: 49, lr: 8.58e-03, grad_scale: 8.0 +2023-02-06 07:08:43,401 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3867, 4.3524, 3.9052, 1.9934, 3.8667, 3.8918, 4.0329, 3.6554], + device='cuda:3'), covar=tensor([0.0742, 0.0590, 0.0987, 0.4600, 0.0754, 0.1012, 0.1305, 0.0904], + device='cuda:3'), in_proj_covar=tensor([0.0435, 0.0344, 0.0359, 0.0451, 0.0356, 0.0336, 0.0354, 0.0300], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 07:08:44,734 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67875.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:09:12,186 INFO [train.py:901] (3/4) Epoch 9, batch 3250, loss[loss=0.2209, simple_loss=0.2963, pruned_loss=0.07274, over 7789.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3221, pruned_loss=0.08993, over 1611347.16 frames. ], batch size: 19, lr: 8.58e-03, grad_scale: 8.0 +2023-02-06 07:09:29,464 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.446e+02 2.709e+02 3.362e+02 4.203e+02 8.128e+02, threshold=6.724e+02, percent-clipped=5.0 +2023-02-06 07:09:46,666 INFO [train.py:901] (3/4) Epoch 9, batch 3300, loss[loss=0.2314, simple_loss=0.2948, pruned_loss=0.08398, over 7259.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3209, pruned_loss=0.08917, over 1608617.31 frames. ], batch size: 16, lr: 8.57e-03, grad_scale: 8.0 +2023-02-06 07:10:04,408 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67990.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:10:10,480 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67999.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:10:22,543 INFO [train.py:901] (3/4) Epoch 9, batch 3350, loss[loss=0.2916, simple_loss=0.35, pruned_loss=0.1166, over 8567.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3213, pruned_loss=0.08916, over 1612179.76 frames. ], batch size: 31, lr: 8.57e-03, grad_scale: 16.0 +2023-02-06 07:10:22,685 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3195, 1.3020, 4.5159, 1.6688, 3.9093, 3.7626, 4.0380, 3.9188], + device='cuda:3'), covar=tensor([0.0535, 0.4279, 0.0469, 0.3465, 0.1234, 0.0844, 0.0513, 0.0621], + device='cuda:3'), in_proj_covar=tensor([0.0447, 0.0553, 0.0546, 0.0515, 0.0576, 0.0489, 0.0478, 0.0546], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 07:10:39,235 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.754e+02 2.493e+02 3.108e+02 4.287e+02 1.101e+03, threshold=6.217e+02, percent-clipped=5.0 +2023-02-06 07:10:40,594 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68042.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:10:49,106 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68054.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:10:56,400 INFO [train.py:901] (3/4) Epoch 9, batch 3400, loss[loss=0.3245, simple_loss=0.3664, pruned_loss=0.1413, over 6930.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3212, pruned_loss=0.08935, over 1610769.84 frames. ], batch size: 72, lr: 8.57e-03, grad_scale: 16.0 +2023-02-06 07:11:24,316 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68105.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 07:11:26,290 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6102, 2.8931, 1.8608, 2.1662, 2.3199, 1.5913, 1.9987, 2.1068], + device='cuda:3'), covar=tensor([0.1189, 0.0272, 0.0924, 0.0565, 0.0553, 0.1223, 0.0895, 0.0861], + device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0235, 0.0310, 0.0295, 0.0305, 0.0321, 0.0343, 0.0313], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 07:11:30,826 INFO [train.py:901] (3/4) Epoch 9, batch 3450, loss[loss=0.2081, simple_loss=0.2803, pruned_loss=0.06793, over 7977.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3218, pruned_loss=0.09003, over 1607125.55 frames. ], batch size: 21, lr: 8.56e-03, grad_scale: 16.0 +2023-02-06 07:11:47,216 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.94 vs. limit=5.0 +2023-02-06 07:11:48,144 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.537e+02 2.585e+02 3.242e+02 3.955e+02 1.617e+03, threshold=6.484e+02, percent-clipped=7.0 +2023-02-06 07:11:48,434 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8603, 2.3253, 3.6508, 2.8159, 3.1555, 2.4596, 2.0328, 1.7926], + device='cuda:3'), covar=tensor([0.2821, 0.3363, 0.0858, 0.2143, 0.1729, 0.1736, 0.1363, 0.3807], + device='cuda:3'), in_proj_covar=tensor([0.0856, 0.0816, 0.0691, 0.0809, 0.0900, 0.0760, 0.0687, 0.0736], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 07:11:57,309 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 07:11:59,840 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68157.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:12:05,799 INFO [train.py:901] (3/4) Epoch 9, batch 3500, loss[loss=0.2783, simple_loss=0.3384, pruned_loss=0.1091, over 7805.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3223, pruned_loss=0.09024, over 1604772.22 frames. ], batch size: 20, lr: 8.56e-03, grad_scale: 8.0 +2023-02-06 07:12:08,688 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68169.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:12:11,464 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5129, 2.0582, 3.5993, 1.2456, 2.5966, 1.9741, 1.6149, 2.4566], + device='cuda:3'), covar=tensor([0.1575, 0.2057, 0.0593, 0.3752, 0.1433, 0.2733, 0.1638, 0.2074], + device='cuda:3'), in_proj_covar=tensor([0.0482, 0.0496, 0.0530, 0.0574, 0.0603, 0.0545, 0.0463, 0.0604], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 07:12:17,932 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 07:12:40,959 INFO [train.py:901] (3/4) Epoch 9, batch 3550, loss[loss=0.3037, simple_loss=0.365, pruned_loss=0.1211, over 8338.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3225, pruned_loss=0.0902, over 1604781.32 frames. ], batch size: 26, lr: 8.56e-03, grad_scale: 8.0 +2023-02-06 07:12:58,952 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.662e+02 2.838e+02 3.387e+02 4.304e+02 7.616e+02, threshold=6.774e+02, percent-clipped=6.0 +2023-02-06 07:13:02,582 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68246.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:13:14,901 INFO [train.py:901] (3/4) Epoch 9, batch 3600, loss[loss=0.2375, simple_loss=0.3107, pruned_loss=0.08217, over 8080.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3221, pruned_loss=0.08991, over 1608283.04 frames. ], batch size: 21, lr: 8.56e-03, grad_scale: 8.0 +2023-02-06 07:13:19,106 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68271.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:13:49,589 INFO [train.py:901] (3/4) Epoch 9, batch 3650, loss[loss=0.2427, simple_loss=0.316, pruned_loss=0.08469, over 8298.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3226, pruned_loss=0.08982, over 1610754.82 frames. ], batch size: 23, lr: 8.55e-03, grad_scale: 8.0 +2023-02-06 07:14:08,230 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.755e+02 2.637e+02 3.214e+02 4.100e+02 7.421e+02, threshold=6.428e+02, percent-clipped=2.0 +2023-02-06 07:14:09,703 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68343.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:14:18,268 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 07:14:25,009 INFO [train.py:901] (3/4) Epoch 9, batch 3700, loss[loss=0.2572, simple_loss=0.3229, pruned_loss=0.09579, over 8245.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3235, pruned_loss=0.09024, over 1613011.38 frames. ], batch size: 22, lr: 8.55e-03, grad_scale: 8.0 +2023-02-06 07:14:26,544 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0987, 1.4187, 1.5327, 1.3057, 1.1612, 1.4369, 1.8172, 1.4701], + device='cuda:3'), covar=tensor([0.0500, 0.1189, 0.1748, 0.1375, 0.0562, 0.1458, 0.0627, 0.0612], + device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0158, 0.0197, 0.0161, 0.0109, 0.0166, 0.0120, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 07:14:57,598 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68413.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:14:58,733 INFO [train.py:901] (3/4) Epoch 9, batch 3750, loss[loss=0.2637, simple_loss=0.3183, pruned_loss=0.1045, over 7423.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3234, pruned_loss=0.09017, over 1616159.12 frames. ], batch size: 17, lr: 8.55e-03, grad_scale: 8.0 +2023-02-06 07:15:00,195 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68417.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:15:06,070 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68425.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:15:08,022 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2915, 1.4387, 1.3090, 1.8627, 0.6776, 1.1496, 1.2606, 1.4684], + device='cuda:3'), covar=tensor([0.0997, 0.0936, 0.1247, 0.0563, 0.1330, 0.1581, 0.0962, 0.0852], + device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0225, 0.0263, 0.0219, 0.0226, 0.0261, 0.0267, 0.0228], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 07:15:14,911 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68438.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:15:16,797 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.858e+02 2.868e+02 3.639e+02 4.960e+02 1.282e+03, threshold=7.278e+02, percent-clipped=8.0 +2023-02-06 07:15:22,920 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68449.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 07:15:23,637 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68450.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:15:29,205 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68458.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:15:33,817 INFO [train.py:901] (3/4) Epoch 9, batch 3800, loss[loss=0.2413, simple_loss=0.3212, pruned_loss=0.08067, over 8293.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3227, pruned_loss=0.08947, over 1616453.06 frames. ], batch size: 23, lr: 8.54e-03, grad_scale: 8.0 +2023-02-06 07:16:07,820 INFO [train.py:901] (3/4) Epoch 9, batch 3850, loss[loss=0.2104, simple_loss=0.2814, pruned_loss=0.06968, over 7723.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3204, pruned_loss=0.08801, over 1616368.57 frames. ], batch size: 18, lr: 8.54e-03, grad_scale: 8.0 +2023-02-06 07:16:24,991 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 07:16:25,653 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.907e+02 2.582e+02 3.048e+02 3.724e+02 6.674e+02, threshold=6.096e+02, percent-clipped=0.0 +2023-02-06 07:16:28,000 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.0395, 1.2437, 3.1577, 1.0242, 2.7839, 2.6764, 2.8670, 2.7774], + device='cuda:3'), covar=tensor([0.0655, 0.3442, 0.0758, 0.3139, 0.1374, 0.0930, 0.0625, 0.0764], + device='cuda:3'), in_proj_covar=tensor([0.0443, 0.0546, 0.0542, 0.0509, 0.0574, 0.0479, 0.0476, 0.0539], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 07:16:42,274 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68564.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 07:16:42,685 INFO [train.py:901] (3/4) Epoch 9, batch 3900, loss[loss=0.3139, simple_loss=0.3921, pruned_loss=0.1178, over 8676.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3221, pruned_loss=0.08885, over 1618879.97 frames. ], batch size: 34, lr: 8.54e-03, grad_scale: 8.0 +2023-02-06 07:17:02,959 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6099, 1.4115, 2.7938, 1.2664, 2.0307, 3.0304, 3.0924, 2.5404], + device='cuda:3'), covar=tensor([0.1052, 0.1470, 0.0409, 0.2057, 0.0873, 0.0305, 0.0475, 0.0718], + device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0285, 0.0246, 0.0276, 0.0262, 0.0228, 0.0304, 0.0290], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 07:17:04,307 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68596.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:17:17,701 INFO [train.py:901] (3/4) Epoch 9, batch 3950, loss[loss=0.256, simple_loss=0.3252, pruned_loss=0.0934, over 8252.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3217, pruned_loss=0.08862, over 1614562.16 frames. ], batch size: 22, lr: 8.53e-03, grad_scale: 8.0 +2023-02-06 07:17:35,328 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.544e+02 2.589e+02 3.045e+02 4.133e+02 1.084e+03, threshold=6.090e+02, percent-clipped=3.0 +2023-02-06 07:17:51,768 INFO [train.py:901] (3/4) Epoch 9, batch 4000, loss[loss=0.229, simple_loss=0.3126, pruned_loss=0.07273, over 8250.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.322, pruned_loss=0.08877, over 1613702.44 frames. ], batch size: 22, lr: 8.53e-03, grad_scale: 8.0 +2023-02-06 07:18:00,060 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.64 vs. limit=5.0 +2023-02-06 07:18:13,674 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.90 vs. limit=5.0 +2023-02-06 07:18:16,919 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-06 07:18:17,951 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68703.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:18:25,239 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68714.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:18:25,677 INFO [train.py:901] (3/4) Epoch 9, batch 4050, loss[loss=0.2609, simple_loss=0.326, pruned_loss=0.09789, over 8488.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3207, pruned_loss=0.08771, over 1615946.31 frames. ], batch size: 29, lr: 8.53e-03, grad_scale: 8.0 +2023-02-06 07:18:28,565 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3813, 1.9440, 2.9470, 2.2683, 2.5609, 2.1481, 1.7112, 1.2824], + device='cuda:3'), covar=tensor([0.3032, 0.3295, 0.0827, 0.2109, 0.1633, 0.2023, 0.1546, 0.3377], + device='cuda:3'), in_proj_covar=tensor([0.0853, 0.0814, 0.0697, 0.0808, 0.0904, 0.0758, 0.0687, 0.0731], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 07:18:42,613 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68739.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:18:43,698 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.898e+02 2.478e+02 3.133e+02 3.692e+02 8.585e+02, threshold=6.266e+02, percent-clipped=3.0 +2023-02-06 07:18:56,669 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5872, 1.9391, 2.0200, 1.3991, 2.0948, 1.4174, 0.5833, 1.9148], + device='cuda:3'), covar=tensor([0.0318, 0.0178, 0.0123, 0.0259, 0.0191, 0.0521, 0.0503, 0.0124], + device='cuda:3'), in_proj_covar=tensor([0.0371, 0.0297, 0.0248, 0.0357, 0.0288, 0.0448, 0.0343, 0.0322], + device='cuda:3'), out_proj_covar=tensor([1.1048e-04, 8.5746e-05, 7.2775e-05, 1.0398e-04, 8.5422e-05, 1.4315e-04, + 1.0210e-04, 9.5259e-05], device='cuda:3') +2023-02-06 07:18:57,743 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68761.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:19:00,372 INFO [train.py:901] (3/4) Epoch 9, batch 4100, loss[loss=0.2386, simple_loss=0.3172, pruned_loss=0.07998, over 8517.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3205, pruned_loss=0.08722, over 1618875.52 frames. ], batch size: 28, lr: 8.52e-03, grad_scale: 8.0 +2023-02-06 07:19:08,015 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 +2023-02-06 07:19:34,811 INFO [train.py:901] (3/4) Epoch 9, batch 4150, loss[loss=0.2289, simple_loss=0.3077, pruned_loss=0.07505, over 8135.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3204, pruned_loss=0.08743, over 1614539.68 frames. ], batch size: 22, lr: 8.52e-03, grad_scale: 8.0 +2023-02-06 07:19:38,350 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68820.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 07:19:52,100 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.821e+02 2.558e+02 3.576e+02 4.352e+02 8.740e+02, threshold=7.151e+02, percent-clipped=5.0 +2023-02-06 07:19:55,817 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68845.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 07:20:02,871 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 07:20:09,486 INFO [train.py:901] (3/4) Epoch 9, batch 4200, loss[loss=0.2337, simple_loss=0.2968, pruned_loss=0.08523, over 7809.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3204, pruned_loss=0.08736, over 1614435.60 frames. ], batch size: 20, lr: 8.52e-03, grad_scale: 8.0 +2023-02-06 07:20:12,588 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-06 07:20:17,183 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68876.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:20:23,072 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 07:20:44,847 INFO [train.py:901] (3/4) Epoch 9, batch 4250, loss[loss=0.2296, simple_loss=0.3132, pruned_loss=0.07301, over 8243.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3208, pruned_loss=0.08782, over 1616667.98 frames. ], batch size: 24, lr: 8.52e-03, grad_scale: 8.0 +2023-02-06 07:20:45,558 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 07:20:51,272 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9564, 1.6481, 2.3261, 1.9142, 2.0394, 1.8272, 1.5503, 0.6426], + device='cuda:3'), covar=tensor([0.3180, 0.2948, 0.0939, 0.1858, 0.1394, 0.1852, 0.1443, 0.3105], + device='cuda:3'), in_proj_covar=tensor([0.0854, 0.0816, 0.0698, 0.0808, 0.0902, 0.0761, 0.0686, 0.0733], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 07:21:02,845 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68940.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:21:03,447 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.856e+02 3.701e+02 4.402e+02 9.379e+02, threshold=7.403e+02, percent-clipped=2.0 +2023-02-06 07:21:20,741 INFO [train.py:901] (3/4) Epoch 9, batch 4300, loss[loss=0.223, simple_loss=0.2931, pruned_loss=0.07648, over 7652.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3217, pruned_loss=0.08879, over 1617456.26 frames. ], batch size: 19, lr: 8.51e-03, grad_scale: 8.0 +2023-02-06 07:21:55,038 INFO [train.py:901] (3/4) Epoch 9, batch 4350, loss[loss=0.2348, simple_loss=0.3126, pruned_loss=0.0785, over 8463.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3199, pruned_loss=0.08795, over 1615237.90 frames. ], batch size: 27, lr: 8.51e-03, grad_scale: 8.0 +2023-02-06 07:22:12,979 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.677e+02 2.761e+02 3.203e+02 3.985e+02 6.558e+02, threshold=6.405e+02, percent-clipped=0.0 +2023-02-06 07:22:16,276 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 07:22:17,724 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69047.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:22:23,149 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69055.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:22:27,151 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69061.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:22:29,684 INFO [train.py:901] (3/4) Epoch 9, batch 4400, loss[loss=0.2296, simple_loss=0.3083, pruned_loss=0.07545, over 8333.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3198, pruned_loss=0.08761, over 1613019.98 frames. ], batch size: 26, lr: 8.51e-03, grad_scale: 8.0 +2023-02-06 07:22:30,556 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9719, 2.0811, 1.6417, 2.8876, 1.3085, 1.3879, 1.8626, 2.4392], + device='cuda:3'), covar=tensor([0.0898, 0.0999, 0.1213, 0.0388, 0.1329, 0.1627, 0.1147, 0.0682], + device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0226, 0.0267, 0.0222, 0.0228, 0.0263, 0.0268, 0.0228], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 07:22:40,710 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8280, 1.4771, 3.3440, 1.4259, 2.1632, 3.5705, 3.6545, 2.9080], + device='cuda:3'), covar=tensor([0.1170, 0.1660, 0.0417, 0.2120, 0.1072, 0.0347, 0.0449, 0.0761], + device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0285, 0.0246, 0.0275, 0.0260, 0.0229, 0.0301, 0.0284], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 07:22:55,804 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 07:23:04,523 INFO [train.py:901] (3/4) Epoch 9, batch 4450, loss[loss=0.1926, simple_loss=0.268, pruned_loss=0.05857, over 7519.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3207, pruned_loss=0.08858, over 1613070.98 frames. ], batch size: 18, lr: 8.50e-03, grad_scale: 8.0 +2023-02-06 07:23:16,659 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69132.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:23:22,378 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.755e+02 2.746e+02 3.298e+02 3.852e+02 8.052e+02, threshold=6.596e+02, percent-clipped=4.0 +2023-02-06 07:23:33,874 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69157.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:23:37,253 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69162.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:23:39,008 INFO [train.py:901] (3/4) Epoch 9, batch 4500, loss[loss=0.2332, simple_loss=0.3157, pruned_loss=0.07528, over 8579.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3212, pruned_loss=0.08921, over 1611474.47 frames. ], batch size: 31, lr: 8.50e-03, grad_scale: 8.0 +2023-02-06 07:23:49,144 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 07:23:49,293 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69180.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:23:53,180 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69186.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:24:12,700 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4163, 1.7943, 1.3486, 2.3100, 1.0270, 1.1621, 1.6248, 1.8177], + device='cuda:3'), covar=tensor([0.1295, 0.0994, 0.1484, 0.0587, 0.1347, 0.1795, 0.1061, 0.0904], + device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0223, 0.0264, 0.0222, 0.0226, 0.0263, 0.0268, 0.0230], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 07:24:13,186 INFO [train.py:901] (3/4) Epoch 9, batch 4550, loss[loss=0.2817, simple_loss=0.3499, pruned_loss=0.1068, over 8553.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3205, pruned_loss=0.08862, over 1606787.16 frames. ], batch size: 31, lr: 8.50e-03, grad_scale: 8.0 +2023-02-06 07:24:16,801 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2028, 1.6542, 4.2333, 1.7822, 2.3132, 4.7423, 4.7466, 4.0793], + device='cuda:3'), covar=tensor([0.1146, 0.1570, 0.0283, 0.2019, 0.1024, 0.0253, 0.0381, 0.0620], + device='cuda:3'), in_proj_covar=tensor([0.0258, 0.0290, 0.0251, 0.0280, 0.0266, 0.0235, 0.0309, 0.0291], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 07:24:31,944 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.836e+02 2.522e+02 2.943e+02 3.743e+02 5.945e+02, threshold=5.886e+02, percent-clipped=0.0 +2023-02-06 07:24:33,333 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69243.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:24:47,702 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69263.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 07:24:48,921 INFO [train.py:901] (3/4) Epoch 9, batch 4600, loss[loss=0.2329, simple_loss=0.3141, pruned_loss=0.07588, over 8512.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3209, pruned_loss=0.08864, over 1610091.85 frames. ], batch size: 26, lr: 8.49e-03, grad_scale: 8.0 +2023-02-06 07:25:07,533 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69291.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:25:22,107 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69311.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:25:24,600 INFO [train.py:901] (3/4) Epoch 9, batch 4650, loss[loss=0.2242, simple_loss=0.3134, pruned_loss=0.06755, over 8105.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3199, pruned_loss=0.08798, over 1609548.49 frames. ], batch size: 23, lr: 8.49e-03, grad_scale: 8.0 +2023-02-06 07:25:38,791 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69336.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:25:40,113 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69338.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:25:42,601 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.490e+02 2.666e+02 3.298e+02 3.900e+02 8.712e+02, threshold=6.595e+02, percent-clipped=8.0 +2023-02-06 07:25:58,527 INFO [train.py:901] (3/4) Epoch 9, batch 4700, loss[loss=0.2349, simple_loss=0.3121, pruned_loss=0.07887, over 8106.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3194, pruned_loss=0.0881, over 1604956.71 frames. ], batch size: 23, lr: 8.49e-03, grad_scale: 8.0 +2023-02-06 07:26:14,068 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69386.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:26:26,910 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69405.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:26:31,850 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69412.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:26:33,682 INFO [train.py:901] (3/4) Epoch 9, batch 4750, loss[loss=0.2788, simple_loss=0.3414, pruned_loss=0.1082, over 7125.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3177, pruned_loss=0.08717, over 1602535.80 frames. ], batch size: 71, lr: 8.48e-03, grad_scale: 8.0 +2023-02-06 07:26:35,887 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69418.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:26:43,653 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-06 07:26:49,111 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 07:26:51,004 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 07:26:51,650 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.677e+02 2.564e+02 3.173e+02 4.227e+02 9.736e+02, threshold=6.346e+02, percent-clipped=4.0 +2023-02-06 07:26:53,207 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69443.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:27:08,529 INFO [train.py:901] (3/4) Epoch 9, batch 4800, loss[loss=0.2073, simple_loss=0.2849, pruned_loss=0.06484, over 7528.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3174, pruned_loss=0.08686, over 1597114.82 frames. ], batch size: 18, lr: 8.48e-03, grad_scale: 8.0 +2023-02-06 07:27:41,300 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 07:27:43,201 INFO [train.py:901] (3/4) Epoch 9, batch 4850, loss[loss=0.225, simple_loss=0.2998, pruned_loss=0.07513, over 8238.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3171, pruned_loss=0.08684, over 1594354.00 frames. ], batch size: 22, lr: 8.48e-03, grad_scale: 8.0 +2023-02-06 07:27:46,760 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69520.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:27:49,417 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69524.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:27:53,339 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69530.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:28:00,637 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.666e+02 2.717e+02 3.193e+02 3.973e+02 8.915e+02, threshold=6.387e+02, percent-clipped=1.0 +2023-02-06 07:28:17,599 INFO [train.py:901] (3/4) Epoch 9, batch 4900, loss[loss=0.1989, simple_loss=0.2625, pruned_loss=0.06762, over 7934.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3163, pruned_loss=0.08655, over 1595566.08 frames. ], batch size: 20, lr: 8.48e-03, grad_scale: 8.0 +2023-02-06 07:28:33,123 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69587.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:28:47,483 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69607.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 07:28:53,430 INFO [train.py:901] (3/4) Epoch 9, batch 4950, loss[loss=0.2334, simple_loss=0.3076, pruned_loss=0.07958, over 7817.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3165, pruned_loss=0.08637, over 1604732.48 frames. ], batch size: 20, lr: 8.47e-03, grad_scale: 8.0 +2023-02-06 07:29:06,610 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69635.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:29:09,344 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69639.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:29:10,455 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.516e+02 2.734e+02 3.225e+02 4.131e+02 8.295e+02, threshold=6.450e+02, percent-clipped=5.0 +2023-02-06 07:29:10,647 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.0232, 1.6411, 1.7813, 1.6693, 1.1944, 1.7765, 2.1638, 1.8156], + device='cuda:3'), covar=tensor([0.0436, 0.1231, 0.1703, 0.1314, 0.0619, 0.1433, 0.0632, 0.0617], + device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0160, 0.0198, 0.0163, 0.0110, 0.0168, 0.0122, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 07:29:13,215 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69645.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:29:27,250 INFO [train.py:901] (3/4) Epoch 9, batch 5000, loss[loss=0.2225, simple_loss=0.3087, pruned_loss=0.06816, over 8458.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3177, pruned_loss=0.08692, over 1609459.63 frames. ], batch size: 25, lr: 8.47e-03, grad_scale: 8.0 +2023-02-06 07:29:39,131 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69682.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:29:46,879 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69692.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 07:29:53,593 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69702.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:30:02,817 INFO [train.py:901] (3/4) Epoch 9, batch 5050, loss[loss=0.2817, simple_loss=0.3505, pruned_loss=0.1064, over 8506.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3181, pruned_loss=0.08741, over 1609417.70 frames. ], batch size: 26, lr: 8.47e-03, grad_scale: 8.0 +2023-02-06 07:30:07,764 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69722.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 07:30:12,770 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69730.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:30:17,055 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3056, 1.9154, 2.9685, 2.3307, 2.5736, 2.0375, 1.6024, 1.3260], + device='cuda:3'), covar=tensor([0.3174, 0.3309, 0.0872, 0.2030, 0.1645, 0.1799, 0.1549, 0.3449], + device='cuda:3'), in_proj_covar=tensor([0.0858, 0.0825, 0.0702, 0.0808, 0.0911, 0.0762, 0.0689, 0.0738], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 07:30:18,113 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 07:30:20,813 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.871e+02 2.703e+02 3.249e+02 3.895e+02 8.845e+02, threshold=6.498e+02, percent-clipped=2.0 +2023-02-06 07:30:26,987 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69750.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:30:30,984 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69756.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:30:36,718 INFO [train.py:901] (3/4) Epoch 9, batch 5100, loss[loss=0.2851, simple_loss=0.3453, pruned_loss=0.1125, over 7973.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3184, pruned_loss=0.08737, over 1608284.54 frames. ], batch size: 21, lr: 8.46e-03, grad_scale: 8.0 +2023-02-06 07:30:44,228 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69776.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:30:59,043 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69797.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:31:01,166 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6104, 1.6385, 4.3853, 1.9857, 2.4877, 4.9894, 4.9586, 4.3099], + device='cuda:3'), covar=tensor([0.1017, 0.1745, 0.0289, 0.1916, 0.1013, 0.0201, 0.0263, 0.0599], + device='cuda:3'), in_proj_covar=tensor([0.0254, 0.0288, 0.0249, 0.0277, 0.0265, 0.0229, 0.0305, 0.0288], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 07:31:01,898 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69801.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:31:11,530 INFO [train.py:901] (3/4) Epoch 9, batch 5150, loss[loss=0.2783, simple_loss=0.3419, pruned_loss=0.1074, over 8321.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3177, pruned_loss=0.08676, over 1607938.79 frames. ], batch size: 25, lr: 8.46e-03, grad_scale: 8.0 +2023-02-06 07:31:29,714 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.853e+02 2.410e+02 3.240e+02 3.896e+02 9.119e+02, threshold=6.481e+02, percent-clipped=3.0 +2023-02-06 07:31:32,782 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69845.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:31:43,080 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 +2023-02-06 07:31:46,671 INFO [train.py:901] (3/4) Epoch 9, batch 5200, loss[loss=0.2155, simple_loss=0.3055, pruned_loss=0.06274, over 8358.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3181, pruned_loss=0.08708, over 1608778.26 frames. ], batch size: 24, lr: 8.46e-03, grad_scale: 8.0 +2023-02-06 07:31:50,876 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69871.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:32:02,093 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3839, 2.0424, 3.0951, 2.4626, 2.8018, 2.1868, 1.7577, 1.5370], + device='cuda:3'), covar=tensor([0.3159, 0.3544, 0.0933, 0.2427, 0.1749, 0.1934, 0.1478, 0.3500], + device='cuda:3'), in_proj_covar=tensor([0.0853, 0.0820, 0.0700, 0.0807, 0.0907, 0.0758, 0.0686, 0.0735], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 07:32:06,754 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69895.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:32:11,374 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69901.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:32:17,761 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 07:32:20,221 INFO [train.py:901] (3/4) Epoch 9, batch 5250, loss[loss=0.2558, simple_loss=0.3176, pruned_loss=0.09696, over 8127.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3173, pruned_loss=0.08648, over 1607742.88 frames. ], batch size: 22, lr: 8.45e-03, grad_scale: 8.0 +2023-02-06 07:32:23,773 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69920.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:32:27,865 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69926.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:32:38,251 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.964e+02 2.909e+02 3.504e+02 4.160e+02 7.603e+02, threshold=7.007e+02, percent-clipped=5.0 +2023-02-06 07:32:50,714 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69958.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:32:55,297 INFO [train.py:901] (3/4) Epoch 9, batch 5300, loss[loss=0.2205, simple_loss=0.3122, pruned_loss=0.06438, over 8201.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.316, pruned_loss=0.08525, over 1609653.45 frames. ], batch size: 23, lr: 8.45e-03, grad_scale: 8.0 +2023-02-06 07:33:05,000 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69978.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 07:33:08,278 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69983.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:33:22,934 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70003.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 07:33:24,857 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70006.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:33:31,223 INFO [train.py:901] (3/4) Epoch 9, batch 5350, loss[loss=0.231, simple_loss=0.3187, pruned_loss=0.07164, over 8302.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3179, pruned_loss=0.0867, over 1610318.94 frames. ], batch size: 23, lr: 8.45e-03, grad_scale: 8.0 +2023-02-06 07:33:34,730 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7474, 1.3408, 1.5657, 1.2517, 0.9410, 1.3696, 1.4245, 1.4334], + device='cuda:3'), covar=tensor([0.0502, 0.1186, 0.1687, 0.1381, 0.0574, 0.1413, 0.0681, 0.0599], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0159, 0.0198, 0.0162, 0.0109, 0.0167, 0.0121, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 07:33:42,009 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70031.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:33:45,285 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70036.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 07:33:48,396 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.807e+02 2.965e+02 3.484e+02 4.155e+02 9.515e+02, threshold=6.968e+02, percent-clipped=2.0 +2023-02-06 07:33:57,317 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70053.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:34:05,195 INFO [train.py:901] (3/4) Epoch 9, batch 5400, loss[loss=0.2297, simple_loss=0.3087, pruned_loss=0.07535, over 8134.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3191, pruned_loss=0.08716, over 1613347.64 frames. ], batch size: 22, lr: 8.45e-03, grad_scale: 8.0 +2023-02-06 07:34:14,822 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70078.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:34:31,560 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70101.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:34:40,952 INFO [train.py:901] (3/4) Epoch 9, batch 5450, loss[loss=0.2312, simple_loss=0.3214, pruned_loss=0.07057, over 8526.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3193, pruned_loss=0.08732, over 1612200.57 frames. ], batch size: 28, lr: 8.44e-03, grad_scale: 8.0 +2023-02-06 07:34:48,368 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70126.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:34:49,037 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70127.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:34:49,627 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70128.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:34:58,814 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.721e+02 2.682e+02 3.191e+02 4.046e+02 1.028e+03, threshold=6.382e+02, percent-clipped=4.0 +2023-02-06 07:35:04,325 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 07:35:05,834 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70151.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 07:35:06,536 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70152.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:35:15,856 INFO [train.py:901] (3/4) Epoch 9, batch 5500, loss[loss=0.2289, simple_loss=0.3064, pruned_loss=0.07565, over 8032.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.319, pruned_loss=0.08692, over 1612632.14 frames. ], batch size: 22, lr: 8.44e-03, grad_scale: 16.0 +2023-02-06 07:35:50,278 INFO [train.py:901] (3/4) Epoch 9, batch 5550, loss[loss=0.2046, simple_loss=0.2654, pruned_loss=0.07187, over 7825.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3191, pruned_loss=0.08741, over 1604741.98 frames. ], batch size: 19, lr: 8.44e-03, grad_scale: 16.0 +2023-02-06 07:36:07,873 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.691e+02 2.484e+02 3.031e+02 3.937e+02 9.276e+02, threshold=6.062e+02, percent-clipped=2.0 +2023-02-06 07:36:14,989 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-06 07:36:24,759 INFO [train.py:901] (3/4) Epoch 9, batch 5600, loss[loss=0.2215, simple_loss=0.3112, pruned_loss=0.06586, over 8772.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3178, pruned_loss=0.08624, over 1606427.36 frames. ], batch size: 30, lr: 8.43e-03, grad_scale: 16.0 +2023-02-06 07:36:34,074 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70278.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:36:35,515 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4376, 1.9076, 3.1307, 1.2357, 2.4180, 1.9012, 1.7539, 2.1065], + device='cuda:3'), covar=tensor([0.1830, 0.2046, 0.0700, 0.3846, 0.1398, 0.2768, 0.1635, 0.1960], + device='cuda:3'), in_proj_covar=tensor([0.0479, 0.0496, 0.0524, 0.0564, 0.0600, 0.0537, 0.0463, 0.0603], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 07:36:59,382 INFO [train.py:901] (3/4) Epoch 9, batch 5650, loss[loss=0.2091, simple_loss=0.2967, pruned_loss=0.06074, over 8336.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.3168, pruned_loss=0.08531, over 1608201.37 frames. ], batch size: 25, lr: 8.43e-03, grad_scale: 8.0 +2023-02-06 07:37:08,098 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 07:37:18,053 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.755e+02 2.609e+02 3.248e+02 4.005e+02 8.106e+02, threshold=6.497e+02, percent-clipped=5.0 +2023-02-06 07:37:32,963 INFO [train.py:901] (3/4) Epoch 9, batch 5700, loss[loss=0.2348, simple_loss=0.3204, pruned_loss=0.07459, over 8354.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3172, pruned_loss=0.08583, over 1608505.09 frames. ], batch size: 24, lr: 8.43e-03, grad_scale: 8.0 +2023-02-06 07:37:56,393 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4986, 1.9737, 3.0514, 2.5826, 2.8320, 2.2576, 1.7896, 1.4267], + device='cuda:3'), covar=tensor([0.2987, 0.3699, 0.0900, 0.1860, 0.1414, 0.1699, 0.1336, 0.3393], + device='cuda:3'), in_proj_covar=tensor([0.0861, 0.0829, 0.0707, 0.0806, 0.0905, 0.0765, 0.0686, 0.0737], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 07:38:02,546 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3907, 4.3666, 3.9422, 1.7798, 3.9078, 3.9656, 3.9607, 3.6345], + device='cuda:3'), covar=tensor([0.0918, 0.0692, 0.1192, 0.5735, 0.0900, 0.0998, 0.1467, 0.0863], + device='cuda:3'), in_proj_covar=tensor([0.0449, 0.0353, 0.0367, 0.0468, 0.0365, 0.0350, 0.0362, 0.0309], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 07:38:02,667 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70407.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 07:38:08,013 INFO [train.py:901] (3/4) Epoch 9, batch 5750, loss[loss=0.2474, simple_loss=0.3282, pruned_loss=0.08335, over 7967.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.3158, pruned_loss=0.08528, over 1608197.02 frames. ], batch size: 21, lr: 8.42e-03, grad_scale: 8.0 +2023-02-06 07:38:12,143 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 07:38:20,280 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70432.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 07:38:27,529 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.028e+02 2.898e+02 3.376e+02 4.229e+02 8.555e+02, threshold=6.753e+02, percent-clipped=3.0 +2023-02-06 07:38:43,381 INFO [train.py:901] (3/4) Epoch 9, batch 5800, loss[loss=0.2056, simple_loss=0.2807, pruned_loss=0.06525, over 7230.00 frames. ], tot_loss[loss=0.2434, simple_loss=0.3158, pruned_loss=0.08548, over 1608727.32 frames. ], batch size: 16, lr: 8.42e-03, grad_scale: 8.0 +2023-02-06 07:38:48,296 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70472.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:38:48,579 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.60 vs. limit=5.0 +2023-02-06 07:39:18,643 INFO [train.py:901] (3/4) Epoch 9, batch 5850, loss[loss=0.2384, simple_loss=0.3154, pruned_loss=0.08076, over 8561.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.3156, pruned_loss=0.08556, over 1611300.28 frames. ], batch size: 49, lr: 8.42e-03, grad_scale: 8.0 +2023-02-06 07:39:28,974 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4293, 1.8240, 1.8591, 1.0135, 2.0406, 1.4563, 0.4423, 1.6526], + device='cuda:3'), covar=tensor([0.0306, 0.0174, 0.0131, 0.0281, 0.0196, 0.0501, 0.0447, 0.0132], + device='cuda:3'), in_proj_covar=tensor([0.0372, 0.0297, 0.0246, 0.0357, 0.0285, 0.0446, 0.0340, 0.0326], + device='cuda:3'), out_proj_covar=tensor([1.1024e-04, 8.5580e-05, 7.1640e-05, 1.0383e-04, 8.3951e-05, 1.4133e-04, + 1.0086e-04, 9.5879e-05], device='cuda:3') +2023-02-06 07:39:37,365 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.629e+02 2.477e+02 3.501e+02 4.376e+02 8.995e+02, threshold=7.001e+02, percent-clipped=4.0 +2023-02-06 07:39:48,209 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6663, 2.2284, 3.7538, 2.8076, 3.0813, 2.5042, 1.9785, 1.7846], + device='cuda:3'), covar=tensor([0.3150, 0.3703, 0.0964, 0.2366, 0.1780, 0.1846, 0.1514, 0.3922], + device='cuda:3'), in_proj_covar=tensor([0.0853, 0.0819, 0.0703, 0.0805, 0.0900, 0.0761, 0.0683, 0.0733], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 07:39:53,322 INFO [train.py:901] (3/4) Epoch 9, batch 5900, loss[loss=0.232, simple_loss=0.3133, pruned_loss=0.07531, over 8244.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3163, pruned_loss=0.0861, over 1611842.07 frames. ], batch size: 24, lr: 8.42e-03, grad_scale: 8.0 +2023-02-06 07:40:08,010 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70587.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:40:27,231 INFO [train.py:901] (3/4) Epoch 9, batch 5950, loss[loss=0.226, simple_loss=0.2933, pruned_loss=0.07936, over 7427.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3169, pruned_loss=0.08652, over 1613804.52 frames. ], batch size: 17, lr: 8.41e-03, grad_scale: 8.0 +2023-02-06 07:40:32,706 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70622.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:40:45,946 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.771e+02 2.730e+02 3.193e+02 3.849e+02 7.953e+02, threshold=6.387e+02, percent-clipped=3.0 +2023-02-06 07:41:02,078 INFO [train.py:901] (3/4) Epoch 9, batch 6000, loss[loss=0.2804, simple_loss=0.327, pruned_loss=0.1168, over 7653.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3174, pruned_loss=0.08712, over 1613437.90 frames. ], batch size: 19, lr: 8.41e-03, grad_scale: 8.0 +2023-02-06 07:41:02,079 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 07:41:14,589 INFO [train.py:935] (3/4) Epoch 9, validation: loss=0.1952, simple_loss=0.2947, pruned_loss=0.0479, over 944034.00 frames. +2023-02-06 07:41:14,590 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 07:41:49,711 INFO [train.py:901] (3/4) Epoch 9, batch 6050, loss[loss=0.1739, simple_loss=0.2633, pruned_loss=0.04229, over 7716.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3182, pruned_loss=0.08779, over 1615986.65 frames. ], batch size: 18, lr: 8.41e-03, grad_scale: 8.0 +2023-02-06 07:42:04,740 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70737.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:42:07,993 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.683e+02 2.822e+02 3.602e+02 4.348e+02 1.269e+03, threshold=7.203e+02, percent-clipped=6.0 +2023-02-06 07:42:10,338 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6266, 1.9746, 2.0503, 1.0682, 2.2595, 1.4308, 0.7064, 1.7749], + device='cuda:3'), covar=tensor([0.0403, 0.0187, 0.0173, 0.0368, 0.0178, 0.0537, 0.0453, 0.0180], + device='cuda:3'), in_proj_covar=tensor([0.0378, 0.0301, 0.0249, 0.0360, 0.0287, 0.0450, 0.0344, 0.0329], + device='cuda:3'), out_proj_covar=tensor([1.1221e-04, 8.6868e-05, 7.2050e-05, 1.0463e-04, 8.4396e-05, 1.4254e-04, + 1.0218e-04, 9.6955e-05], device='cuda:3') +2023-02-06 07:42:19,440 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-06 07:42:20,525 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6407, 1.9292, 3.0358, 1.4075, 2.4197, 2.0482, 1.7563, 2.1208], + device='cuda:3'), covar=tensor([0.1424, 0.1870, 0.0628, 0.3303, 0.1222, 0.2390, 0.1441, 0.1861], + device='cuda:3'), in_proj_covar=tensor([0.0478, 0.0495, 0.0525, 0.0567, 0.0603, 0.0541, 0.0464, 0.0604], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 07:42:24,368 INFO [train.py:901] (3/4) Epoch 9, batch 6100, loss[loss=0.1823, simple_loss=0.2558, pruned_loss=0.05438, over 7432.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3168, pruned_loss=0.08646, over 1610369.73 frames. ], batch size: 17, lr: 8.40e-03, grad_scale: 8.0 +2023-02-06 07:42:42,096 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 07:43:00,374 INFO [train.py:901] (3/4) Epoch 9, batch 6150, loss[loss=0.2216, simple_loss=0.2929, pruned_loss=0.07512, over 7442.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.3157, pruned_loss=0.08574, over 1609391.53 frames. ], batch size: 17, lr: 8.40e-03, grad_scale: 8.0 +2023-02-06 07:43:01,905 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9412, 2.5413, 3.1930, 1.2257, 2.9630, 1.6375, 1.6670, 1.7611], + device='cuda:3'), covar=tensor([0.0473, 0.0241, 0.0146, 0.0475, 0.0368, 0.0607, 0.0514, 0.0335], + device='cuda:3'), in_proj_covar=tensor([0.0376, 0.0300, 0.0247, 0.0360, 0.0288, 0.0449, 0.0343, 0.0328], + device='cuda:3'), out_proj_covar=tensor([1.1154e-04, 8.6542e-05, 7.1460e-05, 1.0460e-04, 8.4599e-05, 1.4198e-04, + 1.0185e-04, 9.6633e-05], device='cuda:3') +2023-02-06 07:43:18,322 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.632e+02 2.692e+02 3.232e+02 3.879e+02 7.941e+02, threshold=6.463e+02, percent-clipped=1.0 +2023-02-06 07:43:19,254 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70843.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:43:33,936 INFO [train.py:901] (3/4) Epoch 9, batch 6200, loss[loss=0.199, simple_loss=0.2858, pruned_loss=0.05606, over 8231.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3158, pruned_loss=0.0856, over 1611370.33 frames. ], batch size: 22, lr: 8.40e-03, grad_scale: 8.0 +2023-02-06 07:43:36,229 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70868.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:44:09,711 INFO [train.py:901] (3/4) Epoch 9, batch 6250, loss[loss=0.2277, simple_loss=0.3124, pruned_loss=0.07153, over 8472.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3168, pruned_loss=0.08643, over 1613639.07 frames. ], batch size: 27, lr: 8.40e-03, grad_scale: 8.0 +2023-02-06 07:44:28,454 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.776e+02 2.792e+02 3.423e+02 4.432e+02 1.474e+03, threshold=6.847e+02, percent-clipped=7.0 +2023-02-06 07:44:43,930 INFO [train.py:901] (3/4) Epoch 9, batch 6300, loss[loss=0.2817, simple_loss=0.3482, pruned_loss=0.1077, over 8809.00 frames. ], tot_loss[loss=0.245, simple_loss=0.3168, pruned_loss=0.08656, over 1614634.69 frames. ], batch size: 40, lr: 8.39e-03, grad_scale: 8.0 +2023-02-06 07:45:03,882 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70993.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:45:16,145 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71010.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:45:19,449 INFO [train.py:901] (3/4) Epoch 9, batch 6350, loss[loss=0.2376, simple_loss=0.3139, pruned_loss=0.08066, over 8661.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3167, pruned_loss=0.08614, over 1615915.82 frames. ], batch size: 34, lr: 8.39e-03, grad_scale: 8.0 +2023-02-06 07:45:21,646 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71018.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:45:23,710 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4348, 1.6901, 2.7321, 1.2619, 1.9245, 1.7835, 1.4957, 1.8343], + device='cuda:3'), covar=tensor([0.1805, 0.2269, 0.0758, 0.3817, 0.1592, 0.2833, 0.1747, 0.2078], + device='cuda:3'), in_proj_covar=tensor([0.0486, 0.0501, 0.0528, 0.0573, 0.0612, 0.0547, 0.0465, 0.0609], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 07:45:38,849 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.775e+02 2.814e+02 3.293e+02 4.210e+02 8.338e+02, threshold=6.585e+02, percent-clipped=5.0 +2023-02-06 07:45:42,976 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71048.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:45:54,196 INFO [train.py:901] (3/4) Epoch 9, batch 6400, loss[loss=0.3193, simple_loss=0.3767, pruned_loss=0.1309, over 8317.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3186, pruned_loss=0.08695, over 1620198.08 frames. ], batch size: 25, lr: 8.39e-03, grad_scale: 8.0 +2023-02-06 07:46:28,849 INFO [train.py:901] (3/4) Epoch 9, batch 6450, loss[loss=0.2938, simple_loss=0.357, pruned_loss=0.1153, over 7166.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3195, pruned_loss=0.08727, over 1621893.98 frames. ], batch size: 72, lr: 8.38e-03, grad_scale: 4.0 +2023-02-06 07:46:48,400 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.820e+02 2.655e+02 3.350e+02 4.272e+02 1.011e+03, threshold=6.701e+02, percent-clipped=3.0 +2023-02-06 07:47:03,598 INFO [train.py:901] (3/4) Epoch 9, batch 6500, loss[loss=0.2776, simple_loss=0.3419, pruned_loss=0.1066, over 8547.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3189, pruned_loss=0.08718, over 1618441.42 frames. ], batch size: 49, lr: 8.38e-03, grad_scale: 4.0 +2023-02-06 07:47:03,834 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.0460, 3.1333, 3.5822, 2.2983, 1.9850, 3.6516, 0.6792, 2.2465], + device='cuda:3'), covar=tensor([0.1828, 0.1137, 0.0396, 0.2738, 0.4301, 0.0291, 0.4360, 0.2031], + device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0158, 0.0094, 0.0208, 0.0242, 0.0096, 0.0158, 0.0156], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 07:47:37,713 INFO [train.py:901] (3/4) Epoch 9, batch 6550, loss[loss=0.2033, simple_loss=0.2749, pruned_loss=0.06583, over 7805.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3176, pruned_loss=0.08627, over 1618187.15 frames. ], batch size: 19, lr: 8.38e-03, grad_scale: 4.0 +2023-02-06 07:47:40,261 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 07:47:50,023 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 07:47:58,021 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.639e+02 2.500e+02 3.444e+02 4.178e+02 7.414e+02, threshold=6.887e+02, percent-clipped=1.0 +2023-02-06 07:48:10,531 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 07:48:13,165 INFO [train.py:901] (3/4) Epoch 9, batch 6600, loss[loss=0.2182, simple_loss=0.302, pruned_loss=0.06726, over 7647.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3175, pruned_loss=0.08673, over 1612670.79 frames. ], batch size: 19, lr: 8.37e-03, grad_scale: 4.0 +2023-02-06 07:48:21,096 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-06 07:48:47,048 INFO [train.py:901] (3/4) Epoch 9, batch 6650, loss[loss=0.1975, simple_loss=0.2725, pruned_loss=0.06125, over 7638.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3163, pruned_loss=0.08599, over 1611421.97 frames. ], batch size: 19, lr: 8.37e-03, grad_scale: 4.0 +2023-02-06 07:49:05,722 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.569e+02 2.641e+02 3.214e+02 4.234e+02 1.005e+03, threshold=6.427e+02, percent-clipped=4.0 +2023-02-06 07:49:13,932 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=71354.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:49:21,688 INFO [train.py:901] (3/4) Epoch 9, batch 6700, loss[loss=0.2443, simple_loss=0.3055, pruned_loss=0.09151, over 7780.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3182, pruned_loss=0.08658, over 1615505.64 frames. ], batch size: 19, lr: 8.37e-03, grad_scale: 4.0 +2023-02-06 07:49:41,042 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=71392.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:49:56,538 INFO [train.py:901] (3/4) Epoch 9, batch 6750, loss[loss=0.2197, simple_loss=0.2859, pruned_loss=0.07676, over 7971.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3189, pruned_loss=0.08745, over 1612816.61 frames. ], batch size: 21, lr: 8.37e-03, grad_scale: 4.0 +2023-02-06 07:50:15,374 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.899e+02 3.032e+02 3.821e+02 4.704e+02 1.129e+03, threshold=7.641e+02, percent-clipped=7.0 +2023-02-06 07:50:23,380 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 07:50:30,842 INFO [train.py:901] (3/4) Epoch 9, batch 6800, loss[loss=0.263, simple_loss=0.3226, pruned_loss=0.1017, over 7932.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3186, pruned_loss=0.08744, over 1613673.97 frames. ], batch size: 20, lr: 8.36e-03, grad_scale: 8.0 +2023-02-06 07:50:32,161 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1969, 4.1062, 3.7817, 1.7854, 3.7366, 3.7593, 3.8164, 3.5428], + device='cuda:3'), covar=tensor([0.0817, 0.0582, 0.1177, 0.5123, 0.0840, 0.0859, 0.1282, 0.0770], + device='cuda:3'), in_proj_covar=tensor([0.0453, 0.0360, 0.0367, 0.0469, 0.0370, 0.0355, 0.0364, 0.0313], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 07:50:33,614 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71469.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:50:55,237 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2178, 1.9294, 2.9005, 2.3276, 2.6060, 2.1213, 1.6555, 1.2853], + device='cuda:3'), covar=tensor([0.3499, 0.3542, 0.0989, 0.2146, 0.1704, 0.1889, 0.1554, 0.3537], + device='cuda:3'), in_proj_covar=tensor([0.0856, 0.0819, 0.0702, 0.0806, 0.0899, 0.0763, 0.0685, 0.0739], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 07:51:01,265 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71507.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:51:06,468 INFO [train.py:901] (3/4) Epoch 9, batch 6850, loss[loss=0.2366, simple_loss=0.3185, pruned_loss=0.07737, over 8497.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3181, pruned_loss=0.08727, over 1610991.19 frames. ], batch size: 28, lr: 8.36e-03, grad_scale: 8.0 +2023-02-06 07:51:14,496 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 07:51:25,243 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.748e+02 2.628e+02 3.217e+02 4.054e+02 6.964e+02, threshold=6.433e+02, percent-clipped=0.0 +2023-02-06 07:51:40,042 INFO [train.py:901] (3/4) Epoch 9, batch 6900, loss[loss=0.3301, simple_loss=0.3946, pruned_loss=0.1328, over 8414.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3193, pruned_loss=0.08797, over 1610634.57 frames. ], batch size: 49, lr: 8.36e-03, grad_scale: 8.0 +2023-02-06 07:51:57,276 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.8290, 5.9095, 5.1423, 2.1481, 5.1838, 5.5818, 5.4829, 5.1943], + device='cuda:3'), covar=tensor([0.0598, 0.0433, 0.0864, 0.4899, 0.0656, 0.0704, 0.1132, 0.0589], + device='cuda:3'), in_proj_covar=tensor([0.0450, 0.0361, 0.0366, 0.0471, 0.0368, 0.0354, 0.0367, 0.0311], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 07:52:15,382 INFO [train.py:901] (3/4) Epoch 9, batch 6950, loss[loss=0.2541, simple_loss=0.3368, pruned_loss=0.08567, over 8333.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.319, pruned_loss=0.08759, over 1606884.54 frames. ], batch size: 25, lr: 8.35e-03, grad_scale: 8.0 +2023-02-06 07:52:22,399 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3384, 2.6344, 3.0086, 1.3990, 3.0962, 1.9526, 1.4954, 2.0246], + device='cuda:3'), covar=tensor([0.0526, 0.0218, 0.0231, 0.0516, 0.0294, 0.0542, 0.0654, 0.0294], + device='cuda:3'), in_proj_covar=tensor([0.0377, 0.0303, 0.0250, 0.0360, 0.0292, 0.0452, 0.0346, 0.0326], + device='cuda:3'), out_proj_covar=tensor([1.1122e-04, 8.6778e-05, 7.2309e-05, 1.0411e-04, 8.5738e-05, 1.4291e-04, + 1.0245e-04, 9.5758e-05], device='cuda:3') +2023-02-06 07:52:23,470 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 07:52:26,278 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.1861, 1.1958, 3.2740, 0.9687, 2.8262, 2.7308, 2.9801, 2.8682], + device='cuda:3'), covar=tensor([0.0687, 0.3959, 0.0856, 0.3610, 0.1512, 0.1069, 0.0735, 0.0880], + device='cuda:3'), in_proj_covar=tensor([0.0448, 0.0558, 0.0559, 0.0515, 0.0585, 0.0496, 0.0491, 0.0546], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 07:52:29,643 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71634.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:52:35,520 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.819e+02 2.622e+02 3.284e+02 3.978e+02 8.428e+02, threshold=6.567e+02, percent-clipped=2.0 +2023-02-06 07:52:50,437 INFO [train.py:901] (3/4) Epoch 9, batch 7000, loss[loss=0.3034, simple_loss=0.3711, pruned_loss=0.1178, over 8471.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3194, pruned_loss=0.08736, over 1608667.76 frames. ], batch size: 29, lr: 8.35e-03, grad_scale: 8.0 +2023-02-06 07:52:56,864 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-06 07:53:24,883 INFO [train.py:901] (3/4) Epoch 9, batch 7050, loss[loss=0.2351, simple_loss=0.3258, pruned_loss=0.0722, over 8463.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3184, pruned_loss=0.0864, over 1612295.51 frames. ], batch size: 25, lr: 8.35e-03, grad_scale: 8.0 +2023-02-06 07:53:32,430 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71725.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:53:45,153 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.687e+02 2.886e+02 3.338e+02 4.007e+02 6.250e+02, threshold=6.676e+02, percent-clipped=0.0 +2023-02-06 07:53:50,708 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71750.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:53:53,323 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6669, 2.0694, 2.1496, 1.1867, 2.3812, 1.5073, 0.6562, 1.8163], + device='cuda:3'), covar=tensor([0.0345, 0.0161, 0.0128, 0.0299, 0.0166, 0.0494, 0.0450, 0.0159], + device='cuda:3'), in_proj_covar=tensor([0.0371, 0.0297, 0.0247, 0.0356, 0.0286, 0.0445, 0.0340, 0.0324], + device='cuda:3'), out_proj_covar=tensor([1.0957e-04, 8.5037e-05, 7.1500e-05, 1.0269e-04, 8.3832e-05, 1.4036e-04, + 1.0047e-04, 9.5186e-05], device='cuda:3') +2023-02-06 07:53:59,291 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71763.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:54:00,448 INFO [train.py:901] (3/4) Epoch 9, batch 7100, loss[loss=0.2072, simple_loss=0.2902, pruned_loss=0.06211, over 7800.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.318, pruned_loss=0.0863, over 1611882.81 frames. ], batch size: 20, lr: 8.35e-03, grad_scale: 8.0 +2023-02-06 07:54:10,698 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3710, 1.5094, 2.1915, 1.1515, 1.5181, 1.5741, 1.3717, 1.2476], + device='cuda:3'), covar=tensor([0.1567, 0.1888, 0.0708, 0.3437, 0.1497, 0.2799, 0.1704, 0.1849], + device='cuda:3'), in_proj_covar=tensor([0.0480, 0.0496, 0.0529, 0.0572, 0.0604, 0.0546, 0.0463, 0.0608], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 07:54:16,145 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71788.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:54:28,670 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5821, 1.8430, 1.9516, 1.0013, 2.1296, 1.3538, 0.5156, 1.6704], + device='cuda:3'), covar=tensor([0.0353, 0.0175, 0.0168, 0.0359, 0.0214, 0.0558, 0.0464, 0.0176], + device='cuda:3'), in_proj_covar=tensor([0.0375, 0.0300, 0.0252, 0.0360, 0.0292, 0.0448, 0.0343, 0.0330], + device='cuda:3'), out_proj_covar=tensor([1.1062e-04, 8.5958e-05, 7.2734e-05, 1.0415e-04, 8.5610e-05, 1.4137e-04, + 1.0129e-04, 9.7004e-05], device='cuda:3') +2023-02-06 07:54:31,426 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1472, 1.0719, 1.2392, 1.1182, 0.8081, 1.2932, 0.0599, 0.9015], + device='cuda:3'), covar=tensor([0.2947, 0.1994, 0.0730, 0.1477, 0.5142, 0.0783, 0.3947, 0.1981], + device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0158, 0.0093, 0.0207, 0.0245, 0.0096, 0.0157, 0.0156], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 07:54:34,526 INFO [train.py:901] (3/4) Epoch 9, batch 7150, loss[loss=0.2449, simple_loss=0.323, pruned_loss=0.08346, over 8326.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3192, pruned_loss=0.08762, over 1609592.56 frames. ], batch size: 25, lr: 8.34e-03, grad_scale: 8.0 +2023-02-06 07:54:54,740 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.819e+02 2.560e+02 3.246e+02 4.043e+02 1.359e+03, threshold=6.493e+02, percent-clipped=7.0 +2023-02-06 07:55:10,759 INFO [train.py:901] (3/4) Epoch 9, batch 7200, loss[loss=0.2671, simple_loss=0.3368, pruned_loss=0.09866, over 6467.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3203, pruned_loss=0.08805, over 1611674.50 frames. ], batch size: 72, lr: 8.34e-03, grad_scale: 8.0 +2023-02-06 07:55:30,900 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 +2023-02-06 07:55:43,959 INFO [train.py:901] (3/4) Epoch 9, batch 7250, loss[loss=0.2332, simple_loss=0.3208, pruned_loss=0.0728, over 8462.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3183, pruned_loss=0.08653, over 1614008.94 frames. ], batch size: 27, lr: 8.34e-03, grad_scale: 8.0 +2023-02-06 07:55:58,954 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71937.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:56:02,878 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.612e+02 2.597e+02 3.277e+02 3.984e+02 9.565e+02, threshold=6.554e+02, percent-clipped=6.0 +2023-02-06 07:56:19,574 INFO [train.py:901] (3/4) Epoch 9, batch 7300, loss[loss=0.2791, simple_loss=0.3432, pruned_loss=0.1075, over 6927.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3172, pruned_loss=0.08624, over 1608311.24 frames. ], batch size: 71, lr: 8.33e-03, grad_scale: 8.0 +2023-02-06 07:56:28,842 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=71978.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:56:29,640 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4880, 2.9715, 1.7911, 2.2255, 2.2624, 1.6031, 2.1336, 2.2257], + device='cuda:3'), covar=tensor([0.1346, 0.0325, 0.1084, 0.0649, 0.0586, 0.1322, 0.0945, 0.0914], + device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0241, 0.0316, 0.0302, 0.0307, 0.0324, 0.0343, 0.0313], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 07:56:48,133 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3982, 2.7901, 1.6904, 2.0232, 2.1279, 1.5202, 1.8685, 1.9336], + device='cuda:3'), covar=tensor([0.1335, 0.0277, 0.0984, 0.0617, 0.0571, 0.1258, 0.0918, 0.0847], + device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0242, 0.0316, 0.0302, 0.0307, 0.0324, 0.0344, 0.0313], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 07:56:52,665 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.1401, 1.2659, 3.2954, 0.9703, 2.8616, 2.7562, 3.0137, 2.8944], + device='cuda:3'), covar=tensor([0.0755, 0.3923, 0.0734, 0.3492, 0.1526, 0.0988, 0.0726, 0.0929], + device='cuda:3'), in_proj_covar=tensor([0.0446, 0.0553, 0.0548, 0.0517, 0.0587, 0.0493, 0.0484, 0.0547], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 07:56:54,487 INFO [train.py:901] (3/4) Epoch 9, batch 7350, loss[loss=0.2613, simple_loss=0.3343, pruned_loss=0.09415, over 8352.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.318, pruned_loss=0.08674, over 1608178.62 frames. ], batch size: 24, lr: 8.33e-03, grad_scale: 8.0 +2023-02-06 07:57:02,505 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 07:57:13,528 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.007e+02 2.925e+02 3.749e+02 4.804e+02 1.068e+03, threshold=7.499e+02, percent-clipped=9.0 +2023-02-06 07:57:22,479 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 07:57:30,036 INFO [train.py:901] (3/4) Epoch 9, batch 7400, loss[loss=0.2491, simple_loss=0.3343, pruned_loss=0.08195, over 8369.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3176, pruned_loss=0.08612, over 1610489.94 frames. ], batch size: 24, lr: 8.33e-03, grad_scale: 8.0 +2023-02-06 07:57:42,992 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.6860, 5.6848, 4.9514, 2.2768, 5.0216, 5.3403, 5.2222, 4.9861], + device='cuda:3'), covar=tensor([0.0589, 0.0487, 0.0876, 0.4526, 0.0648, 0.0686, 0.1218, 0.0589], + device='cuda:3'), in_proj_covar=tensor([0.0450, 0.0354, 0.0368, 0.0468, 0.0364, 0.0352, 0.0363, 0.0308], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 07:57:50,402 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72093.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:57:56,613 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4988, 2.7395, 1.7046, 2.0277, 2.0544, 1.4376, 1.8745, 2.0744], + device='cuda:3'), covar=tensor([0.1116, 0.0258, 0.0962, 0.0590, 0.0613, 0.1270, 0.0909, 0.0813], + device='cuda:3'), in_proj_covar=tensor([0.0342, 0.0238, 0.0313, 0.0299, 0.0303, 0.0320, 0.0339, 0.0308], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 07:58:03,869 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 07:58:05,138 INFO [train.py:901] (3/4) Epoch 9, batch 7450, loss[loss=0.221, simple_loss=0.2979, pruned_loss=0.07205, over 7643.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3164, pruned_loss=0.08491, over 1610822.40 frames. ], batch size: 19, lr: 8.33e-03, grad_scale: 8.0 +2023-02-06 07:58:23,964 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.502e+02 2.410e+02 3.229e+02 3.860e+02 9.903e+02, threshold=6.459e+02, percent-clipped=1.0 +2023-02-06 07:58:26,830 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72147.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 07:58:38,698 INFO [train.py:901] (3/4) Epoch 9, batch 7500, loss[loss=0.2149, simple_loss=0.2893, pruned_loss=0.07028, over 7929.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.3147, pruned_loss=0.08421, over 1610157.63 frames. ], batch size: 20, lr: 8.32e-03, grad_scale: 8.0 +2023-02-06 07:59:09,506 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 +2023-02-06 07:59:15,028 INFO [train.py:901] (3/4) Epoch 9, batch 7550, loss[loss=0.3165, simple_loss=0.3759, pruned_loss=0.1286, over 8462.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.3144, pruned_loss=0.08443, over 1607708.12 frames. ], batch size: 29, lr: 8.32e-03, grad_scale: 8.0 +2023-02-06 07:59:33,688 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.914e+02 2.819e+02 3.433e+02 4.309e+02 8.597e+02, threshold=6.865e+02, percent-clipped=4.0 +2023-02-06 07:59:48,158 INFO [train.py:901] (3/4) Epoch 9, batch 7600, loss[loss=0.2289, simple_loss=0.3143, pruned_loss=0.07172, over 8110.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3166, pruned_loss=0.0858, over 1610736.41 frames. ], batch size: 23, lr: 8.32e-03, grad_scale: 8.0 +2023-02-06 07:59:58,719 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72281.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:00:22,660 INFO [train.py:901] (3/4) Epoch 9, batch 7650, loss[loss=0.2342, simple_loss=0.3015, pruned_loss=0.08351, over 8086.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3171, pruned_loss=0.0865, over 1611476.79 frames. ], batch size: 21, lr: 8.31e-03, grad_scale: 8.0 +2023-02-06 08:00:42,771 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.469e+02 2.706e+02 3.178e+02 3.983e+02 6.818e+02, threshold=6.357e+02, percent-clipped=0.0 +2023-02-06 08:00:43,589 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72344.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:00:46,999 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72349.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:00:57,579 INFO [train.py:901] (3/4) Epoch 9, batch 7700, loss[loss=0.2698, simple_loss=0.3435, pruned_loss=0.09803, over 8518.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3177, pruned_loss=0.08696, over 1612987.43 frames. ], batch size: 29, lr: 8.31e-03, grad_scale: 8.0 +2023-02-06 08:01:03,946 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72374.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:01:09,771 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 08:01:18,751 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72396.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:01:32,218 INFO [train.py:901] (3/4) Epoch 9, batch 7750, loss[loss=0.2535, simple_loss=0.3101, pruned_loss=0.09842, over 7813.00 frames. ], tot_loss[loss=0.2452, simple_loss=0.3174, pruned_loss=0.0865, over 1613389.76 frames. ], batch size: 20, lr: 8.31e-03, grad_scale: 8.0 +2023-02-06 08:01:53,040 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.870e+02 2.670e+02 3.267e+02 4.054e+02 1.108e+03, threshold=6.534e+02, percent-clipped=5.0 +2023-02-06 08:02:08,196 INFO [train.py:901] (3/4) Epoch 9, batch 7800, loss[loss=0.2701, simple_loss=0.3452, pruned_loss=0.09751, over 8472.00 frames. ], tot_loss[loss=0.244, simple_loss=0.3164, pruned_loss=0.0858, over 1613047.18 frames. ], batch size: 27, lr: 8.31e-03, grad_scale: 8.0 +2023-02-06 08:02:08,406 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2657, 1.3891, 1.1990, 1.8600, 0.7823, 1.1068, 1.1644, 1.4558], + device='cuda:3'), covar=tensor([0.1042, 0.0930, 0.1332, 0.0549, 0.1273, 0.1706, 0.0980, 0.0901], + device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0216, 0.0257, 0.0215, 0.0219, 0.0254, 0.0259, 0.0224], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 08:02:25,494 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72491.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:02:37,535 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2095, 1.4248, 1.2202, 1.9323, 0.6544, 1.0761, 1.2786, 1.5471], + device='cuda:3'), covar=tensor([0.1156, 0.0923, 0.1444, 0.0591, 0.1432, 0.1825, 0.0976, 0.0760], + device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0219, 0.0257, 0.0216, 0.0222, 0.0256, 0.0261, 0.0226], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 08:02:41,259 INFO [train.py:901] (3/4) Epoch 9, batch 7850, loss[loss=0.2479, simple_loss=0.329, pruned_loss=0.0834, over 8462.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.3157, pruned_loss=0.08506, over 1614616.11 frames. ], batch size: 25, lr: 8.30e-03, grad_scale: 8.0 +2023-02-06 08:02:59,530 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.780e+02 2.530e+02 3.201e+02 3.890e+02 8.475e+02, threshold=6.403e+02, percent-clipped=6.0 +2023-02-06 08:03:14,039 INFO [train.py:901] (3/4) Epoch 9, batch 7900, loss[loss=0.2572, simple_loss=0.3293, pruned_loss=0.09255, over 8454.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.3163, pruned_loss=0.08517, over 1613624.97 frames. ], batch size: 27, lr: 8.30e-03, grad_scale: 8.0 +2023-02-06 08:03:37,901 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 +2023-02-06 08:03:41,422 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72606.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:03:44,706 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4158, 1.7080, 2.8419, 1.2537, 1.9970, 1.7526, 1.4332, 1.8856], + device='cuda:3'), covar=tensor([0.1697, 0.2115, 0.0626, 0.3756, 0.1425, 0.2795, 0.1910, 0.1998], + device='cuda:3'), in_proj_covar=tensor([0.0480, 0.0496, 0.0531, 0.0572, 0.0609, 0.0540, 0.0469, 0.0611], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 08:03:47,073 INFO [train.py:901] (3/4) Epoch 9, batch 7950, loss[loss=0.2053, simple_loss=0.2902, pruned_loss=0.06025, over 8360.00 frames. ], tot_loss[loss=0.2434, simple_loss=0.3163, pruned_loss=0.08523, over 1611600.58 frames. ], batch size: 24, lr: 8.30e-03, grad_scale: 8.0 +2023-02-06 08:04:05,356 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.729e+02 2.498e+02 3.176e+02 4.184e+02 8.861e+02, threshold=6.353e+02, percent-clipped=6.0 +2023-02-06 08:04:11,429 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72652.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:04:19,759 INFO [train.py:901] (3/4) Epoch 9, batch 8000, loss[loss=0.2327, simple_loss=0.3253, pruned_loss=0.07006, over 8335.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.3162, pruned_loss=0.08519, over 1608277.53 frames. ], batch size: 25, lr: 8.29e-03, grad_scale: 8.0 +2023-02-06 08:04:27,787 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72677.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:04:34,992 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72688.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:04:49,794 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0070, 1.4963, 1.6821, 1.5194, 0.9662, 1.4694, 1.6323, 1.4176], + device='cuda:3'), covar=tensor([0.0523, 0.1240, 0.1659, 0.1356, 0.0592, 0.1467, 0.0691, 0.0630], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0157, 0.0196, 0.0161, 0.0107, 0.0167, 0.0120, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 08:04:52,985 INFO [train.py:901] (3/4) Epoch 9, batch 8050, loss[loss=0.1872, simple_loss=0.2641, pruned_loss=0.0551, over 7796.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.3145, pruned_loss=0.0853, over 1594742.41 frames. ], batch size: 19, lr: 8.29e-03, grad_scale: 8.0 +2023-02-06 08:05:11,528 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.212e+02 2.635e+02 3.102e+02 3.711e+02 7.462e+02, threshold=6.205e+02, percent-clipped=1.0 +2023-02-06 08:05:25,745 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 08:05:31,306 INFO [train.py:901] (3/4) Epoch 10, batch 0, loss[loss=0.2893, simple_loss=0.3549, pruned_loss=0.1119, over 8334.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3549, pruned_loss=0.1119, over 8334.00 frames. ], batch size: 26, lr: 7.88e-03, grad_scale: 8.0 +2023-02-06 08:05:31,306 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 08:05:43,258 INFO [train.py:935] (3/4) Epoch 10, validation: loss=0.1954, simple_loss=0.295, pruned_loss=0.0479, over 944034.00 frames. +2023-02-06 08:05:43,259 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 08:05:57,129 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 08:06:07,501 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-06 08:06:17,960 INFO [train.py:901] (3/4) Epoch 10, batch 50, loss[loss=0.2573, simple_loss=0.3393, pruned_loss=0.08767, over 8370.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3174, pruned_loss=0.08676, over 364930.87 frames. ], batch size: 24, lr: 7.88e-03, grad_scale: 8.0 +2023-02-06 08:06:21,761 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72803.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:06:31,234 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 08:06:49,407 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.516e+02 2.716e+02 3.124e+02 3.887e+02 7.160e+02, threshold=6.248e+02, percent-clipped=5.0 +2023-02-06 08:06:52,303 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 08:06:52,961 INFO [train.py:901] (3/4) Epoch 10, batch 100, loss[loss=0.2313, simple_loss=0.287, pruned_loss=0.08778, over 7540.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3181, pruned_loss=0.08631, over 645951.51 frames. ], batch size: 18, lr: 7.88e-03, grad_scale: 8.0 +2023-02-06 08:07:03,804 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72862.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:07:22,340 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72887.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:07:30,273 INFO [train.py:901] (3/4) Epoch 10, batch 150, loss[loss=0.3054, simple_loss=0.3563, pruned_loss=0.1272, over 6942.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3179, pruned_loss=0.08635, over 858725.85 frames. ], batch size: 72, lr: 7.87e-03, grad_scale: 8.0 +2023-02-06 08:07:33,245 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4276, 1.9590, 3.0648, 2.4452, 2.7128, 2.1886, 1.7219, 1.5277], + device='cuda:3'), covar=tensor([0.3343, 0.3725, 0.0992, 0.2115, 0.1774, 0.1951, 0.1633, 0.3644], + device='cuda:3'), in_proj_covar=tensor([0.0865, 0.0828, 0.0700, 0.0817, 0.0905, 0.0763, 0.0689, 0.0744], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 08:07:36,683 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6625, 1.8115, 4.4198, 1.9189, 2.3901, 5.1683, 4.9663, 4.5097], + device='cuda:3'), covar=tensor([0.0946, 0.1446, 0.0250, 0.1846, 0.1066, 0.0154, 0.0274, 0.0516], + device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0289, 0.0251, 0.0279, 0.0264, 0.0230, 0.0312, 0.0285], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 08:07:40,081 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72912.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:08:01,160 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.888e+02 2.670e+02 3.307e+02 4.288e+02 9.841e+02, threshold=6.614e+02, percent-clipped=3.0 +2023-02-06 08:08:04,559 INFO [train.py:901] (3/4) Epoch 10, batch 200, loss[loss=0.2462, simple_loss=0.3196, pruned_loss=0.08642, over 8480.00 frames. ], tot_loss[loss=0.2452, simple_loss=0.3183, pruned_loss=0.086, over 1024857.49 frames. ], batch size: 26, lr: 7.87e-03, grad_scale: 8.0 +2023-02-06 08:08:29,437 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72982.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 08:08:41,016 INFO [train.py:901] (3/4) Epoch 10, batch 250, loss[loss=0.276, simple_loss=0.3299, pruned_loss=0.1111, over 7234.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3184, pruned_loss=0.08589, over 1160010.25 frames. ], batch size: 71, lr: 7.87e-03, grad_scale: 8.0 +2023-02-06 08:08:47,842 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 08:08:56,862 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 08:09:02,436 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2092, 1.4875, 1.6353, 1.3010, 1.1510, 1.3149, 1.9112, 1.7327], + device='cuda:3'), covar=tensor([0.0537, 0.1309, 0.1726, 0.1432, 0.0590, 0.1669, 0.0657, 0.0596], + device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0155, 0.0196, 0.0160, 0.0107, 0.0167, 0.0120, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 08:09:07,478 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 +2023-02-06 08:09:12,572 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.823e+02 2.688e+02 3.158e+02 3.760e+02 5.735e+02, threshold=6.316e+02, percent-clipped=0.0 +2023-02-06 08:09:16,048 INFO [train.py:901] (3/4) Epoch 10, batch 300, loss[loss=0.2718, simple_loss=0.3469, pruned_loss=0.0984, over 8465.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3194, pruned_loss=0.08667, over 1262735.35 frames. ], batch size: 29, lr: 7.87e-03, grad_scale: 8.0 +2023-02-06 08:09:23,774 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73059.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:09:40,909 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73084.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:09:51,614 INFO [train.py:901] (3/4) Epoch 10, batch 350, loss[loss=0.2571, simple_loss=0.331, pruned_loss=0.0916, over 8485.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3188, pruned_loss=0.08648, over 1341612.61 frames. ], batch size: 28, lr: 7.86e-03, grad_scale: 16.0 +2023-02-06 08:09:55,108 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5779, 2.0039, 3.3170, 1.2913, 2.4886, 1.9879, 1.6003, 2.2716], + device='cuda:3'), covar=tensor([0.1634, 0.2196, 0.0646, 0.3948, 0.1335, 0.2700, 0.1783, 0.2042], + device='cuda:3'), in_proj_covar=tensor([0.0481, 0.0495, 0.0534, 0.0567, 0.0607, 0.0540, 0.0466, 0.0605], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 08:10:23,506 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.360e+02 2.632e+02 3.058e+02 3.924e+02 7.931e+02, threshold=6.116e+02, percent-clipped=5.0 +2023-02-06 08:10:26,911 INFO [train.py:901] (3/4) Epoch 10, batch 400, loss[loss=0.2828, simple_loss=0.3587, pruned_loss=0.1035, over 8281.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3191, pruned_loss=0.08632, over 1405767.97 frames. ], batch size: 23, lr: 7.86e-03, grad_scale: 16.0 +2023-02-06 08:10:49,317 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6423, 2.8483, 2.0756, 2.2507, 2.2970, 1.6824, 2.1599, 2.2509], + device='cuda:3'), covar=tensor([0.1493, 0.0338, 0.0986, 0.0603, 0.0615, 0.1347, 0.1089, 0.1046], + device='cuda:3'), in_proj_covar=tensor([0.0343, 0.0231, 0.0310, 0.0297, 0.0300, 0.0318, 0.0335, 0.0307], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 08:11:01,334 INFO [train.py:901] (3/4) Epoch 10, batch 450, loss[loss=0.2106, simple_loss=0.2936, pruned_loss=0.06377, over 8363.00 frames. ], tot_loss[loss=0.245, simple_loss=0.3185, pruned_loss=0.08574, over 1451257.16 frames. ], batch size: 24, lr: 7.86e-03, grad_scale: 16.0 +2023-02-06 08:11:33,882 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.783e+02 2.629e+02 3.140e+02 3.877e+02 8.143e+02, threshold=6.279e+02, percent-clipped=4.0 +2023-02-06 08:11:35,980 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3387, 4.3707, 3.9569, 1.8801, 3.8700, 3.9728, 4.0107, 3.5145], + device='cuda:3'), covar=tensor([0.0803, 0.0475, 0.0819, 0.4683, 0.0751, 0.0732, 0.1113, 0.0910], + device='cuda:3'), in_proj_covar=tensor([0.0461, 0.0358, 0.0375, 0.0477, 0.0373, 0.0357, 0.0367, 0.0318], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 08:11:37,155 INFO [train.py:901] (3/4) Epoch 10, batch 500, loss[loss=0.2394, simple_loss=0.314, pruned_loss=0.08246, over 8607.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3194, pruned_loss=0.08649, over 1489094.26 frames. ], batch size: 39, lr: 7.86e-03, grad_scale: 16.0 +2023-02-06 08:11:42,548 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73256.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:12:10,509 INFO [train.py:901] (3/4) Epoch 10, batch 550, loss[loss=0.2344, simple_loss=0.3283, pruned_loss=0.07025, over 8345.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3183, pruned_loss=0.08575, over 1517327.81 frames. ], batch size: 26, lr: 7.85e-03, grad_scale: 16.0 +2023-02-06 08:12:18,668 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8724, 1.4507, 3.3626, 1.3260, 2.1776, 3.7091, 3.7486, 3.0972], + device='cuda:3'), covar=tensor([0.1079, 0.1483, 0.0350, 0.1957, 0.1005, 0.0218, 0.0413, 0.0619], + device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0290, 0.0252, 0.0282, 0.0265, 0.0231, 0.0314, 0.0290], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 08:12:19,359 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73311.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:12:19,472 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1424, 1.7776, 2.5613, 2.1647, 2.3666, 2.0106, 1.5754, 0.9963], + device='cuda:3'), covar=tensor([0.3465, 0.3393, 0.0990, 0.1950, 0.1569, 0.1809, 0.1555, 0.3493], + device='cuda:3'), in_proj_covar=tensor([0.0860, 0.0831, 0.0701, 0.0817, 0.0907, 0.0763, 0.0690, 0.0744], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 08:12:23,931 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1592, 1.6813, 1.6202, 1.3534, 1.0703, 1.4270, 1.8789, 1.8116], + device='cuda:3'), covar=tensor([0.0488, 0.1172, 0.1779, 0.1375, 0.0599, 0.1541, 0.0678, 0.0568], + device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0157, 0.0197, 0.0161, 0.0108, 0.0168, 0.0120, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:3') +2023-02-06 08:12:29,157 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73326.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 08:12:41,609 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.623e+02 2.475e+02 3.100e+02 3.629e+02 1.040e+03, threshold=6.201e+02, percent-clipped=3.0 +2023-02-06 08:12:44,822 INFO [train.py:901] (3/4) Epoch 10, batch 600, loss[loss=0.2851, simple_loss=0.3432, pruned_loss=0.1135, over 8498.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3192, pruned_loss=0.08604, over 1542106.81 frames. ], batch size: 28, lr: 7.85e-03, grad_scale: 16.0 +2023-02-06 08:12:56,196 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 08:13:01,725 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73371.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:13:20,007 INFO [train.py:901] (3/4) Epoch 10, batch 650, loss[loss=0.2635, simple_loss=0.3371, pruned_loss=0.09498, over 8099.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3188, pruned_loss=0.08642, over 1557417.10 frames. ], batch size: 23, lr: 7.85e-03, grad_scale: 16.0 +2023-02-06 08:13:50,088 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73441.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 08:13:51,168 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.848e+02 2.469e+02 3.040e+02 3.840e+02 6.530e+02, threshold=6.081e+02, percent-clipped=1.0 +2023-02-06 08:13:54,587 INFO [train.py:901] (3/4) Epoch 10, batch 700, loss[loss=0.2034, simple_loss=0.2714, pruned_loss=0.06769, over 7716.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.3172, pruned_loss=0.08496, over 1572039.56 frames. ], batch size: 18, lr: 7.84e-03, grad_scale: 16.0 +2023-02-06 08:14:31,478 INFO [train.py:901] (3/4) Epoch 10, batch 750, loss[loss=0.2127, simple_loss=0.2878, pruned_loss=0.06875, over 8129.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.3151, pruned_loss=0.08433, over 1573584.20 frames. ], batch size: 22, lr: 7.84e-03, grad_scale: 16.0 +2023-02-06 08:14:45,811 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 08:14:54,756 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 08:15:02,253 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.751e+02 2.756e+02 3.307e+02 3.958e+02 8.111e+02, threshold=6.615e+02, percent-clipped=6.0 +2023-02-06 08:15:02,469 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9652, 2.4020, 1.9216, 2.9245, 1.3617, 1.6049, 1.7550, 2.4539], + device='cuda:3'), covar=tensor([0.0871, 0.0837, 0.1116, 0.0385, 0.1294, 0.1622, 0.1250, 0.0749], + device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0221, 0.0264, 0.0220, 0.0223, 0.0260, 0.0267, 0.0228], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 08:15:05,712 INFO [train.py:901] (3/4) Epoch 10, batch 800, loss[loss=0.2469, simple_loss=0.329, pruned_loss=0.08239, over 8475.00 frames. ], tot_loss[loss=0.2421, simple_loss=0.3152, pruned_loss=0.08446, over 1583977.61 frames. ], batch size: 29, lr: 7.84e-03, grad_scale: 16.0 +2023-02-06 08:15:17,222 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3076, 1.4621, 4.5116, 1.6966, 3.9550, 3.8265, 4.0746, 3.9473], + device='cuda:3'), covar=tensor([0.0534, 0.3752, 0.0448, 0.3161, 0.1188, 0.0892, 0.0537, 0.0651], + device='cuda:3'), in_proj_covar=tensor([0.0444, 0.0556, 0.0557, 0.0510, 0.0585, 0.0496, 0.0488, 0.0552], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 08:15:41,988 INFO [train.py:901] (3/4) Epoch 10, batch 850, loss[loss=0.2432, simple_loss=0.3201, pruned_loss=0.08312, over 8319.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.315, pruned_loss=0.08429, over 1589420.76 frames. ], batch size: 26, lr: 7.84e-03, grad_scale: 16.0 +2023-02-06 08:15:49,886 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73608.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:16:03,001 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73627.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:16:13,771 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.688e+02 2.847e+02 3.470e+02 4.482e+02 1.720e+03, threshold=6.940e+02, percent-clipped=10.0 +2023-02-06 08:16:17,263 INFO [train.py:901] (3/4) Epoch 10, batch 900, loss[loss=0.2108, simple_loss=0.2849, pruned_loss=0.06835, over 7433.00 frames. ], tot_loss[loss=0.2408, simple_loss=0.3142, pruned_loss=0.08373, over 1596054.95 frames. ], batch size: 17, lr: 7.83e-03, grad_scale: 16.0 +2023-02-06 08:16:20,227 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73652.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:16:22,225 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73655.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:16:52,130 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73697.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 08:16:52,448 INFO [train.py:901] (3/4) Epoch 10, batch 950, loss[loss=0.2635, simple_loss=0.3294, pruned_loss=0.09878, over 7973.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3153, pruned_loss=0.08452, over 1601168.40 frames. ], batch size: 21, lr: 7.83e-03, grad_scale: 8.0 +2023-02-06 08:17:10,348 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73722.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 08:17:18,303 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 08:17:22,464 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5648, 1.4651, 4.7039, 1.9797, 4.1250, 3.9310, 4.2771, 4.1150], + device='cuda:3'), covar=tensor([0.0508, 0.4522, 0.0417, 0.3194, 0.1079, 0.0782, 0.0521, 0.0577], + device='cuda:3'), in_proj_covar=tensor([0.0448, 0.0560, 0.0560, 0.0509, 0.0590, 0.0501, 0.0490, 0.0554], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 08:17:24,917 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.904e+02 2.751e+02 3.323e+02 4.211e+02 1.163e+03, threshold=6.645e+02, percent-clipped=9.0 +2023-02-06 08:17:27,464 INFO [train.py:901] (3/4) Epoch 10, batch 1000, loss[loss=0.2259, simple_loss=0.3048, pruned_loss=0.07355, over 8236.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3152, pruned_loss=0.08456, over 1601692.94 frames. ], batch size: 22, lr: 7.83e-03, grad_scale: 8.0 +2023-02-06 08:17:28,907 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73750.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:17:42,245 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73770.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:17:50,797 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 08:18:00,658 INFO [train.py:901] (3/4) Epoch 10, batch 1050, loss[loss=0.2137, simple_loss=0.3016, pruned_loss=0.06288, over 8137.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.3168, pruned_loss=0.08517, over 1607262.95 frames. ], batch size: 22, lr: 7.83e-03, grad_scale: 8.0 +2023-02-06 08:18:01,390 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 08:18:34,182 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.634e+02 3.058e+02 3.903e+02 1.179e+03, threshold=6.116e+02, percent-clipped=2.0 +2023-02-06 08:18:36,809 INFO [train.py:901] (3/4) Epoch 10, batch 1100, loss[loss=0.2692, simple_loss=0.3409, pruned_loss=0.09878, over 8502.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.3172, pruned_loss=0.08531, over 1608590.79 frames. ], batch size: 26, lr: 7.82e-03, grad_scale: 8.0 +2023-02-06 08:19:04,659 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6684, 1.8221, 2.2689, 1.4947, 1.0646, 2.4506, 0.3860, 1.2111], + device='cuda:3'), covar=tensor([0.2590, 0.1628, 0.0539, 0.2781, 0.4683, 0.0430, 0.3624, 0.2414], + device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0162, 0.0095, 0.0215, 0.0252, 0.0097, 0.0160, 0.0162], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 08:19:09,833 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 08:19:10,503 INFO [train.py:901] (3/4) Epoch 10, batch 1150, loss[loss=0.2619, simple_loss=0.3275, pruned_loss=0.09814, over 7819.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.3174, pruned_loss=0.08524, over 1612420.12 frames. ], batch size: 20, lr: 7.82e-03, grad_scale: 8.0 +2023-02-06 08:19:42,449 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.766e+02 2.356e+02 2.791e+02 3.726e+02 1.227e+03, threshold=5.583e+02, percent-clipped=4.0 +2023-02-06 08:19:45,137 INFO [train.py:901] (3/4) Epoch 10, batch 1200, loss[loss=0.2527, simple_loss=0.3283, pruned_loss=0.08855, over 8355.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3178, pruned_loss=0.0849, over 1619916.12 frames. ], batch size: 26, lr: 7.82e-03, grad_scale: 8.0 +2023-02-06 08:19:48,562 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73952.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:20:20,063 INFO [train.py:901] (3/4) Epoch 10, batch 1250, loss[loss=0.2684, simple_loss=0.3378, pruned_loss=0.09955, over 8651.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.3157, pruned_loss=0.0838, over 1616521.38 frames. ], batch size: 34, lr: 7.82e-03, grad_scale: 8.0 +2023-02-06 08:20:25,445 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 +2023-02-06 08:20:39,873 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74026.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:20:48,615 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4670, 2.0140, 3.0592, 2.4509, 2.7987, 2.2015, 1.7767, 1.6208], + device='cuda:3'), covar=tensor([0.3052, 0.3350, 0.0987, 0.2063, 0.1501, 0.1840, 0.1488, 0.3389], + device='cuda:3'), in_proj_covar=tensor([0.0858, 0.0823, 0.0706, 0.0812, 0.0903, 0.0765, 0.0686, 0.0741], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 08:20:51,572 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.914e+02 2.519e+02 3.075e+02 3.983e+02 7.817e+02, threshold=6.150e+02, percent-clipped=4.0 +2023-02-06 08:20:54,941 INFO [train.py:901] (3/4) Epoch 10, batch 1300, loss[loss=0.2394, simple_loss=0.2918, pruned_loss=0.0935, over 7564.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.3147, pruned_loss=0.08341, over 1614318.85 frames. ], batch size: 18, lr: 7.81e-03, grad_scale: 8.0 +2023-02-06 08:20:57,207 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74051.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:21:07,836 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74067.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:21:19,127 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74082.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:21:26,840 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74094.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:21:29,354 INFO [train.py:901] (3/4) Epoch 10, batch 1350, loss[loss=0.2447, simple_loss=0.3144, pruned_loss=0.08754, over 7934.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3159, pruned_loss=0.08424, over 1616930.26 frames. ], batch size: 20, lr: 7.81e-03, grad_scale: 8.0 +2023-02-06 08:21:30,434 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.22 vs. limit=5.0 +2023-02-06 08:21:59,541 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.726e+02 2.773e+02 3.448e+02 4.052e+02 8.675e+02, threshold=6.895e+02, percent-clipped=5.0 +2023-02-06 08:22:02,258 INFO [train.py:901] (3/4) Epoch 10, batch 1400, loss[loss=0.2718, simple_loss=0.3566, pruned_loss=0.09354, over 8499.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3176, pruned_loss=0.08556, over 1616592.74 frames. ], batch size: 26, lr: 7.81e-03, grad_scale: 8.0 +2023-02-06 08:22:02,453 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4191, 1.8563, 1.7866, 1.0533, 1.8047, 1.3870, 0.4856, 1.6794], + device='cuda:3'), covar=tensor([0.0362, 0.0198, 0.0165, 0.0339, 0.0271, 0.0553, 0.0507, 0.0171], + device='cuda:3'), in_proj_covar=tensor([0.0366, 0.0303, 0.0256, 0.0357, 0.0286, 0.0447, 0.0340, 0.0330], + device='cuda:3'), out_proj_covar=tensor([1.0716e-04, 8.6524e-05, 7.3501e-05, 1.0265e-04, 8.3262e-05, 1.4058e-04, + 9.9860e-05, 9.6205e-05], device='cuda:3') +2023-02-06 08:22:38,033 INFO [train.py:901] (3/4) Epoch 10, batch 1450, loss[loss=0.2649, simple_loss=0.3268, pruned_loss=0.1015, over 8230.00 frames. ], tot_loss[loss=0.244, simple_loss=0.3174, pruned_loss=0.08528, over 1617916.94 frames. ], batch size: 22, lr: 7.81e-03, grad_scale: 8.0 +2023-02-06 08:22:41,667 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 08:22:45,872 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74209.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:23:01,080 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2023-02-06 08:23:09,187 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.838e+02 2.525e+02 3.045e+02 3.954e+02 1.310e+03, threshold=6.089e+02, percent-clipped=4.0 +2023-02-06 08:23:11,853 INFO [train.py:901] (3/4) Epoch 10, batch 1500, loss[loss=0.2377, simple_loss=0.3036, pruned_loss=0.08591, over 7647.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.3167, pruned_loss=0.08454, over 1620472.10 frames. ], batch size: 19, lr: 7.80e-03, grad_scale: 8.0 +2023-02-06 08:23:18,344 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74258.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:23:46,601 INFO [train.py:901] (3/4) Epoch 10, batch 1550, loss[loss=0.2011, simple_loss=0.2825, pruned_loss=0.05987, over 8202.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3158, pruned_loss=0.08396, over 1618108.51 frames. ], batch size: 23, lr: 7.80e-03, grad_scale: 8.0 +2023-02-06 08:23:57,170 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.1988, 1.9870, 5.3194, 2.2856, 4.7568, 4.5022, 4.9448, 4.8103], + device='cuda:3'), covar=tensor([0.0494, 0.3975, 0.0389, 0.3031, 0.1008, 0.0843, 0.0441, 0.0497], + device='cuda:3'), in_proj_covar=tensor([0.0455, 0.0564, 0.0567, 0.0517, 0.0588, 0.0505, 0.0491, 0.0556], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 08:24:05,684 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74323.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:24:17,381 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2428, 1.4793, 2.1607, 1.0974, 1.4665, 1.4432, 1.3249, 1.5371], + device='cuda:3'), covar=tensor([0.1712, 0.2091, 0.0803, 0.3578, 0.1666, 0.2929, 0.1805, 0.1980], + device='cuda:3'), in_proj_covar=tensor([0.0480, 0.0500, 0.0527, 0.0565, 0.0608, 0.0540, 0.0463, 0.0604], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 08:24:19,868 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.633e+02 2.540e+02 3.095e+02 3.981e+02 6.537e+02, threshold=6.190e+02, percent-clipped=3.0 +2023-02-06 08:24:22,697 INFO [train.py:901] (3/4) Epoch 10, batch 1600, loss[loss=0.215, simple_loss=0.292, pruned_loss=0.06897, over 7813.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.3144, pruned_loss=0.08297, over 1620529.86 frames. ], batch size: 20, lr: 7.80e-03, grad_scale: 8.0 +2023-02-06 08:24:22,909 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74348.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:24:28,415 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1478, 2.1872, 1.6057, 1.8696, 1.8340, 1.2332, 1.5636, 1.7096], + device='cuda:3'), covar=tensor([0.1179, 0.0332, 0.1032, 0.0524, 0.0645, 0.1388, 0.0830, 0.0751], + device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0239, 0.0316, 0.0303, 0.0307, 0.0323, 0.0342, 0.0311], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 08:24:56,948 INFO [train.py:901] (3/4) Epoch 10, batch 1650, loss[loss=0.2432, simple_loss=0.3124, pruned_loss=0.08704, over 8239.00 frames. ], tot_loss[loss=0.2398, simple_loss=0.3134, pruned_loss=0.08313, over 1619294.20 frames. ], batch size: 22, lr: 7.79e-03, grad_scale: 8.0 +2023-02-06 08:25:00,020 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3922, 1.8744, 2.8610, 2.3186, 2.6666, 2.1692, 1.7836, 1.2072], + device='cuda:3'), covar=tensor([0.3251, 0.3474, 0.0963, 0.2161, 0.1484, 0.1926, 0.1550, 0.3525], + device='cuda:3'), in_proj_covar=tensor([0.0859, 0.0828, 0.0708, 0.0817, 0.0907, 0.0767, 0.0689, 0.0742], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 08:25:12,623 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0981, 2.1467, 1.5839, 1.9353, 1.7417, 1.2529, 1.4724, 1.7238], + device='cuda:3'), covar=tensor([0.1265, 0.0350, 0.1103, 0.0451, 0.0693, 0.1430, 0.0884, 0.0782], + device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0238, 0.0316, 0.0302, 0.0307, 0.0322, 0.0342, 0.0310], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 08:25:18,074 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74426.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:25:30,268 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.456e+02 2.508e+02 3.008e+02 3.971e+02 8.483e+02, threshold=6.016e+02, percent-clipped=6.0 +2023-02-06 08:25:32,863 INFO [train.py:901] (3/4) Epoch 10, batch 1700, loss[loss=0.3005, simple_loss=0.3535, pruned_loss=0.1237, over 8598.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3148, pruned_loss=0.08412, over 1616172.82 frames. ], batch size: 31, lr: 7.79e-03, grad_scale: 8.0 +2023-02-06 08:25:44,218 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74465.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:26:00,776 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74490.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:26:03,007 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 +2023-02-06 08:26:05,906 INFO [train.py:901] (3/4) Epoch 10, batch 1750, loss[loss=0.2115, simple_loss=0.2793, pruned_loss=0.0718, over 7281.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3158, pruned_loss=0.08475, over 1618854.95 frames. ], batch size: 16, lr: 7.79e-03, grad_scale: 8.0 +2023-02-06 08:26:36,857 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74541.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:26:38,773 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.823e+02 2.804e+02 3.517e+02 4.698e+02 1.546e+03, threshold=7.034e+02, percent-clipped=7.0 +2023-02-06 08:26:41,533 INFO [train.py:901] (3/4) Epoch 10, batch 1800, loss[loss=0.2937, simple_loss=0.3526, pruned_loss=0.1173, over 8469.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3172, pruned_loss=0.08592, over 1618589.30 frames. ], batch size: 25, lr: 7.79e-03, grad_scale: 8.0 +2023-02-06 08:26:45,353 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2023-02-06 08:26:48,525 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6259, 4.7035, 4.1805, 1.8592, 4.0957, 4.2373, 4.3348, 3.9276], + device='cuda:3'), covar=tensor([0.0815, 0.0444, 0.0893, 0.4722, 0.0787, 0.0888, 0.0971, 0.0721], + device='cuda:3'), in_proj_covar=tensor([0.0450, 0.0358, 0.0380, 0.0469, 0.0372, 0.0354, 0.0363, 0.0317], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 08:27:14,908 INFO [train.py:901] (3/4) Epoch 10, batch 1850, loss[loss=0.1992, simple_loss=0.2662, pruned_loss=0.06611, over 8024.00 frames. ], tot_loss[loss=0.244, simple_loss=0.317, pruned_loss=0.08549, over 1621349.45 frames. ], batch size: 22, lr: 7.78e-03, grad_scale: 8.0 +2023-02-06 08:27:17,787 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74602.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:27:25,263 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74613.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:27:46,970 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.251e+02 2.722e+02 3.219e+02 4.226e+02 1.097e+03, threshold=6.437e+02, percent-clipped=2.0 +2023-02-06 08:27:50,400 INFO [train.py:901] (3/4) Epoch 10, batch 1900, loss[loss=0.2065, simple_loss=0.2755, pruned_loss=0.06872, over 7784.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.3161, pruned_loss=0.08486, over 1621497.91 frames. ], batch size: 19, lr: 7.78e-03, grad_scale: 8.0 +2023-02-06 08:28:13,885 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 08:28:25,274 INFO [train.py:901] (3/4) Epoch 10, batch 1950, loss[loss=0.2544, simple_loss=0.3148, pruned_loss=0.09699, over 7245.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.316, pruned_loss=0.08489, over 1622243.61 frames. ], batch size: 16, lr: 7.78e-03, grad_scale: 8.0 +2023-02-06 08:28:25,960 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 08:28:38,209 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74717.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:28:43,967 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 08:28:56,744 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.563e+02 2.465e+02 3.030e+02 3.717e+02 6.494e+02, threshold=6.060e+02, percent-clipped=3.0 +2023-02-06 08:28:59,513 INFO [train.py:901] (3/4) Epoch 10, batch 2000, loss[loss=0.2095, simple_loss=0.2924, pruned_loss=0.06328, over 7936.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.317, pruned_loss=0.08555, over 1618654.83 frames. ], batch size: 20, lr: 7.78e-03, grad_scale: 8.0 +2023-02-06 08:29:34,165 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74797.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:29:34,602 INFO [train.py:901] (3/4) Epoch 10, batch 2050, loss[loss=0.2908, simple_loss=0.3453, pruned_loss=0.1181, over 6857.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3178, pruned_loss=0.086, over 1618415.77 frames. ], batch size: 71, lr: 7.77e-03, grad_scale: 8.0 +2023-02-06 08:29:36,988 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.61 vs. limit=5.0 +2023-02-06 08:29:50,348 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74822.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:29:56,002 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-06 08:30:04,619 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.629e+02 2.471e+02 3.084e+02 4.282e+02 1.276e+03, threshold=6.169e+02, percent-clipped=5.0 +2023-02-06 08:30:07,351 INFO [train.py:901] (3/4) Epoch 10, batch 2100, loss[loss=0.2445, simple_loss=0.3322, pruned_loss=0.07845, over 8335.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.317, pruned_loss=0.08538, over 1616129.66 frames. ], batch size: 25, lr: 7.77e-03, grad_scale: 8.0 +2023-02-06 08:30:43,224 INFO [train.py:901] (3/4) Epoch 10, batch 2150, loss[loss=0.2531, simple_loss=0.3368, pruned_loss=0.08466, over 8732.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.318, pruned_loss=0.08577, over 1612130.27 frames. ], batch size: 30, lr: 7.77e-03, grad_scale: 8.0 +2023-02-06 08:31:02,639 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74927.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:31:09,371 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8337, 2.3314, 3.2091, 2.0484, 1.6852, 3.2149, 0.6773, 1.8944], + device='cuda:3'), covar=tensor([0.2489, 0.2272, 0.0420, 0.2691, 0.4667, 0.0424, 0.4540, 0.2141], + device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0165, 0.0092, 0.0213, 0.0255, 0.0097, 0.0161, 0.0158], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0001, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 08:31:13,918 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.764e+02 2.626e+02 3.226e+02 3.775e+02 6.882e+02, threshold=6.451e+02, percent-clipped=1.0 +2023-02-06 08:31:16,707 INFO [train.py:901] (3/4) Epoch 10, batch 2200, loss[loss=0.2626, simple_loss=0.3292, pruned_loss=0.09795, over 8091.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.3171, pruned_loss=0.08515, over 1614239.14 frames. ], batch size: 21, lr: 7.77e-03, grad_scale: 8.0 +2023-02-06 08:31:22,704 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74957.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:31:33,585 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74973.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:31:39,395 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74982.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:31:50,364 INFO [train.py:901] (3/4) Epoch 10, batch 2250, loss[loss=0.1943, simple_loss=0.2654, pruned_loss=0.0616, over 7700.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.316, pruned_loss=0.08457, over 1613989.79 frames. ], batch size: 18, lr: 7.76e-03, grad_scale: 8.0 +2023-02-06 08:31:50,561 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74998.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:32:23,117 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.597e+02 2.748e+02 3.468e+02 4.709e+02 1.048e+03, threshold=6.936e+02, percent-clipped=3.0 +2023-02-06 08:32:25,869 INFO [train.py:901] (3/4) Epoch 10, batch 2300, loss[loss=0.232, simple_loss=0.3194, pruned_loss=0.07235, over 8457.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3148, pruned_loss=0.08434, over 1611517.81 frames. ], batch size: 27, lr: 7.76e-03, grad_scale: 8.0 +2023-02-06 08:32:42,311 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75072.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:32:59,567 INFO [train.py:901] (3/4) Epoch 10, batch 2350, loss[loss=0.2277, simple_loss=0.3139, pruned_loss=0.07079, over 8479.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3148, pruned_loss=0.08408, over 1613106.76 frames. ], batch size: 29, lr: 7.76e-03, grad_scale: 8.0 +2023-02-06 08:33:01,993 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-06 08:33:33,027 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.526e+02 3.215e+02 4.182e+02 1.054e+03, threshold=6.430e+02, percent-clipped=5.0 +2023-02-06 08:33:35,803 INFO [train.py:901] (3/4) Epoch 10, batch 2400, loss[loss=0.2929, simple_loss=0.354, pruned_loss=0.1159, over 8558.00 frames. ], tot_loss[loss=0.2411, simple_loss=0.3147, pruned_loss=0.08375, over 1615803.30 frames. ], batch size: 31, lr: 7.76e-03, grad_scale: 8.0 +2023-02-06 08:34:08,759 INFO [train.py:901] (3/4) Epoch 10, batch 2450, loss[loss=0.1791, simple_loss=0.2525, pruned_loss=0.05284, over 7435.00 frames. ], tot_loss[loss=0.241, simple_loss=0.3142, pruned_loss=0.08395, over 1614042.44 frames. ], batch size: 17, lr: 7.75e-03, grad_scale: 8.0 +2023-02-06 08:34:40,830 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.699e+02 2.613e+02 3.092e+02 4.227e+02 1.037e+03, threshold=6.184e+02, percent-clipped=5.0 +2023-02-06 08:34:43,532 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75246.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:34:44,709 INFO [train.py:901] (3/4) Epoch 10, batch 2500, loss[loss=0.2405, simple_loss=0.3245, pruned_loss=0.07828, over 8327.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3155, pruned_loss=0.08444, over 1617321.47 frames. ], batch size: 26, lr: 7.75e-03, grad_scale: 8.0 +2023-02-06 08:35:00,245 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75271.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:35:11,672 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75288.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:35:18,077 INFO [train.py:901] (3/4) Epoch 10, batch 2550, loss[loss=0.219, simple_loss=0.3004, pruned_loss=0.06874, over 8089.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3161, pruned_loss=0.08463, over 1618243.99 frames. ], batch size: 21, lr: 7.75e-03, grad_scale: 8.0 +2023-02-06 08:35:33,985 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([6.0762, 1.8090, 6.1553, 2.2722, 5.4912, 5.1438, 5.7588, 5.5907], + device='cuda:3'), covar=tensor([0.0306, 0.3633, 0.0249, 0.2810, 0.0852, 0.0594, 0.0339, 0.0382], + device='cuda:3'), in_proj_covar=tensor([0.0448, 0.0559, 0.0566, 0.0516, 0.0589, 0.0506, 0.0492, 0.0555], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 08:35:36,598 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75326.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:35:38,120 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75328.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:35:49,134 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.507e+02 2.664e+02 3.245e+02 3.791e+02 6.757e+02, threshold=6.490e+02, percent-clipped=2.0 +2023-02-06 08:35:51,832 INFO [train.py:901] (3/4) Epoch 10, batch 2600, loss[loss=0.2248, simple_loss=0.3055, pruned_loss=0.07208, over 8032.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.315, pruned_loss=0.08412, over 1616765.70 frames. ], batch size: 22, lr: 7.75e-03, grad_scale: 8.0 +2023-02-06 08:35:55,417 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75353.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:36:02,041 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7759, 3.3924, 2.0943, 2.4757, 2.3785, 1.6668, 2.2819, 2.9004], + device='cuda:3'), covar=tensor([0.1553, 0.0313, 0.1127, 0.0790, 0.0877, 0.1649, 0.1224, 0.0827], + device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0236, 0.0314, 0.0299, 0.0308, 0.0322, 0.0341, 0.0313], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 08:36:08,176 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7406, 2.3621, 4.8716, 1.3838, 3.3047, 2.4743, 1.7964, 2.9993], + device='cuda:3'), covar=tensor([0.1516, 0.2181, 0.0550, 0.3649, 0.1369, 0.2444, 0.1688, 0.2015], + device='cuda:3'), in_proj_covar=tensor([0.0479, 0.0498, 0.0532, 0.0566, 0.0605, 0.0544, 0.0464, 0.0605], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 08:36:19,405 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75386.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:36:27,352 INFO [train.py:901] (3/4) Epoch 10, batch 2650, loss[loss=0.2226, simple_loss=0.3087, pruned_loss=0.06824, over 8335.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.3151, pruned_loss=0.0839, over 1611200.18 frames. ], batch size: 26, lr: 7.74e-03, grad_scale: 8.0 +2023-02-06 08:36:56,752 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75441.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:36:58,551 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.802e+02 2.723e+02 3.413e+02 4.384e+02 8.455e+02, threshold=6.827e+02, percent-clipped=3.0 +2023-02-06 08:37:01,351 INFO [train.py:901] (3/4) Epoch 10, batch 2700, loss[loss=0.2218, simple_loss=0.3061, pruned_loss=0.06872, over 8348.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.315, pruned_loss=0.08408, over 1611299.87 frames. ], batch size: 26, lr: 7.74e-03, grad_scale: 8.0 +2023-02-06 08:37:37,692 INFO [train.py:901] (3/4) Epoch 10, batch 2750, loss[loss=0.2258, simple_loss=0.2863, pruned_loss=0.08269, over 7704.00 frames. ], tot_loss[loss=0.2423, simple_loss=0.3159, pruned_loss=0.08434, over 1612209.58 frames. ], batch size: 18, lr: 7.74e-03, grad_scale: 8.0 +2023-02-06 08:37:47,098 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5190, 1.5507, 1.7463, 1.4181, 1.0536, 1.7764, 0.1128, 1.1393], + device='cuda:3'), covar=tensor([0.3443, 0.2033, 0.0750, 0.1700, 0.4681, 0.0626, 0.3392, 0.1987], + device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0165, 0.0094, 0.0217, 0.0258, 0.0099, 0.0164, 0.0161], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 08:37:55,834 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5685, 2.0803, 3.0805, 2.4039, 2.7122, 2.2636, 1.8141, 1.4458], + device='cuda:3'), covar=tensor([0.3205, 0.3540, 0.0950, 0.2377, 0.1881, 0.2050, 0.1639, 0.3893], + device='cuda:3'), in_proj_covar=tensor([0.0863, 0.0840, 0.0711, 0.0823, 0.0920, 0.0779, 0.0697, 0.0755], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 08:38:01,676 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6985, 1.9127, 1.6417, 2.3382, 1.1649, 1.3771, 1.6587, 1.8568], + device='cuda:3'), covar=tensor([0.0855, 0.0871, 0.1087, 0.0454, 0.1226, 0.1530, 0.0918, 0.0800], + device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0220, 0.0265, 0.0220, 0.0225, 0.0257, 0.0265, 0.0226], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 08:38:06,718 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 +2023-02-06 08:38:08,183 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.897e+02 2.609e+02 3.111e+02 3.957e+02 1.084e+03, threshold=6.223e+02, percent-clipped=3.0 +2023-02-06 08:38:10,727 INFO [train.py:901] (3/4) Epoch 10, batch 2800, loss[loss=0.313, simple_loss=0.3589, pruned_loss=0.1335, over 7511.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3156, pruned_loss=0.08436, over 1610786.07 frames. ], batch size: 71, lr: 7.74e-03, grad_scale: 8.0 +2023-02-06 08:38:39,098 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75590.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:38:45,072 INFO [train.py:901] (3/4) Epoch 10, batch 2850, loss[loss=0.2427, simple_loss=0.3078, pruned_loss=0.08886, over 8241.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3163, pruned_loss=0.08499, over 1612998.06 frames. ], batch size: 22, lr: 7.73e-03, grad_scale: 8.0 +2023-02-06 08:38:48,925 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 08:39:09,362 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75632.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:39:16,276 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75642.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:39:17,406 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.693e+02 2.670e+02 3.172e+02 3.749e+02 6.038e+02, threshold=6.343e+02, percent-clipped=0.0 +2023-02-06 08:39:19,599 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6291, 1.8016, 2.1682, 1.5886, 1.1674, 2.1866, 0.2707, 1.1894], + device='cuda:3'), covar=tensor([0.3079, 0.2000, 0.0551, 0.2451, 0.5213, 0.0558, 0.3963, 0.2325], + device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0164, 0.0093, 0.0217, 0.0257, 0.0099, 0.0163, 0.0160], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 08:39:20,067 INFO [train.py:901] (3/4) Epoch 10, batch 2900, loss[loss=0.2128, simple_loss=0.2894, pruned_loss=0.06808, over 8089.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3163, pruned_loss=0.0853, over 1608894.34 frames. ], batch size: 21, lr: 7.73e-03, grad_scale: 8.0 +2023-02-06 08:39:20,527 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-06 08:39:32,778 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75667.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:39:49,793 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 08:39:53,286 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75697.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:39:53,747 INFO [train.py:901] (3/4) Epoch 10, batch 2950, loss[loss=0.25, simple_loss=0.3234, pruned_loss=0.08824, over 8357.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3164, pruned_loss=0.08526, over 1610770.85 frames. ], batch size: 24, lr: 7.73e-03, grad_scale: 16.0 +2023-02-06 08:39:58,782 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75705.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:40:11,389 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75722.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:40:26,686 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.891e+02 2.480e+02 3.030e+02 3.596e+02 1.304e+03, threshold=6.060e+02, percent-clipped=4.0 +2023-02-06 08:40:28,982 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75747.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:40:29,500 INFO [train.py:901] (3/4) Epoch 10, batch 3000, loss[loss=0.2117, simple_loss=0.2944, pruned_loss=0.06453, over 8446.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3169, pruned_loss=0.08502, over 1617791.09 frames. ], batch size: 25, lr: 7.73e-03, grad_scale: 16.0 +2023-02-06 08:40:29,500 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 08:40:41,876 INFO [train.py:935] (3/4) Epoch 10, validation: loss=0.1918, simple_loss=0.2916, pruned_loss=0.04599, over 944034.00 frames. +2023-02-06 08:40:41,877 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 08:41:14,531 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.52 vs. limit=2.0 +2023-02-06 08:41:15,547 INFO [train.py:901] (3/4) Epoch 10, batch 3050, loss[loss=0.253, simple_loss=0.3287, pruned_loss=0.0887, over 8230.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3166, pruned_loss=0.08479, over 1618255.65 frames. ], batch size: 22, lr: 7.72e-03, grad_scale: 16.0 +2023-02-06 08:41:47,920 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.681e+02 2.518e+02 3.138e+02 4.468e+02 1.006e+03, threshold=6.276e+02, percent-clipped=13.0 +2023-02-06 08:41:50,035 INFO [train.py:901] (3/4) Epoch 10, batch 3100, loss[loss=0.264, simple_loss=0.334, pruned_loss=0.097, over 8449.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3175, pruned_loss=0.08501, over 1618636.20 frames. ], batch size: 27, lr: 7.72e-03, grad_scale: 8.0 +2023-02-06 08:41:55,058 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2018, 1.8264, 2.6671, 2.2186, 2.3815, 2.0467, 1.6678, 1.1922], + device='cuda:3'), covar=tensor([0.3205, 0.3210, 0.0926, 0.2040, 0.1401, 0.1831, 0.1507, 0.3284], + device='cuda:3'), in_proj_covar=tensor([0.0868, 0.0848, 0.0715, 0.0828, 0.0919, 0.0781, 0.0699, 0.0757], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 08:42:25,563 INFO [train.py:901] (3/4) Epoch 10, batch 3150, loss[loss=0.2652, simple_loss=0.3334, pruned_loss=0.0985, over 7025.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3189, pruned_loss=0.08587, over 1616052.43 frames. ], batch size: 71, lr: 7.72e-03, grad_scale: 8.0 +2023-02-06 08:42:48,067 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5198, 1.3147, 4.7204, 1.9027, 4.1354, 3.9185, 4.2148, 4.1434], + device='cuda:3'), covar=tensor([0.0488, 0.4205, 0.0406, 0.2801, 0.0936, 0.0725, 0.0478, 0.0510], + device='cuda:3'), in_proj_covar=tensor([0.0455, 0.0562, 0.0572, 0.0521, 0.0599, 0.0510, 0.0497, 0.0560], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 08:42:57,478 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.807e+02 2.584e+02 3.323e+02 3.941e+02 8.938e+02, threshold=6.646e+02, percent-clipped=3.0 +2023-02-06 08:42:59,542 INFO [train.py:901] (3/4) Epoch 10, batch 3200, loss[loss=0.2417, simple_loss=0.3253, pruned_loss=0.07906, over 8365.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3193, pruned_loss=0.08639, over 1618699.67 frames. ], batch size: 24, lr: 7.72e-03, grad_scale: 8.0 +2023-02-06 08:43:09,325 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75961.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:43:28,326 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75986.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:43:28,984 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4294, 2.7781, 1.9521, 2.2626, 2.3791, 1.5072, 2.1076, 2.2391], + device='cuda:3'), covar=tensor([0.1416, 0.0305, 0.0990, 0.0640, 0.0621, 0.1429, 0.0950, 0.0969], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0237, 0.0314, 0.0301, 0.0312, 0.0326, 0.0343, 0.0317], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 08:43:36,360 INFO [train.py:901] (3/4) Epoch 10, batch 3250, loss[loss=0.2004, simple_loss=0.2906, pruned_loss=0.05512, over 8190.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3174, pruned_loss=0.08475, over 1620868.58 frames. ], batch size: 23, lr: 7.71e-03, grad_scale: 8.0 +2023-02-06 08:43:41,258 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76003.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:43:57,880 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76028.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:44:09,009 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.743e+02 2.688e+02 3.300e+02 3.989e+02 9.835e+02, threshold=6.601e+02, percent-clipped=4.0 +2023-02-06 08:44:11,012 INFO [train.py:901] (3/4) Epoch 10, batch 3300, loss[loss=0.2072, simple_loss=0.2888, pruned_loss=0.06282, over 8242.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3168, pruned_loss=0.08432, over 1620352.65 frames. ], batch size: 22, lr: 7.71e-03, grad_scale: 8.0 +2023-02-06 08:44:20,564 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.0487, 1.2527, 4.3076, 1.6657, 3.7421, 3.6135, 3.8992, 3.7322], + device='cuda:3'), covar=tensor([0.0517, 0.3977, 0.0496, 0.2995, 0.1179, 0.0787, 0.0528, 0.0658], + device='cuda:3'), in_proj_covar=tensor([0.0449, 0.0556, 0.0564, 0.0516, 0.0591, 0.0504, 0.0491, 0.0555], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 08:44:47,577 INFO [train.py:901] (3/4) Epoch 10, batch 3350, loss[loss=0.2267, simple_loss=0.2962, pruned_loss=0.07856, over 8081.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.3161, pruned_loss=0.0838, over 1620714.62 frames. ], batch size: 21, lr: 7.71e-03, grad_scale: 8.0 +2023-02-06 08:44:58,403 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8148, 1.4091, 2.8436, 1.2694, 2.0706, 3.0092, 3.1255, 2.5398], + device='cuda:3'), covar=tensor([0.0941, 0.1362, 0.0393, 0.1992, 0.0821, 0.0311, 0.0571, 0.0724], + device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0293, 0.0252, 0.0283, 0.0268, 0.0233, 0.0318, 0.0288], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 08:45:18,633 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.604e+02 2.652e+02 3.239e+02 4.192e+02 7.352e+02, threshold=6.477e+02, percent-clipped=1.0 +2023-02-06 08:45:20,663 INFO [train.py:901] (3/4) Epoch 10, batch 3400, loss[loss=0.2264, simple_loss=0.2999, pruned_loss=0.07642, over 7979.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.3179, pruned_loss=0.085, over 1622729.81 frames. ], batch size: 21, lr: 7.71e-03, grad_scale: 8.0 +2023-02-06 08:45:55,812 INFO [train.py:901] (3/4) Epoch 10, batch 3450, loss[loss=0.2303, simple_loss=0.318, pruned_loss=0.0713, over 8325.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3182, pruned_loss=0.08558, over 1618876.53 frames. ], batch size: 25, lr: 7.70e-03, grad_scale: 8.0 +2023-02-06 08:46:30,137 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.511e+02 2.394e+02 3.045e+02 3.881e+02 9.338e+02, threshold=6.090e+02, percent-clipped=3.0 +2023-02-06 08:46:32,218 INFO [train.py:901] (3/4) Epoch 10, batch 3500, loss[loss=0.2622, simple_loss=0.3323, pruned_loss=0.09603, over 8145.00 frames. ], tot_loss[loss=0.2424, simple_loss=0.3161, pruned_loss=0.08438, over 1614108.00 frames. ], batch size: 22, lr: 7.70e-03, grad_scale: 8.0 +2023-02-06 08:46:48,062 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 08:46:57,629 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-06 08:46:59,307 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76287.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 08:47:06,735 INFO [train.py:901] (3/4) Epoch 10, batch 3550, loss[loss=0.2477, simple_loss=0.3177, pruned_loss=0.08884, over 8465.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.3156, pruned_loss=0.08389, over 1616847.97 frames. ], batch size: 25, lr: 7.70e-03, grad_scale: 8.0 +2023-02-06 08:47:17,274 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76312.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:47:39,380 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76341.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 08:47:42,024 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.850e+02 2.725e+02 3.470e+02 4.316e+02 7.747e+02, threshold=6.941e+02, percent-clipped=6.0 +2023-02-06 08:47:44,160 INFO [train.py:901] (3/4) Epoch 10, batch 3600, loss[loss=0.2336, simple_loss=0.3176, pruned_loss=0.07477, over 8483.00 frames. ], tot_loss[loss=0.2411, simple_loss=0.3154, pruned_loss=0.08343, over 1618181.90 frames. ], batch size: 25, lr: 7.70e-03, grad_scale: 8.0 +2023-02-06 08:48:18,380 INFO [train.py:901] (3/4) Epoch 10, batch 3650, loss[loss=0.2642, simple_loss=0.3377, pruned_loss=0.09535, over 8461.00 frames. ], tot_loss[loss=0.2405, simple_loss=0.3143, pruned_loss=0.08337, over 1613804.77 frames. ], batch size: 27, lr: 7.69e-03, grad_scale: 8.0 +2023-02-06 08:48:50,989 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.577e+02 2.530e+02 3.057e+02 3.624e+02 8.995e+02, threshold=6.114e+02, percent-clipped=3.0 +2023-02-06 08:48:52,995 INFO [train.py:901] (3/4) Epoch 10, batch 3700, loss[loss=0.2097, simple_loss=0.2816, pruned_loss=0.06894, over 7441.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.3128, pruned_loss=0.08282, over 1611380.40 frames. ], batch size: 17, lr: 7.69e-03, grad_scale: 8.0 +2023-02-06 08:48:54,933 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 08:49:28,890 INFO [train.py:901] (3/4) Epoch 10, batch 3750, loss[loss=0.2752, simple_loss=0.3523, pruned_loss=0.09898, over 8331.00 frames. ], tot_loss[loss=0.2394, simple_loss=0.3131, pruned_loss=0.08286, over 1611842.60 frames. ], batch size: 26, lr: 7.69e-03, grad_scale: 8.0 +2023-02-06 08:50:00,301 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.698e+02 2.765e+02 3.416e+02 4.278e+02 1.031e+03, threshold=6.832e+02, percent-clipped=4.0 +2023-02-06 08:50:02,995 INFO [train.py:901] (3/4) Epoch 10, batch 3800, loss[loss=0.2347, simple_loss=0.3095, pruned_loss=0.07992, over 8696.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3148, pruned_loss=0.08406, over 1612201.04 frames. ], batch size: 34, lr: 7.69e-03, grad_scale: 8.0 +2023-02-06 08:50:19,507 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4247, 2.7478, 1.6708, 2.2350, 2.1334, 1.3666, 1.8651, 2.1654], + device='cuda:3'), covar=tensor([0.1494, 0.0326, 0.1155, 0.0657, 0.0697, 0.1543, 0.1126, 0.0982], + device='cuda:3'), in_proj_covar=tensor([0.0344, 0.0236, 0.0312, 0.0297, 0.0304, 0.0323, 0.0339, 0.0314], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 08:50:38,441 INFO [train.py:901] (3/4) Epoch 10, batch 3850, loss[loss=0.2856, simple_loss=0.3567, pruned_loss=0.1073, over 8517.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.3145, pruned_loss=0.0836, over 1616258.50 frames. ], batch size: 28, lr: 7.68e-03, grad_scale: 8.0 +2023-02-06 08:50:59,044 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 08:51:00,490 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=76631.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 08:51:09,921 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.569e+02 2.563e+02 3.093e+02 4.191e+02 1.151e+03, threshold=6.187e+02, percent-clipped=5.0 +2023-02-06 08:51:11,978 INFO [train.py:901] (3/4) Epoch 10, batch 3900, loss[loss=0.2276, simple_loss=0.2701, pruned_loss=0.09257, over 7690.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.3139, pruned_loss=0.08322, over 1617591.71 frames. ], batch size: 18, lr: 7.68e-03, grad_scale: 8.0 +2023-02-06 08:51:17,437 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=76656.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:51:38,047 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=76685.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 08:51:48,068 INFO [train.py:901] (3/4) Epoch 10, batch 3950, loss[loss=0.2595, simple_loss=0.3429, pruned_loss=0.08801, over 8475.00 frames. ], tot_loss[loss=0.2398, simple_loss=0.3136, pruned_loss=0.08298, over 1615771.14 frames. ], batch size: 27, lr: 7.68e-03, grad_scale: 8.0 +2023-02-06 08:52:19,564 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.796e+02 2.430e+02 3.097e+02 3.693e+02 7.444e+02, threshold=6.193e+02, percent-clipped=4.0 +2023-02-06 08:52:20,454 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76746.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 08:52:21,609 INFO [train.py:901] (3/4) Epoch 10, batch 4000, loss[loss=0.2778, simple_loss=0.3435, pruned_loss=0.106, over 8518.00 frames. ], tot_loss[loss=0.2393, simple_loss=0.3131, pruned_loss=0.08271, over 1614210.20 frames. ], batch size: 26, lr: 7.68e-03, grad_scale: 8.0 +2023-02-06 08:52:37,416 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76771.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:52:56,080 INFO [train.py:901] (3/4) Epoch 10, batch 4050, loss[loss=0.2448, simple_loss=0.3141, pruned_loss=0.08775, over 7814.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3153, pruned_loss=0.08389, over 1617152.86 frames. ], batch size: 20, lr: 7.67e-03, grad_scale: 8.0 +2023-02-06 08:52:57,687 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76800.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 08:53:29,319 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.530e+02 2.637e+02 3.294e+02 4.061e+02 9.505e+02, threshold=6.587e+02, percent-clipped=7.0 +2023-02-06 08:53:31,235 INFO [train.py:901] (3/4) Epoch 10, batch 4100, loss[loss=0.3016, simple_loss=0.3686, pruned_loss=0.1173, over 8528.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.3155, pruned_loss=0.08417, over 1612530.27 frames. ], batch size: 26, lr: 7.67e-03, grad_scale: 8.0 +2023-02-06 08:53:37,245 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76857.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:53:48,071 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3093, 2.1928, 1.5613, 1.8468, 1.7695, 1.3555, 1.5931, 1.6723], + device='cuda:3'), covar=tensor([0.1121, 0.0346, 0.1014, 0.0575, 0.0661, 0.1331, 0.0866, 0.0798], + device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0234, 0.0311, 0.0297, 0.0304, 0.0325, 0.0339, 0.0310], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 08:54:04,757 INFO [train.py:901] (3/4) Epoch 10, batch 4150, loss[loss=0.2521, simple_loss=0.3166, pruned_loss=0.09375, over 8604.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.314, pruned_loss=0.08321, over 1612066.36 frames. ], batch size: 49, lr: 7.67e-03, grad_scale: 8.0 +2023-02-06 08:54:10,375 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.40 vs. limit=5.0 +2023-02-06 08:54:34,616 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.39 vs. limit=5.0 +2023-02-06 08:54:38,768 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.554e+02 2.505e+02 2.967e+02 3.617e+02 8.554e+02, threshold=5.933e+02, percent-clipped=2.0 +2023-02-06 08:54:39,588 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.7885, 5.9240, 5.1105, 2.5127, 5.1098, 5.5245, 5.5325, 5.1544], + device='cuda:3'), covar=tensor([0.0577, 0.0419, 0.0852, 0.4177, 0.0682, 0.0522, 0.1011, 0.0604], + device='cuda:3'), in_proj_covar=tensor([0.0451, 0.0362, 0.0372, 0.0471, 0.0366, 0.0357, 0.0364, 0.0311], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 08:54:40,857 INFO [train.py:901] (3/4) Epoch 10, batch 4200, loss[loss=0.2278, simple_loss=0.2959, pruned_loss=0.07986, over 7921.00 frames. ], tot_loss[loss=0.2404, simple_loss=0.3142, pruned_loss=0.08325, over 1612181.68 frames. ], batch size: 20, lr: 7.67e-03, grad_scale: 8.0 +2023-02-06 08:55:00,913 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 08:55:14,235 INFO [train.py:901] (3/4) Epoch 10, batch 4250, loss[loss=0.2428, simple_loss=0.309, pruned_loss=0.08832, over 7935.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.3141, pruned_loss=0.08362, over 1610166.60 frames. ], batch size: 20, lr: 7.66e-03, grad_scale: 8.0 +2023-02-06 08:55:17,201 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77002.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 08:55:23,802 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 08:55:34,055 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77027.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 08:55:34,065 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77027.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:55:46,496 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.721e+02 2.530e+02 3.131e+02 3.743e+02 6.568e+02, threshold=6.262e+02, percent-clipped=1.0 +2023-02-06 08:55:48,447 INFO [train.py:901] (3/4) Epoch 10, batch 4300, loss[loss=0.2938, simple_loss=0.3419, pruned_loss=0.1228, over 7932.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.3145, pruned_loss=0.08411, over 1607397.57 frames. ], batch size: 20, lr: 7.66e-03, grad_scale: 8.0 +2023-02-06 08:55:52,725 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77052.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:55:55,462 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77056.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 08:56:12,959 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77081.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 08:56:19,608 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7061, 1.4660, 1.6781, 1.5411, 1.1981, 1.5980, 2.2503, 2.0383], + device='cuda:3'), covar=tensor([0.0487, 0.1380, 0.1814, 0.1403, 0.0594, 0.1569, 0.0637, 0.0610], + device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0157, 0.0196, 0.0162, 0.0106, 0.0166, 0.0119, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 08:56:23,978 INFO [train.py:901] (3/4) Epoch 10, batch 4350, loss[loss=0.2562, simple_loss=0.3236, pruned_loss=0.09437, over 8467.00 frames. ], tot_loss[loss=0.2408, simple_loss=0.3143, pruned_loss=0.08362, over 1608527.68 frames. ], batch size: 25, lr: 7.66e-03, grad_scale: 8.0 +2023-02-06 08:56:30,758 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8375, 2.8150, 3.2914, 2.0378, 1.6409, 3.2102, 0.6250, 2.0254], + device='cuda:3'), covar=tensor([0.2296, 0.1192, 0.0380, 0.2496, 0.4393, 0.0488, 0.3918, 0.1942], + device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0160, 0.0091, 0.0207, 0.0251, 0.0099, 0.0159, 0.0154], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 08:56:33,989 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77113.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:56:53,837 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 08:56:54,997 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.789e+02 2.751e+02 3.331e+02 4.491e+02 1.022e+03, threshold=6.663e+02, percent-clipped=8.0 +2023-02-06 08:56:57,038 INFO [train.py:901] (3/4) Epoch 10, batch 4400, loss[loss=0.2145, simple_loss=0.2986, pruned_loss=0.06523, over 8457.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3155, pruned_loss=0.08403, over 1613771.95 frames. ], batch size: 25, lr: 7.66e-03, grad_scale: 8.0 +2023-02-06 08:57:33,159 INFO [train.py:901] (3/4) Epoch 10, batch 4450, loss[loss=0.2032, simple_loss=0.297, pruned_loss=0.05472, over 8102.00 frames. ], tot_loss[loss=0.2408, simple_loss=0.3146, pruned_loss=0.08356, over 1613601.18 frames. ], batch size: 23, lr: 7.65e-03, grad_scale: 8.0 +2023-02-06 08:57:35,381 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77201.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:57:36,009 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 08:57:54,205 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77229.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:58:04,751 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.793e+02 2.784e+02 3.289e+02 4.035e+02 8.452e+02, threshold=6.579e+02, percent-clipped=2.0 +2023-02-06 08:58:06,781 INFO [train.py:901] (3/4) Epoch 10, batch 4500, loss[loss=0.2005, simple_loss=0.2852, pruned_loss=0.05791, over 7815.00 frames. ], tot_loss[loss=0.2406, simple_loss=0.3142, pruned_loss=0.08352, over 1611654.91 frames. ], batch size: 20, lr: 7.65e-03, grad_scale: 8.0 +2023-02-06 08:58:27,560 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 08:58:31,892 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6339, 2.0190, 2.1925, 1.1292, 2.3803, 1.6433, 0.6466, 1.8400], + device='cuda:3'), covar=tensor([0.0398, 0.0199, 0.0189, 0.0366, 0.0226, 0.0494, 0.0512, 0.0176], + device='cuda:3'), in_proj_covar=tensor([0.0373, 0.0305, 0.0262, 0.0370, 0.0299, 0.0456, 0.0347, 0.0338], + device='cuda:3'), out_proj_covar=tensor([1.0826e-04, 8.6417e-05, 7.4624e-05, 1.0589e-04, 8.6720e-05, 1.4185e-04, + 1.0121e-04, 9.8024e-05], device='cuda:3') +2023-02-06 08:58:37,994 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77291.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:58:39,411 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.7990, 1.9578, 5.9218, 2.1209, 5.1425, 4.9236, 5.4706, 5.2578], + device='cuda:3'), covar=tensor([0.0486, 0.4009, 0.0329, 0.3298, 0.1149, 0.0809, 0.0528, 0.0561], + device='cuda:3'), in_proj_covar=tensor([0.0462, 0.0567, 0.0569, 0.0526, 0.0600, 0.0508, 0.0501, 0.0568], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 08:58:41,458 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77295.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:58:43,379 INFO [train.py:901] (3/4) Epoch 10, batch 4550, loss[loss=0.2805, simple_loss=0.3523, pruned_loss=0.1043, over 8464.00 frames. ], tot_loss[loss=0.2388, simple_loss=0.3128, pruned_loss=0.08244, over 1611123.99 frames. ], batch size: 27, lr: 7.65e-03, grad_scale: 8.0 +2023-02-06 08:58:50,337 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0967, 1.2541, 1.2117, 0.6730, 1.2442, 0.9928, 0.1374, 1.1749], + device='cuda:3'), covar=tensor([0.0241, 0.0181, 0.0161, 0.0279, 0.0218, 0.0553, 0.0438, 0.0161], + device='cuda:3'), in_proj_covar=tensor([0.0374, 0.0308, 0.0263, 0.0372, 0.0301, 0.0460, 0.0350, 0.0340], + device='cuda:3'), out_proj_covar=tensor([1.0883e-04, 8.7470e-05, 7.5054e-05, 1.0640e-04, 8.7437e-05, 1.4295e-04, + 1.0188e-04, 9.8406e-05], device='cuda:3') +2023-02-06 08:58:55,664 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77316.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 08:59:14,833 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.361e+02 2.639e+02 3.213e+02 4.072e+02 8.769e+02, threshold=6.426e+02, percent-clipped=3.0 +2023-02-06 08:59:16,946 INFO [train.py:901] (3/4) Epoch 10, batch 4600, loss[loss=0.2316, simple_loss=0.3169, pruned_loss=0.07313, over 8366.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.3147, pruned_loss=0.08399, over 1610798.26 frames. ], batch size: 24, lr: 7.65e-03, grad_scale: 8.0 +2023-02-06 08:59:35,820 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4965, 2.0777, 3.0278, 2.4343, 2.7155, 2.2159, 1.7866, 1.3400], + device='cuda:3'), covar=tensor([0.3326, 0.3489, 0.1050, 0.2146, 0.1687, 0.1866, 0.1521, 0.3760], + device='cuda:3'), in_proj_covar=tensor([0.0854, 0.0835, 0.0703, 0.0822, 0.0908, 0.0767, 0.0691, 0.0746], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 08:59:43,210 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-06 08:59:50,930 INFO [train.py:901] (3/4) Epoch 10, batch 4650, loss[loss=0.2277, simple_loss=0.3019, pruned_loss=0.07678, over 8427.00 frames. ], tot_loss[loss=0.2428, simple_loss=0.3158, pruned_loss=0.08488, over 1615083.75 frames. ], batch size: 48, lr: 7.64e-03, grad_scale: 8.0 +2023-02-06 09:00:25,440 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.953e+02 3.048e+02 3.591e+02 4.434e+02 8.168e+02, threshold=7.182e+02, percent-clipped=8.0 +2023-02-06 09:00:27,540 INFO [train.py:901] (3/4) Epoch 10, batch 4700, loss[loss=0.2478, simple_loss=0.331, pruned_loss=0.08237, over 8491.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.3148, pruned_loss=0.08446, over 1612631.17 frames. ], batch size: 29, lr: 7.64e-03, grad_scale: 8.0 +2023-02-06 09:00:33,885 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77457.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:00:47,328 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2385, 1.6383, 1.6912, 1.3851, 1.0655, 1.5196, 1.7866, 1.7266], + device='cuda:3'), covar=tensor([0.0499, 0.1159, 0.1608, 0.1358, 0.0601, 0.1517, 0.0666, 0.0579], + device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0155, 0.0194, 0.0160, 0.0106, 0.0165, 0.0118, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 09:01:02,748 INFO [train.py:901] (3/4) Epoch 10, batch 4750, loss[loss=0.2268, simple_loss=0.2832, pruned_loss=0.08518, over 7545.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.3148, pruned_loss=0.08397, over 1617681.23 frames. ], batch size: 18, lr: 7.64e-03, grad_scale: 8.0 +2023-02-06 09:01:28,497 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 09:01:30,530 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 09:01:35,881 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.596e+02 2.648e+02 3.312e+02 4.103e+02 1.054e+03, threshold=6.623e+02, percent-clipped=5.0 +2023-02-06 09:01:37,937 INFO [train.py:901] (3/4) Epoch 10, batch 4800, loss[loss=0.2253, simple_loss=0.2937, pruned_loss=0.07846, over 8085.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3156, pruned_loss=0.08479, over 1613865.31 frames. ], batch size: 21, lr: 7.64e-03, grad_scale: 8.0 +2023-02-06 09:01:54,061 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77572.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:01:54,103 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77572.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:01:54,648 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77573.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:02:00,287 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-06 09:02:10,902 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77597.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:02:11,384 INFO [train.py:901] (3/4) Epoch 10, batch 4850, loss[loss=0.2554, simple_loss=0.3345, pruned_loss=0.08814, over 8670.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.3159, pruned_loss=0.08523, over 1611591.02 frames. ], batch size: 39, lr: 7.63e-03, grad_scale: 8.0 +2023-02-06 09:02:16,290 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 09:02:38,007 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77634.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:02:38,638 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77635.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:02:42,018 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77639.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:02:46,052 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.616e+02 3.128e+02 3.870e+02 7.279e+02, threshold=6.256e+02, percent-clipped=1.0 +2023-02-06 09:02:48,055 INFO [train.py:901] (3/4) Epoch 10, batch 4900, loss[loss=0.1961, simple_loss=0.2736, pruned_loss=0.05928, over 7813.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3156, pruned_loss=0.08488, over 1615573.42 frames. ], batch size: 20, lr: 7.63e-03, grad_scale: 8.0 +2023-02-06 09:02:59,929 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.50 vs. limit=5.0 +2023-02-06 09:03:00,785 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.62 vs. limit=5.0 +2023-02-06 09:03:14,724 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1184, 1.6416, 1.6248, 1.2497, 1.1221, 1.4903, 1.8783, 1.6840], + device='cuda:3'), covar=tensor([0.0493, 0.1177, 0.1670, 0.1401, 0.0591, 0.1414, 0.0670, 0.0589], + device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0155, 0.0195, 0.0160, 0.0105, 0.0165, 0.0118, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 09:03:15,420 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77688.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:03:20,005 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77695.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:03:21,727 INFO [train.py:901] (3/4) Epoch 10, batch 4950, loss[loss=0.3122, simple_loss=0.3639, pruned_loss=0.1303, over 7009.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3153, pruned_loss=0.08507, over 1608952.01 frames. ], batch size: 71, lr: 7.63e-03, grad_scale: 8.0 +2023-02-06 09:03:38,878 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-06 09:03:54,524 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.777e+02 2.775e+02 3.348e+02 4.012e+02 9.680e+02, threshold=6.695e+02, percent-clipped=4.0 +2023-02-06 09:03:57,189 INFO [train.py:901] (3/4) Epoch 10, batch 5000, loss[loss=0.2839, simple_loss=0.3513, pruned_loss=0.1082, over 8356.00 frames. ], tot_loss[loss=0.2423, simple_loss=0.3148, pruned_loss=0.08487, over 1605906.24 frames. ], batch size: 24, lr: 7.63e-03, grad_scale: 8.0 +2023-02-06 09:03:58,711 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77750.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:04:01,945 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77754.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:04:14,753 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 09:04:30,716 INFO [train.py:901] (3/4) Epoch 10, batch 5050, loss[loss=0.2161, simple_loss=0.2922, pruned_loss=0.07001, over 7924.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3157, pruned_loss=0.0857, over 1606972.35 frames. ], batch size: 20, lr: 7.62e-03, grad_scale: 8.0 +2023-02-06 09:04:39,315 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 09:04:51,132 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77828.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:04:52,878 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 09:05:02,962 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.592e+02 2.592e+02 3.051e+02 4.098e+02 9.089e+02, threshold=6.102e+02, percent-clipped=4.0 +2023-02-06 09:05:05,646 INFO [train.py:901] (3/4) Epoch 10, batch 5100, loss[loss=0.2765, simple_loss=0.3495, pruned_loss=0.1017, over 8250.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3154, pruned_loss=0.08493, over 1610905.79 frames. ], batch size: 24, lr: 7.62e-03, grad_scale: 16.0 +2023-02-06 09:05:09,145 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77853.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:05:18,408 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6723, 2.3490, 4.4508, 1.2302, 3.0375, 2.0552, 1.6136, 2.6140], + device='cuda:3'), covar=tensor([0.1689, 0.2095, 0.0678, 0.3820, 0.1495, 0.2809, 0.1780, 0.2306], + device='cuda:3'), in_proj_covar=tensor([0.0488, 0.0510, 0.0534, 0.0576, 0.0613, 0.0556, 0.0467, 0.0608], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 09:05:40,062 INFO [train.py:901] (3/4) Epoch 10, batch 5150, loss[loss=0.2871, simple_loss=0.3455, pruned_loss=0.1144, over 8300.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3157, pruned_loss=0.08474, over 1614750.10 frames. ], batch size: 23, lr: 7.62e-03, grad_scale: 16.0 +2023-02-06 09:05:53,326 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.97 vs. limit=5.0 +2023-02-06 09:06:11,276 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77944.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:06:11,738 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.968e+02 2.805e+02 3.349e+02 3.898e+02 8.134e+02, threshold=6.697e+02, percent-clipped=4.0 +2023-02-06 09:06:13,811 INFO [train.py:901] (3/4) Epoch 10, batch 5200, loss[loss=0.2443, simple_loss=0.3241, pruned_loss=0.08218, over 8340.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3174, pruned_loss=0.08554, over 1613068.47 frames. ], batch size: 26, lr: 7.62e-03, grad_scale: 16.0 +2023-02-06 09:06:29,756 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77969.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:06:35,813 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77978.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:06:50,971 INFO [train.py:901] (3/4) Epoch 10, batch 5250, loss[loss=0.2205, simple_loss=0.3037, pruned_loss=0.06862, over 8638.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3161, pruned_loss=0.08458, over 1611551.02 frames. ], batch size: 27, lr: 7.61e-03, grad_scale: 16.0 +2023-02-06 09:06:56,857 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 09:06:57,789 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78006.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:07:00,626 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78010.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:07:14,857 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78031.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:07:17,424 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78035.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:07:19,992 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78039.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:07:23,912 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.508e+02 2.672e+02 3.378e+02 4.041e+02 9.848e+02, threshold=6.756e+02, percent-clipped=3.0 +2023-02-06 09:07:25,980 INFO [train.py:901] (3/4) Epoch 10, batch 5300, loss[loss=0.2556, simple_loss=0.3178, pruned_loss=0.09674, over 7969.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3156, pruned_loss=0.08402, over 1613407.29 frames. ], batch size: 21, lr: 7.61e-03, grad_scale: 16.0 +2023-02-06 09:07:57,654 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78093.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:08:00,804 INFO [train.py:901] (3/4) Epoch 10, batch 5350, loss[loss=0.215, simple_loss=0.2934, pruned_loss=0.06825, over 8083.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3158, pruned_loss=0.08426, over 1613090.67 frames. ], batch size: 21, lr: 7.61e-03, grad_scale: 16.0 +2023-02-06 09:08:14,893 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9184, 2.4017, 2.7082, 1.3964, 3.0455, 1.4914, 1.2621, 1.6903], + device='cuda:3'), covar=tensor([0.0545, 0.0259, 0.0250, 0.0499, 0.0260, 0.0724, 0.0599, 0.0381], + device='cuda:3'), in_proj_covar=tensor([0.0369, 0.0305, 0.0261, 0.0367, 0.0294, 0.0453, 0.0343, 0.0330], + device='cuda:3'), out_proj_covar=tensor([1.0693e-04, 8.6489e-05, 7.4541e-05, 1.0460e-04, 8.5137e-05, 1.4056e-04, + 9.9680e-05, 9.4978e-05], device='cuda:3') +2023-02-06 09:08:34,365 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.399e+02 2.713e+02 3.238e+02 4.266e+02 6.892e+02, threshold=6.476e+02, percent-clipped=1.0 +2023-02-06 09:08:35,735 INFO [train.py:901] (3/4) Epoch 10, batch 5400, loss[loss=0.2412, simple_loss=0.3251, pruned_loss=0.0787, over 8358.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.316, pruned_loss=0.0845, over 1611687.44 frames. ], batch size: 24, lr: 7.61e-03, grad_scale: 8.0 +2023-02-06 09:08:40,013 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78154.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:08:41,605 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.59 vs. limit=5.0 +2023-02-06 09:09:08,700 INFO [train.py:901] (3/4) Epoch 10, batch 5450, loss[loss=0.2441, simple_loss=0.3291, pruned_loss=0.07958, over 8510.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3148, pruned_loss=0.08407, over 1607554.50 frames. ], batch size: 29, lr: 7.60e-03, grad_scale: 8.0 +2023-02-06 09:09:43,432 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.916e+02 2.697e+02 3.396e+02 4.413e+02 8.943e+02, threshold=6.791e+02, percent-clipped=7.0 +2023-02-06 09:09:44,805 INFO [train.py:901] (3/4) Epoch 10, batch 5500, loss[loss=0.2238, simple_loss=0.2935, pruned_loss=0.07703, over 7550.00 frames. ], tot_loss[loss=0.2423, simple_loss=0.3155, pruned_loss=0.08454, over 1610180.59 frames. ], batch size: 18, lr: 7.60e-03, grad_scale: 8.0 +2023-02-06 09:09:46,319 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78250.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:09:46,830 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 09:10:18,553 INFO [train.py:901] (3/4) Epoch 10, batch 5550, loss[loss=0.2959, simple_loss=0.3541, pruned_loss=0.1188, over 8336.00 frames. ], tot_loss[loss=0.242, simple_loss=0.3157, pruned_loss=0.08417, over 1611775.32 frames. ], batch size: 26, lr: 7.60e-03, grad_scale: 8.0 +2023-02-06 09:10:52,577 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.601e+02 2.563e+02 3.102e+02 4.076e+02 7.679e+02, threshold=6.204e+02, percent-clipped=2.0 +2023-02-06 09:10:54,679 INFO [train.py:901] (3/4) Epoch 10, batch 5600, loss[loss=0.2866, simple_loss=0.3487, pruned_loss=0.1122, over 8504.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.3155, pruned_loss=0.08419, over 1615403.37 frames. ], batch size: 26, lr: 7.60e-03, grad_scale: 8.0 +2023-02-06 09:10:55,587 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78349.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:11:07,553 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3601, 4.3484, 3.9148, 2.1086, 3.8130, 3.8497, 3.9472, 3.6474], + device='cuda:3'), covar=tensor([0.0839, 0.0567, 0.1085, 0.4802, 0.0971, 0.1014, 0.1281, 0.0892], + device='cuda:3'), in_proj_covar=tensor([0.0454, 0.0357, 0.0371, 0.0472, 0.0366, 0.0361, 0.0368, 0.0317], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 09:11:12,345 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78374.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:11:28,351 INFO [train.py:901] (3/4) Epoch 10, batch 5650, loss[loss=0.1939, simple_loss=0.2699, pruned_loss=0.05894, over 7974.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.3159, pruned_loss=0.08452, over 1615396.19 frames. ], batch size: 21, lr: 7.59e-03, grad_scale: 8.0 +2023-02-06 09:11:32,098 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1651, 4.1612, 3.7446, 1.7980, 3.6194, 3.7120, 3.7899, 3.4478], + device='cuda:3'), covar=tensor([0.0862, 0.0592, 0.1082, 0.5009, 0.0947, 0.0883, 0.1246, 0.0884], + device='cuda:3'), in_proj_covar=tensor([0.0454, 0.0357, 0.0371, 0.0473, 0.0365, 0.0361, 0.0368, 0.0317], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 09:11:36,902 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78410.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:11:48,304 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 09:11:54,466 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78435.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:12:02,391 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.718e+02 2.443e+02 2.913e+02 3.480e+02 5.594e+02, threshold=5.826e+02, percent-clipped=0.0 +2023-02-06 09:12:03,752 INFO [train.py:901] (3/4) Epoch 10, batch 5700, loss[loss=0.227, simple_loss=0.3079, pruned_loss=0.0731, over 8553.00 frames. ], tot_loss[loss=0.2406, simple_loss=0.3145, pruned_loss=0.08335, over 1616802.01 frames. ], batch size: 31, lr: 7.59e-03, grad_scale: 8.0 +2023-02-06 09:12:16,136 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2719, 1.3394, 2.3408, 1.1599, 1.9579, 2.4648, 2.6130, 2.1186], + device='cuda:3'), covar=tensor([0.1049, 0.1269, 0.0467, 0.2072, 0.0757, 0.0401, 0.0595, 0.0791], + device='cuda:3'), in_proj_covar=tensor([0.0257, 0.0294, 0.0255, 0.0285, 0.0269, 0.0233, 0.0326, 0.0285], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 09:12:25,604 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78478.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:12:39,149 INFO [train.py:901] (3/4) Epoch 10, batch 5750, loss[loss=0.2394, simple_loss=0.3168, pruned_loss=0.08101, over 8640.00 frames. ], tot_loss[loss=0.2412, simple_loss=0.3153, pruned_loss=0.08352, over 1618760.22 frames. ], batch size: 34, lr: 7.59e-03, grad_scale: 8.0 +2023-02-06 09:12:42,075 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4549, 2.6954, 1.7977, 2.2452, 2.2515, 1.3644, 2.0552, 2.2295], + device='cuda:3'), covar=tensor([0.1519, 0.0360, 0.1099, 0.0652, 0.0645, 0.1480, 0.0993, 0.0865], + device='cuda:3'), in_proj_covar=tensor([0.0342, 0.0229, 0.0309, 0.0297, 0.0303, 0.0322, 0.0335, 0.0308], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 09:12:53,268 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 09:13:11,252 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.401e+02 2.793e+02 3.478e+02 4.404e+02 1.244e+03, threshold=6.955e+02, percent-clipped=11.0 +2023-02-06 09:13:12,618 INFO [train.py:901] (3/4) Epoch 10, batch 5800, loss[loss=0.2141, simple_loss=0.2981, pruned_loss=0.06499, over 8117.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3166, pruned_loss=0.08433, over 1619492.49 frames. ], batch size: 23, lr: 7.59e-03, grad_scale: 8.0 +2023-02-06 09:13:17,537 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 +2023-02-06 09:13:29,738 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6286, 2.7733, 1.8584, 2.2147, 2.1678, 1.4898, 1.9914, 2.2811], + device='cuda:3'), covar=tensor([0.1263, 0.0318, 0.0935, 0.0563, 0.0609, 0.1255, 0.0864, 0.0771], + device='cuda:3'), in_proj_covar=tensor([0.0340, 0.0229, 0.0308, 0.0295, 0.0302, 0.0319, 0.0334, 0.0308], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 09:13:31,655 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78574.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:13:45,431 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78594.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:13:48,031 INFO [train.py:901] (3/4) Epoch 10, batch 5850, loss[loss=0.2114, simple_loss=0.2817, pruned_loss=0.07061, over 7538.00 frames. ], tot_loss[loss=0.242, simple_loss=0.3157, pruned_loss=0.08412, over 1620126.42 frames. ], batch size: 18, lr: 7.58e-03, grad_scale: 8.0 +2023-02-06 09:14:19,898 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.740e+02 2.589e+02 3.164e+02 4.281e+02 9.296e+02, threshold=6.329e+02, percent-clipped=5.0 +2023-02-06 09:14:21,268 INFO [train.py:901] (3/4) Epoch 10, batch 5900, loss[loss=0.1951, simple_loss=0.279, pruned_loss=0.05565, over 8132.00 frames. ], tot_loss[loss=0.2411, simple_loss=0.3147, pruned_loss=0.0838, over 1618878.88 frames. ], batch size: 22, lr: 7.58e-03, grad_scale: 8.0 +2023-02-06 09:14:24,992 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-06 09:14:29,629 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-06 09:14:30,103 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.0698, 1.4140, 4.2799, 1.6386, 3.6624, 3.5626, 3.8462, 3.7061], + device='cuda:3'), covar=tensor([0.0618, 0.4139, 0.0490, 0.3134, 0.1316, 0.0809, 0.0596, 0.0707], + device='cuda:3'), in_proj_covar=tensor([0.0464, 0.0557, 0.0561, 0.0514, 0.0590, 0.0497, 0.0495, 0.0562], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 09:14:49,658 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0336, 1.2557, 1.1691, 0.5706, 1.1973, 0.9332, 0.1554, 1.1824], + device='cuda:3'), covar=tensor([0.0226, 0.0190, 0.0160, 0.0315, 0.0209, 0.0566, 0.0448, 0.0166], + device='cuda:3'), in_proj_covar=tensor([0.0379, 0.0308, 0.0265, 0.0376, 0.0303, 0.0462, 0.0351, 0.0336], + device='cuda:3'), out_proj_covar=tensor([1.0992e-04, 8.7306e-05, 7.5411e-05, 1.0730e-04, 8.7486e-05, 1.4334e-04, + 1.0197e-04, 9.6504e-05], device='cuda:3') +2023-02-06 09:14:57,574 INFO [train.py:901] (3/4) Epoch 10, batch 5950, loss[loss=0.1658, simple_loss=0.2408, pruned_loss=0.04544, over 7645.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.3151, pruned_loss=0.08386, over 1618133.08 frames. ], batch size: 19, lr: 7.58e-03, grad_scale: 8.0 +2023-02-06 09:15:05,422 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78709.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:15:30,045 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.711e+02 2.430e+02 2.939e+02 3.954e+02 7.661e+02, threshold=5.878e+02, percent-clipped=3.0 +2023-02-06 09:15:31,441 INFO [train.py:901] (3/4) Epoch 10, batch 6000, loss[loss=0.2146, simple_loss=0.3078, pruned_loss=0.06066, over 8290.00 frames. ], tot_loss[loss=0.2396, simple_loss=0.3136, pruned_loss=0.08286, over 1615553.46 frames. ], batch size: 23, lr: 7.58e-03, grad_scale: 8.0 +2023-02-06 09:15:31,441 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 09:15:43,950 INFO [train.py:935] (3/4) Epoch 10, validation: loss=0.1914, simple_loss=0.2907, pruned_loss=0.04604, over 944034.00 frames. +2023-02-06 09:15:43,951 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 09:16:18,415 INFO [train.py:901] (3/4) Epoch 10, batch 6050, loss[loss=0.1765, simple_loss=0.2605, pruned_loss=0.0462, over 7925.00 frames. ], tot_loss[loss=0.2406, simple_loss=0.3142, pruned_loss=0.08345, over 1613989.02 frames. ], batch size: 20, lr: 7.58e-03, grad_scale: 8.0 +2023-02-06 09:16:35,968 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78822.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:16:52,866 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 2.842e+02 3.348e+02 4.641e+02 9.072e+02, threshold=6.696e+02, percent-clipped=15.0 +2023-02-06 09:16:54,176 INFO [train.py:901] (3/4) Epoch 10, batch 6100, loss[loss=0.2606, simple_loss=0.338, pruned_loss=0.09159, over 8120.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.3149, pruned_loss=0.0834, over 1618608.48 frames. ], batch size: 22, lr: 7.57e-03, grad_scale: 8.0 +2023-02-06 09:17:16,950 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 09:17:24,368 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 09:17:27,721 INFO [train.py:901] (3/4) Epoch 10, batch 6150, loss[loss=0.181, simple_loss=0.2641, pruned_loss=0.04892, over 8254.00 frames. ], tot_loss[loss=0.241, simple_loss=0.3149, pruned_loss=0.08354, over 1619933.04 frames. ], batch size: 22, lr: 7.57e-03, grad_scale: 8.0 +2023-02-06 09:17:41,309 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78918.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:17:54,629 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78937.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:18:01,033 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.685e+02 2.486e+02 3.076e+02 3.632e+02 7.166e+02, threshold=6.152e+02, percent-clipped=1.0 +2023-02-06 09:18:02,457 INFO [train.py:901] (3/4) Epoch 10, batch 6200, loss[loss=0.1862, simple_loss=0.2557, pruned_loss=0.05838, over 7431.00 frames. ], tot_loss[loss=0.2406, simple_loss=0.3146, pruned_loss=0.08331, over 1619322.62 frames. ], batch size: 17, lr: 7.57e-03, grad_scale: 8.0 +2023-02-06 09:18:15,656 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78965.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:18:28,061 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78983.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:18:32,907 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78990.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:18:38,057 INFO [train.py:901] (3/4) Epoch 10, batch 6250, loss[loss=0.2549, simple_loss=0.3253, pruned_loss=0.09231, over 8109.00 frames. ], tot_loss[loss=0.2411, simple_loss=0.315, pruned_loss=0.08362, over 1613486.75 frames. ], batch size: 23, lr: 7.57e-03, grad_scale: 8.0 +2023-02-06 09:19:01,795 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79033.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:19:10,136 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.776e+02 2.509e+02 3.177e+02 4.128e+02 1.006e+03, threshold=6.355e+02, percent-clipped=7.0 +2023-02-06 09:19:11,551 INFO [train.py:901] (3/4) Epoch 10, batch 6300, loss[loss=0.26, simple_loss=0.3389, pruned_loss=0.09051, over 8338.00 frames. ], tot_loss[loss=0.2408, simple_loss=0.3142, pruned_loss=0.08371, over 1610917.19 frames. ], batch size: 25, lr: 7.56e-03, grad_scale: 8.0 +2023-02-06 09:19:31,078 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.8235, 1.6755, 5.9251, 2.2614, 5.2484, 4.8669, 5.4601, 5.2708], + device='cuda:3'), covar=tensor([0.0413, 0.4186, 0.0306, 0.3056, 0.0938, 0.0741, 0.0415, 0.0505], + device='cuda:3'), in_proj_covar=tensor([0.0468, 0.0566, 0.0568, 0.0518, 0.0599, 0.0503, 0.0499, 0.0567], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 09:19:41,955 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1151, 4.0943, 3.7748, 1.8549, 3.6383, 3.8316, 3.7351, 3.5197], + device='cuda:3'), covar=tensor([0.0916, 0.0674, 0.1079, 0.4875, 0.0920, 0.1029, 0.1315, 0.0816], + device='cuda:3'), in_proj_covar=tensor([0.0451, 0.0354, 0.0367, 0.0470, 0.0361, 0.0357, 0.0366, 0.0315], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 09:19:47,657 INFO [train.py:901] (3/4) Epoch 10, batch 6350, loss[loss=0.3072, simple_loss=0.3627, pruned_loss=0.1259, over 8102.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.315, pruned_loss=0.08413, over 1611521.18 frames. ], batch size: 23, lr: 7.56e-03, grad_scale: 8.0 +2023-02-06 09:19:54,818 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.95 vs. limit=5.0 +2023-02-06 09:20:00,072 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9925, 1.6724, 2.1953, 1.8914, 2.0774, 1.9291, 1.5794, 0.7578], + device='cuda:3'), covar=tensor([0.3814, 0.3412, 0.1167, 0.2142, 0.1536, 0.1871, 0.1557, 0.3433], + device='cuda:3'), in_proj_covar=tensor([0.0865, 0.0842, 0.0701, 0.0815, 0.0911, 0.0772, 0.0685, 0.0742], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 09:20:20,611 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.648e+02 2.998e+02 3.636e+02 4.667e+02 1.201e+03, threshold=7.271e+02, percent-clipped=11.0 +2023-02-06 09:20:21,298 INFO [train.py:901] (3/4) Epoch 10, batch 6400, loss[loss=0.2098, simple_loss=0.2833, pruned_loss=0.06812, over 7415.00 frames. ], tot_loss[loss=0.242, simple_loss=0.3152, pruned_loss=0.08436, over 1614061.40 frames. ], batch size: 17, lr: 7.56e-03, grad_scale: 8.0 +2023-02-06 09:20:32,950 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.8779, 6.0307, 5.2259, 2.5411, 5.0615, 5.6144, 5.4407, 5.1855], + device='cuda:3'), covar=tensor([0.0569, 0.0423, 0.1095, 0.4661, 0.0755, 0.0645, 0.1109, 0.0581], + device='cuda:3'), in_proj_covar=tensor([0.0447, 0.0351, 0.0364, 0.0465, 0.0360, 0.0353, 0.0362, 0.0312], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 09:20:54,147 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79193.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:20:57,417 INFO [train.py:901] (3/4) Epoch 10, batch 6450, loss[loss=0.2219, simple_loss=0.3027, pruned_loss=0.07056, over 8342.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.3139, pruned_loss=0.08372, over 1611603.25 frames. ], batch size: 26, lr: 7.56e-03, grad_scale: 8.0 +2023-02-06 09:21:12,209 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79218.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:21:31,626 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.735e+02 2.577e+02 3.130e+02 4.050e+02 7.383e+02, threshold=6.260e+02, percent-clipped=1.0 +2023-02-06 09:21:32,338 INFO [train.py:901] (3/4) Epoch 10, batch 6500, loss[loss=0.2421, simple_loss=0.3205, pruned_loss=0.08184, over 8508.00 frames. ], tot_loss[loss=0.2401, simple_loss=0.314, pruned_loss=0.08312, over 1614545.54 frames. ], batch size: 26, lr: 7.55e-03, grad_scale: 8.0 +2023-02-06 09:21:59,871 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79289.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:22:06,998 INFO [train.py:901] (3/4) Epoch 10, batch 6550, loss[loss=0.2657, simple_loss=0.3392, pruned_loss=0.09607, over 8337.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.3152, pruned_loss=0.08364, over 1615622.67 frames. ], batch size: 26, lr: 7.55e-03, grad_scale: 8.0 +2023-02-06 09:22:11,158 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79303.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 09:22:18,373 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79314.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:22:27,892 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79327.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:22:36,559 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 09:22:41,238 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.597e+02 2.767e+02 3.312e+02 4.239e+02 1.073e+03, threshold=6.623e+02, percent-clipped=3.0 +2023-02-06 09:22:41,948 INFO [train.py:901] (3/4) Epoch 10, batch 6600, loss[loss=0.2251, simple_loss=0.3, pruned_loss=0.07511, over 8078.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3164, pruned_loss=0.08436, over 1616324.95 frames. ], batch size: 21, lr: 7.55e-03, grad_scale: 8.0 +2023-02-06 09:22:44,149 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79351.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:22:53,889 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 09:23:04,711 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79382.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:23:11,911 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79393.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 09:23:15,196 INFO [train.py:901] (3/4) Epoch 10, batch 6650, loss[loss=0.2344, simple_loss=0.3173, pruned_loss=0.07572, over 8574.00 frames. ], tot_loss[loss=0.241, simple_loss=0.3152, pruned_loss=0.08337, over 1616442.05 frames. ], batch size: 34, lr: 7.55e-03, grad_scale: 8.0 +2023-02-06 09:23:18,285 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 +2023-02-06 09:23:47,677 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79442.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:23:50,865 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.700e+02 2.666e+02 3.220e+02 4.193e+02 8.839e+02, threshold=6.440e+02, percent-clipped=3.0 +2023-02-06 09:23:51,577 INFO [train.py:901] (3/4) Epoch 10, batch 6700, loss[loss=0.2094, simple_loss=0.2745, pruned_loss=0.07215, over 7686.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.3147, pruned_loss=0.08335, over 1612674.17 frames. ], batch size: 18, lr: 7.54e-03, grad_scale: 8.0 +2023-02-06 09:24:24,667 INFO [train.py:901] (3/4) Epoch 10, batch 6750, loss[loss=0.2257, simple_loss=0.3049, pruned_loss=0.07324, over 8083.00 frames. ], tot_loss[loss=0.2411, simple_loss=0.3144, pruned_loss=0.08384, over 1614028.12 frames. ], batch size: 21, lr: 7.54e-03, grad_scale: 8.0 +2023-02-06 09:24:34,604 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9538, 2.1219, 1.6216, 2.6216, 1.0834, 1.4189, 1.7543, 2.0803], + device='cuda:3'), covar=tensor([0.0668, 0.0772, 0.1094, 0.0359, 0.1318, 0.1507, 0.1013, 0.0770], + device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0212, 0.0255, 0.0216, 0.0219, 0.0252, 0.0259, 0.0224], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 09:24:35,571 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 +2023-02-06 09:25:00,368 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.601e+02 2.662e+02 3.188e+02 4.113e+02 8.575e+02, threshold=6.376e+02, percent-clipped=4.0 +2023-02-06 09:25:01,059 INFO [train.py:901] (3/4) Epoch 10, batch 6800, loss[loss=0.1946, simple_loss=0.269, pruned_loss=0.06007, over 8242.00 frames. ], tot_loss[loss=0.2401, simple_loss=0.3136, pruned_loss=0.08331, over 1613503.05 frames. ], batch size: 22, lr: 7.54e-03, grad_scale: 8.0 +2023-02-06 09:25:11,664 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 09:25:13,797 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5251, 4.4781, 4.0663, 1.8746, 3.9859, 4.1472, 4.1831, 3.8277], + device='cuda:3'), covar=tensor([0.0770, 0.0539, 0.0921, 0.5337, 0.0815, 0.1042, 0.1218, 0.0734], + device='cuda:3'), in_proj_covar=tensor([0.0453, 0.0355, 0.0371, 0.0474, 0.0368, 0.0358, 0.0366, 0.0317], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 09:25:35,811 INFO [train.py:901] (3/4) Epoch 10, batch 6850, loss[loss=0.2854, simple_loss=0.3483, pruned_loss=0.1112, over 8626.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.3146, pruned_loss=0.08347, over 1617290.15 frames. ], batch size: 49, lr: 7.54e-03, grad_scale: 8.0 +2023-02-06 09:25:39,361 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79603.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:25:48,391 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5549, 1.9603, 3.5254, 1.1972, 2.4195, 2.0737, 1.6275, 2.1386], + device='cuda:3'), covar=tensor([0.1672, 0.2045, 0.0675, 0.3817, 0.1554, 0.2556, 0.1688, 0.2261], + device='cuda:3'), in_proj_covar=tensor([0.0481, 0.0504, 0.0526, 0.0569, 0.0613, 0.0548, 0.0461, 0.0603], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 09:25:59,641 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 09:26:10,489 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.544e+02 2.480e+02 2.958e+02 3.519e+02 6.592e+02, threshold=5.916e+02, percent-clipped=1.0 +2023-02-06 09:26:10,589 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79647.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 09:26:11,119 INFO [train.py:901] (3/4) Epoch 10, batch 6900, loss[loss=0.261, simple_loss=0.3168, pruned_loss=0.1026, over 7920.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.3167, pruned_loss=0.08485, over 1622498.91 frames. ], batch size: 20, lr: 7.53e-03, grad_scale: 8.0 +2023-02-06 09:26:30,984 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79675.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:26:44,404 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79695.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:26:46,400 INFO [train.py:901] (3/4) Epoch 10, batch 6950, loss[loss=0.2537, simple_loss=0.3254, pruned_loss=0.09095, over 8101.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3161, pruned_loss=0.08464, over 1621646.21 frames. ], batch size: 23, lr: 7.53e-03, grad_scale: 8.0 +2023-02-06 09:26:46,619 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79698.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:27:03,561 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79723.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:27:05,486 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79726.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:27:10,652 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 09:27:12,700 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79737.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 09:27:19,257 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.682e+02 2.770e+02 3.379e+02 4.019e+02 1.115e+03, threshold=6.759e+02, percent-clipped=8.0 +2023-02-06 09:27:19,981 INFO [train.py:901] (3/4) Epoch 10, batch 7000, loss[loss=0.2291, simple_loss=0.316, pruned_loss=0.07108, over 8640.00 frames. ], tot_loss[loss=0.2424, simple_loss=0.316, pruned_loss=0.08443, over 1622557.69 frames. ], batch size: 39, lr: 7.53e-03, grad_scale: 8.0 +2023-02-06 09:27:30,363 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79762.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 09:27:39,405 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.0827, 3.9727, 3.6987, 1.8993, 3.6664, 3.8004, 3.7657, 3.4459], + device='cuda:3'), covar=tensor([0.0924, 0.0711, 0.1072, 0.4937, 0.0857, 0.0928, 0.1492, 0.0838], + device='cuda:3'), in_proj_covar=tensor([0.0457, 0.0357, 0.0373, 0.0471, 0.0367, 0.0356, 0.0363, 0.0316], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 09:27:43,455 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79780.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:27:55,335 INFO [train.py:901] (3/4) Epoch 10, batch 7050, loss[loss=0.224, simple_loss=0.2963, pruned_loss=0.07587, over 8232.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.3156, pruned_loss=0.0839, over 1623112.98 frames. ], batch size: 22, lr: 7.53e-03, grad_scale: 8.0 +2023-02-06 09:28:04,460 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6255, 1.3599, 1.6620, 1.2551, 0.8866, 1.4158, 1.4067, 1.4046], + device='cuda:3'), covar=tensor([0.0538, 0.1247, 0.1642, 0.1402, 0.0617, 0.1498, 0.0714, 0.0613], + device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0152, 0.0193, 0.0159, 0.0105, 0.0164, 0.0118, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0006, 0.0006], + device='cuda:3') +2023-02-06 09:28:04,467 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79810.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:28:11,471 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-06 09:28:25,536 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79841.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:28:29,357 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.530e+02 2.704e+02 3.361e+02 4.306e+02 1.362e+03, threshold=6.722e+02, percent-clipped=5.0 +2023-02-06 09:28:30,073 INFO [train.py:901] (3/4) Epoch 10, batch 7100, loss[loss=0.1777, simple_loss=0.2549, pruned_loss=0.05028, over 7427.00 frames. ], tot_loss[loss=0.2411, simple_loss=0.3148, pruned_loss=0.08371, over 1617922.62 frames. ], batch size: 17, lr: 7.53e-03, grad_scale: 8.0 +2023-02-06 09:28:32,888 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79852.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 09:29:06,027 INFO [train.py:901] (3/4) Epoch 10, batch 7150, loss[loss=0.2949, simple_loss=0.3698, pruned_loss=0.1101, over 8541.00 frames. ], tot_loss[loss=0.2398, simple_loss=0.3136, pruned_loss=0.08297, over 1616942.04 frames. ], batch size: 49, lr: 7.52e-03, grad_scale: 8.0 +2023-02-06 09:29:39,481 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.706e+02 2.702e+02 3.262e+02 4.332e+02 1.613e+03, threshold=6.525e+02, percent-clipped=3.0 +2023-02-06 09:29:39,577 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79947.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:29:40,186 INFO [train.py:901] (3/4) Epoch 10, batch 7200, loss[loss=0.2505, simple_loss=0.3039, pruned_loss=0.09858, over 7534.00 frames. ], tot_loss[loss=0.239, simple_loss=0.3133, pruned_loss=0.08232, over 1619910.39 frames. ], batch size: 18, lr: 7.52e-03, grad_scale: 8.0 +2023-02-06 09:30:06,714 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 +2023-02-06 09:30:13,856 INFO [train.py:901] (3/4) Epoch 10, batch 7250, loss[loss=0.2428, simple_loss=0.317, pruned_loss=0.08433, over 8292.00 frames. ], tot_loss[loss=0.239, simple_loss=0.3133, pruned_loss=0.08232, over 1618802.37 frames. ], batch size: 23, lr: 7.52e-03, grad_scale: 8.0 +2023-02-06 09:30:30,524 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80018.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 09:30:31,018 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80019.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:30:47,885 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80043.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 09:30:50,288 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.474e+02 2.755e+02 3.243e+02 3.993e+02 1.489e+03, threshold=6.485e+02, percent-clipped=9.0 +2023-02-06 09:30:50,942 INFO [train.py:901] (3/4) Epoch 10, batch 7300, loss[loss=0.2269, simple_loss=0.2978, pruned_loss=0.07801, over 7694.00 frames. ], tot_loss[loss=0.2406, simple_loss=0.3147, pruned_loss=0.08323, over 1617509.15 frames. ], batch size: 18, lr: 7.52e-03, grad_scale: 8.0 +2023-02-06 09:31:00,591 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80062.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:31:03,354 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80066.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:31:20,005 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80091.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:31:24,179 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80097.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:31:24,595 INFO [train.py:901] (3/4) Epoch 10, batch 7350, loss[loss=0.2319, simple_loss=0.3075, pruned_loss=0.0782, over 8109.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.3144, pruned_loss=0.08311, over 1615758.41 frames. ], batch size: 23, lr: 7.51e-03, grad_scale: 8.0 +2023-02-06 09:31:31,673 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80108.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 09:31:32,533 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-06 09:31:42,400 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80122.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:31:44,300 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80124.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:31:50,390 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80133.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 09:31:50,992 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80134.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:31:52,503 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 09:31:56,728 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 09:31:59,432 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.599e+02 2.519e+02 3.343e+02 4.224e+02 9.659e+02, threshold=6.686e+02, percent-clipped=6.0 +2023-02-06 09:32:00,139 INFO [train.py:901] (3/4) Epoch 10, batch 7400, loss[loss=0.266, simple_loss=0.3294, pruned_loss=0.1013, over 6380.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3154, pruned_loss=0.08375, over 1612460.84 frames. ], batch size: 14, lr: 7.51e-03, grad_scale: 8.0 +2023-02-06 09:32:16,188 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 09:32:29,723 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5231, 1.5811, 4.7177, 2.0032, 4.1198, 3.9323, 4.3101, 4.1576], + device='cuda:3'), covar=tensor([0.0498, 0.3921, 0.0404, 0.2981, 0.1012, 0.0770, 0.0466, 0.0561], + device='cuda:3'), in_proj_covar=tensor([0.0462, 0.0559, 0.0562, 0.0511, 0.0584, 0.0493, 0.0490, 0.0556], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 09:32:34,297 INFO [train.py:901] (3/4) Epoch 10, batch 7450, loss[loss=0.2043, simple_loss=0.2771, pruned_loss=0.06576, over 7207.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.316, pruned_loss=0.08418, over 1613483.01 frames. ], batch size: 16, lr: 7.51e-03, grad_scale: 8.0 +2023-02-06 09:32:54,347 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 09:33:02,512 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80239.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:33:09,145 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.495e+02 2.637e+02 3.217e+02 3.901e+02 6.824e+02, threshold=6.433e+02, percent-clipped=2.0 +2023-02-06 09:33:09,864 INFO [train.py:901] (3/4) Epoch 10, batch 7500, loss[loss=0.1906, simple_loss=0.2682, pruned_loss=0.05656, over 7427.00 frames. ], tot_loss[loss=0.2406, simple_loss=0.3146, pruned_loss=0.0833, over 1615449.83 frames. ], batch size: 17, lr: 7.51e-03, grad_scale: 8.0 +2023-02-06 09:33:43,942 INFO [train.py:901] (3/4) Epoch 10, batch 7550, loss[loss=0.2583, simple_loss=0.3269, pruned_loss=0.09488, over 8133.00 frames. ], tot_loss[loss=0.2398, simple_loss=0.3141, pruned_loss=0.0828, over 1615152.80 frames. ], batch size: 22, lr: 7.50e-03, grad_scale: 8.0 +2023-02-06 09:33:46,247 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80301.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:33:57,962 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80318.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:34:15,056 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80343.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:34:17,459 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.537e+02 2.399e+02 2.908e+02 3.933e+02 1.078e+03, threshold=5.816e+02, percent-clipped=3.0 +2023-02-06 09:34:18,143 INFO [train.py:901] (3/4) Epoch 10, batch 7600, loss[loss=0.22, simple_loss=0.303, pruned_loss=0.06847, over 8081.00 frames. ], tot_loss[loss=0.2395, simple_loss=0.3135, pruned_loss=0.08269, over 1613173.15 frames. ], batch size: 21, lr: 7.50e-03, grad_scale: 8.0 +2023-02-06 09:34:25,955 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-06 09:34:49,203 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80390.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:34:54,344 INFO [train.py:901] (3/4) Epoch 10, batch 7650, loss[loss=0.2389, simple_loss=0.3257, pruned_loss=0.07609, over 8618.00 frames. ], tot_loss[loss=0.2395, simple_loss=0.3139, pruned_loss=0.08258, over 1616075.91 frames. ], batch size: 34, lr: 7.50e-03, grad_scale: 8.0 +2023-02-06 09:35:03,333 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80411.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:35:03,594 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.58 vs. limit=5.0 +2023-02-06 09:35:06,123 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80415.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:35:27,280 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.658e+02 2.621e+02 3.149e+02 3.913e+02 9.838e+02, threshold=6.298e+02, percent-clipped=6.0 +2023-02-06 09:35:27,987 INFO [train.py:901] (3/4) Epoch 10, batch 7700, loss[loss=0.2406, simple_loss=0.3168, pruned_loss=0.08217, over 8575.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.3138, pruned_loss=0.08231, over 1620990.37 frames. ], batch size: 49, lr: 7.50e-03, grad_scale: 8.0 +2023-02-06 09:35:51,541 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80481.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:35:52,936 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0093, 2.3667, 1.8152, 2.8703, 1.3238, 1.6092, 2.1483, 2.5088], + device='cuda:3'), covar=tensor([0.0816, 0.0896, 0.1142, 0.0471, 0.1137, 0.1471, 0.0927, 0.0751], + device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0215, 0.0256, 0.0219, 0.0219, 0.0254, 0.0257, 0.0228], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 09:36:01,534 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80495.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:36:03,351 INFO [train.py:901] (3/4) Epoch 10, batch 7750, loss[loss=0.261, simple_loss=0.3342, pruned_loss=0.09389, over 8746.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3145, pruned_loss=0.08265, over 1623546.82 frames. ], batch size: 39, lr: 7.49e-03, grad_scale: 8.0 +2023-02-06 09:36:06,761 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 09:36:13,279 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80512.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:36:18,618 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80520.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:36:32,807 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2023-02-06 09:36:36,357 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.491e+02 2.713e+02 3.406e+02 4.090e+02 8.759e+02, threshold=6.812e+02, percent-clipped=3.0 +2023-02-06 09:36:37,066 INFO [train.py:901] (3/4) Epoch 10, batch 7800, loss[loss=0.2379, simple_loss=0.3, pruned_loss=0.08789, over 7524.00 frames. ], tot_loss[loss=0.2397, simple_loss=0.3137, pruned_loss=0.0828, over 1620645.37 frames. ], batch size: 18, lr: 7.49e-03, grad_scale: 8.0 +2023-02-06 09:37:09,869 INFO [train.py:901] (3/4) Epoch 10, batch 7850, loss[loss=0.2356, simple_loss=0.319, pruned_loss=0.07607, over 8632.00 frames. ], tot_loss[loss=0.2404, simple_loss=0.3143, pruned_loss=0.08319, over 1619221.71 frames. ], batch size: 34, lr: 7.49e-03, grad_scale: 8.0 +2023-02-06 09:37:41,490 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80645.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:37:42,730 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.605e+02 2.556e+02 3.372e+02 4.255e+02 7.191e+02, threshold=6.744e+02, percent-clipped=1.0 +2023-02-06 09:37:43,428 INFO [train.py:901] (3/4) Epoch 10, batch 7900, loss[loss=0.2666, simple_loss=0.3262, pruned_loss=0.1035, over 6799.00 frames. ], tot_loss[loss=0.2394, simple_loss=0.3135, pruned_loss=0.08267, over 1618880.98 frames. ], batch size: 72, lr: 7.49e-03, grad_scale: 8.0 +2023-02-06 09:38:16,939 INFO [train.py:901] (3/4) Epoch 10, batch 7950, loss[loss=0.1894, simple_loss=0.2666, pruned_loss=0.0561, over 7812.00 frames. ], tot_loss[loss=0.2387, simple_loss=0.3131, pruned_loss=0.08215, over 1617566.85 frames. ], batch size: 20, lr: 7.49e-03, grad_scale: 8.0 +2023-02-06 09:38:26,479 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6273, 2.9033, 2.1997, 2.3204, 2.3881, 1.7411, 2.3686, 2.3159], + device='cuda:3'), covar=tensor([0.1198, 0.0261, 0.0766, 0.0535, 0.0561, 0.1180, 0.0757, 0.0727], + device='cuda:3'), in_proj_covar=tensor([0.0341, 0.0229, 0.0305, 0.0291, 0.0300, 0.0319, 0.0336, 0.0301], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 09:38:46,817 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6267, 1.3291, 4.8127, 1.6636, 4.2036, 3.9934, 4.3395, 4.1875], + device='cuda:3'), covar=tensor([0.0466, 0.4446, 0.0473, 0.3400, 0.1134, 0.0823, 0.0449, 0.0577], + device='cuda:3'), in_proj_covar=tensor([0.0468, 0.0560, 0.0566, 0.0514, 0.0591, 0.0494, 0.0493, 0.0561], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 09:38:50,733 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.443e+02 2.660e+02 3.023e+02 3.700e+02 9.606e+02, threshold=6.046e+02, percent-clipped=2.0 +2023-02-06 09:38:51,443 INFO [train.py:901] (3/4) Epoch 10, batch 8000, loss[loss=0.2476, simple_loss=0.3326, pruned_loss=0.08126, over 8513.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.313, pruned_loss=0.08204, over 1614850.59 frames. ], batch size: 26, lr: 7.48e-03, grad_scale: 8.0 +2023-02-06 09:38:55,688 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80754.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:38:56,306 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80755.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:38:59,816 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80760.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:39:20,375 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80791.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:39:24,989 INFO [train.py:901] (3/4) Epoch 10, batch 8050, loss[loss=0.2111, simple_loss=0.2742, pruned_loss=0.07403, over 7191.00 frames. ], tot_loss[loss=0.239, simple_loss=0.3129, pruned_loss=0.0825, over 1609155.26 frames. ], batch size: 16, lr: 7.48e-03, grad_scale: 8.0 +2023-02-06 09:39:43,570 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80825.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:39:58,215 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 09:40:01,797 INFO [train.py:901] (3/4) Epoch 11, batch 0, loss[loss=0.225, simple_loss=0.3047, pruned_loss=0.07262, over 8187.00 frames. ], tot_loss[loss=0.225, simple_loss=0.3047, pruned_loss=0.07262, over 8187.00 frames. ], batch size: 23, lr: 7.14e-03, grad_scale: 8.0 +2023-02-06 09:40:01,798 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 09:40:13,089 INFO [train.py:935] (3/4) Epoch 11, validation: loss=0.1907, simple_loss=0.2907, pruned_loss=0.04534, over 944034.00 frames. +2023-02-06 09:40:13,090 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 09:40:22,706 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6534, 1.3598, 1.5494, 1.2320, 0.8954, 1.2778, 1.4888, 1.2497], + device='cuda:3'), covar=tensor([0.0567, 0.1317, 0.1830, 0.1515, 0.0631, 0.1656, 0.0750, 0.0696], + device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0155, 0.0197, 0.0161, 0.0107, 0.0166, 0.0119, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:3') +2023-02-06 09:40:23,924 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.799e+02 2.899e+02 3.439e+02 4.416e+02 1.589e+03, threshold=6.879e+02, percent-clipped=9.0 +2023-02-06 09:40:27,456 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 09:40:30,153 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80856.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:40:39,880 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80870.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:40:47,941 INFO [train.py:901] (3/4) Epoch 11, batch 50, loss[loss=0.2576, simple_loss=0.3266, pruned_loss=0.09428, over 8104.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3227, pruned_loss=0.08785, over 365636.04 frames. ], batch size: 23, lr: 7.14e-03, grad_scale: 8.0 +2023-02-06 09:40:52,491 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 +2023-02-06 09:41:03,915 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 09:41:04,376 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2023-02-06 09:41:24,348 INFO [train.py:901] (3/4) Epoch 11, batch 100, loss[loss=0.3025, simple_loss=0.3559, pruned_loss=0.1246, over 6834.00 frames. ], tot_loss[loss=0.244, simple_loss=0.3164, pruned_loss=0.08583, over 639730.86 frames. ], batch size: 71, lr: 7.14e-03, grad_scale: 8.0 +2023-02-06 09:41:29,236 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 09:41:30,717 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80940.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:41:35,311 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.925e+02 2.679e+02 3.187e+02 3.933e+02 1.063e+03, threshold=6.374e+02, percent-clipped=2.0 +2023-02-06 09:41:51,853 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80971.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:41:58,391 INFO [train.py:901] (3/4) Epoch 11, batch 150, loss[loss=0.2523, simple_loss=0.3155, pruned_loss=0.09455, over 7781.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.3163, pruned_loss=0.08502, over 857947.16 frames. ], batch size: 19, lr: 7.13e-03, grad_scale: 8.0 +2023-02-06 09:42:23,722 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81016.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:42:34,641 INFO [train.py:901] (3/4) Epoch 11, batch 200, loss[loss=0.2, simple_loss=0.2715, pruned_loss=0.06421, over 7234.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.3152, pruned_loss=0.08404, over 1029717.70 frames. ], batch size: 16, lr: 7.13e-03, grad_scale: 8.0 +2023-02-06 09:42:37,864 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 +2023-02-06 09:42:42,984 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81041.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:42:47,008 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.935e+02 2.662e+02 3.186e+02 4.005e+02 8.686e+02, threshold=6.371e+02, percent-clipped=5.0 +2023-02-06 09:43:10,553 INFO [train.py:901] (3/4) Epoch 11, batch 250, loss[loss=0.2235, simple_loss=0.3097, pruned_loss=0.06862, over 8496.00 frames. ], tot_loss[loss=0.2387, simple_loss=0.3125, pruned_loss=0.08242, over 1155030.50 frames. ], batch size: 28, lr: 7.13e-03, grad_scale: 8.0 +2023-02-06 09:43:21,546 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 09:43:22,306 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81098.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:43:31,350 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 09:43:42,708 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81126.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:43:46,010 INFO [train.py:901] (3/4) Epoch 11, batch 300, loss[loss=0.2591, simple_loss=0.3385, pruned_loss=0.08983, over 8337.00 frames. ], tot_loss[loss=0.2379, simple_loss=0.3114, pruned_loss=0.08218, over 1252914.29 frames. ], batch size: 26, lr: 7.13e-03, grad_scale: 16.0 +2023-02-06 09:43:48,256 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81134.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:43:48,874 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81135.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:43:57,132 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.630e+02 2.697e+02 3.136e+02 4.054e+02 9.565e+02, threshold=6.271e+02, percent-clipped=1.0 +2023-02-06 09:44:00,772 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81151.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:44:22,513 INFO [train.py:901] (3/4) Epoch 11, batch 350, loss[loss=0.2383, simple_loss=0.3092, pruned_loss=0.08364, over 8297.00 frames. ], tot_loss[loss=0.2369, simple_loss=0.3108, pruned_loss=0.08149, over 1333622.79 frames. ], batch size: 23, lr: 7.13e-03, grad_scale: 8.0 +2023-02-06 09:44:32,495 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2023-02-06 09:44:33,017 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81196.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:44:39,230 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6449, 2.1736, 3.4061, 2.5706, 2.9781, 2.3903, 1.9563, 1.7628], + device='cuda:3'), covar=tensor([0.3868, 0.4305, 0.1191, 0.2749, 0.2100, 0.2078, 0.1671, 0.4466], + device='cuda:3'), in_proj_covar=tensor([0.0892, 0.0863, 0.0720, 0.0833, 0.0935, 0.0794, 0.0704, 0.0762], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 09:44:44,354 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81213.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:44:46,358 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8999, 1.5767, 1.7299, 1.4951, 1.0280, 1.5015, 1.7516, 1.4664], + device='cuda:3'), covar=tensor([0.0483, 0.1079, 0.1544, 0.1210, 0.0539, 0.1306, 0.0647, 0.0566], + device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0155, 0.0195, 0.0159, 0.0106, 0.0164, 0.0118, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:3') +2023-02-06 09:44:49,728 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81221.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:44:53,754 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81227.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:44:56,288 INFO [train.py:901] (3/4) Epoch 11, batch 400, loss[loss=0.1798, simple_loss=0.2576, pruned_loss=0.05099, over 7433.00 frames. ], tot_loss[loss=0.2372, simple_loss=0.3109, pruned_loss=0.08169, over 1390377.65 frames. ], batch size: 17, lr: 7.12e-03, grad_scale: 8.0 +2023-02-06 09:45:00,800 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2023-02-06 09:45:08,650 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.792e+02 2.601e+02 3.216e+02 4.274e+02 6.931e+02, threshold=6.433e+02, percent-clipped=2.0 +2023-02-06 09:45:10,286 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81250.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:45:11,745 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81252.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:45:23,639 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0302, 1.2231, 1.1934, 0.4990, 1.1566, 1.0142, 0.0877, 1.1693], + device='cuda:3'), covar=tensor([0.0259, 0.0225, 0.0194, 0.0375, 0.0226, 0.0685, 0.0512, 0.0208], + device='cuda:3'), in_proj_covar=tensor([0.0373, 0.0304, 0.0259, 0.0372, 0.0293, 0.0457, 0.0345, 0.0339], + device='cuda:3'), out_proj_covar=tensor([1.0753e-04, 8.5703e-05, 7.3022e-05, 1.0561e-04, 8.3944e-05, 1.4109e-04, + 9.9503e-05, 9.6983e-05], device='cuda:3') +2023-02-06 09:45:29,082 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4654, 2.6907, 2.0021, 2.1983, 2.2158, 1.6664, 2.1498, 2.1968], + device='cuda:3'), covar=tensor([0.1366, 0.0355, 0.0879, 0.0532, 0.0549, 0.1249, 0.0831, 0.0782], + device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0233, 0.0308, 0.0296, 0.0303, 0.0325, 0.0338, 0.0304], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 09:45:32,912 INFO [train.py:901] (3/4) Epoch 11, batch 450, loss[loss=0.2424, simple_loss=0.3194, pruned_loss=0.08265, over 8357.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3106, pruned_loss=0.08139, over 1438840.23 frames. ], batch size: 24, lr: 7.12e-03, grad_scale: 8.0 +2023-02-06 09:45:57,085 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81317.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:45:58,455 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81319.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:46:06,255 INFO [train.py:901] (3/4) Epoch 11, batch 500, loss[loss=0.2014, simple_loss=0.2876, pruned_loss=0.05762, over 8287.00 frames. ], tot_loss[loss=0.2369, simple_loss=0.3114, pruned_loss=0.08118, over 1483854.84 frames. ], batch size: 23, lr: 7.12e-03, grad_scale: 8.0 +2023-02-06 09:46:07,904 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.91 vs. limit=5.0 +2023-02-06 09:46:17,547 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.597e+02 2.501e+02 3.364e+02 4.069e+02 6.845e+02, threshold=6.728e+02, percent-clipped=2.0 +2023-02-06 09:46:40,092 INFO [train.py:901] (3/4) Epoch 11, batch 550, loss[loss=0.2561, simple_loss=0.3327, pruned_loss=0.08972, over 8101.00 frames. ], tot_loss[loss=0.2383, simple_loss=0.3123, pruned_loss=0.08212, over 1512111.99 frames. ], batch size: 23, lr: 7.12e-03, grad_scale: 8.0 +2023-02-06 09:47:04,542 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3174, 2.0021, 2.9504, 2.3722, 2.7488, 2.1589, 1.8077, 1.5979], + device='cuda:3'), covar=tensor([0.3705, 0.3710, 0.1076, 0.2199, 0.1676, 0.2161, 0.1696, 0.3643], + device='cuda:3'), in_proj_covar=tensor([0.0886, 0.0853, 0.0717, 0.0828, 0.0927, 0.0789, 0.0701, 0.0759], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 09:47:15,699 INFO [train.py:901] (3/4) Epoch 11, batch 600, loss[loss=0.252, simple_loss=0.3368, pruned_loss=0.08353, over 8701.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3138, pruned_loss=0.08296, over 1538012.94 frames. ], batch size: 30, lr: 7.11e-03, grad_scale: 8.0 +2023-02-06 09:47:27,354 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.440e+02 2.633e+02 3.080e+02 3.885e+02 6.931e+02, threshold=6.160e+02, percent-clipped=1.0 +2023-02-06 09:47:27,538 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0654, 1.4802, 3.4083, 1.5396, 2.0991, 3.7727, 3.7573, 3.1910], + device='cuda:3'), covar=tensor([0.0948, 0.1403, 0.0308, 0.1707, 0.0970, 0.0197, 0.0434, 0.0530], + device='cuda:3'), in_proj_covar=tensor([0.0257, 0.0286, 0.0247, 0.0278, 0.0264, 0.0226, 0.0322, 0.0280], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2023-02-06 09:47:27,561 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1910, 1.4819, 1.5399, 1.3157, 1.1298, 1.3378, 1.7501, 1.5815], + device='cuda:3'), covar=tensor([0.0472, 0.1232, 0.1755, 0.1418, 0.0614, 0.1524, 0.0710, 0.0588], + device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0155, 0.0196, 0.0159, 0.0106, 0.0165, 0.0118, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:3') +2023-02-06 09:47:27,976 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 09:47:35,462 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 09:47:42,472 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81469.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:47:48,472 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81478.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:47:50,439 INFO [train.py:901] (3/4) Epoch 11, batch 650, loss[loss=0.2011, simple_loss=0.2827, pruned_loss=0.05973, over 7659.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.3126, pruned_loss=0.08233, over 1553284.53 frames. ], batch size: 19, lr: 7.11e-03, grad_scale: 8.0 +2023-02-06 09:47:59,450 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81494.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:48:05,537 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1027, 2.1790, 1.6680, 1.9632, 1.7408, 1.3274, 1.6584, 1.6490], + device='cuda:3'), covar=tensor([0.1208, 0.0345, 0.1004, 0.0470, 0.0559, 0.1322, 0.0878, 0.0749], + device='cuda:3'), in_proj_covar=tensor([0.0340, 0.0228, 0.0304, 0.0295, 0.0297, 0.0318, 0.0332, 0.0300], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 09:48:08,335 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81506.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:48:26,931 INFO [train.py:901] (3/4) Epoch 11, batch 700, loss[loss=0.1942, simple_loss=0.2867, pruned_loss=0.05085, over 7974.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3117, pruned_loss=0.08087, over 1567765.15 frames. ], batch size: 21, lr: 7.11e-03, grad_scale: 8.0 +2023-02-06 09:48:27,091 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81531.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:48:28,479 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6499, 2.1947, 4.4205, 1.2362, 3.1392, 2.2144, 1.8686, 2.6481], + device='cuda:3'), covar=tensor([0.2013, 0.2590, 0.0877, 0.4747, 0.1882, 0.3189, 0.1973, 0.3174], + device='cuda:3'), in_proj_covar=tensor([0.0480, 0.0504, 0.0526, 0.0567, 0.0609, 0.0541, 0.0466, 0.0608], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 09:48:38,763 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.613e+02 2.541e+02 3.049e+02 3.626e+02 6.264e+02, threshold=6.097e+02, percent-clipped=1.0 +2023-02-06 09:49:01,392 INFO [train.py:901] (3/4) Epoch 11, batch 750, loss[loss=0.2964, simple_loss=0.3665, pruned_loss=0.1131, over 8487.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3112, pruned_loss=0.08095, over 1579232.54 frames. ], batch size: 25, lr: 7.11e-03, grad_scale: 8.0 +2023-02-06 09:49:07,835 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 +2023-02-06 09:49:10,214 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81593.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:49:10,588 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-06 09:49:21,647 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1770, 1.8451, 2.8891, 2.2043, 2.5301, 2.0575, 1.5646, 1.2575], + device='cuda:3'), covar=tensor([0.4059, 0.4036, 0.1005, 0.2583, 0.1972, 0.2185, 0.1846, 0.3929], + device='cuda:3'), in_proj_covar=tensor([0.0885, 0.0855, 0.0718, 0.0833, 0.0931, 0.0790, 0.0701, 0.0761], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 09:49:24,851 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 09:49:25,693 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0279, 2.4530, 1.9142, 2.9684, 1.3005, 1.6386, 2.0013, 2.3925], + device='cuda:3'), covar=tensor([0.0814, 0.0758, 0.1068, 0.0396, 0.1296, 0.1505, 0.1036, 0.0870], + device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0212, 0.0254, 0.0216, 0.0216, 0.0254, 0.0257, 0.0226], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 09:49:29,029 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.7910, 6.0302, 5.0784, 2.4316, 5.2542, 5.5888, 5.6593, 5.2912], + device='cuda:3'), covar=tensor([0.0583, 0.0323, 0.0863, 0.4476, 0.0630, 0.0581, 0.0833, 0.0492], + device='cuda:3'), in_proj_covar=tensor([0.0465, 0.0368, 0.0379, 0.0479, 0.0373, 0.0365, 0.0367, 0.0322], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 09:49:33,727 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 09:49:36,455 INFO [train.py:901] (3/4) Epoch 11, batch 800, loss[loss=0.2155, simple_loss=0.2924, pruned_loss=0.06937, over 7939.00 frames. ], tot_loss[loss=0.2368, simple_loss=0.3111, pruned_loss=0.08119, over 1583001.09 frames. ], batch size: 20, lr: 7.11e-03, grad_scale: 8.0 +2023-02-06 09:49:49,270 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.638e+02 2.494e+02 2.971e+02 3.970e+02 9.403e+02, threshold=5.941e+02, percent-clipped=2.0 +2023-02-06 09:49:58,275 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81661.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:49:59,657 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81663.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:50:11,545 INFO [train.py:901] (3/4) Epoch 11, batch 850, loss[loss=0.2052, simple_loss=0.2888, pruned_loss=0.06083, over 8384.00 frames. ], tot_loss[loss=0.2375, simple_loss=0.3118, pruned_loss=0.0816, over 1594676.96 frames. ], batch size: 49, lr: 7.10e-03, grad_scale: 8.0 +2023-02-06 09:50:16,502 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81688.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:50:46,069 INFO [train.py:901] (3/4) Epoch 11, batch 900, loss[loss=0.2129, simple_loss=0.2903, pruned_loss=0.06771, over 7809.00 frames. ], tot_loss[loss=0.2378, simple_loss=0.3125, pruned_loss=0.08156, over 1599236.56 frames. ], batch size: 20, lr: 7.10e-03, grad_scale: 8.0 +2023-02-06 09:50:58,840 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.318e+02 2.543e+02 3.289e+02 4.286e+02 9.063e+02, threshold=6.577e+02, percent-clipped=7.0 +2023-02-06 09:51:08,746 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2108, 1.2419, 1.4928, 1.1132, 0.7880, 1.2958, 1.1356, 0.9238], + device='cuda:3'), covar=tensor([0.0608, 0.1330, 0.1772, 0.1581, 0.0616, 0.1617, 0.0769, 0.0734], + device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0154, 0.0195, 0.0159, 0.0106, 0.0165, 0.0118, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:3') +2023-02-06 09:51:18,686 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81776.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:51:20,088 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81778.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:51:21,842 INFO [train.py:901] (3/4) Epoch 11, batch 950, loss[loss=0.2431, simple_loss=0.3187, pruned_loss=0.08373, over 8470.00 frames. ], tot_loss[loss=0.2374, simple_loss=0.3125, pruned_loss=0.0811, over 1604467.00 frames. ], batch size: 49, lr: 7.10e-03, grad_scale: 8.0 +2023-02-06 09:51:37,578 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3897, 1.4915, 1.6818, 1.2885, 0.8943, 1.8169, 0.1272, 1.0957], + device='cuda:3'), covar=tensor([0.3137, 0.1797, 0.0580, 0.1897, 0.4455, 0.0490, 0.3237, 0.1825], + device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0164, 0.0094, 0.0210, 0.0253, 0.0102, 0.0162, 0.0159], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 09:51:51,848 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 09:51:56,006 INFO [train.py:901] (3/4) Epoch 11, batch 1000, loss[loss=0.2455, simple_loss=0.3206, pruned_loss=0.08517, over 8424.00 frames. ], tot_loss[loss=0.237, simple_loss=0.3124, pruned_loss=0.08076, over 1610248.77 frames. ], batch size: 49, lr: 7.10e-03, grad_scale: 8.0 +2023-02-06 09:52:06,324 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5837, 1.9478, 3.1758, 1.3985, 2.2479, 1.9928, 1.6575, 2.0878], + device='cuda:3'), covar=tensor([0.1568, 0.2063, 0.0599, 0.3821, 0.1498, 0.2640, 0.1705, 0.2076], + device='cuda:3'), in_proj_covar=tensor([0.0485, 0.0515, 0.0537, 0.0582, 0.0618, 0.0552, 0.0474, 0.0614], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 09:52:07,432 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.790e+02 2.713e+02 3.211e+02 4.023e+02 7.481e+02, threshold=6.422e+02, percent-clipped=3.0 +2023-02-06 09:52:08,395 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81849.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:52:13,878 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-06 09:52:22,679 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-06 09:52:27,183 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 09:52:27,397 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81874.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:52:31,947 INFO [train.py:901] (3/4) Epoch 11, batch 1050, loss[loss=0.319, simple_loss=0.3738, pruned_loss=0.1321, over 8693.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3116, pruned_loss=0.08077, over 1611270.98 frames. ], batch size: 34, lr: 7.09e-03, grad_scale: 8.0 +2023-02-06 09:52:39,042 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 09:53:06,135 INFO [train.py:901] (3/4) Epoch 11, batch 1100, loss[loss=0.3085, simple_loss=0.3606, pruned_loss=0.1282, over 8244.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3117, pruned_loss=0.08085, over 1613527.67 frames. ], batch size: 24, lr: 7.09e-03, grad_scale: 8.0 +2023-02-06 09:53:07,788 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3908, 2.0717, 3.0042, 2.3414, 2.7489, 2.2595, 1.7951, 1.4688], + device='cuda:3'), covar=tensor([0.3520, 0.3754, 0.1132, 0.2417, 0.1727, 0.2004, 0.1590, 0.3868], + device='cuda:3'), in_proj_covar=tensor([0.0872, 0.0841, 0.0711, 0.0821, 0.0912, 0.0780, 0.0692, 0.0752], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 09:53:18,507 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.736e+02 2.436e+02 2.887e+02 3.709e+02 9.106e+02, threshold=5.774e+02, percent-clipped=2.0 +2023-02-06 09:53:38,028 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-06 09:53:41,608 INFO [train.py:901] (3/4) Epoch 11, batch 1150, loss[loss=0.2456, simple_loss=0.3054, pruned_loss=0.09286, over 7540.00 frames. ], tot_loss[loss=0.236, simple_loss=0.3112, pruned_loss=0.08046, over 1614746.75 frames. ], batch size: 18, lr: 7.09e-03, grad_scale: 8.0 +2023-02-06 09:53:51,771 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 09:54:00,459 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82005.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:54:10,936 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-06 09:54:14,795 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7325, 2.2057, 3.6091, 2.7413, 3.1646, 2.3823, 1.8904, 1.9776], + device='cuda:3'), covar=tensor([0.3574, 0.4423, 0.1151, 0.2470, 0.1798, 0.2006, 0.1591, 0.3981], + device='cuda:3'), in_proj_covar=tensor([0.0868, 0.0838, 0.0708, 0.0816, 0.0905, 0.0776, 0.0688, 0.0747], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 09:54:17,074 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.06 vs. limit=2.0 +2023-02-06 09:54:17,962 INFO [train.py:901] (3/4) Epoch 11, batch 1200, loss[loss=0.1825, simple_loss=0.2516, pruned_loss=0.05669, over 7541.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.311, pruned_loss=0.08044, over 1617763.92 frames. ], batch size: 18, lr: 7.09e-03, grad_scale: 8.0 +2023-02-06 09:54:18,727 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82032.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:54:18,848 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82032.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:54:20,238 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82034.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:54:29,493 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.543e+02 2.664e+02 3.172e+02 3.772e+02 1.117e+03, threshold=6.345e+02, percent-clipped=5.0 +2023-02-06 09:54:36,791 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82057.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:54:38,209 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82059.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:54:48,002 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82073.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:54:53,473 INFO [train.py:901] (3/4) Epoch 11, batch 1250, loss[loss=0.2148, simple_loss=0.2837, pruned_loss=0.07296, over 7982.00 frames. ], tot_loss[loss=0.2354, simple_loss=0.3102, pruned_loss=0.08034, over 1612508.71 frames. ], batch size: 21, lr: 7.09e-03, grad_scale: 8.0 +2023-02-06 09:55:29,351 INFO [train.py:901] (3/4) Epoch 11, batch 1300, loss[loss=0.2203, simple_loss=0.3017, pruned_loss=0.06948, over 8460.00 frames. ], tot_loss[loss=0.2357, simple_loss=0.3107, pruned_loss=0.08038, over 1614594.79 frames. ], batch size: 25, lr: 7.08e-03, grad_scale: 8.0 +2023-02-06 09:55:32,254 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3291, 1.8509, 2.7581, 2.1767, 2.4933, 2.1484, 1.7650, 1.1028], + device='cuda:3'), covar=tensor([0.3734, 0.3773, 0.1026, 0.2352, 0.1757, 0.2145, 0.1589, 0.3953], + device='cuda:3'), in_proj_covar=tensor([0.0880, 0.0849, 0.0713, 0.0824, 0.0914, 0.0785, 0.0696, 0.0754], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 09:55:40,431 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82147.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:55:40,862 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.630e+02 2.583e+02 3.223e+02 4.179e+02 7.623e+02, threshold=6.447e+02, percent-clipped=2.0 +2023-02-06 09:55:46,226 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1411, 4.1224, 3.7376, 2.0275, 3.6906, 3.6491, 3.8235, 3.4041], + device='cuda:3'), covar=tensor([0.0796, 0.0549, 0.0883, 0.4576, 0.0846, 0.1179, 0.1068, 0.0914], + device='cuda:3'), in_proj_covar=tensor([0.0460, 0.0366, 0.0375, 0.0476, 0.0370, 0.0363, 0.0366, 0.0320], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 09:56:03,684 INFO [train.py:901] (3/4) Epoch 11, batch 1350, loss[loss=0.2358, simple_loss=0.3033, pruned_loss=0.08415, over 7546.00 frames. ], tot_loss[loss=0.2365, simple_loss=0.3111, pruned_loss=0.08089, over 1615073.11 frames. ], batch size: 18, lr: 7.08e-03, grad_scale: 8.0 +2023-02-06 09:56:38,846 INFO [train.py:901] (3/4) Epoch 11, batch 1400, loss[loss=0.232, simple_loss=0.3186, pruned_loss=0.07274, over 8257.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3116, pruned_loss=0.08081, over 1619317.89 frames. ], batch size: 24, lr: 7.08e-03, grad_scale: 8.0 +2023-02-06 09:56:51,062 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.653e+02 2.627e+02 3.119e+02 3.954e+02 1.224e+03, threshold=6.238e+02, percent-clipped=1.0 +2023-02-06 09:57:13,115 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5285, 2.9762, 2.6060, 4.0183, 1.7814, 1.9793, 2.4934, 3.1341], + device='cuda:3'), covar=tensor([0.0780, 0.0916, 0.0985, 0.0273, 0.1263, 0.1549, 0.1133, 0.0825], + device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0214, 0.0255, 0.0219, 0.0218, 0.0256, 0.0254, 0.0225], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 09:57:13,607 INFO [train.py:901] (3/4) Epoch 11, batch 1450, loss[loss=0.1769, simple_loss=0.2494, pruned_loss=0.05217, over 8051.00 frames. ], tot_loss[loss=0.2357, simple_loss=0.3109, pruned_loss=0.08023, over 1619072.70 frames. ], batch size: 20, lr: 7.08e-03, grad_scale: 8.0 +2023-02-06 09:57:27,756 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 09:57:46,369 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.72 vs. limit=5.0 +2023-02-06 09:57:48,797 INFO [train.py:901] (3/4) Epoch 11, batch 1500, loss[loss=0.2498, simple_loss=0.3246, pruned_loss=0.0875, over 8686.00 frames. ], tot_loss[loss=0.2369, simple_loss=0.3115, pruned_loss=0.08111, over 1615053.81 frames. ], batch size: 39, lr: 7.08e-03, grad_scale: 8.0 +2023-02-06 09:58:01,381 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.943e+02 2.743e+02 3.193e+02 4.270e+02 9.879e+02, threshold=6.387e+02, percent-clipped=7.0 +2023-02-06 09:58:02,224 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82349.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:58:13,199 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82364.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:58:24,634 INFO [train.py:901] (3/4) Epoch 11, batch 1550, loss[loss=0.2275, simple_loss=0.3119, pruned_loss=0.0716, over 8357.00 frames. ], tot_loss[loss=0.2355, simple_loss=0.3104, pruned_loss=0.08029, over 1614918.40 frames. ], batch size: 24, lr: 7.07e-03, grad_scale: 8.0 +2023-02-06 09:58:39,990 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82403.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:58:50,129 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82417.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:58:50,255 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9734, 2.3106, 1.9199, 2.9556, 1.4449, 1.6103, 1.9668, 2.3744], + device='cuda:3'), covar=tensor([0.0825, 0.0903, 0.1069, 0.0438, 0.1184, 0.1521, 0.1059, 0.0866], + device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0212, 0.0251, 0.0216, 0.0216, 0.0252, 0.0253, 0.0223], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 09:58:57,865 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82428.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:58:59,756 INFO [train.py:901] (3/4) Epoch 11, batch 1600, loss[loss=0.1958, simple_loss=0.282, pruned_loss=0.05478, over 8342.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.3109, pruned_loss=0.08037, over 1620563.34 frames. ], batch size: 26, lr: 7.07e-03, grad_scale: 8.0 +2023-02-06 09:59:13,002 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.709e+02 2.328e+02 2.878e+02 3.468e+02 7.869e+02, threshold=5.757e+02, percent-clipped=2.0 +2023-02-06 09:59:24,128 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82464.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 09:59:36,407 INFO [train.py:901] (3/4) Epoch 11, batch 1650, loss[loss=0.2993, simple_loss=0.3511, pruned_loss=0.1237, over 7084.00 frames. ], tot_loss[loss=0.237, simple_loss=0.3115, pruned_loss=0.08123, over 1617160.41 frames. ], batch size: 71, lr: 7.07e-03, grad_scale: 8.0 +2023-02-06 09:59:57,314 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82511.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 10:00:11,612 INFO [train.py:901] (3/4) Epoch 11, batch 1700, loss[loss=0.2211, simple_loss=0.3096, pruned_loss=0.06629, over 8330.00 frames. ], tot_loss[loss=0.2364, simple_loss=0.3114, pruned_loss=0.08076, over 1616480.18 frames. ], batch size: 25, lr: 7.07e-03, grad_scale: 8.0 +2023-02-06 10:00:12,448 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82532.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:00:17,635 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-06 10:00:23,205 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.722e+02 2.517e+02 3.185e+02 4.066e+02 8.085e+02, threshold=6.370e+02, percent-clipped=5.0 +2023-02-06 10:00:47,531 INFO [train.py:901] (3/4) Epoch 11, batch 1750, loss[loss=0.2437, simple_loss=0.31, pruned_loss=0.08871, over 8101.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.3103, pruned_loss=0.07977, over 1614846.84 frames. ], batch size: 23, lr: 7.06e-03, grad_scale: 8.0 +2023-02-06 10:00:52,546 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82587.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:01:00,903 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5835, 2.6284, 1.8516, 2.3696, 2.3068, 1.5328, 2.1350, 2.1850], + device='cuda:3'), covar=tensor([0.1328, 0.0344, 0.0993, 0.0559, 0.0578, 0.1390, 0.0882, 0.1025], + device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0234, 0.0313, 0.0298, 0.0300, 0.0324, 0.0340, 0.0307], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 10:01:23,316 INFO [train.py:901] (3/4) Epoch 11, batch 1800, loss[loss=0.2496, simple_loss=0.3124, pruned_loss=0.09344, over 7660.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.3113, pruned_loss=0.08045, over 1620898.46 frames. ], batch size: 19, lr: 7.06e-03, grad_scale: 8.0 +2023-02-06 10:01:35,825 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 2.598e+02 3.107e+02 4.193e+02 1.199e+03, threshold=6.213e+02, percent-clipped=8.0 +2023-02-06 10:01:58,586 INFO [train.py:901] (3/4) Epoch 11, batch 1850, loss[loss=0.2222, simple_loss=0.2831, pruned_loss=0.08068, over 7540.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.3113, pruned_loss=0.08046, over 1620457.39 frames. ], batch size: 18, lr: 7.06e-03, grad_scale: 8.0 +2023-02-06 10:02:18,370 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82708.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:02:21,804 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1708, 1.7396, 3.5177, 1.3570, 2.2372, 3.8566, 3.9452, 3.3205], + device='cuda:3'), covar=tensor([0.0938, 0.1414, 0.0302, 0.2139, 0.1010, 0.0222, 0.0433, 0.0559], + device='cuda:3'), in_proj_covar=tensor([0.0258, 0.0294, 0.0255, 0.0285, 0.0270, 0.0232, 0.0330, 0.0288], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 10:02:26,598 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82720.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:02:32,178 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7302, 2.3016, 3.7062, 2.7124, 3.0467, 2.4883, 1.9347, 1.7813], + device='cuda:3'), covar=tensor([0.3586, 0.4223, 0.1041, 0.2713, 0.2033, 0.2022, 0.1680, 0.4483], + device='cuda:3'), in_proj_covar=tensor([0.0887, 0.0862, 0.0727, 0.0836, 0.0932, 0.0793, 0.0704, 0.0761], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 10:02:34,011 INFO [train.py:901] (3/4) Epoch 11, batch 1900, loss[loss=0.2613, simple_loss=0.3358, pruned_loss=0.09337, over 8189.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.3114, pruned_loss=0.08043, over 1618784.17 frames. ], batch size: 23, lr: 7.06e-03, grad_scale: 8.0 +2023-02-06 10:02:43,169 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-06 10:02:43,697 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82745.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:02:45,528 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.672e+02 2.442e+02 3.142e+02 3.936e+02 6.780e+02, threshold=6.284e+02, percent-clipped=1.0 +2023-02-06 10:03:08,869 INFO [train.py:901] (3/4) Epoch 11, batch 1950, loss[loss=0.2185, simple_loss=0.2824, pruned_loss=0.07733, over 7219.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3111, pruned_loss=0.08108, over 1616772.43 frames. ], batch size: 16, lr: 7.06e-03, grad_scale: 8.0 +2023-02-06 10:03:12,343 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 10:03:13,861 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82788.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:03:17,231 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2825, 2.6651, 3.3086, 1.2897, 3.2717, 1.9563, 1.5596, 1.9596], + device='cuda:3'), covar=tensor([0.0495, 0.0222, 0.0161, 0.0527, 0.0253, 0.0597, 0.0644, 0.0355], + device='cuda:3'), in_proj_covar=tensor([0.0382, 0.0317, 0.0266, 0.0376, 0.0304, 0.0468, 0.0352, 0.0347], + device='cuda:3'), out_proj_covar=tensor([1.0988e-04, 8.9128e-05, 7.4872e-05, 1.0638e-04, 8.6597e-05, 1.4414e-04, + 1.0123e-04, 9.9186e-05], device='cuda:3') +2023-02-06 10:03:26,496 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 10:03:32,228 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82813.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:03:39,053 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82823.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:03:45,036 INFO [train.py:901] (3/4) Epoch 11, batch 2000, loss[loss=0.1845, simple_loss=0.2569, pruned_loss=0.05609, over 7232.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3108, pruned_loss=0.08119, over 1617263.28 frames. ], batch size: 16, lr: 7.05e-03, grad_scale: 8.0 +2023-02-06 10:03:47,029 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 10:03:56,671 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.644e+02 2.675e+02 3.279e+02 3.987e+02 1.082e+03, threshold=6.559e+02, percent-clipped=7.0 +2023-02-06 10:04:01,562 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82855.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 10:04:09,121 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4032, 1.5936, 4.2324, 1.8957, 2.1880, 4.9950, 4.9699, 4.2482], + device='cuda:3'), covar=tensor([0.0960, 0.1639, 0.0299, 0.1888, 0.1243, 0.0164, 0.0309, 0.0584], + device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0289, 0.0249, 0.0279, 0.0266, 0.0228, 0.0323, 0.0283], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2023-02-06 10:04:12,622 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5913, 4.5005, 3.9710, 2.1390, 4.0446, 4.1952, 4.1550, 3.8852], + device='cuda:3'), covar=tensor([0.0678, 0.0596, 0.1098, 0.4451, 0.0794, 0.0767, 0.1157, 0.0908], + device='cuda:3'), in_proj_covar=tensor([0.0466, 0.0375, 0.0382, 0.0483, 0.0375, 0.0366, 0.0369, 0.0325], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 10:04:16,807 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5234, 2.1808, 4.1012, 1.3543, 3.0201, 2.0957, 1.6974, 2.6984], + device='cuda:3'), covar=tensor([0.1712, 0.1986, 0.0758, 0.3873, 0.1461, 0.2704, 0.1811, 0.2374], + device='cuda:3'), in_proj_covar=tensor([0.0480, 0.0512, 0.0534, 0.0574, 0.0613, 0.0545, 0.0468, 0.0607], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 10:04:19,320 INFO [train.py:901] (3/4) Epoch 11, batch 2050, loss[loss=0.2378, simple_loss=0.3076, pruned_loss=0.08402, over 7924.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3111, pruned_loss=0.08156, over 1613740.11 frames. ], batch size: 20, lr: 7.05e-03, grad_scale: 8.0 +2023-02-06 10:04:55,299 INFO [train.py:901] (3/4) Epoch 11, batch 2100, loss[loss=0.2094, simple_loss=0.2811, pruned_loss=0.06887, over 7521.00 frames. ], tot_loss[loss=0.2363, simple_loss=0.3108, pruned_loss=0.08092, over 1617179.46 frames. ], batch size: 18, lr: 7.05e-03, grad_scale: 8.0 +2023-02-06 10:04:55,378 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82931.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:05:07,191 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.793e+02 2.489e+02 3.174e+02 3.706e+02 9.083e+02, threshold=6.348e+02, percent-clipped=2.0 +2023-02-06 10:05:22,111 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82970.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 10:05:29,198 INFO [train.py:901] (3/4) Epoch 11, batch 2150, loss[loss=0.2569, simple_loss=0.3321, pruned_loss=0.09084, over 8242.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3116, pruned_loss=0.0808, over 1618964.47 frames. ], batch size: 24, lr: 7.05e-03, grad_scale: 8.0 +2023-02-06 10:05:29,349 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82981.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:06:04,069 INFO [train.py:901] (3/4) Epoch 11, batch 2200, loss[loss=0.2199, simple_loss=0.308, pruned_loss=0.06595, over 8333.00 frames. ], tot_loss[loss=0.236, simple_loss=0.3107, pruned_loss=0.08069, over 1615142.97 frames. ], batch size: 26, lr: 7.05e-03, grad_scale: 8.0 +2023-02-06 10:06:15,764 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83046.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:06:16,953 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.616e+02 2.510e+02 3.092e+02 4.104e+02 1.639e+03, threshold=6.185e+02, percent-clipped=4.0 +2023-02-06 10:06:31,480 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4783, 1.7633, 1.8712, 1.1696, 1.9312, 1.2755, 0.4757, 1.6638], + device='cuda:3'), covar=tensor([0.0306, 0.0199, 0.0164, 0.0292, 0.0230, 0.0553, 0.0502, 0.0158], + device='cuda:3'), in_proj_covar=tensor([0.0381, 0.0316, 0.0264, 0.0374, 0.0302, 0.0467, 0.0350, 0.0346], + device='cuda:3'), out_proj_covar=tensor([1.0941e-04, 8.8922e-05, 7.4357e-05, 1.0567e-04, 8.5977e-05, 1.4398e-04, + 1.0076e-04, 9.8853e-05], device='cuda:3') +2023-02-06 10:06:38,957 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83079.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:06:40,009 INFO [train.py:901] (3/4) Epoch 11, batch 2250, loss[loss=0.2529, simple_loss=0.3332, pruned_loss=0.08631, over 8574.00 frames. ], tot_loss[loss=0.2387, simple_loss=0.313, pruned_loss=0.08223, over 1616867.10 frames. ], batch size: 39, lr: 7.04e-03, grad_scale: 8.0 +2023-02-06 10:06:55,779 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83104.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:07:13,675 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5053, 1.8658, 2.8178, 1.2950, 2.0712, 1.9439, 1.6109, 1.7876], + device='cuda:3'), covar=tensor([0.1707, 0.2021, 0.0745, 0.3961, 0.1523, 0.2661, 0.1859, 0.2077], + device='cuda:3'), in_proj_covar=tensor([0.0476, 0.0507, 0.0529, 0.0568, 0.0608, 0.0538, 0.0463, 0.0600], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 10:07:14,109 INFO [train.py:901] (3/4) Epoch 11, batch 2300, loss[loss=0.2296, simple_loss=0.2952, pruned_loss=0.08201, over 7437.00 frames. ], tot_loss[loss=0.2379, simple_loss=0.312, pruned_loss=0.08191, over 1614060.53 frames. ], batch size: 17, lr: 7.04e-03, grad_scale: 8.0 +2023-02-06 10:07:25,657 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.556e+02 2.480e+02 3.199e+02 4.275e+02 9.806e+02, threshold=6.398e+02, percent-clipped=6.0 +2023-02-06 10:07:48,927 INFO [train.py:901] (3/4) Epoch 11, batch 2350, loss[loss=0.2265, simple_loss=0.3103, pruned_loss=0.07129, over 8458.00 frames. ], tot_loss[loss=0.2388, simple_loss=0.3132, pruned_loss=0.08223, over 1614734.94 frames. ], batch size: 27, lr: 7.04e-03, grad_scale: 16.0 +2023-02-06 10:08:12,077 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83214.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:08:20,049 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83226.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 10:08:23,038 INFO [train.py:901] (3/4) Epoch 11, batch 2400, loss[loss=0.244, simple_loss=0.3142, pruned_loss=0.08688, over 8246.00 frames. ], tot_loss[loss=0.2395, simple_loss=0.3137, pruned_loss=0.08267, over 1610794.40 frames. ], batch size: 24, lr: 7.04e-03, grad_scale: 16.0 +2023-02-06 10:08:35,105 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.638e+02 2.359e+02 2.853e+02 3.666e+02 7.740e+02, threshold=5.706e+02, percent-clipped=1.0 +2023-02-06 10:08:37,260 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83251.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 10:08:58,651 INFO [train.py:901] (3/4) Epoch 11, batch 2450, loss[loss=0.2923, simple_loss=0.3546, pruned_loss=0.115, over 7048.00 frames. ], tot_loss[loss=0.239, simple_loss=0.3134, pruned_loss=0.08232, over 1612673.01 frames. ], batch size: 72, lr: 7.04e-03, grad_scale: 16.0 +2023-02-06 10:09:13,733 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83302.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:09:13,800 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83302.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:09:29,008 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83325.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:09:30,530 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83327.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:09:33,054 INFO [train.py:901] (3/4) Epoch 11, batch 2500, loss[loss=0.1914, simple_loss=0.2702, pruned_loss=0.05628, over 7211.00 frames. ], tot_loss[loss=0.2379, simple_loss=0.3123, pruned_loss=0.08178, over 1612254.02 frames. ], batch size: 16, lr: 7.03e-03, grad_scale: 16.0 +2023-02-06 10:09:44,607 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.719e+02 2.634e+02 3.143e+02 3.904e+02 7.323e+02, threshold=6.285e+02, percent-clipped=4.0 +2023-02-06 10:10:05,062 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2309, 1.6152, 4.3742, 1.9974, 2.5066, 5.0682, 4.9881, 4.4969], + device='cuda:3'), covar=tensor([0.1046, 0.1591, 0.0271, 0.1772, 0.1012, 0.0169, 0.0390, 0.0455], + device='cuda:3'), in_proj_covar=tensor([0.0260, 0.0297, 0.0255, 0.0289, 0.0273, 0.0235, 0.0334, 0.0287], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 10:10:07,382 INFO [train.py:901] (3/4) Epoch 11, batch 2550, loss[loss=0.2063, simple_loss=0.2706, pruned_loss=0.07107, over 7249.00 frames. ], tot_loss[loss=0.2384, simple_loss=0.3125, pruned_loss=0.08212, over 1611312.19 frames. ], batch size: 16, lr: 7.03e-03, grad_scale: 16.0 +2023-02-06 10:10:11,154 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-06 10:10:13,590 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3431, 2.1422, 1.6772, 1.9357, 1.7255, 1.2510, 1.6469, 1.7536], + device='cuda:3'), covar=tensor([0.1111, 0.0345, 0.0941, 0.0465, 0.0637, 0.1341, 0.0907, 0.0712], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0236, 0.0315, 0.0297, 0.0304, 0.0327, 0.0343, 0.0312], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 10:10:43,103 INFO [train.py:901] (3/4) Epoch 11, batch 2600, loss[loss=0.2336, simple_loss=0.3192, pruned_loss=0.07403, over 8425.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.3132, pruned_loss=0.08259, over 1613997.88 frames. ], batch size: 27, lr: 7.03e-03, grad_scale: 16.0 +2023-02-06 10:10:49,500 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83440.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:10:54,705 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.898e+02 2.607e+02 3.192e+02 4.372e+02 8.439e+02, threshold=6.384e+02, percent-clipped=10.0 +2023-02-06 10:11:17,510 INFO [train.py:901] (3/4) Epoch 11, batch 2650, loss[loss=0.224, simple_loss=0.3085, pruned_loss=0.06973, over 8299.00 frames. ], tot_loss[loss=0.2381, simple_loss=0.3122, pruned_loss=0.08201, over 1616291.17 frames. ], batch size: 23, lr: 7.03e-03, grad_scale: 16.0 +2023-02-06 10:11:25,738 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5378, 1.5010, 2.8592, 1.2980, 2.0259, 3.0046, 3.1136, 2.5775], + device='cuda:3'), covar=tensor([0.1027, 0.1362, 0.0350, 0.1945, 0.0822, 0.0288, 0.0499, 0.0623], + device='cuda:3'), in_proj_covar=tensor([0.0257, 0.0295, 0.0252, 0.0286, 0.0268, 0.0232, 0.0330, 0.0284], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], + device='cuda:3') +2023-02-06 10:11:33,051 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.0246, 1.3066, 3.1753, 1.0406, 2.7531, 2.6073, 2.8782, 2.7462], + device='cuda:3'), covar=tensor([0.0797, 0.3659, 0.0759, 0.3591, 0.1409, 0.1011, 0.0713, 0.0905], + device='cuda:3'), in_proj_covar=tensor([0.0482, 0.0572, 0.0584, 0.0530, 0.0604, 0.0506, 0.0502, 0.0577], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 10:11:49,224 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1199, 4.0890, 3.6991, 2.0786, 3.6998, 3.7164, 3.7480, 3.2752], + device='cuda:3'), covar=tensor([0.0955, 0.0710, 0.1249, 0.4211, 0.0906, 0.0785, 0.1429, 0.0900], + device='cuda:3'), in_proj_covar=tensor([0.0455, 0.0368, 0.0376, 0.0477, 0.0370, 0.0366, 0.0367, 0.0323], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 10:11:52,412 INFO [train.py:901] (3/4) Epoch 11, batch 2700, loss[loss=0.2704, simple_loss=0.3231, pruned_loss=0.1088, over 7697.00 frames. ], tot_loss[loss=0.238, simple_loss=0.3123, pruned_loss=0.08184, over 1614263.94 frames. ], batch size: 18, lr: 7.02e-03, grad_scale: 16.0 +2023-02-06 10:12:04,666 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.775e+02 2.578e+02 3.131e+02 4.095e+02 6.916e+02, threshold=6.263e+02, percent-clipped=2.0 +2023-02-06 10:12:11,629 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83558.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:12:23,378 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5902, 2.0263, 2.1834, 1.0665, 2.1342, 1.4808, 0.6239, 1.7592], + device='cuda:3'), covar=tensor([0.0448, 0.0216, 0.0170, 0.0382, 0.0298, 0.0651, 0.0574, 0.0219], + device='cuda:3'), in_proj_covar=tensor([0.0375, 0.0315, 0.0259, 0.0372, 0.0299, 0.0461, 0.0349, 0.0341], + device='cuda:3'), out_proj_covar=tensor([1.0767e-04, 8.8357e-05, 7.2768e-05, 1.0509e-04, 8.5232e-05, 1.4174e-04, + 1.0048e-04, 9.7331e-05], device='cuda:3') +2023-02-06 10:12:27,318 INFO [train.py:901] (3/4) Epoch 11, batch 2750, loss[loss=0.2198, simple_loss=0.2987, pruned_loss=0.07047, over 8254.00 frames. ], tot_loss[loss=0.2395, simple_loss=0.3135, pruned_loss=0.08275, over 1610403.99 frames. ], batch size: 24, lr: 7.02e-03, grad_scale: 16.0 +2023-02-06 10:13:03,334 INFO [train.py:901] (3/4) Epoch 11, batch 2800, loss[loss=0.2523, simple_loss=0.3315, pruned_loss=0.08654, over 8112.00 frames. ], tot_loss[loss=0.24, simple_loss=0.3139, pruned_loss=0.08305, over 1611515.62 frames. ], batch size: 23, lr: 7.02e-03, grad_scale: 16.0 +2023-02-06 10:13:13,811 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83646.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:13:15,055 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.592e+02 2.535e+02 3.136e+02 3.769e+02 1.201e+03, threshold=6.273e+02, percent-clipped=3.0 +2023-02-06 10:13:32,688 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83673.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:13:35,279 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7089, 3.7057, 1.8691, 2.2741, 2.5163, 1.6123, 2.3215, 2.7677], + device='cuda:3'), covar=tensor([0.1937, 0.0467, 0.1406, 0.1000, 0.0880, 0.1797, 0.1462, 0.1188], + device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0235, 0.0313, 0.0294, 0.0301, 0.0323, 0.0338, 0.0309], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 10:13:37,605 INFO [train.py:901] (3/4) Epoch 11, batch 2850, loss[loss=0.3247, simple_loss=0.402, pruned_loss=0.1237, over 8315.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3151, pruned_loss=0.08397, over 1609640.78 frames. ], batch size: 25, lr: 7.02e-03, grad_scale: 16.0 +2023-02-06 10:13:47,914 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83696.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:14:05,642 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83721.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:14:12,967 INFO [train.py:901] (3/4) Epoch 11, batch 2900, loss[loss=0.2449, simple_loss=0.3237, pruned_loss=0.08304, over 8344.00 frames. ], tot_loss[loss=0.2401, simple_loss=0.3138, pruned_loss=0.08324, over 1610778.11 frames. ], batch size: 26, lr: 7.02e-03, grad_scale: 16.0 +2023-02-06 10:14:25,269 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.718e+02 2.545e+02 3.159e+02 4.165e+02 9.643e+02, threshold=6.318e+02, percent-clipped=5.0 +2023-02-06 10:14:27,059 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 10:14:34,296 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83761.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:14:46,314 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83778.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:14:48,174 INFO [train.py:901] (3/4) Epoch 11, batch 2950, loss[loss=0.2822, simple_loss=0.346, pruned_loss=0.1092, over 8556.00 frames. ], tot_loss[loss=0.2398, simple_loss=0.3141, pruned_loss=0.08275, over 1613006.23 frames. ], batch size: 49, lr: 7.01e-03, grad_scale: 16.0 +2023-02-06 10:14:53,615 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 10:15:22,303 INFO [train.py:901] (3/4) Epoch 11, batch 3000, loss[loss=0.2464, simple_loss=0.319, pruned_loss=0.08692, over 7969.00 frames. ], tot_loss[loss=0.2391, simple_loss=0.3136, pruned_loss=0.08232, over 1609388.08 frames. ], batch size: 21, lr: 7.01e-03, grad_scale: 16.0 +2023-02-06 10:15:22,303 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 10:15:34,550 INFO [train.py:935] (3/4) Epoch 11, validation: loss=0.1889, simple_loss=0.2886, pruned_loss=0.04461, over 944034.00 frames. +2023-02-06 10:15:34,551 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 10:15:46,618 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.677e+02 2.511e+02 2.977e+02 3.600e+02 5.313e+02, threshold=5.953e+02, percent-clipped=0.0 +2023-02-06 10:16:10,357 INFO [train.py:901] (3/4) Epoch 11, batch 3050, loss[loss=0.224, simple_loss=0.3039, pruned_loss=0.072, over 7982.00 frames. ], tot_loss[loss=0.2382, simple_loss=0.3134, pruned_loss=0.08153, over 1612731.94 frames. ], batch size: 21, lr: 7.01e-03, grad_scale: 16.0 +2023-02-06 10:16:30,257 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 +2023-02-06 10:16:43,146 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83929.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:16:44,283 INFO [train.py:901] (3/4) Epoch 11, batch 3100, loss[loss=0.2415, simple_loss=0.3245, pruned_loss=0.0792, over 8426.00 frames. ], tot_loss[loss=0.2382, simple_loss=0.3135, pruned_loss=0.08141, over 1613180.70 frames. ], batch size: 27, lr: 7.01e-03, grad_scale: 16.0 +2023-02-06 10:16:55,423 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.027e+02 2.748e+02 3.262e+02 3.755e+02 7.942e+02, threshold=6.525e+02, percent-clipped=1.0 +2023-02-06 10:17:00,086 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83954.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:17:08,766 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83967.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:17:18,437 INFO [train.py:901] (3/4) Epoch 11, batch 3150, loss[loss=0.242, simple_loss=0.311, pruned_loss=0.08651, over 7513.00 frames. ], tot_loss[loss=0.2384, simple_loss=0.3133, pruned_loss=0.08175, over 1615799.30 frames. ], batch size: 18, lr: 7.01e-03, grad_scale: 16.0 +2023-02-06 10:17:24,696 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9406, 1.5614, 2.1707, 1.8499, 2.0328, 1.8711, 1.5473, 0.7085], + device='cuda:3'), covar=tensor([0.3966, 0.3710, 0.1223, 0.2162, 0.1587, 0.2084, 0.1588, 0.3699], + device='cuda:3'), in_proj_covar=tensor([0.0884, 0.0865, 0.0726, 0.0837, 0.0935, 0.0794, 0.0701, 0.0763], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 10:17:34,200 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84003.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:17:44,270 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84017.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:17:53,260 INFO [train.py:901] (3/4) Epoch 11, batch 3200, loss[loss=0.2409, simple_loss=0.3197, pruned_loss=0.08103, over 8357.00 frames. ], tot_loss[loss=0.239, simple_loss=0.3132, pruned_loss=0.08235, over 1615350.62 frames. ], batch size: 24, lr: 7.00e-03, grad_scale: 8.0 +2023-02-06 10:18:01,384 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84042.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:18:05,771 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.534e+02 2.726e+02 3.369e+02 4.220e+02 9.302e+02, threshold=6.739e+02, percent-clipped=4.0 +2023-02-06 10:18:27,190 INFO [train.py:901] (3/4) Epoch 11, batch 3250, loss[loss=0.2758, simple_loss=0.3401, pruned_loss=0.1057, over 8418.00 frames. ], tot_loss[loss=0.2395, simple_loss=0.3139, pruned_loss=0.08255, over 1615838.16 frames. ], batch size: 39, lr: 7.00e-03, grad_scale: 8.0 +2023-02-06 10:18:50,444 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84115.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:18:55,082 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84122.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:18:55,570 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2023-02-06 10:19:01,829 INFO [train.py:901] (3/4) Epoch 11, batch 3300, loss[loss=0.2185, simple_loss=0.2944, pruned_loss=0.07124, over 8131.00 frames. ], tot_loss[loss=0.2384, simple_loss=0.3125, pruned_loss=0.08214, over 1614642.98 frames. ], batch size: 22, lr: 7.00e-03, grad_scale: 8.0 +2023-02-06 10:19:13,378 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.547e+02 2.729e+02 3.101e+02 4.103e+02 8.191e+02, threshold=6.202e+02, percent-clipped=3.0 +2023-02-06 10:19:29,807 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3552, 2.4085, 1.6963, 2.0339, 1.8745, 1.3680, 1.7447, 1.8826], + device='cuda:3'), covar=tensor([0.1300, 0.0309, 0.1030, 0.0539, 0.0616, 0.1351, 0.0921, 0.0872], + device='cuda:3'), in_proj_covar=tensor([0.0344, 0.0233, 0.0314, 0.0293, 0.0301, 0.0322, 0.0337, 0.0307], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 10:19:35,417 INFO [train.py:901] (3/4) Epoch 11, batch 3350, loss[loss=0.2058, simple_loss=0.2931, pruned_loss=0.05929, over 7421.00 frames. ], tot_loss[loss=0.2382, simple_loss=0.3123, pruned_loss=0.08205, over 1612880.07 frames. ], batch size: 17, lr: 7.00e-03, grad_scale: 8.0 +2023-02-06 10:20:01,020 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84217.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:20:10,209 INFO [train.py:901] (3/4) Epoch 11, batch 3400, loss[loss=0.2111, simple_loss=0.2953, pruned_loss=0.06344, over 8021.00 frames. ], tot_loss[loss=0.2383, simple_loss=0.3125, pruned_loss=0.08201, over 1612729.09 frames. ], batch size: 22, lr: 7.00e-03, grad_scale: 8.0 +2023-02-06 10:20:15,228 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84237.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:20:23,162 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.356e+02 2.553e+02 3.068e+02 3.977e+02 7.727e+02, threshold=6.137e+02, percent-clipped=2.0 +2023-02-06 10:20:26,657 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84254.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:20:28,773 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7117, 2.0700, 2.3469, 1.4968, 2.3637, 1.6635, 0.7921, 1.8619], + device='cuda:3'), covar=tensor([0.0419, 0.0228, 0.0159, 0.0371, 0.0246, 0.0572, 0.0512, 0.0200], + device='cuda:3'), in_proj_covar=tensor([0.0389, 0.0323, 0.0268, 0.0386, 0.0309, 0.0472, 0.0358, 0.0352], + device='cuda:3'), out_proj_covar=tensor([1.1150e-04, 9.0472e-05, 7.4890e-05, 1.0921e-04, 8.7879e-05, 1.4481e-04, + 1.0276e-04, 1.0054e-04], device='cuda:3') +2023-02-06 10:20:32,809 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0144, 2.4370, 2.9105, 1.4340, 2.8654, 1.7872, 1.5916, 1.9714], + device='cuda:3'), covar=tensor([0.0542, 0.0280, 0.0170, 0.0502, 0.0317, 0.0653, 0.0544, 0.0329], + device='cuda:3'), in_proj_covar=tensor([0.0389, 0.0323, 0.0267, 0.0386, 0.0309, 0.0472, 0.0358, 0.0352], + device='cuda:3'), out_proj_covar=tensor([1.1149e-04, 9.0398e-05, 7.4871e-05, 1.0912e-04, 8.7882e-05, 1.4486e-04, + 1.0264e-04, 1.0054e-04], device='cuda:3') +2023-02-06 10:20:34,982 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-06 10:20:45,376 INFO [train.py:901] (3/4) Epoch 11, batch 3450, loss[loss=0.2299, simple_loss=0.2999, pruned_loss=0.07996, over 8106.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.3125, pruned_loss=0.08232, over 1612559.94 frames. ], batch size: 23, lr: 6.99e-03, grad_scale: 8.0 +2023-02-06 10:21:06,407 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84311.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:21:20,277 INFO [train.py:901] (3/4) Epoch 11, batch 3500, loss[loss=0.2402, simple_loss=0.3099, pruned_loss=0.08529, over 8237.00 frames. ], tot_loss[loss=0.2387, simple_loss=0.3134, pruned_loss=0.082, over 1617697.96 frames. ], batch size: 22, lr: 6.99e-03, grad_scale: 8.0 +2023-02-06 10:21:31,078 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84347.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:21:32,273 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.912e+02 2.703e+02 3.166e+02 4.187e+02 8.001e+02, threshold=6.332e+02, percent-clipped=6.0 +2023-02-06 10:21:36,488 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84354.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:21:48,769 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 10:21:49,639 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0208, 2.2910, 1.6838, 2.7455, 1.2696, 1.3904, 1.8729, 2.3012], + device='cuda:3'), covar=tensor([0.0757, 0.0867, 0.1055, 0.0447, 0.1246, 0.1637, 0.1010, 0.0752], + device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0219, 0.0258, 0.0222, 0.0221, 0.0255, 0.0259, 0.0225], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 10:21:54,138 INFO [train.py:901] (3/4) Epoch 11, batch 3550, loss[loss=0.2616, simple_loss=0.3313, pruned_loss=0.0959, over 8320.00 frames. ], tot_loss[loss=0.2393, simple_loss=0.3142, pruned_loss=0.08214, over 1621565.26 frames. ], batch size: 26, lr: 6.99e-03, grad_scale: 8.0 +2023-02-06 10:22:25,867 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84426.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:22:28,970 INFO [train.py:901] (3/4) Epoch 11, batch 3600, loss[loss=0.2478, simple_loss=0.306, pruned_loss=0.09484, over 8083.00 frames. ], tot_loss[loss=0.238, simple_loss=0.313, pruned_loss=0.08156, over 1619071.27 frames. ], batch size: 21, lr: 6.99e-03, grad_scale: 8.0 +2023-02-06 10:22:41,776 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.826e+02 2.788e+02 3.447e+02 4.179e+02 1.001e+03, threshold=6.895e+02, percent-clipped=4.0 +2023-02-06 10:22:48,585 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84459.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:22:50,733 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84462.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:23:03,534 INFO [train.py:901] (3/4) Epoch 11, batch 3650, loss[loss=0.2463, simple_loss=0.3128, pruned_loss=0.08991, over 8244.00 frames. ], tot_loss[loss=0.2373, simple_loss=0.3128, pruned_loss=0.0809, over 1621902.35 frames. ], batch size: 22, lr: 6.99e-03, grad_scale: 8.0 +2023-02-06 10:23:10,232 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1844, 1.4697, 1.5459, 1.2929, 0.9941, 1.3998, 1.8024, 1.8560], + device='cuda:3'), covar=tensor([0.0495, 0.1147, 0.1719, 0.1433, 0.0593, 0.1423, 0.0665, 0.0556], + device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0154, 0.0191, 0.0158, 0.0105, 0.0163, 0.0117, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:3') +2023-02-06 10:23:11,597 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84493.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:23:28,893 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84518.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:23:37,223 INFO [train.py:901] (3/4) Epoch 11, batch 3700, loss[loss=0.2013, simple_loss=0.2702, pruned_loss=0.06625, over 7537.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.3138, pruned_loss=0.0817, over 1619676.40 frames. ], batch size: 18, lr: 6.98e-03, grad_scale: 8.0 +2023-02-06 10:23:46,131 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84543.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:23:48,592 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 10:23:49,869 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.734e+02 2.648e+02 3.219e+02 3.938e+02 7.332e+02, threshold=6.437e+02, percent-clipped=1.0 +2023-02-06 10:23:58,482 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84561.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:24:07,287 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84574.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:24:11,840 INFO [train.py:901] (3/4) Epoch 11, batch 3750, loss[loss=0.2225, simple_loss=0.3153, pruned_loss=0.06485, over 8511.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.3132, pruned_loss=0.08108, over 1619146.83 frames. ], batch size: 26, lr: 6.98e-03, grad_scale: 8.0 +2023-02-06 10:24:23,996 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84598.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:24:46,870 INFO [train.py:901] (3/4) Epoch 11, batch 3800, loss[loss=0.2575, simple_loss=0.3342, pruned_loss=0.09044, over 8348.00 frames. ], tot_loss[loss=0.2368, simple_loss=0.3122, pruned_loss=0.08067, over 1617601.01 frames. ], batch size: 26, lr: 6.98e-03, grad_scale: 8.0 +2023-02-06 10:24:58,773 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.733e+02 2.607e+02 3.118e+02 4.251e+02 1.041e+03, threshold=6.237e+02, percent-clipped=4.0 +2023-02-06 10:25:00,935 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84651.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:25:13,124 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.95 vs. limit=5.0 +2023-02-06 10:25:18,146 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84676.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:25:21,362 INFO [train.py:901] (3/4) Epoch 11, batch 3850, loss[loss=0.2207, simple_loss=0.2915, pruned_loss=0.07492, over 8082.00 frames. ], tot_loss[loss=0.2363, simple_loss=0.3116, pruned_loss=0.08049, over 1618621.92 frames. ], batch size: 21, lr: 6.98e-03, grad_scale: 8.0 +2023-02-06 10:25:22,267 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84682.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:25:32,829 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84698.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:25:39,751 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84707.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:25:43,784 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84713.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:25:47,043 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84718.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:25:51,556 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 10:25:55,492 INFO [train.py:901] (3/4) Epoch 11, batch 3900, loss[loss=0.2348, simple_loss=0.3151, pruned_loss=0.07729, over 8506.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3115, pruned_loss=0.08094, over 1615737.72 frames. ], batch size: 49, lr: 6.97e-03, grad_scale: 8.0 +2023-02-06 10:26:03,707 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84743.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:26:08,299 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.891e+02 2.619e+02 3.238e+02 3.926e+02 9.069e+02, threshold=6.476e+02, percent-clipped=5.0 +2023-02-06 10:26:30,324 INFO [train.py:901] (3/4) Epoch 11, batch 3950, loss[loss=0.3095, simple_loss=0.369, pruned_loss=0.125, over 8187.00 frames. ], tot_loss[loss=0.2368, simple_loss=0.3113, pruned_loss=0.08115, over 1610605.15 frames. ], batch size: 23, lr: 6.97e-03, grad_scale: 8.0 +2023-02-06 10:26:34,650 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9746, 3.8898, 2.4758, 2.6841, 2.9394, 1.9685, 2.8417, 2.9739], + device='cuda:3'), covar=tensor([0.1833, 0.0361, 0.1100, 0.0779, 0.0747, 0.1462, 0.1065, 0.1181], + device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0238, 0.0322, 0.0297, 0.0305, 0.0324, 0.0345, 0.0311], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 10:26:52,790 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84813.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:26:56,206 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84818.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:27:04,815 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84830.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:27:05,276 INFO [train.py:901] (3/4) Epoch 11, batch 4000, loss[loss=0.255, simple_loss=0.3286, pruned_loss=0.0907, over 8767.00 frames. ], tot_loss[loss=0.236, simple_loss=0.3106, pruned_loss=0.08066, over 1610095.76 frames. ], batch size: 39, lr: 6.97e-03, grad_scale: 8.0 +2023-02-06 10:27:17,172 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.691e+02 2.608e+02 2.990e+02 3.694e+02 8.393e+02, threshold=5.981e+02, percent-clipped=2.0 +2023-02-06 10:27:21,554 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84855.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:27:39,667 INFO [train.py:901] (3/4) Epoch 11, batch 4050, loss[loss=0.2618, simple_loss=0.3383, pruned_loss=0.0926, over 8677.00 frames. ], tot_loss[loss=0.2368, simple_loss=0.3118, pruned_loss=0.08091, over 1611779.18 frames. ], batch size: 34, lr: 6.97e-03, grad_scale: 8.0 +2023-02-06 10:27:44,487 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84887.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:28:15,554 INFO [train.py:901] (3/4) Epoch 11, batch 4100, loss[loss=0.2366, simple_loss=0.3198, pruned_loss=0.07669, over 8465.00 frames. ], tot_loss[loss=0.2374, simple_loss=0.3122, pruned_loss=0.08129, over 1611141.86 frames. ], batch size: 28, lr: 6.97e-03, grad_scale: 8.0 +2023-02-06 10:28:16,485 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84932.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:28:27,737 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.657e+02 2.518e+02 2.978e+02 3.788e+02 7.594e+02, threshold=5.956e+02, percent-clipped=4.0 +2023-02-06 10:28:33,385 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84957.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:28:41,370 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84969.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:28:41,453 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84969.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:28:43,738 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 10:28:49,610 INFO [train.py:901] (3/4) Epoch 11, batch 4150, loss[loss=0.2737, simple_loss=0.3377, pruned_loss=0.1048, over 8462.00 frames. ], tot_loss[loss=0.2383, simple_loss=0.3132, pruned_loss=0.08175, over 1613792.71 frames. ], batch size: 25, lr: 6.96e-03, grad_scale: 8.0 +2023-02-06 10:28:51,802 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6832, 1.5698, 2.8573, 1.3700, 2.0725, 3.0391, 3.1450, 2.5703], + device='cuda:3'), covar=tensor([0.1021, 0.1345, 0.0335, 0.1873, 0.0786, 0.0294, 0.0516, 0.0687], + device='cuda:3'), in_proj_covar=tensor([0.0262, 0.0298, 0.0257, 0.0289, 0.0271, 0.0237, 0.0337, 0.0293], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 10:28:54,617 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8570, 2.2605, 3.6622, 2.5877, 3.0195, 2.5534, 1.9852, 1.7601], + device='cuda:3'), covar=tensor([0.3264, 0.4194, 0.1141, 0.2797, 0.2065, 0.2039, 0.1590, 0.4306], + device='cuda:3'), in_proj_covar=tensor([0.0878, 0.0862, 0.0728, 0.0829, 0.0931, 0.0790, 0.0703, 0.0760], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 10:28:58,518 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84994.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:28:59,079 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84995.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:29:04,350 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85002.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:29:23,818 INFO [train.py:901] (3/4) Epoch 11, batch 4200, loss[loss=0.2324, simple_loss=0.3017, pruned_loss=0.0816, over 8129.00 frames. ], tot_loss[loss=0.2375, simple_loss=0.3123, pruned_loss=0.08138, over 1609570.69 frames. ], batch size: 22, lr: 6.96e-03, grad_scale: 8.0 +2023-02-06 10:29:36,445 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.682e+02 2.581e+02 3.261e+02 3.967e+02 9.417e+02, threshold=6.523e+02, percent-clipped=7.0 +2023-02-06 10:29:47,918 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 10:29:50,139 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85069.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:29:58,035 INFO [train.py:901] (3/4) Epoch 11, batch 4250, loss[loss=0.3327, simple_loss=0.3779, pruned_loss=0.1438, over 7072.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.312, pruned_loss=0.0811, over 1611045.99 frames. ], batch size: 71, lr: 6.96e-03, grad_scale: 8.0 +2023-02-06 10:30:06,871 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85094.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:30:10,079 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 10:30:18,082 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85110.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:30:25,989 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.0428, 1.2400, 3.2135, 1.0528, 2.7681, 2.6825, 2.8981, 2.7574], + device='cuda:3'), covar=tensor([0.0822, 0.3818, 0.0793, 0.3454, 0.1466, 0.0993, 0.0754, 0.0918], + device='cuda:3'), in_proj_covar=tensor([0.0481, 0.0570, 0.0574, 0.0522, 0.0601, 0.0513, 0.0502, 0.0572], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 10:30:28,613 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85125.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:30:32,505 INFO [train.py:901] (3/4) Epoch 11, batch 4300, loss[loss=0.2045, simple_loss=0.2672, pruned_loss=0.07091, over 7223.00 frames. ], tot_loss[loss=0.2378, simple_loss=0.3125, pruned_loss=0.08153, over 1611768.03 frames. ], batch size: 16, lr: 6.96e-03, grad_scale: 8.0 +2023-02-06 10:30:33,970 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85133.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:30:45,182 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.820e+02 2.616e+02 3.014e+02 4.154e+02 7.931e+02, threshold=6.027e+02, percent-clipped=5.0 +2023-02-06 10:30:54,029 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85162.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:31:06,961 INFO [train.py:901] (3/4) Epoch 11, batch 4350, loss[loss=0.2182, simple_loss=0.2845, pruned_loss=0.07592, over 7784.00 frames. ], tot_loss[loss=0.2364, simple_loss=0.3112, pruned_loss=0.0808, over 1613194.83 frames. ], batch size: 19, lr: 6.96e-03, grad_scale: 8.0 +2023-02-06 10:31:40,296 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 10:31:41,575 INFO [train.py:901] (3/4) Epoch 11, batch 4400, loss[loss=0.2486, simple_loss=0.3152, pruned_loss=0.09097, over 7936.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.3105, pruned_loss=0.0797, over 1612906.63 frames. ], batch size: 20, lr: 6.95e-03, grad_scale: 8.0 +2023-02-06 10:31:54,340 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.706e+02 2.553e+02 3.172e+02 3.669e+02 6.483e+02, threshold=6.345e+02, percent-clipped=4.0 +2023-02-06 10:32:01,485 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85258.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:32:05,695 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 10:32:14,076 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85277.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:32:16,478 INFO [train.py:901] (3/4) Epoch 11, batch 4450, loss[loss=0.263, simple_loss=0.3403, pruned_loss=0.09282, over 8738.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.3096, pruned_loss=0.07931, over 1613831.80 frames. ], batch size: 34, lr: 6.95e-03, grad_scale: 8.0 +2023-02-06 10:32:18,781 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85283.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:32:22,680 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 10:32:38,778 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85313.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:32:41,052 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 +2023-02-06 10:32:50,699 INFO [train.py:901] (3/4) Epoch 11, batch 4500, loss[loss=0.2046, simple_loss=0.2856, pruned_loss=0.0618, over 8082.00 frames. ], tot_loss[loss=0.2328, simple_loss=0.3085, pruned_loss=0.07856, over 1613839.88 frames. ], batch size: 21, lr: 6.95e-03, grad_scale: 8.0 +2023-02-06 10:33:03,395 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.456e+02 2.629e+02 3.227e+02 4.085e+02 1.162e+03, threshold=6.455e+02, percent-clipped=2.0 +2023-02-06 10:33:11,826 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5369, 1.8618, 1.8697, 1.1460, 1.9502, 1.2774, 0.4158, 1.6965], + device='cuda:3'), covar=tensor([0.0306, 0.0186, 0.0169, 0.0299, 0.0251, 0.0572, 0.0521, 0.0166], + device='cuda:3'), in_proj_covar=tensor([0.0385, 0.0321, 0.0268, 0.0378, 0.0307, 0.0467, 0.0353, 0.0351], + device='cuda:3'), out_proj_covar=tensor([1.1011e-04, 8.9668e-05, 7.5148e-05, 1.0646e-04, 8.7415e-05, 1.4280e-04, + 1.0103e-04, 1.0028e-04], device='cuda:3') +2023-02-06 10:33:15,850 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 10:33:16,063 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85366.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:33:26,538 INFO [train.py:901] (3/4) Epoch 11, batch 4550, loss[loss=0.2868, simple_loss=0.3475, pruned_loss=0.1131, over 8274.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.3101, pruned_loss=0.07982, over 1613786.88 frames. ], batch size: 23, lr: 6.95e-03, grad_scale: 8.0 +2023-02-06 10:33:33,518 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85391.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:33:59,382 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85428.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:34:01,129 INFO [train.py:901] (3/4) Epoch 11, batch 4600, loss[loss=0.2304, simple_loss=0.3158, pruned_loss=0.07247, over 7980.00 frames. ], tot_loss[loss=0.237, simple_loss=0.3118, pruned_loss=0.08108, over 1612574.38 frames. ], batch size: 21, lr: 6.95e-03, grad_scale: 8.0 +2023-02-06 10:34:11,935 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85446.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:34:13,789 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.836e+02 2.573e+02 3.214e+02 4.149e+02 1.527e+03, threshold=6.427e+02, percent-clipped=2.0 +2023-02-06 10:34:28,035 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85469.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:34:33,581 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85477.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:34:36,251 INFO [train.py:901] (3/4) Epoch 11, batch 4650, loss[loss=0.2225, simple_loss=0.2976, pruned_loss=0.07372, over 8079.00 frames. ], tot_loss[loss=0.237, simple_loss=0.3115, pruned_loss=0.08119, over 1613811.09 frames. ], batch size: 21, lr: 6.94e-03, grad_scale: 8.0 +2023-02-06 10:34:50,386 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85501.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:34:56,543 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 10:35:11,112 INFO [train.py:901] (3/4) Epoch 11, batch 4700, loss[loss=0.248, simple_loss=0.3213, pruned_loss=0.08729, over 8497.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3111, pruned_loss=0.08107, over 1612232.69 frames. ], batch size: 26, lr: 6.94e-03, grad_scale: 8.0 +2023-02-06 10:35:12,710 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85533.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:35:22,559 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85548.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:35:23,053 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.022e+02 2.812e+02 3.491e+02 4.674e+02 1.006e+03, threshold=6.983e+02, percent-clipped=9.0 +2023-02-06 10:35:27,296 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4854, 1.8595, 3.0988, 1.2602, 2.3331, 1.9380, 1.5759, 2.1265], + device='cuda:3'), covar=tensor([0.1665, 0.2307, 0.0642, 0.3848, 0.1470, 0.2734, 0.1926, 0.2046], + device='cuda:3'), in_proj_covar=tensor([0.0488, 0.0518, 0.0532, 0.0579, 0.0618, 0.0555, 0.0475, 0.0612], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 10:35:30,014 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85558.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:35:45,765 INFO [train.py:901] (3/4) Epoch 11, batch 4750, loss[loss=0.1673, simple_loss=0.2438, pruned_loss=0.0454, over 7420.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.3101, pruned_loss=0.08076, over 1605123.72 frames. ], batch size: 17, lr: 6.94e-03, grad_scale: 8.0 +2023-02-06 10:35:47,965 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85584.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:35:53,311 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85592.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:36:09,830 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 10:36:11,809 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 10:36:13,760 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 +2023-02-06 10:36:20,650 INFO [train.py:901] (3/4) Epoch 11, batch 4800, loss[loss=0.2464, simple_loss=0.3201, pruned_loss=0.08637, over 8583.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.3121, pruned_loss=0.08241, over 1606744.24 frames. ], batch size: 34, lr: 6.94e-03, grad_scale: 8.0 +2023-02-06 10:36:28,782 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85643.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:36:32,777 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.811e+02 2.628e+02 3.255e+02 4.281e+02 8.051e+02, threshold=6.510e+02, percent-clipped=3.0 +2023-02-06 10:36:40,045 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0 +2023-02-06 10:36:55,320 INFO [train.py:901] (3/4) Epoch 11, batch 4850, loss[loss=0.2268, simple_loss=0.2994, pruned_loss=0.07712, over 7420.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3109, pruned_loss=0.08126, over 1609447.42 frames. ], batch size: 17, lr: 6.94e-03, grad_scale: 8.0 +2023-02-06 10:36:57,569 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85684.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:37:01,303 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 10:37:11,264 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6885, 1.6425, 3.0818, 1.2533, 2.1832, 3.3483, 3.3623, 2.8650], + device='cuda:3'), covar=tensor([0.1080, 0.1437, 0.0393, 0.2137, 0.0888, 0.0282, 0.0656, 0.0664], + device='cuda:3'), in_proj_covar=tensor([0.0260, 0.0294, 0.0256, 0.0290, 0.0268, 0.0234, 0.0336, 0.0288], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 10:37:15,342 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85709.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:37:30,196 INFO [train.py:901] (3/4) Epoch 11, batch 4900, loss[loss=0.2581, simple_loss=0.3326, pruned_loss=0.09178, over 8195.00 frames. ], tot_loss[loss=0.2374, simple_loss=0.3116, pruned_loss=0.0816, over 1613422.44 frames. ], batch size: 23, lr: 6.93e-03, grad_scale: 8.0 +2023-02-06 10:37:36,032 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 +2023-02-06 10:37:42,889 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.857e+02 2.544e+02 3.151e+02 4.004e+02 8.063e+02, threshold=6.301e+02, percent-clipped=5.0 +2023-02-06 10:38:04,659 INFO [train.py:901] (3/4) Epoch 11, batch 4950, loss[loss=0.1985, simple_loss=0.2771, pruned_loss=0.05993, over 7915.00 frames. ], tot_loss[loss=0.2373, simple_loss=0.3114, pruned_loss=0.08162, over 1613507.59 frames. ], batch size: 20, lr: 6.93e-03, grad_scale: 8.0 +2023-02-06 10:38:11,395 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85790.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:38:39,606 INFO [train.py:901] (3/4) Epoch 11, batch 5000, loss[loss=0.2056, simple_loss=0.2863, pruned_loss=0.06244, over 7633.00 frames. ], tot_loss[loss=0.2363, simple_loss=0.3104, pruned_loss=0.08105, over 1609294.71 frames. ], batch size: 19, lr: 6.93e-03, grad_scale: 8.0 +2023-02-06 10:38:46,527 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85840.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:38:49,824 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85845.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:38:51,925 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85848.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:38:52,299 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.597e+02 2.585e+02 3.219e+02 4.097e+02 8.363e+02, threshold=6.438e+02, percent-clipped=6.0 +2023-02-06 10:39:03,550 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85865.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:39:08,904 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85873.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:39:13,945 INFO [train.py:901] (3/4) Epoch 11, batch 5050, loss[loss=0.221, simple_loss=0.2949, pruned_loss=0.0736, over 8089.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.311, pruned_loss=0.08161, over 1606055.54 frames. ], batch size: 21, lr: 6.93e-03, grad_scale: 8.0 +2023-02-06 10:39:21,269 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85892.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:39:22,012 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85893.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:39:30,144 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85905.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:39:39,821 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 10:39:48,590 INFO [train.py:901] (3/4) Epoch 11, batch 5100, loss[loss=0.2648, simple_loss=0.3397, pruned_loss=0.09497, over 8730.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3108, pruned_loss=0.0813, over 1611685.29 frames. ], batch size: 30, lr: 6.93e-03, grad_scale: 8.0 +2023-02-06 10:40:00,815 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85948.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 10:40:01,252 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.570e+02 3.113e+02 3.980e+02 6.838e+02, threshold=6.226e+02, percent-clipped=2.0 +2023-02-06 10:40:08,941 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85960.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:40:19,557 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85975.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:40:23,488 INFO [train.py:901] (3/4) Epoch 11, batch 5150, loss[loss=0.2563, simple_loss=0.3262, pruned_loss=0.09324, over 8364.00 frames. ], tot_loss[loss=0.2368, simple_loss=0.311, pruned_loss=0.08137, over 1607387.19 frames. ], batch size: 24, lr: 6.92e-03, grad_scale: 8.0 +2023-02-06 10:40:27,616 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85987.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:40:31,743 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4315, 2.1928, 3.2873, 2.6117, 2.7069, 2.4128, 1.8056, 1.7264], + device='cuda:3'), covar=tensor([0.4352, 0.4518, 0.1300, 0.2592, 0.2418, 0.2452, 0.2045, 0.4559], + device='cuda:3'), in_proj_covar=tensor([0.0886, 0.0864, 0.0734, 0.0834, 0.0938, 0.0796, 0.0701, 0.0770], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 10:40:42,274 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86007.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:40:51,394 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86020.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:40:59,435 INFO [train.py:901] (3/4) Epoch 11, batch 5200, loss[loss=0.2473, simple_loss=0.3123, pruned_loss=0.09116, over 7545.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.3105, pruned_loss=0.08083, over 1611305.09 frames. ], batch size: 18, lr: 6.92e-03, grad_scale: 16.0 +2023-02-06 10:41:12,340 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.535e+02 2.648e+02 3.082e+02 3.913e+02 1.007e+03, threshold=6.165e+02, percent-clipped=5.0 +2023-02-06 10:41:27,743 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86070.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:41:35,327 INFO [train.py:901] (3/4) Epoch 11, batch 5250, loss[loss=0.2176, simple_loss=0.3038, pruned_loss=0.06574, over 8317.00 frames. ], tot_loss[loss=0.2365, simple_loss=0.3107, pruned_loss=0.08116, over 1612051.62 frames. ], batch size: 25, lr: 6.92e-03, grad_scale: 16.0 +2023-02-06 10:41:39,715 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86087.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:41:40,963 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 10:41:50,755 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86102.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:42:10,643 INFO [train.py:901] (3/4) Epoch 11, batch 5300, loss[loss=0.2434, simple_loss=0.3248, pruned_loss=0.08099, over 8347.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.3105, pruned_loss=0.08097, over 1616006.95 frames. ], batch size: 26, lr: 6.92e-03, grad_scale: 16.0 +2023-02-06 10:42:23,760 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.835e+02 2.570e+02 3.118e+02 4.195e+02 8.045e+02, threshold=6.237e+02, percent-clipped=4.0 +2023-02-06 10:42:32,856 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86161.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:42:46,495 INFO [train.py:901] (3/4) Epoch 11, batch 5350, loss[loss=0.2545, simple_loss=0.3225, pruned_loss=0.09326, over 7929.00 frames. ], tot_loss[loss=0.2355, simple_loss=0.3097, pruned_loss=0.08064, over 1612806.94 frames. ], batch size: 20, lr: 6.92e-03, grad_scale: 16.0 +2023-02-06 10:42:50,720 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86186.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:43:12,424 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86216.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:43:22,276 INFO [train.py:901] (3/4) Epoch 11, batch 5400, loss[loss=0.254, simple_loss=0.3277, pruned_loss=0.0902, over 8243.00 frames. ], tot_loss[loss=0.2348, simple_loss=0.3096, pruned_loss=0.08001, over 1618087.00 frames. ], batch size: 22, lr: 6.91e-03, grad_scale: 16.0 +2023-02-06 10:43:26,576 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86237.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:43:29,461 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86241.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:43:34,673 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.524e+02 2.471e+02 3.223e+02 4.268e+02 9.619e+02, threshold=6.446e+02, percent-clipped=7.0 +2023-02-06 10:43:44,504 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86263.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:43:45,132 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86264.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:43:57,388 INFO [train.py:901] (3/4) Epoch 11, batch 5450, loss[loss=0.2291, simple_loss=0.2953, pruned_loss=0.08144, over 7248.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.3089, pruned_loss=0.07979, over 1616216.42 frames. ], batch size: 16, lr: 6.91e-03, grad_scale: 16.0 +2023-02-06 10:44:03,068 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86288.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:44:05,819 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86292.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 10:44:25,136 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86319.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:44:34,081 INFO [train.py:901] (3/4) Epoch 11, batch 5500, loss[loss=0.2835, simple_loss=0.3676, pruned_loss=0.09975, over 8588.00 frames. ], tot_loss[loss=0.2331, simple_loss=0.3079, pruned_loss=0.07916, over 1613226.73 frames. ], batch size: 31, lr: 6.91e-03, grad_scale: 16.0 +2023-02-06 10:44:34,720 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 10:44:39,101 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 +2023-02-06 10:44:40,912 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5287, 1.4304, 2.3091, 1.2206, 2.1145, 2.4792, 2.5651, 2.1028], + device='cuda:3'), covar=tensor([0.0890, 0.1184, 0.0474, 0.1960, 0.0666, 0.0429, 0.0725, 0.0807], + device='cuda:3'), in_proj_covar=tensor([0.0263, 0.0297, 0.0258, 0.0289, 0.0271, 0.0236, 0.0338, 0.0291], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 10:44:46,144 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.790e+02 2.494e+02 3.013e+02 3.770e+02 8.759e+02, threshold=6.025e+02, percent-clipped=3.0 +2023-02-06 10:44:48,463 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86352.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:44:52,690 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86358.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:44:56,794 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86364.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:45:09,176 INFO [train.py:901] (3/4) Epoch 11, batch 5550, loss[loss=0.2619, simple_loss=0.3453, pruned_loss=0.08926, over 8326.00 frames. ], tot_loss[loss=0.2357, simple_loss=0.3102, pruned_loss=0.08056, over 1612774.67 frames. ], batch size: 26, lr: 6.91e-03, grad_scale: 16.0 +2023-02-06 10:45:10,778 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86383.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:45:27,859 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86407.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 10:45:33,154 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86414.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:45:44,325 INFO [train.py:901] (3/4) Epoch 11, batch 5600, loss[loss=0.3065, simple_loss=0.3603, pruned_loss=0.1263, over 8513.00 frames. ], tot_loss[loss=0.236, simple_loss=0.3104, pruned_loss=0.08079, over 1613212.06 frames. ], batch size: 49, lr: 6.91e-03, grad_scale: 16.0 +2023-02-06 10:45:44,405 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86431.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:45:46,520 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86434.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:45:57,241 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.591e+02 2.527e+02 3.003e+02 3.802e+02 9.548e+02, threshold=6.005e+02, percent-clipped=4.0 +2023-02-06 10:46:03,247 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86458.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:46:06,613 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86463.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:46:17,346 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86479.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:46:18,539 INFO [train.py:901] (3/4) Epoch 11, batch 5650, loss[loss=0.2277, simple_loss=0.307, pruned_loss=0.07418, over 8254.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3118, pruned_loss=0.08117, over 1615966.06 frames. ], batch size: 24, lr: 6.90e-03, grad_scale: 16.0 +2023-02-06 10:46:39,891 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 10:46:52,899 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86529.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:46:53,988 INFO [train.py:901] (3/4) Epoch 11, batch 5700, loss[loss=0.2452, simple_loss=0.3246, pruned_loss=0.08293, over 8452.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3113, pruned_loss=0.08096, over 1619823.43 frames. ], batch size: 27, lr: 6.90e-03, grad_scale: 16.0 +2023-02-06 10:47:04,208 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86546.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:47:06,030 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.688e+02 2.473e+02 3.032e+02 3.837e+02 8.433e+02, threshold=6.065e+02, percent-clipped=5.0 +2023-02-06 10:47:13,106 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0048, 1.6344, 1.3373, 1.6056, 1.3242, 1.1586, 1.2548, 1.3635], + device='cuda:3'), covar=tensor([0.1018, 0.0368, 0.1107, 0.0455, 0.0642, 0.1334, 0.0839, 0.0679], + device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0233, 0.0314, 0.0296, 0.0304, 0.0322, 0.0341, 0.0306], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 10:47:28,587 INFO [train.py:901] (3/4) Epoch 11, batch 5750, loss[loss=0.2138, simple_loss=0.2903, pruned_loss=0.06865, over 7539.00 frames. ], tot_loss[loss=0.2363, simple_loss=0.3112, pruned_loss=0.08067, over 1619279.43 frames. ], batch size: 18, lr: 6.90e-03, grad_scale: 16.0 +2023-02-06 10:47:40,235 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86598.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:47:42,134 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 10:47:47,613 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86608.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:47:47,754 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86608.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:48:03,810 INFO [train.py:901] (3/4) Epoch 11, batch 5800, loss[loss=0.2144, simple_loss=0.2794, pruned_loss=0.07475, over 7191.00 frames. ], tot_loss[loss=0.236, simple_loss=0.3105, pruned_loss=0.0808, over 1615149.33 frames. ], batch size: 16, lr: 6.90e-03, grad_scale: 16.0 +2023-02-06 10:48:05,413 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86633.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:48:17,063 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.802e+02 2.625e+02 3.434e+02 4.363e+02 1.044e+03, threshold=6.867e+02, percent-clipped=16.0 +2023-02-06 10:48:26,847 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86663.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 10:48:39,354 INFO [train.py:901] (3/4) Epoch 11, batch 5850, loss[loss=0.2511, simple_loss=0.3309, pruned_loss=0.08564, over 7966.00 frames. ], tot_loss[loss=0.235, simple_loss=0.3097, pruned_loss=0.08015, over 1616679.73 frames. ], batch size: 21, lr: 6.90e-03, grad_scale: 16.0 +2023-02-06 10:48:44,304 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86688.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 10:48:45,663 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86690.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:49:02,117 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86715.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:49:07,944 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86723.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:49:13,225 INFO [train.py:901] (3/4) Epoch 11, batch 5900, loss[loss=0.2882, simple_loss=0.3503, pruned_loss=0.113, over 8516.00 frames. ], tot_loss[loss=0.2363, simple_loss=0.3109, pruned_loss=0.08083, over 1614196.58 frames. ], batch size: 26, lr: 6.89e-03, grad_scale: 16.0 +2023-02-06 10:49:16,684 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86735.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:49:25,725 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.878e+02 2.650e+02 3.002e+02 3.837e+02 8.505e+02, threshold=6.004e+02, percent-clipped=1.0 +2023-02-06 10:49:33,361 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86760.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:49:48,257 INFO [train.py:901] (3/4) Epoch 11, batch 5950, loss[loss=0.1979, simple_loss=0.2751, pruned_loss=0.06039, over 7535.00 frames. ], tot_loss[loss=0.2369, simple_loss=0.3115, pruned_loss=0.08116, over 1616564.47 frames. ], batch size: 18, lr: 6.89e-03, grad_scale: 16.0 +2023-02-06 10:49:51,983 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86785.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:50:01,806 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.33 vs. limit=5.0 +2023-02-06 10:50:03,643 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86802.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:50:03,789 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86802.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:50:05,873 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-06 10:50:06,871 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86807.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:50:09,091 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86810.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:50:18,541 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6134, 1.3920, 1.7146, 1.3763, 0.8372, 1.4256, 1.4424, 1.4769], + device='cuda:3'), covar=tensor([0.0541, 0.1236, 0.1696, 0.1342, 0.0606, 0.1468, 0.0735, 0.0609], + device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0153, 0.0192, 0.0159, 0.0104, 0.0164, 0.0116, 0.0137], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:3') +2023-02-06 10:50:20,479 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86827.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:50:22,934 INFO [train.py:901] (3/4) Epoch 11, batch 6000, loss[loss=0.278, simple_loss=0.3408, pruned_loss=0.1076, over 7980.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3119, pruned_loss=0.08115, over 1619792.49 frames. ], batch size: 21, lr: 6.89e-03, grad_scale: 16.0 +2023-02-06 10:50:22,934 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 10:50:31,615 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3859, 1.7238, 2.6999, 1.2412, 1.9670, 1.6964, 1.5291, 1.8898], + device='cuda:3'), covar=tensor([0.1796, 0.2731, 0.0709, 0.4337, 0.1793, 0.3026, 0.2039, 0.2408], + device='cuda:3'), in_proj_covar=tensor([0.0490, 0.0522, 0.0538, 0.0586, 0.0620, 0.0555, 0.0473, 0.0616], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 10:50:35,330 INFO [train.py:935] (3/4) Epoch 11, validation: loss=0.1887, simple_loss=0.2887, pruned_loss=0.04439, over 944034.00 frames. +2023-02-06 10:50:35,330 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 10:50:36,208 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86832.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:50:39,275 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-06 10:50:47,363 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.618e+02 2.431e+02 2.934e+02 3.566e+02 7.044e+02, threshold=5.869e+02, percent-clipped=5.0 +2023-02-06 10:51:10,322 INFO [train.py:901] (3/4) Epoch 11, batch 6050, loss[loss=0.2104, simple_loss=0.2995, pruned_loss=0.06066, over 8313.00 frames. ], tot_loss[loss=0.2379, simple_loss=0.3125, pruned_loss=0.08161, over 1613929.42 frames. ], batch size: 25, lr: 6.89e-03, grad_scale: 16.0 +2023-02-06 10:51:20,527 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86896.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:51:35,573 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86917.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:51:39,098 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86922.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:51:45,227 INFO [train.py:901] (3/4) Epoch 11, batch 6100, loss[loss=0.2352, simple_loss=0.3167, pruned_loss=0.0768, over 8368.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3112, pruned_loss=0.08101, over 1611675.43 frames. ], batch size: 24, lr: 6.89e-03, grad_scale: 16.0 +2023-02-06 10:51:53,491 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86942.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:51:58,237 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.604e+02 3.114e+02 3.901e+02 9.212e+02, threshold=6.229e+02, percent-clipped=4.0 +2023-02-06 10:52:06,857 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 10:52:06,982 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.6514, 1.6099, 5.7067, 2.1371, 5.1500, 4.7941, 5.3158, 5.1630], + device='cuda:3'), covar=tensor([0.0362, 0.4588, 0.0318, 0.3392, 0.0892, 0.0797, 0.0386, 0.0451], + device='cuda:3'), in_proj_covar=tensor([0.0495, 0.0584, 0.0590, 0.0543, 0.0612, 0.0523, 0.0515, 0.0586], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 10:52:19,169 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86979.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:52:20,335 INFO [train.py:901] (3/4) Epoch 11, batch 6150, loss[loss=0.2263, simple_loss=0.3024, pruned_loss=0.07508, over 8505.00 frames. ], tot_loss[loss=0.235, simple_loss=0.3099, pruned_loss=0.08002, over 1613193.88 frames. ], batch size: 26, lr: 6.88e-03, grad_scale: 16.0 +2023-02-06 10:52:36,922 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87004.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:52:47,759 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87020.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 10:52:55,888 INFO [train.py:901] (3/4) Epoch 11, batch 6200, loss[loss=0.1907, simple_loss=0.2728, pruned_loss=0.05434, over 7830.00 frames. ], tot_loss[loss=0.2335, simple_loss=0.3088, pruned_loss=0.07911, over 1611398.62 frames. ], batch size: 20, lr: 6.88e-03, grad_scale: 16.0 +2023-02-06 10:53:07,899 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.654e+02 2.592e+02 3.192e+02 4.476e+02 1.804e+03, threshold=6.384e+02, percent-clipped=5.0 +2023-02-06 10:53:14,457 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87057.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:53:30,917 INFO [train.py:901] (3/4) Epoch 11, batch 6250, loss[loss=0.2075, simple_loss=0.2852, pruned_loss=0.06491, over 7941.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.3091, pruned_loss=0.07968, over 1607958.52 frames. ], batch size: 20, lr: 6.88e-03, grad_scale: 16.0 +2023-02-06 10:54:06,554 INFO [train.py:901] (3/4) Epoch 11, batch 6300, loss[loss=0.25, simple_loss=0.3219, pruned_loss=0.08901, over 8080.00 frames. ], tot_loss[loss=0.235, simple_loss=0.3098, pruned_loss=0.0801, over 1610455.68 frames. ], batch size: 21, lr: 6.88e-03, grad_scale: 16.0 +2023-02-06 10:54:11,435 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7790, 2.0552, 3.4982, 1.4377, 2.6832, 2.2522, 1.7813, 2.3896], + device='cuda:3'), covar=tensor([0.1497, 0.2208, 0.0694, 0.3731, 0.1396, 0.2581, 0.1727, 0.2207], + device='cuda:3'), in_proj_covar=tensor([0.0480, 0.0516, 0.0532, 0.0576, 0.0610, 0.0551, 0.0467, 0.0610], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 10:54:17,459 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8675, 1.5376, 3.3349, 1.3863, 2.2188, 3.5407, 3.6327, 3.0300], + device='cuda:3'), covar=tensor([0.1123, 0.1520, 0.0298, 0.1894, 0.0949, 0.0260, 0.0460, 0.0646], + device='cuda:3'), in_proj_covar=tensor([0.0260, 0.0293, 0.0255, 0.0288, 0.0269, 0.0231, 0.0334, 0.0289], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 10:54:19,292 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.796e+02 2.563e+02 3.017e+02 3.734e+02 8.364e+02, threshold=6.034e+02, percent-clipped=3.0 +2023-02-06 10:54:36,427 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87173.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:54:38,325 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87176.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:54:39,797 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87178.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:54:41,627 INFO [train.py:901] (3/4) Epoch 11, batch 6350, loss[loss=0.2036, simple_loss=0.2933, pruned_loss=0.05697, over 7812.00 frames. ], tot_loss[loss=0.2337, simple_loss=0.3085, pruned_loss=0.07942, over 1607227.91 frames. ], batch size: 20, lr: 6.88e-03, grad_scale: 16.0 +2023-02-06 10:54:41,989 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 +2023-02-06 10:54:53,187 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87198.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:54:57,265 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87203.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:55:16,800 INFO [train.py:901] (3/4) Epoch 11, batch 6400, loss[loss=0.2484, simple_loss=0.3203, pruned_loss=0.08828, over 8466.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.3075, pruned_loss=0.07844, over 1609806.79 frames. ], batch size: 25, lr: 6.87e-03, grad_scale: 16.0 +2023-02-06 10:55:23,182 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87240.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:55:28,864 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87248.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:55:29,363 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.810e+02 2.577e+02 3.020e+02 3.786e+02 7.428e+02, threshold=6.041e+02, percent-clipped=2.0 +2023-02-06 10:55:35,650 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1282, 1.4926, 1.5870, 1.2252, 0.9135, 1.3353, 1.6698, 1.5996], + device='cuda:3'), covar=tensor([0.0481, 0.1192, 0.1709, 0.1440, 0.0614, 0.1537, 0.0685, 0.0594], + device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0153, 0.0192, 0.0158, 0.0104, 0.0163, 0.0116, 0.0137], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:3') +2023-02-06 10:55:42,987 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6526, 2.0305, 2.0985, 1.3532, 2.3074, 1.4173, 0.7383, 1.7557], + device='cuda:3'), covar=tensor([0.0452, 0.0243, 0.0221, 0.0405, 0.0262, 0.0699, 0.0544, 0.0243], + device='cuda:3'), in_proj_covar=tensor([0.0383, 0.0322, 0.0266, 0.0378, 0.0308, 0.0464, 0.0349, 0.0341], + device='cuda:3'), out_proj_covar=tensor([1.0949e-04, 8.9885e-05, 7.4330e-05, 1.0629e-04, 8.7286e-05, 1.4163e-04, + 9.9369e-05, 9.6785e-05], device='cuda:3') +2023-02-06 10:55:51,534 INFO [train.py:901] (3/4) Epoch 11, batch 6450, loss[loss=0.248, simple_loss=0.3329, pruned_loss=0.08152, over 8570.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.3077, pruned_loss=0.07838, over 1614161.53 frames. ], batch size: 31, lr: 6.87e-03, grad_scale: 16.0 +2023-02-06 10:55:59,190 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87291.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:56:14,184 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87313.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:56:27,324 INFO [train.py:901] (3/4) Epoch 11, batch 6500, loss[loss=0.2256, simple_loss=0.3118, pruned_loss=0.06971, over 8481.00 frames. ], tot_loss[loss=0.232, simple_loss=0.3075, pruned_loss=0.07819, over 1612095.63 frames. ], batch size: 28, lr: 6.87e-03, grad_scale: 16.0 +2023-02-06 10:56:32,334 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87338.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:56:39,860 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.474e+02 2.605e+02 3.245e+02 4.169e+02 7.875e+02, threshold=6.489e+02, percent-clipped=5.0 +2023-02-06 10:56:44,216 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87355.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:56:50,424 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87364.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 10:57:01,889 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87380.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:57:02,433 INFO [train.py:901] (3/4) Epoch 11, batch 6550, loss[loss=0.2715, simple_loss=0.3395, pruned_loss=0.1017, over 8561.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3075, pruned_loss=0.07881, over 1612589.78 frames. ], batch size: 31, lr: 6.87e-03, grad_scale: 16.0 +2023-02-06 10:57:17,754 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 10:57:37,071 INFO [train.py:901] (3/4) Epoch 11, batch 6600, loss[loss=0.2212, simple_loss=0.3002, pruned_loss=0.07105, over 8022.00 frames. ], tot_loss[loss=0.2335, simple_loss=0.3086, pruned_loss=0.07918, over 1613197.22 frames. ], batch size: 22, lr: 6.87e-03, grad_scale: 16.0 +2023-02-06 10:57:37,776 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 10:57:50,090 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.510e+02 2.293e+02 2.790e+02 3.732e+02 8.562e+02, threshold=5.581e+02, percent-clipped=1.0 +2023-02-06 10:58:11,506 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87479.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:58:11,542 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87479.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 10:58:12,690 INFO [train.py:901] (3/4) Epoch 11, batch 6650, loss[loss=0.1821, simple_loss=0.2706, pruned_loss=0.04685, over 7975.00 frames. ], tot_loss[loss=0.233, simple_loss=0.3082, pruned_loss=0.07886, over 1617182.27 frames. ], batch size: 21, lr: 6.86e-03, grad_scale: 16.0 +2023-02-06 10:58:13,572 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8878, 1.7780, 1.7289, 1.6112, 1.1398, 1.6939, 2.0674, 1.8912], + device='cuda:3'), covar=tensor([0.0441, 0.1124, 0.1766, 0.1341, 0.0583, 0.1411, 0.0650, 0.0597], + device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0153, 0.0192, 0.0159, 0.0104, 0.0164, 0.0116, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:3') +2023-02-06 10:58:35,737 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.56 vs. limit=5.0 +2023-02-06 10:58:41,528 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87523.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:58:47,482 INFO [train.py:901] (3/4) Epoch 11, batch 6700, loss[loss=0.2897, simple_loss=0.365, pruned_loss=0.1072, over 8676.00 frames. ], tot_loss[loss=0.2335, simple_loss=0.3084, pruned_loss=0.07927, over 1618180.05 frames. ], batch size: 34, lr: 6.86e-03, grad_scale: 16.0 +2023-02-06 10:58:58,345 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87547.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:58:59,467 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.658e+02 2.493e+02 3.158e+02 4.170e+02 8.693e+02, threshold=6.316e+02, percent-clipped=8.0 +2023-02-06 10:59:16,930 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87572.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:59:22,970 INFO [train.py:901] (3/4) Epoch 11, batch 6750, loss[loss=0.2349, simple_loss=0.3057, pruned_loss=0.08199, over 7975.00 frames. ], tot_loss[loss=0.2336, simple_loss=0.3082, pruned_loss=0.07949, over 1613605.21 frames. ], batch size: 21, lr: 6.86e-03, grad_scale: 16.0 +2023-02-06 10:59:30,579 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87592.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:59:37,571 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87602.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:59:40,790 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87607.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:59:41,382 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.2084, 4.1569, 3.8240, 1.9203, 3.7103, 3.7744, 3.7622, 3.4085], + device='cuda:3'), covar=tensor([0.0781, 0.0690, 0.1053, 0.4864, 0.0837, 0.1038, 0.1473, 0.0916], + device='cuda:3'), in_proj_covar=tensor([0.0460, 0.0374, 0.0380, 0.0480, 0.0376, 0.0376, 0.0377, 0.0334], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 10:59:43,424 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87611.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:59:52,295 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87623.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 10:59:56,922 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 10:59:57,597 INFO [train.py:901] (3/4) Epoch 11, batch 6800, loss[loss=0.2119, simple_loss=0.2833, pruned_loss=0.07026, over 7803.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.3069, pruned_loss=0.07895, over 1609012.00 frames. ], batch size: 19, lr: 6.86e-03, grad_scale: 16.0 +2023-02-06 11:00:01,251 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87636.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:00:10,523 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.561e+02 2.375e+02 2.980e+02 3.798e+02 7.616e+02, threshold=5.961e+02, percent-clipped=2.0 +2023-02-06 11:00:32,377 INFO [train.py:901] (3/4) Epoch 11, batch 6850, loss[loss=0.2637, simple_loss=0.3307, pruned_loss=0.09838, over 7811.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.308, pruned_loss=0.07934, over 1606045.10 frames. ], batch size: 20, lr: 6.86e-03, grad_scale: 16.0 +2023-02-06 11:00:45,155 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 11:00:50,748 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87707.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:01:01,874 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87724.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:01:06,421 INFO [train.py:901] (3/4) Epoch 11, batch 6900, loss[loss=0.1973, simple_loss=0.2801, pruned_loss=0.05723, over 7431.00 frames. ], tot_loss[loss=0.2346, simple_loss=0.3095, pruned_loss=0.07987, over 1611507.17 frames. ], batch size: 17, lr: 6.86e-03, grad_scale: 16.0 +2023-02-06 11:01:10,025 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87735.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:01:19,188 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 2.628e+02 3.043e+02 4.130e+02 7.700e+02, threshold=6.086e+02, percent-clipped=2.0 +2023-02-06 11:01:26,844 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87760.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 11:01:41,614 INFO [train.py:901] (3/4) Epoch 11, batch 6950, loss[loss=0.2524, simple_loss=0.323, pruned_loss=0.09093, over 8358.00 frames. ], tot_loss[loss=0.234, simple_loss=0.3091, pruned_loss=0.07946, over 1614634.32 frames. ], batch size: 24, lr: 6.85e-03, grad_scale: 16.0 +2023-02-06 11:01:52,562 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 11:02:11,618 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87823.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:02:16,819 INFO [train.py:901] (3/4) Epoch 11, batch 7000, loss[loss=0.2325, simple_loss=0.3116, pruned_loss=0.07674, over 8329.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.3101, pruned_loss=0.07982, over 1616561.07 frames. ], batch size: 25, lr: 6.85e-03, grad_scale: 16.0 +2023-02-06 11:02:22,314 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87839.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:02:29,501 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.643e+02 2.548e+02 3.185e+02 4.052e+02 9.283e+02, threshold=6.369e+02, percent-clipped=6.0 +2023-02-06 11:02:39,540 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3190, 1.6900, 4.5116, 1.8880, 2.6639, 5.2775, 5.1678, 4.5062], + device='cuda:3'), covar=tensor([0.1120, 0.1581, 0.0249, 0.1995, 0.0986, 0.0134, 0.0243, 0.0560], + device='cuda:3'), in_proj_covar=tensor([0.0261, 0.0295, 0.0259, 0.0292, 0.0271, 0.0232, 0.0338, 0.0288], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 11:02:41,527 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87867.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:02:42,338 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8939, 1.4508, 1.5565, 1.3416, 0.8683, 1.2632, 1.4806, 1.4542], + device='cuda:3'), covar=tensor([0.0482, 0.1268, 0.1616, 0.1386, 0.0606, 0.1525, 0.0720, 0.0622], + device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0153, 0.0192, 0.0159, 0.0104, 0.0164, 0.0115, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:3') +2023-02-06 11:02:51,623 INFO [train.py:901] (3/4) Epoch 11, batch 7050, loss[loss=0.2642, simple_loss=0.324, pruned_loss=0.1022, over 7920.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.3095, pruned_loss=0.07948, over 1616932.34 frames. ], batch size: 20, lr: 6.85e-03, grad_scale: 16.0 +2023-02-06 11:03:26,705 INFO [train.py:901] (3/4) Epoch 11, batch 7100, loss[loss=0.2036, simple_loss=0.2783, pruned_loss=0.06439, over 7814.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.3082, pruned_loss=0.07859, over 1614051.93 frames. ], batch size: 20, lr: 6.85e-03, grad_scale: 16.0 +2023-02-06 11:03:30,183 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6746, 4.7588, 4.1974, 1.9870, 4.1164, 4.2744, 4.2407, 3.8967], + device='cuda:3'), covar=tensor([0.0780, 0.0586, 0.1066, 0.4993, 0.0800, 0.0723, 0.1374, 0.0805], + device='cuda:3'), in_proj_covar=tensor([0.0466, 0.0376, 0.0384, 0.0484, 0.0379, 0.0377, 0.0380, 0.0337], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 11:03:31,608 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87938.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:03:36,826 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87946.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:03:38,774 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.637e+02 2.732e+02 3.356e+02 4.654e+02 1.650e+03, threshold=6.712e+02, percent-clipped=12.0 +2023-02-06 11:03:40,145 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87951.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:03:48,965 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87963.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:03:51,559 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87967.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:04:00,730 INFO [train.py:901] (3/4) Epoch 11, batch 7150, loss[loss=0.2106, simple_loss=0.3023, pruned_loss=0.05946, over 8284.00 frames. ], tot_loss[loss=0.2326, simple_loss=0.3083, pruned_loss=0.07849, over 1614082.69 frames. ], batch size: 23, lr: 6.85e-03, grad_scale: 16.0 +2023-02-06 11:04:01,630 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87982.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:04:05,830 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87988.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:04:09,952 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6212, 2.6765, 1.7103, 2.1412, 2.1842, 1.4015, 2.1207, 2.2012], + device='cuda:3'), covar=tensor([0.1240, 0.0325, 0.1148, 0.0617, 0.0640, 0.1370, 0.0856, 0.0748], + device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0234, 0.0317, 0.0297, 0.0300, 0.0317, 0.0341, 0.0302], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 11:04:36,664 INFO [train.py:901] (3/4) Epoch 11, batch 7200, loss[loss=0.206, simple_loss=0.2773, pruned_loss=0.06732, over 8097.00 frames. ], tot_loss[loss=0.2326, simple_loss=0.3085, pruned_loss=0.07835, over 1619459.53 frames. ], batch size: 21, lr: 6.84e-03, grad_scale: 32.0 +2023-02-06 11:04:49,434 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.733e+02 2.591e+02 3.086e+02 3.706e+02 9.715e+02, threshold=6.172e+02, percent-clipped=2.0 +2023-02-06 11:04:57,782 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88061.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:05:00,587 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3544, 1.2746, 1.4926, 1.1588, 0.7761, 1.2931, 1.1630, 1.0793], + device='cuda:3'), covar=tensor([0.0518, 0.1204, 0.1699, 0.1406, 0.0588, 0.1475, 0.0696, 0.0645], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0152, 0.0191, 0.0158, 0.0103, 0.0164, 0.0115, 0.0137], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0006], + device='cuda:3') +2023-02-06 11:05:01,290 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88066.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:05:11,823 INFO [train.py:901] (3/4) Epoch 11, batch 7250, loss[loss=0.2104, simple_loss=0.2928, pruned_loss=0.06403, over 8320.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.3097, pruned_loss=0.07939, over 1619329.39 frames. ], batch size: 25, lr: 6.84e-03, grad_scale: 32.0 +2023-02-06 11:05:12,643 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88082.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:05:21,492 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88095.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:05:37,915 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88118.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:05:39,357 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88120.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:05:46,919 INFO [train.py:901] (3/4) Epoch 11, batch 7300, loss[loss=0.2018, simple_loss=0.2902, pruned_loss=0.05665, over 7537.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3088, pruned_loss=0.07886, over 1616693.02 frames. ], batch size: 18, lr: 6.84e-03, grad_scale: 32.0 +2023-02-06 11:06:00,699 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.613e+02 2.414e+02 2.958e+02 3.757e+02 7.369e+02, threshold=5.915e+02, percent-clipped=2.0 +2023-02-06 11:06:22,821 INFO [train.py:901] (3/4) Epoch 11, batch 7350, loss[loss=0.2379, simple_loss=0.3229, pruned_loss=0.07644, over 8522.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.3098, pruned_loss=0.07943, over 1615140.94 frames. ], batch size: 39, lr: 6.84e-03, grad_scale: 32.0 +2023-02-06 11:06:32,251 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 11:06:32,470 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88194.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:06:33,350 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.59 vs. limit=5.0 +2023-02-06 11:06:50,343 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88219.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:06:51,484 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 11:06:58,201 INFO [train.py:901] (3/4) Epoch 11, batch 7400, loss[loss=0.2645, simple_loss=0.3369, pruned_loss=0.09599, over 8507.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.312, pruned_loss=0.08063, over 1616548.88 frames. ], batch size: 26, lr: 6.84e-03, grad_scale: 16.0 +2023-02-06 11:07:03,172 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88238.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:07:11,883 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.780e+02 2.577e+02 3.074e+02 3.691e+02 9.024e+02, threshold=6.148e+02, percent-clipped=4.0 +2023-02-06 11:07:20,903 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88263.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:07:32,879 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 11:07:33,591 INFO [train.py:901] (3/4) Epoch 11, batch 7450, loss[loss=0.21, simple_loss=0.2915, pruned_loss=0.0642, over 7927.00 frames. ], tot_loss[loss=0.2353, simple_loss=0.311, pruned_loss=0.07974, over 1618069.45 frames. ], batch size: 20, lr: 6.83e-03, grad_scale: 16.0 +2023-02-06 11:07:37,182 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1276, 2.4100, 1.9176, 2.9112, 1.3158, 1.6194, 1.7540, 2.5014], + device='cuda:3'), covar=tensor([0.0750, 0.0731, 0.0996, 0.0391, 0.1224, 0.1405, 0.1140, 0.0726], + device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0212, 0.0256, 0.0217, 0.0218, 0.0255, 0.0255, 0.0223], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 11:07:53,288 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88309.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:07:58,566 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88317.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:08:01,839 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88322.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:08:05,927 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3888, 1.9917, 2.8914, 2.3053, 2.6534, 2.1514, 1.7777, 1.3893], + device='cuda:3'), covar=tensor([0.3802, 0.4016, 0.1220, 0.2769, 0.1791, 0.2305, 0.1633, 0.4245], + device='cuda:3'), in_proj_covar=tensor([0.0888, 0.0864, 0.0732, 0.0843, 0.0929, 0.0801, 0.0703, 0.0765], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 11:08:07,741 INFO [train.py:901] (3/4) Epoch 11, batch 7500, loss[loss=0.2512, simple_loss=0.3212, pruned_loss=0.09063, over 7701.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.3098, pruned_loss=0.0792, over 1613576.29 frames. ], batch size: 18, lr: 6.83e-03, grad_scale: 16.0 +2023-02-06 11:08:13,316 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88338.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:08:15,896 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88342.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:08:19,204 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88347.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:08:20,935 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.783e+02 2.827e+02 3.509e+02 4.304e+02 1.282e+03, threshold=7.018e+02, percent-clipped=8.0 +2023-02-06 11:08:29,998 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88363.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:08:42,590 INFO [train.py:901] (3/4) Epoch 11, batch 7550, loss[loss=0.1941, simple_loss=0.2822, pruned_loss=0.053, over 8105.00 frames. ], tot_loss[loss=0.2351, simple_loss=0.3101, pruned_loss=0.07998, over 1611676.33 frames. ], batch size: 23, lr: 6.83e-03, grad_scale: 16.0 +2023-02-06 11:08:46,446 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2023-02-06 11:09:08,829 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5983, 4.5608, 4.1573, 2.1379, 4.0580, 4.0161, 4.2373, 3.6686], + device='cuda:3'), covar=tensor([0.0822, 0.0617, 0.0996, 0.4906, 0.0956, 0.0954, 0.1397, 0.1025], + device='cuda:3'), in_proj_covar=tensor([0.0469, 0.0374, 0.0388, 0.0483, 0.0381, 0.0379, 0.0383, 0.0337], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 11:09:17,377 INFO [train.py:901] (3/4) Epoch 11, batch 7600, loss[loss=0.2204, simple_loss=0.2943, pruned_loss=0.07324, over 7812.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3116, pruned_loss=0.08083, over 1613642.04 frames. ], batch size: 20, lr: 6.83e-03, grad_scale: 16.0 +2023-02-06 11:09:31,032 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.301e+02 2.441e+02 2.975e+02 3.888e+02 6.138e+02, threshold=5.951e+02, percent-clipped=0.0 +2023-02-06 11:09:39,152 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88462.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:09:51,512 INFO [train.py:901] (3/4) Epoch 11, batch 7650, loss[loss=0.2308, simple_loss=0.3088, pruned_loss=0.07641, over 8623.00 frames. ], tot_loss[loss=0.2364, simple_loss=0.3115, pruned_loss=0.08068, over 1614356.49 frames. ], batch size: 34, lr: 6.83e-03, grad_scale: 16.0 +2023-02-06 11:10:26,450 INFO [train.py:901] (3/4) Epoch 11, batch 7700, loss[loss=0.2607, simple_loss=0.3277, pruned_loss=0.09682, over 8460.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3114, pruned_loss=0.08089, over 1615219.94 frames. ], batch size: 25, lr: 6.82e-03, grad_scale: 16.0 +2023-02-06 11:10:39,159 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 11:10:39,709 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.632e+02 2.472e+02 3.053e+02 3.571e+02 8.603e+02, threshold=6.105e+02, percent-clipped=3.0 +2023-02-06 11:10:58,417 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88577.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:11:00,813 INFO [train.py:901] (3/4) Epoch 11, batch 7750, loss[loss=0.1868, simple_loss=0.2576, pruned_loss=0.05798, over 7803.00 frames. ], tot_loss[loss=0.238, simple_loss=0.3124, pruned_loss=0.08181, over 1613220.44 frames. ], batch size: 19, lr: 6.82e-03, grad_scale: 16.0 +2023-02-06 11:11:36,333 INFO [train.py:901] (3/4) Epoch 11, batch 7800, loss[loss=0.2474, simple_loss=0.3277, pruned_loss=0.08353, over 8338.00 frames. ], tot_loss[loss=0.2372, simple_loss=0.3118, pruned_loss=0.08131, over 1612927.92 frames. ], batch size: 25, lr: 6.82e-03, grad_scale: 16.0 +2023-02-06 11:11:48,834 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.685e+02 3.345e+02 4.152e+02 1.012e+03, threshold=6.690e+02, percent-clipped=6.0 +2023-02-06 11:11:50,872 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=88653.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 11:12:09,486 INFO [train.py:901] (3/4) Epoch 11, batch 7850, loss[loss=0.282, simple_loss=0.3406, pruned_loss=0.1117, over 7047.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.311, pruned_loss=0.08072, over 1612579.49 frames. ], batch size: 71, lr: 6.82e-03, grad_scale: 16.0 +2023-02-06 11:12:42,901 INFO [train.py:901] (3/4) Epoch 11, batch 7900, loss[loss=0.2174, simple_loss=0.2964, pruned_loss=0.06923, over 7656.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.3107, pruned_loss=0.0802, over 1612632.21 frames. ], batch size: 19, lr: 6.82e-03, grad_scale: 16.0 +2023-02-06 11:12:55,416 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.727e+02 2.490e+02 3.060e+02 3.735e+02 6.734e+02, threshold=6.120e+02, percent-clipped=1.0 +2023-02-06 11:13:07,262 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88768.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:13:15,806 INFO [train.py:901] (3/4) Epoch 11, batch 7950, loss[loss=0.1753, simple_loss=0.2546, pruned_loss=0.04797, over 7793.00 frames. ], tot_loss[loss=0.235, simple_loss=0.3105, pruned_loss=0.0797, over 1615892.90 frames. ], batch size: 19, lr: 6.81e-03, grad_scale: 16.0 +2023-02-06 11:13:16,456 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-06 11:13:49,354 INFO [train.py:901] (3/4) Epoch 11, batch 8000, loss[loss=0.2674, simple_loss=0.3335, pruned_loss=0.1007, over 7198.00 frames. ], tot_loss[loss=0.2344, simple_loss=0.3098, pruned_loss=0.07948, over 1612802.21 frames. ], batch size: 71, lr: 6.81e-03, grad_scale: 16.0 +2023-02-06 11:13:50,917 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88833.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:14:02,010 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.663e+02 2.603e+02 3.071e+02 3.730e+02 8.421e+02, threshold=6.141e+02, percent-clipped=3.0 +2023-02-06 11:14:07,240 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88858.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:14:22,234 INFO [train.py:901] (3/4) Epoch 11, batch 8050, loss[loss=0.1992, simple_loss=0.2677, pruned_loss=0.06536, over 7547.00 frames. ], tot_loss[loss=0.2314, simple_loss=0.3066, pruned_loss=0.07812, over 1598174.39 frames. ], batch size: 18, lr: 6.81e-03, grad_scale: 16.0 +2023-02-06 11:14:54,629 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 11:14:58,680 INFO [train.py:901] (3/4) Epoch 12, batch 0, loss[loss=0.2195, simple_loss=0.2811, pruned_loss=0.07894, over 7686.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2811, pruned_loss=0.07894, over 7686.00 frames. ], batch size: 18, lr: 6.52e-03, grad_scale: 16.0 +2023-02-06 11:14:58,680 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 11:15:09,781 INFO [train.py:935] (3/4) Epoch 12, validation: loss=0.1897, simple_loss=0.2896, pruned_loss=0.04486, over 944034.00 frames. +2023-02-06 11:15:09,782 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 11:15:23,307 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 11:15:35,202 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.665e+02 2.690e+02 3.540e+02 4.339e+02 7.249e+02, threshold=7.080e+02, percent-clipped=5.0 +2023-02-06 11:15:44,676 INFO [train.py:901] (3/4) Epoch 12, batch 50, loss[loss=0.2552, simple_loss=0.3047, pruned_loss=0.1028, over 7539.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.3161, pruned_loss=0.08526, over 363240.91 frames. ], batch size: 18, lr: 6.52e-03, grad_scale: 16.0 +2023-02-06 11:15:46,145 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6161, 4.6040, 4.1706, 2.0155, 4.1848, 4.2146, 4.1708, 4.0384], + device='cuda:3'), covar=tensor([0.0898, 0.0554, 0.1092, 0.5233, 0.0843, 0.0791, 0.1422, 0.0770], + device='cuda:3'), in_proj_covar=tensor([0.0463, 0.0370, 0.0383, 0.0479, 0.0374, 0.0374, 0.0376, 0.0332], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 11:15:57,435 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 11:16:01,120 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7729, 1.9797, 2.1010, 1.4489, 2.1686, 1.6226, 0.5232, 1.8762], + device='cuda:3'), covar=tensor([0.0297, 0.0184, 0.0157, 0.0273, 0.0229, 0.0514, 0.0547, 0.0134], + device='cuda:3'), in_proj_covar=tensor([0.0388, 0.0327, 0.0272, 0.0382, 0.0315, 0.0474, 0.0358, 0.0353], + device='cuda:3'), out_proj_covar=tensor([1.1061e-04, 9.0904e-05, 7.5933e-05, 1.0699e-04, 8.9075e-05, 1.4421e-04, + 1.0177e-04, 1.0004e-04], device='cuda:3') +2023-02-06 11:16:19,065 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 11:16:19,764 INFO [train.py:901] (3/4) Epoch 12, batch 100, loss[loss=0.2743, simple_loss=0.351, pruned_loss=0.09882, over 8357.00 frames. ], tot_loss[loss=0.2395, simple_loss=0.3143, pruned_loss=0.08239, over 641770.83 frames. ], batch size: 24, lr: 6.52e-03, grad_scale: 16.0 +2023-02-06 11:16:26,515 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89024.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 11:16:32,538 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4058, 3.0644, 2.4566, 4.0021, 1.7122, 2.0324, 2.3172, 3.3624], + device='cuda:3'), covar=tensor([0.0801, 0.0828, 0.0879, 0.0278, 0.1249, 0.1486, 0.1250, 0.0767], + device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0216, 0.0255, 0.0216, 0.0220, 0.0253, 0.0258, 0.0222], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 11:16:33,261 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9067, 2.5040, 4.3484, 1.5502, 3.0707, 2.3745, 1.9405, 2.9502], + device='cuda:3'), covar=tensor([0.1471, 0.1996, 0.0719, 0.3611, 0.1411, 0.2604, 0.1630, 0.2097], + device='cuda:3'), in_proj_covar=tensor([0.0485, 0.0522, 0.0532, 0.0581, 0.0621, 0.0561, 0.0474, 0.0612], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 11:16:40,630 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89045.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:16:43,454 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89049.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:16:43,913 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.547e+02 2.771e+02 3.256e+02 4.152e+02 1.357e+03, threshold=6.512e+02, percent-clipped=1.0 +2023-02-06 11:16:54,736 INFO [train.py:901] (3/4) Epoch 12, batch 150, loss[loss=0.2189, simple_loss=0.3053, pruned_loss=0.0663, over 8289.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.311, pruned_loss=0.08031, over 857116.09 frames. ], batch size: 23, lr: 6.52e-03, grad_scale: 16.0 +2023-02-06 11:17:17,997 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.11 vs. limit=5.0 +2023-02-06 11:17:29,009 INFO [train.py:901] (3/4) Epoch 12, batch 200, loss[loss=0.2425, simple_loss=0.3177, pruned_loss=0.08369, over 8583.00 frames. ], tot_loss[loss=0.2337, simple_loss=0.3095, pruned_loss=0.07898, over 1028140.64 frames. ], batch size: 39, lr: 6.52e-03, grad_scale: 16.0 +2023-02-06 11:17:42,733 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8294, 2.4390, 3.1006, 2.2112, 1.6773, 3.2806, 0.8121, 2.1712], + device='cuda:3'), covar=tensor([0.2246, 0.1440, 0.0415, 0.2049, 0.3504, 0.0324, 0.3434, 0.1699], + device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0168, 0.0101, 0.0212, 0.0253, 0.0104, 0.0164, 0.0163], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 11:17:53,942 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.699e+02 2.712e+02 3.423e+02 4.383e+02 1.008e+03, threshold=6.845e+02, percent-clipped=3.0 +2023-02-06 11:18:03,555 INFO [train.py:901] (3/4) Epoch 12, batch 250, loss[loss=0.2191, simple_loss=0.3131, pruned_loss=0.06259, over 8203.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.311, pruned_loss=0.0794, over 1164819.44 frames. ], batch size: 23, lr: 6.51e-03, grad_scale: 16.0 +2023-02-06 11:18:05,197 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1589, 2.3944, 1.9195, 2.9437, 1.3912, 1.5644, 1.9140, 2.6026], + device='cuda:3'), covar=tensor([0.0657, 0.0802, 0.0921, 0.0408, 0.1069, 0.1426, 0.0984, 0.0656], + device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0215, 0.0254, 0.0214, 0.0218, 0.0252, 0.0258, 0.0222], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 11:18:13,242 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 11:18:17,197 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.30 vs. limit=5.0 +2023-02-06 11:18:20,886 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89187.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:18:22,854 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 11:18:40,051 INFO [train.py:901] (3/4) Epoch 12, batch 300, loss[loss=0.2438, simple_loss=0.3216, pruned_loss=0.08298, over 8200.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.3102, pruned_loss=0.07878, over 1266562.73 frames. ], batch size: 23, lr: 6.51e-03, grad_scale: 16.0 +2023-02-06 11:19:05,002 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.586e+02 2.536e+02 3.052e+02 3.921e+02 6.584e+02, threshold=6.103e+02, percent-clipped=0.0 +2023-02-06 11:19:14,500 INFO [train.py:901] (3/4) Epoch 12, batch 350, loss[loss=0.2452, simple_loss=0.3285, pruned_loss=0.08098, over 8598.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.3102, pruned_loss=0.0791, over 1343223.36 frames. ], batch size: 34, lr: 6.51e-03, grad_scale: 16.0 +2023-02-06 11:19:49,369 INFO [train.py:901] (3/4) Epoch 12, batch 400, loss[loss=0.2435, simple_loss=0.3219, pruned_loss=0.08252, over 8517.00 frames. ], tot_loss[loss=0.2338, simple_loss=0.31, pruned_loss=0.07877, over 1405925.07 frames. ], batch size: 26, lr: 6.51e-03, grad_scale: 16.0 +2023-02-06 11:20:14,268 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.695e+02 2.417e+02 2.965e+02 3.513e+02 5.511e+02, threshold=5.929e+02, percent-clipped=0.0 +2023-02-06 11:20:24,232 INFO [train.py:901] (3/4) Epoch 12, batch 450, loss[loss=0.2592, simple_loss=0.3456, pruned_loss=0.08642, over 8628.00 frames. ], tot_loss[loss=0.2326, simple_loss=0.3091, pruned_loss=0.07806, over 1456342.35 frames. ], batch size: 34, lr: 6.51e-03, grad_scale: 16.0 +2023-02-06 11:20:40,990 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89389.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:20:43,203 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89392.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:20:58,861 INFO [train.py:901] (3/4) Epoch 12, batch 500, loss[loss=0.2457, simple_loss=0.3229, pruned_loss=0.08424, over 8497.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3088, pruned_loss=0.07812, over 1490978.39 frames. ], batch size: 26, lr: 6.51e-03, grad_scale: 16.0 +2023-02-06 11:21:19,380 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89443.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:21:24,108 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.597e+02 2.539e+02 3.031e+02 3.696e+02 8.346e+02, threshold=6.063e+02, percent-clipped=3.0 +2023-02-06 11:21:29,669 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89457.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:21:34,342 INFO [train.py:901] (3/4) Epoch 12, batch 550, loss[loss=0.2301, simple_loss=0.303, pruned_loss=0.07857, over 8598.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.3101, pruned_loss=0.07884, over 1524270.36 frames. ], batch size: 31, lr: 6.50e-03, grad_scale: 16.0 +2023-02-06 11:21:38,942 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-06 11:22:02,591 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89504.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:22:09,201 INFO [train.py:901] (3/4) Epoch 12, batch 600, loss[loss=0.2595, simple_loss=0.3216, pruned_loss=0.09866, over 7821.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.3089, pruned_loss=0.07827, over 1546202.93 frames. ], batch size: 20, lr: 6.50e-03, grad_scale: 16.0 +2023-02-06 11:22:17,854 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89527.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:22:21,115 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89531.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:22:24,200 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 +2023-02-06 11:22:26,581 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 11:22:34,504 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.680e+02 2.630e+02 3.047e+02 3.733e+02 1.036e+03, threshold=6.094e+02, percent-clipped=2.0 +2023-02-06 11:22:44,045 INFO [train.py:901] (3/4) Epoch 12, batch 650, loss[loss=0.2059, simple_loss=0.2787, pruned_loss=0.0665, over 7910.00 frames. ], tot_loss[loss=0.2331, simple_loss=0.309, pruned_loss=0.07864, over 1562382.66 frames. ], batch size: 20, lr: 6.50e-03, grad_scale: 16.0 +2023-02-06 11:22:56,505 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8355, 2.1187, 2.2826, 1.5067, 2.3522, 1.4652, 0.7327, 1.9795], + device='cuda:3'), covar=tensor([0.0390, 0.0208, 0.0146, 0.0352, 0.0252, 0.0579, 0.0630, 0.0176], + device='cuda:3'), in_proj_covar=tensor([0.0390, 0.0326, 0.0274, 0.0381, 0.0315, 0.0470, 0.0355, 0.0351], + device='cuda:3'), out_proj_covar=tensor([1.1118e-04, 9.0633e-05, 7.6560e-05, 1.0674e-04, 8.8918e-05, 1.4283e-04, + 1.0081e-04, 9.8998e-05], device='cuda:3') +2023-02-06 11:23:18,869 INFO [train.py:901] (3/4) Epoch 12, batch 700, loss[loss=0.2446, simple_loss=0.3225, pruned_loss=0.08335, over 8469.00 frames. ], tot_loss[loss=0.2332, simple_loss=0.3091, pruned_loss=0.07863, over 1573933.49 frames. ], batch size: 25, lr: 6.50e-03, grad_scale: 16.0 +2023-02-06 11:23:30,411 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1573, 1.1061, 1.2899, 1.1062, 0.9172, 1.2985, 0.0526, 0.9671], + device='cuda:3'), covar=tensor([0.2563, 0.1823, 0.0598, 0.1319, 0.3719, 0.0622, 0.3230, 0.1634], + device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0171, 0.0102, 0.0217, 0.0258, 0.0107, 0.0166, 0.0169], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 11:23:40,430 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89646.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:23:43,633 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.818e+02 2.821e+02 3.296e+02 4.031e+02 9.579e+02, threshold=6.593e+02, percent-clipped=5.0 +2023-02-06 11:23:51,386 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1943, 1.6239, 4.1432, 1.5804, 2.4900, 4.6174, 4.9666, 3.6526], + device='cuda:3'), covar=tensor([0.1318, 0.1827, 0.0399, 0.2512, 0.1194, 0.0320, 0.0389, 0.0968], + device='cuda:3'), in_proj_covar=tensor([0.0269, 0.0300, 0.0263, 0.0296, 0.0276, 0.0238, 0.0346, 0.0293], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 11:23:53,839 INFO [train.py:901] (3/4) Epoch 12, batch 750, loss[loss=0.2167, simple_loss=0.2898, pruned_loss=0.07185, over 7986.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.3101, pruned_loss=0.07914, over 1586109.63 frames. ], batch size: 21, lr: 6.50e-03, grad_scale: 16.0 +2023-02-06 11:24:11,509 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 11:24:17,640 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89698.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:24:20,275 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 11:24:28,197 INFO [train.py:901] (3/4) Epoch 12, batch 800, loss[loss=0.2443, simple_loss=0.3075, pruned_loss=0.09054, over 7795.00 frames. ], tot_loss[loss=0.2346, simple_loss=0.3104, pruned_loss=0.07942, over 1594438.72 frames. ], batch size: 20, lr: 6.49e-03, grad_scale: 8.0 +2023-02-06 11:24:32,436 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8248, 1.8720, 2.2956, 1.7479, 1.2547, 2.2499, 0.4210, 1.4868], + device='cuda:3'), covar=tensor([0.2377, 0.1771, 0.0440, 0.1949, 0.4031, 0.0473, 0.3415, 0.1999], + device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0171, 0.0103, 0.0218, 0.0259, 0.0107, 0.0167, 0.0169], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 11:24:43,681 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89736.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:24:53,390 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.508e+02 2.628e+02 3.285e+02 4.121e+02 9.349e+02, threshold=6.571e+02, percent-clipped=6.0 +2023-02-06 11:24:59,660 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89760.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:25:02,836 INFO [train.py:901] (3/4) Epoch 12, batch 850, loss[loss=0.2242, simple_loss=0.3061, pruned_loss=0.07114, over 8124.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.31, pruned_loss=0.07925, over 1598303.88 frames. ], batch size: 22, lr: 6.49e-03, grad_scale: 8.0 +2023-02-06 11:25:18,005 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89785.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:25:19,288 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89787.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:25:28,676 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89801.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 11:25:34,025 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2678, 2.7366, 3.1373, 1.4909, 3.1724, 1.8059, 1.5106, 2.2718], + device='cuda:3'), covar=tensor([0.0518, 0.0238, 0.0182, 0.0547, 0.0343, 0.0609, 0.0705, 0.0350], + device='cuda:3'), in_proj_covar=tensor([0.0391, 0.0330, 0.0275, 0.0384, 0.0317, 0.0474, 0.0356, 0.0353], + device='cuda:3'), out_proj_covar=tensor([1.1120e-04, 9.1706e-05, 7.6565e-05, 1.0750e-04, 8.9654e-05, 1.4412e-04, + 1.0104e-04, 9.9689e-05], device='cuda:3') +2023-02-06 11:25:37,790 INFO [train.py:901] (3/4) Epoch 12, batch 900, loss[loss=0.1896, simple_loss=0.2731, pruned_loss=0.05307, over 7419.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.3081, pruned_loss=0.07836, over 1599467.58 frames. ], batch size: 17, lr: 6.49e-03, grad_scale: 8.0 +2023-02-06 11:26:03,293 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.572e+02 2.446e+02 3.021e+02 3.729e+02 6.397e+02, threshold=6.041e+02, percent-clipped=0.0 +2023-02-06 11:26:03,492 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89851.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:26:11,313 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9408, 2.4265, 1.8451, 2.9307, 1.3432, 1.6373, 1.8552, 2.4089], + device='cuda:3'), covar=tensor([0.0847, 0.0802, 0.0929, 0.0376, 0.1214, 0.1487, 0.1201, 0.0783], + device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0213, 0.0255, 0.0215, 0.0218, 0.0252, 0.0258, 0.0218], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 11:26:11,802 INFO [train.py:901] (3/4) Epoch 12, batch 950, loss[loss=0.214, simple_loss=0.2933, pruned_loss=0.06736, over 8248.00 frames. ], tot_loss[loss=0.2329, simple_loss=0.3084, pruned_loss=0.07875, over 1600293.13 frames. ], batch size: 22, lr: 6.49e-03, grad_scale: 8.0 +2023-02-06 11:26:16,404 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=89871.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:26:24,567 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89883.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:26:33,238 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6463, 2.0862, 3.4979, 2.3435, 2.9424, 2.3629, 1.9285, 1.6245], + device='cuda:3'), covar=tensor([0.3956, 0.4685, 0.1283, 0.3046, 0.2112, 0.2317, 0.1732, 0.4852], + device='cuda:3'), in_proj_covar=tensor([0.0895, 0.0876, 0.0743, 0.0853, 0.0936, 0.0806, 0.0708, 0.0772], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 11:26:38,586 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89902.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:26:38,628 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89902.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:26:39,095 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 11:26:46,289 INFO [train.py:901] (3/4) Epoch 12, batch 1000, loss[loss=0.218, simple_loss=0.2997, pruned_loss=0.06814, over 8626.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.3086, pruned_loss=0.07811, over 1606421.94 frames. ], batch size: 34, lr: 6.49e-03, grad_scale: 8.0 +2023-02-06 11:26:47,815 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89916.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:26:48,420 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9029, 2.8816, 3.5416, 2.0510, 1.6141, 3.6204, 0.7300, 2.0845], + device='cuda:3'), covar=tensor([0.2271, 0.1139, 0.0325, 0.2536, 0.4123, 0.0233, 0.3374, 0.1855], + device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0172, 0.0103, 0.0218, 0.0259, 0.0107, 0.0166, 0.0169], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 11:26:55,721 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89927.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:27:11,393 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.442e+02 2.648e+02 3.254e+02 4.081e+02 9.414e+02, threshold=6.507e+02, percent-clipped=7.0 +2023-02-06 11:27:11,438 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 11:27:13,888 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-06 11:27:20,836 INFO [train.py:901] (3/4) Epoch 12, batch 1050, loss[loss=0.252, simple_loss=0.3215, pruned_loss=0.09122, over 7804.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3091, pruned_loss=0.07795, over 1610218.86 frames. ], batch size: 20, lr: 6.49e-03, grad_scale: 8.0 +2023-02-06 11:27:22,382 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.1888, 1.3145, 3.3599, 0.9754, 2.8980, 2.8408, 3.0537, 2.9315], + device='cuda:3'), covar=tensor([0.0748, 0.3676, 0.0802, 0.3505, 0.1475, 0.1007, 0.0695, 0.0837], + device='cuda:3'), in_proj_covar=tensor([0.0501, 0.0586, 0.0593, 0.0543, 0.0620, 0.0528, 0.0520, 0.0592], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 11:27:24,334 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 11:27:35,858 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89986.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:27:56,234 INFO [train.py:901] (3/4) Epoch 12, batch 1100, loss[loss=0.2422, simple_loss=0.3142, pruned_loss=0.08511, over 8109.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3091, pruned_loss=0.07869, over 1608731.16 frames. ], batch size: 23, lr: 6.48e-03, grad_scale: 8.0 +2023-02-06 11:28:04,628 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3516, 2.6791, 1.7758, 2.1158, 2.1074, 1.3839, 2.0033, 1.9898], + device='cuda:3'), covar=tensor([0.1459, 0.0362, 0.1174, 0.0668, 0.0704, 0.1498, 0.0918, 0.0890], + device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0240, 0.0316, 0.0297, 0.0301, 0.0323, 0.0336, 0.0306], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 11:28:16,470 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90042.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:28:22,898 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.744e+02 2.536e+02 3.046e+02 3.976e+02 6.882e+02, threshold=6.092e+02, percent-clipped=1.0 +2023-02-06 11:28:31,021 INFO [train.py:901] (3/4) Epoch 12, batch 1150, loss[loss=0.3008, simple_loss=0.3556, pruned_loss=0.123, over 8374.00 frames. ], tot_loss[loss=0.2323, simple_loss=0.3077, pruned_loss=0.07844, over 1606246.55 frames. ], batch size: 24, lr: 6.48e-03, grad_scale: 4.0 +2023-02-06 11:28:34,438 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 11:28:41,701 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 11:29:01,109 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90107.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:29:03,695 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2404, 3.1180, 2.8948, 1.6450, 2.8758, 2.7797, 2.8689, 2.6597], + device='cuda:3'), covar=tensor([0.1334, 0.0926, 0.1396, 0.4609, 0.1187, 0.1451, 0.1809, 0.1385], + device='cuda:3'), in_proj_covar=tensor([0.0475, 0.0383, 0.0390, 0.0492, 0.0392, 0.0389, 0.0383, 0.0340], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 11:29:05,599 INFO [train.py:901] (3/4) Epoch 12, batch 1200, loss[loss=0.2171, simple_loss=0.2917, pruned_loss=0.07126, over 8024.00 frames. ], tot_loss[loss=0.2351, simple_loss=0.31, pruned_loss=0.08012, over 1605743.99 frames. ], batch size: 22, lr: 6.48e-03, grad_scale: 8.0 +2023-02-06 11:29:19,563 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90132.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:29:30,353 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90148.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:29:32,963 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.678e+02 2.449e+02 3.099e+02 4.282e+02 6.791e+02, threshold=6.197e+02, percent-clipped=4.0 +2023-02-06 11:29:36,564 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90157.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:29:37,303 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90158.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:29:41,831 INFO [train.py:901] (3/4) Epoch 12, batch 1250, loss[loss=0.2557, simple_loss=0.3342, pruned_loss=0.08861, over 8564.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.3092, pruned_loss=0.07956, over 1605837.42 frames. ], batch size: 34, lr: 6.48e-03, grad_scale: 8.0 +2023-02-06 11:29:47,523 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90172.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:29:55,711 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90183.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:30:05,421 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90197.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:30:17,140 INFO [train.py:901] (3/4) Epoch 12, batch 1300, loss[loss=0.2405, simple_loss=0.3268, pruned_loss=0.07712, over 8483.00 frames. ], tot_loss[loss=0.234, simple_loss=0.3096, pruned_loss=0.07922, over 1615102.04 frames. ], batch size: 25, lr: 6.48e-03, grad_scale: 8.0 +2023-02-06 11:30:26,147 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90227.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:30:37,536 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90242.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:30:43,728 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90250.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:30:44,956 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.625e+02 2.408e+02 3.209e+02 4.069e+02 1.568e+03, threshold=6.418e+02, percent-clipped=9.0 +2023-02-06 11:30:47,242 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3950, 4.3374, 3.8743, 2.0545, 3.8447, 3.9102, 3.9359, 3.6049], + device='cuda:3'), covar=tensor([0.0772, 0.0598, 0.1036, 0.4607, 0.0897, 0.0961, 0.1261, 0.0903], + device='cuda:3'), in_proj_covar=tensor([0.0471, 0.0380, 0.0388, 0.0487, 0.0388, 0.0386, 0.0379, 0.0338], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 11:30:53,234 INFO [train.py:901] (3/4) Epoch 12, batch 1350, loss[loss=0.2965, simple_loss=0.3537, pruned_loss=0.1197, over 8445.00 frames. ], tot_loss[loss=0.2348, simple_loss=0.3101, pruned_loss=0.07975, over 1614796.11 frames. ], batch size: 27, lr: 6.47e-03, grad_scale: 8.0 +2023-02-06 11:30:55,570 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90267.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:31:05,020 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8910, 1.5490, 1.8037, 1.6497, 1.0764, 1.6264, 2.0056, 1.9891], + device='cuda:3'), covar=tensor([0.0451, 0.1275, 0.1641, 0.1330, 0.0628, 0.1492, 0.0689, 0.0575], + device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0153, 0.0193, 0.0159, 0.0103, 0.0164, 0.0117, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 11:31:09,729 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3301, 1.4662, 1.3235, 1.8553, 0.7390, 1.2127, 1.2310, 1.5346], + device='cuda:3'), covar=tensor([0.0959, 0.0894, 0.1223, 0.0610, 0.1314, 0.1494, 0.0905, 0.0767], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0212, 0.0257, 0.0215, 0.0217, 0.0250, 0.0257, 0.0217], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 11:31:28,606 INFO [train.py:901] (3/4) Epoch 12, batch 1400, loss[loss=0.2337, simple_loss=0.3122, pruned_loss=0.07756, over 8291.00 frames. ], tot_loss[loss=0.2345, simple_loss=0.3101, pruned_loss=0.0795, over 1618004.21 frames. ], batch size: 23, lr: 6.47e-03, grad_scale: 8.0 +2023-02-06 11:31:36,251 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6528, 1.6784, 2.0206, 1.6633, 1.0561, 2.0353, 0.2286, 1.4061], + device='cuda:3'), covar=tensor([0.2775, 0.1568, 0.0615, 0.1746, 0.4281, 0.0538, 0.3405, 0.1631], + device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0172, 0.0104, 0.0219, 0.0256, 0.0107, 0.0164, 0.0168], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 11:31:47,922 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90342.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:31:54,618 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.507e+02 2.398e+02 2.808e+02 3.540e+02 8.131e+02, threshold=5.617e+02, percent-clipped=1.0 +2023-02-06 11:32:03,601 INFO [train.py:901] (3/4) Epoch 12, batch 1450, loss[loss=0.2197, simple_loss=0.3076, pruned_loss=0.06589, over 8501.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.31, pruned_loss=0.07908, over 1618588.43 frames. ], batch size: 28, lr: 6.47e-03, grad_scale: 8.0 +2023-02-06 11:32:07,901 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90369.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:32:08,430 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 11:32:13,169 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6164, 1.4240, 3.1170, 1.1312, 2.1186, 3.3477, 3.5268, 2.6100], + device='cuda:3'), covar=tensor([0.1393, 0.1783, 0.0490, 0.2592, 0.1245, 0.0375, 0.0566, 0.0978], + device='cuda:3'), in_proj_covar=tensor([0.0267, 0.0296, 0.0260, 0.0293, 0.0275, 0.0236, 0.0346, 0.0292], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 11:32:21,875 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5853, 2.1110, 3.4062, 1.3438, 2.5734, 2.0639, 1.6916, 2.6150], + device='cuda:3'), covar=tensor([0.1668, 0.2243, 0.0731, 0.4018, 0.1417, 0.2684, 0.1819, 0.1872], + device='cuda:3'), in_proj_covar=tensor([0.0487, 0.0519, 0.0533, 0.0582, 0.0622, 0.0558, 0.0473, 0.0618], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 11:32:38,148 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90413.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:32:38,592 INFO [train.py:901] (3/4) Epoch 12, batch 1500, loss[loss=0.2205, simple_loss=0.3019, pruned_loss=0.06957, over 8473.00 frames. ], tot_loss[loss=0.2338, simple_loss=0.3098, pruned_loss=0.07889, over 1615565.12 frames. ], batch size: 25, lr: 6.47e-03, grad_scale: 8.0 +2023-02-06 11:32:55,146 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90438.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:33:04,320 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.331e+02 2.462e+02 2.993e+02 3.898e+02 9.256e+02, threshold=5.985e+02, percent-clipped=2.0 +2023-02-06 11:33:12,494 INFO [train.py:901] (3/4) Epoch 12, batch 1550, loss[loss=0.2241, simple_loss=0.3129, pruned_loss=0.06758, over 8423.00 frames. ], tot_loss[loss=0.2323, simple_loss=0.3081, pruned_loss=0.07829, over 1609254.41 frames. ], batch size: 27, lr: 6.47e-03, grad_scale: 8.0 +2023-02-06 11:33:21,457 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90477.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:33:33,067 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90492.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:33:48,747 INFO [train.py:901] (3/4) Epoch 12, batch 1600, loss[loss=0.2512, simple_loss=0.3299, pruned_loss=0.08624, over 8616.00 frames. ], tot_loss[loss=0.2304, simple_loss=0.3065, pruned_loss=0.07721, over 1605711.43 frames. ], batch size: 34, lr: 6.47e-03, grad_scale: 8.0 +2023-02-06 11:33:48,918 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90514.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:34:15,464 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.379e+02 2.712e+02 3.378e+02 4.197e+02 8.231e+02, threshold=6.755e+02, percent-clipped=6.0 +2023-02-06 11:34:23,534 INFO [train.py:901] (3/4) Epoch 12, batch 1650, loss[loss=0.2356, simple_loss=0.3181, pruned_loss=0.0766, over 8482.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3077, pruned_loss=0.07735, over 1609752.09 frames. ], batch size: 49, lr: 6.46e-03, grad_scale: 8.0 +2023-02-06 11:34:23,750 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1996, 1.1375, 1.3108, 1.1736, 0.9551, 1.3239, 0.0892, 1.0164], + device='cuda:3'), covar=tensor([0.2276, 0.1757, 0.0709, 0.1297, 0.4060, 0.0675, 0.3034, 0.1684], + device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0172, 0.0104, 0.0219, 0.0256, 0.0108, 0.0165, 0.0168], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 11:34:29,142 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1393, 2.5278, 2.9058, 1.3837, 3.0622, 1.7367, 1.4227, 2.1150], + device='cuda:3'), covar=tensor([0.0564, 0.0238, 0.0225, 0.0561, 0.0324, 0.0594, 0.0630, 0.0349], + device='cuda:3'), in_proj_covar=tensor([0.0394, 0.0333, 0.0277, 0.0386, 0.0317, 0.0475, 0.0358, 0.0356], + device='cuda:3'), out_proj_covar=tensor([1.1205e-04, 9.2375e-05, 7.7212e-05, 1.0820e-04, 8.9284e-05, 1.4384e-04, + 1.0188e-04, 1.0050e-04], device='cuda:3') +2023-02-06 11:34:44,125 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90594.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:34:47,058 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90598.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:34:49,232 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5017, 2.1400, 3.5194, 1.3764, 2.5639, 1.9125, 1.7765, 2.3335], + device='cuda:3'), covar=tensor([0.1609, 0.1872, 0.0614, 0.3545, 0.1386, 0.2761, 0.1642, 0.2169], + device='cuda:3'), in_proj_covar=tensor([0.0484, 0.0518, 0.0532, 0.0580, 0.0617, 0.0556, 0.0470, 0.0612], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 11:34:53,826 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90607.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:34:58,209 INFO [train.py:901] (3/4) Epoch 12, batch 1700, loss[loss=0.2525, simple_loss=0.3328, pruned_loss=0.08615, over 8572.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.3081, pruned_loss=0.07816, over 1603162.81 frames. ], batch size: 31, lr: 6.46e-03, grad_scale: 8.0 +2023-02-06 11:35:04,330 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90623.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:35:24,539 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.545e+02 2.486e+02 2.952e+02 3.646e+02 6.764e+02, threshold=5.904e+02, percent-clipped=1.0 +2023-02-06 11:35:33,331 INFO [train.py:901] (3/4) Epoch 12, batch 1750, loss[loss=0.1939, simple_loss=0.2663, pruned_loss=0.06082, over 7794.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.3085, pruned_loss=0.07816, over 1607538.91 frames. ], batch size: 19, lr: 6.46e-03, grad_scale: 8.0 +2023-02-06 11:35:39,571 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1715, 1.8014, 1.3939, 1.6481, 1.5867, 1.2521, 1.4467, 1.5092], + device='cuda:3'), covar=tensor([0.0855, 0.0344, 0.0825, 0.0385, 0.0466, 0.0952, 0.0642, 0.0564], + device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0244, 0.0320, 0.0302, 0.0305, 0.0328, 0.0344, 0.0312], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 11:36:04,662 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90709.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:36:07,401 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90713.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:36:08,038 INFO [train.py:901] (3/4) Epoch 12, batch 1800, loss[loss=0.2526, simple_loss=0.3236, pruned_loss=0.0908, over 8305.00 frames. ], tot_loss[loss=0.2326, simple_loss=0.3088, pruned_loss=0.0782, over 1609135.16 frames. ], batch size: 23, lr: 6.46e-03, grad_scale: 8.0 +2023-02-06 11:36:28,888 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8310, 3.6785, 2.3756, 2.7113, 2.9928, 1.9271, 2.6544, 2.9083], + device='cuda:3'), covar=tensor([0.1794, 0.0356, 0.1101, 0.0786, 0.0626, 0.1362, 0.1084, 0.1011], + device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0244, 0.0321, 0.0303, 0.0305, 0.0328, 0.0344, 0.0312], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 11:36:35,309 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.841e+02 2.625e+02 3.119e+02 3.569e+02 7.012e+02, threshold=6.239e+02, percent-clipped=2.0 +2023-02-06 11:36:43,322 INFO [train.py:901] (3/4) Epoch 12, batch 1850, loss[loss=0.2562, simple_loss=0.3279, pruned_loss=0.09221, over 8517.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.3085, pruned_loss=0.07811, over 1611524.55 frames. ], batch size: 39, lr: 6.46e-03, grad_scale: 8.0 +2023-02-06 11:37:17,708 INFO [train.py:901] (3/4) Epoch 12, batch 1900, loss[loss=0.1762, simple_loss=0.2491, pruned_loss=0.05158, over 7684.00 frames. ], tot_loss[loss=0.2314, simple_loss=0.3081, pruned_loss=0.07739, over 1615083.77 frames. ], batch size: 18, lr: 6.46e-03, grad_scale: 8.0 +2023-02-06 11:37:22,472 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90821.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:37:27,323 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90828.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:37:38,160 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.04 vs. limit=5.0 +2023-02-06 11:37:44,436 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.381e+02 2.569e+02 3.031e+02 3.632e+02 7.649e+02, threshold=6.063e+02, percent-clipped=2.0 +2023-02-06 11:37:47,231 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 11:37:48,699 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=90858.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:37:50,847 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.0510, 2.7015, 3.5246, 2.0702, 1.7102, 3.4099, 0.6147, 2.0542], + device='cuda:3'), covar=tensor([0.1488, 0.1255, 0.0322, 0.2327, 0.4032, 0.0378, 0.3964, 0.2233], + device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0172, 0.0104, 0.0218, 0.0255, 0.0108, 0.0165, 0.0169], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 11:37:52,154 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90863.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:37:52,633 INFO [train.py:901] (3/4) Epoch 12, batch 1950, loss[loss=0.1937, simple_loss=0.2712, pruned_loss=0.05812, over 7514.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3072, pruned_loss=0.07741, over 1612941.91 frames. ], batch size: 18, lr: 6.45e-03, grad_scale: 8.0 +2023-02-06 11:37:55,453 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90867.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:37:59,332 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 11:38:10,328 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90888.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:38:19,030 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 11:38:27,978 INFO [train.py:901] (3/4) Epoch 12, batch 2000, loss[loss=0.3009, simple_loss=0.349, pruned_loss=0.1264, over 7081.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3073, pruned_loss=0.07733, over 1617756.95 frames. ], batch size: 72, lr: 6.45e-03, grad_scale: 8.0 +2023-02-06 11:38:43,400 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90936.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:38:54,924 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.754e+02 2.641e+02 3.163e+02 4.034e+02 9.087e+02, threshold=6.326e+02, percent-clipped=9.0 +2023-02-06 11:39:02,896 INFO [train.py:901] (3/4) Epoch 12, batch 2050, loss[loss=0.1899, simple_loss=0.2707, pruned_loss=0.05452, over 7241.00 frames. ], tot_loss[loss=0.2319, simple_loss=0.3082, pruned_loss=0.07779, over 1618323.00 frames. ], batch size: 16, lr: 6.45e-03, grad_scale: 8.0 +2023-02-06 11:39:03,757 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90965.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:39:09,926 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90973.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:39:21,822 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90990.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:39:27,126 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9301, 1.4286, 1.5764, 1.3308, 0.8767, 1.3736, 1.5269, 1.3470], + device='cuda:3'), covar=tensor([0.0486, 0.1240, 0.1687, 0.1390, 0.0590, 0.1545, 0.0710, 0.0645], + device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0153, 0.0190, 0.0159, 0.0103, 0.0162, 0.0116, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 11:39:34,308 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.63 vs. limit=2.0 +2023-02-06 11:39:38,682 INFO [train.py:901] (3/4) Epoch 12, batch 2100, loss[loss=0.2121, simple_loss=0.2963, pruned_loss=0.06389, over 8190.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.3074, pruned_loss=0.07706, over 1620459.31 frames. ], batch size: 23, lr: 6.45e-03, grad_scale: 8.0 +2023-02-06 11:40:04,173 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.877e+02 2.659e+02 3.265e+02 4.247e+02 8.349e+02, threshold=6.531e+02, percent-clipped=2.0 +2023-02-06 11:40:12,109 INFO [train.py:901] (3/4) Epoch 12, batch 2150, loss[loss=0.219, simple_loss=0.3069, pruned_loss=0.06559, over 8458.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.3079, pruned_loss=0.07753, over 1619251.18 frames. ], batch size: 25, lr: 6.45e-03, grad_scale: 8.0 +2023-02-06 11:40:26,819 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91084.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:40:44,045 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91109.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:40:47,223 INFO [train.py:901] (3/4) Epoch 12, batch 2200, loss[loss=0.1672, simple_loss=0.2493, pruned_loss=0.04256, over 7929.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.3061, pruned_loss=0.07642, over 1618347.49 frames. ], batch size: 20, lr: 6.44e-03, grad_scale: 8.0 +2023-02-06 11:41:13,716 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.817e+02 2.751e+02 3.546e+02 4.173e+02 9.054e+02, threshold=7.092e+02, percent-clipped=3.0 +2023-02-06 11:41:21,758 INFO [train.py:901] (3/4) Epoch 12, batch 2250, loss[loss=0.2154, simple_loss=0.2958, pruned_loss=0.06756, over 8091.00 frames. ], tot_loss[loss=0.2299, simple_loss=0.3062, pruned_loss=0.07681, over 1616142.74 frames. ], batch size: 21, lr: 6.44e-03, grad_scale: 8.0 +2023-02-06 11:41:41,125 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91192.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:41:54,522 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91211.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:41:56,410 INFO [train.py:901] (3/4) Epoch 12, batch 2300, loss[loss=0.1988, simple_loss=0.2662, pruned_loss=0.06576, over 7817.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.3078, pruned_loss=0.07825, over 1614617.83 frames. ], batch size: 20, lr: 6.44e-03, grad_scale: 8.0 +2023-02-06 11:41:58,521 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91217.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:42:07,362 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91229.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:42:23,427 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.568e+02 2.635e+02 3.142e+02 4.194e+02 9.102e+02, threshold=6.284e+02, percent-clipped=2.0 +2023-02-06 11:42:24,994 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91254.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:42:31,681 INFO [train.py:901] (3/4) Epoch 12, batch 2350, loss[loss=0.2164, simple_loss=0.3036, pruned_loss=0.06466, over 8675.00 frames. ], tot_loss[loss=0.2311, simple_loss=0.3073, pruned_loss=0.07747, over 1613314.46 frames. ], batch size: 34, lr: 6.44e-03, grad_scale: 8.0 +2023-02-06 11:42:57,737 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91303.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:43:05,701 INFO [train.py:901] (3/4) Epoch 12, batch 2400, loss[loss=0.1972, simple_loss=0.2702, pruned_loss=0.06212, over 7430.00 frames. ], tot_loss[loss=0.2314, simple_loss=0.3073, pruned_loss=0.07782, over 1612965.58 frames. ], batch size: 17, lr: 6.44e-03, grad_scale: 8.0 +2023-02-06 11:43:14,307 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91326.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:43:32,246 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.793e+02 2.547e+02 3.046e+02 3.774e+02 7.420e+02, threshold=6.092e+02, percent-clipped=3.0 +2023-02-06 11:43:35,851 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.6764, 1.3500, 3.9265, 1.3822, 3.3287, 3.2033, 3.5120, 3.3263], + device='cuda:3'), covar=tensor([0.0783, 0.4432, 0.0584, 0.3844, 0.1262, 0.1006, 0.0716, 0.0854], + device='cuda:3'), in_proj_covar=tensor([0.0498, 0.0580, 0.0588, 0.0536, 0.0609, 0.0524, 0.0518, 0.0583], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 11:43:41,055 INFO [train.py:901] (3/4) Epoch 12, batch 2450, loss[loss=0.1805, simple_loss=0.2678, pruned_loss=0.04654, over 8029.00 frames. ], tot_loss[loss=0.2316, simple_loss=0.3068, pruned_loss=0.07819, over 1606146.66 frames. ], batch size: 22, lr: 6.44e-03, grad_scale: 8.0 +2023-02-06 11:44:06,913 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6209, 1.9626, 2.0254, 1.2314, 2.2008, 1.4680, 0.4702, 1.8566], + device='cuda:3'), covar=tensor([0.0402, 0.0248, 0.0197, 0.0396, 0.0244, 0.0688, 0.0633, 0.0186], + device='cuda:3'), in_proj_covar=tensor([0.0397, 0.0332, 0.0280, 0.0384, 0.0317, 0.0473, 0.0362, 0.0359], + device='cuda:3'), out_proj_covar=tensor([1.1284e-04, 9.1987e-05, 7.7832e-05, 1.0725e-04, 8.9322e-05, 1.4282e-04, + 1.0273e-04, 1.0109e-04], device='cuda:3') +2023-02-06 11:44:11,801 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4701, 2.0938, 2.9892, 2.3800, 2.7899, 2.2855, 1.8690, 1.5290], + device='cuda:3'), covar=tensor([0.3880, 0.3774, 0.1264, 0.2729, 0.1837, 0.2213, 0.1733, 0.4100], + device='cuda:3'), in_proj_covar=tensor([0.0883, 0.0871, 0.0740, 0.0850, 0.0937, 0.0804, 0.0704, 0.0765], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 11:44:15,047 INFO [train.py:901] (3/4) Epoch 12, batch 2500, loss[loss=0.2654, simple_loss=0.329, pruned_loss=0.1009, over 8593.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.307, pruned_loss=0.07871, over 1604798.23 frames. ], batch size: 49, lr: 6.43e-03, grad_scale: 8.0 +2023-02-06 11:44:41,736 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.863e+02 2.576e+02 3.186e+02 4.386e+02 8.083e+02, threshold=6.372e+02, percent-clipped=11.0 +2023-02-06 11:44:50,272 INFO [train.py:901] (3/4) Epoch 12, batch 2550, loss[loss=0.2294, simple_loss=0.2995, pruned_loss=0.07966, over 7976.00 frames. ], tot_loss[loss=0.2328, simple_loss=0.3076, pruned_loss=0.07904, over 1606017.66 frames. ], batch size: 21, lr: 6.43e-03, grad_scale: 8.0 +2023-02-06 11:45:24,404 INFO [train.py:901] (3/4) Epoch 12, batch 2600, loss[loss=0.1621, simple_loss=0.2467, pruned_loss=0.03873, over 7555.00 frames. ], tot_loss[loss=0.2326, simple_loss=0.308, pruned_loss=0.07863, over 1609568.04 frames. ], batch size: 18, lr: 6.43e-03, grad_scale: 8.0 +2023-02-06 11:45:50,006 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.921e+02 2.880e+02 3.430e+02 4.544e+02 8.443e+02, threshold=6.860e+02, percent-clipped=9.0 +2023-02-06 11:45:56,378 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91560.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:45:58,883 INFO [train.py:901] (3/4) Epoch 12, batch 2650, loss[loss=0.2537, simple_loss=0.3233, pruned_loss=0.0921, over 8089.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3089, pruned_loss=0.07886, over 1615121.60 frames. ], batch size: 21, lr: 6.43e-03, grad_scale: 8.0 +2023-02-06 11:46:11,921 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91582.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:46:29,376 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91607.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:46:33,855 INFO [train.py:901] (3/4) Epoch 12, batch 2700, loss[loss=0.2168, simple_loss=0.3025, pruned_loss=0.06555, over 8126.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.3081, pruned_loss=0.07867, over 1611848.66 frames. ], batch size: 22, lr: 6.43e-03, grad_scale: 8.0 +2023-02-06 11:46:35,604 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 +2023-02-06 11:46:55,953 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91647.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:46:59,292 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.823e+02 2.691e+02 3.205e+02 3.908e+02 7.628e+02, threshold=6.410e+02, percent-clipped=2.0 +2023-02-06 11:47:08,035 INFO [train.py:901] (3/4) Epoch 12, batch 2750, loss[loss=0.2977, simple_loss=0.3587, pruned_loss=0.1184, over 7258.00 frames. ], tot_loss[loss=0.2309, simple_loss=0.3073, pruned_loss=0.07722, over 1614581.64 frames. ], batch size: 73, lr: 6.43e-03, grad_scale: 8.0 +2023-02-06 11:47:43,514 INFO [train.py:901] (3/4) Epoch 12, batch 2800, loss[loss=0.2386, simple_loss=0.3065, pruned_loss=0.08533, over 8360.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3091, pruned_loss=0.078, over 1619633.41 frames. ], batch size: 24, lr: 6.42e-03, grad_scale: 8.0 +2023-02-06 11:48:08,849 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.765e+02 2.634e+02 3.181e+02 3.784e+02 9.192e+02, threshold=6.362e+02, percent-clipped=3.0 +2023-02-06 11:48:13,228 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2023-02-06 11:48:15,803 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91762.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:48:16,916 INFO [train.py:901] (3/4) Epoch 12, batch 2850, loss[loss=0.2717, simple_loss=0.332, pruned_loss=0.1057, over 8598.00 frames. ], tot_loss[loss=0.2332, simple_loss=0.3096, pruned_loss=0.07834, over 1619388.01 frames. ], batch size: 34, lr: 6.42e-03, grad_scale: 8.0 +2023-02-06 11:48:52,947 INFO [train.py:901] (3/4) Epoch 12, batch 2900, loss[loss=0.222, simple_loss=0.3172, pruned_loss=0.06341, over 8341.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.3086, pruned_loss=0.07795, over 1615387.24 frames. ], batch size: 26, lr: 6.42e-03, grad_scale: 8.0 +2023-02-06 11:49:18,830 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.727e+02 2.538e+02 3.175e+02 3.875e+02 8.885e+02, threshold=6.349e+02, percent-clipped=4.0 +2023-02-06 11:49:22,155 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 11:49:26,763 INFO [train.py:901] (3/4) Epoch 12, batch 2950, loss[loss=0.1864, simple_loss=0.2533, pruned_loss=0.0597, over 7693.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.3086, pruned_loss=0.07778, over 1616601.37 frames. ], batch size: 18, lr: 6.42e-03, grad_scale: 8.0 +2023-02-06 11:49:49,031 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.58 vs. limit=2.0 +2023-02-06 11:49:54,045 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=91904.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 11:49:57,399 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91909.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:50:00,614 INFO [train.py:901] (3/4) Epoch 12, batch 3000, loss[loss=0.2086, simple_loss=0.2852, pruned_loss=0.06606, over 7967.00 frames. ], tot_loss[loss=0.2329, simple_loss=0.3087, pruned_loss=0.07859, over 1612258.12 frames. ], batch size: 21, lr: 6.42e-03, grad_scale: 8.0 +2023-02-06 11:50:00,614 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 11:50:12,504 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.7803, 3.7573, 3.4949, 1.8087, 3.4380, 3.4474, 3.5115, 3.1931], + device='cuda:3'), covar=tensor([0.1039, 0.0604, 0.0927, 0.5351, 0.0853, 0.0995, 0.1122, 0.0993], + device='cuda:3'), in_proj_covar=tensor([0.0466, 0.0379, 0.0388, 0.0489, 0.0379, 0.0383, 0.0379, 0.0330], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 11:50:13,628 INFO [train.py:935] (3/4) Epoch 12, validation: loss=0.1868, simple_loss=0.2871, pruned_loss=0.04323, over 944034.00 frames. +2023-02-06 11:50:13,628 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 11:50:40,667 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.585e+02 2.361e+02 2.883e+02 3.802e+02 7.578e+02, threshold=5.767e+02, percent-clipped=3.0 +2023-02-06 11:50:49,092 INFO [train.py:901] (3/4) Epoch 12, batch 3050, loss[loss=0.1932, simple_loss=0.2815, pruned_loss=0.05244, over 8243.00 frames. ], tot_loss[loss=0.2319, simple_loss=0.3075, pruned_loss=0.07815, over 1607443.91 frames. ], batch size: 22, lr: 6.41e-03, grad_scale: 8.0 +2023-02-06 11:50:56,133 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91973.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:51:14,056 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91999.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 11:51:25,124 INFO [train.py:901] (3/4) Epoch 12, batch 3100, loss[loss=0.2327, simple_loss=0.3139, pruned_loss=0.07571, over 7964.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3081, pruned_loss=0.07771, over 1609432.02 frames. ], batch size: 21, lr: 6.41e-03, grad_scale: 8.0 +2023-02-06 11:51:28,114 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92018.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:51:28,796 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92019.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:51:30,841 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.0357, 1.6173, 4.1655, 1.8514, 3.7039, 3.5074, 3.7946, 3.6520], + device='cuda:3'), covar=tensor([0.0557, 0.3816, 0.0564, 0.3187, 0.1053, 0.0794, 0.0514, 0.0604], + device='cuda:3'), in_proj_covar=tensor([0.0497, 0.0578, 0.0590, 0.0535, 0.0616, 0.0527, 0.0519, 0.0581], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 11:51:45,786 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92043.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:51:47,899 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5434, 1.9067, 2.1538, 1.1062, 2.2239, 1.2954, 0.6700, 1.6528], + device='cuda:3'), covar=tensor([0.0533, 0.0248, 0.0166, 0.0443, 0.0270, 0.0689, 0.0673, 0.0234], + device='cuda:3'), in_proj_covar=tensor([0.0395, 0.0330, 0.0280, 0.0388, 0.0318, 0.0476, 0.0358, 0.0357], + device='cuda:3'), out_proj_covar=tensor([1.1205e-04, 9.1684e-05, 7.7773e-05, 1.0840e-04, 8.9401e-05, 1.4369e-04, + 1.0151e-04, 1.0049e-04], device='cuda:3') +2023-02-06 11:51:51,734 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.560e+02 2.663e+02 3.347e+02 4.142e+02 7.838e+02, threshold=6.695e+02, percent-clipped=5.0 +2023-02-06 11:52:01,127 INFO [train.py:901] (3/4) Epoch 12, batch 3150, loss[loss=0.3046, simple_loss=0.3541, pruned_loss=0.1275, over 6891.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.3088, pruned_loss=0.07826, over 1610812.40 frames. ], batch size: 72, lr: 6.41e-03, grad_scale: 16.0 +2023-02-06 11:52:35,766 INFO [train.py:901] (3/4) Epoch 12, batch 3200, loss[loss=0.2236, simple_loss=0.3068, pruned_loss=0.07016, over 8329.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3073, pruned_loss=0.07718, over 1609383.88 frames. ], batch size: 26, lr: 6.41e-03, grad_scale: 16.0 +2023-02-06 11:52:55,510 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.20 vs. limit=5.0 +2023-02-06 11:53:02,008 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.744e+02 2.674e+02 3.226e+02 3.971e+02 7.397e+02, threshold=6.453e+02, percent-clipped=3.0 +2023-02-06 11:53:10,361 INFO [train.py:901] (3/4) Epoch 12, batch 3250, loss[loss=0.189, simple_loss=0.2593, pruned_loss=0.05933, over 7540.00 frames. ], tot_loss[loss=0.2305, simple_loss=0.3064, pruned_loss=0.0773, over 1608353.74 frames. ], batch size: 18, lr: 6.41e-03, grad_scale: 16.0 +2023-02-06 11:53:46,148 INFO [train.py:901] (3/4) Epoch 12, batch 3300, loss[loss=0.2335, simple_loss=0.3086, pruned_loss=0.07919, over 8037.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.3058, pruned_loss=0.07665, over 1606857.81 frames. ], batch size: 22, lr: 6.41e-03, grad_scale: 16.0 +2023-02-06 11:53:53,879 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 11:54:11,039 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.662e+02 2.357e+02 2.935e+02 3.680e+02 6.719e+02, threshold=5.870e+02, percent-clipped=1.0 +2023-02-06 11:54:11,767 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92253.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:54:19,108 INFO [train.py:901] (3/4) Epoch 12, batch 3350, loss[loss=0.2548, simple_loss=0.3216, pruned_loss=0.09404, over 8599.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.3057, pruned_loss=0.07691, over 1610332.07 frames. ], batch size: 31, lr: 6.40e-03, grad_scale: 16.0 +2023-02-06 11:54:27,348 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92275.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:54:27,949 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9915, 1.4370, 4.2659, 1.7144, 2.3151, 4.8244, 4.8726, 4.1062], + device='cuda:3'), covar=tensor([0.1407, 0.1963, 0.0319, 0.2338, 0.1303, 0.0217, 0.0471, 0.0718], + device='cuda:3'), in_proj_covar=tensor([0.0270, 0.0299, 0.0265, 0.0297, 0.0280, 0.0238, 0.0351, 0.0294], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 11:54:45,173 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92300.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 11:54:50,801 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3982, 1.5794, 2.3075, 1.2405, 1.6079, 1.6754, 1.4566, 1.5578], + device='cuda:3'), covar=tensor([0.1741, 0.2324, 0.0723, 0.3919, 0.1560, 0.2831, 0.1984, 0.1822], + device='cuda:3'), in_proj_covar=tensor([0.0494, 0.0532, 0.0540, 0.0585, 0.0624, 0.0563, 0.0481, 0.0616], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 11:54:55,335 INFO [train.py:901] (3/4) Epoch 12, batch 3400, loss[loss=0.2145, simple_loss=0.2989, pruned_loss=0.06503, over 8469.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3062, pruned_loss=0.07718, over 1609575.19 frames. ], batch size: 39, lr: 6.40e-03, grad_scale: 16.0 +2023-02-06 11:54:57,465 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92317.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:55:03,718 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4870, 1.8689, 4.3596, 1.8758, 2.4779, 4.8896, 4.9879, 4.2636], + device='cuda:3'), covar=tensor([0.0890, 0.1312, 0.0239, 0.1898, 0.1012, 0.0184, 0.0305, 0.0574], + device='cuda:3'), in_proj_covar=tensor([0.0269, 0.0299, 0.0265, 0.0296, 0.0279, 0.0239, 0.0351, 0.0294], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 11:55:15,994 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92343.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 11:55:21,861 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.667e+02 2.364e+02 2.893e+02 3.659e+02 6.777e+02, threshold=5.785e+02, percent-clipped=2.0 +2023-02-06 11:55:29,921 INFO [train.py:901] (3/4) Epoch 12, batch 3450, loss[loss=0.2144, simple_loss=0.3063, pruned_loss=0.0612, over 8650.00 frames. ], tot_loss[loss=0.2311, simple_loss=0.3074, pruned_loss=0.0774, over 1613524.99 frames. ], batch size: 39, lr: 6.40e-03, grad_scale: 16.0 +2023-02-06 11:55:30,197 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0935, 1.7462, 2.4279, 2.0376, 2.3593, 2.0065, 1.6724, 0.9435], + device='cuda:3'), covar=tensor([0.4231, 0.3776, 0.1238, 0.2391, 0.1697, 0.2240, 0.1655, 0.4000], + device='cuda:3'), in_proj_covar=tensor([0.0892, 0.0872, 0.0733, 0.0847, 0.0936, 0.0805, 0.0706, 0.0767], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 11:55:32,885 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92368.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:56:04,899 INFO [train.py:901] (3/4) Epoch 12, batch 3500, loss[loss=0.2007, simple_loss=0.2799, pruned_loss=0.06077, over 7809.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.3062, pruned_loss=0.07704, over 1612175.71 frames. ], batch size: 19, lr: 6.40e-03, grad_scale: 8.0 +2023-02-06 11:56:18,412 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92432.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:56:29,127 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 11:56:33,046 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.547e+02 2.438e+02 2.928e+02 3.742e+02 8.211e+02, threshold=5.856e+02, percent-clipped=5.0 +2023-02-06 11:56:36,516 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92458.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 11:56:40,251 INFO [train.py:901] (3/4) Epoch 12, batch 3550, loss[loss=0.2419, simple_loss=0.3112, pruned_loss=0.08629, over 6879.00 frames. ], tot_loss[loss=0.2311, simple_loss=0.307, pruned_loss=0.07756, over 1610731.89 frames. ], batch size: 71, lr: 6.40e-03, grad_scale: 8.0 +2023-02-06 11:56:55,154 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0869, 1.2648, 1.1495, 0.6811, 1.2087, 0.9679, 0.1269, 1.2034], + device='cuda:3'), covar=tensor([0.0254, 0.0232, 0.0198, 0.0346, 0.0260, 0.0687, 0.0531, 0.0212], + device='cuda:3'), in_proj_covar=tensor([0.0400, 0.0332, 0.0282, 0.0393, 0.0325, 0.0481, 0.0359, 0.0362], + device='cuda:3'), out_proj_covar=tensor([1.1358e-04, 9.2051e-05, 7.8060e-05, 1.0973e-04, 9.1301e-05, 1.4537e-04, + 1.0191e-04, 1.0198e-04], device='cuda:3') +2023-02-06 11:57:12,359 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 +2023-02-06 11:57:14,696 INFO [train.py:901] (3/4) Epoch 12, batch 3600, loss[loss=0.2259, simple_loss=0.3069, pruned_loss=0.07241, over 8289.00 frames. ], tot_loss[loss=0.232, simple_loss=0.308, pruned_loss=0.07799, over 1613765.00 frames. ], batch size: 23, lr: 6.40e-03, grad_scale: 8.0 +2023-02-06 11:57:39,233 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.6191, 5.7111, 5.0174, 2.1753, 5.1108, 5.2870, 5.2999, 4.9150], + device='cuda:3'), covar=tensor([0.0558, 0.0394, 0.0831, 0.4851, 0.0628, 0.0838, 0.1051, 0.0737], + device='cuda:3'), in_proj_covar=tensor([0.0465, 0.0376, 0.0387, 0.0486, 0.0380, 0.0383, 0.0376, 0.0331], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 11:57:42,455 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.674e+02 2.628e+02 3.055e+02 4.234e+02 9.851e+02, threshold=6.109e+02, percent-clipped=7.0 +2023-02-06 11:57:50,884 INFO [train.py:901] (3/4) Epoch 12, batch 3650, loss[loss=0.2332, simple_loss=0.3152, pruned_loss=0.0756, over 8540.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.308, pruned_loss=0.07751, over 1615221.80 frames. ], batch size: 49, lr: 6.39e-03, grad_scale: 8.0 +2023-02-06 11:58:23,906 INFO [train.py:901] (3/4) Epoch 12, batch 3700, loss[loss=0.2292, simple_loss=0.3092, pruned_loss=0.07458, over 7936.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.308, pruned_loss=0.07779, over 1614774.30 frames. ], batch size: 20, lr: 6.39e-03, grad_scale: 8.0 +2023-02-06 11:58:28,529 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 11:58:31,414 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92624.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:58:44,258 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92643.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:58:48,450 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92649.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:58:50,894 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.564e+02 2.476e+02 3.116e+02 4.152e+02 8.400e+02, threshold=6.233e+02, percent-clipped=9.0 +2023-02-06 11:58:54,573 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.62 vs. limit=2.0 +2023-02-06 11:58:59,706 INFO [train.py:901] (3/4) Epoch 12, batch 3750, loss[loss=0.2503, simple_loss=0.3322, pruned_loss=0.08419, over 8193.00 frames. ], tot_loss[loss=0.2331, simple_loss=0.3096, pruned_loss=0.07829, over 1618219.85 frames. ], batch size: 23, lr: 6.39e-03, grad_scale: 8.0 +2023-02-06 11:59:02,538 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.7153, 4.7171, 4.2409, 2.0082, 4.2838, 4.1806, 4.3835, 3.9339], + device='cuda:3'), covar=tensor([0.0677, 0.0465, 0.0994, 0.4808, 0.0738, 0.0918, 0.1125, 0.0829], + device='cuda:3'), in_proj_covar=tensor([0.0474, 0.0382, 0.0393, 0.0493, 0.0382, 0.0385, 0.0381, 0.0334], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 11:59:17,077 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92688.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:59:33,692 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.9882, 1.9167, 6.0215, 2.2751, 5.4626, 5.0208, 5.5550, 5.5081], + device='cuda:3'), covar=tensor([0.0384, 0.3948, 0.0360, 0.3100, 0.0865, 0.0739, 0.0414, 0.0404], + device='cuda:3'), in_proj_covar=tensor([0.0505, 0.0581, 0.0604, 0.0542, 0.0623, 0.0535, 0.0527, 0.0587], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 11:59:34,461 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92713.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:59:34,944 INFO [train.py:901] (3/4) Epoch 12, batch 3800, loss[loss=0.2279, simple_loss=0.3106, pruned_loss=0.07256, over 8585.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3075, pruned_loss=0.07703, over 1615119.66 frames. ], batch size: 39, lr: 6.39e-03, grad_scale: 8.0 +2023-02-06 11:59:35,191 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92714.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 11:59:52,028 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92738.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 11:59:52,783 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92739.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 12:00:02,126 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.663e+02 2.558e+02 2.972e+02 3.756e+02 9.318e+02, threshold=5.944e+02, percent-clipped=5.0 +2023-02-06 12:00:09,485 INFO [train.py:901] (3/4) Epoch 12, batch 3850, loss[loss=0.1991, simple_loss=0.2744, pruned_loss=0.06191, over 7965.00 frames. ], tot_loss[loss=0.2311, simple_loss=0.308, pruned_loss=0.07714, over 1616262.66 frames. ], batch size: 21, lr: 6.39e-03, grad_scale: 8.0 +2023-02-06 12:00:32,435 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4036, 1.9697, 2.9917, 2.3210, 2.6821, 2.2136, 1.7724, 1.2835], + device='cuda:3'), covar=tensor([0.3965, 0.4014, 0.1178, 0.2837, 0.2038, 0.2369, 0.1739, 0.4474], + device='cuda:3'), in_proj_covar=tensor([0.0897, 0.0870, 0.0728, 0.0852, 0.0935, 0.0804, 0.0705, 0.0765], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 12:00:33,553 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 12:00:45,118 INFO [train.py:901] (3/4) Epoch 12, batch 3900, loss[loss=0.2259, simple_loss=0.3074, pruned_loss=0.07221, over 8439.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3084, pruned_loss=0.07751, over 1621003.44 frames. ], batch size: 27, lr: 6.39e-03, grad_scale: 8.0 +2023-02-06 12:01:08,875 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92849.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:01:11,288 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.559e+02 2.538e+02 2.989e+02 3.922e+02 7.912e+02, threshold=5.979e+02, percent-clipped=3.0 +2023-02-06 12:01:18,864 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6249, 4.6207, 4.1306, 1.9746, 4.1487, 4.1858, 4.3050, 3.8537], + device='cuda:3'), covar=tensor([0.0746, 0.0576, 0.1035, 0.4819, 0.0740, 0.0975, 0.1326, 0.0816], + device='cuda:3'), in_proj_covar=tensor([0.0464, 0.0378, 0.0388, 0.0484, 0.0377, 0.0380, 0.0376, 0.0333], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 12:01:19,449 INFO [train.py:901] (3/4) Epoch 12, batch 3950, loss[loss=0.2519, simple_loss=0.3331, pruned_loss=0.08541, over 8626.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3076, pruned_loss=0.07705, over 1622705.39 frames. ], batch size: 34, lr: 6.38e-03, grad_scale: 8.0 +2023-02-06 12:01:25,440 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9958, 1.5981, 1.7020, 1.2224, 0.9259, 1.4931, 1.8098, 1.5944], + device='cuda:3'), covar=tensor([0.0466, 0.1181, 0.1690, 0.1428, 0.0583, 0.1489, 0.0657, 0.0624], + device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0153, 0.0192, 0.0159, 0.0104, 0.0163, 0.0116, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 12:01:54,556 INFO [train.py:901] (3/4) Epoch 12, batch 4000, loss[loss=0.2248, simple_loss=0.3161, pruned_loss=0.06675, over 8234.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.3061, pruned_loss=0.07648, over 1618385.56 frames. ], batch size: 24, lr: 6.38e-03, grad_scale: 8.0 +2023-02-06 12:01:56,842 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92917.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:02:18,353 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92949.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:02:18,569 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 +2023-02-06 12:02:20,900 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.592e+02 2.373e+02 3.059e+02 3.649e+02 8.513e+02, threshold=6.118e+02, percent-clipped=6.0 +2023-02-06 12:02:28,381 INFO [train.py:901] (3/4) Epoch 12, batch 4050, loss[loss=0.2306, simple_loss=0.3155, pruned_loss=0.07287, over 8103.00 frames. ], tot_loss[loss=0.2304, simple_loss=0.3064, pruned_loss=0.07719, over 1619045.98 frames. ], batch size: 23, lr: 6.38e-03, grad_scale: 8.0 +2023-02-06 12:02:44,203 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=92987.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:02:48,271 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92993.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:03:03,713 INFO [train.py:901] (3/4) Epoch 12, batch 4100, loss[loss=0.2223, simple_loss=0.3026, pruned_loss=0.07103, over 8362.00 frames. ], tot_loss[loss=0.2319, simple_loss=0.3078, pruned_loss=0.07806, over 1617956.59 frames. ], batch size: 24, lr: 6.38e-03, grad_scale: 8.0 +2023-02-06 12:03:13,908 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93028.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:03:21,395 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8073, 2.4988, 3.2132, 1.9499, 1.7003, 3.1587, 0.6211, 1.9671], + device='cuda:3'), covar=tensor([0.2129, 0.1614, 0.0357, 0.2481, 0.4230, 0.0419, 0.3636, 0.1957], + device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0172, 0.0102, 0.0214, 0.0251, 0.0106, 0.0160, 0.0166], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 12:03:30,605 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.685e+02 2.418e+02 3.048e+02 3.757e+02 7.047e+02, threshold=6.097e+02, percent-clipped=3.0 +2023-02-06 12:03:31,419 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93054.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:03:37,929 INFO [train.py:901] (3/4) Epoch 12, batch 4150, loss[loss=0.2526, simple_loss=0.3319, pruned_loss=0.08663, over 8558.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.308, pruned_loss=0.07813, over 1616028.34 frames. ], batch size: 49, lr: 6.38e-03, grad_scale: 8.0 +2023-02-06 12:03:50,874 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93082.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:04:04,576 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93102.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:04:12,238 INFO [train.py:901] (3/4) Epoch 12, batch 4200, loss[loss=0.2157, simple_loss=0.2912, pruned_loss=0.07008, over 8132.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.306, pruned_loss=0.07709, over 1612246.82 frames. ], batch size: 22, lr: 6.38e-03, grad_scale: 8.0 +2023-02-06 12:04:24,958 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 12:04:26,416 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93133.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:04:40,123 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.730e+02 2.563e+02 2.943e+02 3.717e+02 8.503e+02, threshold=5.885e+02, percent-clipped=3.0 +2023-02-06 12:04:47,435 INFO [train.py:901] (3/4) Epoch 12, batch 4250, loss[loss=0.2186, simple_loss=0.2946, pruned_loss=0.07129, over 7970.00 frames. ], tot_loss[loss=0.2331, simple_loss=0.3089, pruned_loss=0.07864, over 1617290.31 frames. ], batch size: 21, lr: 6.37e-03, grad_scale: 8.0 +2023-02-06 12:04:48,801 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 12:05:06,772 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93193.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:05:09,360 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93197.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:05:21,311 INFO [train.py:901] (3/4) Epoch 12, batch 4300, loss[loss=0.234, simple_loss=0.3166, pruned_loss=0.07569, over 8352.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.3087, pruned_loss=0.07829, over 1617002.96 frames. ], batch size: 24, lr: 6.37e-03, grad_scale: 8.0 +2023-02-06 12:05:48,569 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.706e+02 2.718e+02 3.236e+02 4.116e+02 1.260e+03, threshold=6.473e+02, percent-clipped=7.0 +2023-02-06 12:05:54,515 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93261.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:05:56,509 INFO [train.py:901] (3/4) Epoch 12, batch 4350, loss[loss=0.2282, simple_loss=0.301, pruned_loss=0.07767, over 8073.00 frames. ], tot_loss[loss=0.2329, simple_loss=0.3089, pruned_loss=0.07845, over 1615941.44 frames. ], batch size: 21, lr: 6.37e-03, grad_scale: 8.0 +2023-02-06 12:06:15,792 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93293.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:06:16,412 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 12:06:25,937 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93308.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:06:29,909 INFO [train.py:901] (3/4) Epoch 12, batch 4400, loss[loss=0.2405, simple_loss=0.3144, pruned_loss=0.08332, over 7944.00 frames. ], tot_loss[loss=0.2337, simple_loss=0.3095, pruned_loss=0.07897, over 1617044.88 frames. ], batch size: 20, lr: 6.37e-03, grad_scale: 8.0 +2023-02-06 12:06:41,630 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4970, 2.6102, 1.8106, 2.2980, 2.1380, 1.4248, 1.9869, 2.0989], + device='cuda:3'), covar=tensor([0.1290, 0.0355, 0.1016, 0.0546, 0.0622, 0.1339, 0.0928, 0.0851], + device='cuda:3'), in_proj_covar=tensor([0.0341, 0.0234, 0.0313, 0.0296, 0.0297, 0.0323, 0.0341, 0.0301], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 12:06:46,156 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93337.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:06:58,339 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.705e+02 2.551e+02 2.995e+02 3.715e+02 7.484e+02, threshold=5.990e+02, percent-clipped=1.0 +2023-02-06 12:06:58,364 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 12:07:01,783 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93358.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:07:05,646 INFO [train.py:901] (3/4) Epoch 12, batch 4450, loss[loss=0.1933, simple_loss=0.2639, pruned_loss=0.06133, over 7265.00 frames. ], tot_loss[loss=0.2323, simple_loss=0.3081, pruned_loss=0.07826, over 1615678.26 frames. ], batch size: 16, lr: 6.37e-03, grad_scale: 8.0 +2023-02-06 12:07:11,755 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93372.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:07:12,583 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8276, 2.3191, 3.5414, 2.7117, 3.1265, 2.4666, 2.0427, 1.6181], + device='cuda:3'), covar=tensor([0.3547, 0.4274, 0.1233, 0.2613, 0.1900, 0.2177, 0.1596, 0.4622], + device='cuda:3'), in_proj_covar=tensor([0.0886, 0.0868, 0.0725, 0.0847, 0.0927, 0.0798, 0.0700, 0.0762], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 12:07:14,530 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93376.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:07:19,344 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93383.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:07:29,365 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93398.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:07:36,075 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93408.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:07:39,914 INFO [train.py:901] (3/4) Epoch 12, batch 4500, loss[loss=0.2799, simple_loss=0.3513, pruned_loss=0.1043, over 8037.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3075, pruned_loss=0.07791, over 1611494.23 frames. ], batch size: 22, lr: 6.37e-03, grad_scale: 8.0 +2023-02-06 12:07:50,717 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 12:08:06,003 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93452.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:08:06,478 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.678e+02 2.576e+02 3.193e+02 4.187e+02 6.619e+02, threshold=6.386e+02, percent-clipped=4.0 +2023-02-06 12:08:06,708 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93453.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:08:13,869 INFO [train.py:901] (3/4) Epoch 12, batch 4550, loss[loss=0.2153, simple_loss=0.2819, pruned_loss=0.07439, over 7805.00 frames. ], tot_loss[loss=0.2298, simple_loss=0.3059, pruned_loss=0.07686, over 1612718.60 frames. ], batch size: 20, lr: 6.36e-03, grad_scale: 8.0 +2023-02-06 12:08:24,172 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=93477.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:08:24,998 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93478.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:08:31,900 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93487.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:08:49,671 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93513.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:08:50,192 INFO [train.py:901] (3/4) Epoch 12, batch 4600, loss[loss=0.2502, simple_loss=0.3415, pruned_loss=0.07947, over 8521.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3063, pruned_loss=0.07715, over 1612632.35 frames. ], batch size: 28, lr: 6.36e-03, grad_scale: 8.0 +2023-02-06 12:09:16,614 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.735e+02 2.455e+02 3.020e+02 4.052e+02 9.299e+02, threshold=6.041e+02, percent-clipped=5.0 +2023-02-06 12:09:24,869 INFO [train.py:901] (3/4) Epoch 12, batch 4650, loss[loss=0.2691, simple_loss=0.3224, pruned_loss=0.1079, over 7428.00 frames. ], tot_loss[loss=0.2299, simple_loss=0.3061, pruned_loss=0.07688, over 1610803.71 frames. ], batch size: 17, lr: 6.36e-03, grad_scale: 8.0 +2023-02-06 12:09:25,068 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93564.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:09:41,914 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93589.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:09:45,159 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93592.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:09:45,361 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-06 12:09:47,217 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7625, 2.0680, 2.2287, 1.2634, 2.3246, 1.5674, 0.6279, 1.8997], + device='cuda:3'), covar=tensor([0.0426, 0.0209, 0.0158, 0.0437, 0.0236, 0.0666, 0.0651, 0.0189], + device='cuda:3'), in_proj_covar=tensor([0.0398, 0.0332, 0.0277, 0.0394, 0.0322, 0.0481, 0.0361, 0.0363], + device='cuda:3'), out_proj_covar=tensor([1.1275e-04, 9.1637e-05, 7.6628e-05, 1.0989e-04, 9.0315e-05, 1.4519e-04, + 1.0213e-04, 1.0190e-04], device='cuda:3') +2023-02-06 12:09:59,879 INFO [train.py:901] (3/4) Epoch 12, batch 4700, loss[loss=0.2087, simple_loss=0.2738, pruned_loss=0.07179, over 7556.00 frames. ], tot_loss[loss=0.2279, simple_loss=0.304, pruned_loss=0.07588, over 1606825.85 frames. ], batch size: 18, lr: 6.36e-03, grad_scale: 8.0 +2023-02-06 12:10:12,702 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93632.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:10:20,659 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6977, 4.6703, 4.1840, 1.8683, 4.2274, 4.2540, 4.3006, 4.0374], + device='cuda:3'), covar=tensor([0.0752, 0.0618, 0.1073, 0.5621, 0.0790, 0.0889, 0.1487, 0.0743], + device='cuda:3'), in_proj_covar=tensor([0.0472, 0.0389, 0.0395, 0.0493, 0.0387, 0.0390, 0.0384, 0.0336], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 12:10:25,039 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 +2023-02-06 12:10:26,592 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.440e+02 2.370e+02 2.939e+02 3.568e+02 8.447e+02, threshold=5.879e+02, percent-clipped=4.0 +2023-02-06 12:10:29,424 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93657.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:10:33,909 INFO [train.py:901] (3/4) Epoch 12, batch 4750, loss[loss=0.2375, simple_loss=0.3221, pruned_loss=0.07648, over 8330.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.3045, pruned_loss=0.07585, over 1605647.63 frames. ], batch size: 26, lr: 6.36e-03, grad_scale: 8.0 +2023-02-06 12:10:34,129 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93664.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:10:36,648 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4173, 1.4585, 2.4097, 1.2722, 2.3046, 2.5580, 2.6300, 2.1512], + device='cuda:3'), covar=tensor([0.0933, 0.1140, 0.0380, 0.1788, 0.0611, 0.0356, 0.0638, 0.0737], + device='cuda:3'), in_proj_covar=tensor([0.0269, 0.0300, 0.0264, 0.0295, 0.0278, 0.0240, 0.0354, 0.0294], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 12:10:51,743 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93689.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:11:00,451 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 12:11:02,479 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 12:11:04,768 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93708.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:11:09,462 INFO [train.py:901] (3/4) Epoch 12, batch 4800, loss[loss=0.2908, simple_loss=0.3593, pruned_loss=0.1111, over 8505.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3046, pruned_loss=0.07617, over 1603277.35 frames. ], batch size: 28, lr: 6.35e-03, grad_scale: 8.0 +2023-02-06 12:11:23,301 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93733.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:11:24,216 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-06 12:11:30,156 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93743.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:11:36,832 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.527e+02 2.519e+02 2.967e+02 3.635e+02 7.460e+02, threshold=5.934e+02, percent-clipped=2.0 +2023-02-06 12:11:44,120 INFO [train.py:901] (3/4) Epoch 12, batch 4850, loss[loss=0.1997, simple_loss=0.271, pruned_loss=0.06418, over 7794.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.305, pruned_loss=0.07603, over 1607404.29 frames. ], batch size: 19, lr: 6.35e-03, grad_scale: 8.0 +2023-02-06 12:11:47,089 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93768.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:11:47,776 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93769.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:11:53,081 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 12:12:00,547 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93788.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:12:04,712 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93794.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:12:18,639 INFO [train.py:901] (3/4) Epoch 12, batch 4900, loss[loss=0.1961, simple_loss=0.2853, pruned_loss=0.05349, over 8249.00 frames. ], tot_loss[loss=0.2294, simple_loss=0.3056, pruned_loss=0.07655, over 1607564.41 frames. ], batch size: 22, lr: 6.35e-03, grad_scale: 8.0 +2023-02-06 12:12:42,683 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93848.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:12:45,726 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.559e+02 2.389e+02 2.920e+02 3.679e+02 7.315e+02, threshold=5.841e+02, percent-clipped=3.0 +2023-02-06 12:12:53,910 INFO [train.py:901] (3/4) Epoch 12, batch 4950, loss[loss=0.2473, simple_loss=0.3271, pruned_loss=0.08375, over 8353.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.3058, pruned_loss=0.07674, over 1611113.37 frames. ], batch size: 24, lr: 6.35e-03, grad_scale: 8.0 +2023-02-06 12:13:00,036 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93873.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:13:20,749 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4208, 1.5762, 4.5512, 2.0270, 4.0430, 3.8550, 4.1757, 4.1064], + device='cuda:3'), covar=tensor([0.0485, 0.4123, 0.0477, 0.3200, 0.0955, 0.0822, 0.0509, 0.0508], + device='cuda:3'), in_proj_covar=tensor([0.0497, 0.0569, 0.0584, 0.0528, 0.0603, 0.0516, 0.0506, 0.0569], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 12:13:27,059 INFO [train.py:901] (3/4) Epoch 12, batch 5000, loss[loss=0.1851, simple_loss=0.2635, pruned_loss=0.05336, over 7242.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3075, pruned_loss=0.07796, over 1609804.80 frames. ], batch size: 16, lr: 6.35e-03, grad_scale: 8.0 +2023-02-06 12:13:55,385 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.648e+02 2.574e+02 3.082e+02 3.748e+02 7.333e+02, threshold=6.165e+02, percent-clipped=4.0 +2023-02-06 12:13:56,536 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-06 12:14:02,955 INFO [train.py:901] (3/4) Epoch 12, batch 5050, loss[loss=0.2004, simple_loss=0.2695, pruned_loss=0.06565, over 7645.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3068, pruned_loss=0.07783, over 1607569.87 frames. ], batch size: 19, lr: 6.35e-03, grad_scale: 8.0 +2023-02-06 12:14:03,320 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.44 vs. limit=5.0 +2023-02-06 12:14:27,670 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93999.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:14:31,165 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 12:14:38,587 INFO [train.py:901] (3/4) Epoch 12, batch 5100, loss[loss=0.26, simple_loss=0.3308, pruned_loss=0.09461, over 8434.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3065, pruned_loss=0.07751, over 1605829.58 frames. ], batch size: 49, lr: 6.34e-03, grad_scale: 8.0 +2023-02-06 12:15:02,451 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2023-02-06 12:15:04,423 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 12:15:05,345 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.659e+02 2.576e+02 2.962e+02 4.029e+02 5.912e+02, threshold=5.924e+02, percent-clipped=0.0 +2023-02-06 12:15:10,266 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4502, 4.3446, 3.9718, 1.7993, 4.0055, 3.9614, 4.0140, 3.6560], + device='cuda:3'), covar=tensor([0.0680, 0.0561, 0.0895, 0.4722, 0.0749, 0.0914, 0.1185, 0.0819], + device='cuda:3'), in_proj_covar=tensor([0.0463, 0.0379, 0.0389, 0.0486, 0.0382, 0.0386, 0.0380, 0.0334], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 12:15:13,505 INFO [train.py:901] (3/4) Epoch 12, batch 5150, loss[loss=0.2666, simple_loss=0.3305, pruned_loss=0.1014, over 6635.00 frames. ], tot_loss[loss=0.233, simple_loss=0.3082, pruned_loss=0.07896, over 1604655.59 frames. ], batch size: 71, lr: 6.34e-03, grad_scale: 8.0 +2023-02-06 12:15:47,480 INFO [train.py:901] (3/4) Epoch 12, batch 5200, loss[loss=0.2296, simple_loss=0.3024, pruned_loss=0.07836, over 8069.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3087, pruned_loss=0.0789, over 1610807.30 frames. ], batch size: 21, lr: 6.34e-03, grad_scale: 8.0 +2023-02-06 12:15:53,864 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4960, 1.5020, 2.5946, 1.0542, 2.0121, 2.7670, 3.0128, 2.0112], + device='cuda:3'), covar=tensor([0.1469, 0.1676, 0.0588, 0.2856, 0.1066, 0.0564, 0.0760, 0.1338], + device='cuda:3'), in_proj_covar=tensor([0.0267, 0.0297, 0.0262, 0.0292, 0.0276, 0.0238, 0.0349, 0.0293], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 12:15:54,504 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.7350, 1.3157, 3.9497, 1.4108, 3.4511, 3.3229, 3.5423, 3.4434], + device='cuda:3'), covar=tensor([0.0663, 0.3928, 0.0577, 0.3379, 0.1287, 0.0876, 0.0610, 0.0715], + device='cuda:3'), in_proj_covar=tensor([0.0496, 0.0573, 0.0586, 0.0529, 0.0608, 0.0518, 0.0512, 0.0570], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 12:15:59,924 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=94132.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:16:14,667 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.474e+02 2.571e+02 3.074e+02 4.467e+02 8.286e+02, threshold=6.149e+02, percent-clipped=7.0 +2023-02-06 12:16:21,915 INFO [train.py:901] (3/4) Epoch 12, batch 5250, loss[loss=0.2052, simple_loss=0.2734, pruned_loss=0.06854, over 7693.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3085, pruned_loss=0.07828, over 1609630.80 frames. ], batch size: 18, lr: 6.34e-03, grad_scale: 8.0 +2023-02-06 12:16:25,903 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 12:16:57,676 INFO [train.py:901] (3/4) Epoch 12, batch 5300, loss[loss=0.2343, simple_loss=0.3103, pruned_loss=0.07914, over 8370.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3086, pruned_loss=0.0782, over 1612314.35 frames. ], batch size: 24, lr: 6.34e-03, grad_scale: 8.0 +2023-02-06 12:17:13,244 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-06 12:17:15,489 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94241.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 12:17:19,507 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94247.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:17:23,393 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.710e+02 2.628e+02 3.237e+02 4.138e+02 9.258e+02, threshold=6.473e+02, percent-clipped=5.0 +2023-02-06 12:17:31,605 INFO [train.py:901] (3/4) Epoch 12, batch 5350, loss[loss=0.1978, simple_loss=0.283, pruned_loss=0.05631, over 8024.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3076, pruned_loss=0.07704, over 1617347.41 frames. ], batch size: 22, lr: 6.34e-03, grad_scale: 8.0 +2023-02-06 12:17:42,450 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6736, 2.0557, 3.5063, 1.4325, 2.5802, 2.1867, 1.7113, 2.4312], + device='cuda:3'), covar=tensor([0.1620, 0.2132, 0.0659, 0.3846, 0.1454, 0.2568, 0.1734, 0.2098], + device='cuda:3'), in_proj_covar=tensor([0.0487, 0.0521, 0.0532, 0.0582, 0.0614, 0.0555, 0.0475, 0.0613], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 12:17:55,700 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0625, 1.3083, 1.3787, 1.0823, 0.8666, 1.1714, 1.6240, 1.2757], + device='cuda:3'), covar=tensor([0.0588, 0.1898, 0.2682, 0.2081, 0.0793, 0.2273, 0.0897, 0.0888], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0152, 0.0193, 0.0158, 0.0103, 0.0163, 0.0116, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 12:18:05,214 INFO [train.py:901] (3/4) Epoch 12, batch 5400, loss[loss=0.1988, simple_loss=0.2791, pruned_loss=0.05926, over 7929.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.3089, pruned_loss=0.07774, over 1616931.17 frames. ], batch size: 20, lr: 6.33e-03, grad_scale: 8.0 +2023-02-06 12:18:25,503 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=94343.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:18:32,249 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.760e+02 2.485e+02 2.978e+02 4.110e+02 9.009e+02, threshold=5.957e+02, percent-clipped=6.0 +2023-02-06 12:18:39,968 INFO [train.py:901] (3/4) Epoch 12, batch 5450, loss[loss=0.2108, simple_loss=0.2739, pruned_loss=0.0739, over 7711.00 frames. ], tot_loss[loss=0.232, simple_loss=0.3081, pruned_loss=0.07795, over 1612718.89 frames. ], batch size: 18, lr: 6.33e-03, grad_scale: 8.0 +2023-02-06 12:18:40,821 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2339, 2.1901, 1.7267, 2.0089, 1.7514, 1.3726, 1.6521, 1.6649], + device='cuda:3'), covar=tensor([0.1209, 0.0334, 0.1005, 0.0473, 0.0641, 0.1360, 0.0849, 0.0739], + device='cuda:3'), in_proj_covar=tensor([0.0344, 0.0232, 0.0311, 0.0293, 0.0296, 0.0320, 0.0334, 0.0301], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 12:18:41,427 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94366.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:19:12,394 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 12:19:15,761 INFO [train.py:901] (3/4) Epoch 12, batch 5500, loss[loss=0.2448, simple_loss=0.3164, pruned_loss=0.08664, over 8249.00 frames. ], tot_loss[loss=0.2306, simple_loss=0.3072, pruned_loss=0.07707, over 1611766.26 frames. ], batch size: 24, lr: 6.33e-03, grad_scale: 16.0 +2023-02-06 12:19:28,082 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1878, 1.6324, 3.4929, 1.4795, 2.4264, 3.9516, 3.8927, 3.4107], + device='cuda:3'), covar=tensor([0.0868, 0.1419, 0.0317, 0.1866, 0.0909, 0.0202, 0.0417, 0.0532], + device='cuda:3'), in_proj_covar=tensor([0.0267, 0.0297, 0.0260, 0.0291, 0.0274, 0.0235, 0.0347, 0.0290], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 12:19:43,163 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.661e+02 2.406e+02 2.798e+02 3.361e+02 6.650e+02, threshold=5.597e+02, percent-clipped=1.0 +2023-02-06 12:19:45,385 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94458.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:19:46,245 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 12:19:49,116 INFO [train.py:901] (3/4) Epoch 12, batch 5550, loss[loss=0.2703, simple_loss=0.3452, pruned_loss=0.09769, over 8298.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.3068, pruned_loss=0.07668, over 1611619.59 frames. ], batch size: 23, lr: 6.33e-03, grad_scale: 4.0 +2023-02-06 12:20:16,695 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94503.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:20:17,403 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3274, 1.9453, 2.8588, 2.2408, 2.5821, 2.1698, 1.7880, 1.3455], + device='cuda:3'), covar=tensor([0.4125, 0.4154, 0.1268, 0.2807, 0.2025, 0.2317, 0.1728, 0.4355], + device='cuda:3'), in_proj_covar=tensor([0.0893, 0.0874, 0.0730, 0.0856, 0.0945, 0.0804, 0.0705, 0.0767], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 12:20:24,042 INFO [train.py:901] (3/4) Epoch 12, batch 5600, loss[loss=0.2323, simple_loss=0.3148, pruned_loss=0.07492, over 8100.00 frames. ], tot_loss[loss=0.2306, simple_loss=0.3074, pruned_loss=0.07687, over 1612365.68 frames. ], batch size: 23, lr: 6.33e-03, grad_scale: 8.0 +2023-02-06 12:20:35,252 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94528.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:20:54,482 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.995e+02 2.675e+02 3.313e+02 4.214e+02 1.006e+03, threshold=6.626e+02, percent-clipped=7.0 +2023-02-06 12:20:58,134 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1181, 3.8851, 2.2404, 2.7781, 3.0002, 1.6844, 2.7335, 2.8778], + device='cuda:3'), covar=tensor([0.1496, 0.0263, 0.1107, 0.0714, 0.0561, 0.1425, 0.0924, 0.0915], + device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0234, 0.0314, 0.0296, 0.0299, 0.0322, 0.0336, 0.0302], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 12:21:00,666 INFO [train.py:901] (3/4) Epoch 12, batch 5650, loss[loss=0.2094, simple_loss=0.301, pruned_loss=0.05891, over 8326.00 frames. ], tot_loss[loss=0.2314, simple_loss=0.3077, pruned_loss=0.07753, over 1609013.58 frames. ], batch size: 26, lr: 6.33e-03, grad_scale: 8.0 +2023-02-06 12:21:15,258 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=94585.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 12:21:21,330 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 12:21:35,767 INFO [train.py:901] (3/4) Epoch 12, batch 5700, loss[loss=0.1897, simple_loss=0.2637, pruned_loss=0.05784, over 7710.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3072, pruned_loss=0.07762, over 1610659.43 frames. ], batch size: 18, lr: 6.32e-03, grad_scale: 8.0 +2023-02-06 12:22:04,739 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.632e+02 2.327e+02 3.036e+02 3.801e+02 7.493e+02, threshold=6.072e+02, percent-clipped=2.0 +2023-02-06 12:22:10,807 INFO [train.py:901] (3/4) Epoch 12, batch 5750, loss[loss=0.2184, simple_loss=0.3012, pruned_loss=0.06784, over 8494.00 frames. ], tot_loss[loss=0.2313, simple_loss=0.3076, pruned_loss=0.07749, over 1612818.34 frames. ], batch size: 28, lr: 6.32e-03, grad_scale: 8.0 +2023-02-06 12:22:26,193 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 12:22:35,722 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94700.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 12:22:42,476 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=94710.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:22:45,147 INFO [train.py:901] (3/4) Epoch 12, batch 5800, loss[loss=0.2521, simple_loss=0.3381, pruned_loss=0.08303, over 8097.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.3081, pruned_loss=0.07777, over 1620594.85 frames. ], batch size: 23, lr: 6.32e-03, grad_scale: 8.0 +2023-02-06 12:22:45,333 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94714.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:22:53,246 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0457, 2.5360, 3.0082, 1.2213, 3.2041, 1.7252, 1.4095, 1.9940], + device='cuda:3'), covar=tensor([0.0657, 0.0292, 0.0218, 0.0579, 0.0259, 0.0679, 0.0715, 0.0404], + device='cuda:3'), in_proj_covar=tensor([0.0396, 0.0330, 0.0278, 0.0388, 0.0322, 0.0477, 0.0359, 0.0360], + device='cuda:3'), out_proj_covar=tensor([1.1189e-04, 9.0728e-05, 7.6748e-05, 1.0780e-04, 9.0071e-05, 1.4371e-04, + 1.0137e-04, 1.0096e-04], device='cuda:3') +2023-02-06 12:23:02,554 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94739.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:23:10,572 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7012, 2.2224, 3.5906, 2.7971, 3.2809, 2.3575, 1.9010, 1.7302], + device='cuda:3'), covar=tensor([0.3845, 0.4470, 0.1205, 0.2470, 0.1702, 0.2265, 0.1814, 0.4310], + device='cuda:3'), in_proj_covar=tensor([0.0891, 0.0875, 0.0728, 0.0853, 0.0935, 0.0802, 0.0704, 0.0768], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 12:23:13,636 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.691e+02 2.575e+02 3.288e+02 4.021e+02 7.847e+02, threshold=6.576e+02, percent-clipped=2.0 +2023-02-06 12:23:19,964 INFO [train.py:901] (3/4) Epoch 12, batch 5850, loss[loss=0.29, simple_loss=0.36, pruned_loss=0.11, over 8355.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3087, pruned_loss=0.079, over 1617058.94 frames. ], batch size: 24, lr: 6.32e-03, grad_scale: 8.0 +2023-02-06 12:23:33,717 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.1343, 1.7580, 1.8979, 1.7086, 1.1861, 1.8171, 2.2780, 2.2956], + device='cuda:3'), covar=tensor([0.0357, 0.1184, 0.1616, 0.1290, 0.0591, 0.1401, 0.0572, 0.0493], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0151, 0.0190, 0.0157, 0.0103, 0.0160, 0.0113, 0.0137], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 12:23:54,261 INFO [train.py:901] (3/4) Epoch 12, batch 5900, loss[loss=0.3139, simple_loss=0.3649, pruned_loss=0.1315, over 6886.00 frames. ], tot_loss[loss=0.2335, simple_loss=0.309, pruned_loss=0.07899, over 1617019.59 frames. ], batch size: 71, lr: 6.32e-03, grad_scale: 8.0 +2023-02-06 12:24:01,764 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94825.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:24:22,275 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.554e+02 2.604e+02 3.248e+02 4.213e+02 6.479e+02, threshold=6.496e+02, percent-clipped=0.0 +2023-02-06 12:24:28,378 INFO [train.py:901] (3/4) Epoch 12, batch 5950, loss[loss=0.1842, simple_loss=0.2704, pruned_loss=0.049, over 8105.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.308, pruned_loss=0.07824, over 1618281.03 frames. ], batch size: 23, lr: 6.32e-03, grad_scale: 8.0 +2023-02-06 12:24:29,416 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 +2023-02-06 12:25:03,798 INFO [train.py:901] (3/4) Epoch 12, batch 6000, loss[loss=0.2338, simple_loss=0.3091, pruned_loss=0.0792, over 8029.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.3074, pruned_loss=0.07813, over 1611925.32 frames. ], batch size: 22, lr: 6.31e-03, grad_scale: 8.0 +2023-02-06 12:25:03,799 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 12:25:16,946 INFO [train.py:935] (3/4) Epoch 12, validation: loss=0.1862, simple_loss=0.286, pruned_loss=0.04318, over 944034.00 frames. +2023-02-06 12:25:16,947 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 12:25:18,116 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-06 12:25:44,724 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.831e+02 2.442e+02 2.970e+02 3.787e+02 9.017e+02, threshold=5.940e+02, percent-clipped=3.0 +2023-02-06 12:25:45,524 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94956.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 12:25:50,750 INFO [train.py:901] (3/4) Epoch 12, batch 6050, loss[loss=0.2169, simple_loss=0.3009, pruned_loss=0.06643, over 8109.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.3064, pruned_loss=0.07744, over 1612213.20 frames. ], batch size: 23, lr: 6.31e-03, grad_scale: 8.0 +2023-02-06 12:26:02,541 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94981.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 12:26:25,539 INFO [train.py:901] (3/4) Epoch 12, batch 6100, loss[loss=0.2364, simple_loss=0.3154, pruned_loss=0.07867, over 8332.00 frames. ], tot_loss[loss=0.2294, simple_loss=0.3055, pruned_loss=0.07662, over 1609790.52 frames. ], batch size: 26, lr: 6.31e-03, grad_scale: 8.0 +2023-02-06 12:26:50,863 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2291, 1.1348, 1.2510, 1.1681, 0.9671, 1.3085, 0.0377, 0.8575], + device='cuda:3'), covar=tensor([0.2362, 0.1682, 0.0646, 0.1291, 0.3990, 0.0718, 0.3303, 0.1678], + device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0172, 0.0101, 0.0213, 0.0253, 0.0106, 0.0164, 0.0165], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 12:26:54,005 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 12:26:54,667 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.733e+02 2.590e+02 3.216e+02 4.301e+02 8.648e+02, threshold=6.433e+02, percent-clipped=2.0 +2023-02-06 12:27:00,777 INFO [train.py:901] (3/4) Epoch 12, batch 6150, loss[loss=0.1982, simple_loss=0.2824, pruned_loss=0.05701, over 7461.00 frames. ], tot_loss[loss=0.2304, simple_loss=0.3065, pruned_loss=0.07715, over 1612513.01 frames. ], batch size: 17, lr: 6.31e-03, grad_scale: 8.0 +2023-02-06 12:27:12,227 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95081.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:27:29,627 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95106.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:27:34,680 INFO [train.py:901] (3/4) Epoch 12, batch 6200, loss[loss=0.1985, simple_loss=0.2895, pruned_loss=0.05369, over 8363.00 frames. ], tot_loss[loss=0.2298, simple_loss=0.3061, pruned_loss=0.07674, over 1610719.47 frames. ], batch size: 24, lr: 6.31e-03, grad_scale: 8.0 +2023-02-06 12:27:41,101 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95123.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 12:28:04,342 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.821e+02 2.519e+02 2.980e+02 3.798e+02 7.393e+02, threshold=5.961e+02, percent-clipped=2.0 +2023-02-06 12:28:10,299 INFO [train.py:901] (3/4) Epoch 12, batch 6250, loss[loss=0.1977, simple_loss=0.2777, pruned_loss=0.05889, over 7921.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.3063, pruned_loss=0.07702, over 1611432.34 frames. ], batch size: 20, lr: 6.31e-03, grad_scale: 8.0 +2023-02-06 12:28:13,980 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-06 12:28:19,684 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.9548, 1.3009, 6.0890, 1.8797, 5.2893, 5.1425, 5.6211, 5.4562], + device='cuda:3'), covar=tensor([0.0424, 0.4883, 0.0348, 0.3499, 0.1078, 0.0766, 0.0430, 0.0489], + device='cuda:3'), in_proj_covar=tensor([0.0511, 0.0583, 0.0596, 0.0543, 0.0625, 0.0534, 0.0525, 0.0586], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 12:28:21,778 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2422, 1.8678, 2.5667, 2.0488, 2.3352, 2.1267, 1.7204, 1.0751], + device='cuda:3'), covar=tensor([0.3798, 0.3696, 0.1137, 0.2398, 0.1685, 0.2146, 0.1724, 0.3796], + device='cuda:3'), in_proj_covar=tensor([0.0899, 0.0884, 0.0740, 0.0867, 0.0947, 0.0813, 0.0713, 0.0776], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 12:28:43,827 INFO [train.py:901] (3/4) Epoch 12, batch 6300, loss[loss=0.1777, simple_loss=0.2541, pruned_loss=0.05071, over 6792.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3069, pruned_loss=0.07735, over 1610744.92 frames. ], batch size: 15, lr: 6.30e-03, grad_scale: 8.0 +2023-02-06 12:28:46,414 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.81 vs. limit=5.0 +2023-02-06 12:29:13,414 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.551e+02 2.657e+02 3.224e+02 4.358e+02 1.571e+03, threshold=6.448e+02, percent-clipped=5.0 +2023-02-06 12:29:20,982 INFO [train.py:901] (3/4) Epoch 12, batch 6350, loss[loss=0.1971, simple_loss=0.2783, pruned_loss=0.05792, over 7811.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.3059, pruned_loss=0.07674, over 1610991.49 frames. ], batch size: 20, lr: 6.30e-03, grad_scale: 8.0 +2023-02-06 12:29:30,724 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95278.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:29:36,279 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95286.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:29:50,022 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7003, 2.2473, 4.4569, 1.3815, 3.0658, 2.1979, 1.7208, 2.9461], + device='cuda:3'), covar=tensor([0.1747, 0.2423, 0.0622, 0.4133, 0.1583, 0.2896, 0.1940, 0.2168], + device='cuda:3'), in_proj_covar=tensor([0.0493, 0.0531, 0.0537, 0.0586, 0.0621, 0.0559, 0.0478, 0.0618], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 12:29:55,395 INFO [train.py:901] (3/4) Epoch 12, batch 6400, loss[loss=0.1848, simple_loss=0.2542, pruned_loss=0.05771, over 7548.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.3048, pruned_loss=0.07636, over 1613477.46 frames. ], batch size: 18, lr: 6.30e-03, grad_scale: 8.0 +2023-02-06 12:30:23,572 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.685e+02 2.403e+02 2.937e+02 3.904e+02 6.682e+02, threshold=5.874e+02, percent-clipped=3.0 +2023-02-06 12:30:25,105 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95357.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:30:27,148 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4907, 1.8567, 3.1499, 1.2608, 2.2969, 1.8913, 1.6033, 2.1055], + device='cuda:3'), covar=tensor([0.1700, 0.2514, 0.0807, 0.4079, 0.1716, 0.2843, 0.1936, 0.2265], + device='cuda:3'), in_proj_covar=tensor([0.0487, 0.0526, 0.0533, 0.0579, 0.0617, 0.0555, 0.0472, 0.0610], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 12:30:29,662 INFO [train.py:901] (3/4) Epoch 12, batch 6450, loss[loss=0.1783, simple_loss=0.2508, pruned_loss=0.05288, over 7711.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.3054, pruned_loss=0.07678, over 1614171.83 frames. ], batch size: 18, lr: 6.30e-03, grad_scale: 8.0 +2023-02-06 12:31:05,024 INFO [train.py:901] (3/4) Epoch 12, batch 6500, loss[loss=0.245, simple_loss=0.3271, pruned_loss=0.08149, over 8352.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3045, pruned_loss=0.07606, over 1615696.21 frames. ], batch size: 26, lr: 6.30e-03, grad_scale: 8.0 +2023-02-06 12:31:14,613 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2993, 1.5124, 2.1958, 1.1947, 1.5286, 1.5736, 1.4098, 1.4440], + device='cuda:3'), covar=tensor([0.1776, 0.2251, 0.0803, 0.3942, 0.1593, 0.2929, 0.1886, 0.1886], + device='cuda:3'), in_proj_covar=tensor([0.0489, 0.0527, 0.0533, 0.0580, 0.0618, 0.0555, 0.0473, 0.0611], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 12:31:31,884 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.523e+02 2.431e+02 2.857e+02 3.846e+02 1.801e+03, threshold=5.713e+02, percent-clipped=8.0 +2023-02-06 12:31:35,004 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 +2023-02-06 12:31:37,946 INFO [train.py:901] (3/4) Epoch 12, batch 6550, loss[loss=0.1958, simple_loss=0.2798, pruned_loss=0.05587, over 7818.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.3048, pruned_loss=0.07642, over 1613534.05 frames. ], batch size: 20, lr: 6.30e-03, grad_scale: 8.0 +2023-02-06 12:31:40,440 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.27 vs. limit=5.0 +2023-02-06 12:31:40,702 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95467.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 12:31:59,683 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.6090, 1.5353, 3.7856, 1.5347, 3.3233, 3.1801, 3.4455, 3.3266], + device='cuda:3'), covar=tensor([0.0646, 0.3952, 0.0646, 0.3480, 0.1168, 0.0922, 0.0599, 0.0667], + device='cuda:3'), in_proj_covar=tensor([0.0512, 0.0581, 0.0594, 0.0544, 0.0622, 0.0532, 0.0520, 0.0582], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 12:32:06,855 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 12:32:13,528 INFO [train.py:901] (3/4) Epoch 12, batch 6600, loss[loss=0.2668, simple_loss=0.3366, pruned_loss=0.09851, over 8695.00 frames. ], tot_loss[loss=0.2304, simple_loss=0.306, pruned_loss=0.07737, over 1610824.17 frames. ], batch size: 49, lr: 6.29e-03, grad_scale: 8.0 +2023-02-06 12:32:25,471 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-06 12:32:25,784 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 12:32:40,261 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.745e+02 2.522e+02 3.078e+02 3.913e+02 8.021e+02, threshold=6.157e+02, percent-clipped=7.0 +2023-02-06 12:32:43,802 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4036, 1.4585, 1.7728, 1.3284, 0.9833, 1.8000, 0.0973, 1.1471], + device='cuda:3'), covar=tensor([0.2774, 0.1591, 0.0491, 0.1559, 0.4159, 0.0495, 0.2727, 0.1554], + device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0173, 0.0103, 0.0216, 0.0256, 0.0108, 0.0163, 0.0166], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 12:32:46,202 INFO [train.py:901] (3/4) Epoch 12, batch 6650, loss[loss=0.1996, simple_loss=0.2724, pruned_loss=0.06343, over 7784.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.3069, pruned_loss=0.07808, over 1608351.10 frames. ], batch size: 19, lr: 6.29e-03, grad_scale: 8.0 +2023-02-06 12:32:59,165 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95582.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 12:33:15,566 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0 +2023-02-06 12:33:21,233 INFO [train.py:901] (3/4) Epoch 12, batch 6700, loss[loss=0.2725, simple_loss=0.352, pruned_loss=0.09652, over 8253.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3064, pruned_loss=0.07714, over 1612442.56 frames. ], batch size: 24, lr: 6.29e-03, grad_scale: 8.0 +2023-02-06 12:33:27,481 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95622.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:33:33,703 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95630.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:33:35,042 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7400, 1.4583, 2.7492, 1.2271, 2.0572, 3.0201, 3.1263, 2.5812], + device='cuda:3'), covar=tensor([0.1048, 0.1433, 0.0413, 0.2104, 0.0875, 0.0279, 0.0486, 0.0620], + device='cuda:3'), in_proj_covar=tensor([0.0269, 0.0297, 0.0261, 0.0291, 0.0272, 0.0236, 0.0347, 0.0288], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 12:33:37,106 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95635.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:33:50,489 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.832e+02 2.656e+02 3.142e+02 4.011e+02 7.522e+02, threshold=6.284e+02, percent-clipped=4.0 +2023-02-06 12:33:56,567 INFO [train.py:901] (3/4) Epoch 12, batch 6750, loss[loss=0.2856, simple_loss=0.3566, pruned_loss=0.1073, over 8112.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3081, pruned_loss=0.07768, over 1617907.48 frames. ], batch size: 23, lr: 6.29e-03, grad_scale: 8.0 +2023-02-06 12:34:22,169 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95701.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:34:30,630 INFO [train.py:901] (3/4) Epoch 12, batch 6800, loss[loss=0.2312, simple_loss=0.3012, pruned_loss=0.08056, over 8042.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.3068, pruned_loss=0.07666, over 1616799.33 frames. ], batch size: 20, lr: 6.29e-03, grad_scale: 8.0 +2023-02-06 12:34:40,734 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 12:34:47,118 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95737.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:34:53,250 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95745.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:35:00,257 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.685e+02 2.510e+02 2.822e+02 3.564e+02 9.162e+02, threshold=5.644e+02, percent-clipped=3.0 +2023-02-06 12:35:06,300 INFO [train.py:901] (3/4) Epoch 12, batch 6850, loss[loss=0.2682, simple_loss=0.3369, pruned_loss=0.09976, over 8287.00 frames. ], tot_loss[loss=0.2311, simple_loss=0.3079, pruned_loss=0.07718, over 1617618.21 frames. ], batch size: 23, lr: 6.29e-03, grad_scale: 8.0 +2023-02-06 12:35:19,168 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5142, 1.9533, 2.0645, 1.0885, 2.1615, 1.3626, 0.5381, 1.7969], + device='cuda:3'), covar=tensor([0.0454, 0.0231, 0.0181, 0.0476, 0.0295, 0.0758, 0.0616, 0.0213], + device='cuda:3'), in_proj_covar=tensor([0.0394, 0.0329, 0.0282, 0.0393, 0.0325, 0.0483, 0.0360, 0.0360], + device='cuda:3'), out_proj_covar=tensor([1.1132e-04, 8.9943e-05, 7.7810e-05, 1.0918e-04, 9.0896e-05, 1.4543e-04, + 1.0159e-04, 1.0083e-04], device='cuda:3') +2023-02-06 12:35:25,954 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.13 vs. limit=5.0 +2023-02-06 12:35:26,864 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 12:35:40,362 INFO [train.py:901] (3/4) Epoch 12, batch 6900, loss[loss=0.2359, simple_loss=0.3192, pruned_loss=0.0763, over 8198.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3077, pruned_loss=0.07716, over 1614838.83 frames. ], batch size: 23, lr: 6.29e-03, grad_scale: 8.0 +2023-02-06 12:35:41,923 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95816.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:35:43,310 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4298, 1.9556, 3.2263, 1.2407, 2.4173, 1.8936, 1.5310, 2.1895], + device='cuda:3'), covar=tensor([0.1792, 0.2302, 0.0715, 0.4229, 0.1573, 0.3055, 0.2004, 0.2188], + device='cuda:3'), in_proj_covar=tensor([0.0488, 0.0526, 0.0531, 0.0580, 0.0616, 0.0554, 0.0476, 0.0610], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 12:35:48,560 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3033, 1.7393, 2.7281, 1.1676, 2.0214, 1.7081, 1.4783, 1.8323], + device='cuda:3'), covar=tensor([0.1938, 0.2251, 0.0827, 0.4202, 0.1630, 0.3076, 0.1989, 0.2228], + device='cuda:3'), in_proj_covar=tensor([0.0487, 0.0525, 0.0530, 0.0579, 0.0615, 0.0554, 0.0475, 0.0609], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 12:35:51,215 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95830.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 12:35:57,285 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95838.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 12:36:08,028 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.682e+02 2.693e+02 3.422e+02 4.342e+02 1.062e+03, threshold=6.843e+02, percent-clipped=12.0 +2023-02-06 12:36:14,248 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95863.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 12:36:14,702 INFO [train.py:901] (3/4) Epoch 12, batch 6950, loss[loss=0.1962, simple_loss=0.2653, pruned_loss=0.06351, over 7204.00 frames. ], tot_loss[loss=0.2309, simple_loss=0.3074, pruned_loss=0.07717, over 1609990.38 frames. ], batch size: 16, lr: 6.28e-03, grad_scale: 8.0 +2023-02-06 12:36:15,458 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95865.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:36:34,590 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 12:36:48,737 INFO [train.py:901] (3/4) Epoch 12, batch 7000, loss[loss=0.2182, simple_loss=0.2852, pruned_loss=0.07558, over 7659.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3075, pruned_loss=0.07709, over 1610088.68 frames. ], batch size: 19, lr: 6.28e-03, grad_scale: 8.0 +2023-02-06 12:37:17,422 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.755e+02 2.501e+02 3.116e+02 3.850e+02 8.001e+02, threshold=6.232e+02, percent-clipped=2.0 +2023-02-06 12:37:19,590 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.88 vs. limit=5.0 +2023-02-06 12:37:23,279 INFO [train.py:901] (3/4) Epoch 12, batch 7050, loss[loss=0.2128, simple_loss=0.2895, pruned_loss=0.06806, over 7813.00 frames. ], tot_loss[loss=0.2292, simple_loss=0.3059, pruned_loss=0.07626, over 1609214.86 frames. ], batch size: 20, lr: 6.28e-03, grad_scale: 8.0 +2023-02-06 12:37:24,105 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3328, 1.2573, 4.5483, 1.7853, 3.9650, 3.8053, 4.0922, 3.9460], + device='cuda:3'), covar=tensor([0.0531, 0.4811, 0.0462, 0.3358, 0.1139, 0.0806, 0.0519, 0.0616], + device='cuda:3'), in_proj_covar=tensor([0.0504, 0.0580, 0.0592, 0.0543, 0.0618, 0.0530, 0.0521, 0.0582], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 12:37:34,014 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=95979.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:37:44,076 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95993.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:37:45,560 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=5.99 vs. limit=5.0 +2023-02-06 12:37:47,413 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95998.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:37:50,529 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96001.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:37:58,492 INFO [train.py:901] (3/4) Epoch 12, batch 7100, loss[loss=0.1877, simple_loss=0.2732, pruned_loss=0.05106, over 7814.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.3057, pruned_loss=0.07668, over 1610754.82 frames. ], batch size: 20, lr: 6.28e-03, grad_scale: 8.0 +2023-02-06 12:38:01,410 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96018.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:38:07,003 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96026.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:38:26,966 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.650e+02 2.539e+02 3.029e+02 4.080e+02 8.783e+02, threshold=6.058e+02, percent-clipped=4.0 +2023-02-06 12:38:33,147 INFO [train.py:901] (3/4) Epoch 12, batch 7150, loss[loss=0.2597, simple_loss=0.3281, pruned_loss=0.09564, over 6863.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.3064, pruned_loss=0.07703, over 1612535.97 frames. ], batch size: 72, lr: 6.28e-03, grad_scale: 8.0 +2023-02-06 12:38:38,725 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96072.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:38:54,802 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96094.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:38:56,840 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96097.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:38:58,153 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96098.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:39:08,660 INFO [train.py:901] (3/4) Epoch 12, batch 7200, loss[loss=0.2409, simple_loss=0.3013, pruned_loss=0.09023, over 7648.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3069, pruned_loss=0.07732, over 1614158.01 frames. ], batch size: 19, lr: 6.28e-03, grad_scale: 8.0 +2023-02-06 12:39:36,231 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.658e+02 2.446e+02 3.002e+02 3.633e+02 6.248e+02, threshold=6.005e+02, percent-clipped=1.0 +2023-02-06 12:39:42,864 INFO [train.py:901] (3/4) Epoch 12, batch 7250, loss[loss=0.2726, simple_loss=0.3313, pruned_loss=0.1069, over 7053.00 frames. ], tot_loss[loss=0.2305, simple_loss=0.3067, pruned_loss=0.07716, over 1612808.38 frames. ], batch size: 71, lr: 6.27e-03, grad_scale: 8.0 +2023-02-06 12:39:49,510 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96174.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 12:39:55,436 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96183.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:39:56,073 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.3712, 5.4501, 4.8297, 2.3942, 4.8252, 5.0049, 5.1298, 4.5741], + device='cuda:3'), covar=tensor([0.0643, 0.0429, 0.0869, 0.4640, 0.0730, 0.0891, 0.0897, 0.0861], + device='cuda:3'), in_proj_covar=tensor([0.0471, 0.0381, 0.0393, 0.0484, 0.0385, 0.0385, 0.0382, 0.0334], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 12:40:14,154 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96209.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:40:17,295 INFO [train.py:901] (3/4) Epoch 12, batch 7300, loss[loss=0.1525, simple_loss=0.2327, pruned_loss=0.03609, over 7451.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.307, pruned_loss=0.07729, over 1610761.53 frames. ], batch size: 17, lr: 6.27e-03, grad_scale: 8.0 +2023-02-06 12:40:44,977 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8803, 2.2958, 3.1687, 1.6942, 2.6925, 2.1997, 2.0891, 2.5223], + device='cuda:3'), covar=tensor([0.1308, 0.1742, 0.0555, 0.3080, 0.1224, 0.2105, 0.1361, 0.1672], + device='cuda:3'), in_proj_covar=tensor([0.0488, 0.0526, 0.0535, 0.0578, 0.0620, 0.0556, 0.0475, 0.0610], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 12:40:45,404 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.670e+02 2.567e+02 3.297e+02 4.044e+02 1.170e+03, threshold=6.593e+02, percent-clipped=7.0 +2023-02-06 12:40:51,427 INFO [train.py:901] (3/4) Epoch 12, batch 7350, loss[loss=0.2264, simple_loss=0.3096, pruned_loss=0.07157, over 8676.00 frames. ], tot_loss[loss=0.2306, simple_loss=0.3069, pruned_loss=0.07719, over 1610913.59 frames. ], batch size: 34, lr: 6.27e-03, grad_scale: 8.0 +2023-02-06 12:41:09,304 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96289.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 12:41:15,842 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 12:41:26,495 INFO [train.py:901] (3/4) Epoch 12, batch 7400, loss[loss=0.2319, simple_loss=0.3012, pruned_loss=0.08132, over 5953.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3049, pruned_loss=0.07582, over 1603073.81 frames. ], batch size: 13, lr: 6.27e-03, grad_scale: 8.0 +2023-02-06 12:41:33,417 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96324.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:41:36,481 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 12:41:40,821 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.2151, 4.1925, 3.7369, 1.7999, 3.6612, 3.6653, 3.7835, 3.3967], + device='cuda:3'), covar=tensor([0.0742, 0.0588, 0.1058, 0.4461, 0.0832, 0.0959, 0.1215, 0.0996], + device='cuda:3'), in_proj_covar=tensor([0.0467, 0.0381, 0.0390, 0.0483, 0.0381, 0.0384, 0.0380, 0.0333], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 12:41:42,992 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4725, 1.2222, 4.6670, 1.6821, 4.0911, 3.8708, 4.1990, 4.0948], + device='cuda:3'), covar=tensor([0.0554, 0.5206, 0.0477, 0.3836, 0.1156, 0.0911, 0.0575, 0.0614], + device='cuda:3'), in_proj_covar=tensor([0.0512, 0.0589, 0.0604, 0.0549, 0.0627, 0.0535, 0.0530, 0.0592], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 12:41:47,013 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96342.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:41:47,081 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96342.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:41:52,329 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96350.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:41:55,367 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.734e+02 2.475e+02 3.182e+02 4.307e+02 9.281e+02, threshold=6.365e+02, percent-clipped=3.0 +2023-02-06 12:42:01,560 INFO [train.py:901] (3/4) Epoch 12, batch 7450, loss[loss=0.2839, simple_loss=0.3562, pruned_loss=0.1058, over 8593.00 frames. ], tot_loss[loss=0.2289, simple_loss=0.3055, pruned_loss=0.07619, over 1606269.78 frames. ], batch size: 39, lr: 6.27e-03, grad_scale: 8.0 +2023-02-06 12:42:09,242 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96375.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:42:15,394 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 12:42:31,442 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96408.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:42:35,242 INFO [train.py:901] (3/4) Epoch 12, batch 7500, loss[loss=0.2087, simple_loss=0.2845, pruned_loss=0.06642, over 7446.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.3061, pruned_loss=0.07647, over 1605040.49 frames. ], batch size: 17, lr: 6.27e-03, grad_scale: 8.0 +2023-02-06 12:42:54,779 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96442.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:43:02,270 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9825, 3.9371, 2.5234, 2.7884, 3.0862, 2.1596, 2.6499, 2.8535], + device='cuda:3'), covar=tensor([0.1560, 0.0352, 0.0911, 0.0708, 0.0578, 0.1178, 0.0988, 0.1045], + device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0240, 0.0323, 0.0301, 0.0306, 0.0325, 0.0343, 0.0307], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 12:43:03,968 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.737e+02 2.684e+02 3.354e+02 4.069e+02 8.964e+02, threshold=6.707e+02, percent-clipped=7.0 +2023-02-06 12:43:05,486 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96457.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:43:09,792 INFO [train.py:901] (3/4) Epoch 12, batch 7550, loss[loss=0.2089, simple_loss=0.2827, pruned_loss=0.06755, over 7712.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.3073, pruned_loss=0.07709, over 1607447.25 frames. ], batch size: 18, lr: 6.26e-03, grad_scale: 16.0 +2023-02-06 12:43:15,331 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2297, 1.1140, 1.3010, 1.1424, 0.9524, 1.3005, 0.1093, 1.0364], + device='cuda:3'), covar=tensor([0.2282, 0.1694, 0.0614, 0.1129, 0.3634, 0.0648, 0.2861, 0.1479], + device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0173, 0.0103, 0.0216, 0.0254, 0.0109, 0.0164, 0.0166], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 12:43:42,990 INFO [train.py:901] (3/4) Epoch 12, batch 7600, loss[loss=0.2354, simple_loss=0.3006, pruned_loss=0.08517, over 7969.00 frames. ], tot_loss[loss=0.2313, simple_loss=0.3078, pruned_loss=0.07742, over 1608080.55 frames. ], batch size: 21, lr: 6.26e-03, grad_scale: 16.0 +2023-02-06 12:43:52,531 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96527.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:44:05,572 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96545.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 12:44:11,799 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.342e+02 2.706e+02 3.173e+02 4.121e+02 9.971e+02, threshold=6.345e+02, percent-clipped=8.0 +2023-02-06 12:44:13,340 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96557.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:44:18,542 INFO [train.py:901] (3/4) Epoch 12, batch 7650, loss[loss=0.1737, simple_loss=0.2489, pruned_loss=0.04926, over 7307.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3079, pruned_loss=0.07727, over 1608262.98 frames. ], batch size: 16, lr: 6.26e-03, grad_scale: 16.0 +2023-02-06 12:44:23,310 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96570.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 12:44:29,851 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96580.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:44:47,080 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96605.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:44:53,025 INFO [train.py:901] (3/4) Epoch 12, batch 7700, loss[loss=0.2051, simple_loss=0.2955, pruned_loss=0.0573, over 8246.00 frames. ], tot_loss[loss=0.2309, simple_loss=0.3077, pruned_loss=0.07701, over 1611791.08 frames. ], batch size: 24, lr: 6.26e-03, grad_scale: 16.0 +2023-02-06 12:45:02,659 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5304, 1.5354, 1.8010, 1.3933, 1.0914, 1.7802, 0.1228, 1.0994], + device='cuda:3'), covar=tensor([0.2212, 0.1503, 0.0551, 0.1439, 0.3617, 0.0505, 0.3326, 0.1711], + device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0172, 0.0103, 0.0215, 0.0255, 0.0109, 0.0163, 0.0166], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 12:45:12,730 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96642.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:45:21,241 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.670e+02 2.521e+02 3.004e+02 3.630e+02 7.905e+02, threshold=6.007e+02, percent-clipped=3.0 +2023-02-06 12:45:23,931 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 12:45:27,894 INFO [train.py:901] (3/4) Epoch 12, batch 7750, loss[loss=0.2858, simple_loss=0.3618, pruned_loss=0.1048, over 8578.00 frames. ], tot_loss[loss=0.2311, simple_loss=0.308, pruned_loss=0.07709, over 1615568.58 frames. ], batch size: 31, lr: 6.26e-03, grad_scale: 16.0 +2023-02-06 12:45:42,858 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96686.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:46:02,093 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96713.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:46:02,573 INFO [train.py:901] (3/4) Epoch 12, batch 7800, loss[loss=0.2254, simple_loss=0.3129, pruned_loss=0.06895, over 8712.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.3065, pruned_loss=0.07646, over 1615695.03 frames. ], batch size: 34, lr: 6.26e-03, grad_scale: 16.0 +2023-02-06 12:46:15,972 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.80 vs. limit=5.0 +2023-02-06 12:46:19,264 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96738.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:46:23,938 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3268, 1.3055, 2.1454, 0.9648, 2.0054, 2.3087, 2.5568, 1.6859], + device='cuda:3'), covar=tensor([0.1258, 0.1468, 0.0697, 0.2732, 0.0932, 0.0596, 0.0782, 0.1216], + device='cuda:3'), in_proj_covar=tensor([0.0268, 0.0299, 0.0263, 0.0291, 0.0274, 0.0238, 0.0354, 0.0289], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 12:46:28,441 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=96752.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:46:30,315 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.674e+02 2.519e+02 3.193e+02 4.174e+02 8.059e+02, threshold=6.386e+02, percent-clipped=4.0 +2023-02-06 12:46:31,382 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-06 12:46:36,566 INFO [train.py:901] (3/4) Epoch 12, batch 7850, loss[loss=0.2426, simple_loss=0.3124, pruned_loss=0.08645, over 7474.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.306, pruned_loss=0.0765, over 1614826.01 frames. ], batch size: 71, lr: 6.25e-03, grad_scale: 16.0 +2023-02-06 12:46:56,866 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96794.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:47:01,900 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96801.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:47:10,034 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96813.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:47:10,493 INFO [train.py:901] (3/4) Epoch 12, batch 7900, loss[loss=0.3025, simple_loss=0.3599, pruned_loss=0.1226, over 6763.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.3061, pruned_loss=0.0767, over 1615799.17 frames. ], batch size: 72, lr: 6.25e-03, grad_scale: 16.0 +2023-02-06 12:47:27,488 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96838.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:47:38,783 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.595e+02 3.080e+02 3.878e+02 8.124e+02, threshold=6.160e+02, percent-clipped=3.0 +2023-02-06 12:47:44,792 INFO [train.py:901] (3/4) Epoch 12, batch 7950, loss[loss=0.2161, simple_loss=0.2815, pruned_loss=0.07537, over 7798.00 frames. ], tot_loss[loss=0.2319, simple_loss=0.3077, pruned_loss=0.07805, over 1615856.77 frames. ], batch size: 19, lr: 6.25e-03, grad_scale: 16.0 +2023-02-06 12:47:47,003 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96867.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:48:07,435 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96898.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:48:17,698 INFO [train.py:901] (3/4) Epoch 12, batch 8000, loss[loss=0.2301, simple_loss=0.3054, pruned_loss=0.07738, over 8131.00 frames. ], tot_loss[loss=0.2323, simple_loss=0.3078, pruned_loss=0.0784, over 1615500.14 frames. ], batch size: 22, lr: 6.25e-03, grad_scale: 16.0 +2023-02-06 12:48:23,698 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96923.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:48:36,034 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-06 12:48:45,040 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.593e+02 2.377e+02 3.280e+02 4.266e+02 7.100e+02, threshold=6.559e+02, percent-clipped=4.0 +2023-02-06 12:48:51,263 INFO [train.py:901] (3/4) Epoch 12, batch 8050, loss[loss=0.2045, simple_loss=0.2737, pruned_loss=0.06764, over 7542.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.3054, pruned_loss=0.07745, over 1606441.35 frames. ], batch size: 18, lr: 6.25e-03, grad_scale: 16.0 +2023-02-06 12:49:06,772 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.88 vs. limit=5.0 +2023-02-06 12:49:24,805 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 12:49:29,791 INFO [train.py:901] (3/4) Epoch 13, batch 0, loss[loss=0.2443, simple_loss=0.3268, pruned_loss=0.08092, over 8279.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3268, pruned_loss=0.08092, over 8279.00 frames. ], batch size: 23, lr: 6.00e-03, grad_scale: 16.0 +2023-02-06 12:49:29,791 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 12:49:40,735 INFO [train.py:935] (3/4) Epoch 13, validation: loss=0.1867, simple_loss=0.2865, pruned_loss=0.04345, over 944034.00 frames. +2023-02-06 12:49:40,737 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 12:49:41,811 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-06 12:49:55,391 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 12:49:55,526 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97018.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:50:14,915 INFO [train.py:901] (3/4) Epoch 13, batch 50, loss[loss=0.2891, simple_loss=0.3588, pruned_loss=0.1097, over 8461.00 frames. ], tot_loss[loss=0.2347, simple_loss=0.3118, pruned_loss=0.07875, over 368319.19 frames. ], batch size: 25, lr: 6.00e-03, grad_scale: 16.0 +2023-02-06 12:50:20,337 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.491e+02 2.833e+02 3.357e+02 4.758e+02 6.927e+02, threshold=6.715e+02, percent-clipped=2.0 +2023-02-06 12:50:21,954 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97057.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:50:29,188 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 12:50:41,097 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97082.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:50:50,959 INFO [train.py:901] (3/4) Epoch 13, batch 100, loss[loss=0.192, simple_loss=0.2661, pruned_loss=0.05891, over 7645.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.3076, pruned_loss=0.07576, over 638995.58 frames. ], batch size: 19, lr: 6.00e-03, grad_scale: 16.0 +2023-02-06 12:50:52,984 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 12:51:09,074 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97123.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:51:18,986 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97138.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:51:24,701 INFO [train.py:901] (3/4) Epoch 13, batch 150, loss[loss=0.3121, simple_loss=0.3723, pruned_loss=0.1259, over 7965.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3084, pruned_loss=0.07698, over 853655.18 frames. ], batch size: 21, lr: 6.00e-03, grad_scale: 16.0 +2023-02-06 12:51:25,593 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97148.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:51:30,111 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.637e+02 2.477e+02 2.848e+02 3.342e+02 7.997e+02, threshold=5.696e+02, percent-clipped=2.0 +2023-02-06 12:51:39,289 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 +2023-02-06 12:51:42,869 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2196, 1.3318, 4.3649, 1.9613, 2.4477, 4.9879, 4.8606, 4.2995], + device='cuda:3'), covar=tensor([0.1139, 0.1786, 0.0265, 0.1880, 0.1052, 0.0175, 0.0380, 0.0553], + device='cuda:3'), in_proj_covar=tensor([0.0271, 0.0302, 0.0265, 0.0293, 0.0275, 0.0239, 0.0359, 0.0292], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 12:51:58,451 INFO [train.py:901] (3/4) Epoch 13, batch 200, loss[loss=0.2113, simple_loss=0.3023, pruned_loss=0.06012, over 8502.00 frames. ], tot_loss[loss=0.2323, simple_loss=0.3091, pruned_loss=0.07772, over 1024861.43 frames. ], batch size: 26, lr: 6.00e-03, grad_scale: 16.0 +2023-02-06 12:52:09,220 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8791, 2.0406, 1.7347, 2.5880, 1.2185, 1.5438, 1.6695, 2.1109], + device='cuda:3'), covar=tensor([0.0702, 0.0833, 0.0914, 0.0418, 0.1135, 0.1379, 0.1008, 0.0682], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0211, 0.0251, 0.0214, 0.0213, 0.0251, 0.0254, 0.0218], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 12:52:18,286 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2879, 1.8430, 4.2585, 1.6778, 2.2979, 4.7124, 4.8725, 3.8358], + device='cuda:3'), covar=tensor([0.1229, 0.1624, 0.0375, 0.2413, 0.1284, 0.0295, 0.0532, 0.0840], + device='cuda:3'), in_proj_covar=tensor([0.0269, 0.0301, 0.0263, 0.0292, 0.0274, 0.0238, 0.0356, 0.0291], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 12:52:23,162 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 +2023-02-06 12:52:33,410 INFO [train.py:901] (3/4) Epoch 13, batch 250, loss[loss=0.262, simple_loss=0.3433, pruned_loss=0.09033, over 8136.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3081, pruned_loss=0.07699, over 1155770.70 frames. ], batch size: 22, lr: 6.00e-03, grad_scale: 16.0 +2023-02-06 12:52:37,629 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97253.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:52:38,759 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.741e+02 2.455e+02 3.117e+02 3.819e+02 7.824e+02, threshold=6.233e+02, percent-clipped=7.0 +2023-02-06 12:52:46,021 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 12:52:53,683 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-06 12:52:54,016 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97278.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:52:54,531 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 12:53:06,365 INFO [train.py:901] (3/4) Epoch 13, batch 300, loss[loss=0.2122, simple_loss=0.2949, pruned_loss=0.06474, over 8033.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3076, pruned_loss=0.07723, over 1254459.02 frames. ], batch size: 22, lr: 5.99e-03, grad_scale: 16.0 +2023-02-06 12:53:06,760 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 +2023-02-06 12:53:41,549 INFO [train.py:901] (3/4) Epoch 13, batch 350, loss[loss=0.212, simple_loss=0.2755, pruned_loss=0.07425, over 7223.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3088, pruned_loss=0.07806, over 1338238.96 frames. ], batch size: 16, lr: 5.99e-03, grad_scale: 16.0 +2023-02-06 12:53:46,922 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.701e+02 2.508e+02 3.076e+02 3.709e+02 6.548e+02, threshold=6.153e+02, percent-clipped=1.0 +2023-02-06 12:53:50,484 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97360.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:53:51,768 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97362.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:53:53,849 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97365.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:54:11,903 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97392.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:54:15,154 INFO [train.py:901] (3/4) Epoch 13, batch 400, loss[loss=0.2378, simple_loss=0.3155, pruned_loss=0.0801, over 8354.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.3084, pruned_loss=0.07791, over 1401299.61 frames. ], batch size: 24, lr: 5.99e-03, grad_scale: 16.0 +2023-02-06 12:54:34,010 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6294, 1.9554, 3.0811, 1.3921, 2.2014, 1.9449, 1.7403, 2.0855], + device='cuda:3'), covar=tensor([0.1576, 0.2087, 0.0713, 0.3890, 0.1662, 0.2861, 0.1728, 0.2190], + device='cuda:3'), in_proj_covar=tensor([0.0486, 0.0525, 0.0531, 0.0577, 0.0620, 0.0554, 0.0476, 0.0611], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 12:54:45,026 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3119, 1.9570, 2.8687, 2.3532, 2.5764, 2.2264, 1.8486, 1.3288], + device='cuda:3'), covar=tensor([0.4214, 0.4219, 0.1354, 0.2781, 0.2100, 0.2457, 0.1709, 0.4649], + device='cuda:3'), in_proj_covar=tensor([0.0890, 0.0878, 0.0730, 0.0857, 0.0935, 0.0807, 0.0702, 0.0768], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 12:54:51,721 INFO [train.py:901] (3/4) Epoch 13, batch 450, loss[loss=0.2239, simple_loss=0.3078, pruned_loss=0.06997, over 8468.00 frames. ], tot_loss[loss=0.2319, simple_loss=0.3084, pruned_loss=0.07768, over 1452629.52 frames. ], batch size: 29, lr: 5.99e-03, grad_scale: 16.0 +2023-02-06 12:54:57,098 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 2.318e+02 2.836e+02 3.756e+02 7.381e+02, threshold=5.672e+02, percent-clipped=3.0 +2023-02-06 12:55:13,262 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97477.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:55:26,761 INFO [train.py:901] (3/4) Epoch 13, batch 500, loss[loss=0.2442, simple_loss=0.3143, pruned_loss=0.08702, over 8494.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3077, pruned_loss=0.07711, over 1479630.95 frames. ], batch size: 28, lr: 5.99e-03, grad_scale: 16.0 +2023-02-06 12:55:29,814 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6676, 1.6793, 2.0572, 1.5650, 1.0717, 2.0488, 0.2267, 1.3067], + device='cuda:3'), covar=tensor([0.2053, 0.1727, 0.0465, 0.1739, 0.4392, 0.0478, 0.3140, 0.1858], + device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0173, 0.0103, 0.0220, 0.0259, 0.0110, 0.0164, 0.0170], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 12:55:35,492 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97509.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:55:52,969 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97534.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:56:01,768 INFO [train.py:901] (3/4) Epoch 13, batch 550, loss[loss=0.2071, simple_loss=0.2784, pruned_loss=0.06795, over 7220.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.307, pruned_loss=0.07666, over 1510812.06 frames. ], batch size: 16, lr: 5.99e-03, grad_scale: 16.0 +2023-02-06 12:56:07,720 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.562e+02 2.533e+02 3.037e+02 3.770e+02 9.997e+02, threshold=6.074e+02, percent-clipped=4.0 +2023-02-06 12:56:20,094 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97573.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:56:36,760 INFO [train.py:901] (3/4) Epoch 13, batch 600, loss[loss=0.2453, simple_loss=0.3163, pruned_loss=0.08718, over 8329.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.3063, pruned_loss=0.07596, over 1536447.55 frames. ], batch size: 25, lr: 5.98e-03, grad_scale: 16.0 +2023-02-06 12:56:53,704 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97622.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:56:55,695 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 12:56:55,847 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8103, 1.4041, 2.8084, 1.3680, 2.0539, 2.9893, 3.1072, 2.5624], + device='cuda:3'), covar=tensor([0.0934, 0.1425, 0.0388, 0.1967, 0.0843, 0.0312, 0.0563, 0.0669], + device='cuda:3'), in_proj_covar=tensor([0.0267, 0.0299, 0.0264, 0.0292, 0.0275, 0.0239, 0.0356, 0.0291], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 12:57:10,256 INFO [train.py:901] (3/4) Epoch 13, batch 650, loss[loss=0.2082, simple_loss=0.2971, pruned_loss=0.05968, over 8486.00 frames. ], tot_loss[loss=0.2298, simple_loss=0.3074, pruned_loss=0.07612, over 1557918.55 frames. ], batch size: 29, lr: 5.98e-03, grad_scale: 16.0 +2023-02-06 12:57:16,275 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.422e+02 2.537e+02 2.925e+02 3.842e+02 7.324e+02, threshold=5.850e+02, percent-clipped=4.0 +2023-02-06 12:57:37,848 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5588, 2.6390, 1.8416, 2.1740, 2.0637, 1.3505, 1.8773, 2.1150], + device='cuda:3'), covar=tensor([0.1427, 0.0351, 0.1069, 0.0647, 0.0783, 0.1489, 0.1077, 0.0957], + device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0238, 0.0323, 0.0303, 0.0306, 0.0327, 0.0345, 0.0305], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 12:57:42,539 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97692.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 12:57:45,737 INFO [train.py:901] (3/4) Epoch 13, batch 700, loss[loss=0.2292, simple_loss=0.3041, pruned_loss=0.07719, over 8134.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.307, pruned_loss=0.07599, over 1572071.72 frames. ], batch size: 22, lr: 5.98e-03, grad_scale: 16.0 +2023-02-06 12:57:50,154 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.51 vs. limit=5.0 +2023-02-06 12:57:51,262 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97704.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:57:54,515 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97709.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:57:59,092 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2023-02-06 12:58:10,718 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97733.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:58:12,619 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97736.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:58:13,391 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97737.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:58:19,716 INFO [train.py:901] (3/4) Epoch 13, batch 750, loss[loss=0.214, simple_loss=0.2956, pruned_loss=0.06621, over 7967.00 frames. ], tot_loss[loss=0.229, simple_loss=0.3059, pruned_loss=0.07604, over 1580195.51 frames. ], batch size: 21, lr: 5.98e-03, grad_scale: 16.0 +2023-02-06 12:58:25,054 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.619e+02 2.478e+02 2.997e+02 3.995e+02 8.399e+02, threshold=5.994e+02, percent-clipped=5.0 +2023-02-06 12:58:27,379 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97758.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:58:39,716 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 12:58:49,000 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 12:58:51,777 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9459, 1.4653, 3.3285, 1.3771, 2.3818, 3.6763, 3.7254, 2.9850], + device='cuda:3'), covar=tensor([0.1144, 0.1679, 0.0397, 0.2135, 0.0953, 0.0304, 0.0561, 0.0829], + device='cuda:3'), in_proj_covar=tensor([0.0271, 0.0302, 0.0267, 0.0294, 0.0278, 0.0242, 0.0359, 0.0295], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 12:58:54,231 INFO [train.py:901] (3/4) Epoch 13, batch 800, loss[loss=0.2621, simple_loss=0.3194, pruned_loss=0.1024, over 7822.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3053, pruned_loss=0.07564, over 1589825.79 frames. ], batch size: 20, lr: 5.98e-03, grad_scale: 16.0 +2023-02-06 12:59:10,089 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97819.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:59:14,047 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97824.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:59:29,481 INFO [train.py:901] (3/4) Epoch 13, batch 850, loss[loss=0.1745, simple_loss=0.2632, pruned_loss=0.04293, over 8079.00 frames. ], tot_loss[loss=0.227, simple_loss=0.3041, pruned_loss=0.07492, over 1593674.50 frames. ], batch size: 21, lr: 5.98e-03, grad_scale: 16.0 +2023-02-06 12:59:32,323 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97851.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 12:59:35,502 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.775e+02 2.627e+02 3.254e+02 4.246e+02 9.834e+02, threshold=6.507e+02, percent-clipped=8.0 +2023-02-06 13:00:03,798 INFO [train.py:901] (3/4) Epoch 13, batch 900, loss[loss=0.2025, simple_loss=0.2765, pruned_loss=0.06425, over 7424.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.3036, pruned_loss=0.07465, over 1596221.18 frames. ], batch size: 17, lr: 5.98e-03, grad_scale: 8.0 +2023-02-06 13:00:05,028 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 +2023-02-06 13:00:09,705 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97906.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:00:18,218 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=97917.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:00:39,219 INFO [train.py:901] (3/4) Epoch 13, batch 950, loss[loss=0.2344, simple_loss=0.3091, pruned_loss=0.07988, over 8580.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.3051, pruned_loss=0.07559, over 1604868.05 frames. ], batch size: 39, lr: 5.97e-03, grad_scale: 8.0 +2023-02-06 13:00:45,305 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.497e+02 2.593e+02 3.202e+02 4.020e+02 7.231e+02, threshold=6.403e+02, percent-clipped=2.0 +2023-02-06 13:00:55,368 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1193, 1.9304, 3.1044, 1.5900, 2.3529, 3.3990, 3.4343, 2.9883], + device='cuda:3'), covar=tensor([0.0919, 0.1265, 0.0438, 0.1851, 0.1027, 0.0249, 0.0556, 0.0522], + device='cuda:3'), in_proj_covar=tensor([0.0269, 0.0300, 0.0266, 0.0292, 0.0276, 0.0241, 0.0358, 0.0292], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 13:01:08,779 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 13:01:11,125 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97993.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:01:13,715 INFO [train.py:901] (3/4) Epoch 13, batch 1000, loss[loss=0.2718, simple_loss=0.3475, pruned_loss=0.098, over 8287.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.305, pruned_loss=0.07566, over 1606662.24 frames. ], batch size: 23, lr: 5.97e-03, grad_scale: 8.0 +2023-02-06 13:01:29,781 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98018.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:01:39,957 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98032.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:01:43,183 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98036.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:01:44,385 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 13:01:50,454 INFO [train.py:901] (3/4) Epoch 13, batch 1050, loss[loss=0.2593, simple_loss=0.3272, pruned_loss=0.09568, over 8438.00 frames. ], tot_loss[loss=0.228, simple_loss=0.3048, pruned_loss=0.07559, over 1606945.93 frames. ], batch size: 27, lr: 5.97e-03, grad_scale: 8.0 +2023-02-06 13:01:56,534 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.707e+02 2.365e+02 2.893e+02 3.782e+02 5.594e+02, threshold=5.785e+02, percent-clipped=0.0 +2023-02-06 13:01:57,234 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 13:02:10,160 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98075.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:02:13,453 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98080.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:02:24,727 INFO [train.py:901] (3/4) Epoch 13, batch 1100, loss[loss=0.2318, simple_loss=0.313, pruned_loss=0.07526, over 8330.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3032, pruned_loss=0.07495, over 1602087.35 frames. ], batch size: 25, lr: 5.97e-03, grad_scale: 8.0 +2023-02-06 13:02:26,971 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98100.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:02:30,280 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98105.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:02:31,655 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98107.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:02:43,618 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98124.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:02:49,146 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98132.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:02:59,486 INFO [train.py:901] (3/4) Epoch 13, batch 1150, loss[loss=0.2216, simple_loss=0.3007, pruned_loss=0.07125, over 7797.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3025, pruned_loss=0.07476, over 1595818.88 frames. ], batch size: 20, lr: 5.97e-03, grad_scale: 8.0 +2023-02-06 13:03:02,371 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98151.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:03:05,443 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 13:03:06,089 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.693e+02 2.603e+02 3.101e+02 3.825e+02 7.832e+02, threshold=6.203e+02, percent-clipped=6.0 +2023-02-06 13:03:14,955 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7495, 2.0904, 2.2223, 1.2843, 2.3562, 1.5883, 0.6715, 1.8264], + device='cuda:3'), covar=tensor([0.0424, 0.0202, 0.0178, 0.0393, 0.0233, 0.0608, 0.0540, 0.0208], + device='cuda:3'), in_proj_covar=tensor([0.0402, 0.0337, 0.0289, 0.0397, 0.0325, 0.0484, 0.0360, 0.0362], + device='cuda:3'), out_proj_covar=tensor([1.1293e-04, 9.2114e-05, 7.9625e-05, 1.0987e-04, 9.0458e-05, 1.4467e-04, + 1.0148e-04, 1.0106e-04], device='cuda:3') +2023-02-06 13:03:34,172 INFO [train.py:901] (3/4) Epoch 13, batch 1200, loss[loss=0.2169, simple_loss=0.3035, pruned_loss=0.06513, over 8197.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.303, pruned_loss=0.07489, over 1599434.78 frames. ], batch size: 23, lr: 5.97e-03, grad_scale: 8.0 +2023-02-06 13:03:46,029 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 +2023-02-06 13:04:06,227 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2360, 1.1279, 3.3161, 1.0154, 2.8992, 2.7873, 2.9858, 2.9069], + device='cuda:3'), covar=tensor([0.0808, 0.4294, 0.0814, 0.4007, 0.1443, 0.1142, 0.0821, 0.0922], + device='cuda:3'), in_proj_covar=tensor([0.0510, 0.0587, 0.0598, 0.0544, 0.0618, 0.0531, 0.0522, 0.0579], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 13:04:08,153 INFO [train.py:901] (3/4) Epoch 13, batch 1250, loss[loss=0.1888, simple_loss=0.2716, pruned_loss=0.05304, over 8107.00 frames. ], tot_loss[loss=0.2271, simple_loss=0.3043, pruned_loss=0.07499, over 1607816.71 frames. ], batch size: 21, lr: 5.96e-03, grad_scale: 8.0 +2023-02-06 13:04:10,209 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98250.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:04:14,079 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.614e+02 2.568e+02 3.066e+02 4.053e+02 1.440e+03, threshold=6.132e+02, percent-clipped=8.0 +2023-02-06 13:04:36,959 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98288.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:04:43,430 INFO [train.py:901] (3/4) Epoch 13, batch 1300, loss[loss=0.3315, simple_loss=0.3863, pruned_loss=0.1383, over 8759.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.3049, pruned_loss=0.07514, over 1611686.98 frames. ], batch size: 30, lr: 5.96e-03, grad_scale: 8.0 +2023-02-06 13:04:44,922 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6068, 1.9531, 3.3564, 2.4926, 2.7591, 2.2891, 1.8838, 1.6206], + device='cuda:3'), covar=tensor([0.4203, 0.4749, 0.1264, 0.2908, 0.2221, 0.2517, 0.2001, 0.4584], + device='cuda:3'), in_proj_covar=tensor([0.0893, 0.0883, 0.0741, 0.0861, 0.0939, 0.0810, 0.0706, 0.0771], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 13:04:54,307 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6807, 1.3593, 1.5193, 1.2235, 0.9275, 1.2782, 1.4881, 1.5585], + device='cuda:3'), covar=tensor([0.0485, 0.1263, 0.1743, 0.1398, 0.0565, 0.1523, 0.0684, 0.0564], + device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0153, 0.0193, 0.0158, 0.0102, 0.0164, 0.0116, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 13:04:54,336 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98313.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:05:12,876 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4252, 2.7277, 1.9510, 2.2405, 2.1811, 1.6113, 1.8965, 2.1924], + device='cuda:3'), covar=tensor([0.1194, 0.0285, 0.0986, 0.0537, 0.0575, 0.1203, 0.0864, 0.0696], + device='cuda:3'), in_proj_covar=tensor([0.0338, 0.0231, 0.0313, 0.0295, 0.0296, 0.0317, 0.0334, 0.0298], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 13:05:16,646 INFO [train.py:901] (3/4) Epoch 13, batch 1350, loss[loss=0.2697, simple_loss=0.3451, pruned_loss=0.09716, over 8135.00 frames. ], tot_loss[loss=0.2272, simple_loss=0.304, pruned_loss=0.07519, over 1608375.17 frames. ], batch size: 22, lr: 5.96e-03, grad_scale: 8.0 +2023-02-06 13:05:23,233 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.808e+02 2.492e+02 3.102e+02 3.697e+02 5.327e+02, threshold=6.205e+02, percent-clipped=0.0 +2023-02-06 13:05:29,597 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98365.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:05:52,230 INFO [train.py:901] (3/4) Epoch 13, batch 1400, loss[loss=0.1912, simple_loss=0.2783, pruned_loss=0.05206, over 7659.00 frames. ], tot_loss[loss=0.2279, simple_loss=0.3041, pruned_loss=0.07587, over 1606972.15 frames. ], batch size: 19, lr: 5.96e-03, grad_scale: 8.0 +2023-02-06 13:05:59,420 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98407.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:06:14,075 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98428.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:06:16,859 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98432.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:06:26,901 INFO [train.py:901] (3/4) Epoch 13, batch 1450, loss[loss=0.2429, simple_loss=0.3159, pruned_loss=0.08494, over 8368.00 frames. ], tot_loss[loss=0.2294, simple_loss=0.3057, pruned_loss=0.07653, over 1610212.84 frames. ], batch size: 24, lr: 5.96e-03, grad_scale: 8.0 +2023-02-06 13:06:32,939 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.357e+02 2.293e+02 2.812e+02 3.491e+02 8.118e+02, threshold=5.625e+02, percent-clipped=1.0 +2023-02-06 13:06:34,308 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 13:06:41,227 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98468.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:07:02,158 INFO [train.py:901] (3/4) Epoch 13, batch 1500, loss[loss=0.1805, simple_loss=0.2612, pruned_loss=0.04989, over 7801.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.3051, pruned_loss=0.07618, over 1611751.02 frames. ], batch size: 20, lr: 5.96e-03, grad_scale: 8.0 +2023-02-06 13:07:28,254 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4710, 2.0045, 3.4034, 1.2493, 2.6813, 1.9756, 1.5726, 2.4062], + device='cuda:3'), covar=tensor([0.1853, 0.2211, 0.0753, 0.4215, 0.1461, 0.2978, 0.2013, 0.2055], + device='cuda:3'), in_proj_covar=tensor([0.0492, 0.0532, 0.0538, 0.0590, 0.0626, 0.0561, 0.0483, 0.0613], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 13:07:36,072 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8568, 3.8280, 3.4577, 1.9397, 3.3880, 3.4124, 3.4671, 3.2818], + device='cuda:3'), covar=tensor([0.0995, 0.0677, 0.1162, 0.4813, 0.1131, 0.1162, 0.1436, 0.1002], + device='cuda:3'), in_proj_covar=tensor([0.0466, 0.0382, 0.0391, 0.0483, 0.0383, 0.0388, 0.0380, 0.0337], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 13:07:37,387 INFO [train.py:901] (3/4) Epoch 13, batch 1550, loss[loss=0.2425, simple_loss=0.3082, pruned_loss=0.08843, over 8247.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.3033, pruned_loss=0.07519, over 1609914.85 frames. ], batch size: 22, lr: 5.96e-03, grad_scale: 8.0 +2023-02-06 13:07:43,378 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.767e+02 2.583e+02 3.208e+02 4.119e+02 6.608e+02, threshold=6.417e+02, percent-clipped=3.0 +2023-02-06 13:07:46,952 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98561.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:07:47,836 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.33 vs. limit=5.0 +2023-02-06 13:08:02,054 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98583.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:08:11,762 INFO [train.py:901] (3/4) Epoch 13, batch 1600, loss[loss=0.2201, simple_loss=0.3035, pruned_loss=0.06832, over 8345.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.303, pruned_loss=0.07458, over 1614030.91 frames. ], batch size: 26, lr: 5.95e-03, grad_scale: 8.0 +2023-02-06 13:08:27,713 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6983, 1.6161, 2.1721, 1.5981, 1.2522, 2.1569, 0.3905, 1.3040], + device='cuda:3'), covar=tensor([0.2253, 0.1595, 0.0553, 0.1719, 0.3787, 0.0467, 0.3247, 0.1911], + device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0175, 0.0104, 0.0221, 0.0262, 0.0112, 0.0164, 0.0168], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 13:08:29,744 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98621.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:08:46,812 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98646.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:08:47,299 INFO [train.py:901] (3/4) Epoch 13, batch 1650, loss[loss=0.1996, simple_loss=0.2817, pruned_loss=0.05882, over 7931.00 frames. ], tot_loss[loss=0.2273, simple_loss=0.3042, pruned_loss=0.07518, over 1615401.44 frames. ], batch size: 20, lr: 5.95e-03, grad_scale: 8.0 +2023-02-06 13:08:53,412 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.639e+02 2.462e+02 2.942e+02 3.707e+02 8.113e+02, threshold=5.885e+02, percent-clipped=6.0 +2023-02-06 13:09:07,752 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.69 vs. limit=5.0 +2023-02-06 13:09:20,792 INFO [train.py:901] (3/4) Epoch 13, batch 1700, loss[loss=0.2178, simple_loss=0.301, pruned_loss=0.06725, over 8467.00 frames. ], tot_loss[loss=0.2273, simple_loss=0.3043, pruned_loss=0.07519, over 1616510.00 frames. ], batch size: 25, lr: 5.95e-03, grad_scale: 8.0 +2023-02-06 13:09:49,369 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1171, 1.0593, 1.2627, 1.1790, 0.9048, 1.2453, 0.0665, 0.9368], + device='cuda:3'), covar=tensor([0.2627, 0.1875, 0.0737, 0.1498, 0.3815, 0.0776, 0.3122, 0.1676], + device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0175, 0.0104, 0.0220, 0.0260, 0.0112, 0.0163, 0.0167], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 13:09:57,388 INFO [train.py:901] (3/4) Epoch 13, batch 1750, loss[loss=0.2097, simple_loss=0.2884, pruned_loss=0.06552, over 8102.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.3036, pruned_loss=0.0747, over 1617124.54 frames. ], batch size: 23, lr: 5.95e-03, grad_scale: 8.0 +2023-02-06 13:10:03,433 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.629e+02 2.663e+02 3.311e+02 3.905e+02 7.561e+02, threshold=6.622e+02, percent-clipped=6.0 +2023-02-06 13:10:15,123 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98772.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:10:27,290 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4961, 1.1617, 4.7080, 1.6897, 4.0695, 3.9597, 4.2709, 4.1059], + device='cuda:3'), covar=tensor([0.0579, 0.4843, 0.0413, 0.3708, 0.1164, 0.0907, 0.0526, 0.0615], + device='cuda:3'), in_proj_covar=tensor([0.0510, 0.0584, 0.0597, 0.0550, 0.0620, 0.0532, 0.0522, 0.0585], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 13:10:31,875 INFO [train.py:901] (3/4) Epoch 13, batch 1800, loss[loss=0.2155, simple_loss=0.2927, pruned_loss=0.06918, over 8648.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3034, pruned_loss=0.07488, over 1614983.67 frames. ], batch size: 34, lr: 5.95e-03, grad_scale: 8.0 +2023-02-06 13:11:01,380 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98839.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:11:06,994 INFO [train.py:901] (3/4) Epoch 13, batch 1850, loss[loss=0.288, simple_loss=0.35, pruned_loss=0.113, over 8040.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.3046, pruned_loss=0.0759, over 1616533.89 frames. ], batch size: 22, lr: 5.95e-03, grad_scale: 8.0 +2023-02-06 13:11:13,499 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.643e+02 2.419e+02 2.914e+02 4.067e+02 1.078e+03, threshold=5.828e+02, percent-clipped=2.0 +2023-02-06 13:11:18,923 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98864.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:11:34,565 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98887.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:11:41,190 INFO [train.py:901] (3/4) Epoch 13, batch 1900, loss[loss=0.2791, simple_loss=0.3504, pruned_loss=0.1039, over 8727.00 frames. ], tot_loss[loss=0.227, simple_loss=0.3041, pruned_loss=0.07501, over 1617408.44 frames. ], batch size: 30, lr: 5.95e-03, grad_scale: 8.0 +2023-02-06 13:11:46,604 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=98905.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:11:48,684 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98908.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:11:51,822 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98913.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:12:02,109 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0693, 2.4922, 2.8431, 1.3775, 3.0546, 1.7781, 1.4855, 2.0655], + device='cuda:3'), covar=tensor([0.0697, 0.0284, 0.0229, 0.0609, 0.0351, 0.0726, 0.0710, 0.0404], + device='cuda:3'), in_proj_covar=tensor([0.0400, 0.0335, 0.0291, 0.0399, 0.0327, 0.0485, 0.0361, 0.0364], + device='cuda:3'), out_proj_covar=tensor([1.1216e-04, 9.1455e-05, 8.0008e-05, 1.1021e-04, 9.0817e-05, 1.4455e-04, + 1.0175e-04, 1.0132e-04], device='cuda:3') +2023-02-06 13:12:04,112 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9202, 1.3815, 1.6020, 1.2226, 0.9618, 1.5121, 2.1214, 1.8572], + device='cuda:3'), covar=tensor([0.0454, 0.1749, 0.2420, 0.1875, 0.0668, 0.2089, 0.0745, 0.0668], + device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0153, 0.0192, 0.0158, 0.0102, 0.0164, 0.0115, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 13:12:11,197 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 13:12:15,142 INFO [train.py:901] (3/4) Epoch 13, batch 1950, loss[loss=0.2295, simple_loss=0.2948, pruned_loss=0.08212, over 7698.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.305, pruned_loss=0.07528, over 1621599.16 frames. ], batch size: 18, lr: 5.94e-03, grad_scale: 8.0 +2023-02-06 13:12:21,303 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.605e+02 2.445e+02 3.079e+02 3.874e+02 6.986e+02, threshold=6.158e+02, percent-clipped=4.0 +2023-02-06 13:12:23,945 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 13:12:33,016 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2223, 1.5460, 1.6062, 1.4674, 1.0731, 1.4408, 1.7546, 1.4104], + device='cuda:3'), covar=tensor([0.0518, 0.1158, 0.1618, 0.1286, 0.0607, 0.1422, 0.0670, 0.0615], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0152, 0.0191, 0.0157, 0.0102, 0.0163, 0.0115, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 13:12:44,575 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 13:12:45,395 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98989.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:12:50,409 INFO [train.py:901] (3/4) Epoch 13, batch 2000, loss[loss=0.2684, simple_loss=0.3502, pruned_loss=0.09331, over 8488.00 frames. ], tot_loss[loss=0.2289, simple_loss=0.3055, pruned_loss=0.0761, over 1619339.03 frames. ], batch size: 49, lr: 5.94e-03, grad_scale: 8.0 +2023-02-06 13:12:51,992 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7567, 2.0136, 2.1923, 1.2083, 2.4054, 1.5355, 0.7095, 1.9520], + device='cuda:3'), covar=tensor([0.0484, 0.0243, 0.0195, 0.0448, 0.0250, 0.0663, 0.0612, 0.0233], + device='cuda:3'), in_proj_covar=tensor([0.0398, 0.0334, 0.0289, 0.0398, 0.0324, 0.0482, 0.0360, 0.0363], + device='cuda:3'), out_proj_covar=tensor([1.1172e-04, 9.1195e-05, 7.9623e-05, 1.0982e-04, 9.0171e-05, 1.4381e-04, + 1.0126e-04, 1.0103e-04], device='cuda:3') +2023-02-06 13:13:06,441 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99020.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:13:24,250 INFO [train.py:901] (3/4) Epoch 13, batch 2050, loss[loss=0.197, simple_loss=0.2699, pruned_loss=0.06208, over 7705.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3054, pruned_loss=0.07563, over 1621362.50 frames. ], batch size: 18, lr: 5.94e-03, grad_scale: 8.0 +2023-02-06 13:13:30,116 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.519e+02 2.527e+02 3.267e+02 4.166e+02 9.227e+02, threshold=6.535e+02, percent-clipped=8.0 +2023-02-06 13:13:39,430 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99069.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:13:58,771 INFO [train.py:901] (3/4) Epoch 13, batch 2100, loss[loss=0.2024, simple_loss=0.2897, pruned_loss=0.05752, over 8024.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.3047, pruned_loss=0.07526, over 1618748.19 frames. ], batch size: 22, lr: 5.94e-03, grad_scale: 8.0 +2023-02-06 13:14:00,877 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3436, 1.6991, 2.7029, 1.1487, 2.0320, 1.6173, 1.4158, 1.9488], + device='cuda:3'), covar=tensor([0.1820, 0.2051, 0.0719, 0.3989, 0.1514, 0.2993, 0.1943, 0.1955], + device='cuda:3'), in_proj_covar=tensor([0.0491, 0.0533, 0.0539, 0.0592, 0.0624, 0.0560, 0.0485, 0.0613], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 13:14:17,686 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2331, 1.6570, 4.4151, 2.1878, 2.4457, 5.0811, 5.0039, 4.3230], + device='cuda:3'), covar=tensor([0.1093, 0.1655, 0.0272, 0.1752, 0.1055, 0.0158, 0.0329, 0.0582], + device='cuda:3'), in_proj_covar=tensor([0.0269, 0.0300, 0.0268, 0.0291, 0.0276, 0.0237, 0.0357, 0.0293], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 13:14:31,273 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99143.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:14:33,883 INFO [train.py:901] (3/4) Epoch 13, batch 2150, loss[loss=0.2427, simple_loss=0.3212, pruned_loss=0.08212, over 8674.00 frames. ], tot_loss[loss=0.2289, simple_loss=0.306, pruned_loss=0.07593, over 1620410.05 frames. ], batch size: 39, lr: 5.94e-03, grad_scale: 8.0 +2023-02-06 13:14:39,911 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.727e+02 2.393e+02 2.767e+02 3.323e+02 5.467e+02, threshold=5.533e+02, percent-clipped=0.0 +2023-02-06 13:14:48,073 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99168.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:15:07,844 INFO [train.py:901] (3/4) Epoch 13, batch 2200, loss[loss=0.2114, simple_loss=0.2817, pruned_loss=0.0705, over 7254.00 frames. ], tot_loss[loss=0.2304, simple_loss=0.3067, pruned_loss=0.07705, over 1621057.73 frames. ], batch size: 16, lr: 5.94e-03, grad_scale: 8.0 +2023-02-06 13:15:39,425 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 +2023-02-06 13:15:43,036 INFO [train.py:901] (3/4) Epoch 13, batch 2250, loss[loss=0.2335, simple_loss=0.3068, pruned_loss=0.08006, over 7969.00 frames. ], tot_loss[loss=0.23, simple_loss=0.3066, pruned_loss=0.07673, over 1621761.25 frames. ], batch size: 21, lr: 5.93e-03, grad_scale: 8.0 +2023-02-06 13:15:43,157 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99247.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:15:46,460 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99252.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:15:48,887 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.852e+02 2.623e+02 3.377e+02 4.135e+02 6.545e+02, threshold=6.753e+02, percent-clipped=6.0 +2023-02-06 13:15:49,683 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99257.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:16:02,600 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99276.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:16:09,216 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99286.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:16:16,493 INFO [train.py:901] (3/4) Epoch 13, batch 2300, loss[loss=0.2243, simple_loss=0.3169, pruned_loss=0.06588, over 8451.00 frames. ], tot_loss[loss=0.2292, simple_loss=0.3059, pruned_loss=0.07621, over 1619694.83 frames. ], batch size: 27, lr: 5.93e-03, grad_scale: 8.0 +2023-02-06 13:16:19,363 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99301.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:16:43,129 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99333.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:16:49,191 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99342.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:16:52,410 INFO [train.py:901] (3/4) Epoch 13, batch 2350, loss[loss=0.2131, simple_loss=0.2829, pruned_loss=0.07164, over 7968.00 frames. ], tot_loss[loss=0.229, simple_loss=0.3061, pruned_loss=0.07596, over 1622109.76 frames. ], batch size: 21, lr: 5.93e-03, grad_scale: 8.0 +2023-02-06 13:16:58,394 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.470e+02 3.074e+02 3.865e+02 1.080e+03, threshold=6.149e+02, percent-clipped=3.0 +2023-02-06 13:17:06,750 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99367.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:17:10,177 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99372.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:17:26,585 INFO [train.py:901] (3/4) Epoch 13, batch 2400, loss[loss=0.1901, simple_loss=0.2643, pruned_loss=0.05794, over 7802.00 frames. ], tot_loss[loss=0.2279, simple_loss=0.3051, pruned_loss=0.07541, over 1622690.43 frames. ], batch size: 19, lr: 5.93e-03, grad_scale: 8.0 +2023-02-06 13:17:26,707 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4593, 4.4578, 4.0334, 1.8967, 4.0300, 3.9797, 4.0854, 3.8506], + device='cuda:3'), covar=tensor([0.0765, 0.0573, 0.1038, 0.4563, 0.0804, 0.0861, 0.1182, 0.0721], + device='cuda:3'), in_proj_covar=tensor([0.0479, 0.0393, 0.0397, 0.0497, 0.0391, 0.0394, 0.0384, 0.0343], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 13:17:37,540 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99413.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:17:56,001 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99438.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:18:01,826 INFO [train.py:901] (3/4) Epoch 13, batch 2450, loss[loss=0.2314, simple_loss=0.3063, pruned_loss=0.07831, over 7973.00 frames. ], tot_loss[loss=0.2277, simple_loss=0.3046, pruned_loss=0.07537, over 1619027.38 frames. ], batch size: 21, lr: 5.93e-03, grad_scale: 8.0 +2023-02-06 13:18:02,728 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99448.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:18:08,608 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.521e+02 2.541e+02 3.157e+02 3.793e+02 6.756e+02, threshold=6.314e+02, percent-clipped=3.0 +2023-02-06 13:18:29,568 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99486.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:18:36,513 INFO [train.py:901] (3/4) Epoch 13, batch 2500, loss[loss=0.203, simple_loss=0.2849, pruned_loss=0.0606, over 8075.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.3055, pruned_loss=0.07581, over 1621318.05 frames. ], batch size: 21, lr: 5.93e-03, grad_scale: 8.0 +2023-02-06 13:18:57,697 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99528.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:19:10,261 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0725, 1.5749, 3.2589, 1.5242, 2.3058, 3.5581, 3.6377, 3.0154], + device='cuda:3'), covar=tensor([0.0963, 0.1477, 0.0335, 0.1888, 0.0970, 0.0265, 0.0496, 0.0648], + device='cuda:3'), in_proj_covar=tensor([0.0274, 0.0304, 0.0271, 0.0294, 0.0282, 0.0242, 0.0361, 0.0298], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 13:19:10,744 INFO [train.py:901] (3/4) Epoch 13, batch 2550, loss[loss=0.2609, simple_loss=0.3275, pruned_loss=0.09717, over 8351.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3046, pruned_loss=0.0752, over 1620802.08 frames. ], batch size: 24, lr: 5.93e-03, grad_scale: 8.0 +2023-02-06 13:19:17,196 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.560e+02 2.420e+02 2.977e+02 3.875e+02 7.325e+02, threshold=5.954e+02, percent-clipped=4.0 +2023-02-06 13:19:37,759 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99586.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:19:41,078 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99591.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:19:45,636 INFO [train.py:901] (3/4) Epoch 13, batch 2600, loss[loss=0.2017, simple_loss=0.2703, pruned_loss=0.06653, over 7534.00 frames. ], tot_loss[loss=0.2289, simple_loss=0.3052, pruned_loss=0.07633, over 1617758.13 frames. ], batch size: 18, lr: 5.92e-03, grad_scale: 8.0 +2023-02-06 13:19:47,529 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 +2023-02-06 13:19:52,602 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99607.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:20:03,523 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99623.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:20:06,915 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99628.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:20:08,162 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99630.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:20:19,558 INFO [train.py:901] (3/4) Epoch 13, batch 2650, loss[loss=0.2444, simple_loss=0.3318, pruned_loss=0.07853, over 8357.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.3067, pruned_loss=0.07634, over 1619094.69 frames. ], batch size: 24, lr: 5.92e-03, grad_scale: 8.0 +2023-02-06 13:20:20,455 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99648.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:20:23,645 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99653.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:20:25,432 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.509e+02 2.403e+02 3.099e+02 4.031e+02 8.160e+02, threshold=6.198e+02, percent-clipped=1.0 +2023-02-06 13:20:47,384 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99686.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:20:54,454 INFO [train.py:901] (3/4) Epoch 13, batch 2700, loss[loss=0.2454, simple_loss=0.3164, pruned_loss=0.08719, over 8460.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.3056, pruned_loss=0.07625, over 1615711.49 frames. ], batch size: 29, lr: 5.92e-03, grad_scale: 8.0 +2023-02-06 13:20:55,216 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99698.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:20:59,197 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99704.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:21:00,473 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99706.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:21:16,804 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99729.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:21:27,698 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99745.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:21:28,866 INFO [train.py:901] (3/4) Epoch 13, batch 2750, loss[loss=0.2145, simple_loss=0.2952, pruned_loss=0.0669, over 7972.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.3044, pruned_loss=0.07541, over 1613968.45 frames. ], batch size: 21, lr: 5.92e-03, grad_scale: 8.0 +2023-02-06 13:21:34,775 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 2.508e+02 3.194e+02 3.866e+02 8.318e+02, threshold=6.387e+02, percent-clipped=3.0 +2023-02-06 13:21:49,875 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5484, 2.0050, 3.2217, 1.3132, 2.4767, 1.9078, 1.6576, 2.2513], + device='cuda:3'), covar=tensor([0.1718, 0.2109, 0.0693, 0.3962, 0.1510, 0.2855, 0.1782, 0.2177], + device='cuda:3'), in_proj_covar=tensor([0.0489, 0.0530, 0.0539, 0.0588, 0.0622, 0.0559, 0.0481, 0.0612], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 13:21:52,465 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99782.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:21:53,980 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99784.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:22:03,192 INFO [train.py:901] (3/4) Epoch 13, batch 2800, loss[loss=0.1977, simple_loss=0.2655, pruned_loss=0.06501, over 7552.00 frames. ], tot_loss[loss=0.2271, simple_loss=0.3036, pruned_loss=0.07527, over 1611382.66 frames. ], batch size: 18, lr: 5.92e-03, grad_scale: 8.0 +2023-02-06 13:22:06,162 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99801.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:22:11,458 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99809.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:22:26,108 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99830.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:22:37,608 INFO [train.py:901] (3/4) Epoch 13, batch 2850, loss[loss=0.2261, simple_loss=0.3013, pruned_loss=0.07546, over 8194.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.304, pruned_loss=0.07554, over 1613099.33 frames. ], batch size: 23, lr: 5.92e-03, grad_scale: 8.0 +2023-02-06 13:22:43,878 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.542e+02 2.486e+02 2.909e+02 3.673e+02 9.445e+02, threshold=5.818e+02, percent-clipped=3.0 +2023-02-06 13:23:11,490 INFO [train.py:901] (3/4) Epoch 13, batch 2900, loss[loss=0.2389, simple_loss=0.2998, pruned_loss=0.08894, over 7934.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.3041, pruned_loss=0.07611, over 1611832.31 frames. ], batch size: 20, lr: 5.92e-03, grad_scale: 16.0 +2023-02-06 13:23:11,689 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99897.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:23:34,895 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99930.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:23:45,024 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99945.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:23:46,142 INFO [train.py:901] (3/4) Epoch 13, batch 2950, loss[loss=0.2422, simple_loss=0.3164, pruned_loss=0.08398, over 8249.00 frames. ], tot_loss[loss=0.2287, simple_loss=0.3051, pruned_loss=0.07614, over 1615572.31 frames. ], batch size: 24, lr: 5.91e-03, grad_scale: 16.0 +2023-02-06 13:23:48,998 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=99951.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:23:52,274 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 13:23:52,905 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.593e+02 2.762e+02 3.280e+02 4.150e+02 8.176e+02, threshold=6.560e+02, percent-clipped=12.0 +2023-02-06 13:23:57,303 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99962.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:24:14,500 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99987.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:24:21,167 INFO [train.py:901] (3/4) Epoch 13, batch 3000, loss[loss=0.2449, simple_loss=0.3257, pruned_loss=0.08205, over 8505.00 frames. ], tot_loss[loss=0.2272, simple_loss=0.3044, pruned_loss=0.07498, over 1616562.02 frames. ], batch size: 26, lr: 5.91e-03, grad_scale: 16.0 +2023-02-06 13:24:21,167 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 13:24:33,564 INFO [train.py:935] (3/4) Epoch 13, validation: loss=0.1841, simple_loss=0.2841, pruned_loss=0.04204, over 944034.00 frames. +2023-02-06 13:24:33,565 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 13:24:37,726 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100001.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:24:42,517 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2548, 1.2473, 1.4249, 1.3184, 0.7209, 1.3012, 1.2310, 0.9432], + device='cuda:3'), covar=tensor([0.0523, 0.1246, 0.1704, 0.1293, 0.0579, 0.1478, 0.0664, 0.0689], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0153, 0.0192, 0.0158, 0.0102, 0.0163, 0.0115, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 13:24:54,616 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100025.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:24:55,295 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100026.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:25:05,694 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100042.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:25:07,828 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100045.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:25:09,024 INFO [train.py:901] (3/4) Epoch 13, batch 3050, loss[loss=0.2146, simple_loss=0.3007, pruned_loss=0.06424, over 8516.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.3041, pruned_loss=0.07467, over 1617547.09 frames. ], batch size: 28, lr: 5.91e-03, grad_scale: 16.0 +2023-02-06 13:25:15,849 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.507e+02 2.522e+02 3.008e+02 4.207e+02 1.157e+03, threshold=6.017e+02, percent-clipped=6.0 +2023-02-06 13:25:16,790 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100057.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:25:22,893 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100066.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:25:24,274 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4077, 1.4132, 4.5142, 1.6677, 3.9290, 3.7039, 4.0566, 3.9632], + device='cuda:3'), covar=tensor([0.0499, 0.4631, 0.0491, 0.3779, 0.1209, 0.0992, 0.0595, 0.0601], + device='cuda:3'), in_proj_covar=tensor([0.0516, 0.0591, 0.0609, 0.0558, 0.0634, 0.0546, 0.0538, 0.0591], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 13:25:34,607 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100082.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:25:41,470 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5726, 1.5598, 2.0316, 1.5314, 1.1694, 2.0494, 0.3045, 1.2675], + device='cuda:3'), covar=tensor([0.2260, 0.1800, 0.0483, 0.1554, 0.3869, 0.0588, 0.2871, 0.1612], + device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0177, 0.0106, 0.0222, 0.0260, 0.0112, 0.0166, 0.0170], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 13:25:44,733 INFO [train.py:901] (3/4) Epoch 13, batch 3100, loss[loss=0.2613, simple_loss=0.3435, pruned_loss=0.08953, over 8337.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3056, pruned_loss=0.07571, over 1621455.52 frames. ], batch size: 25, lr: 5.91e-03, grad_scale: 16.0 +2023-02-06 13:26:19,849 INFO [train.py:901] (3/4) Epoch 13, batch 3150, loss[loss=0.1956, simple_loss=0.2684, pruned_loss=0.06135, over 7798.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3058, pruned_loss=0.07561, over 1622238.61 frames. ], batch size: 20, lr: 5.91e-03, grad_scale: 16.0 +2023-02-06 13:26:24,151 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100153.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:26:25,928 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.665e+02 2.491e+02 3.036e+02 4.077e+02 6.258e+02, threshold=6.072e+02, percent-clipped=1.0 +2023-02-06 13:26:26,821 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100157.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:26:41,739 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100178.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:26:55,157 INFO [train.py:901] (3/4) Epoch 13, batch 3200, loss[loss=0.1974, simple_loss=0.277, pruned_loss=0.05894, over 7800.00 frames. ], tot_loss[loss=0.2277, simple_loss=0.3052, pruned_loss=0.07512, over 1623995.09 frames. ], batch size: 19, lr: 5.91e-03, grad_scale: 16.0 +2023-02-06 13:26:58,262 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100201.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:27:05,247 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0656, 1.5003, 1.5873, 1.4132, 0.8729, 1.3586, 1.6882, 1.7437], + device='cuda:3'), covar=tensor([0.0496, 0.1298, 0.1749, 0.1397, 0.0628, 0.1594, 0.0719, 0.0574], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0153, 0.0191, 0.0158, 0.0102, 0.0163, 0.0114, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 13:27:15,352 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100226.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:27:29,402 INFO [train.py:901] (3/4) Epoch 13, batch 3250, loss[loss=0.232, simple_loss=0.2918, pruned_loss=0.08609, over 7769.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.3057, pruned_loss=0.07553, over 1621393.27 frames. ], batch size: 19, lr: 5.91e-03, grad_scale: 16.0 +2023-02-06 13:27:34,944 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.7865, 5.8909, 5.0720, 2.2439, 5.1893, 5.5666, 5.3079, 5.2733], + device='cuda:3'), covar=tensor([0.0463, 0.0341, 0.0833, 0.4390, 0.0599, 0.0577, 0.1022, 0.0482], + device='cuda:3'), in_proj_covar=tensor([0.0475, 0.0387, 0.0391, 0.0490, 0.0388, 0.0392, 0.0385, 0.0342], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 13:27:35,465 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.691e+02 2.430e+02 2.991e+02 3.670e+02 7.489e+02, threshold=5.982e+02, percent-clipped=4.0 +2023-02-06 13:27:35,622 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100256.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:28:04,524 INFO [train.py:901] (3/4) Epoch 13, batch 3300, loss[loss=0.1854, simple_loss=0.2661, pruned_loss=0.05228, over 8239.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.3048, pruned_loss=0.075, over 1619080.53 frames. ], batch size: 22, lr: 5.90e-03, grad_scale: 16.0 +2023-02-06 13:28:07,518 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100301.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:28:22,246 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100322.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:28:24,891 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100326.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:28:35,233 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-06 13:28:39,553 INFO [train.py:901] (3/4) Epoch 13, batch 3350, loss[loss=0.2702, simple_loss=0.3298, pruned_loss=0.1053, over 6775.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.3045, pruned_loss=0.07536, over 1614020.67 frames. ], batch size: 71, lr: 5.90e-03, grad_scale: 16.0 +2023-02-06 13:28:39,774 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100347.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:28:45,562 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.501e+02 2.599e+02 3.166e+02 3.997e+02 7.990e+02, threshold=6.333e+02, percent-clipped=6.0 +2023-02-06 13:28:54,365 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100369.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:29:13,504 INFO [train.py:901] (3/4) Epoch 13, batch 3400, loss[loss=0.2413, simple_loss=0.321, pruned_loss=0.08082, over 8459.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3047, pruned_loss=0.07513, over 1617238.24 frames. ], batch size: 25, lr: 5.90e-03, grad_scale: 16.0 +2023-02-06 13:29:25,260 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100413.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:29:29,923 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100420.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:29:42,788 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100438.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:29:44,445 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 +2023-02-06 13:29:48,800 INFO [train.py:901] (3/4) Epoch 13, batch 3450, loss[loss=0.2127, simple_loss=0.3004, pruned_loss=0.06256, over 7807.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3047, pruned_loss=0.07518, over 1618293.91 frames. ], batch size: 20, lr: 5.90e-03, grad_scale: 16.0 +2023-02-06 13:29:54,814 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.653e+02 2.523e+02 3.011e+02 4.006e+02 7.808e+02, threshold=6.023e+02, percent-clipped=2.0 +2023-02-06 13:30:14,320 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100484.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:30:23,422 INFO [train.py:901] (3/4) Epoch 13, batch 3500, loss[loss=0.2387, simple_loss=0.3194, pruned_loss=0.079, over 8134.00 frames. ], tot_loss[loss=0.227, simple_loss=0.3042, pruned_loss=0.07488, over 1618786.10 frames. ], batch size: 22, lr: 5.90e-03, grad_scale: 16.0 +2023-02-06 13:30:53,068 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 13:30:58,421 INFO [train.py:901] (3/4) Epoch 13, batch 3550, loss[loss=0.2231, simple_loss=0.3106, pruned_loss=0.06777, over 8254.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.3047, pruned_loss=0.07498, over 1621386.10 frames. ], batch size: 24, lr: 5.90e-03, grad_scale: 16.0 +2023-02-06 13:31:04,451 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.567e+02 2.564e+02 3.091e+02 3.906e+02 9.185e+02, threshold=6.182e+02, percent-clipped=3.0 +2023-02-06 13:31:33,168 INFO [train.py:901] (3/4) Epoch 13, batch 3600, loss[loss=0.2172, simple_loss=0.3007, pruned_loss=0.06689, over 8283.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3058, pruned_loss=0.07538, over 1625399.75 frames. ], batch size: 23, lr: 5.89e-03, grad_scale: 16.0 +2023-02-06 13:31:35,334 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100600.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:31:56,141 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2395, 1.7909, 1.4323, 1.6352, 1.4869, 1.2687, 1.4671, 1.5239], + device='cuda:3'), covar=tensor([0.0864, 0.0334, 0.0709, 0.0373, 0.0509, 0.0945, 0.0640, 0.0594], + device='cuda:3'), in_proj_covar=tensor([0.0341, 0.0233, 0.0317, 0.0297, 0.0296, 0.0322, 0.0339, 0.0301], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 13:32:08,080 INFO [train.py:901] (3/4) Epoch 13, batch 3650, loss[loss=0.2869, simple_loss=0.3476, pruned_loss=0.1131, over 8710.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.3049, pruned_loss=0.07534, over 1619566.21 frames. ], batch size: 34, lr: 5.89e-03, grad_scale: 16.0 +2023-02-06 13:32:12,319 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8478, 1.6415, 1.8281, 1.6396, 1.1271, 1.6457, 2.2850, 1.9637], + device='cuda:3'), covar=tensor([0.0462, 0.1282, 0.1669, 0.1361, 0.0638, 0.1517, 0.0630, 0.0626], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0152, 0.0191, 0.0158, 0.0102, 0.0163, 0.0115, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 13:32:14,110 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.507e+02 2.362e+02 3.080e+02 3.827e+02 7.938e+02, threshold=6.161e+02, percent-clipped=3.0 +2023-02-06 13:32:17,467 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-06 13:32:34,201 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.74 vs. limit=5.0 +2023-02-06 13:32:43,157 INFO [train.py:901] (3/4) Epoch 13, batch 3700, loss[loss=0.2397, simple_loss=0.3162, pruned_loss=0.08158, over 8030.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.3053, pruned_loss=0.07557, over 1619830.35 frames. ], batch size: 22, lr: 5.89e-03, grad_scale: 8.0 +2023-02-06 13:32:52,912 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-06 13:32:55,266 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100715.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:32:57,152 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 13:33:13,284 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100740.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:33:16,645 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4090, 2.6976, 2.0342, 2.1881, 2.0584, 1.5926, 1.8904, 2.0627], + device='cuda:3'), covar=tensor([0.1454, 0.0335, 0.0894, 0.0586, 0.0743, 0.1388, 0.1058, 0.0911], + device='cuda:3'), in_proj_covar=tensor([0.0344, 0.0234, 0.0318, 0.0299, 0.0300, 0.0322, 0.0340, 0.0303], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 13:33:17,849 INFO [train.py:901] (3/4) Epoch 13, batch 3750, loss[loss=0.2234, simple_loss=0.3044, pruned_loss=0.07119, over 8295.00 frames. ], tot_loss[loss=0.227, simple_loss=0.3042, pruned_loss=0.07491, over 1611969.65 frames. ], batch size: 23, lr: 5.89e-03, grad_scale: 8.0 +2023-02-06 13:33:18,689 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100748.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:33:24,691 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.669e+02 2.543e+02 3.029e+02 3.909e+02 6.778e+02, threshold=6.059e+02, percent-clipped=2.0 +2023-02-06 13:33:30,137 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=100764.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:33:30,993 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100765.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:33:52,954 INFO [train.py:901] (3/4) Epoch 13, batch 3800, loss[loss=0.1928, simple_loss=0.271, pruned_loss=0.05733, over 7930.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.3048, pruned_loss=0.07499, over 1613769.44 frames. ], batch size: 20, lr: 5.89e-03, grad_scale: 8.0 +2023-02-06 13:33:53,763 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3932, 1.3845, 4.5598, 1.7442, 4.1263, 3.8271, 4.0993, 4.0075], + device='cuda:3'), covar=tensor([0.0522, 0.4133, 0.0422, 0.3303, 0.0888, 0.0775, 0.0480, 0.0536], + device='cuda:3'), in_proj_covar=tensor([0.0518, 0.0595, 0.0619, 0.0557, 0.0633, 0.0546, 0.0537, 0.0597], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 13:34:00,247 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2023-02-06 13:34:27,590 INFO [train.py:901] (3/4) Epoch 13, batch 3850, loss[loss=0.2164, simple_loss=0.2895, pruned_loss=0.07168, over 7711.00 frames. ], tot_loss[loss=0.2279, simple_loss=0.305, pruned_loss=0.07539, over 1612504.06 frames. ], batch size: 18, lr: 5.89e-03, grad_scale: 8.0 +2023-02-06 13:34:34,506 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.985e+02 2.830e+02 3.312e+02 3.730e+02 7.453e+02, threshold=6.624e+02, percent-clipped=3.0 +2023-02-06 13:34:50,099 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100879.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:35:01,753 INFO [train.py:901] (3/4) Epoch 13, batch 3900, loss[loss=0.2117, simple_loss=0.2886, pruned_loss=0.06741, over 7816.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.3051, pruned_loss=0.07528, over 1611872.97 frames. ], batch size: 20, lr: 5.89e-03, grad_scale: 8.0 +2023-02-06 13:35:01,759 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 13:35:33,290 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1735, 1.8785, 2.6404, 2.0880, 2.4468, 2.0670, 1.7852, 1.3088], + device='cuda:3'), covar=tensor([0.4175, 0.3976, 0.1330, 0.2818, 0.1987, 0.2467, 0.1681, 0.4246], + device='cuda:3'), in_proj_covar=tensor([0.0891, 0.0887, 0.0734, 0.0862, 0.0941, 0.0813, 0.0705, 0.0776], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 13:35:37,125 INFO [train.py:901] (3/4) Epoch 13, batch 3950, loss[loss=0.2036, simple_loss=0.2689, pruned_loss=0.06913, over 7793.00 frames. ], tot_loss[loss=0.2272, simple_loss=0.3044, pruned_loss=0.07494, over 1605479.46 frames. ], batch size: 19, lr: 5.88e-03, grad_scale: 8.0 +2023-02-06 13:35:44,004 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.582e+02 2.489e+02 3.011e+02 3.855e+02 9.802e+02, threshold=6.021e+02, percent-clipped=2.0 +2023-02-06 13:35:54,142 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100971.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:36:10,612 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.7894, 2.2705, 4.8835, 2.7370, 4.4809, 4.2453, 4.5772, 4.5191], + device='cuda:3'), covar=tensor([0.0413, 0.3369, 0.0515, 0.2784, 0.0829, 0.0726, 0.0424, 0.0463], + device='cuda:3'), in_proj_covar=tensor([0.0513, 0.0588, 0.0612, 0.0551, 0.0630, 0.0543, 0.0533, 0.0593], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 13:36:11,899 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100996.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:36:12,365 INFO [train.py:901] (3/4) Epoch 13, batch 4000, loss[loss=0.2331, simple_loss=0.3103, pruned_loss=0.07793, over 8338.00 frames. ], tot_loss[loss=0.228, simple_loss=0.305, pruned_loss=0.0755, over 1606342.97 frames. ], batch size: 26, lr: 5.88e-03, grad_scale: 8.0 +2023-02-06 13:36:20,492 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.10 vs. limit=5.0 +2023-02-06 13:36:47,684 INFO [train.py:901] (3/4) Epoch 13, batch 4050, loss[loss=0.2185, simple_loss=0.3019, pruned_loss=0.06758, over 8743.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3054, pruned_loss=0.07582, over 1608819.07 frames. ], batch size: 30, lr: 5.88e-03, grad_scale: 8.0 +2023-02-06 13:36:54,254 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.767e+02 2.645e+02 3.184e+02 3.816e+02 9.518e+02, threshold=6.368e+02, percent-clipped=3.0 +2023-02-06 13:37:18,351 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=101092.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:37:21,619 INFO [train.py:901] (3/4) Epoch 13, batch 4100, loss[loss=0.2464, simple_loss=0.3148, pruned_loss=0.08899, over 8081.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3056, pruned_loss=0.07568, over 1611857.03 frames. ], batch size: 21, lr: 5.88e-03, grad_scale: 8.0 +2023-02-06 13:37:41,404 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101125.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:37:47,946 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101135.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:37:55,651 INFO [train.py:901] (3/4) Epoch 13, batch 4150, loss[loss=0.2367, simple_loss=0.3249, pruned_loss=0.07423, over 8192.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3056, pruned_loss=0.07573, over 1610008.59 frames. ], batch size: 23, lr: 5.88e-03, grad_scale: 8.0 +2023-02-06 13:38:02,992 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.662e+02 2.783e+02 3.401e+02 4.642e+02 1.010e+03, threshold=6.803e+02, percent-clipped=7.0 +2023-02-06 13:38:05,283 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101160.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:38:05,891 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.2825, 4.1699, 3.7782, 1.8403, 3.7992, 3.7811, 3.7676, 3.4950], + device='cuda:3'), covar=tensor([0.0670, 0.0485, 0.0895, 0.4393, 0.0713, 0.0828, 0.1173, 0.0840], + device='cuda:3'), in_proj_covar=tensor([0.0479, 0.0391, 0.0401, 0.0500, 0.0394, 0.0395, 0.0389, 0.0344], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 13:38:06,622 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3080, 1.1640, 1.4252, 1.1442, 0.6540, 1.2030, 1.1315, 1.1511], + device='cuda:3'), covar=tensor([0.0581, 0.1780, 0.2350, 0.1761, 0.0671, 0.2025, 0.0796, 0.0723], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0153, 0.0192, 0.0157, 0.0102, 0.0164, 0.0116, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 13:38:30,242 INFO [train.py:901] (3/4) Epoch 13, batch 4200, loss[loss=0.22, simple_loss=0.2949, pruned_loss=0.07257, over 8355.00 frames. ], tot_loss[loss=0.228, simple_loss=0.3051, pruned_loss=0.0755, over 1614327.42 frames. ], batch size: 24, lr: 5.88e-03, grad_scale: 8.0 +2023-02-06 13:38:37,998 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101207.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:38:39,766 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 +2023-02-06 13:38:54,606 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 13:39:05,859 INFO [train.py:901] (3/4) Epoch 13, batch 4250, loss[loss=0.2054, simple_loss=0.2831, pruned_loss=0.06381, over 7932.00 frames. ], tot_loss[loss=0.2293, simple_loss=0.3057, pruned_loss=0.07648, over 1610592.33 frames. ], batch size: 20, lr: 5.88e-03, grad_scale: 8.0 +2023-02-06 13:39:12,480 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.424e+02 2.514e+02 3.154e+02 3.992e+02 7.648e+02, threshold=6.307e+02, percent-clipped=3.0 +2023-02-06 13:39:16,501 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 13:39:40,012 INFO [train.py:901] (3/4) Epoch 13, batch 4300, loss[loss=0.2229, simple_loss=0.3046, pruned_loss=0.0706, over 8503.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.3054, pruned_loss=0.07592, over 1613145.40 frames. ], batch size: 26, lr: 5.87e-03, grad_scale: 8.0 +2023-02-06 13:39:58,266 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 13:40:01,018 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 +2023-02-06 13:40:14,581 INFO [train.py:901] (3/4) Epoch 13, batch 4350, loss[loss=0.2213, simple_loss=0.2922, pruned_loss=0.07517, over 7807.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.3046, pruned_loss=0.07551, over 1610940.74 frames. ], batch size: 20, lr: 5.87e-03, grad_scale: 8.0 +2023-02-06 13:40:14,813 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1293, 1.9405, 2.4209, 1.8341, 1.6675, 2.4740, 1.1199, 2.0625], + device='cuda:3'), covar=tensor([0.2603, 0.1502, 0.0479, 0.1799, 0.2995, 0.0428, 0.2515, 0.1446], + device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0176, 0.0107, 0.0219, 0.0256, 0.0111, 0.0164, 0.0170], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 13:40:21,344 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.618e+02 2.682e+02 3.184e+02 4.441e+02 9.358e+02, threshold=6.368e+02, percent-clipped=11.0 +2023-02-06 13:40:49,426 INFO [train.py:901] (3/4) Epoch 13, batch 4400, loss[loss=0.2267, simple_loss=0.3053, pruned_loss=0.07407, over 7814.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.306, pruned_loss=0.07615, over 1614401.06 frames. ], batch size: 20, lr: 5.87e-03, grad_scale: 8.0 +2023-02-06 13:40:49,433 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 13:41:23,441 INFO [train.py:901] (3/4) Epoch 13, batch 4450, loss[loss=0.2073, simple_loss=0.302, pruned_loss=0.0563, over 8107.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3053, pruned_loss=0.07583, over 1613655.63 frames. ], batch size: 23, lr: 5.87e-03, grad_scale: 8.0 +2023-02-06 13:41:28,823 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 13:41:30,657 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.969e+02 2.700e+02 3.319e+02 4.103e+02 1.285e+03, threshold=6.638e+02, percent-clipped=3.0 +2023-02-06 13:41:34,866 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101463.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:41:38,577 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=101469.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:41:52,069 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101488.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:41:57,886 INFO [train.py:901] (3/4) Epoch 13, batch 4500, loss[loss=0.2583, simple_loss=0.3251, pruned_loss=0.0957, over 8445.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.3043, pruned_loss=0.07546, over 1610539.74 frames. ], batch size: 27, lr: 5.87e-03, grad_scale: 8.0 +2023-02-06 13:42:22,192 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 13:42:33,107 INFO [train.py:901] (3/4) Epoch 13, batch 4550, loss[loss=0.2018, simple_loss=0.2866, pruned_loss=0.0585, over 7808.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.3028, pruned_loss=0.07495, over 1605907.53 frames. ], batch size: 20, lr: 5.87e-03, grad_scale: 8.0 +2023-02-06 13:42:39,887 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.533e+02 2.403e+02 2.986e+02 3.546e+02 6.918e+02, threshold=5.973e+02, percent-clipped=1.0 +2023-02-06 13:42:50,291 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9708, 1.6854, 2.1663, 1.8544, 2.0014, 1.9421, 1.7198, 0.7386], + device='cuda:3'), covar=tensor([0.4874, 0.4022, 0.1529, 0.2607, 0.1927, 0.2481, 0.1863, 0.4157], + device='cuda:3'), in_proj_covar=tensor([0.0894, 0.0893, 0.0747, 0.0869, 0.0947, 0.0819, 0.0710, 0.0779], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 13:42:58,932 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101584.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:43:08,193 INFO [train.py:901] (3/4) Epoch 13, batch 4600, loss[loss=0.2331, simple_loss=0.311, pruned_loss=0.0776, over 8086.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.303, pruned_loss=0.07484, over 1611063.44 frames. ], batch size: 21, lr: 5.87e-03, grad_scale: 8.0 +2023-02-06 13:43:42,601 INFO [train.py:901] (3/4) Epoch 13, batch 4650, loss[loss=0.194, simple_loss=0.2786, pruned_loss=0.05473, over 8289.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3026, pruned_loss=0.07469, over 1602826.94 frames. ], batch size: 23, lr: 5.86e-03, grad_scale: 8.0 +2023-02-06 13:43:49,456 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.782e+02 2.500e+02 2.989e+02 3.844e+02 7.619e+02, threshold=5.978e+02, percent-clipped=4.0 +2023-02-06 13:43:53,084 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1774, 1.4815, 1.7274, 1.4315, 1.0002, 1.4253, 1.6361, 1.7376], + device='cuda:3'), covar=tensor([0.0445, 0.1264, 0.1594, 0.1328, 0.0573, 0.1489, 0.0681, 0.0591], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0152, 0.0191, 0.0157, 0.0102, 0.0162, 0.0115, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 13:44:17,684 INFO [train.py:901] (3/4) Epoch 13, batch 4700, loss[loss=0.2233, simple_loss=0.3124, pruned_loss=0.06709, over 8322.00 frames. ], tot_loss[loss=0.225, simple_loss=0.3019, pruned_loss=0.07409, over 1605434.04 frames. ], batch size: 25, lr: 5.86e-03, grad_scale: 8.0 +2023-02-06 13:44:21,754 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101702.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:44:52,092 INFO [train.py:901] (3/4) Epoch 13, batch 4750, loss[loss=0.259, simple_loss=0.3229, pruned_loss=0.09756, over 8024.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3031, pruned_loss=0.07444, over 1613947.95 frames. ], batch size: 22, lr: 5.86e-03, grad_scale: 8.0 +2023-02-06 13:44:59,504 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.642e+02 2.558e+02 3.081e+02 3.778e+02 8.564e+02, threshold=6.162e+02, percent-clipped=2.0 +2023-02-06 13:45:21,976 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 13:45:24,392 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 13:45:27,097 INFO [train.py:901] (3/4) Epoch 13, batch 4800, loss[loss=0.1806, simple_loss=0.2507, pruned_loss=0.05524, over 7720.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3033, pruned_loss=0.07494, over 1613788.09 frames. ], batch size: 18, lr: 5.86e-03, grad_scale: 8.0 +2023-02-06 13:45:57,597 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101840.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:46:01,949 INFO [train.py:901] (3/4) Epoch 13, batch 4850, loss[loss=0.2153, simple_loss=0.2859, pruned_loss=0.07233, over 7544.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.304, pruned_loss=0.07541, over 1613478.21 frames. ], batch size: 18, lr: 5.86e-03, grad_scale: 8.0 +2023-02-06 13:46:08,645 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 2.589e+02 3.137e+02 3.918e+02 7.572e+02, threshold=6.274e+02, percent-clipped=4.0 +2023-02-06 13:46:14,079 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 13:46:14,950 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101865.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:46:36,897 INFO [train.py:901] (3/4) Epoch 13, batch 4900, loss[loss=0.2701, simple_loss=0.3404, pruned_loss=0.09988, over 8320.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.3048, pruned_loss=0.07536, over 1617843.54 frames. ], batch size: 25, lr: 5.86e-03, grad_scale: 8.0 +2023-02-06 13:46:52,175 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2023-02-06 13:47:06,384 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101938.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 13:47:11,284 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 13:47:12,287 INFO [train.py:901] (3/4) Epoch 13, batch 4950, loss[loss=0.2503, simple_loss=0.3121, pruned_loss=0.0943, over 8133.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.3036, pruned_loss=0.07475, over 1615943.77 frames. ], batch size: 22, lr: 5.86e-03, grad_scale: 8.0 +2023-02-06 13:47:15,216 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3568, 2.4712, 1.7360, 2.0981, 1.9437, 1.3744, 1.7630, 1.8690], + device='cuda:3'), covar=tensor([0.1258, 0.0347, 0.0974, 0.0501, 0.0558, 0.1347, 0.0873, 0.0757], + device='cuda:3'), in_proj_covar=tensor([0.0343, 0.0233, 0.0314, 0.0293, 0.0296, 0.0321, 0.0339, 0.0302], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 13:47:19,104 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.579e+02 2.432e+02 3.023e+02 3.670e+02 7.494e+02, threshold=6.046e+02, percent-clipped=3.0 +2023-02-06 13:47:30,008 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9846, 1.7684, 1.9721, 1.7739, 0.9893, 1.9108, 2.0279, 2.1629], + device='cuda:3'), covar=tensor([0.0424, 0.1135, 0.1513, 0.1276, 0.0583, 0.1337, 0.0623, 0.0527], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0152, 0.0191, 0.0157, 0.0102, 0.0163, 0.0114, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 13:47:46,752 INFO [train.py:901] (3/4) Epoch 13, batch 5000, loss[loss=0.2388, simple_loss=0.3151, pruned_loss=0.08122, over 8135.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3031, pruned_loss=0.07405, over 1618446.45 frames. ], batch size: 22, lr: 5.85e-03, grad_scale: 8.0 +2023-02-06 13:47:53,376 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102005.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:48:22,479 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=102046.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:48:23,077 INFO [train.py:901] (3/4) Epoch 13, batch 5050, loss[loss=0.1836, simple_loss=0.2569, pruned_loss=0.05518, over 7640.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.3026, pruned_loss=0.07449, over 1613346.98 frames. ], batch size: 19, lr: 5.85e-03, grad_scale: 8.0 +2023-02-06 13:48:29,968 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.678e+02 2.626e+02 3.300e+02 4.185e+02 9.088e+02, threshold=6.599e+02, percent-clipped=3.0 +2023-02-06 13:48:50,354 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0 +2023-02-06 13:48:54,012 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 13:48:56,651 INFO [train.py:901] (3/4) Epoch 13, batch 5100, loss[loss=0.2231, simple_loss=0.3093, pruned_loss=0.06845, over 8503.00 frames. ], tot_loss[loss=0.2254, simple_loss=0.302, pruned_loss=0.07436, over 1611603.82 frames. ], batch size: 26, lr: 5.85e-03, grad_scale: 8.0 +2023-02-06 13:49:31,561 INFO [train.py:901] (3/4) Epoch 13, batch 5150, loss[loss=0.2204, simple_loss=0.3026, pruned_loss=0.06907, over 8612.00 frames. ], tot_loss[loss=0.2255, simple_loss=0.3021, pruned_loss=0.07446, over 1611954.01 frames. ], batch size: 31, lr: 5.85e-03, grad_scale: 8.0 +2023-02-06 13:49:38,299 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.486e+02 2.413e+02 2.853e+02 3.425e+02 7.647e+02, threshold=5.706e+02, percent-clipped=3.0 +2023-02-06 13:49:42,467 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102161.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:49:50,399 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 13:50:06,641 INFO [train.py:901] (3/4) Epoch 13, batch 5200, loss[loss=0.2053, simple_loss=0.2975, pruned_loss=0.05661, over 8503.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.302, pruned_loss=0.07476, over 1607808.63 frames. ], batch size: 28, lr: 5.85e-03, grad_scale: 8.0 +2023-02-06 13:50:41,849 INFO [train.py:901] (3/4) Epoch 13, batch 5250, loss[loss=0.209, simple_loss=0.2899, pruned_loss=0.06406, over 8462.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.303, pruned_loss=0.07476, over 1610898.26 frames. ], batch size: 25, lr: 5.85e-03, grad_scale: 8.0 +2023-02-06 13:50:48,577 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.646e+02 2.565e+02 3.047e+02 3.925e+02 1.157e+03, threshold=6.094e+02, percent-clipped=6.0 +2023-02-06 13:50:51,017 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.91 vs. limit=5.0 +2023-02-06 13:50:53,900 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 13:51:06,840 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=102282.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:51:16,471 INFO [train.py:901] (3/4) Epoch 13, batch 5300, loss[loss=0.2186, simple_loss=0.3013, pruned_loss=0.0679, over 8498.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.3029, pruned_loss=0.07494, over 1609070.48 frames. ], batch size: 28, lr: 5.85e-03, grad_scale: 8.0 +2023-02-06 13:51:51,008 INFO [train.py:901] (3/4) Epoch 13, batch 5350, loss[loss=0.2105, simple_loss=0.2916, pruned_loss=0.06477, over 7927.00 frames. ], tot_loss[loss=0.2255, simple_loss=0.3023, pruned_loss=0.07438, over 1613046.94 frames. ], batch size: 20, lr: 5.84e-03, grad_scale: 8.0 +2023-02-06 13:51:52,484 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=102349.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:51:52,613 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0792, 1.6458, 1.3778, 1.5342, 1.3170, 1.2369, 1.2616, 1.3869], + device='cuda:3'), covar=tensor([0.1048, 0.0482, 0.1225, 0.0551, 0.0761, 0.1446, 0.0913, 0.0671], + device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0236, 0.0318, 0.0297, 0.0298, 0.0324, 0.0345, 0.0304], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 13:51:57,792 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.587e+02 2.535e+02 3.049e+02 3.805e+02 7.372e+02, threshold=6.098e+02, percent-clipped=2.0 +2023-02-06 13:52:08,202 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.53 vs. limit=2.0 +2023-02-06 13:52:26,058 INFO [train.py:901] (3/4) Epoch 13, batch 5400, loss[loss=0.2176, simple_loss=0.2987, pruned_loss=0.06827, over 8505.00 frames. ], tot_loss[loss=0.226, simple_loss=0.303, pruned_loss=0.0745, over 1611697.67 frames. ], batch size: 39, lr: 5.84e-03, grad_scale: 8.0 +2023-02-06 13:52:26,255 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102397.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:52:40,303 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102417.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:52:56,887 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102442.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:53:00,140 INFO [train.py:901] (3/4) Epoch 13, batch 5450, loss[loss=0.1985, simple_loss=0.2747, pruned_loss=0.0612, over 7664.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.3036, pruned_loss=0.07512, over 1612910.73 frames. ], batch size: 19, lr: 5.84e-03, grad_scale: 8.0 +2023-02-06 13:53:07,658 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.617e+02 2.724e+02 3.222e+02 3.900e+02 7.023e+02, threshold=6.444e+02, percent-clipped=3.0 +2023-02-06 13:53:12,602 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102464.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:53:34,976 INFO [train.py:901] (3/4) Epoch 13, batch 5500, loss[loss=0.2134, simple_loss=0.3011, pruned_loss=0.06284, over 8572.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.303, pruned_loss=0.07434, over 1617046.64 frames. ], batch size: 31, lr: 5.84e-03, grad_scale: 8.0 +2023-02-06 13:53:41,588 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 13:54:03,104 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7895, 1.6289, 2.8478, 1.3005, 2.1314, 3.1236, 3.1766, 2.6567], + device='cuda:3'), covar=tensor([0.1002, 0.1249, 0.0401, 0.2035, 0.0883, 0.0283, 0.0535, 0.0613], + device='cuda:3'), in_proj_covar=tensor([0.0268, 0.0300, 0.0265, 0.0296, 0.0277, 0.0237, 0.0360, 0.0295], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 13:54:09,472 INFO [train.py:901] (3/4) Epoch 13, batch 5550, loss[loss=0.2326, simple_loss=0.3037, pruned_loss=0.0807, over 7787.00 frames. ], tot_loss[loss=0.2273, simple_loss=0.3039, pruned_loss=0.07537, over 1617659.32 frames. ], batch size: 19, lr: 5.84e-03, grad_scale: 8.0 +2023-02-06 13:54:15,944 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.373e+02 2.279e+02 3.010e+02 3.933e+02 6.976e+02, threshold=6.019e+02, percent-clipped=1.0 +2023-02-06 13:54:43,194 INFO [train.py:901] (3/4) Epoch 13, batch 5600, loss[loss=0.2004, simple_loss=0.2882, pruned_loss=0.05624, over 8105.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3047, pruned_loss=0.07597, over 1615506.73 frames. ], batch size: 23, lr: 5.84e-03, grad_scale: 8.0 +2023-02-06 13:54:44,705 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8219, 1.1465, 1.3444, 1.1078, 0.9728, 1.1021, 1.6591, 1.5532], + device='cuda:3'), covar=tensor([0.0609, 0.1861, 0.2455, 0.1893, 0.0738, 0.2194, 0.0825, 0.0733], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0151, 0.0190, 0.0156, 0.0101, 0.0162, 0.0113, 0.0137], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 13:55:04,665 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5794, 1.9012, 1.9876, 1.0410, 2.0470, 1.4711, 0.4060, 1.8645], + device='cuda:3'), covar=tensor([0.0285, 0.0191, 0.0144, 0.0340, 0.0216, 0.0564, 0.0527, 0.0141], + device='cuda:3'), in_proj_covar=tensor([0.0406, 0.0341, 0.0293, 0.0401, 0.0332, 0.0490, 0.0368, 0.0373], + device='cuda:3'), out_proj_covar=tensor([1.1347e-04, 9.3025e-05, 7.9922e-05, 1.1031e-04, 9.1562e-05, 1.4528e-04, + 1.0327e-04, 1.0324e-04], device='cuda:3') +2023-02-06 13:55:18,276 INFO [train.py:901] (3/4) Epoch 13, batch 5650, loss[loss=0.2872, simple_loss=0.3502, pruned_loss=0.1121, over 8100.00 frames. ], tot_loss[loss=0.23, simple_loss=0.3059, pruned_loss=0.07701, over 1615927.48 frames. ], batch size: 23, lr: 5.84e-03, grad_scale: 8.0 +2023-02-06 13:55:22,473 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102653.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:55:24,851 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.521e+02 2.730e+02 3.267e+02 4.266e+02 8.129e+02, threshold=6.534e+02, percent-clipped=5.0 +2023-02-06 13:55:39,202 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102678.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 13:55:43,627 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 13:55:52,498 INFO [train.py:901] (3/4) Epoch 13, batch 5700, loss[loss=0.1974, simple_loss=0.2809, pruned_loss=0.057, over 7657.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3048, pruned_loss=0.07614, over 1619413.13 frames. ], batch size: 19, lr: 5.83e-03, grad_scale: 16.0 +2023-02-06 13:56:08,740 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102720.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:56:25,991 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102745.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:56:27,756 INFO [train.py:901] (3/4) Epoch 13, batch 5750, loss[loss=0.2045, simple_loss=0.2775, pruned_loss=0.06572, over 7639.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3063, pruned_loss=0.07715, over 1620874.27 frames. ], batch size: 19, lr: 5.83e-03, grad_scale: 16.0 +2023-02-06 13:56:34,421 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.889e+02 2.514e+02 3.075e+02 4.012e+02 7.214e+02, threshold=6.150e+02, percent-clipped=2.0 +2023-02-06 13:56:45,826 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.39 vs. limit=5.0 +2023-02-06 13:56:47,305 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 13:56:49,636 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6399, 2.2588, 3.2867, 2.7333, 2.9792, 2.3982, 2.1057, 2.0409], + device='cuda:3'), covar=tensor([0.3747, 0.4084, 0.1501, 0.2753, 0.2310, 0.2442, 0.1642, 0.4338], + device='cuda:3'), in_proj_covar=tensor([0.0894, 0.0889, 0.0744, 0.0870, 0.0943, 0.0815, 0.0705, 0.0778], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 13:56:52,945 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7605, 2.0583, 2.2514, 1.2527, 2.3119, 1.6350, 0.6490, 1.9400], + device='cuda:3'), covar=tensor([0.0459, 0.0233, 0.0153, 0.0448, 0.0266, 0.0650, 0.0624, 0.0213], + device='cuda:3'), in_proj_covar=tensor([0.0407, 0.0342, 0.0295, 0.0403, 0.0334, 0.0492, 0.0369, 0.0374], + device='cuda:3'), out_proj_covar=tensor([1.1393e-04, 9.3301e-05, 8.0550e-05, 1.1094e-04, 9.2128e-05, 1.4583e-04, + 1.0362e-04, 1.0363e-04], device='cuda:3') +2023-02-06 13:57:01,391 INFO [train.py:901] (3/4) Epoch 13, batch 5800, loss[loss=0.2846, simple_loss=0.3564, pruned_loss=0.1064, over 8280.00 frames. ], tot_loss[loss=0.2287, simple_loss=0.3053, pruned_loss=0.07602, over 1619032.90 frames. ], batch size: 23, lr: 5.83e-03, grad_scale: 16.0 +2023-02-06 13:57:36,616 INFO [train.py:901] (3/4) Epoch 13, batch 5850, loss[loss=0.2241, simple_loss=0.2985, pruned_loss=0.07489, over 8237.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.3044, pruned_loss=0.07522, over 1616303.32 frames. ], batch size: 22, lr: 5.83e-03, grad_scale: 16.0 +2023-02-06 13:57:43,166 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.557e+02 2.218e+02 2.874e+02 3.517e+02 7.476e+02, threshold=5.748e+02, percent-clipped=3.0 +2023-02-06 13:57:52,584 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102869.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:58:11,324 INFO [train.py:901] (3/4) Epoch 13, batch 5900, loss[loss=0.1878, simple_loss=0.2724, pruned_loss=0.05157, over 7430.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.3044, pruned_loss=0.07525, over 1614830.41 frames. ], batch size: 17, lr: 5.83e-03, grad_scale: 16.0 +2023-02-06 13:58:16,412 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.09 vs. limit=5.0 +2023-02-06 13:58:16,872 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102905.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 13:58:42,656 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.08 vs. limit=5.0 +2023-02-06 13:58:46,267 INFO [train.py:901] (3/4) Epoch 13, batch 5950, loss[loss=0.2513, simple_loss=0.3211, pruned_loss=0.09077, over 7813.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.3048, pruned_loss=0.07535, over 1616623.52 frames. ], batch size: 20, lr: 5.83e-03, grad_scale: 16.0 +2023-02-06 13:58:52,894 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.606e+02 2.537e+02 3.124e+02 4.010e+02 1.248e+03, threshold=6.247e+02, percent-clipped=9.0 +2023-02-06 13:58:59,020 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0013, 2.5798, 2.9731, 1.3530, 3.1329, 1.7155, 1.5145, 1.9367], + device='cuda:3'), covar=tensor([0.0756, 0.0327, 0.0243, 0.0637, 0.0439, 0.0780, 0.0799, 0.0508], + device='cuda:3'), in_proj_covar=tensor([0.0410, 0.0346, 0.0296, 0.0406, 0.0334, 0.0496, 0.0372, 0.0376], + device='cuda:3'), out_proj_covar=tensor([1.1459e-04, 9.4281e-05, 8.0591e-05, 1.1148e-04, 9.2291e-05, 1.4726e-04, + 1.0437e-04, 1.0401e-04], device='cuda:3') +2023-02-06 13:59:21,495 INFO [train.py:901] (3/4) Epoch 13, batch 6000, loss[loss=0.201, simple_loss=0.2854, pruned_loss=0.05833, over 8569.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.3051, pruned_loss=0.07521, over 1618365.07 frames. ], batch size: 39, lr: 5.83e-03, grad_scale: 16.0 +2023-02-06 13:59:21,495 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 13:59:36,606 INFO [train.py:935] (3/4) Epoch 13, validation: loss=0.1836, simple_loss=0.2836, pruned_loss=0.04176, over 944034.00 frames. +2023-02-06 13:59:36,607 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 14:00:11,030 INFO [train.py:901] (3/4) Epoch 13, batch 6050, loss[loss=0.1824, simple_loss=0.2691, pruned_loss=0.04785, over 8142.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.3039, pruned_loss=0.07452, over 1616545.17 frames. ], batch size: 22, lr: 5.82e-03, grad_scale: 16.0 +2023-02-06 14:00:14,465 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9622, 2.6055, 3.6925, 1.8973, 1.5600, 3.6286, 0.6700, 2.1874], + device='cuda:3'), covar=tensor([0.2033, 0.1578, 0.0341, 0.3040, 0.4093, 0.0385, 0.3335, 0.1883], + device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0173, 0.0106, 0.0218, 0.0254, 0.0110, 0.0164, 0.0170], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 14:00:18,307 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.701e+02 2.480e+02 3.014e+02 3.999e+02 8.436e+02, threshold=6.027e+02, percent-clipped=4.0 +2023-02-06 14:00:21,130 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.8318, 5.8201, 5.1099, 2.2076, 5.1363, 5.6698, 5.4283, 5.2389], + device='cuda:3'), covar=tensor([0.0567, 0.0422, 0.0910, 0.5203, 0.0732, 0.0704, 0.1154, 0.0641], + device='cuda:3'), in_proj_covar=tensor([0.0479, 0.0391, 0.0397, 0.0495, 0.0391, 0.0392, 0.0389, 0.0343], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 14:00:45,627 INFO [train.py:901] (3/4) Epoch 13, batch 6100, loss[loss=0.2337, simple_loss=0.3124, pruned_loss=0.07752, over 8289.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.3038, pruned_loss=0.07423, over 1611879.41 frames. ], batch size: 23, lr: 5.82e-03, grad_scale: 16.0 +2023-02-06 14:00:58,589 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103116.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:01:14,184 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 14:01:19,569 INFO [train.py:901] (3/4) Epoch 13, batch 6150, loss[loss=0.2231, simple_loss=0.3069, pruned_loss=0.06966, over 8472.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.304, pruned_loss=0.0748, over 1611259.49 frames. ], batch size: 29, lr: 5.82e-03, grad_scale: 16.0 +2023-02-06 14:01:21,559 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103150.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:01:26,195 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.695e+02 2.434e+02 3.117e+02 4.172e+02 7.466e+02, threshold=6.235e+02, percent-clipped=2.0 +2023-02-06 14:01:46,190 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7871, 1.5065, 2.7516, 1.3005, 2.0098, 2.9485, 2.9966, 2.4841], + device='cuda:3'), covar=tensor([0.1044, 0.1468, 0.0463, 0.2101, 0.0988, 0.0326, 0.0688, 0.0722], + device='cuda:3'), in_proj_covar=tensor([0.0268, 0.0303, 0.0265, 0.0297, 0.0278, 0.0239, 0.0361, 0.0295], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 14:01:51,720 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2969, 1.5950, 4.3056, 2.0908, 2.3578, 4.9692, 4.9713, 4.2574], + device='cuda:3'), covar=tensor([0.1103, 0.1621, 0.0297, 0.1738, 0.1153, 0.0177, 0.0404, 0.0568], + device='cuda:3'), in_proj_covar=tensor([0.0267, 0.0301, 0.0264, 0.0296, 0.0277, 0.0239, 0.0359, 0.0294], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 14:01:55,079 INFO [train.py:901] (3/4) Epoch 13, batch 6200, loss[loss=0.1981, simple_loss=0.2838, pruned_loss=0.05622, over 7972.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.3036, pruned_loss=0.07486, over 1614531.89 frames. ], batch size: 21, lr: 5.82e-03, grad_scale: 16.0 +2023-02-06 14:02:06,162 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103213.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:02:30,996 INFO [train.py:901] (3/4) Epoch 13, batch 6250, loss[loss=0.2414, simple_loss=0.3128, pruned_loss=0.08499, over 7977.00 frames. ], tot_loss[loss=0.225, simple_loss=0.3021, pruned_loss=0.07396, over 1609100.67 frames. ], batch size: 21, lr: 5.82e-03, grad_scale: 16.0 +2023-02-06 14:02:32,364 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103249.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:02:37,840 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.552e+02 2.482e+02 2.950e+02 3.630e+02 6.819e+02, threshold=5.900e+02, percent-clipped=4.0 +2023-02-06 14:03:06,056 INFO [train.py:901] (3/4) Epoch 13, batch 6300, loss[loss=0.2121, simple_loss=0.2828, pruned_loss=0.07069, over 7921.00 frames. ], tot_loss[loss=0.2255, simple_loss=0.3023, pruned_loss=0.0744, over 1607620.57 frames. ], batch size: 20, lr: 5.82e-03, grad_scale: 16.0 +2023-02-06 14:03:15,730 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6991, 1.6925, 2.3100, 1.6318, 1.2381, 2.2767, 0.2585, 1.3065], + device='cuda:3'), covar=tensor([0.2329, 0.1396, 0.0384, 0.1632, 0.3415, 0.0393, 0.2876, 0.1749], + device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0173, 0.0107, 0.0221, 0.0256, 0.0110, 0.0166, 0.0173], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 14:03:21,245 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 14:03:27,928 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103328.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:03:40,545 INFO [train.py:901] (3/4) Epoch 13, batch 6350, loss[loss=0.2249, simple_loss=0.2829, pruned_loss=0.08346, over 7216.00 frames. ], tot_loss[loss=0.2273, simple_loss=0.3037, pruned_loss=0.07538, over 1606373.56 frames. ], batch size: 16, lr: 5.82e-03, grad_scale: 16.0 +2023-02-06 14:03:48,225 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.865e+02 2.547e+02 3.093e+02 3.716e+02 8.603e+02, threshold=6.185e+02, percent-clipped=3.0 +2023-02-06 14:03:53,004 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103364.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:04:06,349 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103384.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 14:04:14,878 INFO [train.py:901] (3/4) Epoch 13, batch 6400, loss[loss=0.2029, simple_loss=0.2883, pruned_loss=0.05874, over 8652.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.3032, pruned_loss=0.07508, over 1607335.36 frames. ], batch size: 34, lr: 5.81e-03, grad_scale: 16.0 +2023-02-06 14:04:49,443 INFO [train.py:901] (3/4) Epoch 13, batch 6450, loss[loss=0.2406, simple_loss=0.3028, pruned_loss=0.08922, over 7929.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3024, pruned_loss=0.07473, over 1606250.33 frames. ], batch size: 20, lr: 5.81e-03, grad_scale: 16.0 +2023-02-06 14:04:56,168 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.890e+02 2.528e+02 3.186e+02 3.863e+02 6.544e+02, threshold=6.372e+02, percent-clipped=1.0 +2023-02-06 14:04:58,241 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103460.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:05:13,647 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1337, 1.3480, 1.6735, 1.2600, 0.9632, 1.4540, 1.7768, 1.9002], + device='cuda:3'), covar=tensor([0.0465, 0.1339, 0.1722, 0.1460, 0.0625, 0.1541, 0.0693, 0.0562], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0150, 0.0189, 0.0155, 0.0100, 0.0161, 0.0113, 0.0137], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 14:05:22,247 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103494.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:05:24,240 INFO [train.py:901] (3/4) Epoch 13, batch 6500, loss[loss=0.2099, simple_loss=0.2728, pruned_loss=0.07346, over 7654.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.3031, pruned_loss=0.07497, over 1610821.17 frames. ], batch size: 19, lr: 5.81e-03, grad_scale: 16.0 +2023-02-06 14:05:58,663 INFO [train.py:901] (3/4) Epoch 13, batch 6550, loss[loss=0.2568, simple_loss=0.3385, pruned_loss=0.08757, over 8519.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3051, pruned_loss=0.07579, over 1614521.93 frames. ], batch size: 28, lr: 5.81e-03, grad_scale: 16.0 +2023-02-06 14:06:05,482 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.593e+02 2.442e+02 3.089e+02 4.027e+02 9.292e+02, threshold=6.177e+02, percent-clipped=8.0 +2023-02-06 14:06:12,688 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1665, 4.1179, 3.7717, 1.8509, 3.6687, 3.6915, 3.7393, 3.5703], + device='cuda:3'), covar=tensor([0.0743, 0.0643, 0.1134, 0.4844, 0.0909, 0.1060, 0.1341, 0.0882], + device='cuda:3'), in_proj_covar=tensor([0.0476, 0.0389, 0.0399, 0.0498, 0.0394, 0.0392, 0.0385, 0.0343], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 14:06:18,304 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103575.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:06:24,184 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 14:06:24,394 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103584.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:06:34,118 INFO [train.py:901] (3/4) Epoch 13, batch 6600, loss[loss=0.2249, simple_loss=0.3045, pruned_loss=0.07267, over 8499.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.3055, pruned_loss=0.07609, over 1617221.88 frames. ], batch size: 28, lr: 5.81e-03, grad_scale: 16.0 +2023-02-06 14:06:36,425 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9306, 1.3090, 1.5124, 1.2818, 0.8941, 1.3296, 1.6834, 1.3148], + device='cuda:3'), covar=tensor([0.0494, 0.1275, 0.1676, 0.1374, 0.0609, 0.1563, 0.0681, 0.0721], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0150, 0.0189, 0.0154, 0.0100, 0.0161, 0.0114, 0.0137], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 14:06:37,032 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9964, 1.8706, 2.0584, 1.9109, 1.1002, 1.9155, 2.4064, 2.5282], + device='cuda:3'), covar=tensor([0.0434, 0.1071, 0.1510, 0.1146, 0.0582, 0.1308, 0.0569, 0.0495], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0150, 0.0189, 0.0154, 0.0100, 0.0161, 0.0114, 0.0137], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0008, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 14:06:42,479 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103609.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:06:42,503 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103609.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:06:43,680 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 14:06:49,834 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103620.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:07:02,408 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1997, 1.8436, 2.7486, 2.1631, 2.4787, 2.1147, 1.7960, 1.3589], + device='cuda:3'), covar=tensor([0.4381, 0.4014, 0.1287, 0.2951, 0.2116, 0.2436, 0.1796, 0.4211], + device='cuda:3'), in_proj_covar=tensor([0.0896, 0.0887, 0.0742, 0.0866, 0.0945, 0.0822, 0.0706, 0.0774], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 14:07:07,875 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103645.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:07:09,056 INFO [train.py:901] (3/4) Epoch 13, batch 6650, loss[loss=0.2265, simple_loss=0.3235, pruned_loss=0.06472, over 8370.00 frames. ], tot_loss[loss=0.2292, simple_loss=0.3061, pruned_loss=0.0762, over 1615849.45 frames. ], batch size: 24, lr: 5.81e-03, grad_scale: 16.0 +2023-02-06 14:07:16,596 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.654e+02 2.459e+02 2.800e+02 3.637e+02 6.016e+02, threshold=5.600e+02, percent-clipped=0.0 +2023-02-06 14:07:43,988 INFO [train.py:901] (3/4) Epoch 13, batch 6700, loss[loss=0.2391, simple_loss=0.3033, pruned_loss=0.08742, over 7799.00 frames. ], tot_loss[loss=0.2279, simple_loss=0.3046, pruned_loss=0.07562, over 1617243.71 frames. ], batch size: 19, lr: 5.81e-03, grad_scale: 16.0 +2023-02-06 14:08:06,486 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=103728.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 14:08:18,549 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-06 14:08:19,450 INFO [train.py:901] (3/4) Epoch 13, batch 6750, loss[loss=0.3149, simple_loss=0.3723, pruned_loss=0.1288, over 6986.00 frames. ], tot_loss[loss=0.2287, simple_loss=0.3054, pruned_loss=0.07603, over 1613600.67 frames. ], batch size: 72, lr: 5.80e-03, grad_scale: 16.0 +2023-02-06 14:08:24,989 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0240, 2.1888, 1.7721, 2.7399, 1.2875, 1.6759, 1.8839, 2.2350], + device='cuda:3'), covar=tensor([0.0714, 0.0809, 0.1036, 0.0359, 0.1168, 0.1367, 0.0958, 0.0773], + device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0205, 0.0250, 0.0208, 0.0213, 0.0249, 0.0252, 0.0213], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 14:08:26,133 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.761e+02 2.551e+02 3.234e+02 3.983e+02 1.044e+03, threshold=6.469e+02, percent-clipped=6.0 +2023-02-06 14:08:26,993 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103758.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:08:44,326 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103783.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:08:53,558 INFO [train.py:901] (3/4) Epoch 13, batch 6800, loss[loss=0.239, simple_loss=0.3295, pruned_loss=0.07427, over 8462.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.3041, pruned_loss=0.07542, over 1610263.69 frames. ], batch size: 27, lr: 5.80e-03, grad_scale: 16.0 +2023-02-06 14:08:58,344 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 14:09:17,167 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103831.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:09:19,856 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7840, 1.8954, 1.6255, 2.2802, 1.0653, 1.5168, 1.6154, 1.9413], + device='cuda:3'), covar=tensor([0.0683, 0.0769, 0.0965, 0.0418, 0.1175, 0.1272, 0.0899, 0.0750], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0205, 0.0251, 0.0208, 0.0213, 0.0249, 0.0253, 0.0213], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 14:09:25,231 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103843.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 14:09:27,549 INFO [train.py:901] (3/4) Epoch 13, batch 6850, loss[loss=0.2351, simple_loss=0.2998, pruned_loss=0.08518, over 7306.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.3043, pruned_loss=0.0756, over 1609375.63 frames. ], batch size: 16, lr: 5.80e-03, grad_scale: 16.0 +2023-02-06 14:09:33,735 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103856.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:09:34,154 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.955e+02 2.670e+02 3.153e+02 3.957e+02 9.275e+02, threshold=6.306e+02, percent-clipped=2.0 +2023-02-06 14:09:40,250 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103865.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:09:44,804 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 14:09:46,354 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5149, 2.2537, 3.3628, 2.4197, 2.9927, 2.4981, 2.1913, 1.8290], + device='cuda:3'), covar=tensor([0.4572, 0.4857, 0.1515, 0.3349, 0.2489, 0.2589, 0.1830, 0.5036], + device='cuda:3'), in_proj_covar=tensor([0.0899, 0.0887, 0.0742, 0.0866, 0.0948, 0.0822, 0.0705, 0.0777], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 14:09:57,532 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103890.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:10:00,226 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103894.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:10:02,112 INFO [train.py:901] (3/4) Epoch 13, batch 6900, loss[loss=0.2107, simple_loss=0.2838, pruned_loss=0.06885, over 8141.00 frames. ], tot_loss[loss=0.2273, simple_loss=0.3044, pruned_loss=0.07513, over 1615526.52 frames. ], batch size: 22, lr: 5.80e-03, grad_scale: 8.0 +2023-02-06 14:10:35,839 INFO [train.py:901] (3/4) Epoch 13, batch 6950, loss[loss=0.2845, simple_loss=0.3425, pruned_loss=0.1133, over 7207.00 frames. ], tot_loss[loss=0.227, simple_loss=0.3045, pruned_loss=0.07473, over 1619128.58 frames. ], batch size: 72, lr: 5.80e-03, grad_scale: 8.0 +2023-02-06 14:10:43,033 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.669e+02 2.539e+02 3.074e+02 3.917e+02 9.810e+02, threshold=6.147e+02, percent-clipped=9.0 +2023-02-06 14:10:53,174 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 14:11:00,841 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4803, 1.8819, 2.7402, 1.3211, 1.9788, 1.8422, 1.6186, 1.8311], + device='cuda:3'), covar=tensor([0.1706, 0.2038, 0.0712, 0.3750, 0.1614, 0.2734, 0.1855, 0.2096], + device='cuda:3'), in_proj_covar=tensor([0.0496, 0.0537, 0.0531, 0.0588, 0.0620, 0.0559, 0.0483, 0.0613], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 14:11:09,734 INFO [train.py:901] (3/4) Epoch 13, batch 7000, loss[loss=0.23, simple_loss=0.311, pruned_loss=0.07446, over 8429.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.3027, pruned_loss=0.07354, over 1616851.66 frames. ], batch size: 27, lr: 5.80e-03, grad_scale: 8.0 +2023-02-06 14:11:18,202 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104008.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:11:19,230 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2023-02-06 14:11:23,547 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104015.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:11:35,611 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7800, 1.9041, 1.5897, 2.3279, 1.0481, 1.3614, 1.6014, 1.9076], + device='cuda:3'), covar=tensor([0.0698, 0.0731, 0.0979, 0.0418, 0.1121, 0.1412, 0.0895, 0.0728], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0208, 0.0252, 0.0210, 0.0215, 0.0252, 0.0256, 0.0215], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 14:11:43,224 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.45 vs. limit=5.0 +2023-02-06 14:11:44,810 INFO [train.py:901] (3/4) Epoch 13, batch 7050, loss[loss=0.2237, simple_loss=0.293, pruned_loss=0.0772, over 7547.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.3026, pruned_loss=0.07326, over 1614163.76 frames. ], batch size: 18, lr: 5.80e-03, grad_scale: 8.0 +2023-02-06 14:11:52,789 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.596e+02 2.385e+02 2.879e+02 3.637e+02 6.044e+02, threshold=5.759e+02, percent-clipped=0.0 +2023-02-06 14:11:58,854 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104067.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:12:01,822 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-06 14:12:18,820 INFO [train.py:901] (3/4) Epoch 13, batch 7100, loss[loss=0.1841, simple_loss=0.2632, pruned_loss=0.05247, over 7436.00 frames. ], tot_loss[loss=0.2251, simple_loss=0.303, pruned_loss=0.07366, over 1612508.69 frames. ], batch size: 17, lr: 5.80e-03, grad_scale: 8.0 +2023-02-06 14:12:21,102 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104099.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 14:12:22,946 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104102.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:12:37,519 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104124.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 14:12:39,433 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104127.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:12:46,609 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-06 14:12:47,820 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 +2023-02-06 14:12:52,697 INFO [train.py:901] (3/4) Epoch 13, batch 7150, loss[loss=0.1754, simple_loss=0.2468, pruned_loss=0.05196, over 7524.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.305, pruned_loss=0.07498, over 1612450.15 frames. ], batch size: 18, lr: 5.79e-03, grad_scale: 8.0 +2023-02-06 14:13:00,089 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.449e+02 2.541e+02 2.991e+02 4.071e+02 7.912e+02, threshold=5.982e+02, percent-clipped=4.0 +2023-02-06 14:13:27,755 INFO [train.py:901] (3/4) Epoch 13, batch 7200, loss[loss=0.1597, simple_loss=0.2408, pruned_loss=0.03925, over 7539.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3039, pruned_loss=0.07401, over 1613099.40 frames. ], batch size: 18, lr: 5.79e-03, grad_scale: 8.0 +2023-02-06 14:13:42,140 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104217.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:13:55,691 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104238.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:13:58,506 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104242.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:14:01,688 INFO [train.py:901] (3/4) Epoch 13, batch 7250, loss[loss=0.2294, simple_loss=0.3075, pruned_loss=0.07566, over 7928.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.3037, pruned_loss=0.07448, over 1611642.13 frames. ], batch size: 20, lr: 5.79e-03, grad_scale: 8.0 +2023-02-06 14:14:09,625 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.469e+02 3.063e+02 3.939e+02 8.277e+02, threshold=6.126e+02, percent-clipped=7.0 +2023-02-06 14:14:32,858 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0496, 2.3369, 1.8356, 2.8817, 1.5123, 1.4846, 1.9000, 2.2587], + device='cuda:3'), covar=tensor([0.0695, 0.0797, 0.0974, 0.0370, 0.1173, 0.1502, 0.0975, 0.0791], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0207, 0.0251, 0.0209, 0.0213, 0.0251, 0.0255, 0.0214], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 14:14:37,320 INFO [train.py:901] (3/4) Epoch 13, batch 7300, loss[loss=0.2071, simple_loss=0.2748, pruned_loss=0.06974, over 7713.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3033, pruned_loss=0.07427, over 1611629.58 frames. ], batch size: 18, lr: 5.79e-03, grad_scale: 8.0 +2023-02-06 14:15:11,544 INFO [train.py:901] (3/4) Epoch 13, batch 7350, loss[loss=0.1929, simple_loss=0.2715, pruned_loss=0.05712, over 7431.00 frames. ], tot_loss[loss=0.2254, simple_loss=0.3028, pruned_loss=0.07396, over 1612466.07 frames. ], batch size: 17, lr: 5.79e-03, grad_scale: 8.0 +2023-02-06 14:15:14,977 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104352.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:15:15,799 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104353.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:15:19,092 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.782e+02 2.439e+02 3.043e+02 3.823e+02 6.373e+02, threshold=6.086e+02, percent-clipped=2.0 +2023-02-06 14:15:19,894 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104359.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:15:33,260 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 14:15:46,158 INFO [train.py:901] (3/4) Epoch 13, batch 7400, loss[loss=0.2319, simple_loss=0.3047, pruned_loss=0.07957, over 8099.00 frames. ], tot_loss[loss=0.225, simple_loss=0.3029, pruned_loss=0.07357, over 1613699.89 frames. ], batch size: 23, lr: 5.79e-03, grad_scale: 8.0 +2023-02-06 14:15:53,249 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 14:15:56,728 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=104411.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:16:21,099 INFO [train.py:901] (3/4) Epoch 13, batch 7450, loss[loss=0.2449, simple_loss=0.3215, pruned_loss=0.08414, over 8345.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3038, pruned_loss=0.07408, over 1616158.83 frames. ], batch size: 26, lr: 5.79e-03, grad_scale: 8.0 +2023-02-06 14:16:29,258 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.488e+02 2.494e+02 3.000e+02 3.814e+02 1.100e+03, threshold=5.999e+02, percent-clipped=4.0 +2023-02-06 14:16:33,242 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 14:16:35,471 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104467.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:16:39,584 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104473.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:16:40,181 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104474.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:16:55,894 INFO [train.py:901] (3/4) Epoch 13, batch 7500, loss[loss=0.2974, simple_loss=0.3452, pruned_loss=0.1248, over 6752.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.305, pruned_loss=0.07512, over 1617465.80 frames. ], batch size: 71, lr: 5.78e-03, grad_scale: 8.0 +2023-02-06 14:16:56,789 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104498.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:16:56,807 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104498.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:17:14,584 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104523.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:17:15,883 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2688, 3.1481, 2.9087, 1.5703, 2.8594, 2.8558, 2.9305, 2.7393], + device='cuda:3'), covar=tensor([0.1046, 0.0866, 0.1357, 0.4578, 0.1121, 0.1341, 0.1565, 0.1098], + device='cuda:3'), in_proj_covar=tensor([0.0470, 0.0385, 0.0394, 0.0487, 0.0386, 0.0389, 0.0378, 0.0338], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 14:17:16,626 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104526.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:17:30,659 INFO [train.py:901] (3/4) Epoch 13, batch 7550, loss[loss=0.2104, simple_loss=0.2751, pruned_loss=0.07284, over 7545.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.3044, pruned_loss=0.07523, over 1613562.03 frames. ], batch size: 18, lr: 5.78e-03, grad_scale: 8.0 +2023-02-06 14:17:37,880 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.501e+02 2.482e+02 3.042e+02 4.105e+02 9.709e+02, threshold=6.085e+02, percent-clipped=7.0 +2023-02-06 14:18:03,967 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6276, 1.5185, 2.8760, 1.1037, 2.1624, 3.1724, 3.4033, 2.2901], + device='cuda:3'), covar=tensor([0.1565, 0.1745, 0.0505, 0.2756, 0.1003, 0.0417, 0.0575, 0.1264], + device='cuda:3'), in_proj_covar=tensor([0.0271, 0.0303, 0.0265, 0.0295, 0.0278, 0.0239, 0.0359, 0.0295], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 14:18:05,105 INFO [train.py:901] (3/4) Epoch 13, batch 7600, loss[loss=0.2201, simple_loss=0.2954, pruned_loss=0.07237, over 7647.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.3048, pruned_loss=0.07543, over 1614857.04 frames. ], batch size: 19, lr: 5.78e-03, grad_scale: 8.0 +2023-02-06 14:18:13,227 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104609.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:18:19,840 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.0771, 1.2808, 4.2848, 1.5278, 3.7086, 3.5071, 3.8343, 3.7173], + device='cuda:3'), covar=tensor([0.0606, 0.4834, 0.0519, 0.3959, 0.1201, 0.1109, 0.0610, 0.0694], + device='cuda:3'), in_proj_covar=tensor([0.0522, 0.0585, 0.0609, 0.0555, 0.0632, 0.0542, 0.0528, 0.0591], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 14:18:30,785 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104634.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:18:34,913 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1644, 2.1642, 1.5354, 1.8157, 1.7760, 1.3803, 1.5263, 1.6746], + device='cuda:3'), covar=tensor([0.1389, 0.0339, 0.1178, 0.0579, 0.0745, 0.1389, 0.0961, 0.0850], + device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0238, 0.0323, 0.0299, 0.0302, 0.0325, 0.0344, 0.0309], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 14:18:40,154 INFO [train.py:901] (3/4) Epoch 13, batch 7650, loss[loss=0.2528, simple_loss=0.3386, pruned_loss=0.08348, over 8261.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.3055, pruned_loss=0.07542, over 1619919.85 frames. ], batch size: 24, lr: 5.78e-03, grad_scale: 8.0 +2023-02-06 14:18:47,609 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.372e+02 2.638e+02 3.280e+02 4.340e+02 1.130e+03, threshold=6.560e+02, percent-clipped=9.0 +2023-02-06 14:19:14,843 INFO [train.py:901] (3/4) Epoch 13, batch 7700, loss[loss=0.1951, simple_loss=0.2709, pruned_loss=0.05964, over 7778.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.306, pruned_loss=0.07576, over 1615491.87 frames. ], batch size: 19, lr: 5.78e-03, grad_scale: 8.0 +2023-02-06 14:19:33,193 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104723.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:19:37,764 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 14:19:37,984 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104730.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:19:49,832 INFO [train.py:901] (3/4) Epoch 13, batch 7750, loss[loss=0.3057, simple_loss=0.3737, pruned_loss=0.1189, over 8185.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.3063, pruned_loss=0.0759, over 1621867.44 frames. ], batch size: 23, lr: 5.78e-03, grad_scale: 8.0 +2023-02-06 14:19:50,602 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104748.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:19:55,980 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104755.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:19:57,811 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.403e+02 2.530e+02 2.944e+02 3.392e+02 9.198e+02, threshold=5.888e+02, percent-clipped=3.0 +2023-02-06 14:20:09,967 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5093, 2.6021, 1.7688, 2.1893, 2.2515, 1.5936, 2.0421, 2.1101], + device='cuda:3'), covar=tensor([0.1535, 0.0395, 0.1174, 0.0682, 0.0673, 0.1490, 0.1006, 0.0942], + device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0233, 0.0316, 0.0292, 0.0296, 0.0319, 0.0335, 0.0302], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 14:20:12,809 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-06 14:20:14,006 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104782.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:20:24,830 INFO [train.py:901] (3/4) Epoch 13, batch 7800, loss[loss=0.1805, simple_loss=0.2632, pruned_loss=0.04894, over 7654.00 frames. ], tot_loss[loss=0.2271, simple_loss=0.3046, pruned_loss=0.07482, over 1619880.65 frames. ], batch size: 19, lr: 5.78e-03, grad_scale: 8.0 +2023-02-06 14:20:31,894 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104807.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:20:57,827 INFO [train.py:901] (3/4) Epoch 13, batch 7850, loss[loss=0.2354, simple_loss=0.3231, pruned_loss=0.07382, over 8239.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.3055, pruned_loss=0.07561, over 1617147.08 frames. ], batch size: 24, lr: 5.77e-03, grad_scale: 8.0 +2023-02-06 14:21:05,227 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.456e+02 2.619e+02 3.074e+02 4.074e+02 1.012e+03, threshold=6.148e+02, percent-clipped=5.0 +2023-02-06 14:21:30,905 INFO [train.py:901] (3/4) Epoch 13, batch 7900, loss[loss=0.1728, simple_loss=0.2524, pruned_loss=0.04666, over 7563.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.3049, pruned_loss=0.07562, over 1611263.32 frames. ], batch size: 18, lr: 5.77e-03, grad_scale: 8.0 +2023-02-06 14:21:36,645 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.52 vs. limit=5.0 +2023-02-06 14:22:04,229 INFO [train.py:901] (3/4) Epoch 13, batch 7950, loss[loss=0.2152, simple_loss=0.3011, pruned_loss=0.06462, over 8501.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.3048, pruned_loss=0.07569, over 1612172.30 frames. ], batch size: 26, lr: 5.77e-03, grad_scale: 8.0 +2023-02-06 14:22:11,312 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.657e+02 2.521e+02 3.027e+02 3.866e+02 6.555e+02, threshold=6.053e+02, percent-clipped=2.0 +2023-02-06 14:22:37,770 INFO [train.py:901] (3/4) Epoch 13, batch 8000, loss[loss=0.1953, simple_loss=0.2765, pruned_loss=0.05709, over 8196.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.3043, pruned_loss=0.07548, over 1612784.90 frames. ], batch size: 23, lr: 5.77e-03, grad_scale: 8.0 +2023-02-06 14:23:03,700 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-06 14:23:10,573 INFO [train.py:901] (3/4) Epoch 13, batch 8050, loss[loss=0.2229, simple_loss=0.2908, pruned_loss=0.07754, over 7921.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.3033, pruned_loss=0.07518, over 1607230.97 frames. ], batch size: 20, lr: 5.77e-03, grad_scale: 8.0 +2023-02-06 14:23:18,080 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.602e+02 2.417e+02 2.946e+02 3.621e+02 6.025e+02, threshold=5.892e+02, percent-clipped=0.0 +2023-02-06 14:23:50,191 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 14:23:54,152 INFO [train.py:901] (3/4) Epoch 14, batch 0, loss[loss=0.2044, simple_loss=0.2758, pruned_loss=0.06653, over 6362.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2758, pruned_loss=0.06653, over 6362.00 frames. ], batch size: 14, lr: 5.56e-03, grad_scale: 8.0 +2023-02-06 14:23:54,152 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 14:24:01,350 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4844, 1.7656, 2.6542, 1.2601, 2.0029, 1.7611, 1.5451, 1.9716], + device='cuda:3'), covar=tensor([0.1835, 0.2770, 0.0878, 0.4598, 0.1903, 0.3282, 0.2268, 0.2249], + device='cuda:3'), in_proj_covar=tensor([0.0499, 0.0545, 0.0539, 0.0600, 0.0623, 0.0566, 0.0493, 0.0619], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 14:24:05,196 INFO [train.py:935] (3/4) Epoch 14, validation: loss=0.184, simple_loss=0.2839, pruned_loss=0.04201, over 944034.00 frames. +2023-02-06 14:24:05,196 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 14:24:16,397 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.78 vs. limit=2.0 +2023-02-06 14:24:21,240 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 14:24:22,036 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.9531, 1.9720, 6.0110, 2.1719, 5.4394, 5.0932, 5.6257, 5.5358], + device='cuda:3'), covar=tensor([0.0446, 0.3965, 0.0315, 0.3144, 0.0857, 0.0745, 0.0427, 0.0432], + device='cuda:3'), in_proj_covar=tensor([0.0520, 0.0580, 0.0608, 0.0551, 0.0632, 0.0536, 0.0523, 0.0589], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 14:24:38,591 INFO [train.py:901] (3/4) Epoch 14, batch 50, loss[loss=0.2136, simple_loss=0.2973, pruned_loss=0.06495, over 8132.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3038, pruned_loss=0.07469, over 362428.86 frames. ], batch size: 22, lr: 5.56e-03, grad_scale: 8.0 +2023-02-06 14:24:54,207 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5074, 1.4227, 2.8409, 1.2459, 1.9743, 3.0760, 3.1405, 2.5635], + device='cuda:3'), covar=tensor([0.1145, 0.1479, 0.0363, 0.2033, 0.0916, 0.0277, 0.0542, 0.0660], + device='cuda:3'), in_proj_covar=tensor([0.0272, 0.0304, 0.0265, 0.0295, 0.0281, 0.0241, 0.0362, 0.0295], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 14:24:54,762 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 14:24:58,151 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.784e+02 2.684e+02 3.092e+02 3.835e+02 7.852e+02, threshold=6.183e+02, percent-clipped=3.0 +2023-02-06 14:25:02,467 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4093, 1.9497, 3.4299, 1.2591, 2.3934, 1.8706, 1.5633, 2.3986], + device='cuda:3'), covar=tensor([0.1879, 0.2389, 0.0704, 0.4016, 0.1793, 0.3059, 0.1953, 0.2245], + device='cuda:3'), in_proj_covar=tensor([0.0498, 0.0541, 0.0535, 0.0597, 0.0623, 0.0565, 0.0490, 0.0615], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 14:25:14,422 INFO [train.py:901] (3/4) Epoch 14, batch 100, loss[loss=0.1892, simple_loss=0.2811, pruned_loss=0.04869, over 8245.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.3024, pruned_loss=0.07311, over 639188.60 frames. ], batch size: 24, lr: 5.56e-03, grad_scale: 8.0 +2023-02-06 14:25:17,795 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 14:25:48,643 INFO [train.py:901] (3/4) Epoch 14, batch 150, loss[loss=0.2633, simple_loss=0.3435, pruned_loss=0.09158, over 8439.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.3038, pruned_loss=0.07482, over 854952.57 frames. ], batch size: 49, lr: 5.55e-03, grad_scale: 8.0 +2023-02-06 14:26:08,300 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.716e+02 2.384e+02 2.990e+02 3.742e+02 5.781e+02, threshold=5.980e+02, percent-clipped=0.0 +2023-02-06 14:26:14,015 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-06 14:26:22,055 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 +2023-02-06 14:26:23,065 INFO [train.py:901] (3/4) Epoch 14, batch 200, loss[loss=0.1992, simple_loss=0.2745, pruned_loss=0.06188, over 7505.00 frames. ], tot_loss[loss=0.2294, simple_loss=0.3056, pruned_loss=0.07658, over 1022460.62 frames. ], batch size: 18, lr: 5.55e-03, grad_scale: 8.0 +2023-02-06 14:26:44,058 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-06 14:26:47,394 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2637, 1.2555, 1.5220, 1.2052, 0.6965, 1.3536, 1.2151, 1.1551], + device='cuda:3'), covar=tensor([0.0572, 0.1227, 0.1650, 0.1400, 0.0582, 0.1435, 0.0645, 0.0646], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0150, 0.0188, 0.0155, 0.0100, 0.0160, 0.0112, 0.0137], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 14:26:58,940 INFO [train.py:901] (3/4) Epoch 14, batch 250, loss[loss=0.2046, simple_loss=0.2933, pruned_loss=0.05798, over 8097.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.305, pruned_loss=0.07613, over 1154638.52 frames. ], batch size: 23, lr: 5.55e-03, grad_scale: 8.0 +2023-02-06 14:27:07,605 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 14:27:15,946 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 14:27:18,050 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.724e+02 2.546e+02 3.157e+02 4.204e+02 9.163e+02, threshold=6.313e+02, percent-clipped=6.0 +2023-02-06 14:27:31,869 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3632, 1.5628, 1.6668, 0.9219, 1.7273, 1.3087, 0.2403, 1.5681], + device='cuda:3'), covar=tensor([0.0357, 0.0272, 0.0187, 0.0379, 0.0246, 0.0695, 0.0629, 0.0204], + device='cuda:3'), in_proj_covar=tensor([0.0406, 0.0344, 0.0296, 0.0400, 0.0330, 0.0488, 0.0365, 0.0372], + device='cuda:3'), out_proj_covar=tensor([1.1341e-04, 9.3573e-05, 8.0427e-05, 1.0972e-04, 9.0763e-05, 1.4409e-04, + 1.0227e-04, 1.0270e-04], device='cuda:3') +2023-02-06 14:27:33,663 INFO [train.py:901] (3/4) Epoch 14, batch 300, loss[loss=0.2387, simple_loss=0.3248, pruned_loss=0.07635, over 8112.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.3054, pruned_loss=0.07589, over 1257718.26 frames. ], batch size: 23, lr: 5.55e-03, grad_scale: 8.0 +2023-02-06 14:27:52,816 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105406.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:27:58,028 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.1906, 1.1977, 3.3236, 0.9394, 2.9199, 2.7720, 3.0000, 2.9347], + device='cuda:3'), covar=tensor([0.0763, 0.4188, 0.0898, 0.3959, 0.1458, 0.1239, 0.0785, 0.0864], + device='cuda:3'), in_proj_covar=tensor([0.0520, 0.0586, 0.0618, 0.0557, 0.0635, 0.0539, 0.0530, 0.0597], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 14:28:04,217 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8089, 1.3026, 3.9411, 1.2733, 3.4617, 3.2885, 3.5078, 3.4515], + device='cuda:3'), covar=tensor([0.0551, 0.4403, 0.0620, 0.3909, 0.1185, 0.1010, 0.0637, 0.0669], + device='cuda:3'), in_proj_covar=tensor([0.0520, 0.0586, 0.0617, 0.0557, 0.0634, 0.0538, 0.0529, 0.0596], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 14:28:09,627 INFO [train.py:901] (3/4) Epoch 14, batch 350, loss[loss=0.2041, simple_loss=0.2766, pruned_loss=0.06579, over 8077.00 frames. ], tot_loss[loss=0.2273, simple_loss=0.3045, pruned_loss=0.07504, over 1338813.07 frames. ], batch size: 21, lr: 5.55e-03, grad_scale: 8.0 +2023-02-06 14:28:28,604 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.329e+02 2.437e+02 2.818e+02 3.446e+02 5.751e+02, threshold=5.636e+02, percent-clipped=0.0 +2023-02-06 14:28:43,599 INFO [train.py:901] (3/4) Epoch 14, batch 400, loss[loss=0.2792, simple_loss=0.3478, pruned_loss=0.1053, over 8495.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.3055, pruned_loss=0.07534, over 1402855.13 frames. ], batch size: 26, lr: 5.55e-03, grad_scale: 8.0 +2023-02-06 14:29:00,993 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105504.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:29:13,246 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105520.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:29:20,741 INFO [train.py:901] (3/4) Epoch 14, batch 450, loss[loss=0.2617, simple_loss=0.3354, pruned_loss=0.09393, over 8738.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3043, pruned_loss=0.07386, over 1452541.12 frames. ], batch size: 30, lr: 5.55e-03, grad_scale: 8.0 +2023-02-06 14:29:36,118 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6555, 1.3112, 1.5372, 1.2647, 0.8657, 1.3252, 1.4651, 1.2558], + device='cuda:3'), covar=tensor([0.0528, 0.1288, 0.1782, 0.1431, 0.0619, 0.1540, 0.0723, 0.0689], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0190, 0.0156, 0.0101, 0.0161, 0.0114, 0.0137], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 14:29:40,039 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.863e+02 2.497e+02 2.804e+02 3.770e+02 6.336e+02, threshold=5.609e+02, percent-clipped=1.0 +2023-02-06 14:29:55,206 INFO [train.py:901] (3/4) Epoch 14, batch 500, loss[loss=0.2225, simple_loss=0.3013, pruned_loss=0.07184, over 8578.00 frames. ], tot_loss[loss=0.2255, simple_loss=0.3034, pruned_loss=0.07377, over 1485579.46 frames. ], batch size: 39, lr: 5.54e-03, grad_scale: 8.0 +2023-02-06 14:30:29,396 INFO [train.py:901] (3/4) Epoch 14, batch 550, loss[loss=0.1938, simple_loss=0.2857, pruned_loss=0.05099, over 8519.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.3048, pruned_loss=0.07441, over 1519036.30 frames. ], batch size: 39, lr: 5.54e-03, grad_scale: 8.0 +2023-02-06 14:30:50,301 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.592e+02 2.442e+02 2.933e+02 3.700e+02 8.163e+02, threshold=5.867e+02, percent-clipped=3.0 +2023-02-06 14:31:05,192 INFO [train.py:901] (3/4) Epoch 14, batch 600, loss[loss=0.2358, simple_loss=0.3162, pruned_loss=0.07765, over 7118.00 frames. ], tot_loss[loss=0.2272, simple_loss=0.305, pruned_loss=0.07466, over 1538016.19 frames. ], batch size: 71, lr: 5.54e-03, grad_scale: 8.0 +2023-02-06 14:31:18,452 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 14:31:39,816 INFO [train.py:901] (3/4) Epoch 14, batch 650, loss[loss=0.1663, simple_loss=0.2495, pruned_loss=0.04155, over 7424.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.3042, pruned_loss=0.074, over 1555848.64 frames. ], batch size: 17, lr: 5.54e-03, grad_scale: 8.0 +2023-02-06 14:31:54,391 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=105750.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:32:01,339 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.581e+02 2.402e+02 3.000e+02 3.711e+02 7.109e+02, threshold=6.000e+02, percent-clipped=4.0 +2023-02-06 14:32:17,053 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.04 vs. limit=5.0 +2023-02-06 14:32:17,354 INFO [train.py:901] (3/4) Epoch 14, batch 700, loss[loss=0.2617, simple_loss=0.3267, pruned_loss=0.09833, over 8807.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.3032, pruned_loss=0.07317, over 1570987.51 frames. ], batch size: 40, lr: 5.54e-03, grad_scale: 8.0 +2023-02-06 14:32:37,894 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105810.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:32:42,734 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105817.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:32:51,388 INFO [train.py:901] (3/4) Epoch 14, batch 750, loss[loss=0.2316, simple_loss=0.3137, pruned_loss=0.07474, over 8036.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.3032, pruned_loss=0.07322, over 1583808.93 frames. ], batch size: 22, lr: 5.54e-03, grad_scale: 8.0 +2023-02-06 14:32:51,599 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4914, 2.9059, 2.4308, 4.0247, 1.6558, 2.1925, 2.4732, 3.2536], + device='cuda:3'), covar=tensor([0.0785, 0.0874, 0.0933, 0.0270, 0.1248, 0.1385, 0.1104, 0.0725], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0204, 0.0252, 0.0210, 0.0212, 0.0253, 0.0258, 0.0216], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 14:33:03,871 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=105848.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:33:06,443 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 14:33:11,301 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.389e+02 2.459e+02 2.898e+02 3.725e+02 7.154e+02, threshold=5.796e+02, percent-clipped=4.0 +2023-02-06 14:33:15,493 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=105864.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:33:16,057 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 14:33:16,241 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=105865.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:33:27,180 INFO [train.py:901] (3/4) Epoch 14, batch 800, loss[loss=0.1943, simple_loss=0.2736, pruned_loss=0.05749, over 7527.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.3027, pruned_loss=0.07315, over 1585298.99 frames. ], batch size: 18, lr: 5.54e-03, grad_scale: 16.0 +2023-02-06 14:33:40,346 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2518, 3.1608, 2.9432, 1.6536, 2.8665, 2.8783, 2.9169, 2.7276], + device='cuda:3'), covar=tensor([0.1117, 0.0838, 0.1376, 0.4587, 0.1159, 0.1294, 0.1578, 0.1193], + device='cuda:3'), in_proj_covar=tensor([0.0477, 0.0391, 0.0405, 0.0504, 0.0396, 0.0398, 0.0384, 0.0346], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 14:33:48,448 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.46 vs. limit=5.0 +2023-02-06 14:33:50,560 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 +2023-02-06 14:33:55,100 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2023-02-06 14:34:02,175 INFO [train.py:901] (3/4) Epoch 14, batch 850, loss[loss=0.2161, simple_loss=0.2899, pruned_loss=0.07118, over 7986.00 frames. ], tot_loss[loss=0.2253, simple_loss=0.3036, pruned_loss=0.0735, over 1598706.51 frames. ], batch size: 21, lr: 5.54e-03, grad_scale: 16.0 +2023-02-06 14:34:20,961 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.803e+02 2.477e+02 2.961e+02 4.061e+02 6.411e+02, threshold=5.921e+02, percent-clipped=4.0 +2023-02-06 14:34:24,608 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=105963.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:34:36,588 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=105979.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:34:37,083 INFO [train.py:901] (3/4) Epoch 14, batch 900, loss[loss=0.2055, simple_loss=0.2894, pruned_loss=0.06075, over 8479.00 frames. ], tot_loss[loss=0.2255, simple_loss=0.3038, pruned_loss=0.07363, over 1602307.45 frames. ], batch size: 25, lr: 5.53e-03, grad_scale: 16.0 +2023-02-06 14:35:03,388 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7760, 1.7323, 2.2112, 1.7119, 1.4160, 2.2612, 0.6290, 1.5945], + device='cuda:3'), covar=tensor([0.2566, 0.1247, 0.0467, 0.1426, 0.3073, 0.0425, 0.2960, 0.1678], + device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0177, 0.0108, 0.0220, 0.0256, 0.0111, 0.0164, 0.0174], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 14:35:14,895 INFO [train.py:901] (3/4) Epoch 14, batch 950, loss[loss=0.2093, simple_loss=0.2938, pruned_loss=0.06241, over 8361.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3037, pruned_loss=0.07375, over 1606652.06 frames. ], batch size: 24, lr: 5.53e-03, grad_scale: 16.0 +2023-02-06 14:35:33,973 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.531e+02 2.617e+02 3.202e+02 4.119e+02 6.844e+02, threshold=6.403e+02, percent-clipped=3.0 +2023-02-06 14:35:38,937 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 14:35:49,299 INFO [train.py:901] (3/4) Epoch 14, batch 1000, loss[loss=0.1975, simple_loss=0.2817, pruned_loss=0.05666, over 8618.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.304, pruned_loss=0.07379, over 1612763.74 frames. ], batch size: 34, lr: 5.53e-03, grad_scale: 16.0 +2023-02-06 14:36:00,606 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106095.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:36:14,311 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 14:36:20,078 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106121.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:36:26,949 INFO [train.py:901] (3/4) Epoch 14, batch 1050, loss[loss=0.208, simple_loss=0.2824, pruned_loss=0.06674, over 7807.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.303, pruned_loss=0.0728, over 1607502.69 frames. ], batch size: 20, lr: 5.53e-03, grad_scale: 16.0 +2023-02-06 14:36:26,961 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 14:36:37,977 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106146.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:36:43,499 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106154.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:36:46,250 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.492e+02 2.417e+02 2.951e+02 3.593e+02 9.096e+02, threshold=5.903e+02, percent-clipped=2.0 +2023-02-06 14:36:48,432 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106161.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:37:01,617 INFO [train.py:901] (3/4) Epoch 14, batch 1100, loss[loss=0.2059, simple_loss=0.2837, pruned_loss=0.06405, over 8473.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.3027, pruned_loss=0.07285, over 1606087.23 frames. ], batch size: 25, lr: 5.53e-03, grad_scale: 16.0 +2023-02-06 14:37:29,710 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106219.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:37:35,892 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 14:37:38,717 INFO [train.py:901] (3/4) Epoch 14, batch 1150, loss[loss=0.204, simple_loss=0.2839, pruned_loss=0.06209, over 8372.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3029, pruned_loss=0.07315, over 1614533.42 frames. ], batch size: 24, lr: 5.53e-03, grad_scale: 16.0 +2023-02-06 14:37:42,396 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106235.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:37:46,539 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 +2023-02-06 14:37:49,101 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106244.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:37:58,390 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.753e+02 2.475e+02 3.133e+02 3.919e+02 6.906e+02, threshold=6.266e+02, percent-clipped=3.0 +2023-02-06 14:37:59,994 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106260.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:38:06,207 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106269.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:38:10,866 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106276.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:38:13,355 INFO [train.py:901] (3/4) Epoch 14, batch 1200, loss[loss=0.2157, simple_loss=0.297, pruned_loss=0.06723, over 8256.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3035, pruned_loss=0.07388, over 1615856.25 frames. ], batch size: 24, lr: 5.53e-03, grad_scale: 16.0 +2023-02-06 14:38:17,561 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106286.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:38:42,813 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9232, 1.5422, 3.4700, 1.3393, 2.3965, 3.8419, 3.7639, 3.2493], + device='cuda:3'), covar=tensor([0.1030, 0.1543, 0.0318, 0.2034, 0.0934, 0.0214, 0.0597, 0.0639], + device='cuda:3'), in_proj_covar=tensor([0.0275, 0.0307, 0.0267, 0.0296, 0.0284, 0.0245, 0.0365, 0.0298], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 14:38:47,946 INFO [train.py:901] (3/4) Epoch 14, batch 1250, loss[loss=0.2282, simple_loss=0.3121, pruned_loss=0.07215, over 8495.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3023, pruned_loss=0.07345, over 1615104.39 frames. ], batch size: 26, lr: 5.53e-03, grad_scale: 16.0 +2023-02-06 14:39:05,980 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106354.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:39:08,474 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.620e+02 2.539e+02 3.303e+02 4.386e+02 1.450e+03, threshold=6.607e+02, percent-clipped=4.0 +2023-02-06 14:39:24,634 INFO [train.py:901] (3/4) Epoch 14, batch 1300, loss[loss=0.2168, simple_loss=0.3046, pruned_loss=0.06446, over 8101.00 frames. ], tot_loss[loss=0.224, simple_loss=0.3023, pruned_loss=0.07283, over 1618841.12 frames. ], batch size: 23, lr: 5.52e-03, grad_scale: 16.0 +2023-02-06 14:39:55,747 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4600, 1.6950, 1.6817, 1.1576, 1.7529, 1.3771, 0.3390, 1.5647], + device='cuda:3'), covar=tensor([0.0331, 0.0228, 0.0188, 0.0299, 0.0252, 0.0565, 0.0575, 0.0169], + device='cuda:3'), in_proj_covar=tensor([0.0410, 0.0346, 0.0302, 0.0404, 0.0337, 0.0491, 0.0366, 0.0376], + device='cuda:3'), out_proj_covar=tensor([1.1432e-04, 9.3765e-05, 8.2173e-05, 1.1042e-04, 9.2543e-05, 1.4467e-04, + 1.0230e-04, 1.0355e-04], device='cuda:3') +2023-02-06 14:39:58,993 INFO [train.py:901] (3/4) Epoch 14, batch 1350, loss[loss=0.2185, simple_loss=0.2878, pruned_loss=0.07459, over 7547.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.3025, pruned_loss=0.07266, over 1618455.37 frames. ], batch size: 18, lr: 5.52e-03, grad_scale: 16.0 +2023-02-06 14:40:05,441 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106439.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:40:19,201 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.550e+02 3.060e+02 3.665e+02 8.767e+02, threshold=6.121e+02, percent-clipped=1.0 +2023-02-06 14:40:29,790 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106472.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:40:35,138 INFO [train.py:901] (3/4) Epoch 14, batch 1400, loss[loss=0.2346, simple_loss=0.3128, pruned_loss=0.07819, over 8071.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.3027, pruned_loss=0.07286, over 1617245.75 frames. ], batch size: 21, lr: 5.52e-03, grad_scale: 16.0 +2023-02-06 14:41:05,449 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-06 14:41:07,387 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106525.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:41:10,539 INFO [train.py:901] (3/4) Epoch 14, batch 1450, loss[loss=0.1767, simple_loss=0.2541, pruned_loss=0.04965, over 7433.00 frames. ], tot_loss[loss=0.225, simple_loss=0.3031, pruned_loss=0.07351, over 1616624.79 frames. ], batch size: 17, lr: 5.52e-03, grad_scale: 16.0 +2023-02-06 14:41:11,257 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 14:41:12,179 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106532.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:41:24,665 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106550.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:41:27,430 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106554.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:41:29,522 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106557.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:41:29,949 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.831e+02 2.546e+02 3.123e+02 4.151e+02 8.254e+02, threshold=6.246e+02, percent-clipped=6.0 +2023-02-06 14:41:47,575 INFO [train.py:901] (3/4) Epoch 14, batch 1500, loss[loss=0.2989, simple_loss=0.3668, pruned_loss=0.1155, over 8321.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.3044, pruned_loss=0.0741, over 1617376.11 frames. ], batch size: 26, lr: 5.52e-03, grad_scale: 16.0 +2023-02-06 14:42:22,569 INFO [train.py:901] (3/4) Epoch 14, batch 1550, loss[loss=0.2041, simple_loss=0.2901, pruned_loss=0.0591, over 8577.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.3053, pruned_loss=0.07472, over 1618240.17 frames. ], batch size: 31, lr: 5.52e-03, grad_scale: 16.0 +2023-02-06 14:42:22,652 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106630.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:42:41,334 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.688e+02 2.593e+02 3.196e+02 4.114e+02 8.054e+02, threshold=6.391e+02, percent-clipped=4.0 +2023-02-06 14:42:52,897 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7645, 1.5509, 3.1884, 1.4433, 2.2536, 3.3791, 3.4973, 2.8722], + device='cuda:3'), covar=tensor([0.1136, 0.1521, 0.0318, 0.1989, 0.0889, 0.0270, 0.0511, 0.0631], + device='cuda:3'), in_proj_covar=tensor([0.0270, 0.0303, 0.0263, 0.0293, 0.0281, 0.0242, 0.0362, 0.0293], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 14:42:56,721 INFO [train.py:901] (3/4) Epoch 14, batch 1600, loss[loss=0.3067, simple_loss=0.361, pruned_loss=0.1262, over 7246.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.3042, pruned_loss=0.07466, over 1616960.37 frames. ], batch size: 73, lr: 5.52e-03, grad_scale: 16.0 +2023-02-06 14:43:10,342 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106698.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:43:32,335 INFO [train.py:901] (3/4) Epoch 14, batch 1650, loss[loss=0.2642, simple_loss=0.3261, pruned_loss=0.1011, over 7925.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3033, pruned_loss=0.07423, over 1616434.82 frames. ], batch size: 20, lr: 5.51e-03, grad_scale: 8.0 +2023-02-06 14:43:35,230 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2103, 1.4391, 1.6079, 1.3092, 0.8929, 1.3430, 1.8061, 1.5602], + device='cuda:3'), covar=tensor([0.0501, 0.1297, 0.1836, 0.1503, 0.0634, 0.1599, 0.0671, 0.0678], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0190, 0.0157, 0.0101, 0.0162, 0.0113, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 14:43:42,575 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106745.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:43:51,899 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.679e+02 2.493e+02 3.038e+02 4.078e+02 1.080e+03, threshold=6.076e+02, percent-clipped=3.0 +2023-02-06 14:43:52,347 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.16 vs. limit=5.0 +2023-02-06 14:43:53,551 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3324, 1.5564, 1.6296, 0.8554, 1.6841, 1.2210, 0.2867, 1.4929], + device='cuda:3'), covar=tensor([0.0337, 0.0277, 0.0230, 0.0374, 0.0246, 0.0666, 0.0591, 0.0195], + device='cuda:3'), in_proj_covar=tensor([0.0412, 0.0349, 0.0304, 0.0406, 0.0338, 0.0492, 0.0370, 0.0375], + device='cuda:3'), out_proj_covar=tensor([1.1478e-04, 9.4922e-05, 8.2619e-05, 1.1079e-04, 9.2894e-05, 1.4505e-04, + 1.0328e-04, 1.0334e-04], device='cuda:3') +2023-02-06 14:44:06,421 INFO [train.py:901] (3/4) Epoch 14, batch 1700, loss[loss=0.2305, simple_loss=0.2847, pruned_loss=0.08818, over 4742.00 frames. ], tot_loss[loss=0.225, simple_loss=0.3032, pruned_loss=0.07342, over 1619272.64 frames. ], batch size: 10, lr: 5.51e-03, grad_scale: 8.0 +2023-02-06 14:44:28,951 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106810.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:44:31,562 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106813.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:44:33,565 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=106816.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:44:43,438 INFO [train.py:901] (3/4) Epoch 14, batch 1750, loss[loss=0.2058, simple_loss=0.2902, pruned_loss=0.06067, over 8033.00 frames. ], tot_loss[loss=0.223, simple_loss=0.3016, pruned_loss=0.07219, over 1618360.16 frames. ], batch size: 22, lr: 5.51e-03, grad_scale: 8.0 +2023-02-06 14:44:47,852 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106835.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:45:04,124 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.709e+02 2.358e+02 2.865e+02 3.554e+02 7.426e+02, threshold=5.730e+02, percent-clipped=3.0 +2023-02-06 14:45:18,437 INFO [train.py:901] (3/4) Epoch 14, batch 1800, loss[loss=0.2355, simple_loss=0.3045, pruned_loss=0.08326, over 8091.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.3018, pruned_loss=0.0729, over 1618945.96 frames. ], batch size: 21, lr: 5.51e-03, grad_scale: 8.0 +2023-02-06 14:45:54,581 INFO [train.py:901] (3/4) Epoch 14, batch 1850, loss[loss=0.2453, simple_loss=0.3171, pruned_loss=0.08675, over 8539.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.302, pruned_loss=0.07322, over 1619895.07 frames. ], batch size: 39, lr: 5.51e-03, grad_scale: 4.0 +2023-02-06 14:45:55,506 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106931.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:46:16,024 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.811e+02 2.570e+02 3.068e+02 3.847e+02 1.325e+03, threshold=6.136e+02, percent-clipped=4.0 +2023-02-06 14:46:29,532 INFO [train.py:901] (3/4) Epoch 14, batch 1900, loss[loss=0.2046, simple_loss=0.2919, pruned_loss=0.05869, over 8192.00 frames. ], tot_loss[loss=0.225, simple_loss=0.3023, pruned_loss=0.07381, over 1616337.37 frames. ], batch size: 23, lr: 5.51e-03, grad_scale: 4.0 +2023-02-06 14:46:43,881 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107001.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:46:47,083 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 14:46:59,775 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 14:47:01,258 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107026.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:47:03,728 INFO [train.py:901] (3/4) Epoch 14, batch 1950, loss[loss=0.2684, simple_loss=0.3363, pruned_loss=0.1003, over 7306.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.3027, pruned_loss=0.07443, over 1615000.90 frames. ], batch size: 71, lr: 5.51e-03, grad_scale: 4.0 +2023-02-06 14:47:19,873 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 14:47:26,064 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.458e+02 2.386e+02 2.840e+02 3.483e+02 6.138e+02, threshold=5.681e+02, percent-clipped=1.0 +2023-02-06 14:47:31,992 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107069.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:47:39,106 INFO [train.py:901] (3/4) Epoch 14, batch 2000, loss[loss=0.2838, simple_loss=0.3471, pruned_loss=0.1102, over 6911.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.3022, pruned_loss=0.07384, over 1614702.86 frames. ], batch size: 71, lr: 5.51e-03, grad_scale: 8.0 +2023-02-06 14:47:48,662 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107094.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:48:12,604 INFO [train.py:901] (3/4) Epoch 14, batch 2050, loss[loss=0.2138, simple_loss=0.3003, pruned_loss=0.06368, over 7982.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.3011, pruned_loss=0.07309, over 1612140.16 frames. ], batch size: 21, lr: 5.50e-03, grad_scale: 8.0 +2023-02-06 14:48:19,570 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1142, 2.1226, 1.4805, 1.8694, 1.6709, 1.1875, 1.5346, 1.7518], + device='cuda:3'), covar=tensor([0.1459, 0.0525, 0.1351, 0.0597, 0.0900, 0.1743, 0.1170, 0.0873], + device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0233, 0.0320, 0.0296, 0.0300, 0.0322, 0.0339, 0.0308], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 14:48:23,057 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6546, 1.6071, 2.8238, 1.3266, 2.0662, 3.0054, 3.1201, 2.5189], + device='cuda:3'), covar=tensor([0.1148, 0.1378, 0.0395, 0.2038, 0.0978, 0.0315, 0.0640, 0.0667], + device='cuda:3'), in_proj_covar=tensor([0.0271, 0.0304, 0.0264, 0.0292, 0.0279, 0.0241, 0.0362, 0.0293], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 14:48:26,620 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3863, 2.0853, 2.9523, 2.4140, 2.7481, 2.2648, 2.0185, 1.5179], + device='cuda:3'), covar=tensor([0.4624, 0.4324, 0.1378, 0.2749, 0.2262, 0.2476, 0.1847, 0.4495], + device='cuda:3'), in_proj_covar=tensor([0.0897, 0.0898, 0.0746, 0.0868, 0.0956, 0.0825, 0.0710, 0.0779], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 14:48:27,953 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0482, 2.2690, 1.7945, 2.8680, 1.3564, 1.6281, 1.9536, 2.3844], + device='cuda:3'), covar=tensor([0.0674, 0.0836, 0.0911, 0.0345, 0.1214, 0.1385, 0.1030, 0.0773], + device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0202, 0.0245, 0.0207, 0.0210, 0.0247, 0.0251, 0.0211], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 14:48:33,212 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107158.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 14:48:34,368 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.552e+02 2.411e+02 3.055e+02 3.713e+02 7.642e+02, threshold=6.109e+02, percent-clipped=4.0 +2023-02-06 14:48:42,102 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107170.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:48:49,408 INFO [train.py:901] (3/4) Epoch 14, batch 2100, loss[loss=0.2903, simple_loss=0.3625, pruned_loss=0.1091, over 8479.00 frames. ], tot_loss[loss=0.2252, simple_loss=0.3026, pruned_loss=0.07392, over 1609099.16 frames. ], batch size: 28, lr: 5.50e-03, grad_scale: 8.0 +2023-02-06 14:48:54,366 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107187.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:48:57,041 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7076, 2.2531, 4.1344, 1.5074, 2.8590, 2.2904, 1.9217, 2.8058], + device='cuda:3'), covar=tensor([0.1709, 0.2347, 0.0665, 0.3880, 0.1717, 0.2775, 0.1797, 0.2264], + device='cuda:3'), in_proj_covar=tensor([0.0496, 0.0542, 0.0535, 0.0592, 0.0618, 0.0559, 0.0487, 0.0614], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 14:49:00,227 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1421, 1.9135, 3.2128, 1.4294, 2.3839, 3.5109, 3.5400, 2.9972], + device='cuda:3'), covar=tensor([0.1048, 0.1408, 0.0379, 0.2218, 0.1044, 0.0240, 0.0578, 0.0582], + device='cuda:3'), in_proj_covar=tensor([0.0272, 0.0305, 0.0266, 0.0293, 0.0281, 0.0242, 0.0363, 0.0293], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 14:49:11,188 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107212.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:49:23,196 INFO [train.py:901] (3/4) Epoch 14, batch 2150, loss[loss=0.1798, simple_loss=0.2525, pruned_loss=0.05357, over 7247.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.303, pruned_loss=0.07424, over 1608142.47 frames. ], batch size: 16, lr: 5.50e-03, grad_scale: 8.0 +2023-02-06 14:49:44,426 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.580e+02 2.615e+02 3.041e+02 3.823e+02 8.460e+02, threshold=6.081e+02, percent-clipped=1.0 +2023-02-06 14:49:58,898 INFO [train.py:901] (3/4) Epoch 14, batch 2200, loss[loss=0.2071, simple_loss=0.286, pruned_loss=0.06412, over 8038.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.3041, pruned_loss=0.07471, over 1609605.19 frames. ], batch size: 22, lr: 5.50e-03, grad_scale: 8.0 +2023-02-06 14:50:23,271 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1015, 2.6548, 3.1281, 1.3803, 3.1913, 1.7991, 1.4588, 2.1736], + device='cuda:3'), covar=tensor([0.0579, 0.0287, 0.0204, 0.0617, 0.0352, 0.0706, 0.0720, 0.0425], + device='cuda:3'), in_proj_covar=tensor([0.0411, 0.0351, 0.0306, 0.0407, 0.0337, 0.0496, 0.0372, 0.0378], + device='cuda:3'), out_proj_covar=tensor([1.1456e-04, 9.5359e-05, 8.3165e-05, 1.1110e-04, 9.2411e-05, 1.4607e-04, + 1.0394e-04, 1.0421e-04], device='cuda:3') +2023-02-06 14:50:34,527 INFO [train.py:901] (3/4) Epoch 14, batch 2250, loss[loss=0.2022, simple_loss=0.2718, pruned_loss=0.06632, over 7976.00 frames. ], tot_loss[loss=0.2273, simple_loss=0.3045, pruned_loss=0.075, over 1609186.06 frames. ], batch size: 21, lr: 5.50e-03, grad_scale: 8.0 +2023-02-06 14:50:54,548 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.642e+02 2.599e+02 3.319e+02 4.071e+02 1.027e+03, threshold=6.637e+02, percent-clipped=7.0 +2023-02-06 14:51:08,893 INFO [train.py:901] (3/4) Epoch 14, batch 2300, loss[loss=0.2956, simple_loss=0.3483, pruned_loss=0.1214, over 6998.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.3036, pruned_loss=0.07457, over 1609464.19 frames. ], batch size: 71, lr: 5.50e-03, grad_scale: 8.0 +2023-02-06 14:51:27,523 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9032, 1.4611, 3.3001, 1.5010, 2.3623, 3.5211, 3.6749, 3.0481], + device='cuda:3'), covar=tensor([0.1098, 0.1649, 0.0341, 0.1976, 0.0991, 0.0241, 0.0440, 0.0567], + device='cuda:3'), in_proj_covar=tensor([0.0271, 0.0303, 0.0266, 0.0292, 0.0280, 0.0242, 0.0363, 0.0293], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 14:51:33,161 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 +2023-02-06 14:51:33,643 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0211, 2.2877, 1.7719, 2.8012, 1.2754, 1.5706, 1.9432, 2.3707], + device='cuda:3'), covar=tensor([0.0714, 0.0784, 0.0930, 0.0361, 0.1187, 0.1385, 0.0965, 0.0790], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0206, 0.0249, 0.0211, 0.0214, 0.0252, 0.0254, 0.0214], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 14:51:44,729 INFO [train.py:901] (3/4) Epoch 14, batch 2350, loss[loss=0.2203, simple_loss=0.297, pruned_loss=0.07176, over 7977.00 frames. ], tot_loss[loss=0.225, simple_loss=0.3025, pruned_loss=0.07375, over 1610720.60 frames. ], batch size: 21, lr: 5.50e-03, grad_scale: 8.0 +2023-02-06 14:52:00,477 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5539, 2.2221, 3.4441, 2.6090, 2.9494, 2.3434, 2.0018, 1.7222], + device='cuda:3'), covar=tensor([0.3948, 0.4024, 0.1325, 0.3271, 0.2342, 0.2396, 0.1699, 0.4690], + device='cuda:3'), in_proj_covar=tensor([0.0892, 0.0893, 0.0745, 0.0866, 0.0953, 0.0822, 0.0709, 0.0776], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 14:52:04,939 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.535e+02 2.358e+02 2.889e+02 3.449e+02 7.134e+02, threshold=5.779e+02, percent-clipped=1.0 +2023-02-06 14:52:18,380 INFO [train.py:901] (3/4) Epoch 14, batch 2400, loss[loss=0.296, simple_loss=0.3532, pruned_loss=0.1194, over 8338.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.3033, pruned_loss=0.07462, over 1609408.59 frames. ], batch size: 26, lr: 5.50e-03, grad_scale: 8.0 +2023-02-06 14:52:34,429 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=107502.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 14:52:43,465 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=107514.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:52:53,751 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107528.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:52:54,997 INFO [train.py:901] (3/4) Epoch 14, batch 2450, loss[loss=0.2394, simple_loss=0.3163, pruned_loss=0.08127, over 8774.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3033, pruned_loss=0.07439, over 1617939.26 frames. ], batch size: 30, lr: 5.49e-03, grad_scale: 8.0 +2023-02-06 14:53:16,528 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.677e+02 2.477e+02 3.089e+02 4.011e+02 1.178e+03, threshold=6.179e+02, percent-clipped=8.0 +2023-02-06 14:53:29,784 INFO [train.py:901] (3/4) Epoch 14, batch 2500, loss[loss=0.2296, simple_loss=0.3014, pruned_loss=0.07892, over 7220.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.3035, pruned_loss=0.07437, over 1623559.62 frames. ], batch size: 16, lr: 5.49e-03, grad_scale: 8.0 +2023-02-06 14:53:54,964 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7437, 1.6863, 2.1938, 1.5708, 1.3343, 2.2385, 0.4341, 1.3779], + device='cuda:3'), covar=tensor([0.2051, 0.1476, 0.0417, 0.1436, 0.3140, 0.0354, 0.2577, 0.1668], + device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0177, 0.0108, 0.0218, 0.0261, 0.0112, 0.0162, 0.0174], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 14:53:55,583 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107617.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 14:54:03,479 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107629.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:54:03,974 INFO [train.py:901] (3/4) Epoch 14, batch 2550, loss[loss=0.1798, simple_loss=0.2644, pruned_loss=0.04764, over 7972.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.303, pruned_loss=0.07424, over 1622480.55 frames. ], batch size: 21, lr: 5.49e-03, grad_scale: 8.0 +2023-02-06 14:54:12,888 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6594, 1.5313, 2.8518, 1.2899, 2.0190, 3.0166, 3.1217, 2.5803], + device='cuda:3'), covar=tensor([0.1089, 0.1445, 0.0399, 0.2063, 0.0960, 0.0283, 0.0618, 0.0601], + device='cuda:3'), in_proj_covar=tensor([0.0271, 0.0303, 0.0266, 0.0293, 0.0280, 0.0241, 0.0361, 0.0292], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 14:54:17,609 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6681, 2.0425, 2.2260, 1.2356, 2.3093, 1.5180, 0.6655, 1.8248], + device='cuda:3'), covar=tensor([0.0521, 0.0280, 0.0215, 0.0522, 0.0359, 0.0754, 0.0762, 0.0324], + device='cuda:3'), in_proj_covar=tensor([0.0413, 0.0353, 0.0309, 0.0409, 0.0341, 0.0499, 0.0375, 0.0381], + device='cuda:3'), out_proj_covar=tensor([1.1511e-04, 9.5815e-05, 8.3996e-05, 1.1152e-04, 9.3479e-05, 1.4706e-04, + 1.0461e-04, 1.0489e-04], device='cuda:3') +2023-02-06 14:54:26,274 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.626e+02 2.642e+02 3.253e+02 4.518e+02 1.030e+03, threshold=6.506e+02, percent-clipped=5.0 +2023-02-06 14:54:37,784 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107677.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:54:39,579 INFO [train.py:901] (3/4) Epoch 14, batch 2600, loss[loss=0.2387, simple_loss=0.3043, pruned_loss=0.08655, over 7975.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.3039, pruned_loss=0.07491, over 1621971.22 frames. ], batch size: 21, lr: 5.49e-03, grad_scale: 8.0 +2023-02-06 14:55:12,836 INFO [train.py:901] (3/4) Epoch 14, batch 2650, loss[loss=0.259, simple_loss=0.341, pruned_loss=0.08849, over 8260.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.3043, pruned_loss=0.0747, over 1618865.67 frames. ], batch size: 24, lr: 5.49e-03, grad_scale: 8.0 +2023-02-06 14:55:30,857 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107755.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:55:34,865 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.771e+02 2.443e+02 2.980e+02 3.881e+02 9.981e+02, threshold=5.960e+02, percent-clipped=6.0 +2023-02-06 14:55:38,610 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2276, 2.6675, 3.1845, 1.5690, 3.2567, 2.0565, 1.6669, 2.3041], + device='cuda:3'), covar=tensor([0.0660, 0.0325, 0.0166, 0.0591, 0.0283, 0.0614, 0.0651, 0.0395], + device='cuda:3'), in_proj_covar=tensor([0.0415, 0.0353, 0.0308, 0.0411, 0.0341, 0.0499, 0.0376, 0.0381], + device='cuda:3'), out_proj_covar=tensor([1.1548e-04, 9.5921e-05, 8.3857e-05, 1.1200e-04, 9.3445e-05, 1.4695e-04, + 1.0487e-04, 1.0510e-04], device='cuda:3') +2023-02-06 14:55:49,934 INFO [train.py:901] (3/4) Epoch 14, batch 2700, loss[loss=0.2507, simple_loss=0.321, pruned_loss=0.09024, over 6675.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.3036, pruned_loss=0.07431, over 1615728.45 frames. ], batch size: 71, lr: 5.49e-03, grad_scale: 8.0 +2023-02-06 14:56:23,701 INFO [train.py:901] (3/4) Epoch 14, batch 2750, loss[loss=0.2016, simple_loss=0.2858, pruned_loss=0.05874, over 8235.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3038, pruned_loss=0.07473, over 1611110.30 frames. ], batch size: 22, lr: 5.49e-03, grad_scale: 8.0 +2023-02-06 14:56:44,697 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.545e+02 2.404e+02 2.918e+02 3.592e+02 1.217e+03, threshold=5.837e+02, percent-clipped=4.0 +2023-02-06 14:56:53,331 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=107872.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:56:54,149 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107873.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 14:56:59,458 INFO [train.py:901] (3/4) Epoch 14, batch 2800, loss[loss=0.2502, simple_loss=0.3195, pruned_loss=0.09045, over 8293.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.304, pruned_loss=0.07458, over 1614204.77 frames. ], batch size: 23, lr: 5.49e-03, grad_scale: 8.0 +2023-02-06 14:57:03,888 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107885.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:57:12,678 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107898.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 14:57:20,767 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107910.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:57:33,815 INFO [train.py:901] (3/4) Epoch 14, batch 2850, loss[loss=0.2556, simple_loss=0.3184, pruned_loss=0.09643, over 7655.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.3038, pruned_loss=0.07446, over 1615954.75 frames. ], batch size: 19, lr: 5.48e-03, grad_scale: 8.0 +2023-02-06 14:57:54,102 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.641e+02 2.478e+02 3.087e+02 3.919e+02 8.173e+02, threshold=6.173e+02, percent-clipped=5.0 +2023-02-06 14:58:08,145 INFO [train.py:901] (3/4) Epoch 14, batch 2900, loss[loss=0.2139, simple_loss=0.289, pruned_loss=0.06936, over 7805.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.3037, pruned_loss=0.07425, over 1615017.36 frames. ], batch size: 19, lr: 5.48e-03, grad_scale: 8.0 +2023-02-06 14:58:12,782 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107987.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:58:22,939 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6555, 2.2638, 4.3758, 1.5019, 2.8710, 2.2304, 1.7507, 2.9022], + device='cuda:3'), covar=tensor([0.1662, 0.2307, 0.0641, 0.3814, 0.1642, 0.2688, 0.1826, 0.2116], + device='cuda:3'), in_proj_covar=tensor([0.0493, 0.0541, 0.0533, 0.0586, 0.0616, 0.0554, 0.0485, 0.0610], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 14:58:27,879 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 14:58:37,447 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7362, 2.2860, 3.4620, 2.6267, 3.2545, 2.3851, 2.1777, 1.7451], + device='cuda:3'), covar=tensor([0.4050, 0.4415, 0.1397, 0.2985, 0.2104, 0.2407, 0.1635, 0.4807], + device='cuda:3'), in_proj_covar=tensor([0.0900, 0.0898, 0.0742, 0.0868, 0.0951, 0.0828, 0.0710, 0.0780], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 14:58:38,642 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=108021.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 14:58:44,582 INFO [train.py:901] (3/4) Epoch 14, batch 2950, loss[loss=0.261, simple_loss=0.3303, pruned_loss=0.09586, over 7022.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.3043, pruned_loss=0.07431, over 1615759.73 frames. ], batch size: 71, lr: 5.48e-03, grad_scale: 8.0 +2023-02-06 14:59:04,838 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.603e+02 2.696e+02 3.199e+02 4.019e+02 8.231e+02, threshold=6.398e+02, percent-clipped=3.0 +2023-02-06 14:59:06,751 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 14:59:18,173 INFO [train.py:901] (3/4) Epoch 14, batch 3000, loss[loss=0.2092, simple_loss=0.2818, pruned_loss=0.0683, over 7233.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.3038, pruned_loss=0.0739, over 1612876.80 frames. ], batch size: 16, lr: 5.48e-03, grad_scale: 8.0 +2023-02-06 14:59:18,173 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 14:59:23,432 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3310, 1.5503, 1.5647, 0.9796, 1.5631, 1.1875, 0.3254, 1.4875], + device='cuda:3'), covar=tensor([0.0395, 0.0332, 0.0266, 0.0429, 0.0390, 0.0887, 0.0768, 0.0262], + device='cuda:3'), in_proj_covar=tensor([0.0419, 0.0356, 0.0310, 0.0411, 0.0342, 0.0500, 0.0379, 0.0383], + device='cuda:3'), out_proj_covar=tensor([1.1663e-04, 9.6576e-05, 8.4118e-05, 1.1208e-04, 9.3709e-05, 1.4718e-04, + 1.0580e-04, 1.0561e-04], device='cuda:3') +2023-02-06 14:59:26,783 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6903, 1.7671, 1.5901, 2.1931, 1.1565, 1.4240, 1.5784, 1.7851], + device='cuda:3'), covar=tensor([0.0702, 0.0898, 0.0985, 0.0485, 0.1194, 0.1357, 0.0881, 0.0769], + device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0205, 0.0249, 0.0211, 0.0213, 0.0249, 0.0253, 0.0214], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 14:59:30,506 INFO [train.py:935] (3/4) Epoch 14, validation: loss=0.1827, simple_loss=0.283, pruned_loss=0.04121, over 944034.00 frames. +2023-02-06 14:59:30,506 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 14:59:43,702 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=108099.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:00:05,760 INFO [train.py:901] (3/4) Epoch 14, batch 3050, loss[loss=0.1955, simple_loss=0.2572, pruned_loss=0.06691, over 7411.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3033, pruned_loss=0.07398, over 1612510.78 frames. ], batch size: 17, lr: 5.48e-03, grad_scale: 8.0 +2023-02-06 15:00:10,682 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108136.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:00:28,076 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.637e+02 2.670e+02 3.118e+02 3.835e+02 7.160e+02, threshold=6.236e+02, percent-clipped=1.0 +2023-02-06 15:00:41,680 INFO [train.py:901] (3/4) Epoch 14, batch 3100, loss[loss=0.2592, simple_loss=0.332, pruned_loss=0.09315, over 8525.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.3026, pruned_loss=0.07355, over 1613593.60 frames. ], batch size: 49, lr: 5.48e-03, grad_scale: 8.0 +2023-02-06 15:00:55,329 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9783, 4.0946, 2.5475, 2.8650, 2.9461, 2.5033, 2.7372, 2.9781], + device='cuda:3'), covar=tensor([0.1637, 0.0282, 0.0848, 0.0701, 0.0631, 0.1019, 0.1001, 0.1148], + device='cuda:3'), in_proj_covar=tensor([0.0343, 0.0229, 0.0318, 0.0296, 0.0296, 0.0319, 0.0339, 0.0306], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 15:01:04,698 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108214.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:01:15,963 INFO [train.py:901] (3/4) Epoch 14, batch 3150, loss[loss=0.2545, simple_loss=0.3248, pruned_loss=0.09204, over 8553.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.3022, pruned_loss=0.07327, over 1614717.57 frames. ], batch size: 31, lr: 5.48e-03, grad_scale: 8.0 +2023-02-06 15:01:17,031 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 15:01:24,681 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108243.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:01:26,343 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2023-02-06 15:01:37,168 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.660e+02 2.570e+02 3.163e+02 4.155e+02 7.848e+02, threshold=6.326e+02, percent-clipped=5.0 +2023-02-06 15:01:43,480 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108268.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:01:51,515 INFO [train.py:901] (3/4) Epoch 14, batch 3200, loss[loss=0.2172, simple_loss=0.2999, pruned_loss=0.06726, over 8497.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.3024, pruned_loss=0.07346, over 1615418.26 frames. ], batch size: 26, lr: 5.48e-03, grad_scale: 8.0 +2023-02-06 15:02:23,930 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1488, 1.8137, 2.7127, 2.2068, 2.4423, 2.0843, 1.7321, 1.3205], + device='cuda:3'), covar=tensor([0.4553, 0.4258, 0.1336, 0.2731, 0.2088, 0.2598, 0.1788, 0.4236], + device='cuda:3'), in_proj_covar=tensor([0.0901, 0.0900, 0.0740, 0.0868, 0.0958, 0.0830, 0.0712, 0.0779], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 15:02:25,774 INFO [train.py:901] (3/4) Epoch 14, batch 3250, loss[loss=0.2472, simple_loss=0.3252, pruned_loss=0.0846, over 8447.00 frames. ], tot_loss[loss=0.2253, simple_loss=0.3031, pruned_loss=0.07376, over 1618634.78 frames. ], batch size: 29, lr: 5.47e-03, grad_scale: 8.0 +2023-02-06 15:02:35,709 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108343.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:02:47,025 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.807e+02 2.638e+02 3.239e+02 4.086e+02 1.012e+03, threshold=6.478e+02, percent-clipped=4.0 +2023-02-06 15:03:02,204 INFO [train.py:901] (3/4) Epoch 14, batch 3300, loss[loss=0.2116, simple_loss=0.2921, pruned_loss=0.06554, over 8693.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.3029, pruned_loss=0.0733, over 1620438.14 frames. ], batch size: 34, lr: 5.47e-03, grad_scale: 8.0 +2023-02-06 15:03:05,089 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8472, 1.5232, 3.9552, 1.3662, 3.4918, 3.2805, 3.5768, 3.4843], + device='cuda:3'), covar=tensor([0.0587, 0.4162, 0.0630, 0.3958, 0.1203, 0.1027, 0.0637, 0.0701], + device='cuda:3'), in_proj_covar=tensor([0.0528, 0.0601, 0.0616, 0.0567, 0.0641, 0.0550, 0.0535, 0.0606], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 15:03:11,254 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108392.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:03:27,909 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108417.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:03:36,466 INFO [train.py:901] (3/4) Epoch 14, batch 3350, loss[loss=0.2016, simple_loss=0.2823, pruned_loss=0.06041, over 8115.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.3027, pruned_loss=0.07347, over 1617198.64 frames. ], batch size: 23, lr: 5.47e-03, grad_scale: 8.0 +2023-02-06 15:03:49,804 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108450.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:03:57,198 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.535e+02 2.656e+02 3.299e+02 4.467e+02 8.781e+02, threshold=6.597e+02, percent-clipped=5.0 +2023-02-06 15:04:04,263 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108470.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:04:10,800 INFO [train.py:901] (3/4) Epoch 14, batch 3400, loss[loss=0.1845, simple_loss=0.2713, pruned_loss=0.04889, over 7915.00 frames. ], tot_loss[loss=0.2251, simple_loss=0.3028, pruned_loss=0.07368, over 1615581.42 frames. ], batch size: 20, lr: 5.47e-03, grad_scale: 8.0 +2023-02-06 15:04:22,163 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108495.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:04:46,706 INFO [train.py:901] (3/4) Epoch 14, batch 3450, loss[loss=0.2231, simple_loss=0.2956, pruned_loss=0.0753, over 7704.00 frames. ], tot_loss[loss=0.2254, simple_loss=0.3033, pruned_loss=0.07374, over 1617711.69 frames. ], batch size: 18, lr: 5.47e-03, grad_scale: 8.0 +2023-02-06 15:05:07,907 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.548e+02 2.481e+02 3.055e+02 3.627e+02 7.933e+02, threshold=6.110e+02, percent-clipped=3.0 +2023-02-06 15:05:21,068 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.76 vs. limit=5.0 +2023-02-06 15:05:21,985 INFO [train.py:901] (3/4) Epoch 14, batch 3500, loss[loss=0.2236, simple_loss=0.3086, pruned_loss=0.06927, over 8035.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3025, pruned_loss=0.07334, over 1613718.49 frames. ], batch size: 22, lr: 5.47e-03, grad_scale: 8.0 +2023-02-06 15:05:29,155 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 15:05:31,989 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108595.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:05:57,720 INFO [train.py:901] (3/4) Epoch 14, batch 3550, loss[loss=0.2399, simple_loss=0.3208, pruned_loss=0.07951, over 8365.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.3021, pruned_loss=0.07327, over 1614560.83 frames. ], batch size: 24, lr: 5.47e-03, grad_scale: 8.0 +2023-02-06 15:06:17,928 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.549e+02 2.417e+02 3.151e+02 4.175e+02 8.210e+02, threshold=6.301e+02, percent-clipped=3.0 +2023-02-06 15:06:31,412 INFO [train.py:901] (3/4) Epoch 14, batch 3600, loss[loss=0.2386, simple_loss=0.3163, pruned_loss=0.08041, over 8461.00 frames. ], tot_loss[loss=0.224, simple_loss=0.302, pruned_loss=0.07293, over 1616129.61 frames. ], batch size: 25, lr: 5.47e-03, grad_scale: 8.0 +2023-02-06 15:06:36,134 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=108687.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:06:54,657 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-06 15:07:07,042 INFO [train.py:901] (3/4) Epoch 14, batch 3650, loss[loss=0.1844, simple_loss=0.265, pruned_loss=0.0519, over 7217.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.3018, pruned_loss=0.07292, over 1613957.73 frames. ], batch size: 16, lr: 5.46e-03, grad_scale: 8.0 +2023-02-06 15:07:27,807 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.861e+02 2.655e+02 3.191e+02 3.880e+02 8.243e+02, threshold=6.382e+02, percent-clipped=2.0 +2023-02-06 15:07:30,614 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 15:07:37,170 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-06 15:07:41,536 INFO [train.py:901] (3/4) Epoch 14, batch 3700, loss[loss=0.1842, simple_loss=0.2723, pruned_loss=0.04801, over 8240.00 frames. ], tot_loss[loss=0.223, simple_loss=0.3009, pruned_loss=0.07249, over 1615099.04 frames. ], batch size: 22, lr: 5.46e-03, grad_scale: 8.0 +2023-02-06 15:07:41,783 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([0.7121, 0.8143, 0.8171, 0.3574, 0.8171, 0.6461, 0.0897, 0.7642], + device='cuda:3'), covar=tensor([0.0214, 0.0186, 0.0159, 0.0282, 0.0203, 0.0406, 0.0416, 0.0157], + device='cuda:3'), in_proj_covar=tensor([0.0414, 0.0350, 0.0306, 0.0406, 0.0338, 0.0494, 0.0372, 0.0378], + device='cuda:3'), out_proj_covar=tensor([1.1519e-04, 9.4633e-05, 8.3079e-05, 1.1040e-04, 9.2383e-05, 1.4526e-04, + 1.0375e-04, 1.0391e-04], device='cuda:3') +2023-02-06 15:07:50,772 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=108794.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:07:56,400 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108802.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:08:01,083 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108809.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:08:15,867 INFO [train.py:901] (3/4) Epoch 14, batch 3750, loss[loss=0.2286, simple_loss=0.3123, pruned_loss=0.07248, over 8506.00 frames. ], tot_loss[loss=0.2222, simple_loss=0.3001, pruned_loss=0.07212, over 1611623.30 frames. ], batch size: 28, lr: 5.46e-03, grad_scale: 8.0 +2023-02-06 15:08:16,011 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8223, 1.2649, 3.9606, 1.4251, 3.5317, 3.2798, 3.5136, 3.4004], + device='cuda:3'), covar=tensor([0.0620, 0.4378, 0.0577, 0.3705, 0.1146, 0.1079, 0.0668, 0.0701], + device='cuda:3'), in_proj_covar=tensor([0.0532, 0.0606, 0.0621, 0.0567, 0.0644, 0.0553, 0.0543, 0.0606], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 15:08:37,495 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.674e+02 2.414e+02 2.846e+02 3.664e+02 8.039e+02, threshold=5.692e+02, percent-clipped=5.0 +2023-02-06 15:08:51,947 INFO [train.py:901] (3/4) Epoch 14, batch 3800, loss[loss=0.1912, simple_loss=0.2697, pruned_loss=0.05631, over 8239.00 frames. ], tot_loss[loss=0.2228, simple_loss=0.3005, pruned_loss=0.07256, over 1611135.77 frames. ], batch size: 22, lr: 5.46e-03, grad_scale: 8.0 +2023-02-06 15:09:01,185 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8802, 1.6069, 3.1736, 1.1164, 2.2497, 3.4485, 3.7343, 2.6303], + device='cuda:3'), covar=tensor([0.1436, 0.1899, 0.0504, 0.2945, 0.1286, 0.0454, 0.0534, 0.1215], + device='cuda:3'), in_proj_covar=tensor([0.0274, 0.0305, 0.0268, 0.0298, 0.0284, 0.0247, 0.0369, 0.0297], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 15:09:12,261 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108909.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:09:20,730 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4135, 1.9341, 2.8162, 2.3080, 2.6886, 2.2571, 1.9396, 1.3709], + device='cuda:3'), covar=tensor([0.4195, 0.4401, 0.1391, 0.2755, 0.1846, 0.2349, 0.1785, 0.4232], + device='cuda:3'), in_proj_covar=tensor([0.0894, 0.0897, 0.0743, 0.0871, 0.0950, 0.0827, 0.0710, 0.0778], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 15:09:26,380 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6495, 1.7935, 1.6467, 2.2523, 1.0409, 1.3868, 1.6513, 1.9160], + device='cuda:3'), covar=tensor([0.0890, 0.0922, 0.1053, 0.0478, 0.1209, 0.1626, 0.0920, 0.0793], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0208, 0.0252, 0.0214, 0.0216, 0.0252, 0.0258, 0.0215], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 15:09:26,887 INFO [train.py:901] (3/4) Epoch 14, batch 3850, loss[loss=0.206, simple_loss=0.2777, pruned_loss=0.06716, over 7921.00 frames. ], tot_loss[loss=0.2236, simple_loss=0.3016, pruned_loss=0.07282, over 1612618.02 frames. ], batch size: 20, lr: 5.46e-03, grad_scale: 16.0 +2023-02-06 15:09:33,935 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=108939.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:09:35,942 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 15:09:49,061 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.465e+02 2.574e+02 3.020e+02 4.517e+02 9.725e+02, threshold=6.039e+02, percent-clipped=15.0 +2023-02-06 15:10:04,288 INFO [train.py:901] (3/4) Epoch 14, batch 3900, loss[loss=0.2217, simple_loss=0.2962, pruned_loss=0.07355, over 7919.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.3022, pruned_loss=0.07324, over 1614755.34 frames. ], batch size: 20, lr: 5.46e-03, grad_scale: 16.0 +2023-02-06 15:10:39,028 INFO [train.py:901] (3/4) Epoch 14, batch 3950, loss[loss=0.2313, simple_loss=0.3004, pruned_loss=0.08115, over 8088.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.3018, pruned_loss=0.073, over 1613627.44 frames. ], batch size: 21, lr: 5.46e-03, grad_scale: 16.0 +2023-02-06 15:10:49,250 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-06 15:10:51,040 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1105, 1.6364, 3.2913, 1.3920, 2.2517, 3.5429, 3.6599, 3.0362], + device='cuda:3'), covar=tensor([0.0968, 0.1571, 0.0336, 0.2210, 0.0999, 0.0257, 0.0489, 0.0659], + device='cuda:3'), in_proj_covar=tensor([0.0272, 0.0305, 0.0269, 0.0298, 0.0285, 0.0248, 0.0368, 0.0297], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 15:10:53,915 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 15:10:56,306 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109054.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:10:56,962 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109055.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 15:10:59,097 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109058.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:11:00,257 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.734e+02 2.458e+02 2.966e+02 3.777e+02 8.079e+02, threshold=5.932e+02, percent-clipped=4.0 +2023-02-06 15:11:05,994 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7235, 3.0302, 2.5588, 4.0816, 1.7098, 2.2908, 2.5771, 3.3531], + device='cuda:3'), covar=tensor([0.0602, 0.0827, 0.0833, 0.0201, 0.1174, 0.1322, 0.1016, 0.0726], + device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0207, 0.0250, 0.0211, 0.0214, 0.0251, 0.0255, 0.0214], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 15:11:14,753 INFO [train.py:901] (3/4) Epoch 14, batch 4000, loss[loss=0.2444, simple_loss=0.3235, pruned_loss=0.08264, over 8464.00 frames. ], tot_loss[loss=0.2226, simple_loss=0.3011, pruned_loss=0.07206, over 1613731.18 frames. ], batch size: 29, lr: 5.46e-03, grad_scale: 16.0 +2023-02-06 15:11:17,679 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109083.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:11:22,496 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9406, 2.4188, 3.4241, 1.4938, 1.5399, 3.5992, 0.4034, 1.9958], + device='cuda:3'), covar=tensor([0.1738, 0.1510, 0.0469, 0.3241, 0.4012, 0.0283, 0.3273, 0.1752], + device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0176, 0.0108, 0.0217, 0.0259, 0.0111, 0.0162, 0.0173], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 15:11:42,460 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8288, 1.8266, 2.2767, 1.6099, 1.2662, 2.3974, 0.4233, 1.3237], + device='cuda:3'), covar=tensor([0.1986, 0.1329, 0.0466, 0.1725, 0.3688, 0.0334, 0.2919, 0.1856], + device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0175, 0.0107, 0.0217, 0.0258, 0.0111, 0.0161, 0.0172], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 15:11:50,548 INFO [train.py:901] (3/4) Epoch 14, batch 4050, loss[loss=0.1792, simple_loss=0.2579, pruned_loss=0.05031, over 7646.00 frames. ], tot_loss[loss=0.2228, simple_loss=0.3011, pruned_loss=0.07228, over 1611168.79 frames. ], batch size: 19, lr: 5.45e-03, grad_scale: 16.0 +2023-02-06 15:12:06,805 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109153.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:12:11,645 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.858e+02 2.362e+02 2.684e+02 3.543e+02 7.215e+02, threshold=5.369e+02, percent-clipped=4.0 +2023-02-06 15:12:16,025 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109165.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:12:26,279 INFO [train.py:901] (3/4) Epoch 14, batch 4100, loss[loss=0.2269, simple_loss=0.2978, pruned_loss=0.07798, over 8085.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.3025, pruned_loss=0.07251, over 1613973.54 frames. ], batch size: 21, lr: 5.45e-03, grad_scale: 16.0 +2023-02-06 15:12:33,985 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109190.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:12:50,497 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109212.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:13:02,499 INFO [train.py:901] (3/4) Epoch 14, batch 4150, loss[loss=0.1987, simple_loss=0.2654, pruned_loss=0.06599, over 7705.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.3032, pruned_loss=0.07316, over 1617599.16 frames. ], batch size: 18, lr: 5.45e-03, grad_scale: 8.0 +2023-02-06 15:13:06,913 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5490, 2.1386, 3.2965, 1.3425, 2.5427, 1.9926, 1.8047, 2.3087], + device='cuda:3'), covar=tensor([0.1814, 0.2066, 0.0773, 0.3976, 0.1539, 0.2833, 0.1797, 0.2228], + device='cuda:3'), in_proj_covar=tensor([0.0500, 0.0553, 0.0542, 0.0597, 0.0622, 0.0565, 0.0490, 0.0620], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 15:13:20,630 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4776, 1.6567, 4.4461, 1.9859, 2.2719, 5.0303, 5.0526, 4.3496], + device='cuda:3'), covar=tensor([0.0953, 0.1604, 0.0228, 0.1921, 0.1257, 0.0154, 0.0286, 0.0530], + device='cuda:3'), in_proj_covar=tensor([0.0271, 0.0304, 0.0268, 0.0298, 0.0283, 0.0246, 0.0366, 0.0294], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 15:13:23,981 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.711e+02 2.677e+02 3.078e+02 3.893e+02 8.547e+02, threshold=6.157e+02, percent-clipped=10.0 +2023-02-06 15:13:28,941 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109268.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:13:35,716 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 15:13:37,137 INFO [train.py:901] (3/4) Epoch 14, batch 4200, loss[loss=0.2668, simple_loss=0.328, pruned_loss=0.1028, over 7396.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.3017, pruned_loss=0.07237, over 1614320.50 frames. ], batch size: 72, lr: 5.45e-03, grad_scale: 8.0 +2023-02-06 15:13:59,676 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109310.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:14:01,000 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109312.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:14:01,568 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 15:14:14,550 INFO [train.py:901] (3/4) Epoch 14, batch 4250, loss[loss=0.2276, simple_loss=0.3056, pruned_loss=0.07479, over 8420.00 frames. ], tot_loss[loss=0.2226, simple_loss=0.3013, pruned_loss=0.07201, over 1611599.55 frames. ], batch size: 48, lr: 5.45e-03, grad_scale: 8.0 +2023-02-06 15:14:18,150 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109335.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:14:18,812 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109336.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:14:35,684 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.473e+02 2.488e+02 3.016e+02 3.845e+02 8.299e+02, threshold=6.033e+02, percent-clipped=4.0 +2023-02-06 15:14:48,340 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 15:14:48,668 INFO [train.py:901] (3/4) Epoch 14, batch 4300, loss[loss=0.2474, simple_loss=0.3272, pruned_loss=0.08377, over 8557.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3025, pruned_loss=0.07336, over 1609044.75 frames. ], batch size: 31, lr: 5.45e-03, grad_scale: 8.0 +2023-02-06 15:15:01,893 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109399.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 15:15:16,055 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3011, 1.9228, 2.7917, 2.2216, 2.6426, 2.2204, 1.8794, 1.3102], + device='cuda:3'), covar=tensor([0.4567, 0.4334, 0.1335, 0.2979, 0.1963, 0.2473, 0.1752, 0.4548], + device='cuda:3'), in_proj_covar=tensor([0.0902, 0.0899, 0.0745, 0.0877, 0.0953, 0.0829, 0.0714, 0.0784], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 15:15:24,488 INFO [train.py:901] (3/4) Epoch 14, batch 4350, loss[loss=0.198, simple_loss=0.2676, pruned_loss=0.06424, over 7806.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.3018, pruned_loss=0.07318, over 1604305.18 frames. ], batch size: 19, lr: 5.45e-03, grad_scale: 8.0 +2023-02-06 15:15:34,088 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 15:15:47,274 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.768e+02 2.678e+02 3.270e+02 4.253e+02 1.326e+03, threshold=6.540e+02, percent-clipped=8.0 +2023-02-06 15:16:00,545 INFO [train.py:901] (3/4) Epoch 14, batch 4400, loss[loss=0.2214, simple_loss=0.298, pruned_loss=0.07236, over 8357.00 frames. ], tot_loss[loss=0.2236, simple_loss=0.3015, pruned_loss=0.07287, over 1607389.98 frames. ], batch size: 24, lr: 5.45e-03, grad_scale: 8.0 +2023-02-06 15:16:11,101 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6490, 1.4634, 1.5543, 1.2895, 0.9670, 1.3798, 1.5669, 1.2060], + device='cuda:3'), covar=tensor([0.0556, 0.1231, 0.1634, 0.1399, 0.0649, 0.1456, 0.0702, 0.0693], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0189, 0.0156, 0.0101, 0.0162, 0.0113, 0.0137], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 15:16:15,756 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 15:16:24,162 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109514.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 15:16:31,894 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109524.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:16:35,898 INFO [train.py:901] (3/4) Epoch 14, batch 4450, loss[loss=0.2291, simple_loss=0.3112, pruned_loss=0.07352, over 8500.00 frames. ], tot_loss[loss=0.2236, simple_loss=0.3014, pruned_loss=0.07291, over 1608295.26 frames. ], batch size: 26, lr: 5.44e-03, grad_scale: 8.0 +2023-02-06 15:16:45,978 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 15:16:49,990 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109549.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:16:55,370 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109556.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:16:58,634 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.676e+02 2.456e+02 2.864e+02 3.608e+02 1.087e+03, threshold=5.728e+02, percent-clipped=4.0 +2023-02-06 15:17:04,034 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-06 15:17:11,135 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 15:17:12,486 INFO [train.py:901] (3/4) Epoch 14, batch 4500, loss[loss=0.241, simple_loss=0.3154, pruned_loss=0.08325, over 8477.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.3018, pruned_loss=0.07319, over 1611165.48 frames. ], batch size: 27, lr: 5.44e-03, grad_scale: 8.0 +2023-02-06 15:17:24,283 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109597.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:17:47,002 INFO [train.py:901] (3/4) Epoch 14, batch 4550, loss[loss=0.2322, simple_loss=0.3145, pruned_loss=0.075, over 8300.00 frames. ], tot_loss[loss=0.2244, simple_loss=0.3022, pruned_loss=0.07332, over 1607124.78 frames. ], batch size: 49, lr: 5.44e-03, grad_scale: 8.0 +2023-02-06 15:17:57,570 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1075, 1.7513, 2.3800, 1.9820, 2.2705, 2.0595, 1.7891, 1.0720], + device='cuda:3'), covar=tensor([0.5057, 0.4707, 0.1748, 0.3132, 0.2193, 0.2671, 0.1819, 0.4983], + device='cuda:3'), in_proj_covar=tensor([0.0898, 0.0897, 0.0743, 0.0872, 0.0951, 0.0827, 0.0711, 0.0780], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 15:18:05,731 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109656.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:18:09,030 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.742e+02 2.640e+02 3.232e+02 4.162e+02 9.021e+02, threshold=6.464e+02, percent-clipped=8.0 +2023-02-06 15:18:16,725 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109671.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:18:21,359 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109677.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:18:22,045 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3521, 1.8795, 3.3587, 1.4418, 2.3959, 3.6991, 3.6886, 3.1884], + device='cuda:3'), covar=tensor([0.0836, 0.1405, 0.0353, 0.2104, 0.1019, 0.0206, 0.0503, 0.0540], + device='cuda:3'), in_proj_covar=tensor([0.0276, 0.0307, 0.0271, 0.0301, 0.0288, 0.0249, 0.0372, 0.0297], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 15:18:22,792 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3415, 1.6223, 1.6825, 1.0205, 1.7057, 1.2529, 0.3047, 1.5666], + device='cuda:3'), covar=tensor([0.0360, 0.0244, 0.0204, 0.0335, 0.0275, 0.0636, 0.0612, 0.0179], + device='cuda:3'), in_proj_covar=tensor([0.0414, 0.0352, 0.0302, 0.0409, 0.0340, 0.0496, 0.0372, 0.0378], + device='cuda:3'), out_proj_covar=tensor([1.1513e-04, 9.5241e-05, 8.1927e-05, 1.1133e-04, 9.2941e-05, 1.4597e-04, + 1.0355e-04, 1.0382e-04], device='cuda:3') +2023-02-06 15:18:23,263 INFO [train.py:901] (3/4) Epoch 14, batch 4600, loss[loss=0.2655, simple_loss=0.3362, pruned_loss=0.09741, over 8493.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.3019, pruned_loss=0.07324, over 1610804.39 frames. ], batch size: 28, lr: 5.44e-03, grad_scale: 8.0 +2023-02-06 15:18:23,340 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109680.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:18:56,947 INFO [train.py:901] (3/4) Epoch 14, batch 4650, loss[loss=0.2505, simple_loss=0.3361, pruned_loss=0.08246, over 8017.00 frames. ], tot_loss[loss=0.224, simple_loss=0.302, pruned_loss=0.073, over 1608784.80 frames. ], batch size: 22, lr: 5.44e-03, grad_scale: 8.0 +2023-02-06 15:19:18,715 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.564e+02 2.556e+02 3.032e+02 3.907e+02 9.020e+02, threshold=6.065e+02, percent-clipped=4.0 +2023-02-06 15:19:19,811 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 15:19:24,816 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109770.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 15:19:25,451 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109771.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:19:31,312 INFO [train.py:901] (3/4) Epoch 14, batch 4700, loss[loss=0.2474, simple_loss=0.3286, pruned_loss=0.08315, over 8532.00 frames. ], tot_loss[loss=0.224, simple_loss=0.3019, pruned_loss=0.07311, over 1608625.47 frames. ], batch size: 49, lr: 5.44e-03, grad_scale: 8.0 +2023-02-06 15:19:37,765 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 +2023-02-06 15:19:42,837 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109795.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:19:42,862 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109795.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 15:19:56,334 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 +2023-02-06 15:20:06,516 INFO [train.py:901] (3/4) Epoch 14, batch 4750, loss[loss=0.2266, simple_loss=0.3126, pruned_loss=0.07032, over 8607.00 frames. ], tot_loss[loss=0.2255, simple_loss=0.3033, pruned_loss=0.0738, over 1612695.61 frames. ], batch size: 34, lr: 5.44e-03, grad_scale: 8.0 +2023-02-06 15:20:10,490 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 15:20:12,430 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 15:20:26,961 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.786e+02 2.643e+02 3.166e+02 4.371e+02 1.104e+03, threshold=6.332e+02, percent-clipped=5.0 +2023-02-06 15:20:40,299 INFO [train.py:901] (3/4) Epoch 14, batch 4800, loss[loss=0.2721, simple_loss=0.3383, pruned_loss=0.1029, over 7984.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.3035, pruned_loss=0.07402, over 1612750.82 frames. ], batch size: 21, lr: 5.44e-03, grad_scale: 8.0 +2023-02-06 15:20:52,070 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 +2023-02-06 15:20:56,651 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0 +2023-02-06 15:21:03,220 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 15:21:14,262 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109927.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:21:16,047 INFO [train.py:901] (3/4) Epoch 14, batch 4850, loss[loss=0.217, simple_loss=0.302, pruned_loss=0.06604, over 7985.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.3028, pruned_loss=0.07344, over 1615073.12 frames. ], batch size: 21, lr: 5.43e-03, grad_scale: 8.0 +2023-02-06 15:21:23,121 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-06 15:21:23,510 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=109941.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:21:31,143 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109952.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:21:34,569 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3931, 1.7834, 2.6525, 1.2630, 1.8147, 1.7236, 1.5267, 1.8160], + device='cuda:3'), covar=tensor([0.1972, 0.2308, 0.0841, 0.4142, 0.1809, 0.3088, 0.2003, 0.2305], + device='cuda:3'), in_proj_covar=tensor([0.0498, 0.0549, 0.0538, 0.0595, 0.0620, 0.0563, 0.0487, 0.0617], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 15:21:37,039 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.601e+02 2.400e+02 2.854e+02 3.344e+02 7.947e+02, threshold=5.708e+02, percent-clipped=2.0 +2023-02-06 15:21:49,880 INFO [train.py:901] (3/4) Epoch 14, batch 4900, loss[loss=0.2363, simple_loss=0.3183, pruned_loss=0.07714, over 8465.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3023, pruned_loss=0.07348, over 1614453.06 frames. ], batch size: 25, lr: 5.43e-03, grad_scale: 8.0 +2023-02-06 15:22:18,845 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3610, 1.3193, 2.3429, 1.2213, 2.0450, 2.5078, 2.5968, 2.1125], + device='cuda:3'), covar=tensor([0.1037, 0.1243, 0.0453, 0.1963, 0.0743, 0.0401, 0.0697, 0.0727], + device='cuda:3'), in_proj_covar=tensor([0.0271, 0.0303, 0.0267, 0.0296, 0.0283, 0.0245, 0.0367, 0.0292], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 15:22:19,457 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110021.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:22:20,954 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7240, 1.9346, 2.1488, 1.3619, 2.2068, 1.5208, 0.5058, 1.7793], + device='cuda:3'), covar=tensor([0.0397, 0.0262, 0.0194, 0.0377, 0.0280, 0.0719, 0.0670, 0.0233], + device='cuda:3'), in_proj_covar=tensor([0.0419, 0.0356, 0.0305, 0.0411, 0.0343, 0.0503, 0.0376, 0.0381], + device='cuda:3'), out_proj_covar=tensor([1.1645e-04, 9.6243e-05, 8.2705e-05, 1.1197e-04, 9.3788e-05, 1.4787e-04, + 1.0478e-04, 1.0453e-04], device='cuda:3') +2023-02-06 15:22:24,325 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110027.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:22:26,032 INFO [train.py:901] (3/4) Epoch 14, batch 4950, loss[loss=0.1952, simple_loss=0.2641, pruned_loss=0.06319, over 7443.00 frames. ], tot_loss[loss=0.2229, simple_loss=0.3009, pruned_loss=0.07247, over 1613133.76 frames. ], batch size: 17, lr: 5.43e-03, grad_scale: 8.0 +2023-02-06 15:22:30,967 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110035.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:22:41,737 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110051.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:22:42,420 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110052.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:22:45,151 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110056.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:22:48,336 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.646e+02 2.585e+02 3.180e+02 4.032e+02 7.448e+02, threshold=6.360e+02, percent-clipped=3.0 +2023-02-06 15:22:49,155 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110062.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:22:58,272 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110076.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:23:00,771 INFO [train.py:901] (3/4) Epoch 14, batch 5000, loss[loss=0.2423, simple_loss=0.3197, pruned_loss=0.08239, over 8639.00 frames. ], tot_loss[loss=0.2229, simple_loss=0.3006, pruned_loss=0.07262, over 1612177.94 frames. ], batch size: 39, lr: 5.43e-03, grad_scale: 8.0 +2023-02-06 15:23:33,490 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110128.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:23:34,636 INFO [train.py:901] (3/4) Epoch 14, batch 5050, loss[loss=0.237, simple_loss=0.3191, pruned_loss=0.07748, over 8357.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.2993, pruned_loss=0.07195, over 1609763.48 frames. ], batch size: 24, lr: 5.43e-03, grad_scale: 8.0 +2023-02-06 15:23:36,757 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110133.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:23:38,747 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110136.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:23:43,176 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 15:23:57,271 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.368e+02 2.889e+02 3.601e+02 4.263e+02 9.587e+02, threshold=7.203e+02, percent-clipped=6.0 +2023-02-06 15:24:09,951 INFO [train.py:901] (3/4) Epoch 14, batch 5100, loss[loss=0.1834, simple_loss=0.2627, pruned_loss=0.05207, over 7294.00 frames. ], tot_loss[loss=0.221, simple_loss=0.2984, pruned_loss=0.07179, over 1607220.61 frames. ], batch size: 16, lr: 5.43e-03, grad_scale: 8.0 +2023-02-06 15:24:42,801 INFO [train.py:901] (3/4) Epoch 14, batch 5150, loss[loss=0.2204, simple_loss=0.3034, pruned_loss=0.0687, over 8257.00 frames. ], tot_loss[loss=0.2233, simple_loss=0.3005, pruned_loss=0.07306, over 1612423.32 frames. ], batch size: 24, lr: 5.43e-03, grad_scale: 8.0 +2023-02-06 15:25:05,057 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.676e+02 2.455e+02 3.012e+02 3.817e+02 9.599e+02, threshold=6.024e+02, percent-clipped=2.0 +2023-02-06 15:25:17,480 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7429, 1.7245, 2.4194, 1.3566, 1.1567, 2.4234, 0.3445, 1.3936], + device='cuda:3'), covar=tensor([0.1883, 0.1473, 0.0413, 0.2127, 0.4064, 0.0390, 0.2675, 0.1812], + device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0173, 0.0105, 0.0214, 0.0258, 0.0111, 0.0160, 0.0170], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 15:25:19,324 INFO [train.py:901] (3/4) Epoch 14, batch 5200, loss[loss=0.2047, simple_loss=0.2727, pruned_loss=0.06837, over 7784.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.3011, pruned_loss=0.07327, over 1615048.25 frames. ], batch size: 19, lr: 5.43e-03, grad_scale: 8.0 +2023-02-06 15:25:20,641 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1819, 4.1403, 3.7746, 2.0198, 3.7221, 3.7884, 3.7750, 3.4194], + device='cuda:3'), covar=tensor([0.0800, 0.0642, 0.1333, 0.4219, 0.0855, 0.0944, 0.1437, 0.0984], + device='cuda:3'), in_proj_covar=tensor([0.0480, 0.0397, 0.0401, 0.0495, 0.0393, 0.0398, 0.0387, 0.0346], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 15:25:25,793 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 +2023-02-06 15:25:27,241 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 +2023-02-06 15:25:33,778 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110301.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:25:39,486 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 15:25:41,060 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110312.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:25:47,127 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110321.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:25:53,177 INFO [train.py:901] (3/4) Epoch 14, batch 5250, loss[loss=0.2157, simple_loss=0.303, pruned_loss=0.06415, over 8461.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.2996, pruned_loss=0.07248, over 1611630.02 frames. ], batch size: 25, lr: 5.42e-03, grad_scale: 8.0 +2023-02-06 15:25:57,582 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110336.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:25:58,393 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110337.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:26:15,513 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.612e+02 2.532e+02 3.204e+02 3.879e+02 8.466e+02, threshold=6.409e+02, percent-clipped=5.0 +2023-02-06 15:26:28,850 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110379.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:26:29,482 INFO [train.py:901] (3/4) Epoch 14, batch 5300, loss[loss=0.208, simple_loss=0.2754, pruned_loss=0.07033, over 7686.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.3007, pruned_loss=0.07286, over 1615116.32 frames. ], batch size: 18, lr: 5.42e-03, grad_scale: 8.0 +2023-02-06 15:26:37,624 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 +2023-02-06 15:26:39,496 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110392.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:26:48,923 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110406.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:26:56,337 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110417.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:27:03,181 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5145, 2.6893, 1.9311, 2.2258, 2.2755, 1.6579, 2.0189, 2.0635], + device='cuda:3'), covar=tensor([0.1304, 0.0332, 0.1010, 0.0512, 0.0620, 0.1292, 0.0923, 0.0814], + device='cuda:3'), in_proj_covar=tensor([0.0344, 0.0231, 0.0318, 0.0295, 0.0296, 0.0324, 0.0339, 0.0307], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 15:27:05,003 INFO [train.py:901] (3/4) Epoch 14, batch 5350, loss[loss=0.2039, simple_loss=0.2981, pruned_loss=0.05485, over 8327.00 frames. ], tot_loss[loss=0.2231, simple_loss=0.3009, pruned_loss=0.07261, over 1618356.76 frames. ], batch size: 25, lr: 5.42e-03, grad_scale: 8.0 +2023-02-06 15:27:13,486 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-06 15:27:25,505 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.461e+02 2.452e+02 3.047e+02 3.791e+02 6.566e+02, threshold=6.094e+02, percent-clipped=2.0 +2023-02-06 15:27:32,804 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110472.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:27:36,873 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110477.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:27:38,768 INFO [train.py:901] (3/4) Epoch 14, batch 5400, loss[loss=0.211, simple_loss=0.2807, pruned_loss=0.07062, over 7416.00 frames. ], tot_loss[loss=0.2218, simple_loss=0.3001, pruned_loss=0.07173, over 1618698.85 frames. ], batch size: 17, lr: 5.42e-03, grad_scale: 8.0 +2023-02-06 15:27:48,333 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110494.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:28:08,461 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110521.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:28:14,356 INFO [train.py:901] (3/4) Epoch 14, batch 5450, loss[loss=0.2756, simple_loss=0.3545, pruned_loss=0.09832, over 8517.00 frames. ], tot_loss[loss=0.2212, simple_loss=0.2997, pruned_loss=0.07132, over 1618628.98 frames. ], batch size: 28, lr: 5.42e-03, grad_scale: 8.0 +2023-02-06 15:28:30,476 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 15:28:34,934 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.661e+02 2.429e+02 2.846e+02 3.589e+02 7.640e+02, threshold=5.692e+02, percent-clipped=1.0 +2023-02-06 15:28:39,770 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1750, 2.1058, 1.5989, 1.8528, 1.7278, 1.3873, 1.5794, 1.6071], + device='cuda:3'), covar=tensor([0.1145, 0.0412, 0.1107, 0.0489, 0.0717, 0.1427, 0.0889, 0.0753], + device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0235, 0.0322, 0.0297, 0.0298, 0.0328, 0.0344, 0.0311], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 15:28:47,542 INFO [train.py:901] (3/4) Epoch 14, batch 5500, loss[loss=0.2226, simple_loss=0.3066, pruned_loss=0.06933, over 8258.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.3008, pruned_loss=0.07204, over 1614865.28 frames. ], batch size: 24, lr: 5.42e-03, grad_scale: 8.0 +2023-02-06 15:28:50,374 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3996, 2.8374, 1.8687, 2.1561, 2.2578, 1.6966, 2.0903, 2.1484], + device='cuda:3'), covar=tensor([0.1849, 0.0397, 0.1197, 0.0792, 0.0734, 0.1520, 0.1245, 0.0971], + device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0234, 0.0322, 0.0297, 0.0297, 0.0328, 0.0343, 0.0310], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 15:28:52,324 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110587.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:28:55,821 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110592.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:29:23,305 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110629.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:29:23,844 INFO [train.py:901] (3/4) Epoch 14, batch 5550, loss[loss=0.1861, simple_loss=0.2628, pruned_loss=0.0547, over 7785.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.3002, pruned_loss=0.07152, over 1616123.27 frames. ], batch size: 19, lr: 5.42e-03, grad_scale: 8.0 +2023-02-06 15:29:28,291 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.70 vs. limit=5.0 +2023-02-06 15:29:34,020 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110645.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:29:35,489 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4173, 2.7224, 1.9736, 2.1236, 2.2878, 1.5854, 2.0282, 1.9689], + device='cuda:3'), covar=tensor([0.1482, 0.0363, 0.1012, 0.0655, 0.0656, 0.1479, 0.1023, 0.0896], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0235, 0.0323, 0.0298, 0.0298, 0.0329, 0.0345, 0.0312], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 15:29:44,466 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.692e+02 2.421e+02 3.120e+02 3.692e+02 1.093e+03, threshold=6.240e+02, percent-clipped=9.0 +2023-02-06 15:29:47,080 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110665.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:29:56,913 INFO [train.py:901] (3/4) Epoch 14, batch 5600, loss[loss=0.2036, simple_loss=0.2941, pruned_loss=0.05657, over 8110.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.3021, pruned_loss=0.0728, over 1616260.51 frames. ], batch size: 23, lr: 5.42e-03, grad_scale: 8.0 +2023-02-06 15:29:56,987 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110680.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:30:30,943 INFO [train.py:901] (3/4) Epoch 14, batch 5650, loss[loss=0.1914, simple_loss=0.2793, pruned_loss=0.05174, over 8359.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3026, pruned_loss=0.07333, over 1619521.04 frames. ], batch size: 24, lr: 5.41e-03, grad_scale: 8.0 +2023-02-06 15:30:33,090 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 15:30:47,222 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110750.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:30:49,874 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110754.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:30:54,025 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110760.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:30:54,498 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.668e+02 2.502e+02 3.092e+02 3.638e+02 5.778e+02, threshold=6.185e+02, percent-clipped=0.0 +2023-02-06 15:31:04,325 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110775.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:31:05,721 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110777.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:31:07,495 INFO [train.py:901] (3/4) Epoch 14, batch 5700, loss[loss=0.2434, simple_loss=0.3135, pruned_loss=0.08667, over 8346.00 frames. ], tot_loss[loss=0.2227, simple_loss=0.3011, pruned_loss=0.07215, over 1619644.68 frames. ], batch size: 25, lr: 5.41e-03, grad_scale: 8.0 +2023-02-06 15:31:07,675 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110780.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:31:17,913 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110795.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:31:22,870 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110802.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:31:40,945 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 15:31:41,631 INFO [train.py:901] (3/4) Epoch 14, batch 5750, loss[loss=0.1754, simple_loss=0.2643, pruned_loss=0.04329, over 8254.00 frames. ], tot_loss[loss=0.2217, simple_loss=0.3002, pruned_loss=0.07163, over 1618302.37 frames. ], batch size: 22, lr: 5.41e-03, grad_scale: 8.0 +2023-02-06 15:31:51,514 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110843.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:31:54,992 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110848.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:32:04,408 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.594e+02 2.447e+02 3.019e+02 3.853e+02 7.521e+02, threshold=6.038e+02, percent-clipped=3.0 +2023-02-06 15:32:10,146 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110868.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:32:14,049 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110873.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:32:18,667 INFO [train.py:901] (3/4) Epoch 14, batch 5800, loss[loss=0.2543, simple_loss=0.3356, pruned_loss=0.08655, over 8550.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2991, pruned_loss=0.07127, over 1614984.61 frames. ], batch size: 31, lr: 5.41e-03, grad_scale: 8.0 +2023-02-06 15:32:22,950 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110886.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:32:50,169 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5574, 2.7754, 1.7900, 2.2524, 2.3325, 1.5706, 2.1448, 2.1442], + device='cuda:3'), covar=tensor([0.1354, 0.0306, 0.1107, 0.0542, 0.0635, 0.1385, 0.0895, 0.0845], + device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0231, 0.0321, 0.0294, 0.0296, 0.0326, 0.0341, 0.0308], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 15:32:53,261 INFO [train.py:901] (3/4) Epoch 14, batch 5850, loss[loss=0.2118, simple_loss=0.3031, pruned_loss=0.06025, over 8516.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.2988, pruned_loss=0.0713, over 1613642.99 frames. ], batch size: 28, lr: 5.41e-03, grad_scale: 8.0 +2023-02-06 15:33:02,551 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 +2023-02-06 15:33:14,136 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.687e+02 2.498e+02 3.098e+02 4.112e+02 1.106e+03, threshold=6.195e+02, percent-clipped=10.0 +2023-02-06 15:33:22,883 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=110973.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:33:26,427 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110976.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:33:28,981 INFO [train.py:901] (3/4) Epoch 14, batch 5900, loss[loss=0.2309, simple_loss=0.3102, pruned_loss=0.07574, over 7681.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.2986, pruned_loss=0.07121, over 1614222.84 frames. ], batch size: 18, lr: 5.41e-03, grad_scale: 8.0 +2023-02-06 15:33:38,586 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5579, 1.8870, 2.0345, 1.0914, 2.1092, 1.3744, 0.5218, 1.7820], + device='cuda:3'), covar=tensor([0.0470, 0.0271, 0.0203, 0.0453, 0.0298, 0.0676, 0.0695, 0.0225], + device='cuda:3'), in_proj_covar=tensor([0.0415, 0.0358, 0.0305, 0.0406, 0.0345, 0.0499, 0.0372, 0.0377], + device='cuda:3'), out_proj_covar=tensor([1.1526e-04, 9.6830e-05, 8.2572e-05, 1.1029e-04, 9.4174e-05, 1.4647e-04, + 1.0367e-04, 1.0325e-04], device='cuda:3') +2023-02-06 15:33:54,412 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111016.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:34:03,738 INFO [train.py:901] (3/4) Epoch 14, batch 5950, loss[loss=0.2018, simple_loss=0.2814, pruned_loss=0.06109, over 7527.00 frames. ], tot_loss[loss=0.221, simple_loss=0.2996, pruned_loss=0.07119, over 1616337.73 frames. ], batch size: 18, lr: 5.41e-03, grad_scale: 8.0 +2023-02-06 15:34:07,780 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111036.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:34:11,073 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111041.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:34:17,723 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111051.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:34:24,233 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 2.648e+02 3.047e+02 4.016e+02 7.772e+02, threshold=6.093e+02, percent-clipped=5.0 +2023-02-06 15:34:24,461 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111061.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:34:34,724 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111076.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:34:38,001 INFO [train.py:901] (3/4) Epoch 14, batch 6000, loss[loss=0.2026, simple_loss=0.2965, pruned_loss=0.05433, over 8228.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2995, pruned_loss=0.07118, over 1611896.23 frames. ], batch size: 49, lr: 5.41e-03, grad_scale: 8.0 +2023-02-06 15:34:38,001 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 15:34:50,549 INFO [train.py:935] (3/4) Epoch 14, validation: loss=0.1818, simple_loss=0.2816, pruned_loss=0.04094, over 944034.00 frames. +2023-02-06 15:34:50,550 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 15:34:51,683 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-06 15:34:56,293 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111088.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:35:03,698 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111098.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:35:11,768 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-06 15:35:27,231 INFO [train.py:901] (3/4) Epoch 14, batch 6050, loss[loss=0.2393, simple_loss=0.3139, pruned_loss=0.08231, over 8668.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.2987, pruned_loss=0.07115, over 1611114.12 frames. ], batch size: 34, lr: 5.40e-03, grad_scale: 4.0 +2023-02-06 15:35:49,432 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.663e+02 2.432e+02 2.876e+02 3.526e+02 5.542e+02, threshold=5.752e+02, percent-clipped=0.0 +2023-02-06 15:35:59,319 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2023-02-06 15:36:01,590 INFO [train.py:901] (3/4) Epoch 14, batch 6100, loss[loss=0.2448, simple_loss=0.3161, pruned_loss=0.08677, over 8657.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.3001, pruned_loss=0.07149, over 1618487.81 frames. ], batch size: 39, lr: 5.40e-03, grad_scale: 4.0 +2023-02-06 15:36:15,954 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 15:36:24,936 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111213.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:36:37,111 INFO [train.py:901] (3/4) Epoch 14, batch 6150, loss[loss=0.1927, simple_loss=0.279, pruned_loss=0.05315, over 8025.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.2995, pruned_loss=0.07134, over 1617321.79 frames. ], batch size: 22, lr: 5.40e-03, grad_scale: 4.0 +2023-02-06 15:36:37,873 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111230.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:36:46,729 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111243.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:36:59,331 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.449e+02 2.474e+02 3.213e+02 4.029e+02 8.079e+02, threshold=6.426e+02, percent-clipped=5.0 +2023-02-06 15:37:11,859 INFO [train.py:901] (3/4) Epoch 14, batch 6200, loss[loss=0.2423, simple_loss=0.3208, pruned_loss=0.08186, over 8548.00 frames. ], tot_loss[loss=0.2229, simple_loss=0.3006, pruned_loss=0.07262, over 1618676.02 frames. ], batch size: 31, lr: 5.40e-03, grad_scale: 4.0 +2023-02-06 15:37:38,515 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111320.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:37:40,663 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8070, 1.7805, 2.3952, 1.8691, 1.2957, 2.4401, 0.4847, 1.4100], + device='cuda:3'), covar=tensor([0.2221, 0.1555, 0.0387, 0.1432, 0.3680, 0.0435, 0.2939, 0.1715], + device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0175, 0.0108, 0.0216, 0.0259, 0.0112, 0.0163, 0.0172], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 15:37:45,102 INFO [train.py:901] (3/4) Epoch 14, batch 6250, loss[loss=0.24, simple_loss=0.3242, pruned_loss=0.07793, over 8286.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.3015, pruned_loss=0.07333, over 1618090.04 frames. ], batch size: 23, lr: 5.40e-03, grad_scale: 4.0 +2023-02-06 15:37:54,001 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 15:37:55,169 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111344.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:37:55,819 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111345.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:38:08,441 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.594e+02 2.286e+02 2.818e+02 3.691e+02 1.208e+03, threshold=5.637e+02, percent-clipped=2.0 +2023-02-06 15:38:13,382 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111369.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:38:20,509 INFO [train.py:901] (3/4) Epoch 14, batch 6300, loss[loss=0.2063, simple_loss=0.2848, pruned_loss=0.06391, over 8035.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.3011, pruned_loss=0.07288, over 1619928.26 frames. ], batch size: 20, lr: 5.40e-03, grad_scale: 4.0 +2023-02-06 15:38:55,237 INFO [train.py:901] (3/4) Epoch 14, batch 6350, loss[loss=0.2047, simple_loss=0.2838, pruned_loss=0.06276, over 7804.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.3016, pruned_loss=0.07327, over 1616367.43 frames. ], batch size: 20, lr: 5.40e-03, grad_scale: 4.0 +2023-02-06 15:38:56,128 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3884, 2.3974, 1.6140, 2.0972, 2.0935, 1.3247, 1.7959, 1.9521], + device='cuda:3'), covar=tensor([0.1519, 0.0398, 0.1186, 0.0625, 0.0615, 0.1575, 0.0979, 0.0913], + device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0232, 0.0323, 0.0293, 0.0298, 0.0326, 0.0341, 0.0307], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 15:38:58,936 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111435.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:39:05,112 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3728, 1.3681, 2.3064, 1.1901, 2.1363, 2.4975, 2.6198, 2.1153], + device='cuda:3'), covar=tensor([0.1045, 0.1301, 0.0492, 0.2062, 0.0740, 0.0409, 0.0702, 0.0755], + device='cuda:3'), in_proj_covar=tensor([0.0275, 0.0307, 0.0269, 0.0299, 0.0285, 0.0245, 0.0371, 0.0295], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 15:39:10,426 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1003, 2.9025, 2.1376, 2.4754, 2.4328, 1.9268, 2.2800, 2.7069], + device='cuda:3'), covar=tensor([0.1277, 0.0358, 0.0928, 0.0668, 0.0638, 0.1190, 0.0982, 0.0798], + device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0232, 0.0323, 0.0294, 0.0298, 0.0326, 0.0341, 0.0307], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 15:39:17,455 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.700e+02 2.340e+02 2.891e+02 3.552e+02 9.934e+02, threshold=5.783e+02, percent-clipped=8.0 +2023-02-06 15:39:23,083 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111469.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:39:30,958 INFO [train.py:901] (3/4) Epoch 14, batch 6400, loss[loss=0.2246, simple_loss=0.3078, pruned_loss=0.07071, over 8644.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.3013, pruned_loss=0.07315, over 1616004.70 frames. ], batch size: 39, lr: 5.40e-03, grad_scale: 8.0 +2023-02-06 15:39:35,210 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111486.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:39:40,668 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111494.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:40:05,255 INFO [train.py:901] (3/4) Epoch 14, batch 6450, loss[loss=0.1915, simple_loss=0.267, pruned_loss=0.05795, over 6777.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.3, pruned_loss=0.07241, over 1612968.63 frames. ], batch size: 15, lr: 5.40e-03, grad_scale: 8.0 +2023-02-06 15:40:26,319 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.574e+02 2.418e+02 3.184e+02 3.807e+02 1.482e+03, threshold=6.367e+02, percent-clipped=8.0 +2023-02-06 15:40:39,077 INFO [train.py:901] (3/4) Epoch 14, batch 6500, loss[loss=0.1917, simple_loss=0.2666, pruned_loss=0.05842, over 7201.00 frames. ], tot_loss[loss=0.2235, simple_loss=0.3007, pruned_loss=0.07313, over 1607905.26 frames. ], batch size: 16, lr: 5.39e-03, grad_scale: 8.0 +2023-02-06 15:40:44,376 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111587.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:40:55,096 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111601.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:41:12,108 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111626.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:41:14,602 INFO [train.py:901] (3/4) Epoch 14, batch 6550, loss[loss=0.1822, simple_loss=0.2598, pruned_loss=0.0523, over 7252.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.3015, pruned_loss=0.07343, over 1609317.74 frames. ], batch size: 16, lr: 5.39e-03, grad_scale: 8.0 +2023-02-06 15:41:24,419 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 15:41:35,852 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.348e+02 2.515e+02 3.055e+02 3.900e+02 7.605e+02, threshold=6.110e+02, percent-clipped=3.0 +2023-02-06 15:41:43,341 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 15:41:47,955 INFO [train.py:901] (3/4) Epoch 14, batch 6600, loss[loss=0.2141, simple_loss=0.2972, pruned_loss=0.06543, over 8322.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.3021, pruned_loss=0.07374, over 1612303.51 frames. ], batch size: 25, lr: 5.39e-03, grad_scale: 8.0 +2023-02-06 15:41:55,651 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111691.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:42:03,625 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111702.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:42:14,398 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111716.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:42:23,468 INFO [train.py:901] (3/4) Epoch 14, batch 6650, loss[loss=0.1982, simple_loss=0.2682, pruned_loss=0.06408, over 7696.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.3014, pruned_loss=0.0731, over 1613396.19 frames. ], batch size: 18, lr: 5.39e-03, grad_scale: 8.0 +2023-02-06 15:42:45,230 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.442e+02 2.391e+02 3.105e+02 3.860e+02 7.189e+02, threshold=6.209e+02, percent-clipped=3.0 +2023-02-06 15:42:48,109 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3387, 1.4785, 1.3992, 1.7758, 0.8025, 1.2000, 1.3355, 1.5258], + device='cuda:3'), covar=tensor([0.0884, 0.0872, 0.1062, 0.0565, 0.1151, 0.1503, 0.0820, 0.0711], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0207, 0.0256, 0.0215, 0.0216, 0.0253, 0.0261, 0.0217], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 15:42:49,649 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 15:42:51,481 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3815, 1.6061, 1.6305, 0.9985, 1.6769, 1.3095, 0.2509, 1.5786], + device='cuda:3'), covar=tensor([0.0334, 0.0266, 0.0249, 0.0388, 0.0330, 0.0705, 0.0638, 0.0184], + device='cuda:3'), in_proj_covar=tensor([0.0409, 0.0354, 0.0302, 0.0405, 0.0339, 0.0493, 0.0369, 0.0376], + device='cuda:3'), out_proj_covar=tensor([1.1356e-04, 9.5802e-05, 8.1746e-05, 1.1006e-04, 9.2653e-05, 1.4440e-04, + 1.0266e-04, 1.0282e-04], device='cuda:3') +2023-02-06 15:42:57,306 INFO [train.py:901] (3/4) Epoch 14, batch 6700, loss[loss=0.2488, simple_loss=0.3292, pruned_loss=0.08419, over 8256.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.3011, pruned_loss=0.07289, over 1614432.67 frames. ], batch size: 24, lr: 5.39e-03, grad_scale: 8.0 +2023-02-06 15:43:16,632 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111809.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 15:43:32,517 INFO [train.py:901] (3/4) Epoch 14, batch 6750, loss[loss=0.2035, simple_loss=0.2809, pruned_loss=0.06308, over 7656.00 frames. ], tot_loss[loss=0.2225, simple_loss=0.3002, pruned_loss=0.07239, over 1609079.74 frames. ], batch size: 19, lr: 5.39e-03, grad_scale: 8.0 +2023-02-06 15:43:32,588 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=111830.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:43:54,029 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.411e+02 2.559e+02 3.020e+02 4.182e+02 1.269e+03, threshold=6.039e+02, percent-clipped=6.0 +2023-02-06 15:44:02,386 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 15:44:02,882 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 15:44:07,193 INFO [train.py:901] (3/4) Epoch 14, batch 6800, loss[loss=0.1931, simple_loss=0.2707, pruned_loss=0.05776, over 7796.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.2994, pruned_loss=0.07189, over 1610711.76 frames. ], batch size: 19, lr: 5.39e-03, grad_scale: 8.0 +2023-02-06 15:44:40,418 INFO [train.py:901] (3/4) Epoch 14, batch 6850, loss[loss=0.2365, simple_loss=0.308, pruned_loss=0.08253, over 7648.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.2997, pruned_loss=0.07205, over 1610259.52 frames. ], batch size: 19, lr: 5.39e-03, grad_scale: 8.0 +2023-02-06 15:44:51,193 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 15:44:52,737 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111945.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:45:01,428 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111958.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:45:03,874 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.737e+02 2.542e+02 3.126e+02 4.226e+02 8.027e+02, threshold=6.251e+02, percent-clipped=7.0 +2023-02-06 15:45:16,603 INFO [train.py:901] (3/4) Epoch 14, batch 6900, loss[loss=0.214, simple_loss=0.2942, pruned_loss=0.06692, over 8135.00 frames. ], tot_loss[loss=0.223, simple_loss=0.301, pruned_loss=0.07247, over 1616005.79 frames. ], batch size: 22, lr: 5.38e-03, grad_scale: 8.0 +2023-02-06 15:45:18,711 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111983.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:45:50,586 INFO [train.py:901] (3/4) Epoch 14, batch 6950, loss[loss=0.2131, simple_loss=0.2905, pruned_loss=0.06783, over 7435.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.3023, pruned_loss=0.07271, over 1619023.33 frames. ], batch size: 17, lr: 5.38e-03, grad_scale: 8.0 +2023-02-06 15:45:58,670 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 15:46:13,925 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.444e+02 2.419e+02 2.987e+02 3.531e+02 6.552e+02, threshold=5.974e+02, percent-clipped=1.0 +2023-02-06 15:46:25,966 INFO [train.py:901] (3/4) Epoch 14, batch 7000, loss[loss=0.1618, simple_loss=0.2536, pruned_loss=0.03499, over 7817.00 frames. ], tot_loss[loss=0.2233, simple_loss=0.3021, pruned_loss=0.07229, over 1618888.41 frames. ], batch size: 20, lr: 5.38e-03, grad_scale: 8.0 +2023-02-06 15:46:28,776 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112084.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:46:28,795 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9601, 1.5486, 3.2727, 1.3706, 2.2101, 3.5448, 3.6978, 3.0141], + device='cuda:3'), covar=tensor([0.1119, 0.1647, 0.0342, 0.2206, 0.1068, 0.0244, 0.0489, 0.0621], + device='cuda:3'), in_proj_covar=tensor([0.0277, 0.0306, 0.0268, 0.0298, 0.0284, 0.0244, 0.0369, 0.0295], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 15:46:41,421 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.7261, 3.7104, 3.4115, 1.8816, 3.2765, 3.4287, 3.3830, 3.2273], + device='cuda:3'), covar=tensor([0.1019, 0.0787, 0.1147, 0.5031, 0.1064, 0.1233, 0.1432, 0.0961], + device='cuda:3'), in_proj_covar=tensor([0.0485, 0.0397, 0.0403, 0.0500, 0.0396, 0.0400, 0.0390, 0.0346], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 15:46:59,913 INFO [train.py:901] (3/4) Epoch 14, batch 7050, loss[loss=0.2676, simple_loss=0.3407, pruned_loss=0.09726, over 8528.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.3024, pruned_loss=0.07262, over 1618736.87 frames. ], batch size: 28, lr: 5.38e-03, grad_scale: 8.0 +2023-02-06 15:47:15,936 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112153.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 15:47:21,801 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.644e+02 2.608e+02 3.153e+02 4.211e+02 1.237e+03, threshold=6.307e+02, percent-clipped=12.0 +2023-02-06 15:47:35,265 INFO [train.py:901] (3/4) Epoch 14, batch 7100, loss[loss=0.1908, simple_loss=0.277, pruned_loss=0.05236, over 8029.00 frames. ], tot_loss[loss=0.2233, simple_loss=0.3021, pruned_loss=0.0722, over 1620084.34 frames. ], batch size: 22, lr: 5.38e-03, grad_scale: 8.0 +2023-02-06 15:47:50,285 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112201.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:47:59,153 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-06 15:48:07,018 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0384, 1.6476, 1.4590, 1.6509, 1.4181, 1.2473, 1.2675, 1.3756], + device='cuda:3'), covar=tensor([0.1017, 0.0406, 0.1084, 0.0423, 0.0593, 0.1321, 0.0836, 0.0635], + device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0234, 0.0323, 0.0298, 0.0301, 0.0329, 0.0345, 0.0309], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 15:48:07,693 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112226.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:48:10,129 INFO [train.py:901] (3/4) Epoch 14, batch 7150, loss[loss=0.258, simple_loss=0.3345, pruned_loss=0.09075, over 8334.00 frames. ], tot_loss[loss=0.2222, simple_loss=0.3008, pruned_loss=0.07181, over 1618059.39 frames. ], batch size: 26, lr: 5.38e-03, grad_scale: 8.0 +2023-02-06 15:48:16,275 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2875, 1.2592, 3.3618, 1.1052, 2.9969, 2.7585, 3.0479, 2.9630], + device='cuda:3'), covar=tensor([0.0635, 0.3751, 0.0817, 0.3534, 0.1304, 0.1138, 0.0704, 0.0802], + device='cuda:3'), in_proj_covar=tensor([0.0538, 0.0599, 0.0625, 0.0566, 0.0638, 0.0550, 0.0542, 0.0603], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 15:48:31,549 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.680e+02 2.371e+02 2.859e+02 3.664e+02 7.587e+02, threshold=5.717e+02, percent-clipped=3.0 +2023-02-06 15:48:35,310 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-06 15:48:35,597 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112268.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 15:48:43,215 INFO [train.py:901] (3/4) Epoch 14, batch 7200, loss[loss=0.2435, simple_loss=0.3218, pruned_loss=0.08255, over 8028.00 frames. ], tot_loss[loss=0.223, simple_loss=0.3016, pruned_loss=0.07219, over 1618293.46 frames. ], batch size: 22, lr: 5.38e-03, grad_scale: 8.0 +2023-02-06 15:48:51,521 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1421, 1.7214, 1.8261, 1.6144, 1.1200, 1.7126, 2.0916, 1.9072], + device='cuda:3'), covar=tensor([0.0442, 0.1070, 0.1432, 0.1197, 0.0522, 0.1253, 0.0535, 0.0506], + device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0150, 0.0187, 0.0153, 0.0099, 0.0159, 0.0113, 0.0135], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 15:49:18,427 INFO [train.py:901] (3/4) Epoch 14, batch 7250, loss[loss=0.1932, simple_loss=0.2683, pruned_loss=0.05904, over 7710.00 frames. ], tot_loss[loss=0.222, simple_loss=0.3005, pruned_loss=0.07171, over 1613255.26 frames. ], batch size: 18, lr: 5.38e-03, grad_scale: 8.0 +2023-02-06 15:49:26,297 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6142, 1.4119, 1.5097, 1.2667, 0.8803, 1.3427, 1.5431, 1.2321], + device='cuda:3'), covar=tensor([0.0495, 0.1207, 0.1618, 0.1373, 0.0534, 0.1430, 0.0666, 0.0653], + device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0151, 0.0188, 0.0154, 0.0100, 0.0160, 0.0114, 0.0135], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0008, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 15:49:38,652 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5893, 4.5557, 4.1628, 1.9395, 4.0906, 4.2370, 4.1752, 4.0238], + device='cuda:3'), covar=tensor([0.0615, 0.0475, 0.0868, 0.4478, 0.0739, 0.0712, 0.1149, 0.0591], + device='cuda:3'), in_proj_covar=tensor([0.0483, 0.0397, 0.0405, 0.0502, 0.0397, 0.0400, 0.0392, 0.0346], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 15:49:39,904 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.532e+02 2.708e+02 3.212e+02 3.989e+02 8.387e+02, threshold=6.424e+02, percent-clipped=5.0 +2023-02-06 15:49:52,050 INFO [train.py:901] (3/4) Epoch 14, batch 7300, loss[loss=0.2281, simple_loss=0.3107, pruned_loss=0.07272, over 8323.00 frames. ], tot_loss[loss=0.223, simple_loss=0.301, pruned_loss=0.07251, over 1613178.28 frames. ], batch size: 25, lr: 5.37e-03, grad_scale: 8.0 +2023-02-06 15:50:11,942 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112407.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:50:26,720 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112428.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:50:28,014 INFO [train.py:901] (3/4) Epoch 14, batch 7350, loss[loss=0.1941, simple_loss=0.2815, pruned_loss=0.05331, over 7658.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.301, pruned_loss=0.07269, over 1613019.17 frames. ], batch size: 19, lr: 5.37e-03, grad_scale: 8.0 +2023-02-06 15:50:40,030 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 15:50:49,906 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.807e+02 2.462e+02 2.972e+02 3.682e+02 1.093e+03, threshold=5.943e+02, percent-clipped=5.0 +2023-02-06 15:50:59,939 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 15:51:02,056 INFO [train.py:901] (3/4) Epoch 14, batch 7400, loss[loss=0.2346, simple_loss=0.2982, pruned_loss=0.08549, over 7426.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.3011, pruned_loss=0.07267, over 1616631.71 frames. ], batch size: 17, lr: 5.37e-03, grad_scale: 8.0 +2023-02-06 15:51:10,094 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 +2023-02-06 15:51:32,627 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112524.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 15:51:37,189 INFO [train.py:901] (3/4) Epoch 14, batch 7450, loss[loss=0.2167, simple_loss=0.3006, pruned_loss=0.06645, over 8338.00 frames. ], tot_loss[loss=0.2231, simple_loss=0.3014, pruned_loss=0.07246, over 1614955.12 frames. ], batch size: 26, lr: 5.37e-03, grad_scale: 8.0 +2023-02-06 15:51:41,773 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 15:51:46,946 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112543.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:51:51,003 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112549.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 15:51:59,932 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.546e+02 2.473e+02 3.112e+02 3.710e+02 6.215e+02, threshold=6.224e+02, percent-clipped=1.0 +2023-02-06 15:52:13,216 INFO [train.py:901] (3/4) Epoch 14, batch 7500, loss[loss=0.1953, simple_loss=0.2868, pruned_loss=0.05187, over 8283.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.3004, pruned_loss=0.07218, over 1611603.90 frames. ], batch size: 23, lr: 5.37e-03, grad_scale: 8.0 +2023-02-06 15:52:13,389 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112580.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:52:23,793 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112595.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:52:47,704 INFO [train.py:901] (3/4) Epoch 14, batch 7550, loss[loss=0.2116, simple_loss=0.2883, pruned_loss=0.06742, over 7967.00 frames. ], tot_loss[loss=0.2222, simple_loss=0.3004, pruned_loss=0.07199, over 1611660.91 frames. ], batch size: 21, lr: 5.37e-03, grad_scale: 8.0 +2023-02-06 15:52:51,318 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112635.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:53:11,227 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.690e+02 2.402e+02 2.890e+02 3.643e+02 7.164e+02, threshold=5.781e+02, percent-clipped=3.0 +2023-02-06 15:53:23,629 INFO [train.py:901] (3/4) Epoch 14, batch 7600, loss[loss=0.1611, simple_loss=0.2398, pruned_loss=0.04119, over 7544.00 frames. ], tot_loss[loss=0.2212, simple_loss=0.2995, pruned_loss=0.07139, over 1613712.51 frames. ], batch size: 18, lr: 5.37e-03, grad_scale: 8.0 +2023-02-06 15:53:40,436 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6497, 1.3335, 1.5820, 1.2452, 0.8778, 1.3308, 1.5954, 1.3270], + device='cuda:3'), covar=tensor([0.0498, 0.1223, 0.1654, 0.1399, 0.0596, 0.1504, 0.0659, 0.0634], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0152, 0.0190, 0.0155, 0.0101, 0.0161, 0.0115, 0.0137], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 15:53:57,111 INFO [train.py:901] (3/4) Epoch 14, batch 7650, loss[loss=0.2502, simple_loss=0.317, pruned_loss=0.09167, over 7539.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.2998, pruned_loss=0.07172, over 1611825.16 frames. ], batch size: 18, lr: 5.37e-03, grad_scale: 8.0 +2023-02-06 15:54:11,117 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112751.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:54:18,468 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.696e+02 2.546e+02 2.941e+02 3.649e+02 7.123e+02, threshold=5.882e+02, percent-clipped=5.0 +2023-02-06 15:54:32,365 INFO [train.py:901] (3/4) Epoch 14, batch 7700, loss[loss=0.2145, simple_loss=0.3033, pruned_loss=0.06284, over 8304.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.3006, pruned_loss=0.0716, over 1615633.85 frames. ], batch size: 25, lr: 5.37e-03, grad_scale: 8.0 +2023-02-06 15:54:44,372 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 15:54:45,579 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112799.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:54:52,889 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 15:55:03,119 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112824.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:55:06,911 INFO [train.py:901] (3/4) Epoch 14, batch 7750, loss[loss=0.1753, simple_loss=0.2648, pruned_loss=0.04286, over 7798.00 frames. ], tot_loss[loss=0.222, simple_loss=0.3008, pruned_loss=0.07158, over 1618362.72 frames. ], batch size: 19, lr: 5.36e-03, grad_scale: 8.0 +2023-02-06 15:55:28,275 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.458e+02 2.453e+02 3.172e+02 4.245e+02 8.131e+02, threshold=6.343e+02, percent-clipped=10.0 +2023-02-06 15:55:30,956 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112866.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:55:41,029 INFO [train.py:901] (3/4) Epoch 14, batch 7800, loss[loss=0.2146, simple_loss=0.2865, pruned_loss=0.07136, over 5962.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.3003, pruned_loss=0.07147, over 1613811.99 frames. ], batch size: 13, lr: 5.36e-03, grad_scale: 8.0 +2023-02-06 15:56:12,171 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112924.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:56:16,242 INFO [train.py:901] (3/4) Epoch 14, batch 7850, loss[loss=0.2113, simple_loss=0.3042, pruned_loss=0.05914, over 8105.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.3007, pruned_loss=0.07159, over 1615730.69 frames. ], batch size: 23, lr: 5.36e-03, grad_scale: 8.0 +2023-02-06 15:56:22,361 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112939.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:56:22,438 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112939.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:56:29,118 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7186, 1.9662, 2.2491, 1.5717, 2.2693, 1.4986, 0.9113, 1.9008], + device='cuda:3'), covar=tensor([0.0457, 0.0259, 0.0181, 0.0369, 0.0278, 0.0662, 0.0601, 0.0232], + device='cuda:3'), in_proj_covar=tensor([0.0415, 0.0355, 0.0302, 0.0407, 0.0342, 0.0495, 0.0368, 0.0380], + device='cuda:3'), out_proj_covar=tensor([1.1508e-04, 9.5920e-05, 8.1591e-05, 1.1046e-04, 9.3161e-05, 1.4483e-04, + 1.0230e-04, 1.0379e-04], device='cuda:3') +2023-02-06 15:56:37,783 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.757e+02 2.504e+02 3.067e+02 3.726e+02 7.698e+02, threshold=6.135e+02, percent-clipped=2.0 +2023-02-06 15:56:49,070 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=112979.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:56:49,646 INFO [train.py:901] (3/4) Epoch 14, batch 7900, loss[loss=0.2142, simple_loss=0.3034, pruned_loss=0.06249, over 8322.00 frames. ], tot_loss[loss=0.223, simple_loss=0.3014, pruned_loss=0.0723, over 1618686.10 frames. ], batch size: 25, lr: 5.36e-03, grad_scale: 8.0 +2023-02-06 15:57:22,235 INFO [train.py:901] (3/4) Epoch 14, batch 7950, loss[loss=0.2696, simple_loss=0.3409, pruned_loss=0.09918, over 8689.00 frames. ], tot_loss[loss=0.222, simple_loss=0.3004, pruned_loss=0.07179, over 1619926.94 frames. ], batch size: 34, lr: 5.36e-03, grad_scale: 8.0 +2023-02-06 15:57:28,322 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113039.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:57:38,049 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113054.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:57:43,248 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.840e+02 2.654e+02 3.191e+02 4.041e+02 1.304e+03, threshold=6.382e+02, percent-clipped=5.0 +2023-02-06 15:57:55,258 INFO [train.py:901] (3/4) Epoch 14, batch 8000, loss[loss=0.2463, simple_loss=0.3163, pruned_loss=0.08813, over 7109.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.3014, pruned_loss=0.07252, over 1620114.00 frames. ], batch size: 71, lr: 5.36e-03, grad_scale: 8.0 +2023-02-06 15:58:04,704 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113094.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:58:23,492 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113122.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:58:28,568 INFO [train.py:901] (3/4) Epoch 14, batch 8050, loss[loss=0.2123, simple_loss=0.2834, pruned_loss=0.07058, over 7191.00 frames. ], tot_loss[loss=0.2227, simple_loss=0.3001, pruned_loss=0.07267, over 1599384.71 frames. ], batch size: 16, lr: 5.36e-03, grad_scale: 16.0 +2023-02-06 15:58:40,131 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113147.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 15:58:49,778 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.566e+02 2.323e+02 2.856e+02 3.288e+02 8.076e+02, threshold=5.712e+02, percent-clipped=1.0 +2023-02-06 15:59:01,701 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 15:59:06,257 INFO [train.py:901] (3/4) Epoch 15, batch 0, loss[loss=0.2239, simple_loss=0.3013, pruned_loss=0.07324, over 7653.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.3013, pruned_loss=0.07324, over 7653.00 frames. ], batch size: 19, lr: 5.17e-03, grad_scale: 16.0 +2023-02-06 15:59:06,258 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 15:59:17,268 INFO [train.py:935] (3/4) Epoch 15, validation: loss=0.1825, simple_loss=0.283, pruned_loss=0.04098, over 944034.00 frames. +2023-02-06 15:59:17,269 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 15:59:32,293 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 15:59:48,307 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.49 vs. limit=2.0 +2023-02-06 15:59:49,088 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-06 15:59:51,454 INFO [train.py:901] (3/4) Epoch 15, batch 50, loss[loss=0.2187, simple_loss=0.2839, pruned_loss=0.0768, over 7798.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.3091, pruned_loss=0.07564, over 366937.13 frames. ], batch size: 19, lr: 5.17e-03, grad_scale: 16.0 +2023-02-06 16:00:08,694 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 16:00:21,179 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113252.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:00:27,788 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.520e+02 2.549e+02 3.077e+02 3.582e+02 9.445e+02, threshold=6.153e+02, percent-clipped=5.0 +2023-02-06 16:00:28,490 INFO [train.py:901] (3/4) Epoch 15, batch 100, loss[loss=0.2551, simple_loss=0.3256, pruned_loss=0.09233, over 8648.00 frames. ], tot_loss[loss=0.2277, simple_loss=0.3065, pruned_loss=0.07443, over 643929.27 frames. ], batch size: 34, lr: 5.17e-03, grad_scale: 16.0 +2023-02-06 16:00:29,906 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 16:00:39,309 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1489, 4.1223, 3.7704, 1.8390, 3.6618, 3.7192, 3.6675, 3.5576], + device='cuda:3'), covar=tensor([0.0998, 0.0688, 0.1259, 0.5006, 0.1108, 0.1163, 0.1536, 0.0997], + device='cuda:3'), in_proj_covar=tensor([0.0484, 0.0404, 0.0403, 0.0503, 0.0402, 0.0400, 0.0393, 0.0352], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 16:00:41,981 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=113283.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:00:50,244 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113295.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:01:00,331 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113310.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:01:02,087 INFO [train.py:901] (3/4) Epoch 15, batch 150, loss[loss=0.2041, simple_loss=0.2785, pruned_loss=0.06488, over 7436.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.3026, pruned_loss=0.07261, over 861101.22 frames. ], batch size: 17, lr: 5.17e-03, grad_scale: 16.0 +2023-02-06 16:01:06,867 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113320.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:01:17,345 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113335.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:01:28,771 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113350.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:01:32,821 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6673, 1.9836, 2.1819, 1.2116, 2.2587, 1.4563, 0.6638, 1.8959], + device='cuda:3'), covar=tensor([0.0499, 0.0269, 0.0207, 0.0457, 0.0282, 0.0666, 0.0617, 0.0239], + device='cuda:3'), in_proj_covar=tensor([0.0416, 0.0358, 0.0304, 0.0408, 0.0343, 0.0497, 0.0370, 0.0380], + device='cuda:3'), out_proj_covar=tensor([1.1544e-04, 9.6722e-05, 8.2157e-05, 1.1080e-04, 9.3424e-05, 1.4533e-04, + 1.0253e-04, 1.0406e-04], device='cuda:3') +2023-02-06 16:01:37,328 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.393e+02 2.511e+02 3.032e+02 4.146e+02 1.005e+03, threshold=6.064e+02, percent-clipped=3.0 +2023-02-06 16:01:38,025 INFO [train.py:901] (3/4) Epoch 15, batch 200, loss[loss=0.2244, simple_loss=0.3133, pruned_loss=0.06776, over 8485.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.3028, pruned_loss=0.07349, over 1030001.13 frames. ], batch size: 26, lr: 5.17e-03, grad_scale: 16.0 +2023-02-06 16:01:46,251 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113375.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:02:01,351 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113398.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:02:11,081 INFO [train.py:901] (3/4) Epoch 15, batch 250, loss[loss=0.1693, simple_loss=0.2551, pruned_loss=0.0418, over 7451.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.3045, pruned_loss=0.07463, over 1157373.24 frames. ], batch size: 17, lr: 5.17e-03, grad_scale: 16.0 +2023-02-06 16:02:19,373 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 16:02:28,604 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 16:02:43,772 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.727e+02 2.666e+02 3.062e+02 4.026e+02 8.735e+02, threshold=6.124e+02, percent-clipped=4.0 +2023-02-06 16:02:44,413 INFO [train.py:901] (3/4) Epoch 15, batch 300, loss[loss=0.2359, simple_loss=0.3136, pruned_loss=0.07911, over 8022.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3049, pruned_loss=0.07507, over 1261826.14 frames. ], batch size: 22, lr: 5.17e-03, grad_scale: 16.0 +2023-02-06 16:03:15,226 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2023-02-06 16:03:19,322 INFO [train.py:901] (3/4) Epoch 15, batch 350, loss[loss=0.201, simple_loss=0.2877, pruned_loss=0.05716, over 8245.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.3023, pruned_loss=0.07338, over 1339681.15 frames. ], batch size: 24, lr: 5.17e-03, grad_scale: 16.0 +2023-02-06 16:03:52,045 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.590e+02 2.415e+02 3.115e+02 3.728e+02 6.919e+02, threshold=6.229e+02, percent-clipped=2.0 +2023-02-06 16:03:52,739 INFO [train.py:901] (3/4) Epoch 15, batch 400, loss[loss=0.2325, simple_loss=0.3079, pruned_loss=0.07854, over 7804.00 frames. ], tot_loss[loss=0.2214, simple_loss=0.2996, pruned_loss=0.07153, over 1398639.05 frames. ], batch size: 20, lr: 5.16e-03, grad_scale: 16.0 +2023-02-06 16:04:17,492 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=113596.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:04:28,817 INFO [train.py:901] (3/4) Epoch 15, batch 450, loss[loss=0.2063, simple_loss=0.2853, pruned_loss=0.06365, over 7976.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2997, pruned_loss=0.0709, over 1451634.76 frames. ], batch size: 21, lr: 5.16e-03, grad_scale: 16.0 +2023-02-06 16:04:30,185 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7327, 1.3363, 1.5570, 1.2376, 0.7854, 1.3351, 1.4750, 1.4432], + device='cuda:3'), covar=tensor([0.0558, 0.1296, 0.1717, 0.1454, 0.0640, 0.1518, 0.0733, 0.0631], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0189, 0.0155, 0.0102, 0.0162, 0.0115, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 16:04:43,286 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113635.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:04:56,118 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113654.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:05:01,032 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.675e+02 2.616e+02 3.268e+02 4.141e+02 9.119e+02, threshold=6.536e+02, percent-clipped=2.0 +2023-02-06 16:05:01,750 INFO [train.py:901] (3/4) Epoch 15, batch 500, loss[loss=0.2409, simple_loss=0.3182, pruned_loss=0.08182, over 6685.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.301, pruned_loss=0.07188, over 1487024.32 frames. ], batch size: 72, lr: 5.16e-03, grad_scale: 16.0 +2023-02-06 16:05:02,560 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113664.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:05:12,319 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113679.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:05:35,427 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113711.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:05:36,635 INFO [train.py:901] (3/4) Epoch 15, batch 550, loss[loss=0.2547, simple_loss=0.3299, pruned_loss=0.08977, over 8388.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.302, pruned_loss=0.07291, over 1516906.70 frames. ], batch size: 49, lr: 5.16e-03, grad_scale: 16.0 +2023-02-06 16:06:09,825 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.703e+02 2.516e+02 3.119e+02 4.209e+02 9.524e+02, threshold=6.239e+02, percent-clipped=4.0 +2023-02-06 16:06:10,537 INFO [train.py:901] (3/4) Epoch 15, batch 600, loss[loss=0.2092, simple_loss=0.3011, pruned_loss=0.05867, over 8246.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.3022, pruned_loss=0.07306, over 1542550.81 frames. ], batch size: 24, lr: 5.16e-03, grad_scale: 16.0 +2023-02-06 16:06:17,854 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1596, 1.7261, 4.1750, 1.8828, 2.6157, 4.6166, 4.6122, 4.0830], + device='cuda:3'), covar=tensor([0.1065, 0.1578, 0.0234, 0.1837, 0.0976, 0.0180, 0.0432, 0.0469], + device='cuda:3'), in_proj_covar=tensor([0.0278, 0.0308, 0.0273, 0.0301, 0.0288, 0.0247, 0.0377, 0.0299], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 16:06:24,219 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 16:06:44,320 INFO [train.py:901] (3/4) Epoch 15, batch 650, loss[loss=0.2036, simple_loss=0.2778, pruned_loss=0.06473, over 7813.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3027, pruned_loss=0.07325, over 1561481.66 frames. ], batch size: 20, lr: 5.16e-03, grad_scale: 16.0 +2023-02-06 16:06:53,837 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3080, 1.4280, 1.3680, 1.8076, 0.6500, 1.1637, 1.2675, 1.4219], + device='cuda:3'), covar=tensor([0.0892, 0.0799, 0.1113, 0.0558, 0.1180, 0.1462, 0.0808, 0.0732], + device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0201, 0.0247, 0.0207, 0.0208, 0.0244, 0.0249, 0.0208], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 16:07:19,220 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.468e+02 2.270e+02 2.767e+02 3.649e+02 9.673e+02, threshold=5.535e+02, percent-clipped=4.0 +2023-02-06 16:07:19,888 INFO [train.py:901] (3/4) Epoch 15, batch 700, loss[loss=0.2454, simple_loss=0.3319, pruned_loss=0.07948, over 8520.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3027, pruned_loss=0.07331, over 1573998.24 frames. ], batch size: 28, lr: 5.16e-03, grad_scale: 16.0 +2023-02-06 16:07:53,445 INFO [train.py:901] (3/4) Epoch 15, batch 750, loss[loss=0.2053, simple_loss=0.2952, pruned_loss=0.05775, over 8471.00 frames. ], tot_loss[loss=0.2227, simple_loss=0.3013, pruned_loss=0.07204, over 1581497.59 frames. ], batch size: 25, lr: 5.16e-03, grad_scale: 16.0 +2023-02-06 16:08:11,157 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 16:08:14,525 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.8059, 6.0211, 5.0989, 2.6121, 5.2075, 5.6455, 5.5510, 5.1759], + device='cuda:3'), covar=tensor([0.0637, 0.0397, 0.0999, 0.4168, 0.0711, 0.0706, 0.1074, 0.0564], + device='cuda:3'), in_proj_covar=tensor([0.0477, 0.0399, 0.0398, 0.0497, 0.0396, 0.0396, 0.0383, 0.0346], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 16:08:20,441 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 16:08:29,183 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.267e+02 2.237e+02 2.791e+02 3.511e+02 6.350e+02, threshold=5.582e+02, percent-clipped=4.0 +2023-02-06 16:08:29,880 INFO [train.py:901] (3/4) Epoch 15, batch 800, loss[loss=0.2013, simple_loss=0.2758, pruned_loss=0.06334, over 7808.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2996, pruned_loss=0.07108, over 1586189.97 frames. ], batch size: 20, lr: 5.16e-03, grad_scale: 16.0 +2023-02-06 16:08:32,849 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113967.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:08:40,790 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=113979.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:08:45,869 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8783, 1.6255, 2.1374, 1.8035, 1.9738, 1.9022, 1.5874, 0.7036], + device='cuda:3'), covar=tensor([0.4479, 0.3877, 0.1480, 0.2777, 0.1966, 0.2581, 0.1812, 0.4306], + device='cuda:3'), in_proj_covar=tensor([0.0894, 0.0906, 0.0746, 0.0876, 0.0943, 0.0833, 0.0715, 0.0786], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 16:08:50,096 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113992.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:09:01,983 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=114008.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:09:05,302 INFO [train.py:901] (3/4) Epoch 15, batch 850, loss[loss=0.1906, simple_loss=0.2702, pruned_loss=0.05549, over 7698.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2999, pruned_loss=0.07097, over 1596305.92 frames. ], batch size: 18, lr: 5.15e-03, grad_scale: 16.0 +2023-02-06 16:09:17,861 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-06 16:09:39,413 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.618e+02 2.394e+02 2.826e+02 3.443e+02 6.296e+02, threshold=5.653e+02, percent-clipped=1.0 +2023-02-06 16:09:40,794 INFO [train.py:901] (3/4) Epoch 15, batch 900, loss[loss=0.2713, simple_loss=0.335, pruned_loss=0.1039, over 6229.00 frames. ], tot_loss[loss=0.2221, simple_loss=0.3005, pruned_loss=0.07187, over 1591161.08 frames. ], batch size: 72, lr: 5.15e-03, grad_scale: 16.0 +2023-02-06 16:10:02,632 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114094.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:10:15,168 INFO [train.py:901] (3/4) Epoch 15, batch 950, loss[loss=0.2282, simple_loss=0.3092, pruned_loss=0.07355, over 8609.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.3001, pruned_loss=0.07191, over 1596263.70 frames. ], batch size: 39, lr: 5.15e-03, grad_scale: 16.0 +2023-02-06 16:10:20,921 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2023-02-06 16:10:21,911 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114123.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:10:36,823 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114145.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:10:37,014 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-06 16:10:39,440 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 16:10:49,208 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.482e+02 2.449e+02 2.913e+02 3.851e+02 8.356e+02, threshold=5.826e+02, percent-clipped=3.0 +2023-02-06 16:10:49,926 INFO [train.py:901] (3/4) Epoch 15, batch 1000, loss[loss=0.2394, simple_loss=0.3294, pruned_loss=0.07473, over 8598.00 frames. ], tot_loss[loss=0.2214, simple_loss=0.3002, pruned_loss=0.0713, over 1604212.39 frames. ], batch size: 31, lr: 5.15e-03, grad_scale: 16.0 +2023-02-06 16:11:14,211 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 16:11:20,816 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-06 16:11:25,581 INFO [train.py:901] (3/4) Epoch 15, batch 1050, loss[loss=0.2546, simple_loss=0.3324, pruned_loss=0.08836, over 8326.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.3007, pruned_loss=0.07206, over 1606216.54 frames. ], batch size: 25, lr: 5.15e-03, grad_scale: 16.0 +2023-02-06 16:11:25,604 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 16:11:28,873 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0865, 1.1991, 4.1531, 1.8322, 2.3344, 4.7275, 4.7916, 3.9983], + device='cuda:3'), covar=tensor([0.1152, 0.2079, 0.0295, 0.2049, 0.1333, 0.0191, 0.0407, 0.0599], + device='cuda:3'), in_proj_covar=tensor([0.0277, 0.0306, 0.0270, 0.0299, 0.0285, 0.0246, 0.0372, 0.0297], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 16:11:39,884 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6878, 1.3607, 1.4972, 1.2663, 0.7885, 1.2846, 1.4729, 1.3093], + device='cuda:3'), covar=tensor([0.0532, 0.1282, 0.1724, 0.1382, 0.0634, 0.1515, 0.0732, 0.0628], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0190, 0.0156, 0.0102, 0.0161, 0.0115, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 16:11:57,601 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.714e+02 2.504e+02 3.058e+02 3.938e+02 1.189e+03, threshold=6.116e+02, percent-clipped=4.0 +2023-02-06 16:11:58,320 INFO [train.py:901] (3/4) Epoch 15, batch 1100, loss[loss=0.2102, simple_loss=0.2966, pruned_loss=0.06189, over 8603.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.3018, pruned_loss=0.07298, over 1610596.37 frames. ], batch size: 31, lr: 5.15e-03, grad_scale: 16.0 +2023-02-06 16:12:11,298 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-06 16:12:26,182 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 +2023-02-06 16:12:33,000 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 +2023-02-06 16:12:33,901 INFO [train.py:901] (3/4) Epoch 15, batch 1150, loss[loss=0.2077, simple_loss=0.2872, pruned_loss=0.06408, over 7812.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.2999, pruned_loss=0.07134, over 1613223.00 frames. ], batch size: 20, lr: 5.15e-03, grad_scale: 16.0 +2023-02-06 16:12:38,620 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 16:12:59,531 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114350.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:13:07,357 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.318e+02 2.463e+02 3.139e+02 3.955e+02 6.139e+02, threshold=6.277e+02, percent-clipped=1.0 +2023-02-06 16:13:07,982 INFO [train.py:901] (3/4) Epoch 15, batch 1200, loss[loss=0.3208, simple_loss=0.3721, pruned_loss=0.1347, over 6937.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.2991, pruned_loss=0.0709, over 1612044.99 frames. ], batch size: 73, lr: 5.15e-03, grad_scale: 16.0 +2023-02-06 16:13:16,147 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114375.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:13:18,729 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114379.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:13:36,382 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114404.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:13:42,789 INFO [train.py:901] (3/4) Epoch 15, batch 1250, loss[loss=0.2164, simple_loss=0.2826, pruned_loss=0.07513, over 8246.00 frames. ], tot_loss[loss=0.222, simple_loss=0.3001, pruned_loss=0.07196, over 1609625.58 frames. ], batch size: 22, lr: 5.15e-03, grad_scale: 16.0 +2023-02-06 16:14:16,856 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.611e+02 2.591e+02 3.148e+02 4.129e+02 1.085e+03, threshold=6.295e+02, percent-clipped=6.0 +2023-02-06 16:14:17,473 INFO [train.py:901] (3/4) Epoch 15, batch 1300, loss[loss=0.268, simple_loss=0.3309, pruned_loss=0.1025, over 8591.00 frames. ], tot_loss[loss=0.221, simple_loss=0.299, pruned_loss=0.07147, over 1607283.79 frames. ], batch size: 34, lr: 5.14e-03, grad_scale: 16.0 +2023-02-06 16:14:35,225 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=114489.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:14:51,247 INFO [train.py:901] (3/4) Epoch 15, batch 1350, loss[loss=0.2316, simple_loss=0.3001, pruned_loss=0.08159, over 8609.00 frames. ], tot_loss[loss=0.2217, simple_loss=0.2997, pruned_loss=0.07182, over 1611434.71 frames. ], batch size: 34, lr: 5.14e-03, grad_scale: 16.0 +2023-02-06 16:15:08,699 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.0933, 4.1167, 3.7384, 1.8032, 3.6220, 3.7600, 3.7862, 3.5045], + device='cuda:3'), covar=tensor([0.0872, 0.0597, 0.1062, 0.5033, 0.0933, 0.0940, 0.1294, 0.0863], + device='cuda:3'), in_proj_covar=tensor([0.0481, 0.0398, 0.0399, 0.0498, 0.0395, 0.0398, 0.0381, 0.0348], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 16:15:12,372 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 +2023-02-06 16:15:26,462 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.625e+02 2.434e+02 2.903e+02 3.628e+02 5.826e+02, threshold=5.807e+02, percent-clipped=0.0 +2023-02-06 16:15:27,129 INFO [train.py:901] (3/4) Epoch 15, batch 1400, loss[loss=0.2396, simple_loss=0.3192, pruned_loss=0.07998, over 8472.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2989, pruned_loss=0.07139, over 1609640.01 frames. ], batch size: 27, lr: 5.14e-03, grad_scale: 16.0 +2023-02-06 16:15:54,550 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114604.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:16:00,693 INFO [train.py:901] (3/4) Epoch 15, batch 1450, loss[loss=0.2086, simple_loss=0.2977, pruned_loss=0.05975, over 8519.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2992, pruned_loss=0.07115, over 1612665.70 frames. ], batch size: 28, lr: 5.14e-03, grad_scale: 16.0 +2023-02-06 16:16:08,838 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 16:16:36,177 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.534e+02 2.414e+02 3.068e+02 3.744e+02 6.619e+02, threshold=6.136e+02, percent-clipped=3.0 +2023-02-06 16:16:36,891 INFO [train.py:901] (3/4) Epoch 15, batch 1500, loss[loss=0.1618, simple_loss=0.2456, pruned_loss=0.03897, over 7799.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.2997, pruned_loss=0.0714, over 1615042.65 frames. ], batch size: 19, lr: 5.14e-03, grad_scale: 16.0 +2023-02-06 16:16:48,664 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6716, 1.3285, 4.7826, 1.7290, 4.2219, 3.9756, 4.2854, 4.1885], + device='cuda:3'), covar=tensor([0.0548, 0.5103, 0.0462, 0.4144, 0.1126, 0.0851, 0.0624, 0.0562], + device='cuda:3'), in_proj_covar=tensor([0.0549, 0.0609, 0.0634, 0.0576, 0.0651, 0.0554, 0.0548, 0.0614], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 16:16:58,918 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4760, 1.9983, 3.3402, 1.3954, 2.4363, 2.0319, 1.6619, 2.3441], + device='cuda:3'), covar=tensor([0.1800, 0.2301, 0.0639, 0.3978, 0.1506, 0.2858, 0.1971, 0.2029], + device='cuda:3'), in_proj_covar=tensor([0.0498, 0.0551, 0.0537, 0.0601, 0.0622, 0.0565, 0.0493, 0.0621], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 16:17:00,198 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2336, 2.5171, 3.0243, 1.5662, 3.1149, 1.7234, 1.5786, 1.9989], + device='cuda:3'), covar=tensor([0.0638, 0.0317, 0.0224, 0.0542, 0.0339, 0.0665, 0.0720, 0.0440], + device='cuda:3'), in_proj_covar=tensor([0.0415, 0.0354, 0.0302, 0.0406, 0.0338, 0.0493, 0.0366, 0.0378], + device='cuda:3'), out_proj_covar=tensor([1.1498e-04, 9.5558e-05, 8.1249e-05, 1.1026e-04, 9.1970e-05, 1.4356e-04, + 1.0150e-04, 1.0324e-04], device='cuda:3') +2023-02-06 16:17:09,890 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 +2023-02-06 16:17:11,513 INFO [train.py:901] (3/4) Epoch 15, batch 1550, loss[loss=0.1705, simple_loss=0.2634, pruned_loss=0.03884, over 7654.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2981, pruned_loss=0.07097, over 1609582.21 frames. ], batch size: 19, lr: 5.14e-03, grad_scale: 16.0 +2023-02-06 16:17:26,182 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114734.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:17:45,709 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.542e+02 2.278e+02 2.828e+02 3.736e+02 6.971e+02, threshold=5.655e+02, percent-clipped=1.0 +2023-02-06 16:17:46,443 INFO [train.py:901] (3/4) Epoch 15, batch 1600, loss[loss=0.2205, simple_loss=0.3055, pruned_loss=0.06778, over 7971.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.2987, pruned_loss=0.0707, over 1610934.67 frames. ], batch size: 21, lr: 5.14e-03, grad_scale: 16.0 +2023-02-06 16:18:22,446 INFO [train.py:901] (3/4) Epoch 15, batch 1650, loss[loss=0.1925, simple_loss=0.2698, pruned_loss=0.0576, over 7700.00 frames. ], tot_loss[loss=0.2222, simple_loss=0.3005, pruned_loss=0.07191, over 1614095.45 frames. ], batch size: 18, lr: 5.14e-03, grad_scale: 16.0 +2023-02-06 16:18:55,126 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114860.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:18:56,281 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.261e+02 2.429e+02 2.845e+02 3.384e+02 6.803e+02, threshold=5.691e+02, percent-clipped=1.0 +2023-02-06 16:18:56,970 INFO [train.py:901] (3/4) Epoch 15, batch 1700, loss[loss=0.231, simple_loss=0.3176, pruned_loss=0.07222, over 8481.00 frames. ], tot_loss[loss=0.2212, simple_loss=0.3, pruned_loss=0.07115, over 1616157.81 frames. ], batch size: 29, lr: 5.14e-03, grad_scale: 16.0 +2023-02-06 16:19:12,812 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114885.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:19:16,112 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114889.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:19:32,915 INFO [train.py:901] (3/4) Epoch 15, batch 1750, loss[loss=0.34, simple_loss=0.3881, pruned_loss=0.1459, over 6884.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.299, pruned_loss=0.07096, over 1609615.92 frames. ], batch size: 71, lr: 5.13e-03, grad_scale: 16.0 +2023-02-06 16:19:45,243 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.7897, 2.2079, 4.9147, 2.7232, 4.4538, 4.2373, 4.5664, 4.4661], + device='cuda:3'), covar=tensor([0.0508, 0.3758, 0.0459, 0.3028, 0.0898, 0.0726, 0.0511, 0.0467], + device='cuda:3'), in_proj_covar=tensor([0.0553, 0.0611, 0.0639, 0.0581, 0.0653, 0.0560, 0.0553, 0.0618], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 16:20:06,960 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.463e+02 2.435e+02 3.025e+02 3.758e+02 7.531e+02, threshold=6.050e+02, percent-clipped=3.0 +2023-02-06 16:20:07,576 INFO [train.py:901] (3/4) Epoch 15, batch 1800, loss[loss=0.213, simple_loss=0.2797, pruned_loss=0.07315, over 7531.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.2977, pruned_loss=0.07022, over 1609548.06 frames. ], batch size: 18, lr: 5.13e-03, grad_scale: 16.0 +2023-02-06 16:20:43,790 INFO [train.py:901] (3/4) Epoch 15, batch 1850, loss[loss=0.26, simple_loss=0.3252, pruned_loss=0.09739, over 6880.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2988, pruned_loss=0.07152, over 1605210.00 frames. ], batch size: 71, lr: 5.13e-03, grad_scale: 16.0 +2023-02-06 16:21:17,801 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.743e+02 2.659e+02 3.189e+02 4.139e+02 1.250e+03, threshold=6.379e+02, percent-clipped=4.0 +2023-02-06 16:21:18,506 INFO [train.py:901] (3/4) Epoch 15, batch 1900, loss[loss=0.2337, simple_loss=0.3231, pruned_loss=0.07209, over 8288.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2972, pruned_loss=0.07033, over 1601203.31 frames. ], batch size: 49, lr: 5.13e-03, grad_scale: 16.0 +2023-02-06 16:21:28,840 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=115078.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:21:46,755 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115104.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:21:50,085 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 16:21:53,561 INFO [train.py:901] (3/4) Epoch 15, batch 1950, loss[loss=0.1991, simple_loss=0.2829, pruned_loss=0.05765, over 8029.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2972, pruned_loss=0.06984, over 1606899.50 frames. ], batch size: 22, lr: 5.13e-03, grad_scale: 32.0 +2023-02-06 16:22:04,500 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 16:22:11,847 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7177, 1.7371, 3.2811, 1.3457, 2.2960, 3.6407, 3.9731, 2.6709], + device='cuda:3'), covar=tensor([0.1649, 0.1965, 0.0482, 0.2800, 0.1260, 0.0389, 0.0548, 0.1185], + device='cuda:3'), in_proj_covar=tensor([0.0277, 0.0305, 0.0270, 0.0297, 0.0285, 0.0245, 0.0373, 0.0297], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 16:22:23,207 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 16:22:28,430 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.368e+02 2.421e+02 3.112e+02 3.916e+02 6.433e+02, threshold=6.224e+02, percent-clipped=1.0 +2023-02-06 16:22:29,136 INFO [train.py:901] (3/4) Epoch 15, batch 2000, loss[loss=0.182, simple_loss=0.2636, pruned_loss=0.0502, over 7640.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.297, pruned_loss=0.06993, over 1602079.93 frames. ], batch size: 19, lr: 5.13e-03, grad_scale: 32.0 +2023-02-06 16:22:47,905 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 +2023-02-06 16:22:49,770 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115193.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:22:54,580 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1762, 2.1900, 1.5359, 1.7287, 1.7879, 1.3323, 1.5974, 1.5655], + device='cuda:3'), covar=tensor([0.1336, 0.0343, 0.1193, 0.0610, 0.0693, 0.1457, 0.0911, 0.0864], + device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0229, 0.0324, 0.0302, 0.0300, 0.0330, 0.0345, 0.0308], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 16:22:58,872 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9914, 2.0459, 1.9592, 2.6205, 1.1141, 1.5996, 1.8721, 2.0883], + device='cuda:3'), covar=tensor([0.0688, 0.0875, 0.0860, 0.0397, 0.1188, 0.1326, 0.0827, 0.0791], + device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0203, 0.0248, 0.0211, 0.0211, 0.0247, 0.0253, 0.0212], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 16:23:03,505 INFO [train.py:901] (3/4) Epoch 15, batch 2050, loss[loss=0.205, simple_loss=0.2901, pruned_loss=0.05989, over 8610.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.2977, pruned_loss=0.07019, over 1607066.52 frames. ], batch size: 31, lr: 5.13e-03, grad_scale: 16.0 +2023-02-06 16:23:18,038 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=115233.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:23:36,045 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 +2023-02-06 16:23:39,485 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.528e+02 2.382e+02 2.963e+02 3.753e+02 6.860e+02, threshold=5.925e+02, percent-clipped=2.0 +2023-02-06 16:23:39,506 INFO [train.py:901] (3/4) Epoch 15, batch 2100, loss[loss=0.2468, simple_loss=0.3381, pruned_loss=0.07772, over 8510.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.2986, pruned_loss=0.0706, over 1607639.49 frames. ], batch size: 26, lr: 5.13e-03, grad_scale: 16.0 +2023-02-06 16:24:05,868 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3731, 1.4976, 4.6142, 1.7709, 4.0869, 3.8698, 4.1212, 3.9783], + device='cuda:3'), covar=tensor([0.0549, 0.3986, 0.0408, 0.3174, 0.0994, 0.0813, 0.0500, 0.0614], + device='cuda:3'), in_proj_covar=tensor([0.0543, 0.0598, 0.0628, 0.0568, 0.0643, 0.0552, 0.0542, 0.0609], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 16:24:13,868 INFO [train.py:901] (3/4) Epoch 15, batch 2150, loss[loss=0.2091, simple_loss=0.298, pruned_loss=0.06008, over 8377.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.2988, pruned_loss=0.07072, over 1612090.96 frames. ], batch size: 49, lr: 5.13e-03, grad_scale: 16.0 +2023-02-06 16:24:29,064 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0236, 1.2165, 1.2125, 0.6937, 1.1871, 0.9839, 0.0852, 1.2032], + device='cuda:3'), covar=tensor([0.0381, 0.0292, 0.0271, 0.0432, 0.0356, 0.0845, 0.0700, 0.0260], + device='cuda:3'), in_proj_covar=tensor([0.0427, 0.0365, 0.0312, 0.0420, 0.0349, 0.0507, 0.0380, 0.0392], + device='cuda:3'), out_proj_covar=tensor([1.1842e-04, 9.8345e-05, 8.3735e-05, 1.1414e-04, 9.4909e-05, 1.4778e-04, + 1.0527e-04, 1.0719e-04], device='cuda:3') +2023-02-06 16:24:37,887 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115348.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:24:49,119 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.728e+02 2.532e+02 3.093e+02 4.065e+02 1.254e+03, threshold=6.185e+02, percent-clipped=7.0 +2023-02-06 16:24:49,140 INFO [train.py:901] (3/4) Epoch 15, batch 2200, loss[loss=0.2225, simple_loss=0.3027, pruned_loss=0.0711, over 8505.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.2993, pruned_loss=0.07089, over 1615655.59 frames. ], batch size: 26, lr: 5.12e-03, grad_scale: 16.0 +2023-02-06 16:25:07,044 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115388.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:25:07,721 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115389.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:25:12,316 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3352, 1.8791, 4.4324, 1.8835, 2.5576, 4.8668, 5.0431, 4.1926], + device='cuda:3'), covar=tensor([0.1117, 0.1696, 0.0255, 0.2043, 0.1037, 0.0205, 0.0430, 0.0618], + device='cuda:3'), in_proj_covar=tensor([0.0278, 0.0309, 0.0272, 0.0300, 0.0287, 0.0248, 0.0378, 0.0300], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 16:25:24,253 INFO [train.py:901] (3/4) Epoch 15, batch 2250, loss[loss=0.2183, simple_loss=0.3102, pruned_loss=0.06318, over 8251.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.3004, pruned_loss=0.07107, over 1617946.75 frames. ], batch size: 24, lr: 5.12e-03, grad_scale: 8.0 +2023-02-06 16:25:48,108 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=115448.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:25:48,945 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115449.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:25:58,283 INFO [train.py:901] (3/4) Epoch 15, batch 2300, loss[loss=0.2197, simple_loss=0.3003, pruned_loss=0.06952, over 8453.00 frames. ], tot_loss[loss=0.2197, simple_loss=0.2991, pruned_loss=0.07012, over 1615400.37 frames. ], batch size: 25, lr: 5.12e-03, grad_scale: 8.0 +2023-02-06 16:25:58,960 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.315e+02 2.502e+02 3.175e+02 3.927e+02 9.067e+02, threshold=6.350e+02, percent-clipped=5.0 +2023-02-06 16:26:04,769 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.5548, 5.5955, 4.9327, 2.6844, 4.9751, 5.3469, 5.1526, 5.0687], + device='cuda:3'), covar=tensor([0.0553, 0.0376, 0.0947, 0.3858, 0.0690, 0.0735, 0.0999, 0.0626], + device='cuda:3'), in_proj_covar=tensor([0.0491, 0.0405, 0.0407, 0.0506, 0.0406, 0.0405, 0.0392, 0.0353], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 16:26:07,556 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115474.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:26:17,353 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2619, 1.3393, 1.5711, 1.2066, 0.7290, 1.3602, 1.1861, 1.0990], + device='cuda:3'), covar=tensor([0.0541, 0.1215, 0.1625, 0.1404, 0.0565, 0.1445, 0.0713, 0.0651], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0152, 0.0190, 0.0156, 0.0102, 0.0162, 0.0115, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 16:26:34,675 INFO [train.py:901] (3/4) Epoch 15, batch 2350, loss[loss=0.2164, simple_loss=0.309, pruned_loss=0.06196, over 8034.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.299, pruned_loss=0.07037, over 1617308.55 frames. ], batch size: 22, lr: 5.12e-03, grad_scale: 8.0 +2023-02-06 16:26:52,706 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-06 16:27:09,317 INFO [train.py:901] (3/4) Epoch 15, batch 2400, loss[loss=0.2323, simple_loss=0.3151, pruned_loss=0.07468, over 8315.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2986, pruned_loss=0.07047, over 1618432.63 frames. ], batch size: 25, lr: 5.12e-03, grad_scale: 8.0 +2023-02-06 16:27:09,522 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115563.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:27:10,002 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.670e+02 2.542e+02 3.047e+02 3.524e+02 9.073e+02, threshold=6.095e+02, percent-clipped=1.0 +2023-02-06 16:27:14,077 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2263, 3.1311, 2.8626, 1.5364, 2.8279, 2.8368, 2.8919, 2.7490], + device='cuda:3'), covar=tensor([0.1147, 0.0881, 0.1437, 0.4902, 0.1213, 0.1437, 0.1495, 0.1315], + device='cuda:3'), in_proj_covar=tensor([0.0493, 0.0405, 0.0406, 0.0506, 0.0404, 0.0404, 0.0391, 0.0352], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 16:27:39,712 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115604.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:27:42,703 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 +2023-02-06 16:27:45,573 INFO [train.py:901] (3/4) Epoch 15, batch 2450, loss[loss=0.1878, simple_loss=0.2669, pruned_loss=0.05429, over 7929.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2982, pruned_loss=0.07007, over 1616082.71 frames. ], batch size: 20, lr: 5.12e-03, grad_scale: 8.0 +2023-02-06 16:27:56,559 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115629.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:28:19,897 INFO [train.py:901] (3/4) Epoch 15, batch 2500, loss[loss=0.2644, simple_loss=0.3327, pruned_loss=0.098, over 8333.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2986, pruned_loss=0.07052, over 1616592.12 frames. ], batch size: 25, lr: 5.12e-03, grad_scale: 8.0 +2023-02-06 16:28:20,562 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.646e+02 2.367e+02 2.686e+02 3.697e+02 9.165e+02, threshold=5.372e+02, percent-clipped=5.0 +2023-02-06 16:28:21,755 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 +2023-02-06 16:28:36,131 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.30 vs. limit=5.0 +2023-02-06 16:28:40,554 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5712, 1.9536, 2.0226, 1.1269, 2.0992, 1.4782, 0.5292, 1.8419], + device='cuda:3'), covar=tensor([0.0495, 0.0277, 0.0214, 0.0478, 0.0298, 0.0709, 0.0715, 0.0223], + device='cuda:3'), in_proj_covar=tensor([0.0420, 0.0358, 0.0307, 0.0413, 0.0343, 0.0499, 0.0372, 0.0383], + device='cuda:3'), out_proj_covar=tensor([1.1643e-04, 9.6288e-05, 8.2497e-05, 1.1198e-04, 9.3219e-05, 1.4539e-04, + 1.0298e-04, 1.0469e-04], device='cuda:3') +2023-02-06 16:28:41,954 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5721, 1.7693, 2.7540, 1.4342, 1.9663, 1.9077, 1.5761, 1.8240], + device='cuda:3'), covar=tensor([0.1806, 0.2277, 0.0798, 0.4104, 0.1742, 0.3024, 0.2001, 0.2178], + device='cuda:3'), in_proj_covar=tensor([0.0495, 0.0549, 0.0539, 0.0602, 0.0621, 0.0565, 0.0495, 0.0617], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 16:28:55,235 INFO [train.py:901] (3/4) Epoch 15, batch 2550, loss[loss=0.1904, simple_loss=0.2751, pruned_loss=0.05287, over 7971.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2969, pruned_loss=0.06964, over 1614120.90 frames. ], batch size: 21, lr: 5.12e-03, grad_scale: 8.0 +2023-02-06 16:29:04,479 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.67 vs. limit=2.0 +2023-02-06 16:29:08,942 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=115732.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:29:09,204 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.32 vs. limit=5.0 +2023-02-06 16:29:09,603 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=115733.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:29:30,406 INFO [train.py:901] (3/4) Epoch 15, batch 2600, loss[loss=0.1915, simple_loss=0.2871, pruned_loss=0.04792, over 7980.00 frames. ], tot_loss[loss=0.2188, simple_loss=0.2979, pruned_loss=0.06982, over 1620579.57 frames. ], batch size: 21, lr: 5.12e-03, grad_scale: 8.0 +2023-02-06 16:29:31,072 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.554e+02 2.427e+02 3.148e+02 3.839e+02 8.607e+02, threshold=6.296e+02, percent-clipped=3.0 +2023-02-06 16:30:04,163 INFO [train.py:901] (3/4) Epoch 15, batch 2650, loss[loss=0.2828, simple_loss=0.3495, pruned_loss=0.108, over 8137.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.2997, pruned_loss=0.07083, over 1626248.02 frames. ], batch size: 22, lr: 5.11e-03, grad_scale: 8.0 +2023-02-06 16:30:08,503 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115819.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:30:27,409 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115844.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:30:29,377 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115847.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:30:30,054 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115848.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:30:39,733 INFO [train.py:901] (3/4) Epoch 15, batch 2700, loss[loss=0.2095, simple_loss=0.3035, pruned_loss=0.05782, over 8196.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.3, pruned_loss=0.0713, over 1619059.45 frames. ], batch size: 23, lr: 5.11e-03, grad_scale: 8.0 +2023-02-06 16:30:40,394 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.667e+02 2.337e+02 2.718e+02 3.606e+02 6.832e+02, threshold=5.436e+02, percent-clipped=3.0 +2023-02-06 16:31:12,010 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115910.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 16:31:13,864 INFO [train.py:901] (3/4) Epoch 15, batch 2750, loss[loss=0.22, simple_loss=0.3071, pruned_loss=0.06643, over 8494.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.2991, pruned_loss=0.0712, over 1612043.53 frames. ], batch size: 26, lr: 5.11e-03, grad_scale: 8.0 +2023-02-06 16:31:32,055 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.6684, 3.6259, 3.3342, 1.9074, 3.2431, 3.2448, 3.3025, 3.0498], + device='cuda:3'), covar=tensor([0.1010, 0.0760, 0.1123, 0.4665, 0.0951, 0.1305, 0.1404, 0.1115], + device='cuda:3'), in_proj_covar=tensor([0.0485, 0.0401, 0.0404, 0.0503, 0.0397, 0.0400, 0.0388, 0.0349], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 16:31:44,881 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8900, 1.4732, 3.5428, 1.3497, 2.4214, 3.9044, 3.9623, 3.3485], + device='cuda:3'), covar=tensor([0.1100, 0.1691, 0.0298, 0.2114, 0.1040, 0.0222, 0.0483, 0.0593], + device='cuda:3'), in_proj_covar=tensor([0.0277, 0.0304, 0.0270, 0.0296, 0.0287, 0.0247, 0.0376, 0.0298], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 16:31:49,503 INFO [train.py:901] (3/4) Epoch 15, batch 2800, loss[loss=0.2616, simple_loss=0.3244, pruned_loss=0.09943, over 6982.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.3004, pruned_loss=0.07138, over 1618037.19 frames. ], batch size: 71, lr: 5.11e-03, grad_scale: 8.0 +2023-02-06 16:31:50,151 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.450e+02 2.517e+02 2.986e+02 3.677e+02 9.071e+02, threshold=5.972e+02, percent-clipped=5.0 +2023-02-06 16:32:24,937 INFO [train.py:901] (3/4) Epoch 15, batch 2850, loss[loss=0.219, simple_loss=0.3, pruned_loss=0.069, over 8144.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.2985, pruned_loss=0.0708, over 1611430.85 frames. ], batch size: 22, lr: 5.11e-03, grad_scale: 8.0 +2023-02-06 16:32:35,416 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6726, 1.4014, 1.5169, 1.1965, 0.8048, 1.3042, 1.4921, 1.2559], + device='cuda:3'), covar=tensor([0.0541, 0.1272, 0.1734, 0.1514, 0.0623, 0.1593, 0.0756, 0.0702], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0153, 0.0190, 0.0156, 0.0101, 0.0162, 0.0115, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 16:32:38,092 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116032.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:32:46,508 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2164, 2.4458, 2.0273, 2.7612, 1.4548, 1.8509, 2.1437, 2.3798], + device='cuda:3'), covar=tensor([0.0623, 0.0718, 0.0838, 0.0398, 0.1105, 0.1148, 0.0863, 0.0715], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0203, 0.0249, 0.0211, 0.0212, 0.0250, 0.0253, 0.0214], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 16:33:00,841 INFO [train.py:901] (3/4) Epoch 15, batch 2900, loss[loss=0.2272, simple_loss=0.3183, pruned_loss=0.06807, over 8727.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2991, pruned_loss=0.07078, over 1615498.50 frames. ], batch size: 30, lr: 5.11e-03, grad_scale: 8.0 +2023-02-06 16:33:01,416 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.668e+02 2.452e+02 2.959e+02 3.782e+02 6.842e+02, threshold=5.917e+02, percent-clipped=3.0 +2023-02-06 16:33:29,120 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116103.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:33:29,805 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116104.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:33:35,327 INFO [train.py:901] (3/4) Epoch 15, batch 2950, loss[loss=0.1984, simple_loss=0.277, pruned_loss=0.05985, over 8101.00 frames. ], tot_loss[loss=0.2214, simple_loss=0.3, pruned_loss=0.0714, over 1615747.87 frames. ], batch size: 21, lr: 5.11e-03, grad_scale: 8.0 +2023-02-06 16:33:36,706 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 16:33:42,120 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116123.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:33:45,596 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116128.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:33:46,298 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116129.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:33:49,699 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116134.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:34:08,977 INFO [train.py:901] (3/4) Epoch 15, batch 3000, loss[loss=0.2595, simple_loss=0.3236, pruned_loss=0.09766, over 6890.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.2993, pruned_loss=0.07086, over 1617088.55 frames. ], batch size: 72, lr: 5.11e-03, grad_scale: 8.0 +2023-02-06 16:34:08,977 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 16:34:21,680 INFO [train.py:935] (3/4) Epoch 15, validation: loss=0.1808, simple_loss=0.2809, pruned_loss=0.04034, over 944034.00 frames. +2023-02-06 16:34:21,681 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 16:34:22,359 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.826e+02 2.534e+02 3.127e+02 3.845e+02 7.463e+02, threshold=6.253e+02, percent-clipped=8.0 +2023-02-06 16:34:32,180 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4163, 1.8979, 2.9264, 1.2432, 2.0932, 1.7428, 1.6060, 1.9688], + device='cuda:3'), covar=tensor([0.2115, 0.2523, 0.0941, 0.4674, 0.1983, 0.3590, 0.2330, 0.2797], + device='cuda:3'), in_proj_covar=tensor([0.0497, 0.0552, 0.0542, 0.0600, 0.0624, 0.0567, 0.0495, 0.0619], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 16:34:57,899 INFO [train.py:901] (3/4) Epoch 15, batch 3050, loss[loss=0.2213, simple_loss=0.3095, pruned_loss=0.06649, over 8465.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2991, pruned_loss=0.07073, over 1613737.08 frames. ], batch size: 29, lr: 5.11e-03, grad_scale: 8.0 +2023-02-06 16:35:15,576 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4940, 2.9261, 2.3267, 3.5779, 1.8509, 2.0552, 2.3633, 2.9175], + device='cuda:3'), covar=tensor([0.0627, 0.0608, 0.0805, 0.0311, 0.0999, 0.1242, 0.0961, 0.0682], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0202, 0.0248, 0.0211, 0.0211, 0.0249, 0.0254, 0.0213], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 16:35:26,166 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116254.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 16:35:31,957 INFO [train.py:901] (3/4) Epoch 15, batch 3100, loss[loss=0.2059, simple_loss=0.2847, pruned_loss=0.06356, over 8323.00 frames. ], tot_loss[loss=0.2217, simple_loss=0.3003, pruned_loss=0.07151, over 1615393.26 frames. ], batch size: 26, lr: 5.10e-03, grad_scale: 8.0 +2023-02-06 16:35:32,579 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.632e+02 2.573e+02 3.095e+02 3.865e+02 1.142e+03, threshold=6.190e+02, percent-clipped=3.0 +2023-02-06 16:36:06,934 INFO [train.py:901] (3/4) Epoch 15, batch 3150, loss[loss=0.2187, simple_loss=0.3095, pruned_loss=0.06393, over 8244.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.2996, pruned_loss=0.07129, over 1615148.73 frames. ], batch size: 24, lr: 5.10e-03, grad_scale: 8.0 +2023-02-06 16:36:27,210 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116341.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:36:41,980 INFO [train.py:901] (3/4) Epoch 15, batch 3200, loss[loss=0.2379, simple_loss=0.3193, pruned_loss=0.07827, over 8238.00 frames. ], tot_loss[loss=0.2229, simple_loss=0.3013, pruned_loss=0.07222, over 1617128.73 frames. ], batch size: 24, lr: 5.10e-03, grad_scale: 8.0 +2023-02-06 16:36:43,354 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.618e+02 2.524e+02 3.304e+02 3.942e+02 1.206e+03, threshold=6.608e+02, percent-clipped=2.0 +2023-02-06 16:36:46,747 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116369.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 16:36:51,222 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116376.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:36:53,699 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 16:37:06,199 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2065, 2.3764, 1.9313, 2.9154, 1.3228, 1.7236, 1.9179, 2.3453], + device='cuda:3'), covar=tensor([0.0655, 0.0690, 0.0915, 0.0315, 0.1174, 0.1287, 0.1011, 0.0726], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0203, 0.0250, 0.0212, 0.0211, 0.0250, 0.0257, 0.0214], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 16:37:12,416 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9141, 1.5706, 3.0968, 1.3280, 2.0140, 3.3353, 3.4704, 2.8711], + device='cuda:3'), covar=tensor([0.0969, 0.1582, 0.0356, 0.2044, 0.1097, 0.0267, 0.0546, 0.0585], + device='cuda:3'), in_proj_covar=tensor([0.0277, 0.0306, 0.0272, 0.0299, 0.0288, 0.0248, 0.0377, 0.0299], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 16:37:16,507 INFO [train.py:901] (3/4) Epoch 15, batch 3250, loss[loss=0.2101, simple_loss=0.281, pruned_loss=0.06957, over 7929.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.3001, pruned_loss=0.07147, over 1615355.35 frames. ], batch size: 20, lr: 5.10e-03, grad_scale: 8.0 +2023-02-06 16:37:52,491 INFO [train.py:901] (3/4) Epoch 15, batch 3300, loss[loss=0.2014, simple_loss=0.2899, pruned_loss=0.05642, over 8597.00 frames. ], tot_loss[loss=0.2214, simple_loss=0.3, pruned_loss=0.07146, over 1616921.18 frames. ], batch size: 34, lr: 5.10e-03, grad_scale: 8.0 +2023-02-06 16:37:53,156 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.548e+02 2.388e+02 2.875e+02 3.716e+02 9.209e+02, threshold=5.750e+02, percent-clipped=3.0 +2023-02-06 16:37:53,300 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116464.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:37:55,237 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116467.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:38:02,554 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116478.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:38:12,020 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116491.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:38:26,452 INFO [train.py:901] (3/4) Epoch 15, batch 3350, loss[loss=0.2335, simple_loss=0.3062, pruned_loss=0.0804, over 7430.00 frames. ], tot_loss[loss=0.221, simple_loss=0.2998, pruned_loss=0.0711, over 1616201.98 frames. ], batch size: 17, lr: 5.10e-03, grad_scale: 8.0 +2023-02-06 16:38:33,251 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116523.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:39:02,052 INFO [train.py:901] (3/4) Epoch 15, batch 3400, loss[loss=0.2142, simple_loss=0.2882, pruned_loss=0.07006, over 7717.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2999, pruned_loss=0.07087, over 1619205.41 frames. ], batch size: 18, lr: 5.10e-03, grad_scale: 8.0 +2023-02-06 16:39:02,726 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.566e+02 3.149e+02 4.104e+02 8.501e+02, threshold=6.298e+02, percent-clipped=7.0 +2023-02-06 16:39:14,876 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116582.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:39:22,204 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116593.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:39:36,209 INFO [train.py:901] (3/4) Epoch 15, batch 3450, loss[loss=0.2883, simple_loss=0.3446, pruned_loss=0.116, over 7648.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.2997, pruned_loss=0.07086, over 1613626.10 frames. ], batch size: 19, lr: 5.10e-03, grad_scale: 8.0 +2023-02-06 16:39:44,406 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116625.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 16:39:51,718 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116636.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:40:01,115 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116650.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 16:40:03,772 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7471, 1.9021, 1.9238, 1.3507, 2.0447, 1.5649, 1.0602, 1.7764], + device='cuda:3'), covar=tensor([0.0342, 0.0227, 0.0162, 0.0348, 0.0241, 0.0481, 0.0543, 0.0189], + device='cuda:3'), in_proj_covar=tensor([0.0424, 0.0363, 0.0311, 0.0419, 0.0351, 0.0507, 0.0375, 0.0385], + device='cuda:3'), out_proj_covar=tensor([1.1716e-04, 9.7643e-05, 8.3560e-05, 1.1352e-04, 9.5440e-05, 1.4759e-04, + 1.0361e-04, 1.0489e-04], device='cuda:3') +2023-02-06 16:40:10,039 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.06 vs. limit=5.0 +2023-02-06 16:40:10,185 INFO [train.py:901] (3/4) Epoch 15, batch 3500, loss[loss=0.2214, simple_loss=0.3098, pruned_loss=0.06647, over 8458.00 frames. ], tot_loss[loss=0.2197, simple_loss=0.2986, pruned_loss=0.07035, over 1608594.07 frames. ], batch size: 29, lr: 5.10e-03, grad_scale: 8.0 +2023-02-06 16:40:10,850 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.712e+02 2.398e+02 2.936e+02 3.935e+02 9.560e+02, threshold=5.871e+02, percent-clipped=3.0 +2023-02-06 16:40:13,184 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5216, 1.7959, 2.7437, 1.3413, 1.8978, 1.9325, 1.6026, 1.7547], + device='cuda:3'), covar=tensor([0.1820, 0.2484, 0.0874, 0.4404, 0.1749, 0.3059, 0.2146, 0.2244], + device='cuda:3'), in_proj_covar=tensor([0.0503, 0.0557, 0.0547, 0.0609, 0.0628, 0.0571, 0.0502, 0.0627], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 16:40:26,417 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116685.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:40:35,608 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 16:40:44,884 INFO [train.py:901] (3/4) Epoch 15, batch 3550, loss[loss=0.2143, simple_loss=0.2816, pruned_loss=0.07349, over 7974.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.2982, pruned_loss=0.07018, over 1605787.69 frames. ], batch size: 21, lr: 5.09e-03, grad_scale: 8.0 +2023-02-06 16:40:50,947 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116722.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:41:06,496 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8267, 1.7149, 3.1800, 1.4903, 2.2567, 3.3976, 3.4920, 2.9208], + device='cuda:3'), covar=tensor([0.1074, 0.1414, 0.0325, 0.1881, 0.0871, 0.0268, 0.0563, 0.0584], + device='cuda:3'), in_proj_covar=tensor([0.0280, 0.0308, 0.0272, 0.0304, 0.0291, 0.0251, 0.0380, 0.0302], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 16:41:08,580 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116747.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:41:17,775 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-06 16:41:19,408 INFO [train.py:901] (3/4) Epoch 15, batch 3600, loss[loss=0.1525, simple_loss=0.2379, pruned_loss=0.03358, over 7659.00 frames. ], tot_loss[loss=0.2197, simple_loss=0.2986, pruned_loss=0.07045, over 1603994.82 frames. ], batch size: 19, lr: 5.09e-03, grad_scale: 8.0 +2023-02-06 16:41:20,116 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.456e+02 2.627e+02 3.005e+02 3.918e+02 8.490e+02, threshold=6.010e+02, percent-clipped=4.0 +2023-02-06 16:41:25,799 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116772.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:41:45,864 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4198, 1.5138, 4.2007, 1.9824, 2.3786, 4.7154, 4.8110, 3.9344], + device='cuda:3'), covar=tensor([0.1031, 0.1964, 0.0329, 0.2139, 0.1273, 0.0282, 0.0491, 0.0703], + device='cuda:3'), in_proj_covar=tensor([0.0279, 0.0307, 0.0273, 0.0303, 0.0291, 0.0251, 0.0379, 0.0301], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 16:41:47,426 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116800.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:41:52,898 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116808.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:41:56,184 INFO [train.py:901] (3/4) Epoch 15, batch 3650, loss[loss=0.247, simple_loss=0.3118, pruned_loss=0.09112, over 4947.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.2986, pruned_loss=0.07066, over 1602060.76 frames. ], batch size: 11, lr: 5.09e-03, grad_scale: 8.0 +2023-02-06 16:42:00,908 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116820.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:42:13,592 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116838.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:42:21,035 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116849.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:42:30,284 INFO [train.py:901] (3/4) Epoch 15, batch 3700, loss[loss=0.2101, simple_loss=0.2865, pruned_loss=0.06687, over 7805.00 frames. ], tot_loss[loss=0.2217, simple_loss=0.3005, pruned_loss=0.0714, over 1605703.49 frames. ], batch size: 20, lr: 5.09e-03, grad_scale: 8.0 +2023-02-06 16:42:30,494 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116863.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:42:30,966 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.542e+02 2.301e+02 2.797e+02 3.414e+02 8.630e+02, threshold=5.595e+02, percent-clipped=3.0 +2023-02-06 16:42:33,141 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116867.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:42:36,572 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 16:42:38,094 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116874.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:43:06,649 INFO [train.py:901] (3/4) Epoch 15, batch 3750, loss[loss=0.2304, simple_loss=0.3092, pruned_loss=0.07577, over 8484.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.3008, pruned_loss=0.07092, over 1613855.17 frames. ], batch size: 28, lr: 5.09e-03, grad_scale: 8.0 +2023-02-06 16:43:13,700 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116923.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:43:23,707 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8535, 1.6324, 2.4713, 1.6647, 1.1508, 2.4401, 0.5627, 1.4068], + device='cuda:3'), covar=tensor([0.1834, 0.1659, 0.0386, 0.1623, 0.3637, 0.0438, 0.2389, 0.1723], + device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0181, 0.0112, 0.0216, 0.0260, 0.0116, 0.0163, 0.0177], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 16:43:38,613 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-06 16:43:40,820 INFO [train.py:901] (3/4) Epoch 15, batch 3800, loss[loss=0.2364, simple_loss=0.3285, pruned_loss=0.07221, over 8110.00 frames. ], tot_loss[loss=0.2229, simple_loss=0.3022, pruned_loss=0.07175, over 1616116.58 frames. ], batch size: 23, lr: 5.09e-03, grad_scale: 8.0 +2023-02-06 16:43:41,471 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.704e+02 2.512e+02 2.989e+02 3.697e+02 7.171e+02, threshold=5.977e+02, percent-clipped=7.0 +2023-02-06 16:43:52,421 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=116980.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:43:53,910 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116982.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:44:15,608 INFO [train.py:901] (3/4) Epoch 15, batch 3850, loss[loss=0.2417, simple_loss=0.3329, pruned_loss=0.07519, over 8468.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.3023, pruned_loss=0.0722, over 1614950.62 frames. ], batch size: 25, lr: 5.09e-03, grad_scale: 8.0 +2023-02-06 16:44:42,587 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 16:44:46,207 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117056.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:44:46,539 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.81 vs. limit=5.0 +2023-02-06 16:44:50,965 INFO [train.py:901] (3/4) Epoch 15, batch 3900, loss[loss=0.2321, simple_loss=0.3129, pruned_loss=0.07567, over 8518.00 frames. ], tot_loss[loss=0.2229, simple_loss=0.3016, pruned_loss=0.07213, over 1615937.18 frames. ], batch size: 26, lr: 5.09e-03, grad_scale: 8.0 +2023-02-06 16:44:51,623 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.460e+02 2.428e+02 3.027e+02 3.797e+02 6.654e+02, threshold=6.053e+02, percent-clipped=2.0 +2023-02-06 16:44:53,044 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117066.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:45:03,712 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117081.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:45:12,954 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117095.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:45:17,736 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2029, 2.1961, 1.7440, 1.9266, 1.8768, 1.4421, 1.5920, 1.6767], + device='cuda:3'), covar=tensor([0.1304, 0.0372, 0.1092, 0.0542, 0.0637, 0.1389, 0.0954, 0.0747], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0233, 0.0327, 0.0306, 0.0303, 0.0333, 0.0349, 0.0311], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 16:45:24,922 INFO [train.py:901] (3/4) Epoch 15, batch 3950, loss[loss=0.2193, simple_loss=0.2981, pruned_loss=0.07025, over 8104.00 frames. ], tot_loss[loss=0.2228, simple_loss=0.3012, pruned_loss=0.07221, over 1610196.18 frames. ], batch size: 21, lr: 5.09e-03, grad_scale: 8.0 +2023-02-06 16:45:34,216 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 16:46:01,077 INFO [train.py:901] (3/4) Epoch 15, batch 4000, loss[loss=0.19, simple_loss=0.2669, pruned_loss=0.05652, over 8072.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2998, pruned_loss=0.07093, over 1614299.95 frames. ], batch size: 21, lr: 5.08e-03, grad_scale: 8.0 +2023-02-06 16:46:01,776 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.728e+02 2.473e+02 2.992e+02 3.534e+02 5.115e+02, threshold=5.984e+02, percent-clipped=0.0 +2023-02-06 16:46:01,907 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117164.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:46:12,550 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117179.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:46:13,871 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117181.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:46:15,165 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.2244, 2.4470, 4.4150, 2.6890, 3.9708, 3.7364, 4.0914, 3.9953], + device='cuda:3'), covar=tensor([0.0663, 0.3042, 0.0679, 0.2973, 0.1041, 0.0899, 0.0517, 0.0579], + device='cuda:3'), in_proj_covar=tensor([0.0554, 0.0602, 0.0633, 0.0571, 0.0654, 0.0556, 0.0548, 0.0613], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 16:46:17,466 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2023-02-06 16:46:29,855 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117204.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:46:35,569 INFO [train.py:901] (3/4) Epoch 15, batch 4050, loss[loss=0.2034, simple_loss=0.2829, pruned_loss=0.06191, over 8104.00 frames. ], tot_loss[loss=0.2221, simple_loss=0.3007, pruned_loss=0.07181, over 1612428.18 frames. ], batch size: 23, lr: 5.08e-03, grad_scale: 8.0 +2023-02-06 16:46:53,307 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117238.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:46:53,913 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117239.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:47:11,625 INFO [train.py:901] (3/4) Epoch 15, batch 4100, loss[loss=0.1861, simple_loss=0.2833, pruned_loss=0.0445, over 8457.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.2999, pruned_loss=0.07112, over 1613265.32 frames. ], batch size: 27, lr: 5.08e-03, grad_scale: 8.0 +2023-02-06 16:47:11,825 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117263.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:47:12,285 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.581e+02 2.506e+02 3.096e+02 3.742e+02 9.544e+02, threshold=6.191e+02, percent-clipped=4.0 +2023-02-06 16:47:22,932 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117279.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:47:46,643 INFO [train.py:901] (3/4) Epoch 15, batch 4150, loss[loss=0.2104, simple_loss=0.2974, pruned_loss=0.06171, over 8338.00 frames. ], tot_loss[loss=0.2204, simple_loss=0.2992, pruned_loss=0.07078, over 1613485.77 frames. ], batch size: 26, lr: 5.08e-03, grad_scale: 8.0 +2023-02-06 16:48:10,097 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117347.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:48:12,979 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117351.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:48:21,503 INFO [train.py:901] (3/4) Epoch 15, batch 4200, loss[loss=0.1793, simple_loss=0.2668, pruned_loss=0.04591, over 8127.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2975, pruned_loss=0.06977, over 1611728.40 frames. ], batch size: 22, lr: 5.08e-03, grad_scale: 8.0 +2023-02-06 16:48:22,820 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.518e+02 2.404e+02 2.907e+02 3.383e+02 1.073e+03, threshold=5.814e+02, percent-clipped=1.0 +2023-02-06 16:48:31,971 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117376.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:48:40,576 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 16:48:48,352 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9186, 1.1611, 3.1051, 1.1031, 2.7302, 2.6292, 2.8171, 2.7271], + device='cuda:3'), covar=tensor([0.0853, 0.4060, 0.0959, 0.3856, 0.1348, 0.1036, 0.0717, 0.0860], + device='cuda:3'), in_proj_covar=tensor([0.0550, 0.0605, 0.0628, 0.0571, 0.0650, 0.0552, 0.0546, 0.0610], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 16:48:57,040 INFO [train.py:901] (3/4) Epoch 15, batch 4250, loss[loss=0.2109, simple_loss=0.296, pruned_loss=0.0629, over 8348.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.2975, pruned_loss=0.06951, over 1616344.29 frames. ], batch size: 50, lr: 5.08e-03, grad_scale: 16.0 +2023-02-06 16:49:03,724 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 16:49:14,107 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117437.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:49:30,910 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117462.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:49:31,427 INFO [train.py:901] (3/4) Epoch 15, batch 4300, loss[loss=0.1803, simple_loss=0.2688, pruned_loss=0.04591, over 6006.00 frames. ], tot_loss[loss=0.2197, simple_loss=0.2987, pruned_loss=0.0703, over 1620764.32 frames. ], batch size: 13, lr: 5.08e-03, grad_scale: 16.0 +2023-02-06 16:49:32,092 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.472e+02 2.479e+02 3.115e+02 3.892e+02 7.815e+02, threshold=6.229e+02, percent-clipped=5.0 +2023-02-06 16:50:07,599 INFO [train.py:901] (3/4) Epoch 15, batch 4350, loss[loss=0.2488, simple_loss=0.3229, pruned_loss=0.08736, over 8781.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.3, pruned_loss=0.07112, over 1618910.26 frames. ], batch size: 39, lr: 5.08e-03, grad_scale: 16.0 +2023-02-06 16:50:23,038 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117535.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:50:34,438 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-06 16:50:36,329 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 16:50:40,554 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117560.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:50:41,781 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3821, 4.2999, 3.9470, 2.2348, 3.9046, 3.9104, 4.0187, 3.6576], + device='cuda:3'), covar=tensor([0.0775, 0.0614, 0.1027, 0.4406, 0.0873, 0.1279, 0.1267, 0.0853], + device='cuda:3'), in_proj_covar=tensor([0.0489, 0.0408, 0.0411, 0.0509, 0.0402, 0.0409, 0.0393, 0.0355], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 16:50:42,368 INFO [train.py:901] (3/4) Epoch 15, batch 4400, loss[loss=0.246, simple_loss=0.3149, pruned_loss=0.08855, over 8713.00 frames. ], tot_loss[loss=0.2214, simple_loss=0.2999, pruned_loss=0.07145, over 1617784.30 frames. ], batch size: 39, lr: 5.08e-03, grad_scale: 16.0 +2023-02-06 16:50:43,036 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.635e+02 2.383e+02 3.124e+02 3.901e+02 9.506e+02, threshold=6.248e+02, percent-clipped=7.0 +2023-02-06 16:50:55,792 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117583.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:51:08,235 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4944, 4.4134, 3.9946, 1.9257, 4.0292, 3.9449, 4.0329, 3.8037], + device='cuda:3'), covar=tensor([0.0800, 0.0677, 0.1176, 0.4584, 0.0877, 0.0841, 0.1383, 0.0700], + device='cuda:3'), in_proj_covar=tensor([0.0483, 0.0403, 0.0405, 0.0503, 0.0396, 0.0405, 0.0388, 0.0350], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 16:51:17,968 INFO [train.py:901] (3/4) Epoch 15, batch 4450, loss[loss=0.2285, simple_loss=0.3058, pruned_loss=0.07563, over 8483.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.2997, pruned_loss=0.07144, over 1615406.82 frames. ], batch size: 27, lr: 5.07e-03, grad_scale: 16.0 +2023-02-06 16:51:17,982 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 16:51:52,096 INFO [train.py:901] (3/4) Epoch 15, batch 4500, loss[loss=0.2348, simple_loss=0.3204, pruned_loss=0.07459, over 8604.00 frames. ], tot_loss[loss=0.2218, simple_loss=0.3004, pruned_loss=0.0716, over 1617463.45 frames. ], batch size: 49, lr: 5.07e-03, grad_scale: 16.0 +2023-02-06 16:51:52,740 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.719e+02 2.480e+02 2.963e+02 4.043e+02 1.091e+03, threshold=5.927e+02, percent-clipped=5.0 +2023-02-06 16:52:11,249 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=117691.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:52:11,855 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 16:52:16,201 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117698.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:52:17,579 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6489, 2.7541, 2.4023, 3.8202, 1.7045, 1.8377, 2.3206, 3.1148], + device='cuda:3'), covar=tensor([0.0627, 0.0894, 0.0914, 0.0273, 0.1236, 0.1522, 0.1135, 0.0809], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0205, 0.0253, 0.0213, 0.0213, 0.0250, 0.0258, 0.0214], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 16:52:27,011 INFO [train.py:901] (3/4) Epoch 15, batch 4550, loss[loss=0.1734, simple_loss=0.2485, pruned_loss=0.04911, over 7689.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2992, pruned_loss=0.07068, over 1618383.66 frames. ], batch size: 18, lr: 5.07e-03, grad_scale: 8.0 +2023-02-06 16:53:02,114 INFO [train.py:901] (3/4) Epoch 15, batch 4600, loss[loss=0.2254, simple_loss=0.3121, pruned_loss=0.0693, over 8356.00 frames. ], tot_loss[loss=0.2212, simple_loss=0.3, pruned_loss=0.07124, over 1618645.96 frames. ], batch size: 24, lr: 5.07e-03, grad_scale: 8.0 +2023-02-06 16:53:03,483 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.449e+02 2.311e+02 2.848e+02 3.671e+02 5.923e+02, threshold=5.697e+02, percent-clipped=0.0 +2023-02-06 16:53:31,669 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117806.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:53:34,339 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2643, 1.2323, 1.5005, 1.1744, 0.7373, 1.2847, 1.2227, 1.2601], + device='cuda:3'), covar=tensor([0.0535, 0.1382, 0.1692, 0.1456, 0.0572, 0.1582, 0.0717, 0.0630], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0153, 0.0191, 0.0156, 0.0101, 0.0162, 0.0114, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 16:53:36,042 INFO [train.py:901] (3/4) Epoch 15, batch 4650, loss[loss=0.2909, simple_loss=0.3489, pruned_loss=0.1165, over 8355.00 frames. ], tot_loss[loss=0.2226, simple_loss=0.3008, pruned_loss=0.07226, over 1615071.76 frames. ], batch size: 24, lr: 5.07e-03, grad_scale: 8.0 +2023-02-06 16:53:43,323 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-06 16:53:52,453 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7456, 1.8436, 1.8492, 1.5129, 1.9384, 1.5192, 1.1319, 1.8089], + device='cuda:3'), covar=tensor([0.0382, 0.0241, 0.0181, 0.0349, 0.0244, 0.0488, 0.0559, 0.0190], + device='cuda:3'), in_proj_covar=tensor([0.0418, 0.0359, 0.0310, 0.0414, 0.0344, 0.0504, 0.0375, 0.0381], + device='cuda:3'), out_proj_covar=tensor([1.1551e-04, 9.6559e-05, 8.3003e-05, 1.1208e-04, 9.3504e-05, 1.4687e-04, + 1.0354e-04, 1.0346e-04], device='cuda:3') +2023-02-06 16:54:11,642 INFO [train.py:901] (3/4) Epoch 15, batch 4700, loss[loss=0.2454, simple_loss=0.3224, pruned_loss=0.0842, over 8285.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.3004, pruned_loss=0.07168, over 1614551.91 frames. ], batch size: 23, lr: 5.07e-03, grad_scale: 8.0 +2023-02-06 16:54:12,886 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.594e+02 2.509e+02 3.109e+02 4.231e+02 8.316e+02, threshold=6.217e+02, percent-clipped=12.0 +2023-02-06 16:54:46,551 INFO [train.py:901] (3/4) Epoch 15, batch 4750, loss[loss=0.2781, simple_loss=0.3471, pruned_loss=0.1046, over 8461.00 frames. ], tot_loss[loss=0.2227, simple_loss=0.3007, pruned_loss=0.07232, over 1612883.95 frames. ], batch size: 25, lr: 5.07e-03, grad_scale: 8.0 +2023-02-06 16:55:10,056 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3105, 2.8568, 2.2683, 3.8463, 1.9222, 1.8777, 2.3688, 3.0527], + device='cuda:3'), covar=tensor([0.0749, 0.0856, 0.0914, 0.0281, 0.1119, 0.1388, 0.1037, 0.0769], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0204, 0.0251, 0.0212, 0.0211, 0.0248, 0.0255, 0.0213], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 16:55:11,955 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 16:55:15,282 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 16:55:16,043 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117954.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:55:22,513 INFO [train.py:901] (3/4) Epoch 15, batch 4800, loss[loss=0.2298, simple_loss=0.302, pruned_loss=0.07877, over 8596.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.3007, pruned_loss=0.07195, over 1617177.90 frames. ], batch size: 34, lr: 5.07e-03, grad_scale: 8.0 +2023-02-06 16:55:23,938 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.442e+02 2.482e+02 3.121e+02 4.555e+02 1.692e+03, threshold=6.242e+02, percent-clipped=8.0 +2023-02-06 16:55:33,837 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117979.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:55:57,739 INFO [train.py:901] (3/4) Epoch 15, batch 4850, loss[loss=0.2478, simple_loss=0.3228, pruned_loss=0.08641, over 8442.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.3019, pruned_loss=0.07275, over 1622932.73 frames. ], batch size: 29, lr: 5.07e-03, grad_scale: 8.0 +2023-02-06 16:56:07,046 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 16:56:29,212 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118058.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:56:31,989 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118062.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:56:32,449 INFO [train.py:901] (3/4) Epoch 15, batch 4900, loss[loss=0.2246, simple_loss=0.287, pruned_loss=0.0811, over 7800.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.3021, pruned_loss=0.07306, over 1620391.74 frames. ], batch size: 19, lr: 5.07e-03, grad_scale: 8.0 +2023-02-06 16:56:33,728 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.717e+02 2.453e+02 2.951e+02 3.688e+02 9.605e+02, threshold=5.903e+02, percent-clipped=5.0 +2023-02-06 16:56:50,262 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118087.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:57:07,604 INFO [train.py:901] (3/4) Epoch 15, batch 4950, loss[loss=0.2758, simple_loss=0.3409, pruned_loss=0.1053, over 7964.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.3019, pruned_loss=0.0729, over 1619566.35 frames. ], batch size: 21, lr: 5.06e-03, grad_scale: 8.0 +2023-02-06 16:57:09,214 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3143, 2.2271, 1.7137, 1.9750, 1.8475, 1.4149, 1.7138, 1.7156], + device='cuda:3'), covar=tensor([0.1159, 0.0339, 0.1061, 0.0505, 0.0619, 0.1436, 0.0914, 0.0699], + device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0228, 0.0324, 0.0300, 0.0300, 0.0331, 0.0345, 0.0309], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 16:57:42,121 INFO [train.py:901] (3/4) Epoch 15, batch 5000, loss[loss=0.2149, simple_loss=0.3026, pruned_loss=0.06358, over 8289.00 frames. ], tot_loss[loss=0.2228, simple_loss=0.3008, pruned_loss=0.0724, over 1615934.69 frames. ], batch size: 23, lr: 5.06e-03, grad_scale: 8.0 +2023-02-06 16:57:43,378 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.580e+02 2.421e+02 2.910e+02 3.813e+02 6.624e+02, threshold=5.820e+02, percent-clipped=4.0 +2023-02-06 16:57:58,591 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118186.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:58:17,614 INFO [train.py:901] (3/4) Epoch 15, batch 5050, loss[loss=0.2384, simple_loss=0.3224, pruned_loss=0.07724, over 8324.00 frames. ], tot_loss[loss=0.2212, simple_loss=0.2993, pruned_loss=0.07157, over 1610368.79 frames. ], batch size: 25, lr: 5.06e-03, grad_scale: 8.0 +2023-02-06 16:58:23,569 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 16:58:43,443 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 16:58:52,547 INFO [train.py:901] (3/4) Epoch 15, batch 5100, loss[loss=0.2192, simple_loss=0.2975, pruned_loss=0.07047, over 6009.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.3003, pruned_loss=0.0718, over 1613475.21 frames. ], batch size: 13, lr: 5.06e-03, grad_scale: 8.0 +2023-02-06 16:58:53,819 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.797e+02 2.591e+02 3.125e+02 3.877e+02 7.785e+02, threshold=6.249e+02, percent-clipped=4.0 +2023-02-06 16:58:57,519 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-06 16:59:09,115 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118287.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:59:23,817 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118307.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 16:59:27,782 INFO [train.py:901] (3/4) Epoch 15, batch 5150, loss[loss=0.1957, simple_loss=0.2803, pruned_loss=0.05557, over 7930.00 frames. ], tot_loss[loss=0.222, simple_loss=0.3005, pruned_loss=0.07174, over 1615152.28 frames. ], batch size: 20, lr: 5.06e-03, grad_scale: 8.0 +2023-02-06 16:59:42,970 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4794, 1.8063, 2.7146, 1.3359, 2.0055, 1.8236, 1.5625, 1.9587], + device='cuda:3'), covar=tensor([0.1777, 0.2257, 0.0881, 0.4016, 0.1662, 0.2965, 0.2047, 0.2087], + device='cuda:3'), in_proj_covar=tensor([0.0497, 0.0551, 0.0535, 0.0604, 0.0625, 0.0564, 0.0494, 0.0616], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 16:59:51,101 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118347.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:00:02,426 INFO [train.py:901] (3/4) Epoch 15, batch 5200, loss[loss=0.226, simple_loss=0.3111, pruned_loss=0.0705, over 8481.00 frames. ], tot_loss[loss=0.2221, simple_loss=0.3006, pruned_loss=0.07178, over 1617437.89 frames. ], batch size: 25, lr: 5.06e-03, grad_scale: 8.0 +2023-02-06 17:00:03,692 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.390e+02 2.269e+02 2.811e+02 3.673e+02 9.088e+02, threshold=5.623e+02, percent-clipped=2.0 +2023-02-06 17:00:13,700 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8043, 1.6067, 2.0958, 1.7892, 1.9532, 1.8067, 1.5474, 0.8077], + device='cuda:3'), covar=tensor([0.4920, 0.4031, 0.1403, 0.2733, 0.2025, 0.2546, 0.1959, 0.4150], + device='cuda:3'), in_proj_covar=tensor([0.0915, 0.0922, 0.0758, 0.0891, 0.0960, 0.0844, 0.0722, 0.0799], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 17:00:22,811 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 17:00:29,681 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118402.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:00:37,917 INFO [train.py:901] (3/4) Epoch 15, batch 5250, loss[loss=0.213, simple_loss=0.3021, pruned_loss=0.06191, over 8499.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2992, pruned_loss=0.07066, over 1615706.69 frames. ], batch size: 26, lr: 5.06e-03, grad_scale: 8.0 +2023-02-06 17:00:46,149 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 17:00:47,619 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4953, 1.3454, 4.7479, 1.7992, 4.1055, 3.9561, 4.3050, 4.1326], + device='cuda:3'), covar=tensor([0.0619, 0.4678, 0.0456, 0.3545, 0.1113, 0.0929, 0.0536, 0.0659], + device='cuda:3'), in_proj_covar=tensor([0.0556, 0.0609, 0.0639, 0.0580, 0.0655, 0.0560, 0.0554, 0.0616], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 17:01:12,979 INFO [train.py:901] (3/4) Epoch 15, batch 5300, loss[loss=0.1983, simple_loss=0.2907, pruned_loss=0.05293, over 8552.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2981, pruned_loss=0.07048, over 1610343.42 frames. ], batch size: 31, lr: 5.06e-03, grad_scale: 8.0 +2023-02-06 17:01:14,347 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.517e+02 2.534e+02 2.995e+02 3.765e+02 8.916e+02, threshold=5.991e+02, percent-clipped=4.0 +2023-02-06 17:01:24,824 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 +2023-02-06 17:01:47,927 INFO [train.py:901] (3/4) Epoch 15, batch 5350, loss[loss=0.1833, simple_loss=0.2554, pruned_loss=0.05559, over 7694.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2972, pruned_loss=0.06991, over 1607865.93 frames. ], batch size: 18, lr: 5.06e-03, grad_scale: 8.0 +2023-02-06 17:01:50,886 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118517.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:01:52,212 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2343, 3.1707, 2.8550, 1.6433, 2.8870, 2.8930, 2.8879, 2.8009], + device='cuda:3'), covar=tensor([0.1363, 0.0962, 0.1805, 0.4868, 0.1203, 0.1272, 0.1863, 0.1193], + device='cuda:3'), in_proj_covar=tensor([0.0487, 0.0402, 0.0408, 0.0505, 0.0401, 0.0410, 0.0391, 0.0354], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 17:02:01,054 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118530.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:02:24,461 INFO [train.py:901] (3/4) Epoch 15, batch 5400, loss[loss=0.2271, simple_loss=0.3106, pruned_loss=0.07181, over 8518.00 frames. ], tot_loss[loss=0.219, simple_loss=0.2978, pruned_loss=0.07007, over 1609866.61 frames. ], batch size: 28, lr: 5.05e-03, grad_scale: 8.0 +2023-02-06 17:02:25,792 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.551e+02 2.478e+02 2.903e+02 3.717e+02 8.291e+02, threshold=5.806e+02, percent-clipped=5.0 +2023-02-06 17:02:53,355 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.67 vs. limit=2.0 +2023-02-06 17:02:55,830 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1352, 1.6046, 1.7091, 1.4171, 0.9288, 1.4803, 1.7878, 1.5853], + device='cuda:3'), covar=tensor([0.0504, 0.1197, 0.1652, 0.1365, 0.0581, 0.1467, 0.0644, 0.0638], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0153, 0.0190, 0.0157, 0.0101, 0.0162, 0.0113, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 17:02:58,969 INFO [train.py:901] (3/4) Epoch 15, batch 5450, loss[loss=0.2192, simple_loss=0.288, pruned_loss=0.07516, over 7651.00 frames. ], tot_loss[loss=0.2186, simple_loss=0.2974, pruned_loss=0.06993, over 1609763.98 frames. ], batch size: 19, lr: 5.05e-03, grad_scale: 8.0 +2023-02-06 17:03:11,222 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118631.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:03:21,567 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118645.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:03:24,936 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118649.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:03:26,157 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118651.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:03:34,862 INFO [train.py:901] (3/4) Epoch 15, batch 5500, loss[loss=0.1869, simple_loss=0.2712, pruned_loss=0.05123, over 8083.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2962, pruned_loss=0.06968, over 1604268.22 frames. ], batch size: 21, lr: 5.05e-03, grad_scale: 8.0 +2023-02-06 17:03:36,244 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.663e+02 2.592e+02 3.113e+02 3.610e+02 8.755e+02, threshold=6.227e+02, percent-clipped=2.0 +2023-02-06 17:03:38,383 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 17:03:54,400 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118691.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:04:09,096 INFO [train.py:901] (3/4) Epoch 15, batch 5550, loss[loss=0.1868, simple_loss=0.2732, pruned_loss=0.05016, over 8291.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2963, pruned_loss=0.06966, over 1605412.89 frames. ], batch size: 23, lr: 5.05e-03, grad_scale: 8.0 +2023-02-06 17:04:32,310 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118746.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:04:44,981 INFO [train.py:901] (3/4) Epoch 15, batch 5600, loss[loss=0.1943, simple_loss=0.2682, pruned_loss=0.06022, over 7703.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2972, pruned_loss=0.06986, over 1606150.98 frames. ], batch size: 18, lr: 5.05e-03, grad_scale: 8.0 +2023-02-06 17:04:46,293 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.617e+02 2.537e+02 3.218e+02 3.925e+02 9.216e+02, threshold=6.435e+02, percent-clipped=4.0 +2023-02-06 17:04:47,201 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118766.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:04:52,585 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118773.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:05:09,078 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118798.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:05:14,490 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118806.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:05:19,156 INFO [train.py:901] (3/4) Epoch 15, batch 5650, loss[loss=0.1747, simple_loss=0.2549, pruned_loss=0.04726, over 7438.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2962, pruned_loss=0.06915, over 1608552.75 frames. ], batch size: 17, lr: 5.05e-03, grad_scale: 8.0 +2023-02-06 17:05:43,436 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 17:05:53,490 INFO [train.py:901] (3/4) Epoch 15, batch 5700, loss[loss=0.2724, simple_loss=0.3327, pruned_loss=0.106, over 8704.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2975, pruned_loss=0.06974, over 1612944.54 frames. ], batch size: 34, lr: 5.05e-03, grad_scale: 8.0 +2023-02-06 17:05:54,818 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.577e+02 2.491e+02 2.972e+02 3.726e+02 7.690e+02, threshold=5.944e+02, percent-clipped=5.0 +2023-02-06 17:06:00,481 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5782, 1.9761, 2.1534, 1.2919, 2.2577, 1.4939, 0.6516, 1.7905], + device='cuda:3'), covar=tensor([0.0491, 0.0254, 0.0227, 0.0423, 0.0275, 0.0651, 0.0659, 0.0253], + device='cuda:3'), in_proj_covar=tensor([0.0415, 0.0357, 0.0307, 0.0412, 0.0342, 0.0500, 0.0371, 0.0380], + device='cuda:3'), out_proj_covar=tensor([1.1472e-04, 9.6024e-05, 8.2225e-05, 1.1124e-04, 9.2625e-05, 1.4529e-04, + 1.0207e-04, 1.0334e-04], device='cuda:3') +2023-02-06 17:06:21,168 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118901.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:06:29,112 INFO [train.py:901] (3/4) Epoch 15, batch 5750, loss[loss=0.2437, simple_loss=0.3229, pruned_loss=0.08227, over 8525.00 frames. ], tot_loss[loss=0.2196, simple_loss=0.2986, pruned_loss=0.0703, over 1615677.73 frames. ], batch size: 39, lr: 5.05e-03, grad_scale: 8.0 +2023-02-06 17:06:38,203 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118926.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:06:46,365 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 17:06:58,316 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-06 17:06:59,552 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6538, 1.3995, 4.8540, 1.8851, 4.3123, 4.0480, 4.3911, 4.2677], + device='cuda:3'), covar=tensor([0.0599, 0.4492, 0.0433, 0.3469, 0.1000, 0.0855, 0.0515, 0.0590], + device='cuda:3'), in_proj_covar=tensor([0.0554, 0.0602, 0.0632, 0.0572, 0.0649, 0.0552, 0.0549, 0.0614], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 17:07:04,200 INFO [train.py:901] (3/4) Epoch 15, batch 5800, loss[loss=0.2144, simple_loss=0.3, pruned_loss=0.06442, over 8473.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2967, pruned_loss=0.06934, over 1610753.17 frames. ], batch size: 27, lr: 5.05e-03, grad_scale: 8.0 +2023-02-06 17:07:05,539 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.613e+02 2.317e+02 2.944e+02 4.100e+02 6.996e+02, threshold=5.887e+02, percent-clipped=4.0 +2023-02-06 17:07:26,177 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=118993.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:07:32,096 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119002.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:07:39,886 INFO [train.py:901] (3/4) Epoch 15, batch 5850, loss[loss=0.207, simple_loss=0.282, pruned_loss=0.06595, over 8731.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2957, pruned_loss=0.06882, over 1612811.00 frames. ], batch size: 30, lr: 5.04e-03, grad_scale: 8.0 +2023-02-06 17:07:46,173 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119022.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:07:49,457 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119027.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:08:02,813 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119047.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:08:13,684 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119062.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:08:14,146 INFO [train.py:901] (3/4) Epoch 15, batch 5900, loss[loss=0.1863, simple_loss=0.2544, pruned_loss=0.05912, over 7675.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2953, pruned_loss=0.06899, over 1613451.99 frames. ], batch size: 18, lr: 5.04e-03, grad_scale: 8.0 +2023-02-06 17:08:15,366 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.494e+02 2.486e+02 2.938e+02 3.942e+02 7.909e+02, threshold=5.877e+02, percent-clipped=6.0 +2023-02-06 17:08:30,151 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119087.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:08:34,143 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119093.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:08:44,779 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119108.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:08:46,723 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119111.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:08:47,874 INFO [train.py:901] (3/4) Epoch 15, batch 5950, loss[loss=0.1886, simple_loss=0.2746, pruned_loss=0.05127, over 8237.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.2967, pruned_loss=0.06982, over 1613335.42 frames. ], batch size: 22, lr: 5.04e-03, grad_scale: 8.0 +2023-02-06 17:09:22,903 INFO [train.py:901] (3/4) Epoch 15, batch 6000, loss[loss=0.2518, simple_loss=0.3309, pruned_loss=0.08637, over 8027.00 frames. ], tot_loss[loss=0.2197, simple_loss=0.2982, pruned_loss=0.07059, over 1616132.51 frames. ], batch size: 22, lr: 5.04e-03, grad_scale: 8.0 +2023-02-06 17:09:22,904 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 17:09:35,676 INFO [train.py:935] (3/4) Epoch 15, validation: loss=0.181, simple_loss=0.2808, pruned_loss=0.04056, over 944034.00 frames. +2023-02-06 17:09:35,677 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 17:09:37,095 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.469e+02 2.578e+02 3.120e+02 3.956e+02 1.218e+03, threshold=6.240e+02, percent-clipped=5.0 +2023-02-06 17:10:05,848 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7048, 2.0120, 2.1307, 1.3582, 2.2088, 1.5077, 0.6823, 1.9988], + device='cuda:3'), covar=tensor([0.0452, 0.0245, 0.0194, 0.0456, 0.0302, 0.0647, 0.0663, 0.0229], + device='cuda:3'), in_proj_covar=tensor([0.0417, 0.0359, 0.0309, 0.0415, 0.0344, 0.0502, 0.0373, 0.0383], + device='cuda:3'), out_proj_covar=tensor([1.1503e-04, 9.6344e-05, 8.2818e-05, 1.1194e-04, 9.3071e-05, 1.4615e-04, + 1.0260e-04, 1.0407e-04], device='cuda:3') +2023-02-06 17:10:10,481 INFO [train.py:901] (3/4) Epoch 15, batch 6050, loss[loss=0.2087, simple_loss=0.2853, pruned_loss=0.06603, over 7972.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2977, pruned_loss=0.071, over 1611578.35 frames. ], batch size: 21, lr: 5.04e-03, grad_scale: 8.0 +2023-02-06 17:10:30,807 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3125, 2.6059, 2.2356, 3.3410, 1.8787, 2.0586, 2.4669, 2.8539], + device='cuda:3'), covar=tensor([0.0725, 0.0808, 0.0776, 0.0422, 0.0980, 0.1183, 0.0912, 0.0702], + device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0203, 0.0248, 0.0211, 0.0211, 0.0249, 0.0254, 0.0213], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 17:10:44,318 INFO [train.py:901] (3/4) Epoch 15, batch 6100, loss[loss=0.2447, simple_loss=0.3257, pruned_loss=0.08189, over 8543.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.2984, pruned_loss=0.0715, over 1610015.33 frames. ], batch size: 39, lr: 5.04e-03, grad_scale: 8.0 +2023-02-06 17:10:45,657 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.461e+02 2.463e+02 3.114e+02 4.132e+02 8.492e+02, threshold=6.229e+02, percent-clipped=7.0 +2023-02-06 17:11:03,738 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 +2023-02-06 17:11:13,015 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.50 vs. limit=5.0 +2023-02-06 17:11:18,268 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 17:11:20,322 INFO [train.py:901] (3/4) Epoch 15, batch 6150, loss[loss=0.2392, simple_loss=0.3278, pruned_loss=0.07526, over 8457.00 frames. ], tot_loss[loss=0.2218, simple_loss=0.2994, pruned_loss=0.07207, over 1610831.38 frames. ], batch size: 25, lr: 5.04e-03, grad_scale: 8.0 +2023-02-06 17:11:54,707 INFO [train.py:901] (3/4) Epoch 15, batch 6200, loss[loss=0.3297, simple_loss=0.3721, pruned_loss=0.1436, over 7100.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.2997, pruned_loss=0.07258, over 1609740.57 frames. ], batch size: 72, lr: 5.04e-03, grad_scale: 8.0 +2023-02-06 17:11:55,643 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119364.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:11:56,081 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.631e+02 2.347e+02 3.204e+02 3.871e+02 7.576e+02, threshold=6.408e+02, percent-clipped=2.0 +2023-02-06 17:12:14,478 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119389.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:12:30,388 INFO [train.py:901] (3/4) Epoch 15, batch 6250, loss[loss=0.2563, simple_loss=0.334, pruned_loss=0.08927, over 8498.00 frames. ], tot_loss[loss=0.2206, simple_loss=0.298, pruned_loss=0.07163, over 1609946.04 frames. ], batch size: 26, lr: 5.04e-03, grad_scale: 8.0 +2023-02-06 17:12:47,168 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=119437.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:12:59,471 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=119455.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:13:04,852 INFO [train.py:901] (3/4) Epoch 15, batch 6300, loss[loss=0.2117, simple_loss=0.2927, pruned_loss=0.06531, over 8018.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.2985, pruned_loss=0.0713, over 1610985.62 frames. ], batch size: 22, lr: 5.04e-03, grad_scale: 8.0 +2023-02-06 17:13:06,147 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.720e+02 2.517e+02 3.087e+02 3.932e+02 1.134e+03, threshold=6.173e+02, percent-clipped=3.0 +2023-02-06 17:13:41,045 INFO [train.py:901] (3/4) Epoch 15, batch 6350, loss[loss=0.2525, simple_loss=0.3399, pruned_loss=0.08255, over 8510.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.2996, pruned_loss=0.07167, over 1612396.46 frames. ], batch size: 26, lr: 5.03e-03, grad_scale: 8.0 +2023-02-06 17:13:44,744 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-06 17:13:53,143 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2728, 1.4301, 1.6391, 1.3867, 0.9404, 1.5487, 1.7431, 1.9128], + device='cuda:3'), covar=tensor([0.0455, 0.1301, 0.1742, 0.1400, 0.0608, 0.1519, 0.0701, 0.0551], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0153, 0.0191, 0.0157, 0.0101, 0.0163, 0.0114, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 17:13:53,793 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119532.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:14:07,839 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119552.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:14:15,091 INFO [train.py:901] (3/4) Epoch 15, batch 6400, loss[loss=0.2237, simple_loss=0.3107, pruned_loss=0.06836, over 8532.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2989, pruned_loss=0.07142, over 1606997.16 frames. ], batch size: 28, lr: 5.03e-03, grad_scale: 8.0 +2023-02-06 17:14:16,455 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.619e+02 2.423e+02 3.023e+02 3.752e+02 7.818e+02, threshold=6.047e+02, percent-clipped=4.0 +2023-02-06 17:14:20,031 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119570.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:14:49,977 INFO [train.py:901] (3/4) Epoch 15, batch 6450, loss[loss=0.1949, simple_loss=0.2891, pruned_loss=0.05031, over 8287.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.2981, pruned_loss=0.07084, over 1607108.15 frames. ], batch size: 23, lr: 5.03e-03, grad_scale: 8.0 +2023-02-06 17:15:24,245 INFO [train.py:901] (3/4) Epoch 15, batch 6500, loss[loss=0.2417, simple_loss=0.3203, pruned_loss=0.08155, over 8323.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2987, pruned_loss=0.07097, over 1610500.84 frames. ], batch size: 25, lr: 5.03e-03, grad_scale: 8.0 +2023-02-06 17:15:25,570 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.619e+02 2.561e+02 2.888e+02 3.578e+02 6.995e+02, threshold=5.776e+02, percent-clipped=4.0 +2023-02-06 17:15:38,583 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119683.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:15:58,717 INFO [train.py:901] (3/4) Epoch 15, batch 6550, loss[loss=0.1989, simple_loss=0.2833, pruned_loss=0.05726, over 8458.00 frames. ], tot_loss[loss=0.2186, simple_loss=0.2975, pruned_loss=0.06984, over 1616243.91 frames. ], batch size: 25, lr: 5.03e-03, grad_scale: 16.0 +2023-02-06 17:16:03,800 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8732, 1.6076, 2.0465, 1.7394, 1.8072, 1.9003, 1.6507, 0.6721], + device='cuda:3'), covar=tensor([0.5235, 0.4461, 0.1651, 0.2941, 0.2410, 0.2562, 0.1882, 0.4618], + device='cuda:3'), in_proj_covar=tensor([0.0912, 0.0920, 0.0758, 0.0884, 0.0955, 0.0843, 0.0717, 0.0795], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 17:16:04,741 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 +2023-02-06 17:16:29,732 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 17:16:34,457 INFO [train.py:901] (3/4) Epoch 15, batch 6600, loss[loss=0.2211, simple_loss=0.2961, pruned_loss=0.07301, over 7941.00 frames. ], tot_loss[loss=0.219, simple_loss=0.2979, pruned_loss=0.07008, over 1614405.32 frames. ], batch size: 20, lr: 5.03e-03, grad_scale: 16.0 +2023-02-06 17:16:35,799 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.430e+02 2.456e+02 2.938e+02 3.854e+02 9.901e+02, threshold=5.877e+02, percent-clipped=5.0 +2023-02-06 17:16:47,891 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 17:17:05,469 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119808.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:17:08,620 INFO [train.py:901] (3/4) Epoch 15, batch 6650, loss[loss=0.1793, simple_loss=0.2616, pruned_loss=0.04851, over 7810.00 frames. ], tot_loss[loss=0.2196, simple_loss=0.2985, pruned_loss=0.07029, over 1614528.26 frames. ], batch size: 19, lr: 5.03e-03, grad_scale: 16.0 +2023-02-06 17:17:12,198 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.6893, 5.7390, 4.9652, 2.5794, 5.0903, 5.4340, 5.2849, 5.1207], + device='cuda:3'), covar=tensor([0.0557, 0.0411, 0.0988, 0.4240, 0.0724, 0.0769, 0.1118, 0.0621], + device='cuda:3'), in_proj_covar=tensor([0.0494, 0.0409, 0.0414, 0.0508, 0.0404, 0.0411, 0.0394, 0.0358], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 17:17:17,680 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119826.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:17:22,300 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119833.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:17:36,352 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119851.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:17:44,400 INFO [train.py:901] (3/4) Epoch 15, batch 6700, loss[loss=0.2292, simple_loss=0.318, pruned_loss=0.07023, over 8393.00 frames. ], tot_loss[loss=0.2204, simple_loss=0.2993, pruned_loss=0.07073, over 1609346.97 frames. ], batch size: 49, lr: 5.03e-03, grad_scale: 16.0 +2023-02-06 17:17:45,754 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.991e+02 2.601e+02 2.951e+02 3.516e+02 8.618e+02, threshold=5.902e+02, percent-clipped=2.0 +2023-02-06 17:17:51,656 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 17:17:53,450 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=119876.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:18:19,555 INFO [train.py:901] (3/4) Epoch 15, batch 6750, loss[loss=0.1685, simple_loss=0.2481, pruned_loss=0.04447, over 7711.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.2981, pruned_loss=0.07024, over 1609828.02 frames. ], batch size: 18, lr: 5.03e-03, grad_scale: 16.0 +2023-02-06 17:18:55,244 INFO [train.py:901] (3/4) Epoch 15, batch 6800, loss[loss=0.2482, simple_loss=0.3253, pruned_loss=0.08554, over 8241.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.2983, pruned_loss=0.07012, over 1611383.60 frames. ], batch size: 24, lr: 5.02e-03, grad_scale: 16.0 +2023-02-06 17:18:57,356 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.711e+02 2.559e+02 3.032e+02 3.835e+02 7.300e+02, threshold=6.064e+02, percent-clipped=2.0 +2023-02-06 17:19:03,569 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 17:19:06,737 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-06 17:19:15,434 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119991.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:19:24,246 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 +2023-02-06 17:19:32,095 INFO [train.py:901] (3/4) Epoch 15, batch 6850, loss[loss=0.2276, simple_loss=0.307, pruned_loss=0.07411, over 7789.00 frames. ], tot_loss[loss=0.218, simple_loss=0.2974, pruned_loss=0.06934, over 1614221.41 frames. ], batch size: 19, lr: 5.02e-03, grad_scale: 16.0 +2023-02-06 17:19:41,651 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=120027.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:19:53,404 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 17:20:01,658 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9746, 2.2164, 1.8292, 2.6922, 1.2061, 1.5716, 1.9173, 2.1928], + device='cuda:3'), covar=tensor([0.0718, 0.0762, 0.0967, 0.0405, 0.1225, 0.1445, 0.0893, 0.0800], + device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0204, 0.0248, 0.0214, 0.0212, 0.0252, 0.0255, 0.0214], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 17:20:02,347 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4857, 1.4123, 1.7690, 1.4197, 1.0656, 1.7849, 0.2089, 1.1747], + device='cuda:3'), covar=tensor([0.2026, 0.1532, 0.0496, 0.1191, 0.3643, 0.0563, 0.2709, 0.1516], + device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0182, 0.0115, 0.0220, 0.0263, 0.0117, 0.0164, 0.0179], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 17:20:06,210 INFO [train.py:901] (3/4) Epoch 15, batch 6900, loss[loss=0.2098, simple_loss=0.2852, pruned_loss=0.06717, over 7698.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2979, pruned_loss=0.06938, over 1614335.24 frames. ], batch size: 18, lr: 5.02e-03, grad_scale: 16.0 +2023-02-06 17:20:07,524 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.766e+02 2.397e+02 2.973e+02 3.506e+02 9.980e+02, threshold=5.947e+02, percent-clipped=2.0 +2023-02-06 17:20:42,259 INFO [train.py:901] (3/4) Epoch 15, batch 6950, loss[loss=0.2405, simple_loss=0.3241, pruned_loss=0.07846, over 8245.00 frames. ], tot_loss[loss=0.219, simple_loss=0.2987, pruned_loss=0.06967, over 1618585.71 frames. ], batch size: 24, lr: 5.02e-03, grad_scale: 16.0 +2023-02-06 17:21:02,421 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=120142.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:21:03,576 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 17:21:16,267 INFO [train.py:901] (3/4) Epoch 15, batch 7000, loss[loss=0.2174, simple_loss=0.3006, pruned_loss=0.06715, over 8252.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2982, pruned_loss=0.06918, over 1616898.11 frames. ], batch size: 24, lr: 5.02e-03, grad_scale: 16.0 +2023-02-06 17:21:17,612 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.611e+02 2.303e+02 2.879e+02 3.620e+02 6.461e+02, threshold=5.757e+02, percent-clipped=3.0 +2023-02-06 17:21:51,887 INFO [train.py:901] (3/4) Epoch 15, batch 7050, loss[loss=0.2269, simple_loss=0.3184, pruned_loss=0.06769, over 8505.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2974, pruned_loss=0.06904, over 1614052.53 frames. ], batch size: 29, lr: 5.02e-03, grad_scale: 16.0 +2023-02-06 17:22:11,813 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-06 17:22:15,017 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=120247.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:22:26,167 INFO [train.py:901] (3/4) Epoch 15, batch 7100, loss[loss=0.2091, simple_loss=0.296, pruned_loss=0.06113, over 8281.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2971, pruned_loss=0.06909, over 1612572.27 frames. ], batch size: 23, lr: 5.02e-03, grad_scale: 16.0 +2023-02-06 17:22:27,488 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.702e+02 2.428e+02 3.078e+02 4.147e+02 9.225e+02, threshold=6.156e+02, percent-clipped=10.0 +2023-02-06 17:22:32,379 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=120272.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:22:57,826 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.13 vs. limit=5.0 +2023-02-06 17:23:00,855 INFO [train.py:901] (3/4) Epoch 15, batch 7150, loss[loss=0.2271, simple_loss=0.3084, pruned_loss=0.07292, over 8318.00 frames. ], tot_loss[loss=0.2186, simple_loss=0.2982, pruned_loss=0.06949, over 1614562.39 frames. ], batch size: 25, lr: 5.02e-03, grad_scale: 16.0 +2023-02-06 17:23:27,589 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8500, 1.5196, 1.8089, 1.3795, 0.8531, 1.6071, 1.5796, 1.4948], + device='cuda:3'), covar=tensor([0.0466, 0.1116, 0.1495, 0.1272, 0.0586, 0.1323, 0.0646, 0.0577], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0190, 0.0156, 0.0100, 0.0162, 0.0114, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 17:23:27,656 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5744, 1.8630, 3.1093, 1.4402, 2.2574, 2.0184, 1.6615, 2.0920], + device='cuda:3'), covar=tensor([0.1741, 0.2373, 0.0659, 0.4049, 0.1539, 0.2846, 0.2011, 0.2219], + device='cuda:3'), in_proj_covar=tensor([0.0499, 0.0552, 0.0537, 0.0603, 0.0626, 0.0565, 0.0495, 0.0619], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 17:23:31,185 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 +2023-02-06 17:23:35,467 INFO [train.py:901] (3/4) Epoch 15, batch 7200, loss[loss=0.2488, simple_loss=0.3187, pruned_loss=0.08942, over 6748.00 frames. ], tot_loss[loss=0.219, simple_loss=0.2984, pruned_loss=0.06981, over 1613109.16 frames. ], batch size: 72, lr: 5.02e-03, grad_scale: 16.0 +2023-02-06 17:23:36,812 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.418e+02 2.853e+02 3.692e+02 6.645e+02, threshold=5.707e+02, percent-clipped=2.0 +2023-02-06 17:24:00,235 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=120398.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:24:10,185 INFO [train.py:901] (3/4) Epoch 15, batch 7250, loss[loss=0.2677, simple_loss=0.3286, pruned_loss=0.1033, over 6960.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2988, pruned_loss=0.0706, over 1611512.88 frames. ], batch size: 72, lr: 5.02e-03, grad_scale: 16.0 +2023-02-06 17:24:17,872 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=120423.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:24:29,231 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5587, 2.0064, 3.2217, 1.3625, 2.3108, 1.9713, 1.6704, 2.3746], + device='cuda:3'), covar=tensor([0.1823, 0.2368, 0.0744, 0.4220, 0.1669, 0.2878, 0.2019, 0.2152], + device='cuda:3'), in_proj_covar=tensor([0.0501, 0.0554, 0.0539, 0.0609, 0.0629, 0.0568, 0.0498, 0.0621], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 17:24:45,982 INFO [train.py:901] (3/4) Epoch 15, batch 7300, loss[loss=0.1937, simple_loss=0.2764, pruned_loss=0.05554, over 8138.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2988, pruned_loss=0.07092, over 1611218.01 frames. ], batch size: 22, lr: 5.01e-03, grad_scale: 16.0 +2023-02-06 17:24:47,343 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.594e+02 2.423e+02 2.925e+02 3.483e+02 5.889e+02, threshold=5.849e+02, percent-clipped=3.0 +2023-02-06 17:25:18,689 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0046, 1.7202, 2.5919, 1.4356, 2.1268, 2.9097, 2.8342, 2.5904], + device='cuda:3'), covar=tensor([0.0793, 0.1282, 0.0665, 0.1726, 0.1594, 0.0249, 0.0746, 0.0476], + device='cuda:3'), in_proj_covar=tensor([0.0280, 0.0306, 0.0272, 0.0301, 0.0285, 0.0248, 0.0379, 0.0298], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 17:25:20,537 INFO [train.py:901] (3/4) Epoch 15, batch 7350, loss[loss=0.211, simple_loss=0.3065, pruned_loss=0.0577, over 8594.00 frames. ], tot_loss[loss=0.2204, simple_loss=0.2989, pruned_loss=0.07093, over 1610991.64 frames. ], batch size: 31, lr: 5.01e-03, grad_scale: 16.0 +2023-02-06 17:25:27,945 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0432, 1.8336, 3.4366, 1.2609, 2.2922, 3.8738, 3.8743, 3.2560], + device='cuda:3'), covar=tensor([0.1137, 0.1527, 0.0374, 0.2332, 0.1061, 0.0225, 0.0514, 0.0603], + device='cuda:3'), in_proj_covar=tensor([0.0280, 0.0306, 0.0271, 0.0301, 0.0285, 0.0248, 0.0379, 0.0298], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 17:25:45,380 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 17:25:56,247 INFO [train.py:901] (3/4) Epoch 15, batch 7400, loss[loss=0.2447, simple_loss=0.3158, pruned_loss=0.08678, over 8598.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.299, pruned_loss=0.07077, over 1610029.01 frames. ], batch size: 31, lr: 5.01e-03, grad_scale: 16.0 +2023-02-06 17:25:57,541 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.598e+02 2.487e+02 3.190e+02 4.160e+02 9.613e+02, threshold=6.380e+02, percent-clipped=9.0 +2023-02-06 17:26:04,636 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 17:26:30,894 INFO [train.py:901] (3/4) Epoch 15, batch 7450, loss[loss=0.2717, simple_loss=0.3425, pruned_loss=0.1005, over 8505.00 frames. ], tot_loss[loss=0.2202, simple_loss=0.2989, pruned_loss=0.07076, over 1608653.93 frames. ], batch size: 26, lr: 5.01e-03, grad_scale: 16.0 +2023-02-06 17:26:42,796 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 17:26:47,953 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.26 vs. limit=5.0 +2023-02-06 17:27:06,390 INFO [train.py:901] (3/4) Epoch 15, batch 7500, loss[loss=0.2002, simple_loss=0.291, pruned_loss=0.05474, over 8474.00 frames. ], tot_loss[loss=0.2206, simple_loss=0.2994, pruned_loss=0.07092, over 1613187.49 frames. ], batch size: 25, lr: 5.01e-03, grad_scale: 16.0 +2023-02-06 17:27:07,747 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.835e+02 2.388e+02 2.853e+02 3.831e+02 7.536e+02, threshold=5.707e+02, percent-clipped=4.0 +2023-02-06 17:27:27,368 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=120694.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:27:40,252 INFO [train.py:901] (3/4) Epoch 15, batch 7550, loss[loss=0.2088, simple_loss=0.2976, pruned_loss=0.06006, over 8467.00 frames. ], tot_loss[loss=0.2218, simple_loss=0.3002, pruned_loss=0.0717, over 1609858.19 frames. ], batch size: 25, lr: 5.01e-03, grad_scale: 16.0 +2023-02-06 17:27:40,386 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6138, 4.6149, 4.0836, 2.0223, 4.1019, 4.1562, 4.2351, 4.0075], + device='cuda:3'), covar=tensor([0.0793, 0.0579, 0.1216, 0.4346, 0.1003, 0.0857, 0.1338, 0.0730], + device='cuda:3'), in_proj_covar=tensor([0.0494, 0.0409, 0.0411, 0.0510, 0.0403, 0.0411, 0.0397, 0.0358], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 17:28:14,828 INFO [train.py:901] (3/4) Epoch 15, batch 7600, loss[loss=0.2252, simple_loss=0.3061, pruned_loss=0.07212, over 8489.00 frames. ], tot_loss[loss=0.2196, simple_loss=0.2979, pruned_loss=0.07064, over 1603914.31 frames. ], batch size: 26, lr: 5.01e-03, grad_scale: 16.0 +2023-02-06 17:28:16,204 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.717e+02 2.426e+02 3.048e+02 3.965e+02 8.844e+02, threshold=6.096e+02, percent-clipped=6.0 +2023-02-06 17:28:50,143 INFO [train.py:901] (3/4) Epoch 15, batch 7650, loss[loss=0.2081, simple_loss=0.2759, pruned_loss=0.07015, over 7190.00 frames. ], tot_loss[loss=0.2206, simple_loss=0.2985, pruned_loss=0.07136, over 1606228.02 frames. ], batch size: 16, lr: 5.01e-03, grad_scale: 8.0 +2023-02-06 17:29:25,340 INFO [train.py:901] (3/4) Epoch 15, batch 7700, loss[loss=0.2433, simple_loss=0.3222, pruned_loss=0.08221, over 8253.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2982, pruned_loss=0.07115, over 1604575.99 frames. ], batch size: 24, lr: 5.01e-03, grad_scale: 8.0 +2023-02-06 17:29:27,393 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.773e+02 2.503e+02 3.087e+02 4.175e+02 9.539e+02, threshold=6.174e+02, percent-clipped=7.0 +2023-02-06 17:29:52,776 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 17:30:01,557 INFO [train.py:901] (3/4) Epoch 15, batch 7750, loss[loss=0.1843, simple_loss=0.2649, pruned_loss=0.05181, over 7536.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.2984, pruned_loss=0.0707, over 1604781.65 frames. ], batch size: 18, lr: 5.01e-03, grad_scale: 8.0 +2023-02-06 17:30:23,664 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-06 17:30:36,085 INFO [train.py:901] (3/4) Epoch 15, batch 7800, loss[loss=0.212, simple_loss=0.2906, pruned_loss=0.06671, over 6740.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.298, pruned_loss=0.07039, over 1606143.24 frames. ], batch size: 71, lr: 5.00e-03, grad_scale: 8.0 +2023-02-06 17:30:38,105 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.815e+02 2.376e+02 2.783e+02 3.266e+02 5.993e+02, threshold=5.565e+02, percent-clipped=0.0 +2023-02-06 17:30:42,285 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6579, 1.2498, 1.4933, 1.1913, 0.8254, 1.3442, 1.4826, 1.2332], + device='cuda:3'), covar=tensor([0.0514, 0.1329, 0.1727, 0.1446, 0.0610, 0.1621, 0.0686, 0.0706], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0190, 0.0156, 0.0099, 0.0161, 0.0112, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 17:31:09,466 INFO [train.py:901] (3/4) Epoch 15, batch 7850, loss[loss=0.2244, simple_loss=0.3125, pruned_loss=0.06815, over 8515.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.2994, pruned_loss=0.07136, over 1611866.99 frames. ], batch size: 28, lr: 5.00e-03, grad_scale: 8.0 +2023-02-06 17:31:14,865 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121021.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:31:26,044 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=121038.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:31:37,296 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8821, 1.5981, 3.3041, 1.2708, 2.2088, 3.6807, 3.9968, 2.7608], + device='cuda:3'), covar=tensor([0.1362, 0.1943, 0.0460, 0.2630, 0.1256, 0.0348, 0.0594, 0.0984], + device='cuda:3'), in_proj_covar=tensor([0.0278, 0.0307, 0.0271, 0.0299, 0.0286, 0.0247, 0.0375, 0.0297], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 17:31:42,563 INFO [train.py:901] (3/4) Epoch 15, batch 7900, loss[loss=0.2419, simple_loss=0.3313, pruned_loss=0.07628, over 8557.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.2995, pruned_loss=0.07155, over 1609617.04 frames. ], batch size: 31, lr: 5.00e-03, grad_scale: 8.0 +2023-02-06 17:31:44,514 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.597e+02 2.419e+02 3.139e+02 4.114e+02 1.036e+03, threshold=6.279e+02, percent-clipped=8.0 +2023-02-06 17:31:49,953 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.90 vs. limit=5.0 +2023-02-06 17:32:15,975 INFO [train.py:901] (3/4) Epoch 15, batch 7950, loss[loss=0.2101, simple_loss=0.3018, pruned_loss=0.05922, over 8264.00 frames. ], tot_loss[loss=0.2214, simple_loss=0.2996, pruned_loss=0.07159, over 1612012.15 frames. ], batch size: 24, lr: 5.00e-03, grad_scale: 8.0 +2023-02-06 17:32:42,024 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121153.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:32:47,156 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.7880, 5.8799, 5.1164, 2.4483, 5.2191, 5.5134, 5.4241, 5.3719], + device='cuda:3'), covar=tensor([0.0473, 0.0349, 0.0849, 0.4255, 0.0579, 0.0755, 0.0978, 0.0564], + device='cuda:3'), in_proj_covar=tensor([0.0491, 0.0404, 0.0409, 0.0513, 0.0401, 0.0407, 0.0391, 0.0357], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 17:32:48,355 INFO [train.py:901] (3/4) Epoch 15, batch 8000, loss[loss=0.2233, simple_loss=0.312, pruned_loss=0.0673, over 8434.00 frames. ], tot_loss[loss=0.2212, simple_loss=0.2992, pruned_loss=0.07159, over 1611413.95 frames. ], batch size: 27, lr: 5.00e-03, grad_scale: 8.0 +2023-02-06 17:32:50,387 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.632e+02 2.460e+02 2.992e+02 3.696e+02 7.694e+02, threshold=5.984e+02, percent-clipped=2.0 +2023-02-06 17:33:01,515 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121182.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:33:22,842 INFO [train.py:901] (3/4) Epoch 15, batch 8050, loss[loss=0.1762, simple_loss=0.2552, pruned_loss=0.04856, over 7436.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2982, pruned_loss=0.0718, over 1599046.49 frames. ], batch size: 17, lr: 5.00e-03, grad_scale: 8.0 +2023-02-06 17:33:55,969 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 17:34:00,924 INFO [train.py:901] (3/4) Epoch 16, batch 0, loss[loss=0.2111, simple_loss=0.2903, pruned_loss=0.06598, over 7934.00 frames. ], tot_loss[loss=0.2111, simple_loss=0.2903, pruned_loss=0.06598, over 7934.00 frames. ], batch size: 20, lr: 4.84e-03, grad_scale: 8.0 +2023-02-06 17:34:00,924 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 17:34:11,044 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5534, 1.6174, 2.7253, 1.2827, 2.0351, 2.9180, 3.0640, 2.4904], + device='cuda:3'), covar=tensor([0.1363, 0.1608, 0.0449, 0.2492, 0.0954, 0.0374, 0.0562, 0.0779], + device='cuda:3'), in_proj_covar=tensor([0.0280, 0.0309, 0.0271, 0.0301, 0.0288, 0.0248, 0.0377, 0.0297], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 17:34:11,910 INFO [train.py:935] (3/4) Epoch 16, validation: loss=0.1795, simple_loss=0.2801, pruned_loss=0.03944, over 944034.00 frames. +2023-02-06 17:34:11,911 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 17:34:17,802 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3515, 1.2192, 4.5560, 1.7789, 3.9919, 3.7995, 4.0643, 3.9755], + device='cuda:3'), covar=tensor([0.0541, 0.5060, 0.0503, 0.3724, 0.1176, 0.1040, 0.0552, 0.0693], + device='cuda:3'), in_proj_covar=tensor([0.0558, 0.0611, 0.0632, 0.0582, 0.0653, 0.0565, 0.0555, 0.0614], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 17:34:24,911 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.891e+02 2.543e+02 3.194e+02 4.084e+02 8.334e+02, threshold=6.389e+02, percent-clipped=7.0 +2023-02-06 17:34:26,238 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 17:34:41,512 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 +2023-02-06 17:34:47,408 INFO [train.py:901] (3/4) Epoch 16, batch 50, loss[loss=0.1723, simple_loss=0.2499, pruned_loss=0.04737, over 7636.00 frames. ], tot_loss[loss=0.2225, simple_loss=0.3007, pruned_loss=0.07218, over 367337.95 frames. ], batch size: 17, lr: 4.84e-03, grad_scale: 8.0 +2023-02-06 17:35:00,275 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 +2023-02-06 17:35:02,267 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 17:35:09,805 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121329.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 17:35:20,725 INFO [train.py:901] (3/4) Epoch 16, batch 100, loss[loss=0.2265, simple_loss=0.308, pruned_loss=0.07249, over 8707.00 frames. ], tot_loss[loss=0.2196, simple_loss=0.2988, pruned_loss=0.07026, over 642660.39 frames. ], batch size: 34, lr: 4.84e-03, grad_scale: 8.0 +2023-02-06 17:35:24,733 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 17:35:33,289 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=121365.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:35:33,868 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.555e+02 2.470e+02 2.913e+02 3.674e+02 6.203e+02, threshold=5.826e+02, percent-clipped=0.0 +2023-02-06 17:35:52,045 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5543, 1.9568, 3.1784, 1.3876, 2.2633, 1.9641, 1.5747, 2.3887], + device='cuda:3'), covar=tensor([0.1875, 0.2464, 0.0827, 0.4150, 0.1838, 0.3051, 0.2194, 0.2086], + device='cuda:3'), in_proj_covar=tensor([0.0508, 0.0560, 0.0542, 0.0614, 0.0637, 0.0580, 0.0507, 0.0627], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 17:35:53,851 INFO [train.py:901] (3/4) Epoch 16, batch 150, loss[loss=0.2394, simple_loss=0.3268, pruned_loss=0.07605, over 8468.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.299, pruned_loss=0.06941, over 857593.24 frames. ], batch size: 25, lr: 4.84e-03, grad_scale: 8.0 +2023-02-06 17:36:00,966 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5444, 1.4724, 1.7692, 1.3293, 1.1252, 1.8033, 0.1125, 1.1261], + device='cuda:3'), covar=tensor([0.1920, 0.1563, 0.0471, 0.1209, 0.3303, 0.0523, 0.2538, 0.1442], + device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0180, 0.0112, 0.0215, 0.0257, 0.0116, 0.0163, 0.0176], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 17:36:04,299 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121409.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:36:15,673 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121425.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:36:21,674 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121434.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:36:30,081 INFO [train.py:901] (3/4) Epoch 16, batch 200, loss[loss=0.227, simple_loss=0.3058, pruned_loss=0.07411, over 8226.00 frames. ], tot_loss[loss=0.2212, simple_loss=0.3009, pruned_loss=0.07071, over 1027932.51 frames. ], batch size: 22, lr: 4.83e-03, grad_scale: 8.0 +2023-02-06 17:36:33,180 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.05 vs. limit=5.0 +2023-02-06 17:36:43,676 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.611e+02 2.385e+02 2.940e+02 3.661e+02 7.455e+02, threshold=5.881e+02, percent-clipped=4.0 +2023-02-06 17:36:53,376 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121480.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:36:56,052 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7507, 1.9571, 1.5966, 2.2661, 1.0091, 1.3918, 1.6588, 1.9508], + device='cuda:3'), covar=tensor([0.0773, 0.0703, 0.1001, 0.0432, 0.1184, 0.1396, 0.0832, 0.0729], + device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0206, 0.0252, 0.0215, 0.0216, 0.0252, 0.0257, 0.0214], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 17:36:59,470 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3172, 2.4370, 1.6009, 1.9283, 2.0117, 1.3868, 1.8784, 1.7277], + device='cuda:3'), covar=tensor([0.1498, 0.0364, 0.1275, 0.0706, 0.0665, 0.1545, 0.0932, 0.1045], + device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0233, 0.0326, 0.0303, 0.0301, 0.0330, 0.0341, 0.0312], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 17:37:04,013 INFO [train.py:901] (3/4) Epoch 16, batch 250, loss[loss=0.2485, simple_loss=0.3332, pruned_loss=0.08185, over 8108.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.3008, pruned_loss=0.07094, over 1160244.94 frames. ], batch size: 23, lr: 4.83e-03, grad_scale: 8.0 +2023-02-06 17:37:07,578 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5990, 2.2039, 4.3672, 1.4660, 2.9332, 2.2986, 1.6450, 2.7851], + device='cuda:3'), covar=tensor([0.1885, 0.2464, 0.0666, 0.4353, 0.1911, 0.3095, 0.2247, 0.2503], + device='cuda:3'), in_proj_covar=tensor([0.0507, 0.0559, 0.0542, 0.0612, 0.0636, 0.0578, 0.0505, 0.0627], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 17:37:18,650 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 17:37:24,828 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=121526.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:37:28,150 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 17:37:39,702 INFO [train.py:901] (3/4) Epoch 16, batch 300, loss[loss=0.2444, simple_loss=0.3168, pruned_loss=0.08596, over 6738.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.2998, pruned_loss=0.07079, over 1257060.05 frames. ], batch size: 72, lr: 4.83e-03, grad_scale: 8.0 +2023-02-06 17:37:54,073 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.651e+02 2.529e+02 3.079e+02 3.820e+02 7.739e+02, threshold=6.158e+02, percent-clipped=5.0 +2023-02-06 17:38:14,124 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7689, 1.8765, 1.6449, 2.2936, 1.0090, 1.4008, 1.6756, 1.9196], + device='cuda:3'), covar=tensor([0.0744, 0.0754, 0.0995, 0.0498, 0.1178, 0.1499, 0.0784, 0.0763], + device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0205, 0.0250, 0.0214, 0.0214, 0.0251, 0.0254, 0.0213], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 17:38:14,588 INFO [train.py:901] (3/4) Epoch 16, batch 350, loss[loss=0.2161, simple_loss=0.2916, pruned_loss=0.07032, over 8137.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.3006, pruned_loss=0.07086, over 1338692.82 frames. ], batch size: 22, lr: 4.83e-03, grad_scale: 8.0 +2023-02-06 17:38:45,978 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121641.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:38:49,789 INFO [train.py:901] (3/4) Epoch 16, batch 400, loss[loss=0.2135, simple_loss=0.2933, pruned_loss=0.0668, over 8496.00 frames. ], tot_loss[loss=0.2227, simple_loss=0.3019, pruned_loss=0.07177, over 1405372.18 frames. ], batch size: 28, lr: 4.83e-03, grad_scale: 8.0 +2023-02-06 17:39:04,295 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.748e+02 2.467e+02 3.087e+02 3.761e+02 6.357e+02, threshold=6.175e+02, percent-clipped=1.0 +2023-02-06 17:39:09,177 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=121673.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 17:39:25,132 INFO [train.py:901] (3/4) Epoch 16, batch 450, loss[loss=0.2007, simple_loss=0.2859, pruned_loss=0.05777, over 8075.00 frames. ], tot_loss[loss=0.2229, simple_loss=0.302, pruned_loss=0.07189, over 1455911.98 frames. ], batch size: 21, lr: 4.83e-03, grad_scale: 8.0 +2023-02-06 17:39:52,426 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121736.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:39:59,043 INFO [train.py:901] (3/4) Epoch 16, batch 500, loss[loss=0.2294, simple_loss=0.3073, pruned_loss=0.07572, over 7648.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.3017, pruned_loss=0.07148, over 1495433.27 frames. ], batch size: 19, lr: 4.83e-03, grad_scale: 8.0 +2023-02-06 17:40:00,111 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.07 vs. limit=5.0 +2023-02-06 17:40:10,963 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121761.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:40:14,769 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.651e+02 2.435e+02 2.838e+02 3.555e+02 6.989e+02, threshold=5.677e+02, percent-clipped=1.0 +2023-02-06 17:40:17,014 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=121769.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:40:29,902 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121788.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 17:40:35,813 INFO [train.py:901] (3/4) Epoch 16, batch 550, loss[loss=0.2087, simple_loss=0.3011, pruned_loss=0.05811, over 8186.00 frames. ], tot_loss[loss=0.2222, simple_loss=0.3012, pruned_loss=0.0716, over 1523519.43 frames. ], batch size: 23, lr: 4.83e-03, grad_scale: 8.0 +2023-02-06 17:41:02,700 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9622, 1.3578, 3.2984, 1.4503, 2.2825, 3.5974, 3.6939, 3.0414], + device='cuda:3'), covar=tensor([0.1066, 0.1846, 0.0336, 0.2091, 0.1043, 0.0233, 0.0453, 0.0625], + device='cuda:3'), in_proj_covar=tensor([0.0281, 0.0310, 0.0273, 0.0301, 0.0292, 0.0248, 0.0381, 0.0298], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 17:41:09,178 INFO [train.py:901] (3/4) Epoch 16, batch 600, loss[loss=0.2248, simple_loss=0.3091, pruned_loss=0.07024, over 8326.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.2999, pruned_loss=0.07058, over 1547135.38 frames. ], batch size: 25, lr: 4.83e-03, grad_scale: 8.0 +2023-02-06 17:41:15,570 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 17:41:22,439 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.450e+02 2.425e+02 3.086e+02 4.175e+02 1.417e+03, threshold=6.173e+02, percent-clipped=9.0 +2023-02-06 17:41:26,596 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 17:41:36,803 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121884.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:41:44,738 INFO [train.py:901] (3/4) Epoch 16, batch 650, loss[loss=0.2138, simple_loss=0.2879, pruned_loss=0.06988, over 8085.00 frames. ], tot_loss[loss=0.2202, simple_loss=0.2997, pruned_loss=0.07034, over 1565996.41 frames. ], batch size: 21, lr: 4.83e-03, grad_scale: 8.0 +2023-02-06 17:41:45,634 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121897.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:41:48,937 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2023-02-06 17:42:02,839 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121922.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:42:18,674 INFO [train.py:901] (3/4) Epoch 16, batch 700, loss[loss=0.1676, simple_loss=0.2503, pruned_loss=0.04247, over 7534.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2982, pruned_loss=0.07006, over 1572237.64 frames. ], batch size: 18, lr: 4.82e-03, grad_scale: 8.0 +2023-02-06 17:42:32,105 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.351e+02 2.496e+02 2.978e+02 3.542e+02 1.118e+03, threshold=5.957e+02, percent-clipped=1.0 +2023-02-06 17:42:33,588 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121968.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:42:41,116 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-06 17:42:53,716 INFO [train.py:901] (3/4) Epoch 16, batch 750, loss[loss=0.1594, simple_loss=0.2401, pruned_loss=0.03938, over 7546.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2969, pruned_loss=0.06916, over 1584304.15 frames. ], batch size: 18, lr: 4.82e-03, grad_scale: 8.0 +2023-02-06 17:43:14,263 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 17:43:14,505 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0867, 1.2457, 1.1871, 0.6881, 1.2433, 1.0392, 0.0861, 1.2010], + device='cuda:3'), covar=tensor([0.0317, 0.0285, 0.0241, 0.0449, 0.0323, 0.0756, 0.0639, 0.0259], + device='cuda:3'), in_proj_covar=tensor([0.0411, 0.0355, 0.0306, 0.0409, 0.0341, 0.0495, 0.0361, 0.0377], + device='cuda:3'), out_proj_covar=tensor([1.1340e-04, 9.4978e-05, 8.1782e-05, 1.1026e-04, 9.1961e-05, 1.4363e-04, + 9.9360e-05, 1.0221e-04], device='cuda:3') +2023-02-06 17:43:23,862 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 17:43:28,671 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122044.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 17:43:29,778 INFO [train.py:901] (3/4) Epoch 16, batch 800, loss[loss=0.2039, simple_loss=0.2751, pruned_loss=0.06635, over 7640.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2965, pruned_loss=0.06907, over 1592096.67 frames. ], batch size: 19, lr: 4.82e-03, grad_scale: 8.0 +2023-02-06 17:43:43,079 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.703e+02 2.422e+02 2.925e+02 3.576e+02 6.712e+02, threshold=5.851e+02, percent-clipped=2.0 +2023-02-06 17:43:45,458 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122069.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 17:44:03,131 INFO [train.py:901] (3/4) Epoch 16, batch 850, loss[loss=0.2097, simple_loss=0.2877, pruned_loss=0.06589, over 8242.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2972, pruned_loss=0.06974, over 1598232.10 frames. ], batch size: 22, lr: 4.82e-03, grad_scale: 8.0 +2023-02-06 17:44:16,336 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7203, 1.6773, 2.3584, 1.5675, 1.2741, 2.3092, 0.3061, 1.2939], + device='cuda:3'), covar=tensor([0.2087, 0.1463, 0.0347, 0.1565, 0.2962, 0.0451, 0.2667, 0.1694], + device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0179, 0.0112, 0.0212, 0.0255, 0.0115, 0.0161, 0.0176], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 17:44:31,619 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122135.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:44:34,942 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122140.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:44:38,760 INFO [train.py:901] (3/4) Epoch 16, batch 900, loss[loss=0.1967, simple_loss=0.283, pruned_loss=0.0552, over 8297.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2961, pruned_loss=0.06915, over 1599208.65 frames. ], batch size: 23, lr: 4.82e-03, grad_scale: 8.0 +2023-02-06 17:44:48,954 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6269, 1.6029, 2.0885, 1.3849, 1.1584, 2.0807, 0.2537, 1.1739], + device='cuda:3'), covar=tensor([0.2261, 0.1551, 0.0420, 0.1635, 0.3412, 0.0474, 0.3006, 0.1640], + device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0179, 0.0112, 0.0212, 0.0255, 0.0115, 0.0161, 0.0176], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 17:44:52,333 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122165.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:44:52,800 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.628e+02 2.482e+02 3.085e+02 4.013e+02 7.148e+02, threshold=6.170e+02, percent-clipped=4.0 +2023-02-06 17:45:01,233 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-06 17:45:12,901 INFO [train.py:901] (3/4) Epoch 16, batch 950, loss[loss=0.2427, simple_loss=0.3174, pruned_loss=0.08394, over 8338.00 frames. ], tot_loss[loss=0.2169, simple_loss=0.2958, pruned_loss=0.06902, over 1600185.85 frames. ], batch size: 25, lr: 4.82e-03, grad_scale: 8.0 +2023-02-06 17:45:24,863 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122213.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:45:40,122 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 17:45:49,028 INFO [train.py:901] (3/4) Epoch 16, batch 1000, loss[loss=0.211, simple_loss=0.2911, pruned_loss=0.06548, over 8516.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2959, pruned_loss=0.06934, over 1602423.71 frames. ], batch size: 28, lr: 4.82e-03, grad_scale: 8.0 +2023-02-06 17:46:03,410 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.620e+02 2.462e+02 3.004e+02 3.600e+02 8.525e+02, threshold=6.009e+02, percent-clipped=4.0 +2023-02-06 17:46:14,166 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 17:46:23,692 INFO [train.py:901] (3/4) Epoch 16, batch 1050, loss[loss=0.1815, simple_loss=0.2529, pruned_loss=0.05507, over 7281.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2953, pruned_loss=0.06903, over 1601356.06 frames. ], batch size: 16, lr: 4.82e-03, grad_scale: 8.0 +2023-02-06 17:46:26,429 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 17:46:34,620 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=122312.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:46:51,293 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122337.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:46:57,871 INFO [train.py:901] (3/4) Epoch 16, batch 1100, loss[loss=0.186, simple_loss=0.2676, pruned_loss=0.05216, over 7977.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2947, pruned_loss=0.06895, over 1604906.80 frames. ], batch size: 21, lr: 4.82e-03, grad_scale: 8.0 +2023-02-06 17:47:12,620 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.699e+02 3.204e+02 3.982e+02 8.590e+02, threshold=6.408e+02, percent-clipped=5.0 +2023-02-06 17:47:33,538 INFO [train.py:901] (3/4) Epoch 16, batch 1150, loss[loss=0.2489, simple_loss=0.3131, pruned_loss=0.09238, over 6818.00 frames. ], tot_loss[loss=0.2162, simple_loss=0.2953, pruned_loss=0.06852, over 1607366.74 frames. ], batch size: 73, lr: 4.82e-03, grad_scale: 8.0 +2023-02-06 17:47:38,316 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 17:47:54,844 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122427.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:48:07,552 INFO [train.py:901] (3/4) Epoch 16, batch 1200, loss[loss=0.1652, simple_loss=0.2475, pruned_loss=0.04145, over 7199.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2956, pruned_loss=0.06852, over 1606571.85 frames. ], batch size: 16, lr: 4.81e-03, grad_scale: 8.0 +2023-02-06 17:48:21,993 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.603e+02 2.417e+02 3.007e+02 3.779e+02 1.089e+03, threshold=6.013e+02, percent-clipped=2.0 +2023-02-06 17:48:31,806 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=122479.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:48:36,695 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122486.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:48:43,455 INFO [train.py:901] (3/4) Epoch 16, batch 1250, loss[loss=0.2335, simple_loss=0.3147, pruned_loss=0.07616, over 8110.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2967, pruned_loss=0.0689, over 1608966.03 frames. ], batch size: 23, lr: 4.81e-03, grad_scale: 8.0 +2023-02-06 17:49:19,089 INFO [train.py:901] (3/4) Epoch 16, batch 1300, loss[loss=0.1873, simple_loss=0.262, pruned_loss=0.05633, over 7697.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2955, pruned_loss=0.06782, over 1612095.02 frames. ], batch size: 18, lr: 4.81e-03, grad_scale: 8.0 +2023-02-06 17:49:26,970 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=122557.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:49:28,693 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-06 17:49:32,897 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5983, 2.2247, 3.2339, 2.5426, 2.9849, 2.4062, 2.0723, 1.9092], + device='cuda:3'), covar=tensor([0.4495, 0.4728, 0.1570, 0.3248, 0.2437, 0.2512, 0.1751, 0.4834], + device='cuda:3'), in_proj_covar=tensor([0.0902, 0.0916, 0.0754, 0.0887, 0.0950, 0.0835, 0.0716, 0.0792], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 17:49:33,313 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.675e+02 2.571e+02 3.105e+02 3.703e+02 6.719e+02, threshold=6.210e+02, percent-clipped=4.0 +2023-02-06 17:49:54,913 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122594.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:49:56,059 INFO [train.py:901] (3/4) Epoch 16, batch 1350, loss[loss=0.249, simple_loss=0.3228, pruned_loss=0.08767, over 7967.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2955, pruned_loss=0.06779, over 1614676.39 frames. ], batch size: 21, lr: 4.81e-03, grad_scale: 8.0 +2023-02-06 17:50:08,293 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4905, 2.7150, 1.8660, 2.2644, 2.0754, 1.4612, 2.0199, 2.0865], + device='cuda:3'), covar=tensor([0.1522, 0.0396, 0.1199, 0.0630, 0.0789, 0.1612, 0.1034, 0.0883], + device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0238, 0.0332, 0.0307, 0.0305, 0.0332, 0.0348, 0.0316], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 17:50:31,465 INFO [train.py:901] (3/4) Epoch 16, batch 1400, loss[loss=0.1777, simple_loss=0.2573, pruned_loss=0.04906, over 7549.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2949, pruned_loss=0.06763, over 1612035.68 frames. ], batch size: 18, lr: 4.81e-03, grad_scale: 8.0 +2023-02-06 17:50:34,470 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4284, 2.0214, 3.2609, 1.2885, 2.3678, 1.8989, 1.5484, 2.2150], + device='cuda:3'), covar=tensor([0.2118, 0.2384, 0.0777, 0.4559, 0.1841, 0.3237, 0.2321, 0.2449], + device='cuda:3'), in_proj_covar=tensor([0.0508, 0.0560, 0.0542, 0.0610, 0.0631, 0.0575, 0.0503, 0.0624], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 17:50:45,932 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.669e+02 2.319e+02 2.799e+02 3.491e+02 7.123e+02, threshold=5.597e+02, percent-clipped=1.0 +2023-02-06 17:50:49,435 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122672.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:50:55,480 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=122681.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:50:56,987 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122683.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:51:05,777 INFO [train.py:901] (3/4) Epoch 16, batch 1450, loss[loss=0.2843, simple_loss=0.3437, pruned_loss=0.1125, over 7233.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2952, pruned_loss=0.0675, over 1614121.24 frames. ], batch size: 72, lr: 4.81e-03, grad_scale: 4.0 +2023-02-06 17:51:12,730 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 17:51:15,655 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122708.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:51:42,667 INFO [train.py:901] (3/4) Epoch 16, batch 1500, loss[loss=0.2447, simple_loss=0.3146, pruned_loss=0.08739, over 7056.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.297, pruned_loss=0.06864, over 1611746.73 frames. ], batch size: 72, lr: 4.81e-03, grad_scale: 4.0 +2023-02-06 17:51:56,875 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.290e+02 2.515e+02 3.024e+02 4.111e+02 8.238e+02, threshold=6.047e+02, percent-clipped=9.0 +2023-02-06 17:52:05,878 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1281, 1.8832, 2.6330, 2.1749, 2.4811, 2.1582, 1.8369, 1.4413], + device='cuda:3'), covar=tensor([0.4468, 0.4014, 0.1375, 0.2856, 0.2013, 0.2387, 0.1744, 0.4107], + device='cuda:3'), in_proj_covar=tensor([0.0909, 0.0923, 0.0758, 0.0893, 0.0952, 0.0840, 0.0720, 0.0796], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 17:52:16,422 INFO [train.py:901] (3/4) Epoch 16, batch 1550, loss[loss=0.2503, simple_loss=0.315, pruned_loss=0.09281, over 7919.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2973, pruned_loss=0.06917, over 1611185.39 frames. ], batch size: 20, lr: 4.81e-03, grad_scale: 4.0 +2023-02-06 17:52:16,636 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122796.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:52:21,764 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 17:52:41,664 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=122830.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:52:41,855 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7446, 2.3904, 3.4862, 2.6722, 3.2087, 2.6483, 2.3344, 1.8374], + device='cuda:3'), covar=tensor([0.4452, 0.4693, 0.1523, 0.3378, 0.2219, 0.2386, 0.1639, 0.5022], + device='cuda:3'), in_proj_covar=tensor([0.0907, 0.0921, 0.0758, 0.0893, 0.0949, 0.0838, 0.0717, 0.0794], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 17:52:52,388 INFO [train.py:901] (3/4) Epoch 16, batch 1600, loss[loss=0.2556, simple_loss=0.3284, pruned_loss=0.09136, over 8395.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.297, pruned_loss=0.06901, over 1612684.40 frames. ], batch size: 49, lr: 4.81e-03, grad_scale: 8.0 +2023-02-06 17:52:55,439 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122850.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:53:07,647 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.387e+02 2.555e+02 3.178e+02 4.067e+02 1.179e+03, threshold=6.355e+02, percent-clipped=12.0 +2023-02-06 17:53:13,371 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122875.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:53:23,677 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122890.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:53:27,661 INFO [train.py:901] (3/4) Epoch 16, batch 1650, loss[loss=0.1958, simple_loss=0.2712, pruned_loss=0.06025, over 7229.00 frames. ], tot_loss[loss=0.2162, simple_loss=0.2962, pruned_loss=0.06807, over 1615295.78 frames. ], batch size: 16, lr: 4.81e-03, grad_scale: 8.0 +2023-02-06 17:53:28,081 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 17:53:28,525 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7921, 1.7045, 1.8369, 1.5236, 1.1766, 1.5281, 2.1011, 2.1835], + device='cuda:3'), covar=tensor([0.0436, 0.1141, 0.1681, 0.1403, 0.0592, 0.1449, 0.0640, 0.0536], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0150, 0.0188, 0.0155, 0.0099, 0.0161, 0.0113, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 17:53:49,582 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122928.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:54:02,349 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122945.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:54:02,852 INFO [train.py:901] (3/4) Epoch 16, batch 1700, loss[loss=0.215, simple_loss=0.2923, pruned_loss=0.0689, over 7788.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2955, pruned_loss=0.0679, over 1609115.23 frames. ], batch size: 19, lr: 4.80e-03, grad_scale: 8.0 +2023-02-06 17:54:08,407 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122953.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:54:11,198 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1822, 2.5791, 3.0070, 1.6398, 3.2052, 1.7112, 1.3960, 1.9816], + device='cuda:3'), covar=tensor([0.0631, 0.0321, 0.0192, 0.0562, 0.0270, 0.0797, 0.0721, 0.0437], + device='cuda:3'), in_proj_covar=tensor([0.0417, 0.0360, 0.0313, 0.0413, 0.0346, 0.0504, 0.0368, 0.0386], + device='cuda:3'), out_proj_covar=tensor([1.1497e-04, 9.6637e-05, 8.3694e-05, 1.1138e-04, 9.3238e-05, 1.4601e-04, + 1.0098e-04, 1.0457e-04], device='cuda:3') +2023-02-06 17:54:17,605 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.498e+02 2.342e+02 2.881e+02 3.479e+02 7.679e+02, threshold=5.763e+02, percent-clipped=3.0 +2023-02-06 17:54:38,077 INFO [train.py:901] (3/4) Epoch 16, batch 1750, loss[loss=0.2455, simple_loss=0.3244, pruned_loss=0.08327, over 8544.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2949, pruned_loss=0.06772, over 1606917.43 frames. ], batch size: 49, lr: 4.80e-03, grad_scale: 8.0 +2023-02-06 17:55:12,116 INFO [train.py:901] (3/4) Epoch 16, batch 1800, loss[loss=0.1739, simple_loss=0.253, pruned_loss=0.04739, over 7444.00 frames. ], tot_loss[loss=0.2169, simple_loss=0.2966, pruned_loss=0.06863, over 1612886.08 frames. ], batch size: 17, lr: 4.80e-03, grad_scale: 8.0 +2023-02-06 17:55:16,362 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123052.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:55:27,703 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.561e+02 2.489e+02 2.922e+02 3.750e+02 7.056e+02, threshold=5.843e+02, percent-clipped=4.0 +2023-02-06 17:55:35,411 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123077.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:55:48,780 INFO [train.py:901] (3/4) Epoch 16, batch 1850, loss[loss=0.2438, simple_loss=0.3161, pruned_loss=0.08571, over 8134.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.2949, pruned_loss=0.06788, over 1612412.10 frames. ], batch size: 22, lr: 4.80e-03, grad_scale: 8.0 +2023-02-06 17:56:22,235 INFO [train.py:901] (3/4) Epoch 16, batch 1900, loss[loss=0.2435, simple_loss=0.3241, pruned_loss=0.08143, over 8360.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2955, pruned_loss=0.068, over 1617268.57 frames. ], batch size: 24, lr: 4.80e-03, grad_scale: 8.0 +2023-02-06 17:56:36,269 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.572e+02 2.569e+02 3.077e+02 4.069e+02 9.708e+02, threshold=6.154e+02, percent-clipped=7.0 +2023-02-06 17:56:57,742 INFO [train.py:901] (3/4) Epoch 16, batch 1950, loss[loss=0.2626, simple_loss=0.3451, pruned_loss=0.09008, over 8334.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2967, pruned_loss=0.06841, over 1622019.40 frames. ], batch size: 26, lr: 4.80e-03, grad_scale: 8.0 +2023-02-06 17:56:59,132 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 17:57:01,388 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123201.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:57:11,338 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 17:57:18,815 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123226.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:57:24,081 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=123234.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:57:30,732 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 17:57:32,102 INFO [train.py:901] (3/4) Epoch 16, batch 2000, loss[loss=0.2468, simple_loss=0.3212, pruned_loss=0.08617, over 8699.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.297, pruned_loss=0.06895, over 1618324.22 frames. ], batch size: 49, lr: 4.80e-03, grad_scale: 8.0 +2023-02-06 17:57:46,349 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.259e+02 2.202e+02 2.631e+02 3.355e+02 6.225e+02, threshold=5.263e+02, percent-clipped=1.0 +2023-02-06 17:58:05,880 INFO [train.py:901] (3/4) Epoch 16, batch 2050, loss[loss=0.1882, simple_loss=0.2717, pruned_loss=0.05235, over 7975.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2961, pruned_loss=0.06844, over 1615426.84 frames. ], batch size: 21, lr: 4.80e-03, grad_scale: 8.0 +2023-02-06 17:58:31,864 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123332.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:58:41,131 INFO [train.py:901] (3/4) Epoch 16, batch 2100, loss[loss=0.2367, simple_loss=0.3196, pruned_loss=0.07691, over 8555.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2964, pruned_loss=0.06831, over 1616799.98 frames. ], batch size: 39, lr: 4.80e-03, grad_scale: 8.0 +2023-02-06 17:58:43,351 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123349.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 17:58:54,983 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.773e+02 2.517e+02 3.000e+02 3.631e+02 1.037e+03, threshold=6.000e+02, percent-clipped=6.0 +2023-02-06 17:59:14,280 INFO [train.py:901] (3/4) Epoch 16, batch 2150, loss[loss=0.2654, simple_loss=0.3331, pruned_loss=0.09879, over 8644.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2975, pruned_loss=0.06912, over 1615663.64 frames. ], batch size: 34, lr: 4.80e-03, grad_scale: 8.0 +2023-02-06 17:59:19,821 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4155, 2.1045, 2.8701, 2.3416, 2.8812, 2.3871, 2.0711, 1.4895], + device='cuda:3'), covar=tensor([0.4623, 0.4486, 0.1530, 0.3299, 0.2048, 0.2612, 0.1825, 0.4845], + device='cuda:3'), in_proj_covar=tensor([0.0920, 0.0928, 0.0763, 0.0900, 0.0964, 0.0846, 0.0720, 0.0803], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 17:59:22,909 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.9811, 6.1516, 5.3109, 2.3863, 5.3429, 5.6790, 5.7180, 5.3766], + device='cuda:3'), covar=tensor([0.0496, 0.0310, 0.0806, 0.4227, 0.0664, 0.0741, 0.0889, 0.0557], + device='cuda:3'), in_proj_covar=tensor([0.0493, 0.0410, 0.0410, 0.0511, 0.0404, 0.0410, 0.0397, 0.0358], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 17:59:50,133 INFO [train.py:901] (3/4) Epoch 16, batch 2200, loss[loss=0.217, simple_loss=0.3046, pruned_loss=0.06465, over 8620.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.299, pruned_loss=0.07026, over 1612725.72 frames. ], batch size: 31, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 17:59:50,273 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123446.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:00:04,144 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.316e+02 2.694e+02 3.295e+02 4.036e+02 1.292e+03, threshold=6.590e+02, percent-clipped=6.0 +2023-02-06 18:00:23,391 INFO [train.py:901] (3/4) Epoch 16, batch 2250, loss[loss=0.2132, simple_loss=0.2961, pruned_loss=0.06509, over 8521.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.298, pruned_loss=0.06932, over 1616787.91 frames. ], batch size: 39, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 18:00:46,373 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-06 18:00:58,616 INFO [train.py:901] (3/4) Epoch 16, batch 2300, loss[loss=0.2053, simple_loss=0.2738, pruned_loss=0.06843, over 7779.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2974, pruned_loss=0.06938, over 1611716.19 frames. ], batch size: 19, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 18:01:13,226 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.552e+02 2.374e+02 2.935e+02 3.719e+02 2.594e+03, threshold=5.871e+02, percent-clipped=2.0 +2023-02-06 18:01:22,769 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5673, 1.9405, 2.0205, 1.3365, 2.1448, 1.5189, 0.5487, 1.8223], + device='cuda:3'), covar=tensor([0.0476, 0.0269, 0.0215, 0.0420, 0.0328, 0.0769, 0.0701, 0.0244], + device='cuda:3'), in_proj_covar=tensor([0.0423, 0.0364, 0.0313, 0.0416, 0.0350, 0.0507, 0.0372, 0.0389], + device='cuda:3'), out_proj_covar=tensor([1.1640e-04, 9.7522e-05, 8.3455e-05, 1.1214e-04, 9.4408e-05, 1.4701e-04, + 1.0201e-04, 1.0529e-04], device='cuda:3') +2023-02-06 18:01:32,632 INFO [train.py:901] (3/4) Epoch 16, batch 2350, loss[loss=0.199, simple_loss=0.2808, pruned_loss=0.05863, over 8470.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2958, pruned_loss=0.06847, over 1614329.27 frames. ], batch size: 29, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 18:01:38,853 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123605.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:01:55,687 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123630.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:02:06,132 INFO [train.py:901] (3/4) Epoch 16, batch 2400, loss[loss=0.2412, simple_loss=0.3116, pruned_loss=0.08533, over 8027.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2955, pruned_loss=0.06831, over 1615557.47 frames. ], batch size: 22, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 18:02:22,333 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.591e+02 2.321e+02 3.011e+02 3.485e+02 7.740e+02, threshold=6.021e+02, percent-clipped=5.0 +2023-02-06 18:02:28,436 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=123676.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:02:42,480 INFO [train.py:901] (3/4) Epoch 16, batch 2450, loss[loss=0.1905, simple_loss=0.2728, pruned_loss=0.05409, over 8248.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.2946, pruned_loss=0.06785, over 1611394.38 frames. ], batch size: 22, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 18:03:15,350 INFO [train.py:901] (3/4) Epoch 16, batch 2500, loss[loss=0.2233, simple_loss=0.3012, pruned_loss=0.07271, over 8196.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2964, pruned_loss=0.06894, over 1612202.54 frames. ], batch size: 23, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 18:03:29,367 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.811e+02 2.388e+02 3.009e+02 3.987e+02 1.163e+03, threshold=6.019e+02, percent-clipped=7.0 +2023-02-06 18:03:46,951 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=123790.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:03:47,744 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123791.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:03:50,919 INFO [train.py:901] (3/4) Epoch 16, batch 2550, loss[loss=0.2164, simple_loss=0.2907, pruned_loss=0.07103, over 8245.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2964, pruned_loss=0.06927, over 1611271.86 frames. ], batch size: 22, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 18:04:24,938 INFO [train.py:901] (3/4) Epoch 16, batch 2600, loss[loss=0.2248, simple_loss=0.2974, pruned_loss=0.07609, over 7345.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2974, pruned_loss=0.06959, over 1616160.04 frames. ], batch size: 71, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 18:04:38,926 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.835e+02 2.447e+02 2.814e+02 3.524e+02 5.517e+02, threshold=5.629e+02, percent-clipped=0.0 +2023-02-06 18:04:54,351 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123890.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:04:58,763 INFO [train.py:901] (3/4) Epoch 16, batch 2650, loss[loss=0.2489, simple_loss=0.3244, pruned_loss=0.08672, over 8356.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.297, pruned_loss=0.06939, over 1615481.72 frames. ], batch size: 24, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 18:05:01,631 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0451, 1.6423, 1.4557, 1.5388, 1.3707, 1.2762, 1.2873, 1.3011], + device='cuda:3'), covar=tensor([0.1054, 0.0410, 0.1145, 0.0520, 0.0734, 0.1386, 0.0895, 0.0775], + device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0233, 0.0328, 0.0304, 0.0302, 0.0334, 0.0348, 0.0313], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 18:05:06,373 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123905.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:05:34,176 INFO [train.py:901] (3/4) Epoch 16, batch 2700, loss[loss=0.2904, simple_loss=0.3462, pruned_loss=0.1173, over 6890.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.2971, pruned_loss=0.06965, over 1615652.87 frames. ], batch size: 71, lr: 4.79e-03, grad_scale: 8.0 +2023-02-06 18:05:48,214 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.461e+02 2.455e+02 3.188e+02 4.135e+02 8.908e+02, threshold=6.377e+02, percent-clipped=7.0 +2023-02-06 18:06:04,519 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4956, 2.6632, 2.2524, 3.7071, 1.6563, 2.0663, 2.2615, 2.9004], + device='cuda:3'), covar=tensor([0.0668, 0.0951, 0.0813, 0.0355, 0.1137, 0.1237, 0.1117, 0.0748], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0203, 0.0250, 0.0211, 0.0211, 0.0249, 0.0254, 0.0213], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 18:06:07,605 INFO [train.py:901] (3/4) Epoch 16, batch 2750, loss[loss=0.1828, simple_loss=0.2541, pruned_loss=0.05569, over 7401.00 frames. ], tot_loss[loss=0.2169, simple_loss=0.2959, pruned_loss=0.06889, over 1615182.53 frames. ], batch size: 17, lr: 4.78e-03, grad_scale: 8.0 +2023-02-06 18:06:24,199 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7251, 1.6874, 2.2521, 1.6049, 1.1798, 2.3056, 0.4050, 1.2268], + device='cuda:3'), covar=tensor([0.2279, 0.1521, 0.0420, 0.1500, 0.3443, 0.0371, 0.2653, 0.1846], + device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0183, 0.0115, 0.0215, 0.0259, 0.0118, 0.0165, 0.0178], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 18:06:45,091 INFO [train.py:901] (3/4) Epoch 16, batch 2800, loss[loss=0.1845, simple_loss=0.2593, pruned_loss=0.05487, over 7931.00 frames. ], tot_loss[loss=0.2162, simple_loss=0.2951, pruned_loss=0.06861, over 1605325.89 frames. ], batch size: 20, lr: 4.78e-03, grad_scale: 8.0 +2023-02-06 18:06:46,015 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124047.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:06:50,843 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124054.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:06:59,429 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.436e+02 2.578e+02 3.039e+02 4.001e+02 1.196e+03, threshold=6.079e+02, percent-clipped=5.0 +2023-02-06 18:06:59,558 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5184, 4.4482, 4.0015, 2.3614, 3.9963, 4.1703, 4.1252, 3.9249], + device='cuda:3'), covar=tensor([0.0652, 0.0563, 0.0967, 0.4182, 0.0745, 0.0872, 0.1145, 0.0762], + device='cuda:3'), in_proj_covar=tensor([0.0494, 0.0409, 0.0415, 0.0513, 0.0405, 0.0411, 0.0401, 0.0358], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 18:07:03,060 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124072.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:07:08,263 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124080.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:07:18,991 INFO [train.py:901] (3/4) Epoch 16, batch 2850, loss[loss=0.1709, simple_loss=0.2528, pruned_loss=0.04454, over 7807.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2954, pruned_loss=0.06872, over 1605552.89 frames. ], batch size: 20, lr: 4.78e-03, grad_scale: 8.0 +2023-02-06 18:07:55,339 INFO [train.py:901] (3/4) Epoch 16, batch 2900, loss[loss=0.2196, simple_loss=0.2925, pruned_loss=0.0733, over 8083.00 frames. ], tot_loss[loss=0.2162, simple_loss=0.2955, pruned_loss=0.06846, over 1609150.47 frames. ], batch size: 21, lr: 4.78e-03, grad_scale: 8.0 +2023-02-06 18:08:06,271 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124161.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:08:07,977 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 +2023-02-06 18:08:08,495 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.07 vs. limit=5.0 +2023-02-06 18:08:10,038 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.642e+02 2.456e+02 3.206e+02 4.387e+02 8.191e+02, threshold=6.412e+02, percent-clipped=4.0 +2023-02-06 18:08:22,884 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124186.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:08:29,505 INFO [train.py:901] (3/4) Epoch 16, batch 2950, loss[loss=0.1937, simple_loss=0.2589, pruned_loss=0.06426, over 7199.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2968, pruned_loss=0.06975, over 1610834.59 frames. ], batch size: 16, lr: 4.78e-03, grad_scale: 8.0 +2023-02-06 18:08:35,642 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 18:08:50,767 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0465, 1.5379, 1.6570, 1.3965, 0.9538, 1.4907, 1.6393, 1.5421], + device='cuda:3'), covar=tensor([0.0476, 0.1198, 0.1671, 0.1420, 0.0594, 0.1454, 0.0680, 0.0619], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0151, 0.0189, 0.0157, 0.0100, 0.0162, 0.0113, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 18:08:55,342 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124234.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:09:03,841 INFO [train.py:901] (3/4) Epoch 16, batch 3000, loss[loss=0.2062, simple_loss=0.2886, pruned_loss=0.06188, over 8539.00 frames. ], tot_loss[loss=0.2186, simple_loss=0.2971, pruned_loss=0.07009, over 1609297.92 frames. ], batch size: 49, lr: 4.78e-03, grad_scale: 8.0 +2023-02-06 18:09:03,841 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 18:09:16,269 INFO [train.py:935] (3/4) Epoch 16, validation: loss=0.1794, simple_loss=0.2796, pruned_loss=0.03958, over 944034.00 frames. +2023-02-06 18:09:16,270 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 18:09:32,708 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.508e+02 2.393e+02 2.939e+02 3.627e+02 1.404e+03, threshold=5.877e+02, percent-clipped=2.0 +2023-02-06 18:09:49,110 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124290.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:09:51,099 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0880, 1.9272, 3.1759, 1.3869, 2.2438, 3.3944, 3.4440, 2.8864], + device='cuda:3'), covar=tensor([0.0932, 0.1294, 0.0341, 0.1977, 0.0928, 0.0257, 0.0571, 0.0536], + device='cuda:3'), in_proj_covar=tensor([0.0277, 0.0305, 0.0271, 0.0297, 0.0288, 0.0248, 0.0377, 0.0295], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 18:09:52,920 INFO [train.py:901] (3/4) Epoch 16, batch 3050, loss[loss=0.1786, simple_loss=0.2565, pruned_loss=0.05037, over 7533.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.297, pruned_loss=0.07022, over 1609191.00 frames. ], batch size: 18, lr: 4.78e-03, grad_scale: 8.0 +2023-02-06 18:10:24,200 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9423, 1.6991, 2.0870, 1.8296, 2.0010, 1.9423, 1.7153, 0.7502], + device='cuda:3'), covar=tensor([0.4946, 0.4114, 0.1571, 0.2927, 0.1993, 0.2445, 0.1744, 0.4426], + device='cuda:3'), in_proj_covar=tensor([0.0915, 0.0924, 0.0756, 0.0894, 0.0961, 0.0847, 0.0723, 0.0796], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 18:10:25,956 INFO [train.py:901] (3/4) Epoch 16, batch 3100, loss[loss=0.2203, simple_loss=0.2828, pruned_loss=0.07892, over 7240.00 frames. ], tot_loss[loss=0.2188, simple_loss=0.2974, pruned_loss=0.07015, over 1611656.48 frames. ], batch size: 16, lr: 4.78e-03, grad_scale: 8.0 +2023-02-06 18:10:28,053 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124349.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:10:39,328 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124366.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 18:10:39,801 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.590e+02 2.464e+02 2.975e+02 4.095e+02 1.383e+03, threshold=5.950e+02, percent-clipped=6.0 +2023-02-06 18:11:01,473 INFO [train.py:901] (3/4) Epoch 16, batch 3150, loss[loss=0.2041, simple_loss=0.2841, pruned_loss=0.06207, over 8341.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2963, pruned_loss=0.06947, over 1610665.98 frames. ], batch size: 26, lr: 4.78e-03, grad_scale: 8.0 +2023-02-06 18:11:02,939 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124398.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:11:21,450 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124424.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:11:36,597 INFO [train.py:901] (3/4) Epoch 16, batch 3200, loss[loss=0.2413, simple_loss=0.3201, pruned_loss=0.08125, over 8618.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2967, pruned_loss=0.0699, over 1611397.24 frames. ], batch size: 31, lr: 4.78e-03, grad_scale: 8.0 +2023-02-06 18:11:39,414 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124450.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:11:43,517 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.10 vs. limit=5.0 +2023-02-06 18:11:50,435 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.712e+02 2.552e+02 3.102e+02 3.772e+02 6.284e+02, threshold=6.205e+02, percent-clipped=3.0 +2023-02-06 18:12:07,430 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.0811, 1.7756, 1.9944, 1.7257, 0.8817, 1.7987, 2.3463, 2.1964], + device='cuda:3'), covar=tensor([0.0430, 0.1195, 0.1599, 0.1317, 0.0615, 0.1412, 0.0590, 0.0553], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0151, 0.0189, 0.0157, 0.0100, 0.0162, 0.0113, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 18:12:09,966 INFO [train.py:901] (3/4) Epoch 16, batch 3250, loss[loss=0.2215, simple_loss=0.3003, pruned_loss=0.07137, over 8248.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.2967, pruned_loss=0.06979, over 1613791.65 frames. ], batch size: 24, lr: 4.77e-03, grad_scale: 8.0 +2023-02-06 18:12:23,042 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124513.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:12:40,938 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124539.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:12:45,402 INFO [train.py:901] (3/4) Epoch 16, batch 3300, loss[loss=0.193, simple_loss=0.2731, pruned_loss=0.05649, over 7433.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2956, pruned_loss=0.06865, over 1614496.90 frames. ], batch size: 17, lr: 4.77e-03, grad_scale: 8.0 +2023-02-06 18:12:59,420 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.761e+02 2.425e+02 2.919e+02 3.659e+02 6.879e+02, threshold=5.837e+02, percent-clipped=1.0 +2023-02-06 18:13:18,844 INFO [train.py:901] (3/4) Epoch 16, batch 3350, loss[loss=0.2139, simple_loss=0.2945, pruned_loss=0.06662, over 8766.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2958, pruned_loss=0.06885, over 1613320.91 frames. ], batch size: 40, lr: 4.77e-03, grad_scale: 8.0 +2023-02-06 18:13:23,904 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5625, 1.4887, 4.4647, 1.8164, 2.4828, 5.1599, 5.2045, 4.4719], + device='cuda:3'), covar=tensor([0.1025, 0.1927, 0.0308, 0.2032, 0.1245, 0.0173, 0.0379, 0.0549], + device='cuda:3'), in_proj_covar=tensor([0.0279, 0.0308, 0.0275, 0.0300, 0.0292, 0.0249, 0.0382, 0.0299], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 18:13:25,356 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124605.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:13:32,076 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4170, 2.0678, 2.8788, 2.3051, 2.7743, 2.3419, 1.9973, 1.4785], + device='cuda:3'), covar=tensor([0.4703, 0.4700, 0.1564, 0.3138, 0.2066, 0.2705, 0.1805, 0.4986], + device='cuda:3'), in_proj_covar=tensor([0.0913, 0.0923, 0.0758, 0.0894, 0.0960, 0.0846, 0.0722, 0.0797], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 18:13:43,506 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124630.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:13:46,713 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124634.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:13:54,715 INFO [train.py:901] (3/4) Epoch 16, batch 3400, loss[loss=0.1842, simple_loss=0.2585, pruned_loss=0.05494, over 7704.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2958, pruned_loss=0.06865, over 1614230.49 frames. ], batch size: 18, lr: 4.77e-03, grad_scale: 8.0 +2023-02-06 18:14:01,621 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124656.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:14:08,826 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.629e+02 2.420e+02 3.011e+02 3.525e+02 7.222e+02, threshold=6.022e+02, percent-clipped=3.0 +2023-02-06 18:14:15,310 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3009, 1.2363, 2.3815, 1.1717, 1.9868, 2.5004, 2.6338, 2.1202], + device='cuda:3'), covar=tensor([0.1189, 0.1456, 0.0469, 0.2136, 0.0856, 0.0396, 0.0630, 0.0729], + device='cuda:3'), in_proj_covar=tensor([0.0279, 0.0309, 0.0274, 0.0300, 0.0291, 0.0249, 0.0382, 0.0297], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 18:14:18,584 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124681.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:14:19,314 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7180, 1.6056, 2.8059, 1.2887, 2.0994, 3.0166, 3.1337, 2.5291], + device='cuda:3'), covar=tensor([0.1151, 0.1470, 0.0425, 0.2135, 0.0897, 0.0319, 0.0600, 0.0644], + device='cuda:3'), in_proj_covar=tensor([0.0278, 0.0308, 0.0273, 0.0299, 0.0290, 0.0249, 0.0381, 0.0297], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 18:14:28,873 INFO [train.py:901] (3/4) Epoch 16, batch 3450, loss[loss=0.2493, simple_loss=0.3269, pruned_loss=0.08585, over 8486.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.2973, pruned_loss=0.0696, over 1611790.48 frames. ], batch size: 28, lr: 4.77e-03, grad_scale: 16.0 +2023-02-06 18:14:38,586 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124710.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 18:15:05,353 INFO [train.py:901] (3/4) Epoch 16, batch 3500, loss[loss=0.1857, simple_loss=0.2604, pruned_loss=0.05549, over 7441.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2983, pruned_loss=0.06973, over 1617121.06 frames. ], batch size: 17, lr: 4.77e-03, grad_scale: 16.0 +2023-02-06 18:15:07,651 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124749.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:15:20,546 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.462e+02 2.534e+02 3.082e+02 3.894e+02 7.146e+02, threshold=6.164e+02, percent-clipped=3.0 +2023-02-06 18:15:22,093 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124769.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:15:38,229 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 18:15:38,978 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=124794.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:15:39,090 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124794.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:15:39,728 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124795.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:15:40,182 INFO [train.py:901] (3/4) Epoch 16, batch 3550, loss[loss=0.1922, simple_loss=0.2632, pruned_loss=0.06064, over 7714.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2975, pruned_loss=0.06962, over 1610771.07 frames. ], batch size: 18, lr: 4.77e-03, grad_scale: 16.0 +2023-02-06 18:15:56,879 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124820.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:16:00,137 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124825.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 18:16:08,892 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124838.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:16:14,229 INFO [train.py:901] (3/4) Epoch 16, batch 3600, loss[loss=0.1992, simple_loss=0.2878, pruned_loss=0.05526, over 8584.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2969, pruned_loss=0.0692, over 1609920.79 frames. ], batch size: 39, lr: 4.77e-03, grad_scale: 16.0 +2023-02-06 18:16:18,570 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5878, 1.2718, 4.7640, 1.7670, 4.1870, 3.9795, 4.2845, 4.1440], + device='cuda:3'), covar=tensor([0.0623, 0.5202, 0.0511, 0.4057, 0.1246, 0.1029, 0.0604, 0.0750], + device='cuda:3'), in_proj_covar=tensor([0.0561, 0.0615, 0.0638, 0.0590, 0.0663, 0.0570, 0.0562, 0.0626], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 18:16:30,802 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.656e+02 2.338e+02 2.977e+02 3.463e+02 8.977e+02, threshold=5.954e+02, percent-clipped=2.0 +2023-02-06 18:16:50,929 INFO [train.py:901] (3/4) Epoch 16, batch 3650, loss[loss=0.284, simple_loss=0.3517, pruned_loss=0.1081, over 8675.00 frames. ], tot_loss[loss=0.2186, simple_loss=0.298, pruned_loss=0.06961, over 1612717.79 frames. ], batch size: 39, lr: 4.77e-03, grad_scale: 16.0 +2023-02-06 18:16:59,897 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124909.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:17:24,956 INFO [train.py:901] (3/4) Epoch 16, batch 3700, loss[loss=0.2047, simple_loss=0.2746, pruned_loss=0.06739, over 7203.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.2983, pruned_loss=0.0699, over 1611382.90 frames. ], batch size: 16, lr: 4.77e-03, grad_scale: 16.0 +2023-02-06 18:17:38,859 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 18:17:40,140 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.880e+02 2.643e+02 3.299e+02 4.315e+02 1.525e+03, threshold=6.598e+02, percent-clipped=10.0 +2023-02-06 18:18:01,618 INFO [train.py:901] (3/4) Epoch 16, batch 3750, loss[loss=0.2271, simple_loss=0.3081, pruned_loss=0.07304, over 8330.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2978, pruned_loss=0.06946, over 1613973.21 frames. ], batch size: 25, lr: 4.77e-03, grad_scale: 16.0 +2023-02-06 18:18:04,451 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125000.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:18:07,878 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125005.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:18:18,222 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-06 18:18:20,988 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125025.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:18:24,567 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125030.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:18:35,177 INFO [train.py:901] (3/4) Epoch 16, batch 3800, loss[loss=0.23, simple_loss=0.3102, pruned_loss=0.07489, over 8190.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2975, pruned_loss=0.06954, over 1609724.56 frames. ], batch size: 23, lr: 4.76e-03, grad_scale: 16.0 +2023-02-06 18:18:49,286 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.650e+02 2.284e+02 2.854e+02 3.651e+02 7.015e+02, threshold=5.709e+02, percent-clipped=3.0 +2023-02-06 18:18:54,970 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8690, 1.6999, 2.4847, 1.6210, 1.2718, 2.5083, 0.3763, 1.4624], + device='cuda:3'), covar=tensor([0.1970, 0.1534, 0.0376, 0.1727, 0.3242, 0.0433, 0.2879, 0.1534], + device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0180, 0.0112, 0.0211, 0.0256, 0.0115, 0.0161, 0.0175], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 18:18:58,968 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125081.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 18:19:10,722 INFO [train.py:901] (3/4) Epoch 16, batch 3850, loss[loss=0.2316, simple_loss=0.3172, pruned_loss=0.07298, over 8248.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2974, pruned_loss=0.06919, over 1614903.65 frames. ], batch size: 24, lr: 4.76e-03, grad_scale: 16.0 +2023-02-06 18:19:18,425 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125106.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 18:19:24,394 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125115.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:19:41,021 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125140.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:19:45,073 INFO [train.py:901] (3/4) Epoch 16, batch 3900, loss[loss=0.2366, simple_loss=0.3056, pruned_loss=0.08384, over 8596.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2982, pruned_loss=0.06981, over 1617364.02 frames. ], batch size: 31, lr: 4.76e-03, grad_scale: 16.0 +2023-02-06 18:19:45,089 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 18:19:45,237 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125146.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:19:56,173 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([0.9632, 1.1436, 1.5638, 0.8988, 1.1882, 1.1228, 1.0067, 1.1052], + device='cuda:3'), covar=tensor([0.1222, 0.1634, 0.0602, 0.2869, 0.1174, 0.2116, 0.1498, 0.1614], + device='cuda:3'), in_proj_covar=tensor([0.0506, 0.0562, 0.0544, 0.0614, 0.0633, 0.0574, 0.0502, 0.0621], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 18:19:58,196 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125165.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:19:58,452 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 +2023-02-06 18:19:59,306 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.561e+02 2.507e+02 2.888e+02 3.601e+02 7.393e+02, threshold=5.777e+02, percent-clipped=3.0 +2023-02-06 18:20:09,607 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125182.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:20:14,434 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4192, 4.4085, 3.8779, 1.8553, 3.8834, 4.0086, 4.0566, 3.7005], + device='cuda:3'), covar=tensor([0.0800, 0.0593, 0.1169, 0.5230, 0.0938, 0.0860, 0.1206, 0.0934], + device='cuda:3'), in_proj_covar=tensor([0.0492, 0.0410, 0.0409, 0.0514, 0.0402, 0.0410, 0.0398, 0.0357], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 18:20:15,221 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125190.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:20:19,100 INFO [train.py:901] (3/4) Epoch 16, batch 3950, loss[loss=0.2128, simple_loss=0.287, pruned_loss=0.06924, over 7695.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2975, pruned_loss=0.06958, over 1611532.66 frames. ], batch size: 18, lr: 4.76e-03, grad_scale: 16.0 +2023-02-06 18:20:20,129 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.19 vs. limit=5.0 +2023-02-06 18:20:35,589 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.28 vs. limit=5.0 +2023-02-06 18:20:55,580 INFO [train.py:901] (3/4) Epoch 16, batch 4000, loss[loss=0.1904, simple_loss=0.2744, pruned_loss=0.05319, over 8140.00 frames. ], tot_loss[loss=0.2186, simple_loss=0.2979, pruned_loss=0.0696, over 1614315.59 frames. ], batch size: 22, lr: 4.76e-03, grad_scale: 16.0 +2023-02-06 18:21:09,913 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.388e+02 2.424e+02 2.747e+02 3.530e+02 7.172e+02, threshold=5.495e+02, percent-clipped=3.0 +2023-02-06 18:21:29,134 INFO [train.py:901] (3/4) Epoch 16, batch 4050, loss[loss=0.1912, simple_loss=0.289, pruned_loss=0.04676, over 8282.00 frames. ], tot_loss[loss=0.219, simple_loss=0.2982, pruned_loss=0.06995, over 1609653.49 frames. ], batch size: 23, lr: 4.76e-03, grad_scale: 16.0 +2023-02-06 18:21:29,965 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125297.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:22:05,148 INFO [train.py:901] (3/4) Epoch 16, batch 4100, loss[loss=0.2394, simple_loss=0.3149, pruned_loss=0.08195, over 8025.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2984, pruned_loss=0.0703, over 1608796.31 frames. ], batch size: 22, lr: 4.76e-03, grad_scale: 16.0 +2023-02-06 18:22:19,363 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.470e+02 2.458e+02 2.941e+02 3.398e+02 7.943e+02, threshold=5.881e+02, percent-clipped=6.0 +2023-02-06 18:22:22,358 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125371.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:22:24,212 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125374.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:22:37,727 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125394.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:22:38,921 INFO [train.py:901] (3/4) Epoch 16, batch 4150, loss[loss=0.2543, simple_loss=0.3403, pruned_loss=0.08415, over 8328.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2998, pruned_loss=0.07096, over 1612297.64 frames. ], batch size: 48, lr: 4.76e-03, grad_scale: 16.0 +2023-02-06 18:22:39,124 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125396.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:22:39,144 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125396.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:22:55,757 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125421.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:23:09,082 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125439.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:23:14,199 INFO [train.py:901] (3/4) Epoch 16, batch 4200, loss[loss=0.2275, simple_loss=0.3014, pruned_loss=0.07679, over 8194.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.298, pruned_loss=0.07013, over 1608237.03 frames. ], batch size: 23, lr: 4.76e-03, grad_scale: 16.0 +2023-02-06 18:23:21,476 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.78 vs. limit=5.0 +2023-02-06 18:23:29,130 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.614e+02 2.508e+02 2.881e+02 3.373e+02 7.881e+02, threshold=5.761e+02, percent-clipped=2.0 +2023-02-06 18:23:39,930 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 18:23:44,591 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125490.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:23:48,646 INFO [train.py:901] (3/4) Epoch 16, batch 4250, loss[loss=0.2038, simple_loss=0.2911, pruned_loss=0.05824, over 8455.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.2982, pruned_loss=0.06997, over 1609419.43 frames. ], batch size: 25, lr: 4.76e-03, grad_scale: 16.0 +2023-02-06 18:23:51,792 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 +2023-02-06 18:24:01,578 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 18:24:23,121 INFO [train.py:901] (3/4) Epoch 16, batch 4300, loss[loss=0.2056, simple_loss=0.276, pruned_loss=0.06756, over 7930.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2987, pruned_loss=0.07019, over 1612124.65 frames. ], batch size: 20, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:24:28,632 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125553.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:24:37,293 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7192, 2.0450, 2.1638, 1.4136, 2.3070, 1.5729, 0.6222, 1.9859], + device='cuda:3'), covar=tensor([0.0514, 0.0274, 0.0214, 0.0462, 0.0262, 0.0691, 0.0756, 0.0216], + device='cuda:3'), in_proj_covar=tensor([0.0428, 0.0361, 0.0311, 0.0418, 0.0353, 0.0511, 0.0372, 0.0390], + device='cuda:3'), out_proj_covar=tensor([1.1756e-04, 9.6357e-05, 8.2811e-05, 1.1225e-04, 9.5326e-05, 1.4791e-04, + 1.0189e-04, 1.0563e-04], device='cuda:3') +2023-02-06 18:24:38,405 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.705e+02 2.516e+02 3.115e+02 4.119e+02 8.810e+02, threshold=6.231e+02, percent-clipped=6.0 +2023-02-06 18:24:46,754 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125578.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:24:58,872 INFO [train.py:901] (3/4) Epoch 16, batch 4350, loss[loss=0.2086, simple_loss=0.2977, pruned_loss=0.05971, over 8016.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2975, pruned_loss=0.06999, over 1603429.88 frames. ], batch size: 22, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:25:04,019 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6396, 1.9773, 2.1231, 1.3207, 2.2770, 1.5240, 0.6736, 1.8565], + device='cuda:3'), covar=tensor([0.0524, 0.0267, 0.0186, 0.0419, 0.0270, 0.0677, 0.0708, 0.0230], + device='cuda:3'), in_proj_covar=tensor([0.0430, 0.0363, 0.0313, 0.0420, 0.0355, 0.0513, 0.0373, 0.0391], + device='cuda:3'), out_proj_covar=tensor([1.1805e-04, 9.6888e-05, 8.3259e-05, 1.1272e-04, 9.5760e-05, 1.4852e-04, + 1.0234e-04, 1.0598e-04], device='cuda:3') +2023-02-06 18:25:05,371 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125605.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:25:11,536 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6217, 1.8850, 1.9990, 1.2165, 2.1517, 1.3751, 0.5412, 1.8028], + device='cuda:3'), covar=tensor([0.0471, 0.0325, 0.0248, 0.0478, 0.0314, 0.0876, 0.0739, 0.0264], + device='cuda:3'), in_proj_covar=tensor([0.0429, 0.0362, 0.0312, 0.0419, 0.0354, 0.0511, 0.0373, 0.0391], + device='cuda:3'), out_proj_covar=tensor([1.1776e-04, 9.6691e-05, 8.3003e-05, 1.1244e-04, 9.5586e-05, 1.4802e-04, + 1.0216e-04, 1.0578e-04], device='cuda:3') +2023-02-06 18:25:16,539 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125621.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:25:33,364 INFO [train.py:901] (3/4) Epoch 16, batch 4400, loss[loss=0.1712, simple_loss=0.2595, pruned_loss=0.04149, over 7805.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2956, pruned_loss=0.06896, over 1601151.40 frames. ], batch size: 19, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:25:34,016 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 18:25:48,652 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.590e+02 2.489e+02 3.156e+02 3.927e+02 6.760e+02, threshold=6.312e+02, percent-clipped=2.0 +2023-02-06 18:25:53,923 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 +2023-02-06 18:26:09,567 INFO [train.py:901] (3/4) Epoch 16, batch 4450, loss[loss=0.1885, simple_loss=0.2791, pruned_loss=0.04899, over 8475.00 frames. ], tot_loss[loss=0.2166, simple_loss=0.2955, pruned_loss=0.06883, over 1601913.54 frames. ], batch size: 25, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:26:14,207 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 18:26:24,219 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125718.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:26:38,220 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125738.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:26:41,764 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125743.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:26:43,639 INFO [train.py:901] (3/4) Epoch 16, batch 4500, loss[loss=0.1767, simple_loss=0.2672, pruned_loss=0.04311, over 8026.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2961, pruned_loss=0.0687, over 1605178.35 frames. ], batch size: 22, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:26:57,806 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.421e+02 2.361e+02 2.740e+02 3.373e+02 6.169e+02, threshold=5.479e+02, percent-clipped=0.0 +2023-02-06 18:27:04,077 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 18:27:10,715 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125783.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:27:15,515 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5302, 1.5138, 1.8285, 1.3190, 1.1538, 1.8142, 0.1883, 1.1363], + device='cuda:3'), covar=tensor([0.2078, 0.1426, 0.0507, 0.1189, 0.3324, 0.0498, 0.2446, 0.1496], + device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0182, 0.0115, 0.0214, 0.0259, 0.0118, 0.0165, 0.0177], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 18:27:19,273 INFO [train.py:901] (3/4) Epoch 16, batch 4550, loss[loss=0.2111, simple_loss=0.2951, pruned_loss=0.0635, over 8464.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2976, pruned_loss=0.06953, over 1609250.31 frames. ], batch size: 25, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:27:45,704 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125833.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:27:54,454 INFO [train.py:901] (3/4) Epoch 16, batch 4600, loss[loss=0.2735, simple_loss=0.3354, pruned_loss=0.1058, over 6626.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.2985, pruned_loss=0.07005, over 1611905.68 frames. ], batch size: 71, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:27:59,460 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125853.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:28:05,128 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125861.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:28:08,976 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.613e+02 2.490e+02 3.040e+02 3.897e+02 1.241e+03, threshold=6.080e+02, percent-clipped=8.0 +2023-02-06 18:28:22,162 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125886.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:28:30,022 INFO [train.py:901] (3/4) Epoch 16, batch 4650, loss[loss=0.2644, simple_loss=0.3468, pruned_loss=0.091, over 8647.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.2997, pruned_loss=0.07065, over 1612044.42 frames. ], batch size: 39, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:28:31,600 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125898.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:28:40,004 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8711, 1.6409, 1.7015, 1.5358, 1.1289, 1.6368, 1.7903, 1.5468], + device='cuda:3'), covar=tensor([0.0571, 0.0963, 0.1344, 0.1154, 0.0613, 0.1138, 0.0702, 0.0548], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0190, 0.0156, 0.0101, 0.0162, 0.0114, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 18:28:59,490 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3157, 2.0089, 2.8019, 2.1678, 2.6523, 2.2432, 1.9690, 1.2988], + device='cuda:3'), covar=tensor([0.4691, 0.4760, 0.1488, 0.3164, 0.2171, 0.2632, 0.1723, 0.5145], + device='cuda:3'), in_proj_covar=tensor([0.0915, 0.0922, 0.0760, 0.0896, 0.0959, 0.0846, 0.0723, 0.0799], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 18:29:06,099 INFO [train.py:901] (3/4) Epoch 16, batch 4700, loss[loss=0.22, simple_loss=0.3154, pruned_loss=0.06223, over 8324.00 frames. ], tot_loss[loss=0.2202, simple_loss=0.2992, pruned_loss=0.07064, over 1607736.58 frames. ], batch size: 25, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:29:18,975 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=125965.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:29:20,230 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.462e+02 2.580e+02 3.138e+02 4.127e+02 1.212e+03, threshold=6.277e+02, percent-clipped=5.0 +2023-02-06 18:29:39,835 INFO [train.py:901] (3/4) Epoch 16, batch 4750, loss[loss=0.1754, simple_loss=0.2579, pruned_loss=0.04648, over 6816.00 frames. ], tot_loss[loss=0.2206, simple_loss=0.2995, pruned_loss=0.07087, over 1607869.52 frames. ], batch size: 15, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:29:55,979 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126016.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:30:11,192 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 18:30:13,722 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 18:30:16,326 INFO [train.py:901] (3/4) Epoch 16, batch 4800, loss[loss=0.2246, simple_loss=0.3044, pruned_loss=0.07242, over 8474.00 frames. ], tot_loss[loss=0.2196, simple_loss=0.2986, pruned_loss=0.07029, over 1612117.65 frames. ], batch size: 25, lr: 4.75e-03, grad_scale: 16.0 +2023-02-06 18:30:21,713 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-06 18:30:31,319 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.747e+02 2.301e+02 2.788e+02 3.330e+02 6.705e+02, threshold=5.575e+02, percent-clipped=2.0 +2023-02-06 18:30:40,407 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126080.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:30:45,002 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=126087.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:30:46,456 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126089.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:30:51,047 INFO [train.py:901] (3/4) Epoch 16, batch 4850, loss[loss=0.1961, simple_loss=0.2724, pruned_loss=0.05985, over 7654.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2969, pruned_loss=0.06902, over 1612599.38 frames. ], batch size: 19, lr: 4.74e-03, grad_scale: 16.0 +2023-02-06 18:30:59,987 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126109.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:31:01,812 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 18:31:03,309 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126114.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:31:19,031 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126134.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:31:26,988 INFO [train.py:901] (3/4) Epoch 16, batch 4900, loss[loss=0.1868, simple_loss=0.2712, pruned_loss=0.05117, over 7418.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2955, pruned_loss=0.06804, over 1612493.39 frames. ], batch size: 17, lr: 4.74e-03, grad_scale: 16.0 +2023-02-06 18:31:32,561 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126154.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:31:41,755 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.384e+02 3.140e+02 3.836e+02 7.587e+02, threshold=6.281e+02, percent-clipped=5.0 +2023-02-06 18:31:49,039 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-06 18:31:50,106 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126179.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:32:01,581 INFO [train.py:901] (3/4) Epoch 16, batch 4950, loss[loss=0.2073, simple_loss=0.2904, pruned_loss=0.06214, over 8498.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2968, pruned_loss=0.06878, over 1616924.99 frames. ], batch size: 26, lr: 4.74e-03, grad_scale: 16.0 +2023-02-06 18:32:06,047 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126202.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:32:35,836 INFO [train.py:901] (3/4) Epoch 16, batch 5000, loss[loss=0.2675, simple_loss=0.3261, pruned_loss=0.1044, over 6983.00 frames. ], tot_loss[loss=0.2169, simple_loss=0.2964, pruned_loss=0.06873, over 1614959.54 frames. ], batch size: 71, lr: 4.74e-03, grad_scale: 16.0 +2023-02-06 18:32:50,295 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.665e+02 2.421e+02 2.802e+02 3.540e+02 7.456e+02, threshold=5.603e+02, percent-clipped=2.0 +2023-02-06 18:33:10,451 INFO [train.py:901] (3/4) Epoch 16, batch 5050, loss[loss=0.1731, simple_loss=0.2495, pruned_loss=0.04837, over 7777.00 frames. ], tot_loss[loss=0.217, simple_loss=0.2963, pruned_loss=0.06882, over 1612634.22 frames. ], batch size: 19, lr: 4.74e-03, grad_scale: 8.0 +2023-02-06 18:33:38,227 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126336.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:33:41,473 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 18:33:44,843 INFO [train.py:901] (3/4) Epoch 16, batch 5100, loss[loss=0.2461, simple_loss=0.3165, pruned_loss=0.08781, over 8359.00 frames. ], tot_loss[loss=0.216, simple_loss=0.2951, pruned_loss=0.06842, over 1610969.45 frames. ], batch size: 24, lr: 4.74e-03, grad_scale: 8.0 +2023-02-06 18:33:55,181 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=126360.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:33:56,645 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126361.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:34:01,121 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.556e+02 2.574e+02 2.967e+02 3.773e+02 8.448e+02, threshold=5.934e+02, percent-clipped=7.0 +2023-02-06 18:34:20,690 INFO [train.py:901] (3/4) Epoch 16, batch 5150, loss[loss=0.2122, simple_loss=0.2873, pruned_loss=0.06856, over 8092.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2955, pruned_loss=0.06837, over 1614226.16 frames. ], batch size: 21, lr: 4.74e-03, grad_scale: 8.0 +2023-02-06 18:34:22,778 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126398.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:34:40,810 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4925, 1.4455, 1.7957, 1.2278, 1.0852, 1.7873, 0.1369, 1.1403], + device='cuda:3'), covar=tensor([0.2144, 0.1487, 0.0475, 0.1364, 0.3319, 0.0454, 0.2652, 0.1556], + device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0181, 0.0113, 0.0213, 0.0259, 0.0117, 0.0164, 0.0178], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 18:34:42,900 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1970, 2.1217, 1.6684, 1.9053, 1.7682, 1.4306, 1.6178, 1.6417], + device='cuda:3'), covar=tensor([0.1323, 0.0454, 0.1182, 0.0482, 0.0678, 0.1553, 0.0893, 0.0878], + device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0233, 0.0326, 0.0301, 0.0300, 0.0332, 0.0344, 0.0315], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 18:34:54,883 INFO [train.py:901] (3/4) Epoch 16, batch 5200, loss[loss=0.2003, simple_loss=0.2789, pruned_loss=0.06086, over 8128.00 frames. ], tot_loss[loss=0.2166, simple_loss=0.2959, pruned_loss=0.06862, over 1616898.90 frames. ], batch size: 22, lr: 4.74e-03, grad_scale: 8.0 +2023-02-06 18:35:03,386 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126458.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:35:10,023 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.581e+02 2.450e+02 2.961e+02 4.009e+02 9.502e+02, threshold=5.923e+02, percent-clipped=8.0 +2023-02-06 18:35:10,207 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126468.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:35:15,119 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126475.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:35:15,314 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.09 vs. limit=5.0 +2023-02-06 18:35:21,834 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126483.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:35:30,964 INFO [train.py:901] (3/4) Epoch 16, batch 5250, loss[loss=0.2925, simple_loss=0.3518, pruned_loss=0.1166, over 8108.00 frames. ], tot_loss[loss=0.2174, simple_loss=0.2969, pruned_loss=0.06899, over 1616589.31 frames. ], batch size: 23, lr: 4.74e-03, grad_scale: 8.0 +2023-02-06 18:35:39,842 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 18:36:05,600 INFO [train.py:901] (3/4) Epoch 16, batch 5300, loss[loss=0.1902, simple_loss=0.2679, pruned_loss=0.05621, over 7563.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2963, pruned_loss=0.06903, over 1614748.95 frames. ], batch size: 18, lr: 4.74e-03, grad_scale: 8.0 +2023-02-06 18:36:20,896 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.503e+02 2.415e+02 2.951e+02 3.953e+02 1.148e+03, threshold=5.902e+02, percent-clipped=4.0 +2023-02-06 18:36:21,856 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5332, 1.9769, 3.2356, 1.3112, 2.3705, 1.8359, 1.7942, 2.1876], + device='cuda:3'), covar=tensor([0.2028, 0.2677, 0.0963, 0.4947, 0.1890, 0.3540, 0.2414, 0.2654], + device='cuda:3'), in_proj_covar=tensor([0.0500, 0.0554, 0.0541, 0.0612, 0.0627, 0.0566, 0.0497, 0.0617], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 18:36:41,560 INFO [train.py:901] (3/4) Epoch 16, batch 5350, loss[loss=0.1856, simple_loss=0.2637, pruned_loss=0.05377, over 7691.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2961, pruned_loss=0.06908, over 1612034.39 frames. ], batch size: 18, lr: 4.73e-03, grad_scale: 8.0 +2023-02-06 18:37:16,921 INFO [train.py:901] (3/4) Epoch 16, batch 5400, loss[loss=0.202, simple_loss=0.2753, pruned_loss=0.06432, over 7810.00 frames. ], tot_loss[loss=0.217, simple_loss=0.2956, pruned_loss=0.06913, over 1604888.48 frames. ], batch size: 20, lr: 4.73e-03, grad_scale: 8.0 +2023-02-06 18:37:24,787 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6766, 2.2975, 3.3842, 2.6001, 3.1937, 2.5307, 2.3010, 1.8122], + device='cuda:3'), covar=tensor([0.4687, 0.5277, 0.1771, 0.3317, 0.2219, 0.2850, 0.1813, 0.5288], + device='cuda:3'), in_proj_covar=tensor([0.0907, 0.0915, 0.0756, 0.0891, 0.0957, 0.0841, 0.0717, 0.0795], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 18:37:32,197 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.767e+02 2.413e+02 2.875e+02 3.758e+02 9.843e+02, threshold=5.751e+02, percent-clipped=6.0 +2023-02-06 18:37:51,440 INFO [train.py:901] (3/4) Epoch 16, batch 5450, loss[loss=0.2168, simple_loss=0.3073, pruned_loss=0.06311, over 8195.00 frames. ], tot_loss[loss=0.2169, simple_loss=0.296, pruned_loss=0.06896, over 1606738.92 frames. ], batch size: 23, lr: 4.73e-03, grad_scale: 8.0 +2023-02-06 18:38:15,727 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2291, 1.9808, 2.6900, 2.2095, 2.6089, 2.2322, 1.9587, 1.2170], + device='cuda:3'), covar=tensor([0.4607, 0.4177, 0.1592, 0.3008, 0.2020, 0.2410, 0.1604, 0.4844], + device='cuda:3'), in_proj_covar=tensor([0.0912, 0.0922, 0.0760, 0.0898, 0.0962, 0.0847, 0.0721, 0.0799], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 18:38:17,608 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126731.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:38:24,946 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=126742.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:38:28,282 INFO [train.py:901] (3/4) Epoch 16, batch 5500, loss[loss=0.2128, simple_loss=0.2728, pruned_loss=0.07635, over 6832.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2967, pruned_loss=0.06877, over 1609564.76 frames. ], batch size: 15, lr: 4.73e-03, grad_scale: 8.0 +2023-02-06 18:38:29,001 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 18:38:35,373 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126756.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:38:44,227 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.501e+02 2.451e+02 2.886e+02 3.496e+02 8.391e+02, threshold=5.772e+02, percent-clipped=4.0 +2023-02-06 18:39:02,241 INFO [train.py:901] (3/4) Epoch 16, batch 5550, loss[loss=0.2649, simple_loss=0.3487, pruned_loss=0.09059, over 8135.00 frames. ], tot_loss[loss=0.2169, simple_loss=0.2969, pruned_loss=0.06846, over 1613845.89 frames. ], batch size: 22, lr: 4.73e-03, grad_scale: 4.0 +2023-02-06 18:39:13,435 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=126812.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:39:30,333 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126834.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:39:38,318 INFO [train.py:901] (3/4) Epoch 16, batch 5600, loss[loss=0.2655, simple_loss=0.3419, pruned_loss=0.09452, over 8030.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2968, pruned_loss=0.06845, over 1616485.66 frames. ], batch size: 22, lr: 4.73e-03, grad_scale: 8.0 +2023-02-06 18:39:45,819 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126857.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:39:54,359 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.492e+02 2.374e+02 2.959e+02 4.088e+02 8.002e+02, threshold=5.917e+02, percent-clipped=4.0 +2023-02-06 18:40:12,816 INFO [train.py:901] (3/4) Epoch 16, batch 5650, loss[loss=0.2586, simple_loss=0.3162, pruned_loss=0.1004, over 7214.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.297, pruned_loss=0.0688, over 1613461.87 frames. ], batch size: 71, lr: 4.73e-03, grad_scale: 8.0 +2023-02-06 18:40:27,756 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-06 18:40:30,279 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.7921, 1.6437, 5.9298, 2.0847, 5.3118, 4.9304, 5.4746, 5.3434], + device='cuda:3'), covar=tensor([0.0457, 0.4577, 0.0406, 0.3647, 0.1000, 0.0858, 0.0494, 0.0507], + device='cuda:3'), in_proj_covar=tensor([0.0554, 0.0610, 0.0631, 0.0581, 0.0658, 0.0562, 0.0553, 0.0618], + device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 18:40:33,398 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 18:40:33,516 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126927.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:40:48,559 INFO [train.py:901] (3/4) Epoch 16, batch 5700, loss[loss=0.2239, simple_loss=0.3024, pruned_loss=0.07269, over 8494.00 frames. ], tot_loss[loss=0.2174, simple_loss=0.2968, pruned_loss=0.06896, over 1616982.99 frames. ], batch size: 26, lr: 4.73e-03, grad_scale: 8.0 +2023-02-06 18:40:52,882 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6508, 2.1287, 1.6337, 2.7517, 1.3087, 1.4477, 1.9937, 2.1387], + device='cuda:3'), covar=tensor([0.1049, 0.0902, 0.1231, 0.0451, 0.1191, 0.1717, 0.0930, 0.0975], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0202, 0.0250, 0.0213, 0.0210, 0.0249, 0.0254, 0.0213], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 18:40:56,307 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6988, 1.9510, 2.1548, 1.3301, 2.2808, 1.4868, 0.6064, 1.8380], + device='cuda:3'), covar=tensor([0.0495, 0.0303, 0.0251, 0.0479, 0.0278, 0.0696, 0.0746, 0.0258], + device='cuda:3'), in_proj_covar=tensor([0.0431, 0.0369, 0.0318, 0.0423, 0.0356, 0.0518, 0.0376, 0.0396], + device='cuda:3'), out_proj_covar=tensor([1.1826e-04, 9.8715e-05, 8.4573e-05, 1.1356e-04, 9.5934e-05, 1.5007e-04, + 1.0287e-04, 1.0721e-04], device='cuda:3') +2023-02-06 18:41:02,489 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 +2023-02-06 18:41:04,181 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.657e+02 2.583e+02 3.205e+02 4.543e+02 7.570e+02, threshold=6.410e+02, percent-clipped=11.0 +2023-02-06 18:41:20,982 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([0.9536, 1.5031, 1.1205, 1.3455, 1.2174, 1.0158, 1.1760, 1.1098], + device='cuda:3'), covar=tensor([0.1121, 0.0497, 0.1356, 0.0618, 0.0832, 0.1634, 0.0992, 0.0870], + device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0235, 0.0328, 0.0305, 0.0302, 0.0337, 0.0347, 0.0317], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 18:41:22,819 INFO [train.py:901] (3/4) Epoch 16, batch 5750, loss[loss=0.2458, simple_loss=0.3185, pruned_loss=0.08652, over 8109.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2967, pruned_loss=0.06878, over 1613988.59 frames. ], batch size: 21, lr: 4.73e-03, grad_scale: 8.0 +2023-02-06 18:41:39,590 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 18:41:56,555 INFO [train.py:901] (3/4) Epoch 16, batch 5800, loss[loss=0.1785, simple_loss=0.2522, pruned_loss=0.05242, over 7788.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2961, pruned_loss=0.06851, over 1618324.94 frames. ], batch size: 19, lr: 4.73e-03, grad_scale: 8.0 +2023-02-06 18:42:14,367 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.430e+02 2.425e+02 2.951e+02 3.537e+02 6.549e+02, threshold=5.902e+02, percent-clipped=1.0 +2023-02-06 18:42:24,910 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 18:42:33,212 INFO [train.py:901] (3/4) Epoch 16, batch 5850, loss[loss=0.15, simple_loss=0.2246, pruned_loss=0.03772, over 6799.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2948, pruned_loss=0.06763, over 1615190.98 frames. ], batch size: 15, lr: 4.73e-03, grad_scale: 8.0 +2023-02-06 18:42:45,194 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127113.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:43:02,053 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127138.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:43:07,130 INFO [train.py:901] (3/4) Epoch 16, batch 5900, loss[loss=0.2196, simple_loss=0.2943, pruned_loss=0.07244, over 7432.00 frames. ], tot_loss[loss=0.216, simple_loss=0.2954, pruned_loss=0.06833, over 1616155.73 frames. ], batch size: 17, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:43:09,359 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5794, 2.3695, 3.4861, 2.2396, 2.9312, 3.8930, 3.7876, 3.5073], + device='cuda:3'), covar=tensor([0.0786, 0.1227, 0.0567, 0.1555, 0.1327, 0.0209, 0.0543, 0.0453], + device='cuda:3'), in_proj_covar=tensor([0.0280, 0.0312, 0.0275, 0.0301, 0.0293, 0.0250, 0.0384, 0.0298], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 18:43:23,002 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.574e+02 2.337e+02 2.920e+02 3.581e+02 1.365e+03, threshold=5.840e+02, percent-clipped=5.0 +2023-02-06 18:43:27,094 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2023-02-06 18:43:30,612 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=127178.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:43:34,126 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127183.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:43:42,722 INFO [train.py:901] (3/4) Epoch 16, batch 5950, loss[loss=0.1887, simple_loss=0.2821, pruned_loss=0.04764, over 8300.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.2954, pruned_loss=0.06765, over 1618478.47 frames. ], batch size: 23, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:43:51,307 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127208.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:44:17,704 INFO [train.py:901] (3/4) Epoch 16, batch 6000, loss[loss=0.2102, simple_loss=0.2823, pruned_loss=0.06901, over 7541.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.2945, pruned_loss=0.06736, over 1614254.81 frames. ], batch size: 18, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:44:17,704 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 18:44:29,966 INFO [train.py:935] (3/4) Epoch 16, validation: loss=0.1793, simple_loss=0.2799, pruned_loss=0.03935, over 944034.00 frames. +2023-02-06 18:44:29,967 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 18:44:44,467 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127267.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:44:45,671 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.507e+02 2.282e+02 2.976e+02 3.659e+02 8.304e+02, threshold=5.951e+02, percent-clipped=2.0 +2023-02-06 18:44:55,051 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8930, 1.5371, 3.4926, 1.3841, 2.2860, 3.8422, 3.8683, 3.3013], + device='cuda:3'), covar=tensor([0.1161, 0.1767, 0.0303, 0.2167, 0.1061, 0.0215, 0.0526, 0.0552], + device='cuda:3'), in_proj_covar=tensor([0.0280, 0.0311, 0.0274, 0.0301, 0.0293, 0.0250, 0.0383, 0.0298], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 18:45:01,839 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127293.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:45:03,665 INFO [train.py:901] (3/4) Epoch 16, batch 6050, loss[loss=0.2272, simple_loss=0.3111, pruned_loss=0.07164, over 8193.00 frames. ], tot_loss[loss=0.2159, simple_loss=0.2958, pruned_loss=0.06802, over 1617537.25 frames. ], batch size: 23, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:45:08,719 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-06 18:45:27,055 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2023-02-06 18:45:39,308 INFO [train.py:901] (3/4) Epoch 16, batch 6100, loss[loss=0.2516, simple_loss=0.3326, pruned_loss=0.08529, over 7268.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2955, pruned_loss=0.06855, over 1614942.55 frames. ], batch size: 72, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:45:46,283 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.75 vs. limit=2.0 +2023-02-06 18:45:55,482 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.542e+02 2.555e+02 2.947e+02 3.627e+02 8.036e+02, threshold=5.895e+02, percent-clipped=1.0 +2023-02-06 18:46:09,096 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 18:46:13,727 INFO [train.py:901] (3/4) Epoch 16, batch 6150, loss[loss=0.216, simple_loss=0.3053, pruned_loss=0.06338, over 8248.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2952, pruned_loss=0.06883, over 1612805.58 frames. ], batch size: 24, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:46:36,722 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6506, 2.0894, 3.3349, 1.4214, 2.4790, 2.0655, 1.7737, 2.4372], + device='cuda:3'), covar=tensor([0.1816, 0.2457, 0.0764, 0.4287, 0.1704, 0.2862, 0.1970, 0.2107], + device='cuda:3'), in_proj_covar=tensor([0.0504, 0.0557, 0.0543, 0.0612, 0.0625, 0.0567, 0.0500, 0.0618], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 18:46:48,825 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127445.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 18:46:49,327 INFO [train.py:901] (3/4) Epoch 16, batch 6200, loss[loss=0.2228, simple_loss=0.2974, pruned_loss=0.07408, over 8017.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2932, pruned_loss=0.06788, over 1611549.76 frames. ], batch size: 22, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:46:59,667 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-06 18:47:04,720 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.651e+02 2.656e+02 3.320e+02 4.256e+02 8.643e+02, threshold=6.639e+02, percent-clipped=4.0 +2023-02-06 18:47:23,437 INFO [train.py:901] (3/4) Epoch 16, batch 6250, loss[loss=0.2612, simple_loss=0.3364, pruned_loss=0.09299, over 7108.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2927, pruned_loss=0.06762, over 1610018.94 frames. ], batch size: 71, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:47:33,309 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-06 18:47:44,502 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9612, 2.3740, 3.5055, 1.9408, 1.6964, 3.3835, 0.6386, 2.1179], + device='cuda:3'), covar=tensor([0.1867, 0.1610, 0.0308, 0.2332, 0.3285, 0.0479, 0.3070, 0.1788], + device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0186, 0.0116, 0.0215, 0.0261, 0.0121, 0.0167, 0.0181], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 18:47:57,839 INFO [train.py:901] (3/4) Epoch 16, batch 6300, loss[loss=0.2011, simple_loss=0.2894, pruned_loss=0.05636, over 8352.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.293, pruned_loss=0.06767, over 1612899.82 frames. ], batch size: 24, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:47:58,588 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127547.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:48:00,675 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127549.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:48:14,536 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.739e+02 2.653e+02 3.258e+02 3.936e+02 6.732e+02, threshold=6.516e+02, percent-clipped=2.0 +2023-02-06 18:48:17,993 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127574.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:48:32,719 INFO [train.py:901] (3/4) Epoch 16, batch 6350, loss[loss=0.2109, simple_loss=0.2846, pruned_loss=0.06857, over 8192.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.2934, pruned_loss=0.06795, over 1607389.45 frames. ], batch size: 23, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:48:43,692 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=127611.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:49:07,014 INFO [train.py:901] (3/4) Epoch 16, batch 6400, loss[loss=0.1893, simple_loss=0.2686, pruned_loss=0.05498, over 7419.00 frames. ], tot_loss[loss=0.2169, simple_loss=0.2955, pruned_loss=0.06913, over 1605880.67 frames. ], batch size: 17, lr: 4.72e-03, grad_scale: 8.0 +2023-02-06 18:49:13,036 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-06 18:49:24,191 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.659e+02 2.402e+02 3.034e+02 3.710e+02 8.847e+02, threshold=6.069e+02, percent-clipped=1.0 +2023-02-06 18:49:32,559 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127680.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:49:43,220 INFO [train.py:901] (3/4) Epoch 16, batch 6450, loss[loss=0.2165, simple_loss=0.2853, pruned_loss=0.07386, over 5574.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.295, pruned_loss=0.06861, over 1607323.20 frames. ], batch size: 12, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:50:04,152 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127726.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:50:17,013 INFO [train.py:901] (3/4) Epoch 16, batch 6500, loss[loss=0.2687, simple_loss=0.3315, pruned_loss=0.1029, over 7978.00 frames. ], tot_loss[loss=0.218, simple_loss=0.2965, pruned_loss=0.06976, over 1609424.75 frames. ], batch size: 21, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:50:32,625 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.462e+02 2.427e+02 3.150e+02 4.006e+02 1.604e+03, threshold=6.301e+02, percent-clipped=4.0 +2023-02-06 18:50:42,621 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 +2023-02-06 18:50:48,410 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=127789.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 18:50:52,950 INFO [train.py:901] (3/4) Epoch 16, batch 6550, loss[loss=0.2285, simple_loss=0.3189, pruned_loss=0.06909, over 8499.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2977, pruned_loss=0.07006, over 1613551.97 frames. ], batch size: 26, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:50:53,787 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127797.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:50:56,242 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 18:51:17,239 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 18:51:27,459 INFO [train.py:901] (3/4) Epoch 16, batch 6600, loss[loss=0.173, simple_loss=0.2407, pruned_loss=0.05264, over 7541.00 frames. ], tot_loss[loss=0.2186, simple_loss=0.2969, pruned_loss=0.07018, over 1611481.19 frames. ], batch size: 18, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:51:36,809 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 18:51:37,277 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 18:51:42,294 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127868.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:51:42,776 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.771e+02 2.405e+02 2.899e+02 3.574e+02 1.034e+03, threshold=5.799e+02, percent-clipped=3.0 +2023-02-06 18:51:57,085 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127890.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:51:57,613 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=127891.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:52:00,755 INFO [train.py:901] (3/4) Epoch 16, batch 6650, loss[loss=0.2394, simple_loss=0.3247, pruned_loss=0.07702, over 8611.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2971, pruned_loss=0.06992, over 1605710.77 frames. ], batch size: 39, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:52:07,586 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127904.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 18:52:15,487 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.1910, 3.1438, 2.9218, 1.7834, 2.8318, 2.8748, 2.8879, 2.6818], + device='cuda:3'), covar=tensor([0.1286, 0.0891, 0.1325, 0.4359, 0.1210, 0.1282, 0.1702, 0.1246], + device='cuda:3'), in_proj_covar=tensor([0.0498, 0.0411, 0.0415, 0.0515, 0.0406, 0.0415, 0.0406, 0.0357], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 18:52:31,598 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5194, 1.3899, 4.7208, 1.8525, 4.1656, 3.8890, 4.2381, 4.1157], + device='cuda:3'), covar=tensor([0.0573, 0.5243, 0.0449, 0.3940, 0.1076, 0.0919, 0.0568, 0.0693], + device='cuda:3'), in_proj_covar=tensor([0.0566, 0.0622, 0.0644, 0.0591, 0.0672, 0.0572, 0.0567, 0.0631], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 18:52:36,161 INFO [train.py:901] (3/4) Epoch 16, batch 6700, loss[loss=0.2196, simple_loss=0.3032, pruned_loss=0.068, over 8253.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2973, pruned_loss=0.06979, over 1610514.43 frames. ], batch size: 24, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:52:52,429 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.717e+02 2.543e+02 2.898e+02 3.564e+02 8.195e+02, threshold=5.796e+02, percent-clipped=3.0 +2023-02-06 18:53:01,213 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127982.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:53:10,442 INFO [train.py:901] (3/4) Epoch 16, batch 6750, loss[loss=0.2708, simple_loss=0.3292, pruned_loss=0.1063, over 7075.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2975, pruned_loss=0.06975, over 1614410.97 frames. ], batch size: 73, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:53:18,490 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128006.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:53:19,184 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128007.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:53:32,893 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128024.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:53:47,637 INFO [train.py:901] (3/4) Epoch 16, batch 6800, loss[loss=0.2226, simple_loss=0.3091, pruned_loss=0.06805, over 8245.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2982, pruned_loss=0.07011, over 1615388.51 frames. ], batch size: 24, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:53:48,774 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-06 18:53:51,065 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 18:54:04,017 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.610e+02 2.604e+02 3.143e+02 4.008e+02 8.483e+02, threshold=6.287e+02, percent-clipped=3.0 +2023-02-06 18:54:22,246 INFO [train.py:901] (3/4) Epoch 16, batch 6850, loss[loss=0.2094, simple_loss=0.2762, pruned_loss=0.07126, over 7798.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2962, pruned_loss=0.06897, over 1611743.78 frames. ], batch size: 19, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:54:37,690 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0 +2023-02-06 18:54:40,698 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 18:54:45,024 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9277, 2.5807, 2.0425, 2.2804, 2.3344, 1.8908, 2.1624, 2.3787], + device='cuda:3'), covar=tensor([0.0946, 0.0272, 0.0768, 0.0465, 0.0472, 0.1026, 0.0674, 0.0696], + device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0231, 0.0323, 0.0297, 0.0299, 0.0330, 0.0341, 0.0311], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 18:54:49,919 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1544, 1.2402, 1.5212, 1.1720, 0.6724, 1.3204, 1.1067, 1.0433], + device='cuda:3'), covar=tensor([0.0555, 0.1264, 0.1696, 0.1441, 0.0598, 0.1535, 0.0712, 0.0682], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0152, 0.0190, 0.0156, 0.0100, 0.0163, 0.0114, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 18:54:53,397 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128139.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:54:54,710 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128141.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:54:58,077 INFO [train.py:901] (3/4) Epoch 16, batch 6900, loss[loss=0.221, simple_loss=0.3073, pruned_loss=0.06735, over 8320.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2971, pruned_loss=0.0693, over 1615853.23 frames. ], batch size: 25, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:55:08,138 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128160.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 18:55:14,244 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.605e+02 3.172e+02 3.868e+02 9.306e+02, threshold=6.344e+02, percent-clipped=5.0 +2023-02-06 18:55:25,866 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128185.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 18:55:32,799 INFO [train.py:901] (3/4) Epoch 16, batch 6950, loss[loss=0.1889, simple_loss=0.2781, pruned_loss=0.04981, over 7939.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.2976, pruned_loss=0.06943, over 1618139.42 frames. ], batch size: 20, lr: 4.71e-03, grad_scale: 8.0 +2023-02-06 18:55:43,520 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128212.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:55:48,028 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 18:55:58,347 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128234.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:56:07,205 INFO [train.py:901] (3/4) Epoch 16, batch 7000, loss[loss=0.2469, simple_loss=0.321, pruned_loss=0.08643, over 8498.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2963, pruned_loss=0.06865, over 1620656.23 frames. ], batch size: 26, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 18:56:08,134 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5408, 1.9953, 3.3828, 1.3428, 2.6539, 2.1129, 1.6388, 2.4693], + device='cuda:3'), covar=tensor([0.1869, 0.2579, 0.0767, 0.4279, 0.1577, 0.2747, 0.2088, 0.2265], + device='cuda:3'), in_proj_covar=tensor([0.0502, 0.0559, 0.0539, 0.0609, 0.0621, 0.0565, 0.0499, 0.0615], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 18:56:15,612 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128256.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:56:19,641 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128262.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:56:24,001 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.530e+02 2.690e+02 3.457e+02 5.056e+02 8.270e+02, threshold=6.915e+02, percent-clipped=6.0 +2023-02-06 18:56:27,837 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.38 vs. limit=5.0 +2023-02-06 18:56:36,191 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128287.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:56:42,581 INFO [train.py:901] (3/4) Epoch 16, batch 7050, loss[loss=0.2403, simple_loss=0.3049, pruned_loss=0.08783, over 7922.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2977, pruned_loss=0.06905, over 1623695.14 frames. ], batch size: 20, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 18:57:03,876 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128327.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:57:11,324 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128338.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:57:16,466 INFO [train.py:901] (3/4) Epoch 16, batch 7100, loss[loss=0.1929, simple_loss=0.2679, pruned_loss=0.05897, over 7656.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2975, pruned_loss=0.0689, over 1623471.01 frames. ], batch size: 19, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 18:57:18,650 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128349.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:57:33,889 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.397e+02 2.456e+02 3.083e+02 3.766e+02 8.441e+02, threshold=6.166e+02, percent-clipped=2.0 +2023-02-06 18:57:52,034 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128395.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:57:52,518 INFO [train.py:901] (3/4) Epoch 16, batch 7150, loss[loss=0.2293, simple_loss=0.3088, pruned_loss=0.07485, over 8548.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2958, pruned_loss=0.0679, over 1620439.35 frames. ], batch size: 49, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 18:58:09,872 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128420.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:58:17,227 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128431.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:58:27,265 INFO [train.py:901] (3/4) Epoch 16, batch 7200, loss[loss=0.1887, simple_loss=0.2703, pruned_loss=0.05351, over 7913.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.296, pruned_loss=0.06752, over 1623578.65 frames. ], batch size: 20, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 18:58:41,650 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-06 18:58:42,535 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.378e+02 2.905e+02 3.370e+02 6.119e+02, threshold=5.810e+02, percent-clipped=0.0 +2023-02-06 18:58:45,615 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4782, 1.9043, 3.2190, 1.3267, 2.3601, 1.9322, 1.6306, 2.3371], + device='cuda:3'), covar=tensor([0.1873, 0.2572, 0.0701, 0.4403, 0.1672, 0.2994, 0.2168, 0.2072], + device='cuda:3'), in_proj_covar=tensor([0.0511, 0.0569, 0.0550, 0.0619, 0.0635, 0.0577, 0.0510, 0.0625], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 18:58:55,486 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2100, 1.3761, 3.3336, 1.0914, 2.9446, 2.7663, 3.0495, 2.9229], + device='cuda:3'), covar=tensor([0.0798, 0.4016, 0.0872, 0.4089, 0.1421, 0.1167, 0.0739, 0.0944], + device='cuda:3'), in_proj_covar=tensor([0.0563, 0.0622, 0.0644, 0.0591, 0.0672, 0.0575, 0.0566, 0.0633], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 18:58:58,272 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0691, 2.3111, 3.3613, 1.8778, 2.8489, 2.3963, 2.1220, 2.6598], + device='cuda:3'), covar=tensor([0.1382, 0.1978, 0.0593, 0.3258, 0.1252, 0.2191, 0.1631, 0.1789], + device='cuda:3'), in_proj_covar=tensor([0.0511, 0.0569, 0.0549, 0.0619, 0.0634, 0.0577, 0.0510, 0.0625], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 18:59:02,791 INFO [train.py:901] (3/4) Epoch 16, batch 7250, loss[loss=0.2197, simple_loss=0.3085, pruned_loss=0.06546, over 8359.00 frames. ], tot_loss[loss=0.2162, simple_loss=0.2969, pruned_loss=0.06776, over 1625304.70 frames. ], batch size: 24, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 18:59:13,779 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128512.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:59:31,344 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128537.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 18:59:37,227 INFO [train.py:901] (3/4) Epoch 16, batch 7300, loss[loss=0.1927, simple_loss=0.2638, pruned_loss=0.06079, over 7436.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2959, pruned_loss=0.06744, over 1621770.26 frames. ], batch size: 17, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 18:59:42,187 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5002, 2.4065, 1.7210, 2.1337, 2.0125, 1.2941, 1.9391, 2.0829], + device='cuda:3'), covar=tensor([0.1092, 0.0338, 0.1054, 0.0527, 0.0610, 0.1481, 0.0834, 0.0686], + device='cuda:3'), in_proj_covar=tensor([0.0344, 0.0230, 0.0322, 0.0298, 0.0298, 0.0327, 0.0339, 0.0311], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 18:59:52,610 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.751e+02 2.470e+02 2.980e+02 3.722e+02 1.252e+03, threshold=5.960e+02, percent-clipped=4.0 +2023-02-06 19:00:02,294 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128583.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:00:12,483 INFO [train.py:901] (3/4) Epoch 16, batch 7350, loss[loss=0.2658, simple_loss=0.3407, pruned_loss=0.09544, over 8366.00 frames. ], tot_loss[loss=0.2147, simple_loss=0.2952, pruned_loss=0.06713, over 1615328.49 frames. ], batch size: 24, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 19:00:19,523 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128605.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:00:21,507 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128608.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:00:31,403 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 19:00:36,137 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128630.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:00:46,278 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7360, 1.6777, 2.6644, 1.3433, 2.1404, 2.8703, 2.9503, 2.4999], + device='cuda:3'), covar=tensor([0.1025, 0.1349, 0.0540, 0.2022, 0.1032, 0.0335, 0.0699, 0.0620], + device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0310, 0.0273, 0.0302, 0.0293, 0.0252, 0.0385, 0.0297], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 19:00:47,434 INFO [train.py:901] (3/4) Epoch 16, batch 7400, loss[loss=0.2316, simple_loss=0.3141, pruned_loss=0.07453, over 8641.00 frames. ], tot_loss[loss=0.2159, simple_loss=0.2962, pruned_loss=0.06774, over 1619465.79 frames. ], batch size: 34, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 19:00:49,525 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 19:01:02,831 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.626e+02 2.610e+02 3.305e+02 3.788e+02 1.058e+03, threshold=6.610e+02, percent-clipped=7.0 +2023-02-06 19:01:11,772 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128682.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:01:21,079 INFO [train.py:901] (3/4) Epoch 16, batch 7450, loss[loss=0.1917, simple_loss=0.2708, pruned_loss=0.05629, over 7652.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2952, pruned_loss=0.06735, over 1618325.39 frames. ], batch size: 19, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 19:01:30,646 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 19:01:47,212 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 +2023-02-06 19:01:56,807 INFO [train.py:901] (3/4) Epoch 16, batch 7500, loss[loss=0.1798, simple_loss=0.2581, pruned_loss=0.05075, over 7817.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.295, pruned_loss=0.06729, over 1621068.86 frames. ], batch size: 20, lr: 4.70e-03, grad_scale: 8.0 +2023-02-06 19:02:13,138 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.212e+02 2.417e+02 2.923e+02 3.614e+02 6.549e+02, threshold=5.847e+02, percent-clipped=0.0 +2023-02-06 19:02:17,186 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=128775.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:02:31,139 INFO [train.py:901] (3/4) Epoch 16, batch 7550, loss[loss=0.2342, simple_loss=0.3086, pruned_loss=0.0799, over 8453.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2949, pruned_loss=0.06752, over 1619967.48 frames. ], batch size: 27, lr: 4.69e-03, grad_scale: 16.0 +2023-02-06 19:02:32,029 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128797.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:02:33,329 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128799.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:03:07,389 INFO [train.py:901] (3/4) Epoch 16, batch 7600, loss[loss=0.2347, simple_loss=0.3215, pruned_loss=0.07391, over 8323.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2964, pruned_loss=0.06817, over 1621163.88 frames. ], batch size: 25, lr: 4.69e-03, grad_scale: 16.0 +2023-02-06 19:03:23,260 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.700e+02 2.439e+02 3.123e+02 4.017e+02 8.994e+02, threshold=6.245e+02, percent-clipped=5.0 +2023-02-06 19:03:31,022 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7818, 1.4109, 1.5416, 1.2889, 0.9417, 1.3846, 1.6470, 1.3446], + device='cuda:3'), covar=tensor([0.0535, 0.1304, 0.1708, 0.1486, 0.0624, 0.1558, 0.0718, 0.0682], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0152, 0.0190, 0.0156, 0.0100, 0.0161, 0.0113, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 19:03:32,443 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2942, 1.5018, 1.3394, 1.8266, 0.7514, 1.1730, 1.3584, 1.4917], + device='cuda:3'), covar=tensor([0.0981, 0.0774, 0.1038, 0.0545, 0.1191, 0.1527, 0.0755, 0.0774], + device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0200, 0.0246, 0.0209, 0.0207, 0.0246, 0.0249, 0.0211], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 19:03:38,585 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128890.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:03:42,464 INFO [train.py:901] (3/4) Epoch 16, batch 7650, loss[loss=0.2033, simple_loss=0.2885, pruned_loss=0.05908, over 7812.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2965, pruned_loss=0.06842, over 1619754.68 frames. ], batch size: 20, lr: 4.69e-03, grad_scale: 16.0 +2023-02-06 19:03:50,518 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128908.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 19:04:17,494 INFO [train.py:901] (3/4) Epoch 16, batch 7700, loss[loss=0.2145, simple_loss=0.2896, pruned_loss=0.06965, over 7421.00 frames. ], tot_loss[loss=0.218, simple_loss=0.2976, pruned_loss=0.06922, over 1620931.61 frames. ], batch size: 17, lr: 4.69e-03, grad_scale: 8.0 +2023-02-06 19:04:34,543 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.676e+02 2.361e+02 3.016e+02 3.880e+02 7.767e+02, threshold=6.032e+02, percent-clipped=3.0 +2023-02-06 19:04:42,146 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 19:04:52,875 INFO [train.py:901] (3/4) Epoch 16, batch 7750, loss[loss=0.1564, simple_loss=0.2441, pruned_loss=0.03436, over 7972.00 frames. ], tot_loss[loss=0.2159, simple_loss=0.2963, pruned_loss=0.06782, over 1624021.11 frames. ], batch size: 21, lr: 4.69e-03, grad_scale: 8.0 +2023-02-06 19:05:10,498 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3455, 1.6917, 1.6238, 0.9709, 1.6462, 1.3152, 0.3040, 1.4997], + device='cuda:3'), covar=tensor([0.0488, 0.0340, 0.0282, 0.0516, 0.0477, 0.0865, 0.0829, 0.0277], + device='cuda:3'), in_proj_covar=tensor([0.0427, 0.0365, 0.0315, 0.0423, 0.0351, 0.0512, 0.0375, 0.0394], + device='cuda:3'), out_proj_covar=tensor([1.1699e-04, 9.7428e-05, 8.3572e-05, 1.1348e-04, 9.4393e-05, 1.4788e-04, + 1.0258e-04, 1.0646e-04], device='cuda:3') +2023-02-06 19:05:13,130 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2179, 1.0926, 1.3023, 1.0648, 0.9966, 1.3317, 0.0475, 0.9198], + device='cuda:3'), covar=tensor([0.2046, 0.1659, 0.0578, 0.1168, 0.3210, 0.0606, 0.2592, 0.1529], + device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0185, 0.0115, 0.0216, 0.0263, 0.0120, 0.0165, 0.0179], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 19:05:26,231 INFO [train.py:901] (3/4) Epoch 16, batch 7800, loss[loss=0.2281, simple_loss=0.3133, pruned_loss=0.07143, over 8302.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2953, pruned_loss=0.06783, over 1620353.39 frames. ], batch size: 23, lr: 4.69e-03, grad_scale: 8.0 +2023-02-06 19:05:31,078 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129053.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:05:34,341 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5014, 1.8932, 2.0097, 1.1100, 2.0695, 1.2916, 0.5681, 1.5855], + device='cuda:3'), covar=tensor([0.0547, 0.0325, 0.0238, 0.0555, 0.0371, 0.0929, 0.0813, 0.0328], + device='cuda:3'), in_proj_covar=tensor([0.0428, 0.0366, 0.0315, 0.0422, 0.0351, 0.0512, 0.0375, 0.0394], + device='cuda:3'), out_proj_covar=tensor([1.1713e-04, 9.7452e-05, 8.3565e-05, 1.1332e-04, 9.4473e-05, 1.4777e-04, + 1.0254e-04, 1.0635e-04], device='cuda:3') +2023-02-06 19:05:41,933 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.656e+02 2.423e+02 2.949e+02 3.975e+02 9.373e+02, threshold=5.898e+02, percent-clipped=5.0 +2023-02-06 19:05:48,217 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129078.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:06:00,661 INFO [train.py:901] (3/4) Epoch 16, batch 7850, loss[loss=0.1861, simple_loss=0.2688, pruned_loss=0.0517, over 8242.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2951, pruned_loss=0.06743, over 1621604.04 frames. ], batch size: 22, lr: 4.69e-03, grad_scale: 8.0 +2023-02-06 19:06:32,322 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=129143.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:06:34,273 INFO [train.py:901] (3/4) Epoch 16, batch 7900, loss[loss=0.2038, simple_loss=0.2772, pruned_loss=0.0652, over 7550.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2961, pruned_loss=0.06831, over 1616396.93 frames. ], batch size: 18, lr: 4.69e-03, grad_scale: 8.0 +2023-02-06 19:06:34,508 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129146.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:06:51,032 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.551e+02 2.289e+02 2.786e+02 3.620e+02 6.776e+02, threshold=5.572e+02, percent-clipped=2.0 +2023-02-06 19:06:51,881 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129171.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:07:03,904 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3900, 2.9278, 2.2936, 4.0516, 1.6338, 1.9502, 2.4587, 3.0239], + device='cuda:3'), covar=tensor([0.0774, 0.0840, 0.0908, 0.0260, 0.1179, 0.1433, 0.1120, 0.0840], + device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0202, 0.0249, 0.0211, 0.0208, 0.0248, 0.0252, 0.0213], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 19:07:08,442 INFO [train.py:901] (3/4) Epoch 16, batch 7950, loss[loss=0.2726, simple_loss=0.3377, pruned_loss=0.1038, over 7073.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.297, pruned_loss=0.06862, over 1614542.76 frames. ], batch size: 71, lr: 4.69e-03, grad_scale: 8.0 +2023-02-06 19:07:12,208 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5870, 1.3814, 1.5855, 1.2860, 0.8637, 1.4103, 1.4846, 1.3593], + device='cuda:3'), covar=tensor([0.0556, 0.1268, 0.1698, 0.1457, 0.0599, 0.1537, 0.0714, 0.0626], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0151, 0.0189, 0.0156, 0.0100, 0.0161, 0.0113, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 19:07:39,192 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7744, 1.9292, 1.7480, 2.3355, 1.0670, 1.4949, 1.6015, 1.9377], + device='cuda:3'), covar=tensor([0.0732, 0.0723, 0.0863, 0.0421, 0.1053, 0.1363, 0.0823, 0.0788], + device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0202, 0.0248, 0.0211, 0.0208, 0.0246, 0.0252, 0.0212], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 19:07:42,911 INFO [train.py:901] (3/4) Epoch 16, batch 8000, loss[loss=0.2155, simple_loss=0.2896, pruned_loss=0.0707, over 7975.00 frames. ], tot_loss[loss=0.217, simple_loss=0.2969, pruned_loss=0.06858, over 1618291.30 frames. ], batch size: 21, lr: 4.69e-03, grad_scale: 8.0 +2023-02-06 19:07:43,887 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4406, 1.3765, 1.7471, 1.2363, 1.2120, 1.7265, 0.3380, 1.1712], + device='cuda:3'), covar=tensor([0.1915, 0.1673, 0.0515, 0.1323, 0.3266, 0.0583, 0.2657, 0.1734], + device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0185, 0.0115, 0.0217, 0.0265, 0.0121, 0.0166, 0.0181], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 19:07:47,179 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=129252.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 19:07:51,191 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129258.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:07:59,068 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.621e+02 2.511e+02 2.964e+02 3.601e+02 8.820e+02, threshold=5.927e+02, percent-clipped=6.0 +2023-02-06 19:08:16,579 INFO [train.py:901] (3/4) Epoch 16, batch 8050, loss[loss=0.2056, simple_loss=0.2875, pruned_loss=0.06187, over 7932.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.297, pruned_loss=0.06976, over 1600879.01 frames. ], batch size: 20, lr: 4.69e-03, grad_scale: 8.0 +2023-02-06 19:08:52,564 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 19:08:56,566 INFO [train.py:901] (3/4) Epoch 17, batch 0, loss[loss=0.211, simple_loss=0.2939, pruned_loss=0.06409, over 8501.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2939, pruned_loss=0.06409, over 8501.00 frames. ], batch size: 26, lr: 4.54e-03, grad_scale: 8.0 +2023-02-06 19:08:56,566 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 19:09:04,437 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5294, 1.8075, 2.6434, 1.3617, 1.9923, 1.7787, 1.6473, 1.9159], + device='cuda:3'), covar=tensor([0.1749, 0.2552, 0.0806, 0.4270, 0.1841, 0.3175, 0.2158, 0.2310], + device='cuda:3'), in_proj_covar=tensor([0.0504, 0.0561, 0.0543, 0.0612, 0.0631, 0.0570, 0.0501, 0.0620], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 19:09:07,559 INFO [train.py:935] (3/4) Epoch 17, validation: loss=0.1792, simple_loss=0.2794, pruned_loss=0.03944, over 944034.00 frames. +2023-02-06 19:09:07,559 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 19:09:19,459 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-06 19:09:21,130 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 19:09:29,749 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3253, 2.0446, 2.8571, 2.2827, 2.6763, 2.2141, 2.0088, 1.4447], + device='cuda:3'), covar=tensor([0.4332, 0.4471, 0.1583, 0.3347, 0.2167, 0.2697, 0.1698, 0.4935], + device='cuda:3'), in_proj_covar=tensor([0.0906, 0.0918, 0.0758, 0.0892, 0.0954, 0.0845, 0.0719, 0.0792], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 19:09:33,860 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129367.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 19:09:35,633 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.554e+02 2.551e+02 3.127e+02 3.678e+02 8.568e+02, threshold=6.254e+02, percent-clipped=4.0 +2023-02-06 19:09:41,817 INFO [train.py:901] (3/4) Epoch 17, batch 50, loss[loss=0.2647, simple_loss=0.3376, pruned_loss=0.09591, over 8539.00 frames. ], tot_loss[loss=0.2235, simple_loss=0.3011, pruned_loss=0.07295, over 366709.10 frames. ], batch size: 31, lr: 4.54e-03, grad_scale: 8.0 +2023-02-06 19:09:54,005 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 19:10:18,436 INFO [train.py:901] (3/4) Epoch 17, batch 100, loss[loss=0.1771, simple_loss=0.2569, pruned_loss=0.04864, over 7538.00 frames. ], tot_loss[loss=0.2214, simple_loss=0.3, pruned_loss=0.0714, over 648442.29 frames. ], batch size: 18, lr: 4.54e-03, grad_scale: 8.0 +2023-02-06 19:10:18,444 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 19:10:19,955 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129431.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 19:10:32,120 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8311, 3.5636, 2.3212, 2.7993, 2.6851, 1.8648, 2.7401, 2.9734], + device='cuda:3'), covar=tensor([0.1829, 0.0373, 0.1272, 0.0782, 0.0834, 0.1751, 0.1169, 0.1340], + device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0235, 0.0328, 0.0301, 0.0299, 0.0335, 0.0342, 0.0315], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 19:10:44,166 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5781, 1.8779, 1.9934, 1.1489, 2.0953, 1.4733, 0.5638, 1.7397], + device='cuda:3'), covar=tensor([0.0510, 0.0300, 0.0222, 0.0537, 0.0378, 0.0804, 0.0753, 0.0297], + device='cuda:3'), in_proj_covar=tensor([0.0425, 0.0365, 0.0315, 0.0422, 0.0348, 0.0513, 0.0374, 0.0390], + device='cuda:3'), out_proj_covar=tensor([1.1630e-04, 9.7486e-05, 8.3561e-05, 1.1323e-04, 9.3624e-05, 1.4825e-04, + 1.0240e-04, 1.0518e-04], device='cuda:3') +2023-02-06 19:10:46,023 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.389e+02 2.489e+02 3.062e+02 3.657e+02 7.822e+02, threshold=6.124e+02, percent-clipped=4.0 +2023-02-06 19:10:52,173 INFO [train.py:901] (3/4) Epoch 17, batch 150, loss[loss=0.2027, simple_loss=0.2833, pruned_loss=0.06104, over 8239.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2977, pruned_loss=0.06981, over 863154.27 frames. ], batch size: 22, lr: 4.54e-03, grad_scale: 8.0 +2023-02-06 19:11:18,275 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129514.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:11:29,036 INFO [train.py:901] (3/4) Epoch 17, batch 200, loss[loss=0.2129, simple_loss=0.2899, pruned_loss=0.06795, over 8246.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2969, pruned_loss=0.06898, over 1034535.66 frames. ], batch size: 22, lr: 4.54e-03, grad_scale: 8.0 +2023-02-06 19:11:36,256 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129539.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:11:57,079 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.609e+02 2.455e+02 2.902e+02 3.926e+02 7.649e+02, threshold=5.804e+02, percent-clipped=5.0 +2023-02-06 19:12:03,435 INFO [train.py:901] (3/4) Epoch 17, batch 250, loss[loss=0.2356, simple_loss=0.2987, pruned_loss=0.08622, over 7543.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2968, pruned_loss=0.06866, over 1160498.26 frames. ], batch size: 18, lr: 4.54e-03, grad_scale: 8.0 +2023-02-06 19:12:09,641 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 19:12:11,839 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0009, 1.6155, 1.3565, 1.5270, 1.3683, 1.2118, 1.1997, 1.3220], + device='cuda:3'), covar=tensor([0.1156, 0.0464, 0.1296, 0.0514, 0.0694, 0.1540, 0.0924, 0.0822], + device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0236, 0.0330, 0.0303, 0.0301, 0.0337, 0.0344, 0.0316], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 19:12:18,388 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 19:12:33,734 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129623.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 19:12:38,235 INFO [train.py:901] (3/4) Epoch 17, batch 300, loss[loss=0.2222, simple_loss=0.3039, pruned_loss=0.07026, over 8620.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2974, pruned_loss=0.06916, over 1262215.49 frames. ], batch size: 39, lr: 4.54e-03, grad_scale: 8.0 +2023-02-06 19:12:39,095 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129630.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:12:53,545 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129648.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 19:13:08,147 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.705e+02 2.453e+02 3.064e+02 3.747e+02 1.027e+03, threshold=6.129e+02, percent-clipped=5.0 +2023-02-06 19:13:14,342 INFO [train.py:901] (3/4) Epoch 17, batch 350, loss[loss=0.2197, simple_loss=0.302, pruned_loss=0.06864, over 7821.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2972, pruned_loss=0.06902, over 1341007.42 frames. ], batch size: 20, lr: 4.54e-03, grad_scale: 8.0 +2023-02-06 19:13:47,828 INFO [train.py:901] (3/4) Epoch 17, batch 400, loss[loss=0.2034, simple_loss=0.2803, pruned_loss=0.06321, over 7922.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2979, pruned_loss=0.06949, over 1403327.53 frames. ], batch size: 20, lr: 4.54e-03, grad_scale: 8.0 +2023-02-06 19:14:08,552 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([0.9923, 1.0053, 1.1689, 0.9452, 0.6421, 1.0410, 1.0131, 0.9155], + device='cuda:3'), covar=tensor([0.0516, 0.0907, 0.1261, 0.1075, 0.0504, 0.1090, 0.0587, 0.0499], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0151, 0.0189, 0.0155, 0.0100, 0.0161, 0.0114, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 19:14:17,992 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.577e+02 2.355e+02 2.898e+02 3.830e+02 8.224e+02, threshold=5.797e+02, percent-clipped=7.0 +2023-02-06 19:14:21,452 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=129775.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 19:14:24,068 INFO [train.py:901] (3/4) Epoch 17, batch 450, loss[loss=0.2421, simple_loss=0.3231, pruned_loss=0.08053, over 8496.00 frames. ], tot_loss[loss=0.2188, simple_loss=0.2985, pruned_loss=0.06953, over 1456989.13 frames. ], batch size: 26, lr: 4.54e-03, grad_scale: 8.0 +2023-02-06 19:14:26,520 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-06 19:14:39,892 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.23 vs. limit=5.0 +2023-02-06 19:14:44,133 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 19:14:44,535 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129809.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:14:58,031 INFO [train.py:901] (3/4) Epoch 17, batch 500, loss[loss=0.2247, simple_loss=0.3057, pruned_loss=0.07184, over 8027.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.2991, pruned_loss=0.06974, over 1493917.96 frames. ], batch size: 22, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:15:28,005 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.494e+02 2.377e+02 2.910e+02 3.862e+02 1.132e+03, threshold=5.820e+02, percent-clipped=8.0 +2023-02-06 19:15:35,670 INFO [train.py:901] (3/4) Epoch 17, batch 550, loss[loss=0.2625, simple_loss=0.3197, pruned_loss=0.1027, over 5136.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2991, pruned_loss=0.06989, over 1522423.12 frames. ], batch size: 11, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:15:43,353 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129890.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 19:16:10,114 INFO [train.py:901] (3/4) Epoch 17, batch 600, loss[loss=0.2331, simple_loss=0.3133, pruned_loss=0.0764, over 8503.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2981, pruned_loss=0.06928, over 1542727.45 frames. ], batch size: 26, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:16:19,728 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 19:16:24,107 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9763, 1.5262, 1.6402, 1.4143, 0.9762, 1.4784, 1.8029, 1.6941], + device='cuda:3'), covar=tensor([0.0539, 0.1200, 0.1637, 0.1390, 0.0611, 0.1483, 0.0668, 0.0581], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0151, 0.0190, 0.0156, 0.0100, 0.0162, 0.0114, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 19:16:38,509 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.845e+02 2.576e+02 2.936e+02 3.639e+02 7.352e+02, threshold=5.872e+02, percent-clipped=2.0 +2023-02-06 19:16:41,362 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=129974.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:16:44,756 INFO [train.py:901] (3/4) Epoch 17, batch 650, loss[loss=0.2503, simple_loss=0.3264, pruned_loss=0.08707, over 8389.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2994, pruned_loss=0.06972, over 1564566.95 frames. ], batch size: 49, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:17:09,828 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8374, 1.8721, 2.4074, 1.6410, 1.3895, 2.4131, 0.4497, 1.4563], + device='cuda:3'), covar=tensor([0.2032, 0.1371, 0.0340, 0.1512, 0.2952, 0.0380, 0.2431, 0.1576], + device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0185, 0.0115, 0.0217, 0.0265, 0.0121, 0.0165, 0.0179], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 19:17:11,757 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2704, 1.5050, 4.4315, 2.0564, 2.5877, 5.0329, 5.0175, 4.3335], + device='cuda:3'), covar=tensor([0.1173, 0.1802, 0.0326, 0.1910, 0.1070, 0.0201, 0.0491, 0.0572], + device='cuda:3'), in_proj_covar=tensor([0.0279, 0.0306, 0.0270, 0.0298, 0.0291, 0.0251, 0.0384, 0.0296], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 19:17:16,416 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130018.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:17:16,512 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5805, 1.8531, 2.7591, 1.4458, 2.1225, 1.9073, 1.6593, 2.0071], + device='cuda:3'), covar=tensor([0.1861, 0.2314, 0.0814, 0.4103, 0.1571, 0.2898, 0.2059, 0.2012], + device='cuda:3'), in_proj_covar=tensor([0.0505, 0.0560, 0.0542, 0.0610, 0.0628, 0.0568, 0.0502, 0.0617], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 19:17:17,845 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130020.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:17:23,793 INFO [train.py:901] (3/4) Epoch 17, batch 700, loss[loss=0.2163, simple_loss=0.2981, pruned_loss=0.06726, over 8455.00 frames. ], tot_loss[loss=0.218, simple_loss=0.2978, pruned_loss=0.0691, over 1577827.23 frames. ], batch size: 27, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:17:51,871 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.651e+02 2.350e+02 2.811e+02 3.683e+02 1.098e+03, threshold=5.622e+02, percent-clipped=6.0 +2023-02-06 19:17:58,279 INFO [train.py:901] (3/4) Epoch 17, batch 750, loss[loss=0.2099, simple_loss=0.2795, pruned_loss=0.07018, over 8078.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2964, pruned_loss=0.06818, over 1588430.18 frames. ], batch size: 21, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:18:04,874 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3173, 1.6599, 1.7081, 1.0659, 1.7431, 1.2676, 0.2634, 1.5247], + device='cuda:3'), covar=tensor([0.0382, 0.0290, 0.0223, 0.0377, 0.0324, 0.0724, 0.0642, 0.0235], + device='cuda:3'), in_proj_covar=tensor([0.0421, 0.0362, 0.0309, 0.0418, 0.0346, 0.0507, 0.0370, 0.0388], + device='cuda:3'), out_proj_covar=tensor([1.1516e-04, 9.6632e-05, 8.1938e-05, 1.1221e-04, 9.3132e-05, 1.4651e-04, + 1.0125e-04, 1.0452e-04], device='cuda:3') +2023-02-06 19:18:05,572 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130089.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:18:08,253 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 19:18:19,439 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 19:18:36,029 INFO [train.py:901] (3/4) Epoch 17, batch 800, loss[loss=0.1967, simple_loss=0.263, pruned_loss=0.06523, over 7430.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2972, pruned_loss=0.06891, over 1594890.80 frames. ], batch size: 17, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:18:46,672 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7547, 1.7647, 2.3822, 1.6487, 1.3446, 2.3157, 0.4775, 1.4308], + device='cuda:3'), covar=tensor([0.1818, 0.1368, 0.0352, 0.1422, 0.3154, 0.0458, 0.2478, 0.1565], + device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0186, 0.0116, 0.0218, 0.0265, 0.0122, 0.0167, 0.0181], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 19:18:48,084 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130146.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 19:18:52,760 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130153.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:19:04,223 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.543e+02 2.363e+02 2.676e+02 3.408e+02 8.560e+02, threshold=5.353e+02, percent-clipped=3.0 +2023-02-06 19:19:05,151 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130171.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 19:19:10,476 INFO [train.py:901] (3/4) Epoch 17, batch 850, loss[loss=0.1843, simple_loss=0.255, pruned_loss=0.05678, over 7542.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2974, pruned_loss=0.06885, over 1602095.72 frames. ], batch size: 18, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:19:47,566 INFO [train.py:901] (3/4) Epoch 17, batch 900, loss[loss=0.2467, simple_loss=0.3159, pruned_loss=0.08872, over 8104.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2955, pruned_loss=0.06801, over 1605344.34 frames. ], batch size: 23, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:20:15,376 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130268.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:20:16,501 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.765e+02 2.489e+02 3.023e+02 3.878e+02 8.176e+02, threshold=6.045e+02, percent-clipped=7.0 +2023-02-06 19:20:22,805 INFO [train.py:901] (3/4) Epoch 17, batch 950, loss[loss=0.2315, simple_loss=0.3107, pruned_loss=0.07608, over 8500.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.295, pruned_loss=0.06795, over 1609312.87 frames. ], batch size: 26, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:20:29,213 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3286, 1.5530, 2.1803, 1.1824, 1.4831, 1.5535, 1.3553, 1.5950], + device='cuda:3'), covar=tensor([0.2021, 0.2514, 0.0950, 0.4402, 0.2028, 0.3327, 0.2346, 0.2151], + device='cuda:3'), in_proj_covar=tensor([0.0507, 0.0564, 0.0545, 0.0612, 0.0632, 0.0570, 0.0504, 0.0619], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 19:20:43,395 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 19:20:57,189 INFO [train.py:901] (3/4) Epoch 17, batch 1000, loss[loss=0.2128, simple_loss=0.2863, pruned_loss=0.06963, over 7527.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2959, pruned_loss=0.06858, over 1608223.91 frames. ], batch size: 18, lr: 4.53e-03, grad_scale: 8.0 +2023-02-06 19:21:04,876 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0060, 2.2516, 1.9841, 3.0233, 1.4144, 1.7702, 2.0804, 2.4024], + device='cuda:3'), covar=tensor([0.0780, 0.0844, 0.0981, 0.0336, 0.1200, 0.1288, 0.0972, 0.0767], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0202, 0.0252, 0.0213, 0.0210, 0.0250, 0.0255, 0.0214], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 19:21:09,223 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130345.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:21:20,028 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 19:21:21,979 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130362.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:21:23,327 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130364.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:21:27,498 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 2.680e+02 3.059e+02 3.924e+02 8.380e+02, threshold=6.118e+02, percent-clipped=2.0 +2023-02-06 19:21:27,752 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130370.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:21:33,152 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 19:21:33,588 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-06 19:21:33,831 INFO [train.py:901] (3/4) Epoch 17, batch 1050, loss[loss=0.1828, simple_loss=0.2617, pruned_loss=0.05193, over 7434.00 frames. ], tot_loss[loss=0.2159, simple_loss=0.2946, pruned_loss=0.06864, over 1607168.56 frames. ], batch size: 17, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:21:49,962 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130402.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:22:08,452 INFO [train.py:901] (3/4) Epoch 17, batch 1100, loss[loss=0.1871, simple_loss=0.2712, pruned_loss=0.05145, over 8131.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2952, pruned_loss=0.06823, over 1611280.28 frames. ], batch size: 22, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:22:14,705 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8550, 1.6691, 3.3807, 1.5279, 2.4737, 3.7235, 3.8197, 3.1989], + device='cuda:3'), covar=tensor([0.1173, 0.1522, 0.0324, 0.1970, 0.0957, 0.0223, 0.0394, 0.0543], + device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0307, 0.0272, 0.0301, 0.0293, 0.0252, 0.0386, 0.0296], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 19:22:23,047 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130450.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:22:27,202 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130456.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:22:38,665 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.758e+02 2.545e+02 2.978e+02 3.676e+02 6.168e+02, threshold=5.956e+02, percent-clipped=1.0 +2023-02-06 19:22:44,133 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130477.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:22:45,340 INFO [train.py:901] (3/4) Epoch 17, batch 1150, loss[loss=0.176, simple_loss=0.2712, pruned_loss=0.04037, over 8025.00 frames. ], tot_loss[loss=0.215, simple_loss=0.295, pruned_loss=0.0675, over 1613094.99 frames. ], batch size: 22, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:22:45,511 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130479.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:22:45,963 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 19:23:16,172 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130524.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:23:19,416 INFO [train.py:901] (3/4) Epoch 17, batch 1200, loss[loss=0.2156, simple_loss=0.2913, pruned_loss=0.06992, over 8350.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2958, pruned_loss=0.06776, over 1615119.48 frames. ], batch size: 48, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:23:33,403 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130549.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:23:45,146 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130566.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:23:47,779 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.542e+02 2.314e+02 2.862e+02 3.617e+02 1.013e+03, threshold=5.724e+02, percent-clipped=2.0 +2023-02-06 19:23:53,883 INFO [train.py:901] (3/4) Epoch 17, batch 1250, loss[loss=0.2665, simple_loss=0.3346, pruned_loss=0.09926, over 7808.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2955, pruned_loss=0.06789, over 1616631.59 frames. ], batch size: 20, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:23:57,469 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130583.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:24:07,984 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5464, 1.8089, 1.8689, 1.1671, 1.9739, 1.3577, 0.4692, 1.8300], + device='cuda:3'), covar=tensor([0.0435, 0.0272, 0.0208, 0.0425, 0.0289, 0.0778, 0.0714, 0.0237], + device='cuda:3'), in_proj_covar=tensor([0.0427, 0.0365, 0.0312, 0.0424, 0.0350, 0.0516, 0.0377, 0.0394], + device='cuda:3'), out_proj_covar=tensor([1.1675e-04, 9.7352e-05, 8.2703e-05, 1.1359e-04, 9.3933e-05, 1.4911e-04, + 1.0301e-04, 1.0598e-04], device='cuda:3') +2023-02-06 19:24:30,834 INFO [train.py:901] (3/4) Epoch 17, batch 1300, loss[loss=0.2795, simple_loss=0.3565, pruned_loss=0.1013, over 8472.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2955, pruned_loss=0.0679, over 1614970.27 frames. ], batch size: 29, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:24:35,110 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2881, 2.6293, 3.0395, 1.5457, 3.3451, 1.8869, 1.5978, 2.3040], + device='cuda:3'), covar=tensor([0.0688, 0.0361, 0.0213, 0.0637, 0.0304, 0.0719, 0.0755, 0.0470], + device='cuda:3'), in_proj_covar=tensor([0.0426, 0.0365, 0.0313, 0.0424, 0.0348, 0.0516, 0.0376, 0.0393], + device='cuda:3'), out_proj_covar=tensor([1.1640e-04, 9.7338e-05, 8.2955e-05, 1.1361e-04, 9.3577e-05, 1.4901e-04, + 1.0280e-04, 1.0569e-04], device='cuda:3') +2023-02-06 19:24:59,335 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.629e+02 2.380e+02 3.126e+02 3.675e+02 7.509e+02, threshold=6.253e+02, percent-clipped=2.0 +2023-02-06 19:25:05,687 INFO [train.py:901] (3/4) Epoch 17, batch 1350, loss[loss=0.1694, simple_loss=0.2495, pruned_loss=0.0446, over 7650.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2958, pruned_loss=0.06746, over 1617629.26 frames. ], batch size: 19, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:25:43,062 INFO [train.py:901] (3/4) Epoch 17, batch 1400, loss[loss=0.2005, simple_loss=0.2695, pruned_loss=0.06569, over 7786.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.295, pruned_loss=0.06734, over 1616019.58 frames. ], batch size: 19, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:25:46,048 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130733.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:25:47,394 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130735.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:25:54,827 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130746.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:26:03,100 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130758.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:26:04,431 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130760.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:26:11,021 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.401e+02 2.607e+02 3.260e+02 4.191e+02 1.113e+03, threshold=6.520e+02, percent-clipped=3.0 +2023-02-06 19:26:17,374 INFO [train.py:901] (3/4) Epoch 17, batch 1450, loss[loss=0.2223, simple_loss=0.3063, pruned_loss=0.06918, over 8294.00 frames. ], tot_loss[loss=0.216, simple_loss=0.2961, pruned_loss=0.06794, over 1615198.92 frames. ], batch size: 23, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:26:20,718 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 19:26:27,816 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130794.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:26:32,216 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130800.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:26:34,999 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.7542, 1.7420, 3.9742, 1.4235, 3.4967, 3.3667, 3.6523, 3.4932], + device='cuda:3'), covar=tensor([0.0699, 0.3900, 0.0726, 0.3916, 0.1319, 0.1023, 0.0607, 0.0758], + device='cuda:3'), in_proj_covar=tensor([0.0570, 0.0627, 0.0650, 0.0595, 0.0676, 0.0580, 0.0573, 0.0637], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 19:26:54,259 INFO [train.py:901] (3/4) Epoch 17, batch 1500, loss[loss=0.2133, simple_loss=0.2919, pruned_loss=0.06735, over 7528.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2949, pruned_loss=0.06754, over 1613390.70 frames. ], batch size: 18, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:27:17,129 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130861.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:27:22,932 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.580e+02 2.370e+02 2.974e+02 3.638e+02 1.375e+03, threshold=5.949e+02, percent-clipped=1.0 +2023-02-06 19:27:29,127 INFO [train.py:901] (3/4) Epoch 17, batch 1550, loss[loss=0.2656, simple_loss=0.3229, pruned_loss=0.1042, over 7003.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2957, pruned_loss=0.06843, over 1612847.41 frames. ], batch size: 72, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:27:50,134 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130909.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:27:50,699 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130910.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:27:54,327 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130915.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:28:02,601 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=130927.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:28:03,805 INFO [train.py:901] (3/4) Epoch 17, batch 1600, loss[loss=0.2325, simple_loss=0.3082, pruned_loss=0.07838, over 8239.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2962, pruned_loss=0.06823, over 1614994.64 frames. ], batch size: 22, lr: 4.52e-03, grad_scale: 8.0 +2023-02-06 19:28:34,757 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.790e+02 2.345e+02 2.992e+02 3.546e+02 8.486e+02, threshold=5.983e+02, percent-clipped=5.0 +2023-02-06 19:28:40,942 INFO [train.py:901] (3/4) Epoch 17, batch 1650, loss[loss=0.2485, simple_loss=0.3273, pruned_loss=0.08485, over 8356.00 frames. ], tot_loss[loss=0.216, simple_loss=0.2956, pruned_loss=0.06818, over 1614551.44 frames. ], batch size: 24, lr: 4.51e-03, grad_scale: 16.0 +2023-02-06 19:28:57,386 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-06 19:29:13,492 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131025.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:29:16,140 INFO [train.py:901] (3/4) Epoch 17, batch 1700, loss[loss=0.2162, simple_loss=0.2942, pruned_loss=0.06913, over 8083.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2958, pruned_loss=0.06781, over 1617694.35 frames. ], batch size: 21, lr: 4.51e-03, grad_scale: 16.0 +2023-02-06 19:29:25,392 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131042.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:29:46,943 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.737e+02 2.451e+02 3.155e+02 3.823e+02 7.811e+02, threshold=6.311e+02, percent-clipped=3.0 +2023-02-06 19:29:53,069 INFO [train.py:901] (3/4) Epoch 17, batch 1750, loss[loss=0.1907, simple_loss=0.2775, pruned_loss=0.05197, over 7541.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2978, pruned_loss=0.06905, over 1619974.20 frames. ], batch size: 18, lr: 4.51e-03, grad_scale: 16.0 +2023-02-06 19:29:58,739 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.1166, 1.8030, 1.9344, 1.8847, 1.2422, 1.8050, 2.6368, 2.5759], + device='cuda:3'), covar=tensor([0.0434, 0.1115, 0.1613, 0.1322, 0.0579, 0.1399, 0.0510, 0.0495], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0191, 0.0157, 0.0100, 0.0164, 0.0115, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 19:30:19,627 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131117.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:30:27,911 INFO [train.py:901] (3/4) Epoch 17, batch 1800, loss[loss=0.1803, simple_loss=0.2732, pruned_loss=0.04368, over 8467.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2971, pruned_loss=0.06822, over 1617642.03 frames. ], batch size: 25, lr: 4.51e-03, grad_scale: 16.0 +2023-02-06 19:30:37,105 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131142.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:30:52,694 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131165.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:30:53,406 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1035, 1.6419, 1.4801, 1.6594, 1.4499, 1.3981, 1.3038, 1.3834], + device='cuda:3'), covar=tensor([0.1047, 0.0416, 0.1123, 0.0448, 0.0662, 0.1275, 0.0832, 0.0713], + device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0230, 0.0324, 0.0298, 0.0296, 0.0326, 0.0338, 0.0311], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 19:30:55,955 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.715e+02 2.745e+02 3.356e+02 4.683e+02 1.105e+03, threshold=6.712e+02, percent-clipped=11.0 +2023-02-06 19:30:56,921 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131171.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:31:03,685 INFO [train.py:901] (3/4) Epoch 17, batch 1850, loss[loss=0.2085, simple_loss=0.2701, pruned_loss=0.07344, over 7693.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2976, pruned_loss=0.06887, over 1620649.70 frames. ], batch size: 18, lr: 4.51e-03, grad_scale: 16.0 +2023-02-06 19:31:06,774 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0423, 2.2888, 1.8884, 2.7621, 1.4530, 1.6627, 1.9738, 2.3050], + device='cuda:3'), covar=tensor([0.0708, 0.0778, 0.0988, 0.0416, 0.1133, 0.1391, 0.0919, 0.0820], + device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0198, 0.0248, 0.0211, 0.0208, 0.0246, 0.0250, 0.0210], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 19:31:12,470 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131190.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:31:13,837 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131192.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:31:16,719 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131196.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:31:39,542 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5173, 2.2885, 3.2820, 2.6473, 3.1491, 2.5276, 2.2312, 1.8713], + device='cuda:3'), covar=tensor([0.5071, 0.5195, 0.1760, 0.3310, 0.2386, 0.2624, 0.1768, 0.5050], + device='cuda:3'), in_proj_covar=tensor([0.0921, 0.0928, 0.0773, 0.0899, 0.0962, 0.0853, 0.0723, 0.0798], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 19:31:39,955 INFO [train.py:901] (3/4) Epoch 17, batch 1900, loss[loss=0.1919, simple_loss=0.2776, pruned_loss=0.05307, over 8125.00 frames. ], tot_loss[loss=0.2166, simple_loss=0.2967, pruned_loss=0.06828, over 1615387.55 frames. ], batch size: 22, lr: 4.51e-03, grad_scale: 16.0 +2023-02-06 19:32:08,100 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.492e+02 2.313e+02 2.955e+02 3.582e+02 5.685e+02, threshold=5.910e+02, percent-clipped=0.0 +2023-02-06 19:32:08,131 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 19:32:14,127 INFO [train.py:901] (3/4) Epoch 17, batch 1950, loss[loss=0.2129, simple_loss=0.3013, pruned_loss=0.0622, over 8499.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2963, pruned_loss=0.06794, over 1616380.84 frames. ], batch size: 26, lr: 4.51e-03, grad_scale: 16.0 +2023-02-06 19:32:15,756 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131281.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:32:19,626 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 19:32:28,929 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131298.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:32:35,156 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131306.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:32:39,941 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 19:32:47,534 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131323.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:32:51,199 INFO [train.py:901] (3/4) Epoch 17, batch 2000, loss[loss=0.2366, simple_loss=0.3219, pruned_loss=0.07567, over 8247.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2966, pruned_loss=0.06838, over 1616525.93 frames. ], batch size: 24, lr: 4.51e-03, grad_scale: 16.0 +2023-02-06 19:33:19,858 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.770e+02 2.510e+02 3.128e+02 3.622e+02 6.098e+02, threshold=6.257e+02, percent-clipped=1.0 +2023-02-06 19:33:25,352 INFO [train.py:901] (3/4) Epoch 17, batch 2050, loss[loss=0.2181, simple_loss=0.2983, pruned_loss=0.06896, over 8449.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2956, pruned_loss=0.06785, over 1613476.85 frames. ], batch size: 25, lr: 4.51e-03, grad_scale: 8.0 +2023-02-06 19:34:00,685 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131427.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:34:01,952 INFO [train.py:901] (3/4) Epoch 17, batch 2100, loss[loss=0.1735, simple_loss=0.2594, pruned_loss=0.04378, over 7670.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2941, pruned_loss=0.06748, over 1610781.27 frames. ], batch size: 19, lr: 4.51e-03, grad_scale: 8.0 +2023-02-06 19:34:06,129 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131434.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:34:31,410 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.671e+02 2.457e+02 2.884e+02 3.530e+02 8.686e+02, threshold=5.767e+02, percent-clipped=1.0 +2023-02-06 19:34:36,973 INFO [train.py:901] (3/4) Epoch 17, batch 2150, loss[loss=0.2275, simple_loss=0.3225, pruned_loss=0.06627, over 8457.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2952, pruned_loss=0.06754, over 1614854.70 frames. ], batch size: 27, lr: 4.51e-03, grad_scale: 8.0 +2023-02-06 19:34:58,677 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131510.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:35:12,359 INFO [train.py:901] (3/4) Epoch 17, batch 2200, loss[loss=0.2036, simple_loss=0.2869, pruned_loss=0.06015, over 8106.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2953, pruned_loss=0.06748, over 1617402.99 frames. ], batch size: 23, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:35:18,094 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=131536.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:35:36,119 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6287, 2.2649, 1.7649, 4.0719, 1.6589, 1.5764, 2.3884, 2.7373], + device='cuda:3'), covar=tensor([0.1662, 0.1376, 0.2015, 0.0262, 0.1569, 0.2124, 0.1270, 0.0895], + device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0198, 0.0247, 0.0211, 0.0208, 0.0246, 0.0253, 0.0209], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 19:35:43,530 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.685e+02 2.550e+02 3.248e+02 4.465e+02 1.208e+03, threshold=6.496e+02, percent-clipped=6.0 +2023-02-06 19:35:49,218 INFO [train.py:901] (3/4) Epoch 17, batch 2250, loss[loss=0.208, simple_loss=0.2721, pruned_loss=0.07188, over 7670.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2951, pruned_loss=0.06736, over 1617844.40 frames. ], batch size: 18, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:36:23,880 INFO [train.py:901] (3/4) Epoch 17, batch 2300, loss[loss=0.2361, simple_loss=0.316, pruned_loss=0.07815, over 8589.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2956, pruned_loss=0.06778, over 1618239.86 frames. ], batch size: 49, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:36:24,090 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3940, 2.7847, 2.5277, 3.9990, 1.8065, 2.2366, 2.5158, 3.1038], + device='cuda:3'), covar=tensor([0.0733, 0.0813, 0.0844, 0.0326, 0.1174, 0.1225, 0.1034, 0.0728], + device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0199, 0.0249, 0.0212, 0.0209, 0.0247, 0.0254, 0.0210], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 19:36:40,759 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131651.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:36:55,869 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.285e+02 2.553e+02 3.001e+02 3.824e+02 6.268e+02, threshold=6.003e+02, percent-clipped=0.0 +2023-02-06 19:37:01,535 INFO [train.py:901] (3/4) Epoch 17, batch 2350, loss[loss=0.1809, simple_loss=0.2639, pruned_loss=0.049, over 8095.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2947, pruned_loss=0.06763, over 1613317.65 frames. ], batch size: 21, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:37:16,027 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6169, 1.6684, 2.0811, 1.4626, 1.2026, 2.0709, 0.3023, 1.2574], + device='cuda:3'), covar=tensor([0.1972, 0.1230, 0.0374, 0.1265, 0.2920, 0.0425, 0.2449, 0.1440], + device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0183, 0.0114, 0.0216, 0.0260, 0.0121, 0.0166, 0.0179], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 19:37:21,843 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 +2023-02-06 19:37:35,923 INFO [train.py:901] (3/4) Epoch 17, batch 2400, loss[loss=0.1773, simple_loss=0.2753, pruned_loss=0.03964, over 8228.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2957, pruned_loss=0.06825, over 1614344.07 frames. ], batch size: 22, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:37:40,426 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2101, 2.0955, 1.6241, 1.8712, 1.7213, 1.4074, 1.7119, 1.6669], + device='cuda:3'), covar=tensor([0.1184, 0.0421, 0.1185, 0.0523, 0.0667, 0.1355, 0.0831, 0.0815], + device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0233, 0.0324, 0.0299, 0.0297, 0.0328, 0.0340, 0.0312], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 19:38:06,374 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.511e+02 2.467e+02 3.155e+02 3.892e+02 8.269e+02, threshold=6.310e+02, percent-clipped=4.0 +2023-02-06 19:38:06,491 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=131771.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:38:12,217 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=131778.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:38:12,839 INFO [train.py:901] (3/4) Epoch 17, batch 2450, loss[loss=0.1991, simple_loss=0.291, pruned_loss=0.05357, over 8524.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2947, pruned_loss=0.06753, over 1613234.97 frames. ], batch size: 49, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:38:47,888 INFO [train.py:901] (3/4) Epoch 17, batch 2500, loss[loss=0.2254, simple_loss=0.304, pruned_loss=0.07339, over 8361.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2953, pruned_loss=0.0679, over 1612441.81 frames. ], batch size: 24, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:38:58,548 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8229, 1.4194, 1.6071, 1.3538, 0.9564, 1.3725, 1.7327, 1.5522], + device='cuda:3'), covar=tensor([0.0498, 0.1215, 0.1570, 0.1362, 0.0560, 0.1430, 0.0659, 0.0597], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0151, 0.0189, 0.0156, 0.0099, 0.0161, 0.0114, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 19:39:05,482 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=131854.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:39:17,118 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.761e+02 2.481e+02 2.929e+02 3.320e+02 7.417e+02, threshold=5.858e+02, percent-clipped=2.0 +2023-02-06 19:39:20,127 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131875.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:39:22,846 INFO [train.py:901] (3/4) Epoch 17, batch 2550, loss[loss=0.1764, simple_loss=0.2687, pruned_loss=0.04205, over 7968.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2956, pruned_loss=0.06857, over 1608793.91 frames. ], batch size: 21, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:39:29,642 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131886.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:39:34,611 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131893.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:39:42,812 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9878, 1.7735, 3.2824, 1.5438, 2.4676, 3.5448, 3.6580, 2.9738], + device='cuda:3'), covar=tensor([0.1133, 0.1619, 0.0480, 0.2085, 0.1126, 0.0332, 0.0691, 0.0667], + device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0308, 0.0275, 0.0302, 0.0292, 0.0252, 0.0388, 0.0299], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 19:39:45,590 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131907.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:40:00,833 INFO [train.py:901] (3/4) Epoch 17, batch 2600, loss[loss=0.1845, simple_loss=0.2658, pruned_loss=0.05164, over 8601.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.2949, pruned_loss=0.06777, over 1609733.79 frames. ], batch size: 31, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:40:03,026 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131932.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:40:28,771 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131969.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:40:29,953 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.651e+02 2.407e+02 2.887e+02 3.716e+02 6.826e+02, threshold=5.774e+02, percent-clipped=1.0 +2023-02-06 19:40:35,438 INFO [train.py:901] (3/4) Epoch 17, batch 2650, loss[loss=0.1951, simple_loss=0.2942, pruned_loss=0.048, over 8707.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.295, pruned_loss=0.06796, over 1611599.36 frames. ], batch size: 34, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:40:52,328 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0743, 1.2587, 1.2483, 0.6507, 1.2397, 1.0401, 0.0696, 1.2267], + device='cuda:3'), covar=tensor([0.0339, 0.0318, 0.0260, 0.0435, 0.0391, 0.0853, 0.0665, 0.0271], + device='cuda:3'), in_proj_covar=tensor([0.0426, 0.0368, 0.0315, 0.0425, 0.0351, 0.0509, 0.0376, 0.0391], + device='cuda:3'), out_proj_covar=tensor([1.1651e-04, 9.7932e-05, 8.3442e-05, 1.1367e-04, 9.4395e-05, 1.4685e-04, + 1.0248e-04, 1.0514e-04], device='cuda:3') +2023-02-06 19:41:13,503 INFO [train.py:901] (3/4) Epoch 17, batch 2700, loss[loss=0.2512, simple_loss=0.3192, pruned_loss=0.09157, over 7290.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2961, pruned_loss=0.06862, over 1614537.59 frames. ], batch size: 71, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:41:27,368 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132049.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:41:31,122 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 19:41:42,497 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 2.450e+02 3.248e+02 4.102e+02 1.137e+03, threshold=6.496e+02, percent-clipped=12.0 +2023-02-06 19:41:43,394 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132072.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:41:48,263 INFO [train.py:901] (3/4) Epoch 17, batch 2750, loss[loss=0.2625, simple_loss=0.3283, pruned_loss=0.09832, over 7027.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2948, pruned_loss=0.06814, over 1606695.25 frames. ], batch size: 71, lr: 4.50e-03, grad_scale: 8.0 +2023-02-06 19:42:25,027 INFO [train.py:901] (3/4) Epoch 17, batch 2800, loss[loss=0.2451, simple_loss=0.3249, pruned_loss=0.08269, over 8196.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2944, pruned_loss=0.06766, over 1609159.30 frames. ], batch size: 23, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:42:33,150 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.7811, 5.7650, 5.0813, 2.5214, 5.0979, 5.6509, 5.4435, 5.2952], + device='cuda:3'), covar=tensor([0.0587, 0.0467, 0.1030, 0.4846, 0.0790, 0.0795, 0.1099, 0.0616], + device='cuda:3'), in_proj_covar=tensor([0.0505, 0.0414, 0.0418, 0.0512, 0.0407, 0.0416, 0.0402, 0.0360], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 19:42:35,387 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132142.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:42:40,216 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132149.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:42:52,729 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132167.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:42:55,269 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.536e+02 2.217e+02 2.865e+02 3.623e+02 1.020e+03, threshold=5.730e+02, percent-clipped=3.0 +2023-02-06 19:42:57,334 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132174.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:43:00,624 INFO [train.py:901] (3/4) Epoch 17, batch 2850, loss[loss=0.19, simple_loss=0.2716, pruned_loss=0.05421, over 8134.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.2947, pruned_loss=0.06784, over 1603957.43 frames. ], batch size: 22, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:43:10,452 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 19:43:29,199 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=132219.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:43:33,397 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132225.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:43:35,747 INFO [train.py:901] (3/4) Epoch 17, batch 2900, loss[loss=0.2235, simple_loss=0.3001, pruned_loss=0.07344, over 7810.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.2945, pruned_loss=0.06732, over 1604358.02 frames. ], batch size: 20, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:43:52,939 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132250.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:44:08,371 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.523e+02 2.407e+02 2.887e+02 3.454e+02 7.005e+02, threshold=5.774e+02, percent-clipped=2.0 +2023-02-06 19:44:09,818 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 19:44:13,741 INFO [train.py:901] (3/4) Epoch 17, batch 2950, loss[loss=0.2033, simple_loss=0.2787, pruned_loss=0.06398, over 8076.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2941, pruned_loss=0.06736, over 1604501.00 frames. ], batch size: 21, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:44:36,062 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-06 19:44:48,315 INFO [train.py:901] (3/4) Epoch 17, batch 3000, loss[loss=0.1823, simple_loss=0.2776, pruned_loss=0.04353, over 8235.00 frames. ], tot_loss[loss=0.2159, simple_loss=0.2958, pruned_loss=0.06806, over 1609704.87 frames. ], batch size: 22, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:44:48,315 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 19:44:56,140 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5615, 1.8382, 2.5538, 1.4134, 2.1176, 1.8400, 1.6952, 1.9575], + device='cuda:3'), covar=tensor([0.1635, 0.2621, 0.0882, 0.4086, 0.1636, 0.2763, 0.2056, 0.2194], + device='cuda:3'), in_proj_covar=tensor([0.0509, 0.0566, 0.0546, 0.0619, 0.0634, 0.0574, 0.0510, 0.0623], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 19:45:00,591 INFO [train.py:935] (3/4) Epoch 17, validation: loss=0.1786, simple_loss=0.2786, pruned_loss=0.03928, over 944034.00 frames. +2023-02-06 19:45:00,592 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 19:45:04,438 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132334.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:45:07,390 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2420, 1.9449, 2.6472, 2.1425, 2.5084, 2.2126, 1.9167, 1.3171], + device='cuda:3'), covar=tensor([0.4945, 0.4688, 0.1716, 0.3283, 0.2352, 0.2597, 0.1779, 0.5029], + device='cuda:3'), in_proj_covar=tensor([0.0931, 0.0938, 0.0776, 0.0908, 0.0973, 0.0858, 0.0727, 0.0809], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 19:45:12,169 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7014, 1.5161, 1.7091, 1.4608, 0.9044, 1.4948, 1.5055, 1.3479], + device='cuda:3'), covar=tensor([0.0498, 0.1303, 0.1668, 0.1372, 0.0591, 0.1549, 0.0691, 0.0673], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0191, 0.0157, 0.0100, 0.0163, 0.0115, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0008, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 19:45:31,440 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.547e+02 2.492e+02 3.005e+02 3.786e+02 8.313e+02, threshold=6.010e+02, percent-clipped=11.0 +2023-02-06 19:45:37,097 INFO [train.py:901] (3/4) Epoch 17, batch 3050, loss[loss=0.2309, simple_loss=0.3092, pruned_loss=0.07632, over 8337.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2966, pruned_loss=0.06853, over 1609848.85 frames. ], batch size: 26, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:45:48,262 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=132393.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:46:04,209 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=132416.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:46:12,936 INFO [train.py:901] (3/4) Epoch 17, batch 3100, loss[loss=0.2224, simple_loss=0.3015, pruned_loss=0.07163, over 7550.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2971, pruned_loss=0.06864, over 1611601.38 frames. ], batch size: 18, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:46:41,891 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.762e+02 2.340e+02 2.843e+02 3.195e+02 7.960e+02, threshold=5.685e+02, percent-clipped=6.0 +2023-02-06 19:46:47,328 INFO [train.py:901] (3/4) Epoch 17, batch 3150, loss[loss=0.2118, simple_loss=0.2873, pruned_loss=0.06819, over 8025.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2965, pruned_loss=0.06826, over 1615628.02 frames. ], batch size: 22, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:46:48,924 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2933, 2.3478, 1.7253, 2.0679, 2.0290, 1.4948, 1.8877, 1.8946], + device='cuda:3'), covar=tensor([0.1376, 0.0343, 0.1101, 0.0550, 0.0690, 0.1409, 0.0801, 0.0870], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0233, 0.0324, 0.0300, 0.0297, 0.0328, 0.0339, 0.0315], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 19:47:09,742 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132508.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:47:23,940 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.67 vs. limit=5.0 +2023-02-06 19:47:24,993 INFO [train.py:901] (3/4) Epoch 17, batch 3200, loss[loss=0.2131, simple_loss=0.3052, pruned_loss=0.06044, over 8324.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2965, pruned_loss=0.06821, over 1613790.78 frames. ], batch size: 25, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:47:26,576 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132531.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:47:54,177 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.603e+02 2.495e+02 3.112e+02 3.824e+02 1.248e+03, threshold=6.223e+02, percent-clipped=6.0 +2023-02-06 19:47:59,505 INFO [train.py:901] (3/4) Epoch 17, batch 3250, loss[loss=0.2478, simple_loss=0.3214, pruned_loss=0.08712, over 8495.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2973, pruned_loss=0.06883, over 1614726.67 frames. ], batch size: 26, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:48:07,401 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132590.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:48:11,633 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3109, 2.3225, 1.7602, 2.1398, 1.9774, 1.5932, 1.7774, 1.9775], + device='cuda:3'), covar=tensor([0.1391, 0.0399, 0.1191, 0.0545, 0.0674, 0.1467, 0.0907, 0.0826], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0233, 0.0324, 0.0301, 0.0298, 0.0330, 0.0341, 0.0315], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 19:48:26,272 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132615.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:48:37,516 INFO [train.py:901] (3/4) Epoch 17, batch 3300, loss[loss=0.2274, simple_loss=0.316, pruned_loss=0.06941, over 8288.00 frames. ], tot_loss[loss=0.217, simple_loss=0.2971, pruned_loss=0.06841, over 1616571.19 frames. ], batch size: 23, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:49:06,785 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.345e+02 2.532e+02 2.971e+02 3.744e+02 7.972e+02, threshold=5.942e+02, percent-clipped=3.0 +2023-02-06 19:49:11,878 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132678.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:49:12,432 INFO [train.py:901] (3/4) Epoch 17, batch 3350, loss[loss=0.2106, simple_loss=0.2947, pruned_loss=0.06329, over 8129.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2975, pruned_loss=0.06884, over 1619595.22 frames. ], batch size: 22, lr: 4.49e-03, grad_scale: 8.0 +2023-02-06 19:49:49,258 INFO [train.py:901] (3/4) Epoch 17, batch 3400, loss[loss=0.2207, simple_loss=0.2958, pruned_loss=0.07283, over 8511.00 frames. ], tot_loss[loss=0.2162, simple_loss=0.2962, pruned_loss=0.06814, over 1622745.89 frames. ], batch size: 28, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:49:55,926 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4044, 4.3321, 3.9637, 2.1160, 3.8900, 3.9864, 3.9527, 3.7960], + device='cuda:3'), covar=tensor([0.0766, 0.0549, 0.0977, 0.4562, 0.0889, 0.0915, 0.1197, 0.0812], + device='cuda:3'), in_proj_covar=tensor([0.0501, 0.0414, 0.0421, 0.0513, 0.0406, 0.0414, 0.0400, 0.0358], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 19:50:01,004 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-06 19:50:04,442 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.35 vs. limit=5.0 +2023-02-06 19:50:14,691 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132764.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:50:14,772 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132764.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:50:19,439 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.671e+02 2.442e+02 2.969e+02 4.012e+02 9.663e+02, threshold=5.937e+02, percent-clipped=5.0 +2023-02-06 19:50:24,934 INFO [train.py:901] (3/4) Epoch 17, batch 3450, loss[loss=0.1739, simple_loss=0.2536, pruned_loss=0.04708, over 7701.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2967, pruned_loss=0.06881, over 1620167.13 frames. ], batch size: 18, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:50:30,882 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132787.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:50:32,220 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132789.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:50:33,510 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7143, 1.9634, 2.0801, 1.4118, 2.1518, 1.4942, 0.7907, 1.8268], + device='cuda:3'), covar=tensor([0.0488, 0.0314, 0.0225, 0.0469, 0.0322, 0.0715, 0.0696, 0.0277], + device='cuda:3'), in_proj_covar=tensor([0.0431, 0.0370, 0.0315, 0.0427, 0.0353, 0.0510, 0.0378, 0.0392], + device='cuda:3'), out_proj_covar=tensor([1.1769e-04, 9.8406e-05, 8.3563e-05, 1.1444e-04, 9.4984e-05, 1.4697e-04, + 1.0310e-04, 1.0527e-04], device='cuda:3') +2023-02-06 19:50:47,682 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132812.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:50:59,803 INFO [train.py:901] (3/4) Epoch 17, batch 3500, loss[loss=0.1908, simple_loss=0.2672, pruned_loss=0.05718, over 7539.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2953, pruned_loss=0.06842, over 1616883.73 frames. ], batch size: 18, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:51:13,840 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 19:51:31,530 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.520e+02 2.401e+02 3.009e+02 3.970e+02 8.620e+02, threshold=6.019e+02, percent-clipped=6.0 +2023-02-06 19:51:37,025 INFO [train.py:901] (3/4) Epoch 17, batch 3550, loss[loss=0.2904, simple_loss=0.3559, pruned_loss=0.1124, over 8358.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2961, pruned_loss=0.06851, over 1617672.31 frames. ], batch size: 26, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:52:11,133 INFO [train.py:901] (3/4) Epoch 17, batch 3600, loss[loss=0.2082, simple_loss=0.2822, pruned_loss=0.06713, over 7426.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2959, pruned_loss=0.06869, over 1616290.22 frames. ], batch size: 17, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:52:11,369 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6612, 1.9914, 2.1402, 1.2096, 2.2121, 1.4671, 0.6997, 1.8457], + device='cuda:3'), covar=tensor([0.0522, 0.0289, 0.0233, 0.0562, 0.0341, 0.0817, 0.0751, 0.0306], + device='cuda:3'), in_proj_covar=tensor([0.0434, 0.0375, 0.0319, 0.0431, 0.0358, 0.0515, 0.0380, 0.0396], + device='cuda:3'), out_proj_covar=tensor([1.1865e-04, 9.9762e-05, 8.4682e-05, 1.1547e-04, 9.6214e-05, 1.4857e-04, + 1.0381e-04, 1.0618e-04], device='cuda:3') +2023-02-06 19:52:40,020 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8003, 2.1533, 2.3243, 1.4721, 2.3900, 1.6633, 0.7500, 1.9388], + device='cuda:3'), covar=tensor([0.0487, 0.0225, 0.0208, 0.0421, 0.0245, 0.0629, 0.0649, 0.0254], + device='cuda:3'), in_proj_covar=tensor([0.0432, 0.0373, 0.0318, 0.0429, 0.0356, 0.0512, 0.0378, 0.0394], + device='cuda:3'), out_proj_covar=tensor([1.1802e-04, 9.9351e-05, 8.4339e-05, 1.1490e-04, 9.5670e-05, 1.4765e-04, + 1.0323e-04, 1.0562e-04], device='cuda:3') +2023-02-06 19:52:41,877 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.639e+02 2.442e+02 2.775e+02 3.418e+02 6.006e+02, threshold=5.549e+02, percent-clipped=0.0 +2023-02-06 19:52:48,333 INFO [train.py:901] (3/4) Epoch 17, batch 3650, loss[loss=0.2058, simple_loss=0.2795, pruned_loss=0.06608, over 8236.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2954, pruned_loss=0.06815, over 1616327.75 frames. ], batch size: 22, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:53:18,533 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=133022.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:53:21,778 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 19:53:23,087 INFO [train.py:901] (3/4) Epoch 17, batch 3700, loss[loss=0.1959, simple_loss=0.2628, pruned_loss=0.06449, over 7234.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2959, pruned_loss=0.06855, over 1618420.06 frames. ], batch size: 16, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:53:37,038 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 19:53:53,566 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.474e+02 2.419e+02 3.081e+02 4.194e+02 7.364e+02, threshold=6.162e+02, percent-clipped=6.0 +2023-02-06 19:53:59,120 INFO [train.py:901] (3/4) Epoch 17, batch 3750, loss[loss=0.1887, simple_loss=0.2665, pruned_loss=0.05542, over 7702.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2949, pruned_loss=0.06796, over 1613155.23 frames. ], batch size: 18, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:54:10,364 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.4014, 5.3689, 4.8049, 2.3585, 4.7977, 5.1126, 4.9971, 4.7938], + device='cuda:3'), covar=tensor([0.0594, 0.0460, 0.0888, 0.4790, 0.0833, 0.0820, 0.1105, 0.0660], + device='cuda:3'), in_proj_covar=tensor([0.0504, 0.0412, 0.0419, 0.0514, 0.0408, 0.0416, 0.0400, 0.0360], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 19:54:21,508 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=133108.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:54:35,509 INFO [train.py:901] (3/4) Epoch 17, batch 3800, loss[loss=0.2159, simple_loss=0.2972, pruned_loss=0.06727, over 8142.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2966, pruned_loss=0.06905, over 1615119.64 frames. ], batch size: 22, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:54:41,291 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133137.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:55:04,572 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.494e+02 2.594e+02 3.054e+02 3.718e+02 6.772e+02, threshold=6.108e+02, percent-clipped=5.0 +2023-02-06 19:55:09,934 INFO [train.py:901] (3/4) Epoch 17, batch 3850, loss[loss=0.2159, simple_loss=0.2836, pruned_loss=0.07409, over 7935.00 frames. ], tot_loss[loss=0.217, simple_loss=0.2963, pruned_loss=0.06885, over 1614638.64 frames. ], batch size: 20, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:55:31,129 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 19:55:39,721 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2184, 1.9849, 2.7190, 2.1985, 2.5836, 2.1919, 1.9340, 1.4101], + device='cuda:3'), covar=tensor([0.4652, 0.4631, 0.1616, 0.3500, 0.2494, 0.2773, 0.1841, 0.4904], + device='cuda:3'), in_proj_covar=tensor([0.0918, 0.0931, 0.0770, 0.0902, 0.0968, 0.0850, 0.0721, 0.0800], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 19:55:43,019 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133223.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:55:46,893 INFO [train.py:901] (3/4) Epoch 17, batch 3900, loss[loss=0.177, simple_loss=0.2694, pruned_loss=0.04226, over 8192.00 frames. ], tot_loss[loss=0.218, simple_loss=0.2974, pruned_loss=0.06931, over 1614076.07 frames. ], batch size: 23, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:56:15,759 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 2.486e+02 2.968e+02 4.028e+02 1.073e+03, threshold=5.936e+02, percent-clipped=5.0 +2023-02-06 19:56:21,114 INFO [train.py:901] (3/4) Epoch 17, batch 3950, loss[loss=0.2264, simple_loss=0.3012, pruned_loss=0.07583, over 7797.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2967, pruned_loss=0.06914, over 1614163.91 frames. ], batch size: 20, lr: 4.48e-03, grad_scale: 8.0 +2023-02-06 19:56:53,886 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.94 vs. limit=5.0 +2023-02-06 19:56:56,947 INFO [train.py:901] (3/4) Epoch 17, batch 4000, loss[loss=0.2139, simple_loss=0.3, pruned_loss=0.06389, over 8513.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2959, pruned_loss=0.06851, over 1613539.17 frames. ], batch size: 29, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 19:57:27,408 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.588e+02 2.526e+02 3.333e+02 3.995e+02 7.649e+02, threshold=6.666e+02, percent-clipped=5.0 +2023-02-06 19:57:32,346 INFO [train.py:901] (3/4) Epoch 17, batch 4050, loss[loss=0.2041, simple_loss=0.2967, pruned_loss=0.05574, over 8108.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2967, pruned_loss=0.0688, over 1620234.72 frames. ], batch size: 23, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 19:57:41,378 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133392.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 19:57:42,113 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133393.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:57:42,760 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133394.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:57:59,639 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133418.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:58:07,686 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2023-02-06 19:58:07,783 INFO [train.py:901] (3/4) Epoch 17, batch 4100, loss[loss=0.2462, simple_loss=0.3212, pruned_loss=0.08555, over 8245.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2973, pruned_loss=0.0693, over 1621214.22 frames. ], batch size: 22, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 19:58:40,085 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.726e+02 2.513e+02 2.919e+02 3.658e+02 1.440e+03, threshold=5.839e+02, percent-clipped=2.0 +2023-02-06 19:58:45,031 INFO [train.py:901] (3/4) Epoch 17, batch 4150, loss[loss=0.2361, simple_loss=0.3178, pruned_loss=0.07719, over 8339.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2959, pruned_loss=0.06853, over 1618938.61 frames. ], batch size: 26, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 19:58:45,262 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133479.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:58:54,027 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133492.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:59:02,306 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133504.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 19:59:14,062 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5481, 1.2989, 4.6971, 1.7695, 4.1855, 3.8705, 4.3173, 4.1590], + device='cuda:3'), covar=tensor([0.0512, 0.4964, 0.0537, 0.4051, 0.1143, 0.1102, 0.0502, 0.0641], + device='cuda:3'), in_proj_covar=tensor([0.0571, 0.0622, 0.0662, 0.0595, 0.0671, 0.0582, 0.0570, 0.0635], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 19:59:19,507 INFO [train.py:901] (3/4) Epoch 17, batch 4200, loss[loss=0.216, simple_loss=0.2949, pruned_loss=0.06855, over 8032.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2966, pruned_loss=0.06879, over 1620686.58 frames. ], batch size: 22, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 19:59:32,473 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 19:59:39,304 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.7836, 1.3565, 3.9194, 1.4563, 3.5124, 3.2559, 3.5393, 3.4348], + device='cuda:3'), covar=tensor([0.0611, 0.4457, 0.0730, 0.4140, 0.1129, 0.1115, 0.0694, 0.0782], + device='cuda:3'), in_proj_covar=tensor([0.0573, 0.0623, 0.0664, 0.0595, 0.0673, 0.0583, 0.0572, 0.0636], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 19:59:51,074 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.728e+02 2.565e+02 3.135e+02 3.827e+02 1.180e+03, threshold=6.269e+02, percent-clipped=6.0 +2023-02-06 19:59:56,778 INFO [train.py:901] (3/4) Epoch 17, batch 4250, loss[loss=0.1793, simple_loss=0.2718, pruned_loss=0.04337, over 8240.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2962, pruned_loss=0.06832, over 1623159.08 frames. ], batch size: 22, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 19:59:57,447 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 20:00:03,426 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 +2023-02-06 20:00:05,318 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.74 vs. limit=5.0 +2023-02-06 20:00:30,990 INFO [train.py:901] (3/4) Epoch 17, batch 4300, loss[loss=0.2131, simple_loss=0.3088, pruned_loss=0.05871, over 8513.00 frames. ], tot_loss[loss=0.216, simple_loss=0.2956, pruned_loss=0.06823, over 1620805.39 frames. ], batch size: 26, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 20:01:00,697 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133670.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:01:01,925 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.491e+02 2.551e+02 3.118e+02 3.976e+02 6.360e+02, threshold=6.236e+02, percent-clipped=1.0 +2023-02-06 20:01:06,900 INFO [train.py:901] (3/4) Epoch 17, batch 4350, loss[loss=0.2874, simple_loss=0.3548, pruned_loss=0.11, over 6974.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2958, pruned_loss=0.06819, over 1619702.66 frames. ], batch size: 72, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 20:01:18,441 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5316, 2.0159, 3.3476, 1.3606, 2.5541, 2.0222, 1.5561, 2.5582], + device='cuda:3'), covar=tensor([0.1791, 0.2352, 0.0702, 0.4218, 0.1552, 0.2892, 0.2217, 0.2002], + device='cuda:3'), in_proj_covar=tensor([0.0510, 0.0571, 0.0550, 0.0621, 0.0637, 0.0576, 0.0514, 0.0627], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 20:01:25,375 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-06 20:01:31,292 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 20:01:43,135 INFO [train.py:901] (3/4) Epoch 17, batch 4400, loss[loss=0.1984, simple_loss=0.3001, pruned_loss=0.04835, over 8362.00 frames. ], tot_loss[loss=0.2166, simple_loss=0.297, pruned_loss=0.06814, over 1624287.77 frames. ], batch size: 24, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 20:01:48,106 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=133736.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 20:01:49,422 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=133738.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:02:12,878 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.820e+02 2.579e+02 3.148e+02 3.884e+02 8.584e+02, threshold=6.297e+02, percent-clipped=6.0 +2023-02-06 20:02:13,681 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 20:02:18,533 INFO [train.py:901] (3/4) Epoch 17, batch 4450, loss[loss=0.2024, simple_loss=0.2815, pruned_loss=0.06166, over 8191.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2948, pruned_loss=0.06746, over 1621929.07 frames. ], batch size: 23, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 20:02:55,024 INFO [train.py:901] (3/4) Epoch 17, batch 4500, loss[loss=0.1758, simple_loss=0.2511, pruned_loss=0.05028, over 7431.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.2939, pruned_loss=0.06767, over 1615437.56 frames. ], batch size: 17, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 20:03:00,079 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=133836.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:03:10,432 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133851.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 20:03:10,911 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 20:03:11,739 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133853.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:03:24,313 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.703e+02 2.330e+02 2.856e+02 3.592e+02 8.327e+02, threshold=5.711e+02, percent-clipped=1.0 +2023-02-06 20:03:29,183 INFO [train.py:901] (3/4) Epoch 17, batch 4550, loss[loss=0.1821, simple_loss=0.275, pruned_loss=0.04465, over 8136.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2943, pruned_loss=0.06774, over 1612479.25 frames. ], batch size: 22, lr: 4.47e-03, grad_scale: 8.0 +2023-02-06 20:03:32,459 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.68 vs. limit=5.0 +2023-02-06 20:04:04,549 INFO [train.py:901] (3/4) Epoch 17, batch 4600, loss[loss=0.2564, simple_loss=0.3239, pruned_loss=0.0944, over 8620.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2953, pruned_loss=0.06849, over 1613030.42 frames. ], batch size: 49, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:04:19,774 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-06 20:04:21,359 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133951.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:04:35,425 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.660e+02 2.386e+02 2.834e+02 3.772e+02 7.696e+02, threshold=5.668e+02, percent-clipped=3.0 +2023-02-06 20:04:40,234 INFO [train.py:901] (3/4) Epoch 17, batch 4650, loss[loss=0.2239, simple_loss=0.3056, pruned_loss=0.07117, over 8564.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2952, pruned_loss=0.06821, over 1615490.37 frames. ], batch size: 39, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:05:06,465 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=134014.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:05:07,195 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134015.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:05:16,664 INFO [train.py:901] (3/4) Epoch 17, batch 4700, loss[loss=0.1916, simple_loss=0.2752, pruned_loss=0.05395, over 8037.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2935, pruned_loss=0.0673, over 1613786.71 frames. ], batch size: 22, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:05:48,988 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.622e+02 2.331e+02 2.674e+02 3.349e+02 6.559e+02, threshold=5.348e+02, percent-clipped=3.0 +2023-02-06 20:05:53,966 INFO [train.py:901] (3/4) Epoch 17, batch 4750, loss[loss=0.2063, simple_loss=0.2828, pruned_loss=0.06494, over 8089.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2934, pruned_loss=0.06696, over 1612906.66 frames. ], batch size: 21, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:06:13,302 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134107.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 20:06:14,648 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134109.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:06:17,896 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 20:06:20,573 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 20:06:28,183 INFO [train.py:901] (3/4) Epoch 17, batch 4800, loss[loss=0.2413, simple_loss=0.3176, pruned_loss=0.08252, over 8600.00 frames. ], tot_loss[loss=0.2147, simple_loss=0.2942, pruned_loss=0.06761, over 1612609.38 frames. ], batch size: 31, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:06:29,214 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134129.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:06:31,288 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134132.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 20:06:32,691 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134134.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:07:00,734 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.617e+02 2.352e+02 2.869e+02 3.488e+02 8.440e+02, threshold=5.739e+02, percent-clipped=9.0 +2023-02-06 20:07:06,361 INFO [train.py:901] (3/4) Epoch 17, batch 4850, loss[loss=0.2071, simple_loss=0.2968, pruned_loss=0.05865, over 8027.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2944, pruned_loss=0.06782, over 1609679.56 frames. ], batch size: 22, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:07:14,663 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 20:07:26,226 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134207.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:07:41,127 INFO [train.py:901] (3/4) Epoch 17, batch 4900, loss[loss=0.2301, simple_loss=0.3119, pruned_loss=0.07417, over 8600.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2935, pruned_loss=0.06721, over 1611906.62 frames. ], batch size: 31, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:07:43,514 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134232.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:08:13,116 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.679e+02 2.657e+02 3.351e+02 4.707e+02 1.168e+03, threshold=6.701e+02, percent-clipped=12.0 +2023-02-06 20:08:17,820 INFO [train.py:901] (3/4) Epoch 17, batch 4950, loss[loss=0.2425, simple_loss=0.3099, pruned_loss=0.08755, over 7325.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2942, pruned_loss=0.06722, over 1614214.95 frames. ], batch size: 71, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:08:49,559 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134322.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:08:52,984 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134327.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:08:54,217 INFO [train.py:901] (3/4) Epoch 17, batch 5000, loss[loss=0.2131, simple_loss=0.2934, pruned_loss=0.06643, over 7821.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2932, pruned_loss=0.06667, over 1611285.41 frames. ], batch size: 20, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:09:09,866 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5217, 2.6304, 1.7614, 2.2010, 2.1037, 1.6039, 2.0367, 2.0537], + device='cuda:3'), covar=tensor([0.1563, 0.0382, 0.1214, 0.0672, 0.0704, 0.1442, 0.1020, 0.0991], + device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0234, 0.0327, 0.0302, 0.0297, 0.0332, 0.0342, 0.0315], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 20:09:15,213 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=134359.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:09:24,853 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.543e+02 2.361e+02 2.656e+02 3.405e+02 6.362e+02, threshold=5.311e+02, percent-clipped=0.0 +2023-02-06 20:09:30,487 INFO [train.py:901] (3/4) Epoch 17, batch 5050, loss[loss=0.1919, simple_loss=0.2792, pruned_loss=0.05228, over 8727.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2933, pruned_loss=0.06656, over 1613870.92 frames. ], batch size: 34, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:09:31,735 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 20:09:35,089 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134385.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:09:54,066 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134410.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:09:58,691 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 20:10:05,232 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8241, 2.3682, 3.4101, 1.9598, 1.8128, 3.4570, 0.7376, 2.1078], + device='cuda:3'), covar=tensor([0.1733, 0.1544, 0.0338, 0.1983, 0.2931, 0.0284, 0.2654, 0.1527], + device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0186, 0.0117, 0.0221, 0.0264, 0.0124, 0.0168, 0.0184], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 20:10:07,165 INFO [train.py:901] (3/4) Epoch 17, batch 5100, loss[loss=0.2074, simple_loss=0.2963, pruned_loss=0.05927, over 8475.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2936, pruned_loss=0.06662, over 1619237.10 frames. ], batch size: 25, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:10:12,406 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8376, 1.9213, 2.4189, 1.6315, 1.4369, 2.4703, 0.5126, 1.5166], + device='cuda:3'), covar=tensor([0.2203, 0.1264, 0.0577, 0.1751, 0.3115, 0.0631, 0.2619, 0.2023], + device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0185, 0.0116, 0.0220, 0.0263, 0.0123, 0.0167, 0.0183], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 20:10:36,989 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.440e+02 2.384e+02 2.769e+02 3.675e+02 1.185e+03, threshold=5.538e+02, percent-clipped=9.0 +2023-02-06 20:10:38,569 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134474.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:10:42,699 INFO [train.py:901] (3/4) Epoch 17, batch 5150, loss[loss=0.2084, simple_loss=0.2833, pruned_loss=0.06675, over 7803.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2938, pruned_loss=0.06661, over 1616438.96 frames. ], batch size: 20, lr: 4.46e-03, grad_scale: 8.0 +2023-02-06 20:11:20,275 INFO [train.py:901] (3/4) Epoch 17, batch 5200, loss[loss=0.238, simple_loss=0.3203, pruned_loss=0.07779, over 7966.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.2939, pruned_loss=0.0668, over 1615415.32 frames. ], batch size: 21, lr: 4.45e-03, grad_scale: 8.0 +2023-02-06 20:11:49,978 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.450e+02 2.225e+02 2.783e+02 3.706e+02 1.482e+03, threshold=5.567e+02, percent-clipped=8.0 +2023-02-06 20:11:54,881 INFO [train.py:901] (3/4) Epoch 17, batch 5250, loss[loss=0.2142, simple_loss=0.287, pruned_loss=0.07073, over 8749.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2943, pruned_loss=0.06695, over 1617707.43 frames. ], batch size: 39, lr: 4.45e-03, grad_scale: 8.0 +2023-02-06 20:11:57,589 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 20:11:57,995 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 +2023-02-06 20:12:09,111 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8209, 1.5183, 3.9370, 1.4696, 3.4683, 3.2522, 3.6014, 3.4439], + device='cuda:3'), covar=tensor([0.0657, 0.4559, 0.0692, 0.4150, 0.1322, 0.1038, 0.0707, 0.0821], + device='cuda:3'), in_proj_covar=tensor([0.0581, 0.0621, 0.0663, 0.0593, 0.0674, 0.0579, 0.0574, 0.0641], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 20:12:31,033 INFO [train.py:901] (3/4) Epoch 17, batch 5300, loss[loss=0.1668, simple_loss=0.2433, pruned_loss=0.04518, over 8083.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.293, pruned_loss=0.06582, over 1619746.35 frames. ], batch size: 21, lr: 4.45e-03, grad_scale: 8.0 +2023-02-06 20:12:34,158 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5559, 1.6357, 4.7133, 1.8390, 4.2120, 3.9543, 4.3136, 4.1564], + device='cuda:3'), covar=tensor([0.0563, 0.4632, 0.0520, 0.3911, 0.1066, 0.0989, 0.0585, 0.0648], + device='cuda:3'), in_proj_covar=tensor([0.0583, 0.0623, 0.0664, 0.0595, 0.0676, 0.0580, 0.0576, 0.0642], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 20:12:48,147 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3296, 2.0347, 2.7951, 2.2996, 2.7163, 2.2400, 2.0358, 1.5836], + device='cuda:3'), covar=tensor([0.4490, 0.4382, 0.1499, 0.2733, 0.1904, 0.2546, 0.1670, 0.4380], + device='cuda:3'), in_proj_covar=tensor([0.0913, 0.0927, 0.0771, 0.0901, 0.0965, 0.0847, 0.0719, 0.0798], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 20:12:58,386 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=134666.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:13:00,526 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5354, 2.6066, 1.8485, 2.2988, 2.1406, 1.6099, 2.0856, 2.0891], + device='cuda:3'), covar=tensor([0.1349, 0.0342, 0.1142, 0.0558, 0.0727, 0.1519, 0.0965, 0.0890], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0232, 0.0325, 0.0301, 0.0296, 0.0330, 0.0340, 0.0312], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 20:13:01,795 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=134671.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:13:02,357 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.561e+02 2.333e+02 2.884e+02 3.429e+02 1.143e+03, threshold=5.769e+02, percent-clipped=6.0 +2023-02-06 20:13:05,298 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0608, 1.6424, 1.3805, 1.6355, 1.3703, 1.2459, 1.3468, 1.3257], + device='cuda:3'), covar=tensor([0.1027, 0.0420, 0.1168, 0.0483, 0.0670, 0.1356, 0.0785, 0.0817], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0232, 0.0325, 0.0301, 0.0296, 0.0331, 0.0340, 0.0312], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 20:13:07,134 INFO [train.py:901] (3/4) Epoch 17, batch 5350, loss[loss=0.1923, simple_loss=0.2857, pruned_loss=0.04941, over 8328.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.2946, pruned_loss=0.0664, over 1620233.77 frames. ], batch size: 25, lr: 4.45e-03, grad_scale: 8.0 +2023-02-06 20:13:26,623 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6881, 2.0253, 2.2013, 1.2669, 2.3570, 1.5055, 0.6950, 1.8840], + device='cuda:3'), covar=tensor([0.0572, 0.0339, 0.0238, 0.0617, 0.0361, 0.0820, 0.0771, 0.0279], + device='cuda:3'), in_proj_covar=tensor([0.0438, 0.0376, 0.0320, 0.0434, 0.0359, 0.0520, 0.0380, 0.0397], + device='cuda:3'), out_proj_covar=tensor([1.1944e-04, 1.0013e-04, 8.4772e-05, 1.1600e-04, 9.6398e-05, 1.5007e-04, + 1.0359e-04, 1.0650e-04], device='cuda:3') +2023-02-06 20:13:43,340 INFO [train.py:901] (3/4) Epoch 17, batch 5400, loss[loss=0.1863, simple_loss=0.2683, pruned_loss=0.05217, over 7966.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.296, pruned_loss=0.06644, over 1624560.85 frames. ], batch size: 21, lr: 4.45e-03, grad_scale: 8.0 +2023-02-06 20:13:44,314 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134730.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:14:01,981 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134755.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:14:14,310 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.461e+02 2.337e+02 2.988e+02 3.635e+02 1.067e+03, threshold=5.976e+02, percent-clipped=7.0 +2023-02-06 20:14:18,986 INFO [train.py:901] (3/4) Epoch 17, batch 5450, loss[loss=0.2495, simple_loss=0.3262, pruned_loss=0.08637, over 8114.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2964, pruned_loss=0.06739, over 1625101.56 frames. ], batch size: 23, lr: 4.45e-03, grad_scale: 8.0 +2023-02-06 20:14:20,487 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134781.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:14:24,001 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134786.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:14:54,786 INFO [train.py:901] (3/4) Epoch 17, batch 5500, loss[loss=0.2591, simple_loss=0.3253, pruned_loss=0.09649, over 8464.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2959, pruned_loss=0.06717, over 1624671.60 frames. ], batch size: 27, lr: 4.45e-03, grad_scale: 8.0 +2023-02-06 20:14:55,407 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 20:15:00,395 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5458, 1.7794, 1.8423, 1.1823, 1.9423, 1.3355, 0.4458, 1.7254], + device='cuda:3'), covar=tensor([0.0439, 0.0309, 0.0212, 0.0466, 0.0311, 0.0794, 0.0731, 0.0221], + device='cuda:3'), in_proj_covar=tensor([0.0434, 0.0373, 0.0316, 0.0428, 0.0355, 0.0514, 0.0375, 0.0394], + device='cuda:3'), out_proj_covar=tensor([1.1841e-04, 9.9318e-05, 8.3595e-05, 1.1440e-04, 9.4999e-05, 1.4805e-04, + 1.0213e-04, 1.0556e-04], device='cuda:3') +2023-02-06 20:15:25,536 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.381e+02 2.895e+02 3.783e+02 8.489e+02, threshold=5.790e+02, percent-clipped=3.0 +2023-02-06 20:15:31,376 INFO [train.py:901] (3/4) Epoch 17, batch 5550, loss[loss=0.2204, simple_loss=0.3064, pruned_loss=0.06725, over 8193.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2946, pruned_loss=0.06684, over 1621194.61 frames. ], batch size: 23, lr: 4.45e-03, grad_scale: 8.0 +2023-02-06 20:15:32,207 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134880.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:15:42,878 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.92 vs. limit=5.0 +2023-02-06 20:15:58,385 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.99 vs. limit=5.0 +2023-02-06 20:16:06,822 INFO [train.py:901] (3/4) Epoch 17, batch 5600, loss[loss=0.2287, simple_loss=0.3243, pruned_loss=0.06654, over 8200.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2927, pruned_loss=0.06612, over 1613378.42 frames. ], batch size: 23, lr: 4.45e-03, grad_scale: 8.0 +2023-02-06 20:16:12,376 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7991, 1.5461, 3.1347, 1.4668, 2.2472, 3.3293, 3.4894, 2.8430], + device='cuda:3'), covar=tensor([0.1177, 0.1699, 0.0387, 0.2046, 0.0993, 0.0280, 0.0600, 0.0620], + device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0311, 0.0276, 0.0302, 0.0295, 0.0255, 0.0390, 0.0301], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 20:16:23,704 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1942, 1.6711, 1.2816, 1.6319, 1.4202, 1.1197, 1.4553, 1.4826], + device='cuda:3'), covar=tensor([0.0740, 0.0322, 0.0822, 0.0351, 0.0526, 0.1039, 0.0581, 0.0523], + device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0232, 0.0323, 0.0301, 0.0296, 0.0330, 0.0340, 0.0312], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 20:16:32,068 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8936, 1.6042, 2.0412, 1.7622, 1.9864, 1.9169, 1.6966, 0.8155], + device='cuda:3'), covar=tensor([0.5265, 0.4401, 0.1740, 0.3189, 0.2257, 0.2761, 0.1888, 0.4774], + device='cuda:3'), in_proj_covar=tensor([0.0914, 0.0929, 0.0773, 0.0902, 0.0972, 0.0849, 0.0721, 0.0800], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 20:16:38,748 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.780e+02 2.724e+02 3.292e+02 4.135e+02 9.276e+02, threshold=6.584e+02, percent-clipped=7.0 +2023-02-06 20:16:42,891 INFO [train.py:901] (3/4) Epoch 17, batch 5650, loss[loss=0.1782, simple_loss=0.2635, pruned_loss=0.04642, over 7648.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.2939, pruned_loss=0.0668, over 1612821.64 frames. ], batch size: 19, lr: 4.45e-03, grad_scale: 4.0 +2023-02-06 20:17:04,390 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 20:17:18,720 INFO [train.py:901] (3/4) Epoch 17, batch 5700, loss[loss=0.1631, simple_loss=0.25, pruned_loss=0.03813, over 6433.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2941, pruned_loss=0.06744, over 1609699.92 frames. ], batch size: 14, lr: 4.45e-03, grad_scale: 4.0 +2023-02-06 20:17:24,519 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135037.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:17:28,020 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135042.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:17:42,577 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135062.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:17:45,955 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135067.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:17:49,719 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.508e+02 2.516e+02 3.214e+02 3.973e+02 1.283e+03, threshold=6.427e+02, percent-clipped=6.0 +2023-02-06 20:17:53,701 INFO [train.py:901] (3/4) Epoch 17, batch 5750, loss[loss=0.1808, simple_loss=0.2621, pruned_loss=0.04976, over 8146.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2936, pruned_loss=0.06678, over 1609526.79 frames. ], batch size: 22, lr: 4.45e-03, grad_scale: 4.0 +2023-02-06 20:18:11,550 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 20:18:13,321 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 +2023-02-06 20:18:17,698 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4121, 4.3276, 3.9699, 2.1688, 3.8278, 3.9263, 3.9946, 3.6313], + device='cuda:3'), covar=tensor([0.0751, 0.0589, 0.1089, 0.4785, 0.0905, 0.1033, 0.1360, 0.0826], + device='cuda:3'), in_proj_covar=tensor([0.0508, 0.0417, 0.0423, 0.0525, 0.0414, 0.0417, 0.0407, 0.0364], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 20:18:30,185 INFO [train.py:901] (3/4) Epoch 17, batch 5800, loss[loss=0.2541, simple_loss=0.3218, pruned_loss=0.09325, over 7652.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2942, pruned_loss=0.06691, over 1615542.77 frames. ], batch size: 19, lr: 4.44e-03, grad_scale: 4.0 +2023-02-06 20:18:47,013 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5582, 2.2719, 3.2262, 2.6258, 3.0367, 2.4565, 2.1654, 1.7491], + device='cuda:3'), covar=tensor([0.4784, 0.4928, 0.1493, 0.3151, 0.2405, 0.2598, 0.1780, 0.5085], + device='cuda:3'), in_proj_covar=tensor([0.0925, 0.0938, 0.0778, 0.0908, 0.0978, 0.0856, 0.0726, 0.0804], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 20:19:00,545 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.513e+02 2.313e+02 2.882e+02 3.681e+02 6.576e+02, threshold=5.764e+02, percent-clipped=1.0 +2023-02-06 20:19:04,567 INFO [train.py:901] (3/4) Epoch 17, batch 5850, loss[loss=0.2558, simple_loss=0.3363, pruned_loss=0.08762, over 8461.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.2952, pruned_loss=0.06704, over 1618999.52 frames. ], batch size: 29, lr: 4.44e-03, grad_scale: 4.0 +2023-02-06 20:19:12,154 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=135189.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:19:37,646 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=135224.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:19:41,092 INFO [train.py:901] (3/4) Epoch 17, batch 5900, loss[loss=0.2213, simple_loss=0.3112, pruned_loss=0.06569, over 8490.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2945, pruned_loss=0.06693, over 1616004.41 frames. ], batch size: 29, lr: 4.44e-03, grad_scale: 4.0 +2023-02-06 20:20:12,437 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.753e+02 2.357e+02 3.084e+02 3.660e+02 6.807e+02, threshold=6.167e+02, percent-clipped=2.0 +2023-02-06 20:20:16,628 INFO [train.py:901] (3/4) Epoch 17, batch 5950, loss[loss=0.2179, simple_loss=0.3086, pruned_loss=0.06362, over 8448.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2948, pruned_loss=0.06692, over 1621118.28 frames. ], batch size: 27, lr: 4.44e-03, grad_scale: 4.0 +2023-02-06 20:20:52,281 INFO [train.py:901] (3/4) Epoch 17, batch 6000, loss[loss=0.1973, simple_loss=0.2793, pruned_loss=0.05765, over 7924.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.295, pruned_loss=0.06762, over 1623099.20 frames. ], batch size: 20, lr: 4.44e-03, grad_scale: 8.0 +2023-02-06 20:20:52,281 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 20:21:02,022 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8176, 3.7988, 3.4541, 2.1810, 3.3243, 3.4635, 3.5014, 3.2135], + device='cuda:3'), covar=tensor([0.0914, 0.0594, 0.1009, 0.4503, 0.1072, 0.1039, 0.1206, 0.1085], + device='cuda:3'), in_proj_covar=tensor([0.0501, 0.0412, 0.0417, 0.0518, 0.0406, 0.0410, 0.0401, 0.0360], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 20:21:05,418 INFO [train.py:935] (3/4) Epoch 17, validation: loss=0.1774, simple_loss=0.2777, pruned_loss=0.03857, over 944034.00 frames. +2023-02-06 20:21:05,419 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 20:21:12,601 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=135339.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:21:16,156 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5055, 1.4629, 1.8156, 1.3372, 1.1673, 1.8067, 0.1558, 1.1759], + device='cuda:3'), covar=tensor([0.1807, 0.1359, 0.0419, 0.1048, 0.2831, 0.0464, 0.2172, 0.1297], + device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0184, 0.0115, 0.0216, 0.0260, 0.0123, 0.0165, 0.0180], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 20:21:36,686 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.411e+02 3.026e+02 3.580e+02 8.983e+02, threshold=6.051e+02, percent-clipped=2.0 +2023-02-06 20:21:40,882 INFO [train.py:901] (3/4) Epoch 17, batch 6050, loss[loss=0.1723, simple_loss=0.245, pruned_loss=0.04983, over 7222.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2936, pruned_loss=0.06766, over 1611796.47 frames. ], batch size: 16, lr: 4.44e-03, grad_scale: 8.0 +2023-02-06 20:21:41,759 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8007, 3.6668, 3.3783, 1.8870, 3.3123, 3.3082, 3.4261, 3.0040], + device='cuda:3'), covar=tensor([0.0927, 0.0710, 0.1183, 0.4684, 0.1038, 0.1236, 0.1330, 0.1051], + device='cuda:3'), in_proj_covar=tensor([0.0501, 0.0412, 0.0416, 0.0517, 0.0406, 0.0410, 0.0400, 0.0359], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 20:22:16,317 INFO [train.py:901] (3/4) Epoch 17, batch 6100, loss[loss=0.2412, simple_loss=0.3119, pruned_loss=0.08521, over 8251.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2947, pruned_loss=0.06831, over 1609843.55 frames. ], batch size: 22, lr: 4.44e-03, grad_scale: 8.0 +2023-02-06 20:22:47,563 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.454e+02 2.459e+02 2.890e+02 3.783e+02 6.848e+02, threshold=5.780e+02, percent-clipped=3.0 +2023-02-06 20:22:49,621 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 20:22:52,257 INFO [train.py:901] (3/4) Epoch 17, batch 6150, loss[loss=0.1717, simple_loss=0.2487, pruned_loss=0.04739, over 7207.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2958, pruned_loss=0.06885, over 1613224.32 frames. ], batch size: 16, lr: 4.44e-03, grad_scale: 8.0 +2023-02-06 20:22:55,588 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.8227, 5.9181, 5.0377, 2.5738, 5.1921, 5.6083, 5.3850, 5.2298], + device='cuda:3'), covar=tensor([0.0487, 0.0337, 0.0859, 0.3877, 0.0712, 0.0581, 0.0964, 0.0506], + device='cuda:3'), in_proj_covar=tensor([0.0502, 0.0413, 0.0418, 0.0519, 0.0409, 0.0413, 0.0402, 0.0360], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 20:23:26,588 INFO [train.py:901] (3/4) Epoch 17, batch 6200, loss[loss=0.2241, simple_loss=0.3004, pruned_loss=0.07395, over 8246.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2951, pruned_loss=0.06787, over 1615888.49 frames. ], batch size: 24, lr: 4.44e-03, grad_scale: 8.0 +2023-02-06 20:23:29,372 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=135533.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:23:57,601 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.581e+02 2.470e+02 3.035e+02 3.550e+02 6.137e+02, threshold=6.070e+02, percent-clipped=1.0 +2023-02-06 20:24:01,728 INFO [train.py:901] (3/4) Epoch 17, batch 6250, loss[loss=0.2047, simple_loss=0.2718, pruned_loss=0.06881, over 7787.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2955, pruned_loss=0.06803, over 1617948.93 frames. ], batch size: 19, lr: 4.44e-03, grad_scale: 8.0 +2023-02-06 20:24:13,157 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135595.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:24:29,700 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-06 20:24:30,226 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135620.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:24:36,951 INFO [train.py:901] (3/4) Epoch 17, batch 6300, loss[loss=0.1937, simple_loss=0.2791, pruned_loss=0.05409, over 8288.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2946, pruned_loss=0.06777, over 1614732.43 frames. ], batch size: 23, lr: 4.44e-03, grad_scale: 8.0 +2023-02-06 20:24:49,740 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=135648.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:24:49,776 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=135648.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:25:07,385 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.756e+02 2.700e+02 3.426e+02 4.477e+02 8.691e+02, threshold=6.853e+02, percent-clipped=8.0 +2023-02-06 20:25:11,443 INFO [train.py:901] (3/4) Epoch 17, batch 6350, loss[loss=0.192, simple_loss=0.269, pruned_loss=0.0575, over 7925.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2961, pruned_loss=0.06871, over 1619475.13 frames. ], batch size: 20, lr: 4.44e-03, grad_scale: 8.0 +2023-02-06 20:25:46,617 INFO [train.py:901] (3/4) Epoch 17, batch 6400, loss[loss=0.1923, simple_loss=0.2934, pruned_loss=0.04565, over 8615.00 frames. ], tot_loss[loss=0.2166, simple_loss=0.2961, pruned_loss=0.06854, over 1618232.17 frames. ], batch size: 39, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:26:16,681 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 2.216e+02 2.648e+02 3.143e+02 6.334e+02, threshold=5.295e+02, percent-clipped=0.0 +2023-02-06 20:26:20,488 INFO [train.py:901] (3/4) Epoch 17, batch 6450, loss[loss=0.2302, simple_loss=0.327, pruned_loss=0.06671, over 8765.00 frames. ], tot_loss[loss=0.2174, simple_loss=0.2967, pruned_loss=0.06904, over 1620103.19 frames. ], batch size: 40, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:26:39,796 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4075, 1.4852, 1.3913, 1.8570, 0.7958, 1.2171, 1.2660, 1.4597], + device='cuda:3'), covar=tensor([0.0841, 0.0773, 0.1070, 0.0473, 0.1059, 0.1427, 0.0781, 0.0737], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0198, 0.0248, 0.0211, 0.0209, 0.0248, 0.0256, 0.0211], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 20:26:56,417 INFO [train.py:901] (3/4) Epoch 17, batch 6500, loss[loss=0.2046, simple_loss=0.2956, pruned_loss=0.05678, over 8465.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2974, pruned_loss=0.06961, over 1620651.60 frames. ], batch size: 25, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:27:27,349 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.809e+02 2.443e+02 3.095e+02 4.367e+02 8.897e+02, threshold=6.190e+02, percent-clipped=12.0 +2023-02-06 20:27:31,534 INFO [train.py:901] (3/4) Epoch 17, batch 6550, loss[loss=0.1992, simple_loss=0.2863, pruned_loss=0.05607, over 8240.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2961, pruned_loss=0.06879, over 1620039.55 frames. ], batch size: 24, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:27:44,287 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2695, 1.6942, 1.7343, 1.0023, 1.6618, 1.2885, 0.2779, 1.6631], + device='cuda:3'), covar=tensor([0.0415, 0.0257, 0.0199, 0.0445, 0.0317, 0.0686, 0.0672, 0.0199], + device='cuda:3'), in_proj_covar=tensor([0.0430, 0.0372, 0.0320, 0.0425, 0.0353, 0.0512, 0.0374, 0.0395], + device='cuda:3'), out_proj_covar=tensor([1.1723e-04, 9.8929e-05, 8.4846e-05, 1.1351e-04, 9.4554e-05, 1.4739e-04, + 1.0180e-04, 1.0568e-04], device='cuda:3') +2023-02-06 20:27:48,288 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135904.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:27:56,446 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 20:28:06,536 INFO [train.py:901] (3/4) Epoch 17, batch 6600, loss[loss=0.2158, simple_loss=0.2968, pruned_loss=0.06739, over 7652.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2966, pruned_loss=0.06896, over 1620879.90 frames. ], batch size: 19, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:28:06,734 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135929.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:28:16,347 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 20:28:36,543 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.563e+02 2.554e+02 2.943e+02 3.634e+02 1.271e+03, threshold=5.887e+02, percent-clipped=2.0 +2023-02-06 20:28:40,563 INFO [train.py:901] (3/4) Epoch 17, batch 6650, loss[loss=0.193, simple_loss=0.2891, pruned_loss=0.04846, over 5553.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2952, pruned_loss=0.06793, over 1618121.58 frames. ], batch size: 12, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:28:49,903 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=135992.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:28:56,445 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136000.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:29:04,534 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136012.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:29:12,172 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.23 vs. limit=5.0 +2023-02-06 20:29:16,447 INFO [train.py:901] (3/4) Epoch 17, batch 6700, loss[loss=0.2024, simple_loss=0.2829, pruned_loss=0.06097, over 7800.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2947, pruned_loss=0.06839, over 1613887.61 frames. ], batch size: 20, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:29:45,909 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2051, 1.3303, 1.6784, 1.2971, 0.6859, 1.4415, 1.2582, 1.0939], + device='cuda:3'), covar=tensor([0.0519, 0.1224, 0.1531, 0.1359, 0.0568, 0.1437, 0.0634, 0.0657], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0153, 0.0191, 0.0158, 0.0100, 0.0162, 0.0114, 0.0138], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 20:29:47,790 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.623e+02 2.437e+02 3.090e+02 3.837e+02 8.578e+02, threshold=6.181e+02, percent-clipped=4.0 +2023-02-06 20:29:50,647 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5957, 2.0000, 3.3157, 1.4313, 2.3147, 2.0682, 1.6708, 2.4304], + device='cuda:3'), covar=tensor([0.1861, 0.2525, 0.0780, 0.4344, 0.1914, 0.3003, 0.2197, 0.2200], + device='cuda:3'), in_proj_covar=tensor([0.0507, 0.0568, 0.0542, 0.0612, 0.0632, 0.0572, 0.0507, 0.0618], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 20:29:51,766 INFO [train.py:901] (3/4) Epoch 17, batch 6750, loss[loss=0.1974, simple_loss=0.2758, pruned_loss=0.05947, over 7802.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2944, pruned_loss=0.06822, over 1608039.88 frames. ], batch size: 20, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:30:09,234 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-06 20:30:11,742 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136107.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:30:26,348 INFO [train.py:901] (3/4) Epoch 17, batch 6800, loss[loss=0.2335, simple_loss=0.3089, pruned_loss=0.07906, over 8511.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.295, pruned_loss=0.06759, over 1612558.25 frames. ], batch size: 26, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:30:35,093 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 20:30:48,380 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136160.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:30:57,853 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.568e+02 2.691e+02 3.204e+02 3.737e+02 8.793e+02, threshold=6.409e+02, percent-clipped=5.0 +2023-02-06 20:31:01,825 INFO [train.py:901] (3/4) Epoch 17, batch 6850, loss[loss=0.1745, simple_loss=0.2467, pruned_loss=0.05121, over 7436.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2956, pruned_loss=0.06803, over 1607542.84 frames. ], batch size: 17, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:31:22,727 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 20:31:37,000 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-06 20:31:37,305 INFO [train.py:901] (3/4) Epoch 17, batch 6900, loss[loss=0.1769, simple_loss=0.2732, pruned_loss=0.04024, over 8359.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2952, pruned_loss=0.06751, over 1610023.95 frames. ], batch size: 24, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:32:08,492 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.868e+02 2.541e+02 3.415e+02 4.318e+02 7.722e+02, threshold=6.831e+02, percent-clipped=4.0 +2023-02-06 20:32:12,496 INFO [train.py:901] (3/4) Epoch 17, batch 6950, loss[loss=0.2142, simple_loss=0.2872, pruned_loss=0.07058, over 7973.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2958, pruned_loss=0.06864, over 1606719.80 frames. ], batch size: 21, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:32:32,956 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 20:32:48,142 INFO [train.py:901] (3/4) Epoch 17, batch 7000, loss[loss=0.2341, simple_loss=0.3202, pruned_loss=0.07396, over 8198.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2952, pruned_loss=0.06793, over 1609991.69 frames. ], batch size: 23, lr: 4.43e-03, grad_scale: 8.0 +2023-02-06 20:32:54,378 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7176, 1.5892, 2.9021, 1.3893, 2.1545, 3.1031, 3.2111, 2.6794], + device='cuda:3'), covar=tensor([0.1104, 0.1428, 0.0361, 0.2038, 0.0885, 0.0291, 0.0551, 0.0612], + device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0311, 0.0277, 0.0306, 0.0296, 0.0254, 0.0392, 0.0300], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 20:32:58,260 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=136344.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:33:04,331 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136353.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:33:06,280 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=136356.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:33:11,076 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136363.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:33:18,223 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.510e+02 2.499e+02 2.956e+02 3.583e+02 7.307e+02, threshold=5.911e+02, percent-clipped=2.0 +2023-02-06 20:33:22,383 INFO [train.py:901] (3/4) Epoch 17, batch 7050, loss[loss=0.2354, simple_loss=0.2954, pruned_loss=0.08769, over 6658.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2952, pruned_loss=0.06811, over 1606905.48 frames. ], batch size: 71, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:33:29,399 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136388.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:33:41,802 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0751, 1.6419, 1.3577, 1.5827, 1.3255, 1.2200, 1.2486, 1.2977], + device='cuda:3'), covar=tensor([0.1001, 0.0436, 0.1168, 0.0518, 0.0695, 0.1412, 0.0871, 0.0728], + device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0229, 0.0322, 0.0297, 0.0293, 0.0327, 0.0338, 0.0310], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 20:33:57,954 INFO [train.py:901] (3/4) Epoch 17, batch 7100, loss[loss=0.2423, simple_loss=0.33, pruned_loss=0.07729, over 8331.00 frames. ], tot_loss[loss=0.2162, simple_loss=0.2961, pruned_loss=0.06814, over 1612254.20 frames. ], batch size: 26, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:34:18,668 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136459.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:34:26,517 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136471.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:34:27,626 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.706e+02 2.503e+02 2.917e+02 3.905e+02 1.004e+03, threshold=5.834e+02, percent-clipped=4.0 +2023-02-06 20:34:31,744 INFO [train.py:901] (3/4) Epoch 17, batch 7150, loss[loss=0.2379, simple_loss=0.313, pruned_loss=0.08133, over 8601.00 frames. ], tot_loss[loss=0.218, simple_loss=0.2976, pruned_loss=0.06917, over 1616874.37 frames. ], batch size: 34, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:34:50,071 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=136504.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:35:07,636 INFO [train.py:901] (3/4) Epoch 17, batch 7200, loss[loss=0.1931, simple_loss=0.2887, pruned_loss=0.04877, over 8188.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2965, pruned_loss=0.06891, over 1607314.32 frames. ], batch size: 23, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:35:09,680 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.7393, 2.1044, 6.0518, 2.4896, 5.1437, 5.0479, 5.6279, 5.5959], + device='cuda:3'), covar=tensor([0.0870, 0.5658, 0.0649, 0.3975, 0.1710, 0.1254, 0.0701, 0.0621], + device='cuda:3'), in_proj_covar=tensor([0.0570, 0.0615, 0.0648, 0.0586, 0.0667, 0.0565, 0.0567, 0.0629], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 20:35:37,959 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.550e+02 2.498e+02 3.072e+02 3.698e+02 8.742e+02, threshold=6.145e+02, percent-clipped=2.0 +2023-02-06 20:35:42,145 INFO [train.py:901] (3/4) Epoch 17, batch 7250, loss[loss=0.2475, simple_loss=0.3209, pruned_loss=0.08709, over 8515.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2961, pruned_loss=0.06875, over 1610048.71 frames. ], batch size: 49, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:36:11,363 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136619.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:36:17,806 INFO [train.py:901] (3/4) Epoch 17, batch 7300, loss[loss=0.2026, simple_loss=0.2865, pruned_loss=0.05933, over 8198.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2949, pruned_loss=0.06809, over 1608265.43 frames. ], batch size: 23, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:36:40,629 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136661.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:36:42,658 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1499, 1.4233, 3.5065, 1.4176, 2.3037, 3.8393, 3.9423, 3.3164], + device='cuda:3'), covar=tensor([0.1070, 0.1981, 0.0354, 0.2394, 0.1149, 0.0234, 0.0393, 0.0567], + device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0314, 0.0278, 0.0307, 0.0296, 0.0255, 0.0392, 0.0299], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 20:36:48,514 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.729e+02 2.506e+02 2.969e+02 3.762e+02 7.100e+02, threshold=5.939e+02, percent-clipped=2.0 +2023-02-06 20:36:52,585 INFO [train.py:901] (3/4) Epoch 17, batch 7350, loss[loss=0.1927, simple_loss=0.2811, pruned_loss=0.05214, over 8473.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2943, pruned_loss=0.0671, over 1609132.30 frames. ], batch size: 27, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:37:05,621 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=136697.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:37:15,657 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-06 20:37:16,515 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 20:37:17,972 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136715.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:37:27,014 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136727.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:37:28,132 INFO [train.py:901] (3/4) Epoch 17, batch 7400, loss[loss=0.1944, simple_loss=0.2745, pruned_loss=0.05716, over 8037.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2944, pruned_loss=0.06758, over 1610594.77 frames. ], batch size: 22, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:37:35,014 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 20:37:36,608 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136740.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:37:45,396 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136752.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:37:59,292 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.787e+02 2.354e+02 2.898e+02 3.777e+02 7.037e+02, threshold=5.795e+02, percent-clipped=3.0 +2023-02-06 20:38:03,299 INFO [train.py:901] (3/4) Epoch 17, batch 7450, loss[loss=0.2173, simple_loss=0.3047, pruned_loss=0.06493, over 8105.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.2949, pruned_loss=0.0679, over 1610602.62 frames. ], batch size: 23, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:38:16,573 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 20:38:26,057 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136812.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:38:34,159 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136824.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:38:37,330 INFO [train.py:901] (3/4) Epoch 17, batch 7500, loss[loss=0.2639, simple_loss=0.3396, pruned_loss=0.09406, over 8023.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2954, pruned_loss=0.06776, over 1610368.66 frames. ], batch size: 22, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:39:09,459 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.732e+02 2.462e+02 2.866e+02 3.948e+02 7.787e+02, threshold=5.732e+02, percent-clipped=6.0 +2023-02-06 20:39:11,061 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136875.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:39:13,444 INFO [train.py:901] (3/4) Epoch 17, batch 7550, loss[loss=0.2227, simple_loss=0.3093, pruned_loss=0.06803, over 8345.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2962, pruned_loss=0.06834, over 1610565.64 frames. ], batch size: 24, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:39:28,622 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136900.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:39:49,027 INFO [train.py:901] (3/4) Epoch 17, batch 7600, loss[loss=0.1916, simple_loss=0.2595, pruned_loss=0.06185, over 7802.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2968, pruned_loss=0.06834, over 1617522.34 frames. ], batch size: 19, lr: 4.42e-03, grad_scale: 8.0 +2023-02-06 20:40:15,221 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6681, 2.0071, 3.3530, 1.4586, 2.4595, 2.0644, 1.6676, 2.3999], + device='cuda:3'), covar=tensor([0.1757, 0.2364, 0.0700, 0.4227, 0.1657, 0.2886, 0.2123, 0.2172], + device='cuda:3'), in_proj_covar=tensor([0.0509, 0.0571, 0.0545, 0.0614, 0.0635, 0.0576, 0.0509, 0.0619], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 20:40:21,096 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.468e+02 2.516e+02 2.945e+02 3.717e+02 7.457e+02, threshold=5.891e+02, percent-clipped=6.0 +2023-02-06 20:40:22,736 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2608, 1.9705, 2.6926, 2.1662, 2.5828, 2.2735, 1.9641, 1.3063], + device='cuda:3'), covar=tensor([0.4773, 0.4601, 0.1549, 0.3122, 0.2128, 0.2528, 0.1692, 0.4509], + device='cuda:3'), in_proj_covar=tensor([0.0920, 0.0937, 0.0775, 0.0902, 0.0969, 0.0853, 0.0723, 0.0798], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 20:40:25,200 INFO [train.py:901] (3/4) Epoch 17, batch 7650, loss[loss=0.2162, simple_loss=0.2972, pruned_loss=0.0676, over 7659.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2943, pruned_loss=0.06685, over 1616656.88 frames. ], batch size: 19, lr: 4.41e-03, grad_scale: 16.0 +2023-02-06 20:40:38,676 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6656, 2.0539, 3.1985, 1.4227, 2.4051, 2.1356, 1.6966, 2.3841], + device='cuda:3'), covar=tensor([0.1738, 0.2316, 0.0758, 0.4280, 0.1650, 0.2893, 0.2108, 0.2034], + device='cuda:3'), in_proj_covar=tensor([0.0510, 0.0572, 0.0546, 0.0615, 0.0636, 0.0577, 0.0509, 0.0620], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 20:40:43,281 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=137005.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:40:58,549 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 20:41:00,049 INFO [train.py:901] (3/4) Epoch 17, batch 7700, loss[loss=0.2195, simple_loss=0.2825, pruned_loss=0.07826, over 8090.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.2943, pruned_loss=0.06746, over 1612966.80 frames. ], batch size: 21, lr: 4.41e-03, grad_scale: 16.0 +2023-02-06 20:41:26,497 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 20:41:26,688 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137068.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:41:30,557 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.690e+02 2.598e+02 3.111e+02 3.900e+02 8.834e+02, threshold=6.222e+02, percent-clipped=1.0 +2023-02-06 20:41:34,727 INFO [train.py:901] (3/4) Epoch 17, batch 7750, loss[loss=0.2297, simple_loss=0.3055, pruned_loss=0.07695, over 8189.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.2933, pruned_loss=0.06701, over 1615655.54 frames. ], batch size: 23, lr: 4.41e-03, grad_scale: 16.0 +2023-02-06 20:41:36,314 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6534, 1.5843, 2.1462, 1.5642, 1.1661, 2.1762, 0.5345, 1.3088], + device='cuda:3'), covar=tensor([0.1816, 0.1355, 0.0381, 0.1268, 0.3218, 0.0461, 0.2301, 0.1567], + device='cuda:3'), in_proj_covar=tensor([0.0176, 0.0183, 0.0116, 0.0216, 0.0263, 0.0123, 0.0166, 0.0182], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 20:41:44,965 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137093.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:42:03,585 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=137120.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:42:09,491 INFO [train.py:901] (3/4) Epoch 17, batch 7800, loss[loss=0.1976, simple_loss=0.2702, pruned_loss=0.06253, over 7422.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2937, pruned_loss=0.06674, over 1618354.19 frames. ], batch size: 17, lr: 4.41e-03, grad_scale: 16.0 +2023-02-06 20:42:36,424 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=137168.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:42:39,579 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.681e+02 2.454e+02 2.768e+02 3.488e+02 7.043e+02, threshold=5.537e+02, percent-clipped=4.0 +2023-02-06 20:42:41,436 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 +2023-02-06 20:42:43,619 INFO [train.py:901] (3/4) Epoch 17, batch 7850, loss[loss=0.2305, simple_loss=0.313, pruned_loss=0.07403, over 8597.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.293, pruned_loss=0.06633, over 1616998.24 frames. ], batch size: 50, lr: 4.41e-03, grad_scale: 16.0 +2023-02-06 20:43:16,604 INFO [train.py:901] (3/4) Epoch 17, batch 7900, loss[loss=0.178, simple_loss=0.2684, pruned_loss=0.04379, over 7978.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.2938, pruned_loss=0.06688, over 1616699.90 frames. ], batch size: 21, lr: 4.41e-03, grad_scale: 16.0 +2023-02-06 20:43:45,781 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.620e+02 2.478e+02 3.005e+02 3.961e+02 6.905e+02, threshold=6.010e+02, percent-clipped=7.0 +2023-02-06 20:43:49,853 INFO [train.py:901] (3/4) Epoch 17, batch 7950, loss[loss=0.192, simple_loss=0.2676, pruned_loss=0.05822, over 6757.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2931, pruned_loss=0.0666, over 1611583.83 frames. ], batch size: 15, lr: 4.41e-03, grad_scale: 16.0 +2023-02-06 20:43:52,843 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=137283.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:44:23,211 INFO [train.py:901] (3/4) Epoch 17, batch 8000, loss[loss=0.1963, simple_loss=0.2717, pruned_loss=0.06044, over 7707.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2942, pruned_loss=0.06713, over 1614087.75 frames. ], batch size: 18, lr: 4.41e-03, grad_scale: 16.0 +2023-02-06 20:44:52,977 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.715e+02 2.583e+02 3.026e+02 3.684e+02 1.341e+03, threshold=6.053e+02, percent-clipped=4.0 +2023-02-06 20:44:55,368 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137376.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:44:57,246 INFO [train.py:901] (3/4) Epoch 17, batch 8050, loss[loss=0.1723, simple_loss=0.2534, pruned_loss=0.04558, over 7558.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2937, pruned_loss=0.06729, over 1603334.35 frames. ], batch size: 18, lr: 4.41e-03, grad_scale: 16.0 +2023-02-06 20:45:12,517 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137401.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:45:29,328 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 20:45:34,948 INFO [train.py:901] (3/4) Epoch 18, batch 0, loss[loss=0.2158, simple_loss=0.3054, pruned_loss=0.0631, over 8492.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.3054, pruned_loss=0.0631, over 8492.00 frames. ], batch size: 26, lr: 4.28e-03, grad_scale: 16.0 +2023-02-06 20:45:34,948 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 20:45:46,125 INFO [train.py:935] (3/4) Epoch 18, validation: loss=0.1783, simple_loss=0.2784, pruned_loss=0.03907, over 944034.00 frames. +2023-02-06 20:45:46,126 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 20:46:00,859 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 20:46:20,802 INFO [train.py:901] (3/4) Epoch 18, batch 50, loss[loss=0.1835, simple_loss=0.2735, pruned_loss=0.04675, over 8137.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2956, pruned_loss=0.06793, over 361913.60 frames. ], batch size: 22, lr: 4.28e-03, grad_scale: 16.0 +2023-02-06 20:46:29,004 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.854e+02 2.698e+02 3.585e+02 4.414e+02 8.769e+02, threshold=7.169e+02, percent-clipped=9.0 +2023-02-06 20:46:35,902 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 20:46:56,059 INFO [train.py:901] (3/4) Epoch 18, batch 100, loss[loss=0.2028, simple_loss=0.2849, pruned_loss=0.06037, over 7807.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2992, pruned_loss=0.06878, over 641096.18 frames. ], batch size: 19, lr: 4.28e-03, grad_scale: 16.0 +2023-02-06 20:46:58,818 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 20:47:16,436 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137539.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:47:31,280 INFO [train.py:901] (3/4) Epoch 18, batch 150, loss[loss=0.2022, simple_loss=0.2824, pruned_loss=0.06097, over 8506.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.297, pruned_loss=0.06728, over 858558.07 frames. ], batch size: 26, lr: 4.28e-03, grad_scale: 16.0 +2023-02-06 20:47:33,505 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137564.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:47:39,699 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.673e+02 2.369e+02 2.797e+02 3.885e+02 6.122e+02, threshold=5.595e+02, percent-clipped=0.0 +2023-02-06 20:47:49,999 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 +2023-02-06 20:47:53,234 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9776, 1.7295, 2.1315, 1.9315, 2.0739, 1.9802, 1.7779, 0.7718], + device='cuda:3'), covar=tensor([0.5041, 0.4237, 0.1728, 0.2831, 0.1984, 0.2668, 0.1825, 0.4393], + device='cuda:3'), in_proj_covar=tensor([0.0922, 0.0939, 0.0773, 0.0909, 0.0973, 0.0855, 0.0725, 0.0802], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 20:48:07,693 INFO [train.py:901] (3/4) Epoch 18, batch 200, loss[loss=0.225, simple_loss=0.3113, pruned_loss=0.06936, over 8640.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2968, pruned_loss=0.06738, over 1028115.37 frames. ], batch size: 34, lr: 4.28e-03, grad_scale: 16.0 +2023-02-06 20:48:44,082 INFO [train.py:901] (3/4) Epoch 18, batch 250, loss[loss=0.243, simple_loss=0.3131, pruned_loss=0.08649, over 8365.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2982, pruned_loss=0.06856, over 1161983.04 frames. ], batch size: 24, lr: 4.28e-03, grad_scale: 16.0 +2023-02-06 20:48:52,405 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.709e+02 2.467e+02 3.008e+02 3.586e+02 6.135e+02, threshold=6.015e+02, percent-clipped=1.0 +2023-02-06 20:48:55,946 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 20:49:03,701 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 20:49:19,875 INFO [train.py:901] (3/4) Epoch 18, batch 300, loss[loss=0.2411, simple_loss=0.3235, pruned_loss=0.07937, over 8506.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2972, pruned_loss=0.06772, over 1265609.94 frames. ], batch size: 26, lr: 4.28e-03, grad_scale: 8.0 +2023-02-06 20:49:44,086 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 +2023-02-06 20:49:55,786 INFO [train.py:901] (3/4) Epoch 18, batch 350, loss[loss=0.1966, simple_loss=0.282, pruned_loss=0.05559, over 8287.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2978, pruned_loss=0.06775, over 1349319.51 frames. ], batch size: 23, lr: 4.28e-03, grad_scale: 8.0 +2023-02-06 20:50:05,695 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.654e+02 2.555e+02 3.034e+02 3.752e+02 7.695e+02, threshold=6.069e+02, percent-clipped=3.0 +2023-02-06 20:50:32,304 INFO [train.py:901] (3/4) Epoch 18, batch 400, loss[loss=0.2552, simple_loss=0.3346, pruned_loss=0.08789, over 8178.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2976, pruned_loss=0.06764, over 1413554.84 frames. ], batch size: 23, lr: 4.28e-03, grad_scale: 8.0 +2023-02-06 20:50:40,683 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.2947, 4.3901, 3.9359, 2.1268, 3.8163, 3.8926, 3.9472, 3.7090], + device='cuda:3'), covar=tensor([0.0917, 0.0619, 0.1274, 0.5113, 0.1026, 0.1284, 0.1504, 0.0913], + device='cuda:3'), in_proj_covar=tensor([0.0507, 0.0418, 0.0422, 0.0523, 0.0413, 0.0421, 0.0410, 0.0366], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 20:50:46,323 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6265, 1.6176, 2.0528, 1.4473, 1.1033, 2.0925, 0.3403, 1.2086], + device='cuda:3'), covar=tensor([0.1877, 0.1384, 0.0411, 0.1330, 0.3310, 0.0486, 0.2349, 0.1527], + device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0184, 0.0118, 0.0219, 0.0263, 0.0124, 0.0166, 0.0182], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 20:51:08,172 INFO [train.py:901] (3/4) Epoch 18, batch 450, loss[loss=0.2269, simple_loss=0.3078, pruned_loss=0.07295, over 8132.00 frames. ], tot_loss[loss=0.2166, simple_loss=0.2977, pruned_loss=0.06777, over 1461251.89 frames. ], batch size: 22, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:51:16,924 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.786e+02 2.504e+02 3.016e+02 3.557e+02 6.367e+02, threshold=6.032e+02, percent-clipped=3.0 +2023-02-06 20:51:43,038 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=137910.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:51:43,595 INFO [train.py:901] (3/4) Epoch 18, batch 500, loss[loss=0.2517, simple_loss=0.3185, pruned_loss=0.0925, over 7113.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2977, pruned_loss=0.06844, over 1495639.09 frames. ], batch size: 72, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:51:45,197 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=137913.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:52:05,361 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5739, 2.7911, 1.8344, 2.4021, 2.1802, 1.6903, 2.1387, 2.3009], + device='cuda:3'), covar=tensor([0.1454, 0.0326, 0.1133, 0.0626, 0.0727, 0.1381, 0.1025, 0.0953], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0234, 0.0326, 0.0300, 0.0297, 0.0330, 0.0342, 0.0316], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 20:52:20,558 INFO [train.py:901] (3/4) Epoch 18, batch 550, loss[loss=0.2012, simple_loss=0.2688, pruned_loss=0.06682, over 7559.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.297, pruned_loss=0.06799, over 1522339.44 frames. ], batch size: 18, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:52:29,490 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.662e+02 2.606e+02 3.197e+02 3.974e+02 7.545e+02, threshold=6.394e+02, percent-clipped=3.0 +2023-02-06 20:52:56,990 INFO [train.py:901] (3/4) Epoch 18, batch 600, loss[loss=0.2396, simple_loss=0.313, pruned_loss=0.08315, over 8476.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2964, pruned_loss=0.06741, over 1549478.57 frames. ], batch size: 25, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:53:09,287 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9406, 2.1874, 1.9236, 2.5721, 1.3324, 1.5756, 1.9051, 2.1786], + device='cuda:3'), covar=tensor([0.0685, 0.0683, 0.0881, 0.0381, 0.1012, 0.1252, 0.0817, 0.0660], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0199, 0.0251, 0.0212, 0.0208, 0.0249, 0.0254, 0.0210], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 20:53:11,892 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 20:53:32,788 INFO [train.py:901] (3/4) Epoch 18, batch 650, loss[loss=0.2502, simple_loss=0.3306, pruned_loss=0.08483, over 8500.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2959, pruned_loss=0.06717, over 1565460.34 frames. ], batch size: 26, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:53:43,406 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.549e+02 2.461e+02 2.865e+02 3.365e+02 7.739e+02, threshold=5.729e+02, percent-clipped=1.0 +2023-02-06 20:54:00,540 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 20:54:09,454 INFO [train.py:901] (3/4) Epoch 18, batch 700, loss[loss=0.2081, simple_loss=0.2895, pruned_loss=0.06332, over 8345.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2955, pruned_loss=0.06714, over 1581449.69 frames. ], batch size: 26, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:54:29,539 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 20:54:30,790 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6389, 1.5037, 1.9995, 1.4039, 1.1818, 2.1049, 0.2763, 1.2455], + device='cuda:3'), covar=tensor([0.1958, 0.1695, 0.0487, 0.1498, 0.3385, 0.0459, 0.2684, 0.1778], + device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0184, 0.0117, 0.0217, 0.0262, 0.0123, 0.0165, 0.0182], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 20:54:44,128 INFO [train.py:901] (3/4) Epoch 18, batch 750, loss[loss=0.228, simple_loss=0.2956, pruned_loss=0.08018, over 6860.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.2955, pruned_loss=0.06744, over 1589642.44 frames. ], batch size: 72, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:54:53,227 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.516e+02 2.332e+02 3.041e+02 3.730e+02 6.216e+02, threshold=6.081e+02, percent-clipped=3.0 +2023-02-06 20:54:58,184 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 20:55:08,084 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 20:55:19,814 INFO [train.py:901] (3/4) Epoch 18, batch 800, loss[loss=0.2011, simple_loss=0.2997, pruned_loss=0.05127, over 8192.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.295, pruned_loss=0.06663, over 1594952.88 frames. ], batch size: 23, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:55:49,563 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=138254.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:55:51,620 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=138257.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:55:54,224 INFO [train.py:901] (3/4) Epoch 18, batch 850, loss[loss=0.207, simple_loss=0.2753, pruned_loss=0.06941, over 7420.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2957, pruned_loss=0.06642, over 1602594.87 frames. ], batch size: 17, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:56:03,045 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.796e+02 2.308e+02 2.906e+02 3.562e+02 8.427e+02, threshold=5.812e+02, percent-clipped=4.0 +2023-02-06 20:56:13,563 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138288.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:56:30,836 INFO [train.py:901] (3/4) Epoch 18, batch 900, loss[loss=0.2328, simple_loss=0.3045, pruned_loss=0.08056, over 8027.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.2947, pruned_loss=0.06643, over 1600993.87 frames. ], batch size: 22, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:56:35,147 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6134, 1.3773, 1.6638, 1.3281, 0.8626, 1.3857, 1.3964, 1.2684], + device='cuda:3'), covar=tensor([0.0545, 0.1247, 0.1653, 0.1443, 0.0586, 0.1510, 0.0743, 0.0672], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0152, 0.0190, 0.0158, 0.0100, 0.0161, 0.0114, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 20:56:50,631 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138340.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:57:05,377 INFO [train.py:901] (3/4) Epoch 18, batch 950, loss[loss=0.1848, simple_loss=0.2633, pruned_loss=0.05318, over 7440.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.2946, pruned_loss=0.06656, over 1602980.12 frames. ], batch size: 17, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:57:10,976 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138369.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:57:13,023 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138372.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 20:57:14,179 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.844e+02 2.531e+02 3.020e+02 3.937e+02 8.991e+02, threshold=6.039e+02, percent-clipped=7.0 +2023-02-06 20:57:14,413 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3434, 2.4702, 2.3584, 3.9205, 1.6390, 2.1254, 2.4571, 3.1930], + device='cuda:3'), covar=tensor([0.0766, 0.0981, 0.0864, 0.0247, 0.1157, 0.1225, 0.1071, 0.0608], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0200, 0.0251, 0.0212, 0.0207, 0.0249, 0.0253, 0.0208], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 20:57:29,249 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 20:57:40,379 INFO [train.py:901] (3/4) Epoch 18, batch 1000, loss[loss=0.2038, simple_loss=0.276, pruned_loss=0.06582, over 7212.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2942, pruned_loss=0.06623, over 1609995.43 frames. ], batch size: 16, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:57:56,233 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6051, 2.0240, 3.3401, 1.3371, 2.4253, 1.9443, 1.6931, 2.4827], + device='cuda:3'), covar=tensor([0.1895, 0.2566, 0.0938, 0.4490, 0.1854, 0.3162, 0.2126, 0.2265], + device='cuda:3'), in_proj_covar=tensor([0.0511, 0.0574, 0.0544, 0.0618, 0.0636, 0.0577, 0.0510, 0.0624], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 20:58:05,480 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 20:58:16,697 INFO [train.py:901] (3/4) Epoch 18, batch 1050, loss[loss=0.1897, simple_loss=0.2703, pruned_loss=0.05453, over 8081.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.2936, pruned_loss=0.06632, over 1603052.18 frames. ], batch size: 21, lr: 4.27e-03, grad_scale: 8.0 +2023-02-06 20:58:18,832 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 20:58:25,546 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.682e+02 2.454e+02 3.228e+02 4.133e+02 8.765e+02, threshold=6.456e+02, percent-clipped=4.0 +2023-02-06 20:58:51,055 INFO [train.py:901] (3/4) Epoch 18, batch 1100, loss[loss=0.2511, simple_loss=0.3337, pruned_loss=0.08426, over 8587.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2942, pruned_loss=0.0669, over 1604564.23 frames. ], batch size: 31, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 20:58:56,219 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.7870, 5.8938, 5.1480, 2.4869, 5.1133, 5.5669, 5.4438, 5.3661], + device='cuda:3'), covar=tensor([0.0483, 0.0392, 0.0856, 0.4209, 0.0677, 0.0777, 0.1080, 0.0557], + device='cuda:3'), in_proj_covar=tensor([0.0507, 0.0419, 0.0419, 0.0517, 0.0409, 0.0422, 0.0407, 0.0365], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 20:59:26,908 INFO [train.py:901] (3/4) Epoch 18, batch 1150, loss[loss=0.1956, simple_loss=0.2835, pruned_loss=0.05382, over 8491.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2946, pruned_loss=0.06721, over 1604550.44 frames. ], batch size: 26, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 20:59:29,623 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 20:59:35,879 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.371e+02 2.366e+02 2.909e+02 3.553e+02 5.350e+02, threshold=5.817e+02, percent-clipped=0.0 +2023-02-06 20:59:44,962 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5679, 1.9722, 2.9823, 1.3080, 2.1473, 1.8179, 1.7064, 2.0877], + device='cuda:3'), covar=tensor([0.2134, 0.2598, 0.0907, 0.4988, 0.2059, 0.3711, 0.2488, 0.2572], + device='cuda:3'), in_proj_covar=tensor([0.0516, 0.0577, 0.0550, 0.0625, 0.0642, 0.0583, 0.0515, 0.0630], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 21:00:02,035 INFO [train.py:901] (3/4) Epoch 18, batch 1200, loss[loss=0.2239, simple_loss=0.3033, pruned_loss=0.07219, over 8492.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2947, pruned_loss=0.06717, over 1609689.24 frames. ], batch size: 28, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:00:04,276 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.9849, 1.7520, 6.1094, 2.3064, 5.6025, 5.1944, 5.6526, 5.6008], + device='cuda:3'), covar=tensor([0.0407, 0.4113, 0.0274, 0.3323, 0.0775, 0.0742, 0.0454, 0.0426], + device='cuda:3'), in_proj_covar=tensor([0.0582, 0.0618, 0.0662, 0.0591, 0.0674, 0.0577, 0.0569, 0.0642], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 21:00:11,865 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138625.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:00:13,950 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138628.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:00:16,606 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=138632.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:00:29,705 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138650.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:00:31,812 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138653.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:00:37,162 INFO [train.py:901] (3/4) Epoch 18, batch 1250, loss[loss=0.1918, simple_loss=0.2796, pruned_loss=0.05205, over 8031.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.295, pruned_loss=0.06712, over 1613538.39 frames. ], batch size: 22, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:00:46,118 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5122, 4.5686, 4.1135, 1.9228, 4.0357, 4.1232, 4.1930, 3.9734], + device='cuda:3'), covar=tensor([0.0726, 0.0511, 0.0944, 0.4779, 0.0789, 0.0777, 0.1137, 0.0689], + device='cuda:3'), in_proj_covar=tensor([0.0513, 0.0420, 0.0421, 0.0521, 0.0412, 0.0424, 0.0409, 0.0367], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 21:00:47,269 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.636e+02 2.310e+02 2.834e+02 3.613e+02 5.274e+02, threshold=5.668e+02, percent-clipped=0.0 +2023-02-06 21:00:54,247 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=138684.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:01:04,840 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138699.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:01:08,295 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138704.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:01:12,860 INFO [train.py:901] (3/4) Epoch 18, batch 1300, loss[loss=0.2302, simple_loss=0.3125, pruned_loss=0.07395, over 8243.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.295, pruned_loss=0.06698, over 1615570.88 frames. ], batch size: 22, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:01:15,853 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138715.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:01:34,957 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4120, 1.6681, 2.7226, 1.2165, 1.9353, 1.7962, 1.3695, 1.8572], + device='cuda:3'), covar=tensor([0.1887, 0.2362, 0.0731, 0.4470, 0.1796, 0.3151, 0.2324, 0.2135], + device='cuda:3'), in_proj_covar=tensor([0.0514, 0.0577, 0.0549, 0.0624, 0.0643, 0.0583, 0.0513, 0.0630], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 21:01:37,544 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138747.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:01:38,938 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9351, 2.0789, 1.8108, 2.6196, 1.1733, 1.5314, 1.7936, 2.1198], + device='cuda:3'), covar=tensor([0.0757, 0.0741, 0.0949, 0.0361, 0.1098, 0.1350, 0.0882, 0.0738], + device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0202, 0.0254, 0.0214, 0.0210, 0.0251, 0.0256, 0.0210], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 21:01:47,850 INFO [train.py:901] (3/4) Epoch 18, batch 1350, loss[loss=0.1826, simple_loss=0.2669, pruned_loss=0.04914, over 8140.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2948, pruned_loss=0.06687, over 1618504.73 frames. ], batch size: 22, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:01:56,594 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.596e+02 2.404e+02 2.906e+02 3.545e+02 6.613e+02, threshold=5.812e+02, percent-clipped=4.0 +2023-02-06 21:02:15,443 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138799.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:02:23,336 INFO [train.py:901] (3/4) Epoch 18, batch 1400, loss[loss=0.1738, simple_loss=0.2511, pruned_loss=0.04826, over 7444.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2963, pruned_loss=0.06758, over 1622924.09 frames. ], batch size: 17, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:02:43,933 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.26 vs. limit=5.0 +2023-02-06 21:02:57,621 INFO [train.py:901] (3/4) Epoch 18, batch 1450, loss[loss=0.1927, simple_loss=0.2612, pruned_loss=0.06208, over 7685.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2951, pruned_loss=0.06684, over 1622276.55 frames. ], batch size: 18, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:03:06,392 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.331e+02 2.491e+02 3.050e+02 4.246e+02 7.467e+02, threshold=6.100e+02, percent-clipped=3.0 +2023-02-06 21:03:07,079 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 21:03:33,392 INFO [train.py:901] (3/4) Epoch 18, batch 1500, loss[loss=0.2233, simple_loss=0.2885, pruned_loss=0.07909, over 7248.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2943, pruned_loss=0.06645, over 1619186.19 frames. ], batch size: 16, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:03:44,791 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138927.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:04:08,397 INFO [train.py:901] (3/4) Epoch 18, batch 1550, loss[loss=0.2165, simple_loss=0.2993, pruned_loss=0.06685, over 6694.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2941, pruned_loss=0.06654, over 1616480.81 frames. ], batch size: 71, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:04:17,342 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.725e+02 2.366e+02 2.933e+02 3.736e+02 6.367e+02, threshold=5.865e+02, percent-clipped=3.0 +2023-02-06 21:04:38,197 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139003.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:04:43,436 INFO [train.py:901] (3/4) Epoch 18, batch 1600, loss[loss=0.2136, simple_loss=0.2906, pruned_loss=0.06835, over 8471.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2935, pruned_loss=0.06664, over 1615047.16 frames. ], batch size: 27, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:04:54,110 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4013, 1.6978, 4.6063, 1.8080, 4.1115, 3.8583, 4.1632, 4.0426], + device='cuda:3'), covar=tensor([0.0519, 0.3806, 0.0391, 0.3588, 0.0928, 0.0825, 0.0493, 0.0576], + device='cuda:3'), in_proj_covar=tensor([0.0587, 0.0623, 0.0665, 0.0593, 0.0679, 0.0582, 0.0573, 0.0646], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 21:04:56,950 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139028.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:05:07,017 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139043.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:05:10,368 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139048.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:05:15,387 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139055.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:05:18,032 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139059.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:05:19,996 INFO [train.py:901] (3/4) Epoch 18, batch 1650, loss[loss=0.2197, simple_loss=0.304, pruned_loss=0.06766, over 8106.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2941, pruned_loss=0.06709, over 1613933.07 frames. ], batch size: 23, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:05:28,835 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.409e+02 2.375e+02 2.907e+02 3.508e+02 7.626e+02, threshold=5.813e+02, percent-clipped=3.0 +2023-02-06 21:05:33,246 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139080.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:05:37,215 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0476, 1.6291, 1.4347, 1.5850, 1.3303, 1.2500, 1.1932, 1.2435], + device='cuda:3'), covar=tensor([0.1094, 0.0457, 0.1234, 0.0532, 0.0745, 0.1427, 0.0982, 0.0878], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0236, 0.0328, 0.0302, 0.0298, 0.0331, 0.0343, 0.0318], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 21:05:54,418 INFO [train.py:901] (3/4) Epoch 18, batch 1700, loss[loss=0.2467, simple_loss=0.3104, pruned_loss=0.09144, over 8353.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2939, pruned_loss=0.06693, over 1616798.15 frames. ], batch size: 24, lr: 4.26e-03, grad_scale: 8.0 +2023-02-06 21:06:28,707 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139158.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:06:30,617 INFO [train.py:901] (3/4) Epoch 18, batch 1750, loss[loss=0.212, simple_loss=0.2921, pruned_loss=0.06593, over 8258.00 frames. ], tot_loss[loss=0.214, simple_loss=0.294, pruned_loss=0.06704, over 1620804.60 frames. ], batch size: 22, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:06:32,195 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139163.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:06:39,592 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.310e+02 2.913e+02 3.912e+02 7.750e+02, threshold=5.826e+02, percent-clipped=6.0 +2023-02-06 21:06:39,808 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139174.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:07:05,725 INFO [train.py:901] (3/4) Epoch 18, batch 1800, loss[loss=0.215, simple_loss=0.2827, pruned_loss=0.07367, over 8084.00 frames. ], tot_loss[loss=0.213, simple_loss=0.2932, pruned_loss=0.06636, over 1616637.89 frames. ], batch size: 21, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:07:37,630 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139256.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:07:40,842 INFO [train.py:901] (3/4) Epoch 18, batch 1850, loss[loss=0.2437, simple_loss=0.322, pruned_loss=0.08264, over 7466.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2928, pruned_loss=0.06588, over 1614430.04 frames. ], batch size: 71, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:07:47,457 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.8560, 1.6598, 5.9678, 2.1389, 5.3316, 5.0205, 5.5330, 5.4022], + device='cuda:3'), covar=tensor([0.0481, 0.4743, 0.0357, 0.3916, 0.1137, 0.0899, 0.0490, 0.0531], + device='cuda:3'), in_proj_covar=tensor([0.0585, 0.0623, 0.0664, 0.0592, 0.0676, 0.0581, 0.0573, 0.0647], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 21:07:49,485 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139271.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:07:51,406 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.507e+02 2.274e+02 2.776e+02 3.369e+02 8.658e+02, threshold=5.552e+02, percent-clipped=2.0 +2023-02-06 21:07:53,731 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9829, 3.6081, 2.2036, 2.8818, 2.7377, 1.9633, 2.8070, 2.9874], + device='cuda:3'), covar=tensor([0.1686, 0.0446, 0.1252, 0.0849, 0.0764, 0.1486, 0.1161, 0.1274], + device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0234, 0.0324, 0.0299, 0.0297, 0.0329, 0.0340, 0.0314], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 21:07:58,399 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3327, 2.0219, 2.7147, 2.2286, 2.7254, 2.2639, 2.0074, 1.5580], + device='cuda:3'), covar=tensor([0.4753, 0.4583, 0.1662, 0.3605, 0.2235, 0.2593, 0.1820, 0.4734], + device='cuda:3'), in_proj_covar=tensor([0.0919, 0.0938, 0.0777, 0.0909, 0.0976, 0.0856, 0.0726, 0.0802], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 21:08:05,262 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139294.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:08:14,583 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139307.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:08:17,142 INFO [train.py:901] (3/4) Epoch 18, batch 1900, loss[loss=0.2071, simple_loss=0.2823, pruned_loss=0.06599, over 8072.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2935, pruned_loss=0.06643, over 1613442.84 frames. ], batch size: 21, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:08:52,413 INFO [train.py:901] (3/4) Epoch 18, batch 1950, loss[loss=0.2405, simple_loss=0.3209, pruned_loss=0.08002, over 8517.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2926, pruned_loss=0.06611, over 1613621.89 frames. ], batch size: 26, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:08:55,252 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 21:09:01,254 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.789e+02 2.421e+02 2.964e+02 3.877e+02 7.962e+02, threshold=5.927e+02, percent-clipped=5.0 +2023-02-06 21:09:08,117 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 21:09:11,192 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139386.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:09:28,244 INFO [train.py:901] (3/4) Epoch 18, batch 2000, loss[loss=0.2436, simple_loss=0.3302, pruned_loss=0.07848, over 8592.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2931, pruned_loss=0.06665, over 1610072.26 frames. ], batch size: 31, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:09:28,249 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 21:09:30,557 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139414.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:09:33,906 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139419.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:09:42,134 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139430.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:09:48,228 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139439.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:09:51,662 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139444.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:09:54,542 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.44 vs. limit=5.0 +2023-02-06 21:09:59,144 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139455.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:10:02,980 INFO [train.py:901] (3/4) Epoch 18, batch 2050, loss[loss=0.252, simple_loss=0.3192, pruned_loss=0.09236, over 8593.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2947, pruned_loss=0.06766, over 1613512.07 frames. ], batch size: 39, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:10:12,679 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.679e+02 2.515e+02 3.080e+02 3.592e+02 7.733e+02, threshold=6.160e+02, percent-clipped=3.0 +2023-02-06 21:10:39,816 INFO [train.py:901] (3/4) Epoch 18, batch 2100, loss[loss=0.2651, simple_loss=0.3345, pruned_loss=0.09781, over 8090.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.2933, pruned_loss=0.06713, over 1609337.38 frames. ], batch size: 21, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:10:46,071 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1932, 1.2317, 1.5255, 1.2245, 0.7033, 1.2822, 1.2344, 1.0630], + device='cuda:3'), covar=tensor([0.0603, 0.1316, 0.1664, 0.1476, 0.0568, 0.1579, 0.0709, 0.0698], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0190, 0.0158, 0.0099, 0.0161, 0.0114, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 21:10:53,020 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3801, 1.6075, 1.6556, 1.0063, 1.7128, 1.2809, 0.2647, 1.5535], + device='cuda:3'), covar=tensor([0.0448, 0.0290, 0.0269, 0.0445, 0.0374, 0.0840, 0.0778, 0.0252], + device='cuda:3'), in_proj_covar=tensor([0.0437, 0.0377, 0.0325, 0.0432, 0.0365, 0.0522, 0.0384, 0.0405], + device='cuda:3'), out_proj_covar=tensor([1.1910e-04, 9.9955e-05, 8.6009e-05, 1.1499e-04, 9.7570e-05, 1.4966e-04, + 1.0457e-04, 1.0862e-04], device='cuda:3') +2023-02-06 21:10:58,025 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.0152, 2.3799, 3.5288, 2.0819, 1.7838, 3.5234, 0.5530, 2.1821], + device='cuda:3'), covar=tensor([0.1393, 0.1413, 0.0274, 0.2010, 0.3163, 0.0431, 0.2752, 0.1617], + device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0188, 0.0118, 0.0216, 0.0264, 0.0126, 0.0165, 0.0184], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 21:11:01,675 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 21:11:15,316 INFO [train.py:901] (3/4) Epoch 18, batch 2150, loss[loss=0.2117, simple_loss=0.3128, pruned_loss=0.05535, over 8352.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2948, pruned_loss=0.06801, over 1612146.68 frames. ], batch size: 24, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:11:24,962 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.702e+02 2.487e+02 3.024e+02 3.808e+02 9.008e+02, threshold=6.048e+02, percent-clipped=4.0 +2023-02-06 21:11:33,584 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-06 21:11:34,710 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139589.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:11:43,091 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139600.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:11:43,161 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139600.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:11:50,319 INFO [train.py:901] (3/4) Epoch 18, batch 2200, loss[loss=0.1906, simple_loss=0.2738, pruned_loss=0.0537, over 7700.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2938, pruned_loss=0.0674, over 1609357.88 frames. ], batch size: 18, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:12:10,497 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139638.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:12:13,213 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139642.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:12:19,363 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139651.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:12:22,903 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1327, 2.5972, 3.0222, 1.4251, 3.1791, 1.8064, 1.5009, 2.0301], + device='cuda:3'), covar=tensor([0.0741, 0.0344, 0.0223, 0.0741, 0.0317, 0.0782, 0.0855, 0.0570], + device='cuda:3'), in_proj_covar=tensor([0.0428, 0.0369, 0.0320, 0.0424, 0.0357, 0.0512, 0.0377, 0.0398], + device='cuda:3'), out_proj_covar=tensor([1.1676e-04, 9.7882e-05, 8.4682e-05, 1.1285e-04, 9.5403e-05, 1.4668e-04, + 1.0260e-04, 1.0672e-04], device='cuda:3') +2023-02-06 21:12:26,632 INFO [train.py:901] (3/4) Epoch 18, batch 2250, loss[loss=0.2489, simple_loss=0.3259, pruned_loss=0.08596, over 8452.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.293, pruned_loss=0.06673, over 1612440.19 frames. ], batch size: 27, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:12:31,134 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139667.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:12:36,169 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.736e+02 2.519e+02 3.270e+02 4.475e+02 8.912e+02, threshold=6.540e+02, percent-clipped=11.0 +2023-02-06 21:12:53,999 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.2411, 1.6664, 4.4257, 1.9752, 3.9950, 3.6750, 4.0088, 3.8841], + device='cuda:3'), covar=tensor([0.0556, 0.4083, 0.0553, 0.3575, 0.0959, 0.0967, 0.0540, 0.0608], + device='cuda:3'), in_proj_covar=tensor([0.0585, 0.0621, 0.0667, 0.0595, 0.0675, 0.0580, 0.0574, 0.0647], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 21:13:01,644 INFO [train.py:901] (3/4) Epoch 18, batch 2300, loss[loss=0.2409, simple_loss=0.3127, pruned_loss=0.08459, over 6676.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.294, pruned_loss=0.06711, over 1615651.40 frames. ], batch size: 72, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:13:04,669 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139715.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:13:05,410 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5931, 2.0265, 2.1908, 1.4002, 2.2786, 1.4640, 0.6298, 1.8135], + device='cuda:3'), covar=tensor([0.0548, 0.0281, 0.0196, 0.0465, 0.0313, 0.0823, 0.0720, 0.0271], + device='cuda:3'), in_proj_covar=tensor([0.0431, 0.0371, 0.0322, 0.0427, 0.0359, 0.0517, 0.0380, 0.0401], + device='cuda:3'), out_proj_covar=tensor([1.1755e-04, 9.8298e-05, 8.5310e-05, 1.1363e-04, 9.5936e-05, 1.4809e-04, + 1.0338e-04, 1.0728e-04], device='cuda:3') +2023-02-06 21:13:32,006 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139753.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:13:37,207 INFO [train.py:901] (3/4) Epoch 18, batch 2350, loss[loss=0.209, simple_loss=0.3017, pruned_loss=0.05818, over 8327.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2948, pruned_loss=0.06746, over 1617550.88 frames. ], batch size: 25, lr: 4.25e-03, grad_scale: 8.0 +2023-02-06 21:13:40,621 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139766.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:13:47,224 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.421e+02 2.439e+02 2.945e+02 3.859e+02 6.515e+02, threshold=5.891e+02, percent-clipped=0.0 +2023-02-06 21:13:55,509 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139787.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:14:09,008 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139807.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:14:11,597 INFO [train.py:901] (3/4) Epoch 18, batch 2400, loss[loss=0.1913, simple_loss=0.2746, pruned_loss=0.05397, over 8247.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2945, pruned_loss=0.06767, over 1617829.14 frames. ], batch size: 22, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:14:20,253 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139822.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:14:21,638 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139824.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:14:37,679 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5603, 2.6682, 1.8719, 2.3490, 2.3171, 1.6105, 2.3208, 2.3756], + device='cuda:3'), covar=tensor([0.1453, 0.0398, 0.1236, 0.0711, 0.0746, 0.1507, 0.0947, 0.1003], + device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0233, 0.0322, 0.0299, 0.0294, 0.0328, 0.0337, 0.0310], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 21:14:44,092 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2091, 2.5252, 2.9150, 1.6505, 2.9760, 1.7359, 1.5179, 2.0835], + device='cuda:3'), covar=tensor([0.0688, 0.0326, 0.0228, 0.0655, 0.0365, 0.0845, 0.0835, 0.0543], + device='cuda:3'), in_proj_covar=tensor([0.0432, 0.0372, 0.0322, 0.0427, 0.0358, 0.0518, 0.0379, 0.0401], + device='cuda:3'), out_proj_covar=tensor([1.1782e-04, 9.8596e-05, 8.5355e-05, 1.1361e-04, 9.5606e-05, 1.4841e-04, + 1.0307e-04, 1.0736e-04], device='cuda:3') +2023-02-06 21:14:48,492 INFO [train.py:901] (3/4) Epoch 18, batch 2450, loss[loss=0.1911, simple_loss=0.2713, pruned_loss=0.05541, over 7642.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2947, pruned_loss=0.06757, over 1618535.46 frames. ], batch size: 19, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:14:53,495 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139868.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:14:58,179 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.793e+02 2.359e+02 2.854e+02 3.442e+02 8.627e+02, threshold=5.708e+02, percent-clipped=1.0 +2023-02-06 21:15:00,747 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-06 21:15:23,626 INFO [train.py:901] (3/4) Epoch 18, batch 2500, loss[loss=0.2435, simple_loss=0.3157, pruned_loss=0.08564, over 6846.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.295, pruned_loss=0.06772, over 1615627.25 frames. ], batch size: 71, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:15:39,668 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139933.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:15:47,236 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=139944.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:16:00,232 INFO [train.py:901] (3/4) Epoch 18, batch 2550, loss[loss=0.2317, simple_loss=0.3079, pruned_loss=0.07777, over 8087.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2939, pruned_loss=0.06725, over 1618967.82 frames. ], batch size: 21, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:16:07,334 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139971.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:16:09,801 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.528e+02 2.329e+02 2.906e+02 3.594e+02 7.294e+02, threshold=5.811e+02, percent-clipped=3.0 +2023-02-06 21:16:25,070 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139996.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:16:35,499 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140009.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:16:36,613 INFO [train.py:901] (3/4) Epoch 18, batch 2600, loss[loss=0.2226, simple_loss=0.2895, pruned_loss=0.07783, over 7534.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2943, pruned_loss=0.06775, over 1616782.06 frames. ], batch size: 18, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:16:39,140 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-06 21:16:44,583 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140022.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:16:52,715 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140034.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:17:01,557 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140047.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:17:03,005 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140048.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:17:07,127 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5774, 2.5952, 1.8959, 2.2587, 2.1330, 1.5765, 2.1624, 2.1786], + device='cuda:3'), covar=tensor([0.1423, 0.0334, 0.1102, 0.0588, 0.0726, 0.1417, 0.0894, 0.0956], + device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0234, 0.0323, 0.0300, 0.0296, 0.0329, 0.0339, 0.0312], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 21:17:10,539 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140059.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:17:11,785 INFO [train.py:901] (3/4) Epoch 18, batch 2650, loss[loss=0.2191, simple_loss=0.3108, pruned_loss=0.06373, over 8542.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2949, pruned_loss=0.06767, over 1619492.71 frames. ], batch size: 49, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:17:22,349 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.675e+02 2.571e+02 2.973e+02 3.666e+02 6.732e+02, threshold=5.945e+02, percent-clipped=3.0 +2023-02-06 21:17:47,905 INFO [train.py:901] (3/4) Epoch 18, batch 2700, loss[loss=0.2172, simple_loss=0.2825, pruned_loss=0.07595, over 7812.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.2949, pruned_loss=0.06784, over 1616934.02 frames. ], batch size: 20, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:17:55,845 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140121.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:18:00,544 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140128.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:18:02,536 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140131.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:18:16,416 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140151.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:18:23,213 INFO [train.py:901] (3/4) Epoch 18, batch 2750, loss[loss=0.2349, simple_loss=0.312, pruned_loss=0.07889, over 8469.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2943, pruned_loss=0.06768, over 1614067.07 frames. ], batch size: 25, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:18:23,409 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9052, 1.6262, 1.6741, 1.4617, 0.9031, 1.4904, 1.7145, 1.3970], + device='cuda:3'), covar=tensor([0.0487, 0.1184, 0.1612, 0.1315, 0.0601, 0.1472, 0.0679, 0.0638], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0190, 0.0158, 0.0100, 0.0161, 0.0113, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 21:18:27,490 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140166.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:18:28,899 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140168.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:18:33,791 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.559e+02 2.336e+02 2.919e+02 3.807e+02 8.313e+02, threshold=5.838e+02, percent-clipped=5.0 +2023-02-06 21:18:50,343 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3048, 2.1642, 1.6565, 1.9160, 1.7713, 1.4013, 1.6179, 1.6774], + device='cuda:3'), covar=tensor([0.1228, 0.0358, 0.1059, 0.0544, 0.0687, 0.1490, 0.0927, 0.0833], + device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0235, 0.0323, 0.0302, 0.0296, 0.0331, 0.0340, 0.0314], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 21:19:00,759 INFO [train.py:901] (3/4) Epoch 18, batch 2800, loss[loss=0.2009, simple_loss=0.28, pruned_loss=0.06093, over 7808.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2932, pruned_loss=0.06687, over 1611931.92 frames. ], batch size: 20, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:19:01,504 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140212.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:19:25,733 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140246.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:19:35,893 INFO [train.py:901] (3/4) Epoch 18, batch 2850, loss[loss=0.2236, simple_loss=0.3108, pruned_loss=0.06826, over 8186.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2928, pruned_loss=0.06643, over 1611243.79 frames. ], batch size: 23, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:19:39,583 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140266.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:19:45,685 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.859e+02 2.447e+02 2.919e+02 3.574e+02 5.806e+02, threshold=5.838e+02, percent-clipped=0.0 +2023-02-06 21:19:47,313 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140277.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:19:50,917 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140281.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:19:52,321 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140283.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:19:58,486 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([6.0935, 1.4863, 6.2642, 2.3208, 5.6586, 5.3322, 5.7747, 5.6794], + device='cuda:3'), covar=tensor([0.0484, 0.4730, 0.0346, 0.3383, 0.0997, 0.0820, 0.0506, 0.0512], + device='cuda:3'), in_proj_covar=tensor([0.0587, 0.0620, 0.0666, 0.0596, 0.0676, 0.0581, 0.0577, 0.0647], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 21:20:06,056 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3624, 2.0409, 1.6640, 2.0562, 1.7801, 1.3040, 1.7302, 1.8456], + device='cuda:3'), covar=tensor([0.1135, 0.0444, 0.1216, 0.0480, 0.0766, 0.1651, 0.0929, 0.0724], + device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0235, 0.0322, 0.0302, 0.0297, 0.0331, 0.0340, 0.0313], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 21:20:07,428 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140304.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:20:10,080 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5964, 1.4178, 4.7481, 1.8592, 4.2550, 4.0171, 4.3025, 4.1694], + device='cuda:3'), covar=tensor([0.0501, 0.4606, 0.0477, 0.3447, 0.1055, 0.0844, 0.0532, 0.0582], + device='cuda:3'), in_proj_covar=tensor([0.0587, 0.0619, 0.0666, 0.0597, 0.0675, 0.0581, 0.0578, 0.0647], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 21:20:10,419 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-06 21:20:11,947 INFO [train.py:901] (3/4) Epoch 18, batch 2900, loss[loss=0.2337, simple_loss=0.3245, pruned_loss=0.07143, over 8292.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2946, pruned_loss=0.06717, over 1614183.14 frames. ], batch size: 23, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:20:13,912 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 +2023-02-06 21:20:14,963 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140315.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:20:23,768 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140327.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:20:25,189 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140329.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:20:33,504 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140340.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:20:44,300 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 21:20:47,991 INFO [train.py:901] (3/4) Epoch 18, batch 2950, loss[loss=0.2172, simple_loss=0.3044, pruned_loss=0.06496, over 6970.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.2948, pruned_loss=0.06719, over 1612979.23 frames. ], batch size: 71, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:20:57,359 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.671e+02 3.280e+02 4.327e+02 7.160e+02, threshold=6.561e+02, percent-clipped=5.0 +2023-02-06 21:21:07,285 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8792, 1.6273, 2.0345, 1.8003, 1.8249, 1.9083, 1.6209, 0.7655], + device='cuda:3'), covar=tensor([0.4981, 0.4190, 0.1647, 0.2740, 0.2122, 0.2739, 0.1955, 0.4234], + device='cuda:3'), in_proj_covar=tensor([0.0923, 0.0943, 0.0781, 0.0908, 0.0975, 0.0862, 0.0727, 0.0806], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 21:21:18,955 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140405.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:21:23,797 INFO [train.py:901] (3/4) Epoch 18, batch 3000, loss[loss=0.1994, simple_loss=0.2981, pruned_loss=0.05029, over 8296.00 frames. ], tot_loss[loss=0.2166, simple_loss=0.2964, pruned_loss=0.06836, over 1613433.70 frames. ], batch size: 23, lr: 4.24e-03, grad_scale: 8.0 +2023-02-06 21:21:23,798 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 21:21:32,958 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7231, 1.8227, 1.6959, 2.2445, 1.1686, 1.4624, 1.6605, 1.7863], + device='cuda:3'), covar=tensor([0.0753, 0.0846, 0.0925, 0.0438, 0.1124, 0.1317, 0.0829, 0.0853], + device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0201, 0.0252, 0.0214, 0.0209, 0.0251, 0.0258, 0.0214], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 21:21:37,682 INFO [train.py:935] (3/4) Epoch 18, validation: loss=0.1773, simple_loss=0.2774, pruned_loss=0.03861, over 944034.00 frames. +2023-02-06 21:21:37,683 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 21:22:14,087 INFO [train.py:901] (3/4) Epoch 18, batch 3050, loss[loss=0.2374, simple_loss=0.3136, pruned_loss=0.0806, over 8505.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.297, pruned_loss=0.069, over 1615569.84 frames. ], batch size: 26, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:22:16,897 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140465.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:22:21,664 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140472.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:22:24,218 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.478e+02 2.663e+02 3.172e+02 4.119e+02 9.916e+02, threshold=6.345e+02, percent-clipped=7.0 +2023-02-06 21:22:42,899 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140502.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:22:46,271 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140507.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:22:48,875 INFO [train.py:901] (3/4) Epoch 18, batch 3100, loss[loss=0.1947, simple_loss=0.2668, pruned_loss=0.0613, over 7975.00 frames. ], tot_loss[loss=0.2186, simple_loss=0.2979, pruned_loss=0.0697, over 1616591.62 frames. ], batch size: 21, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:22:56,952 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140522.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:00,971 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140527.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:01,034 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140527.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:07,883 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140537.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:09,293 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140539.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:15,567 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140547.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:25,516 INFO [train.py:901] (3/4) Epoch 18, batch 3150, loss[loss=0.2787, simple_loss=0.3473, pruned_loss=0.105, over 8322.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2976, pruned_loss=0.06913, over 1617756.45 frames. ], batch size: 26, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:23:26,371 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140562.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:27,692 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140564.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:34,931 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.450e+02 2.438e+02 2.948e+02 4.263e+02 1.019e+03, threshold=5.895e+02, percent-clipped=4.0 +2023-02-06 21:23:38,588 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140580.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:40,639 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140583.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:43,426 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140587.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:23:59,142 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140608.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:24:01,013 INFO [train.py:901] (3/4) Epoch 18, batch 3200, loss[loss=0.2524, simple_loss=0.3349, pruned_loss=0.08499, over 8299.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2962, pruned_loss=0.06805, over 1617664.63 frames. ], batch size: 23, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:24:06,861 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-06 21:24:07,891 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140621.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:24:11,471 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3929, 2.4014, 1.6415, 2.2370, 2.1595, 1.3464, 1.9928, 2.1175], + device='cuda:3'), covar=tensor([0.1409, 0.0406, 0.1337, 0.0623, 0.0751, 0.1675, 0.0997, 0.0843], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0233, 0.0323, 0.0302, 0.0296, 0.0330, 0.0339, 0.0313], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 21:24:36,825 INFO [train.py:901] (3/4) Epoch 18, batch 3250, loss[loss=0.1738, simple_loss=0.2602, pruned_loss=0.04373, over 8287.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2941, pruned_loss=0.06686, over 1616227.20 frames. ], batch size: 23, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:24:46,453 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.683e+02 2.256e+02 2.889e+02 3.448e+02 6.536e+02, threshold=5.777e+02, percent-clipped=1.0 +2023-02-06 21:25:13,080 INFO [train.py:901] (3/4) Epoch 18, batch 3300, loss[loss=0.2352, simple_loss=0.3357, pruned_loss=0.06736, over 8250.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2949, pruned_loss=0.06688, over 1619962.13 frames. ], batch size: 24, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:25:27,703 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2777, 1.3472, 3.3684, 1.0527, 3.0032, 2.8168, 3.1028, 3.0101], + device='cuda:3'), covar=tensor([0.0820, 0.3922, 0.0812, 0.3967, 0.1369, 0.1105, 0.0757, 0.0836], + device='cuda:3'), in_proj_covar=tensor([0.0589, 0.0621, 0.0667, 0.0596, 0.0677, 0.0579, 0.0575, 0.0644], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 21:25:27,797 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0875, 1.2379, 1.1690, 0.7705, 1.1716, 1.0144, 0.1183, 1.1874], + device='cuda:3'), covar=tensor([0.0334, 0.0331, 0.0288, 0.0432, 0.0373, 0.0896, 0.0706, 0.0273], + device='cuda:3'), in_proj_covar=tensor([0.0431, 0.0371, 0.0319, 0.0427, 0.0356, 0.0516, 0.0375, 0.0395], + device='cuda:3'), out_proj_covar=tensor([1.1729e-04, 9.8323e-05, 8.4337e-05, 1.1388e-04, 9.4896e-05, 1.4785e-04, + 1.0217e-04, 1.0561e-04], device='cuda:3') +2023-02-06 21:25:30,571 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140736.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:25:33,929 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140741.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:25:39,311 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140749.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:25:40,431 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-06 21:25:47,450 INFO [train.py:901] (3/4) Epoch 18, batch 3350, loss[loss=0.2394, simple_loss=0.3233, pruned_loss=0.07773, over 8287.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2948, pruned_loss=0.06661, over 1620544.50 frames. ], batch size: 23, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:25:57,594 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.611e+02 2.504e+02 2.969e+02 3.727e+02 7.020e+02, threshold=5.938e+02, percent-clipped=2.0 +2023-02-06 21:26:23,949 INFO [train.py:901] (3/4) Epoch 18, batch 3400, loss[loss=0.1682, simple_loss=0.2503, pruned_loss=0.04308, over 7967.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2941, pruned_loss=0.0663, over 1615088.82 frames. ], batch size: 21, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:26:35,827 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140827.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:26:42,132 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140836.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:26:46,754 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140843.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:26:52,093 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140851.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:26:58,805 INFO [train.py:901] (3/4) Epoch 18, batch 3450, loss[loss=0.2333, simple_loss=0.3162, pruned_loss=0.07519, over 8522.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2942, pruned_loss=0.06654, over 1614739.84 frames. ], batch size: 28, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:26:59,032 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140861.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:27:01,073 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140864.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:27:03,764 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140868.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:27:05,721 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=140871.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:27:08,268 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.604e+02 2.419e+02 3.065e+02 3.703e+02 6.567e+02, threshold=6.131e+02, percent-clipped=3.0 +2023-02-06 21:27:34,153 INFO [train.py:901] (3/4) Epoch 18, batch 3500, loss[loss=0.2461, simple_loss=0.3134, pruned_loss=0.08939, over 8511.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2935, pruned_loss=0.0665, over 1611611.89 frames. ], batch size: 26, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:27:51,059 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 21:27:51,204 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140935.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:28:09,211 INFO [train.py:901] (3/4) Epoch 18, batch 3550, loss[loss=0.2078, simple_loss=0.2836, pruned_loss=0.066, over 7432.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2923, pruned_loss=0.0658, over 1608069.50 frames. ], batch size: 17, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:28:11,375 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9096, 2.0866, 1.9111, 2.6008, 1.2424, 1.5863, 1.8921, 2.1324], + device='cuda:3'), covar=tensor([0.0784, 0.0836, 0.0844, 0.0374, 0.1016, 0.1314, 0.0780, 0.0750], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0200, 0.0251, 0.0213, 0.0206, 0.0249, 0.0255, 0.0212], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 21:28:12,016 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140965.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:28:12,767 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140966.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:28:18,745 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.584e+02 2.456e+02 3.083e+02 3.681e+02 6.081e+02, threshold=6.167e+02, percent-clipped=0.0 +2023-02-06 21:28:26,462 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140986.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:28:30,642 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140992.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:28:44,285 INFO [train.py:901] (3/4) Epoch 18, batch 3600, loss[loss=0.2023, simple_loss=0.278, pruned_loss=0.06325, over 7235.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2928, pruned_loss=0.06633, over 1608844.65 frames. ], batch size: 16, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:28:49,259 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141017.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:29:20,370 INFO [train.py:901] (3/4) Epoch 18, batch 3650, loss[loss=0.2482, simple_loss=0.3331, pruned_loss=0.08168, over 8341.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2931, pruned_loss=0.0666, over 1612628.65 frames. ], batch size: 26, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:29:30,821 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.558e+02 2.345e+02 2.956e+02 3.633e+02 6.454e+02, threshold=5.912e+02, percent-clipped=1.0 +2023-02-06 21:29:37,715 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141085.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:29:55,733 INFO [train.py:901] (3/4) Epoch 18, batch 3700, loss[loss=0.1511, simple_loss=0.2372, pruned_loss=0.03252, over 7814.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2923, pruned_loss=0.06599, over 1613029.23 frames. ], batch size: 20, lr: 4.23e-03, grad_scale: 8.0 +2023-02-06 21:29:57,139 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 21:30:02,958 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141120.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:30:20,676 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141145.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:30:31,544 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4701, 4.4693, 4.0860, 2.1671, 3.9659, 4.1073, 4.1986, 3.7660], + device='cuda:3'), covar=tensor([0.0726, 0.0603, 0.0976, 0.4473, 0.0771, 0.1100, 0.1183, 0.0904], + device='cuda:3'), in_proj_covar=tensor([0.0507, 0.0422, 0.0420, 0.0522, 0.0412, 0.0422, 0.0408, 0.0369], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 21:30:32,112 INFO [train.py:901] (3/4) Epoch 18, batch 3750, loss[loss=0.2388, simple_loss=0.3242, pruned_loss=0.0767, over 8620.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2936, pruned_loss=0.06653, over 1616286.12 frames. ], batch size: 39, lr: 4.22e-03, grad_scale: 8.0 +2023-02-06 21:30:32,288 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141161.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:30:39,106 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141171.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:30:41,858 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.679e+02 3.309e+02 4.099e+02 7.455e+02, threshold=6.618e+02, percent-clipped=7.0 +2023-02-06 21:31:00,278 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141200.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:31:07,419 INFO [train.py:901] (3/4) Epoch 18, batch 3800, loss[loss=0.2204, simple_loss=0.282, pruned_loss=0.07939, over 7717.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2932, pruned_loss=0.06627, over 1613381.97 frames. ], batch size: 18, lr: 4.22e-03, grad_scale: 8.0 +2023-02-06 21:31:15,017 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141222.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:31:29,266 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141242.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:31:32,639 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141247.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:31:39,638 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.87 vs. limit=5.0 +2023-02-06 21:31:42,608 INFO [train.py:901] (3/4) Epoch 18, batch 3850, loss[loss=0.2165, simple_loss=0.2878, pruned_loss=0.07261, over 7548.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.2936, pruned_loss=0.06625, over 1613411.44 frames. ], batch size: 18, lr: 4.22e-03, grad_scale: 8.0 +2023-02-06 21:31:46,907 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141267.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:31:52,711 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.574e+02 2.500e+02 3.018e+02 3.684e+02 7.912e+02, threshold=6.036e+02, percent-clipped=1.0 +2023-02-06 21:31:52,871 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.2431, 4.2206, 3.8823, 2.1143, 3.7894, 3.9317, 3.8103, 3.5400], + device='cuda:3'), covar=tensor([0.0921, 0.0633, 0.1036, 0.4300, 0.0856, 0.0796, 0.1314, 0.0713], + device='cuda:3'), in_proj_covar=tensor([0.0507, 0.0424, 0.0420, 0.0522, 0.0414, 0.0422, 0.0406, 0.0369], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 21:31:55,473 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141279.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:32:00,516 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141286.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:32:03,623 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-06 21:32:03,896 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 21:32:17,071 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141309.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:32:18,348 INFO [train.py:901] (3/4) Epoch 18, batch 3900, loss[loss=0.2255, simple_loss=0.303, pruned_loss=0.07397, over 7667.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.2935, pruned_loss=0.06629, over 1609317.72 frames. ], batch size: 19, lr: 4.22e-03, grad_scale: 8.0 +2023-02-06 21:32:42,404 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141347.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:32:52,481 INFO [train.py:901] (3/4) Epoch 18, batch 3950, loss[loss=0.198, simple_loss=0.2864, pruned_loss=0.05476, over 8038.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2944, pruned_loss=0.06716, over 1607672.38 frames. ], batch size: 22, lr: 4.22e-03, grad_scale: 8.0 +2023-02-06 21:33:02,707 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.596e+02 2.421e+02 2.990e+02 3.795e+02 7.053e+02, threshold=5.979e+02, percent-clipped=3.0 +2023-02-06 21:33:15,870 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141394.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:33:27,611 INFO [train.py:901] (3/4) Epoch 18, batch 4000, loss[loss=0.21, simple_loss=0.2893, pruned_loss=0.06533, over 8250.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2945, pruned_loss=0.06727, over 1609975.70 frames. ], batch size: 24, lr: 4.22e-03, grad_scale: 16.0 +2023-02-06 21:33:37,287 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141424.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:33:40,604 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4467, 2.0246, 2.8318, 2.2280, 2.7563, 2.3317, 2.1187, 1.4034], + device='cuda:3'), covar=tensor([0.4695, 0.4711, 0.1691, 0.3437, 0.2337, 0.2760, 0.1803, 0.4944], + device='cuda:3'), in_proj_covar=tensor([0.0921, 0.0942, 0.0779, 0.0906, 0.0979, 0.0858, 0.0725, 0.0804], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 21:33:44,517 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141435.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 21:33:45,250 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7547, 2.2944, 4.3595, 1.4503, 3.1233, 2.3369, 1.8232, 2.8410], + device='cuda:3'), covar=tensor([0.1720, 0.2328, 0.0751, 0.4185, 0.1589, 0.2854, 0.1999, 0.2393], + device='cuda:3'), in_proj_covar=tensor([0.0510, 0.0575, 0.0549, 0.0620, 0.0637, 0.0578, 0.0512, 0.0626], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 21:33:58,720 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141456.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:34:00,772 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141459.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:34:02,041 INFO [train.py:901] (3/4) Epoch 18, batch 4050, loss[loss=0.2105, simple_loss=0.282, pruned_loss=0.06945, over 7970.00 frames. ], tot_loss[loss=0.213, simple_loss=0.2931, pruned_loss=0.06644, over 1603521.40 frames. ], batch size: 21, lr: 4.22e-03, grad_scale: 16.0 +2023-02-06 21:34:02,224 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9762, 3.7907, 2.4053, 2.8277, 2.6954, 2.0885, 2.8131, 2.9013], + device='cuda:3'), covar=tensor([0.1575, 0.0333, 0.1041, 0.0735, 0.0840, 0.1355, 0.1036, 0.1098], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0234, 0.0326, 0.0306, 0.0299, 0.0331, 0.0341, 0.0317], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 21:34:02,886 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0530, 1.4626, 1.5936, 1.3612, 0.9845, 1.3908, 1.8034, 1.5561], + device='cuda:3'), covar=tensor([0.0506, 0.1230, 0.1721, 0.1405, 0.0601, 0.1485, 0.0680, 0.0642], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0153, 0.0191, 0.0157, 0.0099, 0.0162, 0.0114, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 21:34:12,698 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.609e+02 2.514e+02 3.146e+02 4.229e+02 8.641e+02, threshold=6.293e+02, percent-clipped=9.0 +2023-02-06 21:34:16,935 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141481.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:34:22,711 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.16 vs. limit=5.0 +2023-02-06 21:34:34,632 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141505.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:34:38,481 INFO [train.py:901] (3/4) Epoch 18, batch 4100, loss[loss=0.2096, simple_loss=0.2933, pruned_loss=0.06298, over 8579.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2929, pruned_loss=0.06619, over 1606361.08 frames. ], batch size: 39, lr: 4.22e-03, grad_scale: 16.0 +2023-02-06 21:34:41,368 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5474, 1.5368, 2.9211, 1.3390, 2.2125, 3.1490, 3.2399, 2.7623], + device='cuda:3'), covar=tensor([0.1158, 0.1460, 0.0346, 0.1903, 0.0798, 0.0268, 0.0557, 0.0582], + device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0313, 0.0276, 0.0309, 0.0297, 0.0256, 0.0398, 0.0298], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 21:34:53,730 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8249, 1.8360, 2.5503, 1.6796, 1.3739, 2.5487, 0.6691, 1.5096], + device='cuda:3'), covar=tensor([0.2025, 0.1493, 0.0364, 0.1493, 0.3151, 0.0386, 0.2057, 0.1531], + device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0188, 0.0119, 0.0216, 0.0261, 0.0127, 0.0164, 0.0183], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 21:35:00,227 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141542.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:35:13,079 INFO [train.py:901] (3/4) Epoch 18, batch 4150, loss[loss=0.238, simple_loss=0.3044, pruned_loss=0.08583, over 8486.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2924, pruned_loss=0.06604, over 1609075.50 frames. ], batch size: 29, lr: 4.22e-03, grad_scale: 16.0 +2023-02-06 21:35:17,346 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141567.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:35:22,680 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.494e+02 2.507e+02 2.964e+02 3.952e+02 7.900e+02, threshold=5.928e+02, percent-clipped=3.0 +2023-02-06 21:35:48,912 INFO [train.py:901] (3/4) Epoch 18, batch 4200, loss[loss=0.2014, simple_loss=0.2843, pruned_loss=0.05928, over 8565.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2931, pruned_loss=0.06606, over 1609364.96 frames. ], batch size: 34, lr: 4.22e-03, grad_scale: 16.0 +2023-02-06 21:35:55,793 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141620.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:36:02,339 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 21:36:16,008 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141650.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:36:23,890 INFO [train.py:901] (3/4) Epoch 18, batch 4250, loss[loss=0.2218, simple_loss=0.2934, pruned_loss=0.07513, over 8458.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.294, pruned_loss=0.06671, over 1611123.31 frames. ], batch size: 27, lr: 4.22e-03, grad_scale: 16.0 +2023-02-06 21:36:24,612 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 21:36:33,262 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.772e+02 2.491e+02 2.994e+02 3.932e+02 8.485e+02, threshold=5.988e+02, percent-clipped=6.0 +2023-02-06 21:36:33,494 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141675.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:36:36,884 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141680.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:36:44,013 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141691.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:36:54,160 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141705.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:36:58,030 INFO [train.py:901] (3/4) Epoch 18, batch 4300, loss[loss=0.219, simple_loss=0.2946, pruned_loss=0.07166, over 8103.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2946, pruned_loss=0.06699, over 1609095.09 frames. ], batch size: 23, lr: 4.22e-03, grad_scale: 16.0 +2023-02-06 21:37:32,932 INFO [train.py:901] (3/4) Epoch 18, batch 4350, loss[loss=0.2452, simple_loss=0.3201, pruned_loss=0.08514, over 8186.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2936, pruned_loss=0.06661, over 1611614.71 frames. ], batch size: 23, lr: 4.22e-03, grad_scale: 16.0 +2023-02-06 21:37:38,873 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 +2023-02-06 21:37:43,208 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.448e+02 2.620e+02 3.197e+02 4.150e+02 9.266e+02, threshold=6.393e+02, percent-clipped=5.0 +2023-02-06 21:37:46,102 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141779.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 21:37:54,218 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 21:38:02,437 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=141803.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:38:04,522 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141806.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:38:07,896 INFO [train.py:901] (3/4) Epoch 18, batch 4400, loss[loss=0.2047, simple_loss=0.2923, pruned_loss=0.05858, over 8653.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.2932, pruned_loss=0.06649, over 1611120.85 frames. ], batch size: 34, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:38:36,615 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 21:38:43,845 INFO [train.py:901] (3/4) Epoch 18, batch 4450, loss[loss=0.1771, simple_loss=0.2761, pruned_loss=0.03908, over 8318.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2938, pruned_loss=0.0672, over 1611500.19 frames. ], batch size: 25, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:38:53,330 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.555e+02 2.507e+02 2.868e+02 3.524e+02 7.777e+02, threshold=5.735e+02, percent-clipped=2.0 +2023-02-06 21:38:54,257 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141876.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:39:07,102 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141894.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 21:39:11,833 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141901.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:39:18,273 INFO [train.py:901] (3/4) Epoch 18, batch 4500, loss[loss=0.2189, simple_loss=0.3004, pruned_loss=0.06868, over 7800.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2931, pruned_loss=0.06635, over 1609240.11 frames. ], batch size: 20, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:39:23,247 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141918.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:39:27,796 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 21:39:34,740 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141934.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:39:42,678 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141946.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:39:53,388 INFO [train.py:901] (3/4) Epoch 18, batch 4550, loss[loss=0.1944, simple_loss=0.2833, pruned_loss=0.05278, over 8693.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2929, pruned_loss=0.06619, over 1608167.75 frames. ], batch size: 39, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:40:03,505 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.745e+02 2.488e+02 2.920e+02 3.454e+02 6.371e+02, threshold=5.840e+02, percent-clipped=2.0 +2023-02-06 21:40:29,745 INFO [train.py:901] (3/4) Epoch 18, batch 4600, loss[loss=0.2331, simple_loss=0.3209, pruned_loss=0.07263, over 8735.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2937, pruned_loss=0.06667, over 1612498.28 frames. ], batch size: 30, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:41:02,726 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2023-02-06 21:41:04,181 INFO [train.py:901] (3/4) Epoch 18, batch 4650, loss[loss=0.256, simple_loss=0.3217, pruned_loss=0.09516, over 8454.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.293, pruned_loss=0.06673, over 1609616.65 frames. ], batch size: 49, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:41:05,109 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142062.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:41:13,897 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.464e+02 2.389e+02 2.901e+02 3.503e+02 7.256e+02, threshold=5.801e+02, percent-clipped=3.0 +2023-02-06 21:41:23,668 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142087.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:41:39,475 INFO [train.py:901] (3/4) Epoch 18, batch 4700, loss[loss=0.1961, simple_loss=0.2822, pruned_loss=0.05503, over 8128.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2935, pruned_loss=0.06675, over 1610213.18 frames. ], batch size: 22, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:41:56,995 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0633, 1.7518, 2.3450, 1.9917, 2.2753, 2.0686, 1.7817, 1.1140], + device='cuda:3'), covar=tensor([0.5105, 0.4439, 0.1817, 0.3108, 0.2143, 0.2627, 0.1874, 0.4687], + device='cuda:3'), in_proj_covar=tensor([0.0927, 0.0947, 0.0783, 0.0910, 0.0981, 0.0863, 0.0731, 0.0810], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 21:42:06,020 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142150.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 21:42:13,363 INFO [train.py:901] (3/4) Epoch 18, batch 4750, loss[loss=0.2112, simple_loss=0.3037, pruned_loss=0.05934, over 8243.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2926, pruned_loss=0.06623, over 1611077.23 frames. ], batch size: 24, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:42:17,759 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8379, 3.8235, 3.4999, 1.8099, 3.3235, 3.4630, 3.4206, 3.2352], + device='cuda:3'), covar=tensor([0.0952, 0.0666, 0.1115, 0.4673, 0.1076, 0.1045, 0.1443, 0.0891], + device='cuda:3'), in_proj_covar=tensor([0.0509, 0.0425, 0.0419, 0.0520, 0.0412, 0.0418, 0.0403, 0.0366], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 21:42:23,435 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142174.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:42:23,891 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.545e+02 2.398e+02 2.792e+02 3.541e+02 9.190e+02, threshold=5.585e+02, percent-clipped=4.0 +2023-02-06 21:42:24,100 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142175.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 21:42:28,133 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.6059, 1.6297, 5.8639, 2.4031, 4.7510, 4.8536, 5.3907, 5.3763], + device='cuda:3'), covar=tensor([0.1154, 0.7281, 0.0870, 0.4435, 0.2693, 0.1476, 0.1018, 0.0936], + device='cuda:3'), in_proj_covar=tensor([0.0594, 0.0625, 0.0670, 0.0600, 0.0682, 0.0582, 0.0575, 0.0642], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 21:42:30,579 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 21:42:32,654 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 21:42:41,670 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142199.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:42:49,348 INFO [train.py:901] (3/4) Epoch 18, batch 4800, loss[loss=0.2325, simple_loss=0.3195, pruned_loss=0.07272, over 8290.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.2934, pruned_loss=0.06642, over 1617924.45 frames. ], batch size: 23, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:43:00,603 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9126, 2.5312, 3.5496, 1.8875, 1.8610, 3.4586, 0.6896, 2.0893], + device='cuda:3'), covar=tensor([0.1477, 0.1384, 0.0268, 0.1898, 0.3149, 0.0357, 0.2675, 0.1391], + device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0189, 0.0121, 0.0219, 0.0266, 0.0128, 0.0165, 0.0184], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 21:43:23,866 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 21:43:24,513 INFO [train.py:901] (3/4) Epoch 18, batch 4850, loss[loss=0.284, simple_loss=0.3437, pruned_loss=0.1121, over 6969.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.293, pruned_loss=0.0666, over 1606278.20 frames. ], batch size: 72, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:43:33,968 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.520e+02 2.522e+02 3.053e+02 3.876e+02 6.315e+02, threshold=6.106e+02, percent-clipped=2.0 +2023-02-06 21:43:36,096 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=142278.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:43:45,106 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=142290.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:43:59,066 INFO [train.py:901] (3/4) Epoch 18, batch 4900, loss[loss=0.1849, simple_loss=0.2698, pruned_loss=0.04999, over 7918.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2926, pruned_loss=0.06641, over 1606670.81 frames. ], batch size: 20, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:44:34,497 INFO [train.py:901] (3/4) Epoch 18, batch 4950, loss[loss=0.1996, simple_loss=0.2915, pruned_loss=0.05387, over 8247.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2924, pruned_loss=0.0662, over 1609049.17 frames. ], batch size: 22, lr: 4.21e-03, grad_scale: 16.0 +2023-02-06 21:44:41,669 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.7789, 1.8649, 5.9070, 2.5511, 5.2288, 5.0057, 5.4233, 5.2965], + device='cuda:3'), covar=tensor([0.0479, 0.4784, 0.0391, 0.3402, 0.1070, 0.0937, 0.0513, 0.0528], + device='cuda:3'), in_proj_covar=tensor([0.0588, 0.0620, 0.0664, 0.0597, 0.0676, 0.0578, 0.0571, 0.0639], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 21:44:44,958 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.704e+02 2.451e+02 2.943e+02 3.789e+02 7.945e+02, threshold=5.886e+02, percent-clipped=1.0 +2023-02-06 21:44:57,238 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142393.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:45:05,916 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142405.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:45:09,747 INFO [train.py:901] (3/4) Epoch 18, batch 5000, loss[loss=0.2227, simple_loss=0.3081, pruned_loss=0.06864, over 8497.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.294, pruned_loss=0.06727, over 1611529.51 frames. ], batch size: 28, lr: 4.21e-03, grad_scale: 8.0 +2023-02-06 21:45:31,134 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.67 vs. limit=2.0 +2023-02-06 21:45:33,586 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1724, 1.4100, 4.3582, 1.5839, 3.8949, 3.6532, 3.9281, 3.8111], + device='cuda:3'), covar=tensor([0.0553, 0.4387, 0.0486, 0.3984, 0.0963, 0.0863, 0.0527, 0.0641], + device='cuda:3'), in_proj_covar=tensor([0.0587, 0.0619, 0.0663, 0.0594, 0.0676, 0.0577, 0.0569, 0.0639], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 21:45:44,314 INFO [train.py:901] (3/4) Epoch 18, batch 5050, loss[loss=0.21, simple_loss=0.3041, pruned_loss=0.05795, over 8471.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2942, pruned_loss=0.06704, over 1608627.92 frames. ], batch size: 25, lr: 4.21e-03, grad_scale: 8.0 +2023-02-06 21:45:54,476 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.667e+02 2.384e+02 2.804e+02 3.417e+02 5.925e+02, threshold=5.609e+02, percent-clipped=1.0 +2023-02-06 21:45:59,248 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9847, 1.7794, 3.3857, 1.5590, 2.3201, 3.6521, 3.7158, 3.1556], + device='cuda:3'), covar=tensor([0.1135, 0.1466, 0.0296, 0.1847, 0.0949, 0.0223, 0.0545, 0.0528], + device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0312, 0.0276, 0.0306, 0.0298, 0.0256, 0.0397, 0.0297], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 21:46:04,022 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 21:46:19,170 INFO [train.py:901] (3/4) Epoch 18, batch 5100, loss[loss=0.2147, simple_loss=0.3057, pruned_loss=0.0618, over 8034.00 frames. ], tot_loss[loss=0.213, simple_loss=0.2939, pruned_loss=0.06604, over 1612545.65 frames. ], batch size: 22, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:46:27,950 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6834, 1.9063, 2.1013, 1.4406, 2.1777, 1.4728, 0.6602, 1.9125], + device='cuda:3'), covar=tensor([0.0523, 0.0322, 0.0246, 0.0476, 0.0353, 0.0814, 0.0743, 0.0253], + device='cuda:3'), in_proj_covar=tensor([0.0427, 0.0369, 0.0316, 0.0425, 0.0357, 0.0514, 0.0370, 0.0393], + device='cuda:3'), out_proj_covar=tensor([1.1599e-04, 9.7765e-05, 8.3674e-05, 1.1317e-04, 9.5103e-05, 1.4723e-04, + 1.0050e-04, 1.0481e-04], device='cuda:3') +2023-02-06 21:46:54,249 INFO [train.py:901] (3/4) Epoch 18, batch 5150, loss[loss=0.244, simple_loss=0.2923, pruned_loss=0.09786, over 7692.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.294, pruned_loss=0.0666, over 1614305.19 frames. ], batch size: 18, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:47:04,404 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.592e+02 2.481e+02 3.004e+02 4.323e+02 1.197e+03, threshold=6.009e+02, percent-clipped=7.0 +2023-02-06 21:47:14,699 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142591.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:47:29,008 INFO [train.py:901] (3/4) Epoch 18, batch 5200, loss[loss=0.1993, simple_loss=0.285, pruned_loss=0.05684, over 8493.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2937, pruned_loss=0.06667, over 1613567.77 frames. ], batch size: 49, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:47:45,306 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.8932, 6.0141, 5.2165, 2.4069, 5.3027, 5.6407, 5.6043, 5.3861], + device='cuda:3'), covar=tensor([0.0494, 0.0396, 0.0841, 0.4331, 0.0656, 0.0670, 0.0919, 0.0647], + device='cuda:3'), in_proj_covar=tensor([0.0507, 0.0421, 0.0419, 0.0519, 0.0411, 0.0417, 0.0401, 0.0368], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 21:47:55,313 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142649.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:48:02,991 INFO [train.py:901] (3/4) Epoch 18, batch 5250, loss[loss=0.1837, simple_loss=0.2729, pruned_loss=0.04723, over 7801.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.2934, pruned_loss=0.06642, over 1615509.85 frames. ], batch size: 20, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:48:03,207 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142661.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:48:03,673 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 21:48:13,210 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142674.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:48:14,366 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.428e+02 2.491e+02 3.102e+02 3.692e+02 6.533e+02, threshold=6.204e+02, percent-clipped=2.0 +2023-02-06 21:48:21,254 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142686.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:48:37,897 INFO [train.py:901] (3/4) Epoch 18, batch 5300, loss[loss=0.2286, simple_loss=0.3092, pruned_loss=0.07399, over 8606.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2936, pruned_loss=0.06642, over 1613757.85 frames. ], batch size: 39, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:48:38,137 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3200, 2.6234, 3.1644, 1.6351, 3.2556, 1.9452, 1.5528, 2.2521], + device='cuda:3'), covar=tensor([0.0704, 0.0386, 0.0202, 0.0715, 0.0383, 0.0738, 0.0829, 0.0495], + device='cuda:3'), in_proj_covar=tensor([0.0429, 0.0369, 0.0318, 0.0427, 0.0357, 0.0516, 0.0370, 0.0395], + device='cuda:3'), out_proj_covar=tensor([1.1662e-04, 9.7769e-05, 8.4124e-05, 1.1369e-04, 9.4975e-05, 1.4790e-04, + 1.0063e-04, 1.0539e-04], device='cuda:3') +2023-02-06 21:48:53,423 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0329, 2.1574, 2.2985, 1.6022, 2.3115, 1.7206, 1.6565, 1.9343], + device='cuda:3'), covar=tensor([0.0520, 0.0348, 0.0231, 0.0480, 0.0366, 0.0571, 0.0632, 0.0363], + device='cuda:3'), in_proj_covar=tensor([0.0431, 0.0371, 0.0320, 0.0429, 0.0359, 0.0518, 0.0372, 0.0396], + device='cuda:3'), out_proj_covar=tensor([1.1717e-04, 9.8130e-05, 8.4633e-05, 1.1421e-04, 9.5457e-05, 1.4854e-04, + 1.0109e-04, 1.0578e-04], device='cuda:3') +2023-02-06 21:48:59,520 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142742.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:49:12,900 INFO [train.py:901] (3/4) Epoch 18, batch 5350, loss[loss=0.181, simple_loss=0.2664, pruned_loss=0.04781, over 8092.00 frames. ], tot_loss[loss=0.213, simple_loss=0.2937, pruned_loss=0.06618, over 1612572.45 frames. ], batch size: 21, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:49:22,760 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.596e+02 2.581e+02 3.011e+02 3.651e+02 7.168e+02, threshold=6.023e+02, percent-clipped=3.0 +2023-02-06 21:49:43,696 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.50 vs. limit=2.0 +2023-02-06 21:49:48,113 INFO [train.py:901] (3/4) Epoch 18, batch 5400, loss[loss=0.1909, simple_loss=0.2744, pruned_loss=0.0537, over 8035.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2927, pruned_loss=0.06587, over 1612939.17 frames. ], batch size: 22, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:49:49,774 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 +2023-02-06 21:50:06,772 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0776, 1.8613, 2.3683, 1.9619, 2.2161, 2.1334, 1.9153, 1.1151], + device='cuda:3'), covar=tensor([0.5215, 0.4523, 0.1760, 0.3762, 0.2694, 0.3042, 0.1908, 0.5179], + device='cuda:3'), in_proj_covar=tensor([0.0927, 0.0943, 0.0776, 0.0911, 0.0984, 0.0863, 0.0731, 0.0811], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 21:50:22,753 INFO [train.py:901] (3/4) Epoch 18, batch 5450, loss[loss=0.1988, simple_loss=0.2742, pruned_loss=0.06165, over 7808.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2912, pruned_loss=0.06524, over 1609336.47 frames. ], batch size: 19, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:50:33,557 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.700e+02 2.381e+02 3.003e+02 4.378e+02 7.690e+02, threshold=6.006e+02, percent-clipped=4.0 +2023-02-06 21:50:50,005 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 21:50:58,894 INFO [train.py:901] (3/4) Epoch 18, batch 5500, loss[loss=0.2456, simple_loss=0.3288, pruned_loss=0.08114, over 8105.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2928, pruned_loss=0.06558, over 1616099.07 frames. ], batch size: 23, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:51:14,989 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=142935.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:51:33,202 INFO [train.py:901] (3/4) Epoch 18, batch 5550, loss[loss=0.1772, simple_loss=0.2581, pruned_loss=0.04818, over 7794.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.2922, pruned_loss=0.06549, over 1615478.62 frames. ], batch size: 19, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:51:43,333 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.602e+02 2.398e+02 2.938e+02 3.826e+02 1.126e+03, threshold=5.876e+02, percent-clipped=10.0 +2023-02-06 21:52:08,277 INFO [train.py:901] (3/4) Epoch 18, batch 5600, loss[loss=0.2323, simple_loss=0.321, pruned_loss=0.07186, over 8745.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2924, pruned_loss=0.06511, over 1616646.92 frames. ], batch size: 30, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:52:19,261 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-06 21:52:23,148 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4079, 1.6268, 2.0960, 1.3101, 1.3602, 1.6222, 1.5251, 1.3743], + device='cuda:3'), covar=tensor([0.1697, 0.2155, 0.0865, 0.3903, 0.1940, 0.3148, 0.1983, 0.2206], + device='cuda:3'), in_proj_covar=tensor([0.0512, 0.0573, 0.0549, 0.0619, 0.0634, 0.0580, 0.0512, 0.0624], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 21:52:36,271 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143050.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:52:43,531 INFO [train.py:901] (3/4) Epoch 18, batch 5650, loss[loss=0.2192, simple_loss=0.3025, pruned_loss=0.06799, over 8504.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2924, pruned_loss=0.06573, over 1614447.80 frames. ], batch size: 26, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:52:54,545 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.751e+02 2.315e+02 3.071e+02 3.627e+02 7.364e+02, threshold=6.141e+02, percent-clipped=4.0 +2023-02-06 21:53:00,411 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 21:53:01,153 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=143086.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:53:18,744 INFO [train.py:901] (3/4) Epoch 18, batch 5700, loss[loss=0.2027, simple_loss=0.2853, pruned_loss=0.0601, over 8597.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2928, pruned_loss=0.06592, over 1610557.73 frames. ], batch size: 34, lr: 4.20e-03, grad_scale: 8.0 +2023-02-06 21:53:37,151 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5621, 3.0140, 2.5252, 4.1493, 1.7803, 2.0983, 2.4482, 3.1943], + device='cuda:3'), covar=tensor([0.0718, 0.0764, 0.0837, 0.0264, 0.1157, 0.1271, 0.1007, 0.0783], + device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0200, 0.0252, 0.0214, 0.0209, 0.0249, 0.0255, 0.0214], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 21:53:41,104 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1070, 2.1148, 1.7255, 1.9611, 1.6263, 1.4329, 1.5815, 1.6560], + device='cuda:3'), covar=tensor([0.1424, 0.0427, 0.1112, 0.0500, 0.0771, 0.1544, 0.0926, 0.0853], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0235, 0.0324, 0.0303, 0.0295, 0.0330, 0.0341, 0.0314], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 21:53:53,719 INFO [train.py:901] (3/4) Epoch 18, batch 5750, loss[loss=0.2217, simple_loss=0.3018, pruned_loss=0.07074, over 8320.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2929, pruned_loss=0.06638, over 1607871.88 frames. ], batch size: 25, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 21:53:56,196 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.88 vs. limit=5.0 +2023-02-06 21:54:04,018 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.527e+02 2.423e+02 2.839e+02 3.621e+02 5.889e+02, threshold=5.677e+02, percent-clipped=0.0 +2023-02-06 21:54:04,717 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 21:54:21,859 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143201.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:54:28,665 INFO [train.py:901] (3/4) Epoch 18, batch 5800, loss[loss=0.2014, simple_loss=0.2681, pruned_loss=0.06729, over 7790.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2922, pruned_loss=0.06635, over 1604627.26 frames. ], batch size: 19, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 21:54:43,951 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.7307, 4.7012, 4.1307, 2.1601, 4.1229, 4.2550, 4.2593, 4.0227], + device='cuda:3'), covar=tensor([0.0638, 0.0465, 0.1063, 0.4132, 0.0825, 0.0867, 0.1132, 0.0765], + device='cuda:3'), in_proj_covar=tensor([0.0506, 0.0421, 0.0420, 0.0517, 0.0412, 0.0419, 0.0402, 0.0368], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 21:55:04,389 INFO [train.py:901] (3/4) Epoch 18, batch 5850, loss[loss=0.2144, simple_loss=0.3069, pruned_loss=0.06101, over 8327.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2909, pruned_loss=0.0653, over 1603942.50 frames. ], batch size: 25, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 21:55:15,559 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.339e+02 2.467e+02 2.892e+02 3.630e+02 6.628e+02, threshold=5.783e+02, percent-clipped=2.0 +2023-02-06 21:55:36,578 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=143306.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:55:39,811 INFO [train.py:901] (3/4) Epoch 18, batch 5900, loss[loss=0.1718, simple_loss=0.2496, pruned_loss=0.04705, over 7814.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.291, pruned_loss=0.06531, over 1605815.93 frames. ], batch size: 20, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 21:55:53,317 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=143331.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:56:03,176 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4175, 2.2042, 2.8650, 2.3740, 2.8075, 2.3665, 2.1212, 1.6175], + device='cuda:3'), covar=tensor([0.4796, 0.4199, 0.1559, 0.3644, 0.2409, 0.2858, 0.1853, 0.4749], + device='cuda:3'), in_proj_covar=tensor([0.0929, 0.0946, 0.0777, 0.0915, 0.0981, 0.0866, 0.0729, 0.0812], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 21:56:14,464 INFO [train.py:901] (3/4) Epoch 18, batch 5950, loss[loss=0.2255, simple_loss=0.2964, pruned_loss=0.07725, over 7938.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2915, pruned_loss=0.06554, over 1609368.30 frames. ], batch size: 20, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 21:56:25,158 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.550e+02 2.388e+02 2.875e+02 3.741e+02 7.794e+02, threshold=5.749e+02, percent-clipped=3.0 +2023-02-06 21:56:49,478 INFO [train.py:901] (3/4) Epoch 18, batch 6000, loss[loss=0.2492, simple_loss=0.3307, pruned_loss=0.0838, over 8191.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.291, pruned_loss=0.06504, over 1609910.24 frames. ], batch size: 23, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 21:56:49,478 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 21:57:03,428 INFO [train.py:935] (3/4) Epoch 18, validation: loss=0.1765, simple_loss=0.2767, pruned_loss=0.03814, over 944034.00 frames. +2023-02-06 21:57:03,430 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 21:57:07,209 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8703, 2.2209, 1.8615, 2.9224, 1.3488, 1.6298, 2.1462, 2.2582], + device='cuda:3'), covar=tensor([0.0794, 0.0849, 0.0901, 0.0394, 0.1267, 0.1465, 0.0960, 0.0924], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0199, 0.0251, 0.0213, 0.0208, 0.0247, 0.0254, 0.0214], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 21:57:08,527 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=143418.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 21:57:33,582 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8268, 3.8303, 3.5193, 1.7805, 3.3813, 3.4574, 3.5239, 3.2794], + device='cuda:3'), covar=tensor([0.1028, 0.0695, 0.1245, 0.4789, 0.1040, 0.1166, 0.1439, 0.1016], + device='cuda:3'), in_proj_covar=tensor([0.0515, 0.0428, 0.0427, 0.0527, 0.0419, 0.0425, 0.0410, 0.0375], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 21:57:35,795 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=143457.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:57:38,434 INFO [train.py:901] (3/4) Epoch 18, batch 6050, loss[loss=0.2196, simple_loss=0.3057, pruned_loss=0.06678, over 8534.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2921, pruned_loss=0.06575, over 1614331.68 frames. ], batch size: 31, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 21:57:47,832 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-06 21:57:48,575 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.525e+02 2.412e+02 3.060e+02 4.409e+02 1.030e+03, threshold=6.120e+02, percent-clipped=9.0 +2023-02-06 21:57:53,565 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=143482.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 21:57:59,612 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=143491.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 21:58:02,442 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8193, 1.3424, 1.6927, 1.2655, 0.9149, 1.3981, 1.6920, 1.6352], + device='cuda:3'), covar=tensor([0.0533, 0.1348, 0.1704, 0.1494, 0.0588, 0.1509, 0.0689, 0.0611], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0190, 0.0158, 0.0099, 0.0161, 0.0113, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 21:58:06,564 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9038, 2.0710, 1.8915, 2.5853, 1.1359, 1.6022, 1.8523, 2.0804], + device='cuda:3'), covar=tensor([0.0731, 0.0811, 0.0863, 0.0393, 0.1110, 0.1308, 0.0786, 0.0750], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0199, 0.0250, 0.0213, 0.0208, 0.0247, 0.0254, 0.0214], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 21:58:13,403 INFO [train.py:901] (3/4) Epoch 18, batch 6100, loss[loss=0.229, simple_loss=0.3109, pruned_loss=0.07354, over 8590.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2906, pruned_loss=0.06511, over 1607848.52 frames. ], batch size: 49, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 21:58:22,902 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6444, 4.6191, 4.1613, 2.2282, 4.1099, 4.1642, 4.1923, 3.9431], + device='cuda:3'), covar=tensor([0.0633, 0.0495, 0.0958, 0.4278, 0.0888, 0.1004, 0.1246, 0.0783], + device='cuda:3'), in_proj_covar=tensor([0.0512, 0.0426, 0.0425, 0.0524, 0.0418, 0.0423, 0.0408, 0.0374], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 21:58:39,433 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 21:58:40,456 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.20 vs. limit=5.0 +2023-02-06 21:58:49,839 INFO [train.py:901] (3/4) Epoch 18, batch 6150, loss[loss=0.2148, simple_loss=0.2997, pruned_loss=0.06496, over 8457.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2906, pruned_loss=0.06515, over 1610795.07 frames. ], batch size: 27, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 21:59:00,211 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.564e+02 2.359e+02 3.030e+02 3.820e+02 7.737e+02, threshold=6.061e+02, percent-clipped=3.0 +2023-02-06 21:59:25,630 INFO [train.py:901] (3/4) Epoch 18, batch 6200, loss[loss=0.2525, simple_loss=0.3095, pruned_loss=0.09777, over 8240.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2911, pruned_loss=0.06549, over 1615935.04 frames. ], batch size: 22, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 21:59:56,290 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8593, 2.1973, 1.7967, 2.7517, 1.9580, 1.8293, 2.3111, 2.4106], + device='cuda:3'), covar=tensor([0.1183, 0.0900, 0.1541, 0.0391, 0.0996, 0.1396, 0.0682, 0.0658], + device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0200, 0.0252, 0.0213, 0.0209, 0.0249, 0.0255, 0.0214], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 22:00:01,325 INFO [train.py:901] (3/4) Epoch 18, batch 6250, loss[loss=0.2071, simple_loss=0.2925, pruned_loss=0.06083, over 8686.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.2914, pruned_loss=0.06577, over 1612670.52 frames. ], batch size: 34, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 22:00:07,083 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9228, 2.3943, 1.9107, 2.3336, 2.1763, 1.7562, 2.0850, 2.2970], + device='cuda:3'), covar=tensor([0.1002, 0.0395, 0.1023, 0.0509, 0.0634, 0.1222, 0.0826, 0.0799], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0234, 0.0325, 0.0304, 0.0295, 0.0330, 0.0341, 0.0315], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 22:00:12,419 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.460e+02 3.089e+02 4.040e+02 1.017e+03, threshold=6.178e+02, percent-clipped=5.0 +2023-02-06 22:00:20,714 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4929, 1.7137, 1.8549, 1.2662, 1.9255, 1.3494, 0.5131, 1.7039], + device='cuda:3'), covar=tensor([0.0465, 0.0318, 0.0234, 0.0437, 0.0297, 0.0808, 0.0686, 0.0219], + device='cuda:3'), in_proj_covar=tensor([0.0436, 0.0372, 0.0322, 0.0433, 0.0362, 0.0523, 0.0380, 0.0398], + device='cuda:3'), out_proj_covar=tensor([1.1826e-04, 9.8424e-05, 8.5340e-05, 1.1531e-04, 9.6311e-05, 1.4985e-04, + 1.0310e-04, 1.0596e-04], device='cuda:3') +2023-02-06 22:00:37,036 INFO [train.py:901] (3/4) Epoch 18, batch 6300, loss[loss=0.1951, simple_loss=0.2771, pruned_loss=0.05649, over 7815.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2915, pruned_loss=0.06555, over 1612197.79 frames. ], batch size: 20, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 22:01:11,904 INFO [train.py:901] (3/4) Epoch 18, batch 6350, loss[loss=0.2105, simple_loss=0.2645, pruned_loss=0.0782, over 7724.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2905, pruned_loss=0.06533, over 1610428.50 frames. ], batch size: 18, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 22:01:13,342 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=143762.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 22:01:22,540 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.598e+02 2.204e+02 2.882e+02 3.589e+02 6.333e+02, threshold=5.763e+02, percent-clipped=1.0 +2023-02-06 22:01:47,363 INFO [train.py:901] (3/4) Epoch 18, batch 6400, loss[loss=0.1841, simple_loss=0.2677, pruned_loss=0.0503, over 8136.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2904, pruned_loss=0.065, over 1612526.11 frames. ], batch size: 22, lr: 4.19e-03, grad_scale: 8.0 +2023-02-06 22:01:49,681 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 +2023-02-06 22:01:55,546 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3924, 1.5980, 1.6181, 1.1039, 1.6401, 1.2757, 0.3194, 1.6246], + device='cuda:3'), covar=tensor([0.0406, 0.0297, 0.0297, 0.0416, 0.0359, 0.0834, 0.0731, 0.0221], + device='cuda:3'), in_proj_covar=tensor([0.0437, 0.0373, 0.0322, 0.0435, 0.0362, 0.0524, 0.0380, 0.0399], + device='cuda:3'), out_proj_covar=tensor([1.1853e-04, 9.8686e-05, 8.5205e-05, 1.1575e-04, 9.6397e-05, 1.5007e-04, + 1.0315e-04, 1.0630e-04], device='cuda:3') +2023-02-06 22:02:04,437 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=143835.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 22:02:20,759 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1598, 1.5234, 3.4009, 1.5094, 2.3034, 3.7997, 3.8881, 3.2563], + device='cuda:3'), covar=tensor([0.1052, 0.1880, 0.0364, 0.2219, 0.1253, 0.0216, 0.0437, 0.0547], + device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0313, 0.0277, 0.0306, 0.0295, 0.0254, 0.0399, 0.0297], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 22:02:21,899 INFO [train.py:901] (3/4) Epoch 18, batch 6450, loss[loss=0.1998, simple_loss=0.2815, pruned_loss=0.05907, over 7787.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2906, pruned_loss=0.06521, over 1613216.58 frames. ], batch size: 19, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:02:33,451 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.760e+02 2.452e+02 2.973e+02 3.704e+02 1.405e+03, threshold=5.946e+02, percent-clipped=1.0 +2023-02-06 22:02:34,284 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143877.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 22:02:39,644 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-06 22:02:57,286 INFO [train.py:901] (3/4) Epoch 18, batch 6500, loss[loss=0.2115, simple_loss=0.2718, pruned_loss=0.07559, over 7705.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2913, pruned_loss=0.06599, over 1611208.41 frames. ], batch size: 18, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:03:24,066 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143950.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 22:03:31,360 INFO [train.py:901] (3/4) Epoch 18, batch 6550, loss[loss=0.1867, simple_loss=0.2593, pruned_loss=0.05706, over 7426.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2927, pruned_loss=0.06684, over 1609791.81 frames. ], batch size: 17, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:03:38,659 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2621, 1.9408, 2.6944, 2.2022, 2.5946, 2.2269, 2.0036, 1.4081], + device='cuda:3'), covar=tensor([0.4851, 0.4585, 0.1626, 0.3363, 0.2297, 0.2761, 0.1850, 0.4904], + device='cuda:3'), in_proj_covar=tensor([0.0928, 0.0945, 0.0779, 0.0913, 0.0982, 0.0866, 0.0731, 0.0809], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 22:03:41,849 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.926e+02 2.526e+02 3.154e+02 3.765e+02 8.734e+02, threshold=6.308e+02, percent-clipped=5.0 +2023-02-06 22:03:48,834 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 22:04:08,861 INFO [train.py:901] (3/4) Epoch 18, batch 6600, loss[loss=0.1825, simple_loss=0.2663, pruned_loss=0.04938, over 7667.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2922, pruned_loss=0.06619, over 1606928.52 frames. ], batch size: 19, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:04:10,898 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 22:04:25,261 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-06 22:04:34,007 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.75 vs. limit=5.0 +2023-02-06 22:04:43,666 INFO [train.py:901] (3/4) Epoch 18, batch 6650, loss[loss=0.2294, simple_loss=0.3053, pruned_loss=0.0768, over 8196.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2925, pruned_loss=0.06605, over 1610402.19 frames. ], batch size: 23, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:04:54,721 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.649e+02 2.298e+02 3.022e+02 3.555e+02 7.360e+02, threshold=6.043e+02, percent-clipped=4.0 +2023-02-06 22:05:19,670 INFO [train.py:901] (3/4) Epoch 18, batch 6700, loss[loss=0.2145, simple_loss=0.3038, pruned_loss=0.06259, over 8511.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2919, pruned_loss=0.06573, over 1607755.48 frames. ], batch size: 28, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:05:34,398 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=144133.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 22:05:51,944 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=144158.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 22:05:53,841 INFO [train.py:901] (3/4) Epoch 18, batch 6750, loss[loss=0.1764, simple_loss=0.2558, pruned_loss=0.04853, over 7552.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2927, pruned_loss=0.06599, over 1613084.50 frames. ], batch size: 18, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:06:03,939 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.630e+02 2.293e+02 3.003e+02 3.717e+02 7.578e+02, threshold=6.007e+02, percent-clipped=1.0 +2023-02-06 22:06:25,435 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=144206.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 22:06:27,268 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 22:06:28,567 INFO [train.py:901] (3/4) Epoch 18, batch 6800, loss[loss=0.1879, simple_loss=0.263, pruned_loss=0.05642, over 7704.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.2919, pruned_loss=0.06564, over 1613504.69 frames. ], batch size: 18, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:06:41,515 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4432, 1.7106, 2.6692, 1.2892, 1.8131, 1.7772, 1.5059, 1.8609], + device='cuda:3'), covar=tensor([0.1999, 0.2517, 0.0848, 0.4400, 0.2070, 0.3159, 0.2307, 0.2342], + device='cuda:3'), in_proj_covar=tensor([0.0514, 0.0580, 0.0551, 0.0625, 0.0641, 0.0586, 0.0517, 0.0633], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 22:06:42,847 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=144231.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 22:07:03,973 INFO [train.py:901] (3/4) Epoch 18, batch 6850, loss[loss=0.2369, simple_loss=0.3097, pruned_loss=0.08203, over 8454.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2921, pruned_loss=0.06569, over 1615932.43 frames. ], batch size: 27, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:07:13,994 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.379e+02 2.937e+02 3.634e+02 6.722e+02, threshold=5.873e+02, percent-clipped=2.0 +2023-02-06 22:07:17,332 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 22:07:38,062 INFO [train.py:901] (3/4) Epoch 18, batch 6900, loss[loss=0.2871, simple_loss=0.3526, pruned_loss=0.1108, over 6692.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2925, pruned_loss=0.06563, over 1614839.12 frames. ], batch size: 71, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:08:13,483 INFO [train.py:901] (3/4) Epoch 18, batch 6950, loss[loss=0.2384, simple_loss=0.3166, pruned_loss=0.0801, over 8323.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2922, pruned_loss=0.06595, over 1610382.18 frames. ], batch size: 25, lr: 4.18e-03, grad_scale: 8.0 +2023-02-06 22:08:24,083 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 2.398e+02 2.919e+02 3.864e+02 7.610e+02, threshold=5.839e+02, percent-clipped=3.0 +2023-02-06 22:08:25,464 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 22:08:47,781 INFO [train.py:901] (3/4) Epoch 18, batch 7000, loss[loss=0.1795, simple_loss=0.2562, pruned_loss=0.05139, over 7939.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2924, pruned_loss=0.06624, over 1605958.43 frames. ], batch size: 20, lr: 4.18e-03, grad_scale: 16.0 +2023-02-06 22:08:49,931 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.7882, 1.5069, 3.9771, 1.5666, 3.5182, 3.2446, 3.6244, 3.5020], + device='cuda:3'), covar=tensor([0.0696, 0.3943, 0.0606, 0.3622, 0.1178, 0.1008, 0.0630, 0.0757], + device='cuda:3'), in_proj_covar=tensor([0.0596, 0.0632, 0.0676, 0.0603, 0.0685, 0.0584, 0.0582, 0.0652], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 22:09:01,097 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=144429.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:09:05,741 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7110, 1.3394, 2.8403, 1.4055, 2.0292, 3.0354, 3.2238, 2.5899], + device='cuda:3'), covar=tensor([0.1111, 0.1758, 0.0385, 0.2105, 0.0952, 0.0289, 0.0570, 0.0625], + device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0314, 0.0278, 0.0307, 0.0295, 0.0255, 0.0400, 0.0298], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 22:09:22,578 INFO [train.py:901] (3/4) Epoch 18, batch 7050, loss[loss=0.242, simple_loss=0.3167, pruned_loss=0.0837, over 8643.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2927, pruned_loss=0.06686, over 1603041.08 frames. ], batch size: 34, lr: 4.18e-03, grad_scale: 16.0 +2023-02-06 22:09:34,230 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.214e+02 2.534e+02 2.937e+02 3.689e+02 8.247e+02, threshold=5.874e+02, percent-clipped=3.0 +2023-02-06 22:09:58,439 INFO [train.py:901] (3/4) Epoch 18, batch 7100, loss[loss=0.2064, simple_loss=0.2892, pruned_loss=0.06182, over 8107.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2936, pruned_loss=0.06714, over 1604518.96 frames. ], batch size: 23, lr: 4.18e-03, grad_scale: 16.0 +2023-02-06 22:10:33,587 INFO [train.py:901] (3/4) Epoch 18, batch 7150, loss[loss=0.1981, simple_loss=0.274, pruned_loss=0.06113, over 7804.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2933, pruned_loss=0.06686, over 1604497.22 frames. ], batch size: 19, lr: 4.17e-03, grad_scale: 16.0 +2023-02-06 22:10:43,975 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.496e+02 2.263e+02 2.906e+02 3.662e+02 1.305e+03, threshold=5.813e+02, percent-clipped=7.0 +2023-02-06 22:11:00,102 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.28 vs. limit=5.0 +2023-02-06 22:11:10,032 INFO [train.py:901] (3/4) Epoch 18, batch 7200, loss[loss=0.2209, simple_loss=0.3124, pruned_loss=0.06463, over 8498.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2932, pruned_loss=0.06666, over 1607566.23 frames. ], batch size: 26, lr: 4.17e-03, grad_scale: 16.0 +2023-02-06 22:11:12,568 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-06 22:11:20,760 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8449, 1.5399, 1.7402, 1.3869, 1.0665, 1.5492, 1.7458, 1.4154], + device='cuda:3'), covar=tensor([0.0555, 0.1214, 0.1588, 0.1384, 0.0586, 0.1420, 0.0684, 0.0667], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0151, 0.0190, 0.0157, 0.0099, 0.0161, 0.0113, 0.0139], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 22:11:44,457 INFO [train.py:901] (3/4) Epoch 18, batch 7250, loss[loss=0.1895, simple_loss=0.2503, pruned_loss=0.06439, over 7433.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2927, pruned_loss=0.06639, over 1606149.18 frames. ], batch size: 17, lr: 4.17e-03, grad_scale: 16.0 +2023-02-06 22:11:54,472 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.514e+02 2.379e+02 2.816e+02 3.627e+02 9.857e+02, threshold=5.632e+02, percent-clipped=4.0 +2023-02-06 22:12:19,766 INFO [train.py:901] (3/4) Epoch 18, batch 7300, loss[loss=0.1723, simple_loss=0.2496, pruned_loss=0.04752, over 7442.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2936, pruned_loss=0.06668, over 1610554.60 frames. ], batch size: 17, lr: 4.17e-03, grad_scale: 16.0 +2023-02-06 22:12:50,210 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4622, 2.5666, 1.7689, 2.3449, 2.0009, 1.4354, 2.0315, 2.0531], + device='cuda:3'), covar=tensor([0.1503, 0.0350, 0.1281, 0.0507, 0.0758, 0.1598, 0.0984, 0.0834], + device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0233, 0.0324, 0.0301, 0.0294, 0.0328, 0.0340, 0.0314], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 22:12:53,982 INFO [train.py:901] (3/4) Epoch 18, batch 7350, loss[loss=0.2056, simple_loss=0.3029, pruned_loss=0.05412, over 8195.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2927, pruned_loss=0.06627, over 1612618.41 frames. ], batch size: 23, lr: 4.17e-03, grad_scale: 16.0 +2023-02-06 22:13:02,927 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=144773.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:13:04,753 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.558e+02 2.484e+02 2.992e+02 3.514e+02 8.978e+02, threshold=5.985e+02, percent-clipped=6.0 +2023-02-06 22:13:08,084 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 22:13:26,895 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 22:13:28,903 INFO [train.py:901] (3/4) Epoch 18, batch 7400, loss[loss=0.1843, simple_loss=0.2531, pruned_loss=0.05778, over 7715.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2929, pruned_loss=0.06579, over 1614646.14 frames. ], batch size: 18, lr: 4.17e-03, grad_scale: 16.0 +2023-02-06 22:14:04,309 INFO [train.py:901] (3/4) Epoch 18, batch 7450, loss[loss=0.1648, simple_loss=0.2499, pruned_loss=0.03989, over 7660.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2942, pruned_loss=0.06641, over 1619195.41 frames. ], batch size: 19, lr: 4.17e-03, grad_scale: 16.0 +2023-02-06 22:14:07,796 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 22:14:07,992 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2286, 2.1732, 1.5901, 1.9522, 1.7977, 1.3284, 1.7028, 1.5497], + device='cuda:3'), covar=tensor([0.1298, 0.0369, 0.1094, 0.0472, 0.0594, 0.1405, 0.0858, 0.0835], + device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0233, 0.0324, 0.0302, 0.0295, 0.0329, 0.0342, 0.0314], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 22:14:14,578 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.437e+02 2.433e+02 3.083e+02 4.140e+02 9.921e+02, threshold=6.167e+02, percent-clipped=3.0 +2023-02-06 22:14:23,403 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=144888.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:14:38,536 INFO [train.py:901] (3/4) Epoch 18, batch 7500, loss[loss=0.1984, simple_loss=0.2949, pruned_loss=0.05092, over 8710.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2943, pruned_loss=0.06619, over 1620002.96 frames. ], batch size: 34, lr: 4.17e-03, grad_scale: 8.0 +2023-02-06 22:15:14,116 INFO [train.py:901] (3/4) Epoch 18, batch 7550, loss[loss=0.2145, simple_loss=0.2886, pruned_loss=0.07026, over 7932.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2941, pruned_loss=0.06618, over 1619682.51 frames. ], batch size: 20, lr: 4.17e-03, grad_scale: 8.0 +2023-02-06 22:15:24,760 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.430e+02 2.361e+02 2.893e+02 3.293e+02 8.578e+02, threshold=5.785e+02, percent-clipped=2.0 +2023-02-06 22:15:48,835 INFO [train.py:901] (3/4) Epoch 18, batch 7600, loss[loss=0.1662, simple_loss=0.2466, pruned_loss=0.04288, over 7656.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2944, pruned_loss=0.06603, over 1619649.96 frames. ], batch size: 19, lr: 4.17e-03, grad_scale: 8.0 +2023-02-06 22:15:51,056 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145014.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:16:12,837 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145045.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:16:24,429 INFO [train.py:901] (3/4) Epoch 18, batch 7650, loss[loss=0.1893, simple_loss=0.2759, pruned_loss=0.05137, over 7194.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2933, pruned_loss=0.06575, over 1617136.09 frames. ], batch size: 16, lr: 4.17e-03, grad_scale: 8.0 +2023-02-06 22:16:35,693 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.643e+02 2.409e+02 3.204e+02 3.806e+02 7.453e+02, threshold=6.408e+02, percent-clipped=5.0 +2023-02-06 22:16:40,604 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145084.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:16:57,519 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6101, 1.8592, 2.0250, 1.2311, 2.1007, 1.3965, 0.7057, 1.8114], + device='cuda:3'), covar=tensor([0.0771, 0.0446, 0.0417, 0.0756, 0.0499, 0.1048, 0.1014, 0.0362], + device='cuda:3'), in_proj_covar=tensor([0.0440, 0.0379, 0.0325, 0.0433, 0.0364, 0.0525, 0.0381, 0.0404], + device='cuda:3'), out_proj_covar=tensor([1.1951e-04, 1.0020e-04, 8.5967e-05, 1.1516e-04, 9.6754e-05, 1.5029e-04, + 1.0339e-04, 1.0806e-04], device='cuda:3') +2023-02-06 22:16:58,566 INFO [train.py:901] (3/4) Epoch 18, batch 7700, loss[loss=0.2211, simple_loss=0.2849, pruned_loss=0.07867, over 7668.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.293, pruned_loss=0.06629, over 1613850.30 frames. ], batch size: 19, lr: 4.17e-03, grad_scale: 8.0 +2023-02-06 22:17:16,304 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 22:17:17,791 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1405, 1.2312, 1.2382, 0.9323, 1.2655, 1.0308, 0.3743, 1.1838], + device='cuda:3'), covar=tensor([0.0335, 0.0273, 0.0226, 0.0340, 0.0310, 0.0567, 0.0597, 0.0218], + device='cuda:3'), in_proj_covar=tensor([0.0438, 0.0377, 0.0323, 0.0431, 0.0362, 0.0523, 0.0379, 0.0403], + device='cuda:3'), out_proj_covar=tensor([1.1892e-04, 9.9597e-05, 8.5574e-05, 1.1453e-04, 9.6450e-05, 1.4958e-04, + 1.0290e-04, 1.0766e-04], device='cuda:3') +2023-02-06 22:17:21,903 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145144.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:17:33,670 INFO [train.py:901] (3/4) Epoch 18, batch 7750, loss[loss=0.2895, simple_loss=0.3512, pruned_loss=0.114, over 8294.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2943, pruned_loss=0.06727, over 1610405.02 frames. ], batch size: 23, lr: 4.17e-03, grad_scale: 8.0 +2023-02-06 22:17:40,052 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145169.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:17:45,202 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.614e+02 2.617e+02 3.101e+02 3.765e+02 9.296e+02, threshold=6.202e+02, percent-clipped=3.0 +2023-02-06 22:18:08,802 INFO [train.py:901] (3/4) Epoch 18, batch 7800, loss[loss=0.2065, simple_loss=0.3028, pruned_loss=0.05513, over 8289.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2941, pruned_loss=0.06697, over 1608694.94 frames. ], batch size: 23, lr: 4.17e-03, grad_scale: 8.0 +2023-02-06 22:18:42,838 INFO [train.py:901] (3/4) Epoch 18, batch 7850, loss[loss=0.1984, simple_loss=0.2813, pruned_loss=0.05774, over 8256.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2929, pruned_loss=0.06643, over 1609611.20 frames. ], batch size: 24, lr: 4.16e-03, grad_scale: 8.0 +2023-02-06 22:18:53,262 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.506e+02 2.477e+02 2.948e+02 3.643e+02 1.044e+03, threshold=5.895e+02, percent-clipped=9.0 +2023-02-06 22:19:16,116 INFO [train.py:901] (3/4) Epoch 18, batch 7900, loss[loss=0.1769, simple_loss=0.252, pruned_loss=0.05088, over 7409.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.2932, pruned_loss=0.06651, over 1610214.09 frames. ], batch size: 17, lr: 4.16e-03, grad_scale: 8.0 +2023-02-06 22:19:47,152 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=145358.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:19:49,082 INFO [train.py:901] (3/4) Epoch 18, batch 7950, loss[loss=0.2181, simple_loss=0.3002, pruned_loss=0.06804, over 8202.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2932, pruned_loss=0.0659, over 1614938.29 frames. ], batch size: 23, lr: 4.16e-03, grad_scale: 8.0 +2023-02-06 22:19:59,831 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.564e+02 2.389e+02 3.012e+02 3.869e+02 1.111e+03, threshold=6.025e+02, percent-clipped=3.0 +2023-02-06 22:20:04,931 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.12 vs. limit=5.0 +2023-02-06 22:20:07,942 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=145389.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:20:08,012 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145389.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:20:23,113 INFO [train.py:901] (3/4) Epoch 18, batch 8000, loss[loss=0.2332, simple_loss=0.3164, pruned_loss=0.07499, over 8469.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2929, pruned_loss=0.06578, over 1616857.62 frames. ], batch size: 27, lr: 4.16e-03, grad_scale: 8.0 +2023-02-06 22:20:25,589 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 +2023-02-06 22:20:31,461 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4496, 1.6331, 2.1508, 1.3343, 1.4948, 1.7080, 1.4948, 1.4456], + device='cuda:3'), covar=tensor([0.1765, 0.2268, 0.0977, 0.4150, 0.1852, 0.3147, 0.2122, 0.1979], + device='cuda:3'), in_proj_covar=tensor([0.0515, 0.0578, 0.0548, 0.0625, 0.0638, 0.0583, 0.0515, 0.0628], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 22:20:34,672 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=145428.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:20:57,109 INFO [train.py:901] (3/4) Epoch 18, batch 8050, loss[loss=0.1799, simple_loss=0.2564, pruned_loss=0.05177, over 7227.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2916, pruned_loss=0.06615, over 1590907.86 frames. ], batch size: 16, lr: 4.16e-03, grad_scale: 8.0 +2023-02-06 22:21:05,665 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145473.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:21:08,161 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.710e+02 2.350e+02 2.866e+02 3.408e+02 5.747e+02, threshold=5.732e+02, percent-clipped=0.0 +2023-02-06 22:21:29,237 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 22:21:34,893 INFO [train.py:901] (3/4) Epoch 19, batch 0, loss[loss=0.244, simple_loss=0.3149, pruned_loss=0.08652, over 8094.00 frames. ], tot_loss[loss=0.244, simple_loss=0.3149, pruned_loss=0.08652, over 8094.00 frames. ], batch size: 23, lr: 4.05e-03, grad_scale: 8.0 +2023-02-06 22:21:34,894 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 22:21:46,548 INFO [train.py:935] (3/4) Epoch 19, validation: loss=0.1782, simple_loss=0.2779, pruned_loss=0.03928, over 944034.00 frames. +2023-02-06 22:21:46,549 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 22:21:54,199 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145504.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:22:03,060 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 22:22:22,466 INFO [train.py:901] (3/4) Epoch 19, batch 50, loss[loss=0.1729, simple_loss=0.2506, pruned_loss=0.04758, over 7437.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2906, pruned_loss=0.06452, over 365881.72 frames. ], batch size: 17, lr: 4.05e-03, grad_scale: 8.0 +2023-02-06 22:22:22,665 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145543.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:22:23,353 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9624, 1.6352, 3.5683, 1.5994, 2.4965, 3.8896, 4.0418, 3.2806], + device='cuda:3'), covar=tensor([0.1178, 0.1704, 0.0322, 0.1954, 0.1017, 0.0227, 0.0548, 0.0549], + device='cuda:3'), in_proj_covar=tensor([0.0281, 0.0312, 0.0275, 0.0303, 0.0293, 0.0253, 0.0396, 0.0296], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 22:22:25,796 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 22:22:40,517 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 22:22:42,317 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 +2023-02-06 22:22:45,196 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.562e+02 2.340e+02 2.977e+02 3.641e+02 7.952e+02, threshold=5.953e+02, percent-clipped=6.0 +2023-02-06 22:22:56,255 INFO [train.py:901] (3/4) Epoch 19, batch 100, loss[loss=0.2519, simple_loss=0.3172, pruned_loss=0.09335, over 7190.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2949, pruned_loss=0.06675, over 644152.78 frames. ], batch size: 71, lr: 4.05e-03, grad_scale: 8.0 +2023-02-06 22:22:59,350 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6888, 2.2384, 4.0837, 1.5684, 3.0325, 2.1772, 1.7843, 2.8595], + device='cuda:3'), covar=tensor([0.1827, 0.2572, 0.0750, 0.4519, 0.1734, 0.3180, 0.2190, 0.2400], + device='cuda:3'), in_proj_covar=tensor([0.0515, 0.0578, 0.0548, 0.0628, 0.0635, 0.0585, 0.0517, 0.0629], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 22:23:01,906 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 22:23:10,096 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145612.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:23:20,686 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.7143, 4.6867, 4.1813, 2.1754, 4.1330, 4.3431, 4.2578, 4.0546], + device='cuda:3'), covar=tensor([0.0823, 0.0504, 0.0949, 0.4601, 0.0820, 0.0932, 0.1165, 0.0786], + device='cuda:3'), in_proj_covar=tensor([0.0513, 0.0426, 0.0429, 0.0527, 0.0414, 0.0426, 0.0408, 0.0373], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 22:23:32,314 INFO [train.py:901] (3/4) Epoch 19, batch 150, loss[loss=0.2341, simple_loss=0.3166, pruned_loss=0.0758, over 8518.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2934, pruned_loss=0.06613, over 859786.24 frames. ], batch size: 39, lr: 4.05e-03, grad_scale: 8.0 +2023-02-06 22:23:40,877 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2843, 1.9879, 2.7535, 2.2431, 2.6228, 2.2465, 1.9931, 1.4403], + device='cuda:3'), covar=tensor([0.5175, 0.4912, 0.1710, 0.3468, 0.2318, 0.2919, 0.2021, 0.5114], + device='cuda:3'), in_proj_covar=tensor([0.0925, 0.0945, 0.0773, 0.0911, 0.0976, 0.0864, 0.0728, 0.0808], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 22:23:46,187 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145661.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:23:57,046 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.454e+02 2.969e+02 3.777e+02 1.176e+03, threshold=5.938e+02, percent-clipped=4.0 +2023-02-06 22:24:07,976 INFO [train.py:901] (3/4) Epoch 19, batch 200, loss[loss=0.2443, simple_loss=0.3178, pruned_loss=0.08538, over 8290.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2936, pruned_loss=0.06582, over 1027557.08 frames. ], batch size: 23, lr: 4.05e-03, grad_scale: 8.0 +2023-02-06 22:24:33,093 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8301, 3.6354, 2.1712, 2.6877, 2.6134, 1.9447, 2.6852, 2.8790], + device='cuda:3'), covar=tensor([0.1778, 0.0322, 0.1137, 0.0735, 0.0771, 0.1356, 0.1078, 0.1096], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0234, 0.0323, 0.0301, 0.0296, 0.0328, 0.0339, 0.0313], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 22:24:33,108 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145729.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:24:35,731 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=145733.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:24:43,207 INFO [train.py:901] (3/4) Epoch 19, batch 250, loss[loss=0.2205, simple_loss=0.315, pruned_loss=0.06298, over 8476.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.293, pruned_loss=0.0654, over 1160192.41 frames. ], batch size: 25, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:24:51,130 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145754.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:24:55,236 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145760.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:24:58,398 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-06 22:25:06,970 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-06 22:25:07,541 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.430e+02 2.432e+02 3.022e+02 3.893e+02 7.688e+02, threshold=6.043e+02, percent-clipped=6.0 +2023-02-06 22:25:13,307 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145785.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:25:18,656 INFO [train.py:901] (3/4) Epoch 19, batch 300, loss[loss=0.1872, simple_loss=0.2658, pruned_loss=0.05429, over 7657.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2926, pruned_loss=0.06565, over 1256809.30 frames. ], batch size: 19, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:25:22,945 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145799.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:25:39,974 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145824.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:25:51,137 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145839.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 22:25:53,701 INFO [train.py:901] (3/4) Epoch 19, batch 350, loss[loss=0.1824, simple_loss=0.2571, pruned_loss=0.05385, over 7237.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2926, pruned_loss=0.06566, over 1332178.87 frames. ], batch size: 16, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:25:57,438 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145848.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:26:17,676 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.604e+02 2.383e+02 2.952e+02 3.795e+02 9.100e+02, threshold=5.904e+02, percent-clipped=6.0 +2023-02-06 22:26:30,044 INFO [train.py:901] (3/4) Epoch 19, batch 400, loss[loss=0.1669, simple_loss=0.2462, pruned_loss=0.04378, over 7265.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2919, pruned_loss=0.0653, over 1391886.34 frames. ], batch size: 16, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:26:32,903 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4785, 4.4931, 4.0582, 2.0804, 3.9984, 4.0311, 4.0399, 3.8046], + device='cuda:3'), covar=tensor([0.0764, 0.0554, 0.1121, 0.4451, 0.0870, 0.1030, 0.1219, 0.0855], + device='cuda:3'), in_proj_covar=tensor([0.0511, 0.0422, 0.0425, 0.0524, 0.0411, 0.0424, 0.0402, 0.0370], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 22:26:44,749 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8481, 1.6002, 1.7718, 1.5984, 1.0511, 1.5789, 2.1593, 2.0253], + device='cuda:3'), covar=tensor([0.0462, 0.1298, 0.1670, 0.1380, 0.0598, 0.1534, 0.0663, 0.0592], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0152, 0.0189, 0.0157, 0.0099, 0.0161, 0.0112, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 22:27:04,033 INFO [train.py:901] (3/4) Epoch 19, batch 450, loss[loss=0.1922, simple_loss=0.2689, pruned_loss=0.05774, over 7655.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.293, pruned_loss=0.06573, over 1443996.00 frames. ], batch size: 19, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:27:12,922 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=145956.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:27:23,355 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.02 vs. limit=5.0 +2023-02-06 22:27:28,517 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.474e+02 2.839e+02 3.457e+02 5.406e+02, threshold=5.677e+02, percent-clipped=0.0 +2023-02-06 22:27:40,188 INFO [train.py:901] (3/4) Epoch 19, batch 500, loss[loss=0.2249, simple_loss=0.3088, pruned_loss=0.07056, over 8335.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2922, pruned_loss=0.06515, over 1482659.05 frames. ], batch size: 25, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:27:50,097 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=146005.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:28:03,432 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5306, 1.4883, 4.7346, 1.7133, 4.1607, 3.9086, 4.2830, 4.1205], + device='cuda:3'), covar=tensor([0.0549, 0.4619, 0.0453, 0.4250, 0.1094, 0.0929, 0.0561, 0.0662], + device='cuda:3'), in_proj_covar=tensor([0.0597, 0.0627, 0.0673, 0.0605, 0.0687, 0.0591, 0.0585, 0.0649], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 22:28:15,840 INFO [train.py:901] (3/4) Epoch 19, batch 550, loss[loss=0.194, simple_loss=0.284, pruned_loss=0.05204, over 7984.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2931, pruned_loss=0.06567, over 1515477.92 frames. ], batch size: 21, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:28:19,298 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8710, 3.8067, 3.5169, 1.7261, 3.4315, 3.4617, 3.4560, 3.1911], + device='cuda:3'), covar=tensor([0.0974, 0.0701, 0.1196, 0.4780, 0.1018, 0.1082, 0.1513, 0.0938], + device='cuda:3'), in_proj_covar=tensor([0.0519, 0.0428, 0.0430, 0.0530, 0.0416, 0.0430, 0.0411, 0.0375], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 22:28:35,122 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146071.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:28:38,933 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.555e+02 2.557e+02 3.049e+02 4.000e+02 8.642e+02, threshold=6.099e+02, percent-clipped=4.0 +2023-02-06 22:28:50,772 INFO [train.py:901] (3/4) Epoch 19, batch 600, loss[loss=0.2025, simple_loss=0.2659, pruned_loss=0.06954, over 7423.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2927, pruned_loss=0.06576, over 1539898.90 frames. ], batch size: 17, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:28:54,564 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8010, 1.4121, 4.0110, 1.3354, 3.5468, 3.3350, 3.6546, 3.5170], + device='cuda:3'), covar=tensor([0.0651, 0.4172, 0.0603, 0.3921, 0.1276, 0.1037, 0.0598, 0.0754], + device='cuda:3'), in_proj_covar=tensor([0.0594, 0.0624, 0.0670, 0.0600, 0.0684, 0.0588, 0.0582, 0.0646], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 22:28:59,452 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146104.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:29:03,473 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.8072, 4.7968, 4.3500, 2.3859, 4.2896, 4.3196, 4.4180, 4.0022], + device='cuda:3'), covar=tensor([0.0626, 0.0468, 0.0923, 0.3826, 0.0740, 0.0762, 0.1097, 0.0711], + device='cuda:3'), in_proj_covar=tensor([0.0517, 0.0425, 0.0427, 0.0526, 0.0413, 0.0427, 0.0407, 0.0373], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 22:29:11,441 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-06 22:29:11,618 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146120.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:29:17,651 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146129.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:29:26,673 INFO [train.py:901] (3/4) Epoch 19, batch 650, loss[loss=0.1869, simple_loss=0.2674, pruned_loss=0.05316, over 7805.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2904, pruned_loss=0.06498, over 1552080.25 frames. ], batch size: 20, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:29:42,925 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146167.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:29:49,766 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.807e+02 2.628e+02 2.995e+02 3.912e+02 8.872e+02, threshold=5.991e+02, percent-clipped=7.0 +2023-02-06 22:29:53,935 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=146183.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 22:30:00,627 INFO [train.py:901] (3/4) Epoch 19, batch 700, loss[loss=0.2086, simple_loss=0.2956, pruned_loss=0.06082, over 8682.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2915, pruned_loss=0.06524, over 1569826.47 frames. ], batch size: 34, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:30:37,726 INFO [train.py:901] (3/4) Epoch 19, batch 750, loss[loss=0.192, simple_loss=0.2838, pruned_loss=0.05012, over 8456.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2906, pruned_loss=0.06472, over 1580215.47 frames. ], batch size: 27, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:30:58,059 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-06 22:31:00,738 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.543e+02 2.187e+02 2.733e+02 3.387e+02 1.037e+03, threshold=5.466e+02, percent-clipped=4.0 +2023-02-06 22:31:06,864 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-06 22:31:11,497 INFO [train.py:901] (3/4) Epoch 19, batch 800, loss[loss=0.2447, simple_loss=0.3123, pruned_loss=0.08851, over 7331.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2914, pruned_loss=0.06498, over 1588505.02 frames. ], batch size: 71, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:31:14,845 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146298.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 22:31:15,544 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1837, 1.2098, 1.6307, 1.1091, 0.7427, 1.3278, 1.2406, 1.0451], + device='cuda:3'), covar=tensor([0.0564, 0.1175, 0.1556, 0.1435, 0.0522, 0.1432, 0.0657, 0.0706], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0153, 0.0190, 0.0158, 0.0099, 0.0162, 0.0113, 0.0141], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 22:31:17,173 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.79 vs. limit=5.0 +2023-02-06 22:31:35,767 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146327.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:31:47,226 INFO [train.py:901] (3/4) Epoch 19, batch 850, loss[loss=0.2151, simple_loss=0.2994, pruned_loss=0.06536, over 7941.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2925, pruned_loss=0.06571, over 1594632.87 frames. ], batch size: 20, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:31:54,229 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146352.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:32:10,850 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146376.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:32:11,293 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.640e+02 2.470e+02 3.071e+02 3.941e+02 1.675e+03, threshold=6.141e+02, percent-clipped=6.0 +2023-02-06 22:32:22,253 INFO [train.py:901] (3/4) Epoch 19, batch 900, loss[loss=0.2146, simple_loss=0.2969, pruned_loss=0.06612, over 8527.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2931, pruned_loss=0.06598, over 1597670.61 frames. ], batch size: 28, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:32:27,825 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146401.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:32:41,220 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7796, 2.1955, 4.0524, 1.5989, 2.9715, 2.3416, 1.8015, 2.7749], + device='cuda:3'), covar=tensor([0.1904, 0.2787, 0.0893, 0.4678, 0.1857, 0.3048, 0.2326, 0.2533], + device='cuda:3'), in_proj_covar=tensor([0.0516, 0.0580, 0.0550, 0.0630, 0.0639, 0.0584, 0.0518, 0.0629], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 22:32:56,373 INFO [train.py:901] (3/4) Epoch 19, batch 950, loss[loss=0.215, simple_loss=0.2834, pruned_loss=0.07327, over 7931.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2937, pruned_loss=0.06599, over 1602730.46 frames. ], batch size: 20, lr: 4.04e-03, grad_scale: 8.0 +2023-02-06 22:33:09,723 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.54 vs. limit=5.0 +2023-02-06 22:33:20,835 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.0297, 1.5541, 1.9206, 1.4919, 1.0539, 1.5143, 2.0479, 2.2620], + device='cuda:3'), covar=tensor([0.0426, 0.1298, 0.1592, 0.1474, 0.0622, 0.1616, 0.0638, 0.0549], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0151, 0.0188, 0.0157, 0.0099, 0.0162, 0.0112, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 22:33:21,314 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.322e+02 2.324e+02 2.987e+02 4.077e+02 9.877e+02, threshold=5.974e+02, percent-clipped=4.0 +2023-02-06 22:33:22,702 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-06 22:33:32,162 INFO [train.py:901] (3/4) Epoch 19, batch 1000, loss[loss=0.1794, simple_loss=0.2614, pruned_loss=0.04874, over 7924.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2918, pruned_loss=0.06462, over 1605038.07 frames. ], batch size: 20, lr: 4.03e-03, grad_scale: 8.0 +2023-02-06 22:33:44,468 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=146511.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:33:54,581 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-06 22:34:00,614 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2023-02-06 22:34:06,370 INFO [train.py:901] (3/4) Epoch 19, batch 1050, loss[loss=0.2237, simple_loss=0.2999, pruned_loss=0.07376, over 8133.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2919, pruned_loss=0.06494, over 1604278.41 frames. ], batch size: 22, lr: 4.03e-03, grad_scale: 8.0 +2023-02-06 22:34:07,093 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-06 22:34:14,961 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146554.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 22:34:31,572 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.392e+02 2.403e+02 2.837e+02 3.508e+02 6.242e+02, threshold=5.674e+02, percent-clipped=1.0 +2023-02-06 22:34:34,618 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146579.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 22:34:44,087 INFO [train.py:901] (3/4) Epoch 19, batch 1100, loss[loss=0.2179, simple_loss=0.3002, pruned_loss=0.06782, over 8475.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2908, pruned_loss=0.06433, over 1606859.40 frames. ], batch size: 29, lr: 4.03e-03, grad_scale: 8.0 +2023-02-06 22:34:55,193 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146609.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:35:06,979 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146626.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:35:18,374 INFO [train.py:901] (3/4) Epoch 19, batch 1150, loss[loss=0.1817, simple_loss=0.2639, pruned_loss=0.04975, over 7919.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2921, pruned_loss=0.06548, over 1606230.17 frames. ], batch size: 20, lr: 4.03e-03, grad_scale: 8.0 +2023-02-06 22:35:19,117 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-06 22:35:19,269 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146644.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:35:42,411 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.477e+02 2.484e+02 2.879e+02 3.755e+02 5.922e+02, threshold=5.758e+02, percent-clipped=3.0 +2023-02-06 22:35:53,854 INFO [train.py:901] (3/4) Epoch 19, batch 1200, loss[loss=0.2106, simple_loss=0.3029, pruned_loss=0.05918, over 8247.00 frames. ], tot_loss[loss=0.213, simple_loss=0.2933, pruned_loss=0.06631, over 1611935.36 frames. ], batch size: 24, lr: 4.03e-03, grad_scale: 8.0 +2023-02-06 22:36:28,994 INFO [train.py:901] (3/4) Epoch 19, batch 1250, loss[loss=0.2095, simple_loss=0.2967, pruned_loss=0.06117, over 8289.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2934, pruned_loss=0.06663, over 1609242.52 frames. ], batch size: 23, lr: 4.03e-03, grad_scale: 8.0 +2023-02-06 22:36:37,770 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 +2023-02-06 22:36:52,651 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.364e+02 2.471e+02 2.976e+02 4.092e+02 7.603e+02, threshold=5.951e+02, percent-clipped=4.0 +2023-02-06 22:37:04,310 INFO [train.py:901] (3/4) Epoch 19, batch 1300, loss[loss=0.2057, simple_loss=0.2755, pruned_loss=0.06793, over 8248.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2938, pruned_loss=0.06631, over 1617078.24 frames. ], batch size: 22, lr: 4.03e-03, grad_scale: 8.0 +2023-02-06 22:37:40,718 INFO [train.py:901] (3/4) Epoch 19, batch 1350, loss[loss=0.2384, simple_loss=0.316, pruned_loss=0.08042, over 8668.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2947, pruned_loss=0.0667, over 1619874.81 frames. ], batch size: 34, lr: 4.03e-03, grad_scale: 8.0 +2023-02-06 22:37:53,742 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146862.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:37:58,968 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 +2023-02-06 22:38:03,895 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.521e+02 2.302e+02 2.844e+02 3.659e+02 6.626e+02, threshold=5.688e+02, percent-clipped=1.0 +2023-02-06 22:38:07,784 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146882.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:38:15,240 INFO [train.py:901] (3/4) Epoch 19, batch 1400, loss[loss=0.1789, simple_loss=0.2504, pruned_loss=0.05367, over 7232.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2927, pruned_loss=0.06549, over 1617376.85 frames. ], batch size: 16, lr: 4.03e-03, grad_scale: 8.0 +2023-02-06 22:38:25,963 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146907.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:38:43,246 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 22:38:52,626 INFO [train.py:901] (3/4) Epoch 19, batch 1450, loss[loss=0.2243, simple_loss=0.3098, pruned_loss=0.06938, over 8493.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2929, pruned_loss=0.06532, over 1618273.54 frames. ], batch size: 28, lr: 4.03e-03, grad_scale: 16.0 +2023-02-06 22:38:56,663 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-06 22:38:59,402 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=146953.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:39:16,190 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.810e+02 2.362e+02 2.962e+02 3.993e+02 1.525e+03, threshold=5.923e+02, percent-clipped=6.0 +2023-02-06 22:39:22,647 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8179, 1.3296, 3.9756, 1.4941, 3.4959, 3.2869, 3.5701, 3.4785], + device='cuda:3'), covar=tensor([0.0674, 0.4620, 0.0630, 0.4001, 0.1261, 0.1125, 0.0732, 0.0761], + device='cuda:3'), in_proj_covar=tensor([0.0602, 0.0635, 0.0677, 0.0609, 0.0692, 0.0594, 0.0590, 0.0651], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 22:39:23,967 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=146988.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:39:27,289 INFO [train.py:901] (3/4) Epoch 19, batch 1500, loss[loss=0.215, simple_loss=0.3107, pruned_loss=0.05965, over 8471.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2933, pruned_loss=0.06534, over 1619336.01 frames. ], batch size: 25, lr: 4.03e-03, grad_scale: 16.0 +2023-02-06 22:40:03,257 INFO [train.py:901] (3/4) Epoch 19, batch 1550, loss[loss=0.1684, simple_loss=0.255, pruned_loss=0.04087, over 8040.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2945, pruned_loss=0.0662, over 1617612.48 frames. ], batch size: 20, lr: 4.03e-03, grad_scale: 16.0 +2023-02-06 22:40:22,655 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147068.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:40:28,398 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.668e+02 2.434e+02 2.984e+02 3.600e+02 8.495e+02, threshold=5.968e+02, percent-clipped=1.0 +2023-02-06 22:40:39,456 INFO [train.py:901] (3/4) Epoch 19, batch 1600, loss[loss=0.157, simple_loss=0.2389, pruned_loss=0.03757, over 7692.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2936, pruned_loss=0.06599, over 1617410.69 frames. ], batch size: 18, lr: 4.03e-03, grad_scale: 16.0 +2023-02-06 22:40:46,379 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147103.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:41:14,549 INFO [train.py:901] (3/4) Epoch 19, batch 1650, loss[loss=0.1974, simple_loss=0.2814, pruned_loss=0.05671, over 8365.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2936, pruned_loss=0.06564, over 1617389.61 frames. ], batch size: 24, lr: 4.03e-03, grad_scale: 8.0 +2023-02-06 22:41:40,974 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.588e+02 2.354e+02 2.709e+02 3.474e+02 7.081e+02, threshold=5.418e+02, percent-clipped=1.0 +2023-02-06 22:41:51,226 INFO [train.py:901] (3/4) Epoch 19, batch 1700, loss[loss=0.2037, simple_loss=0.2819, pruned_loss=0.06282, over 7931.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.2923, pruned_loss=0.06524, over 1613871.44 frames. ], batch size: 20, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:41:52,242 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2810, 1.9985, 2.6667, 2.2142, 2.6295, 2.2693, 1.9671, 1.4201], + device='cuda:3'), covar=tensor([0.5086, 0.4506, 0.1777, 0.3217, 0.2321, 0.2685, 0.1871, 0.4819], + device='cuda:3'), in_proj_covar=tensor([0.0936, 0.0953, 0.0788, 0.0917, 0.0984, 0.0869, 0.0730, 0.0810], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 22:42:00,572 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=147206.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:42:25,937 INFO [train.py:901] (3/4) Epoch 19, batch 1750, loss[loss=0.218, simple_loss=0.286, pruned_loss=0.07494, over 7425.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2927, pruned_loss=0.06543, over 1613518.79 frames. ], batch size: 17, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:42:37,958 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147259.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:42:41,411 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3247, 1.5336, 4.5012, 1.7842, 4.0033, 3.7327, 4.0653, 3.9743], + device='cuda:3'), covar=tensor([0.0629, 0.4704, 0.0547, 0.4224, 0.1059, 0.0999, 0.0597, 0.0677], + device='cuda:3'), in_proj_covar=tensor([0.0601, 0.0636, 0.0677, 0.0612, 0.0692, 0.0596, 0.0591, 0.0653], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 22:42:46,384 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6186, 4.6796, 4.1612, 1.9747, 4.1225, 4.2847, 4.1475, 4.1049], + device='cuda:3'), covar=tensor([0.0702, 0.0456, 0.0985, 0.4706, 0.0833, 0.0776, 0.1288, 0.0716], + device='cuda:3'), in_proj_covar=tensor([0.0513, 0.0423, 0.0428, 0.0526, 0.0415, 0.0427, 0.0410, 0.0375], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 22:42:51,055 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.541e+02 2.529e+02 3.043e+02 3.569e+02 7.736e+02, threshold=6.085e+02, percent-clipped=5.0 +2023-02-06 22:43:03,014 INFO [train.py:901] (3/4) Epoch 19, batch 1800, loss[loss=0.2497, simple_loss=0.3352, pruned_loss=0.08213, over 8580.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2932, pruned_loss=0.06631, over 1613527.12 frames. ], batch size: 31, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:43:22,541 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147321.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:43:24,605 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147324.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:43:27,349 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147328.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:43:27,659 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-06 22:43:37,466 INFO [train.py:901] (3/4) Epoch 19, batch 1850, loss[loss=0.2079, simple_loss=0.2969, pruned_loss=0.05949, over 8606.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2929, pruned_loss=0.06612, over 1617896.35 frames. ], batch size: 34, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:43:41,699 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147349.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:43:48,568 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147359.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:44:02,401 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.550e+02 2.300e+02 2.823e+02 3.606e+02 1.006e+03, threshold=5.645e+02, percent-clipped=2.0 +2023-02-06 22:44:02,738 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-06 22:44:06,673 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147384.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:44:12,517 INFO [train.py:901] (3/4) Epoch 19, batch 1900, loss[loss=0.1888, simple_loss=0.2707, pruned_loss=0.05343, over 7657.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2924, pruned_loss=0.06581, over 1619990.29 frames. ], batch size: 19, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:44:37,287 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147425.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:44:44,937 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-06 22:44:49,664 INFO [train.py:901] (3/4) Epoch 19, batch 1950, loss[loss=0.2447, simple_loss=0.3112, pruned_loss=0.08903, over 7409.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2913, pruned_loss=0.0648, over 1615467.71 frames. ], batch size: 17, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:44:55,965 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147452.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:44:56,513 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-06 22:45:13,743 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.377e+02 2.289e+02 2.862e+02 3.830e+02 8.439e+02, threshold=5.724e+02, percent-clipped=6.0 +2023-02-06 22:45:15,250 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-06 22:45:24,868 INFO [train.py:901] (3/4) Epoch 19, batch 2000, loss[loss=0.254, simple_loss=0.3214, pruned_loss=0.09327, over 8250.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2917, pruned_loss=0.06509, over 1614761.35 frames. ], batch size: 49, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:46:01,777 INFO [train.py:901] (3/4) Epoch 19, batch 2050, loss[loss=0.264, simple_loss=0.3287, pruned_loss=0.09965, over 8116.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2919, pruned_loss=0.0654, over 1610124.62 frames. ], batch size: 23, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:46:09,084 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.88 vs. limit=5.0 +2023-02-06 22:46:25,334 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147577.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:46:25,778 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.652e+02 2.500e+02 2.918e+02 3.445e+02 6.516e+02, threshold=5.836e+02, percent-clipped=2.0 +2023-02-06 22:46:36,254 INFO [train.py:901] (3/4) Epoch 19, batch 2100, loss[loss=0.2196, simple_loss=0.308, pruned_loss=0.06559, over 8337.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.2921, pruned_loss=0.06553, over 1611228.92 frames. ], batch size: 25, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:46:42,936 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147602.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:46:43,441 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=147603.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:46:47,320 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.54 vs. limit=5.0 +2023-02-06 22:47:12,091 INFO [train.py:901] (3/4) Epoch 19, batch 2150, loss[loss=0.2031, simple_loss=0.2967, pruned_loss=0.05478, over 8284.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2919, pruned_loss=0.06523, over 1616067.64 frames. ], batch size: 23, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:47:28,883 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6189, 2.7647, 1.8508, 2.2570, 2.2561, 1.6024, 2.2068, 2.1647], + device='cuda:3'), covar=tensor([0.1460, 0.0320, 0.1095, 0.0689, 0.0773, 0.1423, 0.0894, 0.0920], + device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0234, 0.0325, 0.0303, 0.0301, 0.0331, 0.0341, 0.0316], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 22:47:31,900 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 +2023-02-06 22:47:33,564 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=147672.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:47:36,914 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147677.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:47:37,423 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.398e+02 3.174e+02 3.852e+02 9.466e+02, threshold=6.348e+02, percent-clipped=6.0 +2023-02-06 22:47:47,720 INFO [train.py:901] (3/4) Epoch 19, batch 2200, loss[loss=0.2418, simple_loss=0.3216, pruned_loss=0.081, over 8590.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2918, pruned_loss=0.0654, over 1614156.52 frames. ], batch size: 31, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:48:04,720 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147718.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:48:21,945 INFO [train.py:901] (3/4) Epoch 19, batch 2250, loss[loss=0.2029, simple_loss=0.2936, pruned_loss=0.0561, over 8193.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2928, pruned_loss=0.06631, over 1613077.79 frames. ], batch size: 23, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:48:41,110 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=147769.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:48:47,101 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.454e+02 2.392e+02 3.089e+02 3.849e+02 9.613e+02, threshold=6.179e+02, percent-clipped=2.0 +2023-02-06 22:48:53,337 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147787.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:48:56,992 INFO [train.py:901] (3/4) Epoch 19, batch 2300, loss[loss=0.2321, simple_loss=0.3139, pruned_loss=0.07517, over 8237.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2913, pruned_loss=0.06566, over 1611446.26 frames. ], batch size: 24, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:48:58,974 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=147796.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:49:24,358 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147833.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:49:30,987 INFO [train.py:901] (3/4) Epoch 19, batch 2350, loss[loss=0.1903, simple_loss=0.2835, pruned_loss=0.04861, over 8466.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2923, pruned_loss=0.06595, over 1616105.32 frames. ], batch size: 25, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:49:53,249 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147875.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:49:55,896 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.642e+02 2.451e+02 2.984e+02 3.607e+02 1.132e+03, threshold=5.968e+02, percent-clipped=4.0 +2023-02-06 22:50:01,541 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147884.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:50:05,018 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4158, 2.5909, 1.6685, 2.2295, 2.1190, 1.3880, 2.0405, 2.1437], + device='cuda:3'), covar=tensor([0.1671, 0.0464, 0.1386, 0.0726, 0.0829, 0.1917, 0.1240, 0.1104], + device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0233, 0.0322, 0.0300, 0.0298, 0.0327, 0.0338, 0.0313], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 22:50:07,523 INFO [train.py:901] (3/4) Epoch 19, batch 2400, loss[loss=0.1922, simple_loss=0.2807, pruned_loss=0.0518, over 8242.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2928, pruned_loss=0.06649, over 1616385.85 frames. ], batch size: 24, lr: 4.02e-03, grad_scale: 8.0 +2023-02-06 22:50:17,641 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.74 vs. limit=5.0 +2023-02-06 22:50:20,040 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147911.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:50:32,261 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147929.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:50:41,466 INFO [train.py:901] (3/4) Epoch 19, batch 2450, loss[loss=0.2043, simple_loss=0.2893, pruned_loss=0.05963, over 8254.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2924, pruned_loss=0.06663, over 1614785.27 frames. ], batch size: 24, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:51:03,065 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147974.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:51:05,437 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.749e+02 2.599e+02 2.990e+02 3.557e+02 6.406e+02, threshold=5.981e+02, percent-clipped=1.0 +2023-02-06 22:51:15,599 INFO [train.py:901] (3/4) Epoch 19, batch 2500, loss[loss=0.1793, simple_loss=0.2622, pruned_loss=0.04815, over 7546.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2917, pruned_loss=0.06565, over 1616927.94 frames. ], batch size: 18, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:51:20,621 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147999.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:51:37,737 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148021.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:51:41,237 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5903, 1.8823, 2.6904, 1.5111, 1.9188, 1.9048, 1.7766, 1.7907], + device='cuda:3'), covar=tensor([0.1707, 0.2194, 0.0796, 0.3922, 0.1696, 0.3046, 0.1865, 0.2256], + device='cuda:3'), in_proj_covar=tensor([0.0513, 0.0577, 0.0548, 0.0626, 0.0634, 0.0583, 0.0518, 0.0627], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 22:51:52,315 INFO [train.py:901] (3/4) Epoch 19, batch 2550, loss[loss=0.1865, simple_loss=0.2529, pruned_loss=0.06006, over 7546.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2912, pruned_loss=0.06511, over 1618487.13 frames. ], batch size: 18, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:51:52,571 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148043.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:52:09,288 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148068.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:52:15,629 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.425e+02 2.379e+02 2.867e+02 3.516e+02 7.047e+02, threshold=5.734e+02, percent-clipped=3.0 +2023-02-06 22:52:17,130 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1578, 2.5964, 3.0119, 1.4964, 3.0418, 1.7512, 1.5628, 2.1448], + device='cuda:3'), covar=tensor([0.0871, 0.0380, 0.0287, 0.0929, 0.0449, 0.0993, 0.0921, 0.0517], + device='cuda:3'), in_proj_covar=tensor([0.0441, 0.0380, 0.0329, 0.0435, 0.0363, 0.0526, 0.0383, 0.0404], + device='cuda:3'), out_proj_covar=tensor([1.1934e-04, 1.0026e-04, 8.7018e-05, 1.1568e-04, 9.6228e-05, 1.5023e-04, + 1.0354e-04, 1.0795e-04], device='cuda:3') +2023-02-06 22:52:24,508 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2724, 3.8797, 2.5636, 3.1709, 3.0461, 1.8980, 3.0832, 3.2573], + device='cuda:3'), covar=tensor([0.1404, 0.0335, 0.1032, 0.0597, 0.0785, 0.1549, 0.1023, 0.0781], + device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0236, 0.0326, 0.0305, 0.0302, 0.0333, 0.0343, 0.0318], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 22:52:26,280 INFO [train.py:901] (3/4) Epoch 19, batch 2600, loss[loss=0.181, simple_loss=0.2731, pruned_loss=0.04444, over 8254.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2911, pruned_loss=0.0651, over 1617145.03 frames. ], batch size: 24, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:52:57,240 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148136.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:53:00,042 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148140.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:53:01,854 INFO [train.py:901] (3/4) Epoch 19, batch 2650, loss[loss=0.2345, simple_loss=0.3119, pruned_loss=0.07857, over 7934.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2926, pruned_loss=0.06594, over 1615817.06 frames. ], batch size: 20, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:53:16,797 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148165.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:53:18,195 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148167.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:53:24,870 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148177.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:53:25,453 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.795e+02 2.384e+02 2.853e+02 3.529e+02 7.126e+02, threshold=5.707e+02, percent-clipped=4.0 +2023-02-06 22:53:27,177 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.52 vs. limit=2.0 +2023-02-06 22:53:35,198 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148192.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:53:35,672 INFO [train.py:901] (3/4) Epoch 19, batch 2700, loss[loss=0.1641, simple_loss=0.2347, pruned_loss=0.04679, over 7541.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2927, pruned_loss=0.06575, over 1617604.56 frames. ], batch size: 18, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:53:38,591 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2588, 2.1611, 1.5433, 1.9157, 1.8476, 1.3644, 1.7821, 1.6606], + device='cuda:3'), covar=tensor([0.1352, 0.0372, 0.1234, 0.0519, 0.0711, 0.1511, 0.0890, 0.0954], + device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0237, 0.0327, 0.0306, 0.0302, 0.0333, 0.0344, 0.0318], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 22:53:54,176 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148219.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:54:03,924 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8784, 1.9935, 1.7642, 2.3344, 1.0981, 1.6122, 1.6870, 1.8839], + device='cuda:3'), covar=tensor([0.0704, 0.0713, 0.0939, 0.0435, 0.1110, 0.1297, 0.0868, 0.0749], + device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0198, 0.0252, 0.0214, 0.0208, 0.0248, 0.0256, 0.0213], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 22:54:11,924 INFO [train.py:901] (3/4) Epoch 19, batch 2750, loss[loss=0.1833, simple_loss=0.2526, pruned_loss=0.05701, over 7431.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2908, pruned_loss=0.06504, over 1616599.02 frames. ], batch size: 17, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:54:33,010 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148273.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:54:36,062 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.455e+02 2.484e+02 2.895e+02 4.098e+02 9.310e+02, threshold=5.790e+02, percent-clipped=8.0 +2023-02-06 22:54:45,410 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148292.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:54:45,920 INFO [train.py:901] (3/4) Epoch 19, batch 2800, loss[loss=0.2503, simple_loss=0.3224, pruned_loss=0.08916, over 8603.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2922, pruned_loss=0.06631, over 1617596.09 frames. ], batch size: 39, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:54:54,040 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148305.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:55:13,962 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148334.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:55:19,843 INFO [train.py:901] (3/4) Epoch 19, batch 2850, loss[loss=0.2036, simple_loss=0.2865, pruned_loss=0.06036, over 8301.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2924, pruned_loss=0.06599, over 1620944.46 frames. ], batch size: 23, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:55:29,881 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.81 vs. limit=5.0 +2023-02-06 22:55:46,057 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.455e+02 2.512e+02 2.931e+02 3.824e+02 7.566e+02, threshold=5.862e+02, percent-clipped=4.0 +2023-02-06 22:55:52,986 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148388.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:55:55,653 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148392.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:55:56,118 INFO [train.py:901] (3/4) Epoch 19, batch 2900, loss[loss=0.2341, simple_loss=0.3143, pruned_loss=0.07691, over 8690.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.294, pruned_loss=0.06664, over 1627339.40 frames. ], batch size: 39, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:56:10,221 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 +2023-02-06 22:56:12,664 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148417.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:56:29,357 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-06 22:56:29,949 INFO [train.py:901] (3/4) Epoch 19, batch 2950, loss[loss=0.2347, simple_loss=0.3089, pruned_loss=0.08026, over 8455.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2944, pruned_loss=0.06638, over 1623481.55 frames. ], batch size: 49, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:56:32,823 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148447.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:56:52,418 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9193, 1.5057, 1.7214, 1.3146, 0.9127, 1.4773, 1.6314, 1.3743], + device='cuda:3'), covar=tensor([0.0537, 0.1279, 0.1633, 0.1431, 0.0628, 0.1525, 0.0692, 0.0678], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0153, 0.0190, 0.0158, 0.0100, 0.0162, 0.0113, 0.0141], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 22:56:54,933 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.676e+02 2.514e+02 3.009e+02 3.973e+02 7.443e+02, threshold=6.017e+02, percent-clipped=3.0 +2023-02-06 22:57:06,345 INFO [train.py:901] (3/4) Epoch 19, batch 3000, loss[loss=0.1887, simple_loss=0.283, pruned_loss=0.04726, over 8094.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.2943, pruned_loss=0.06654, over 1617063.78 frames. ], batch size: 23, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:57:06,345 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 22:57:22,669 INFO [train.py:935] (3/4) Epoch 19, validation: loss=0.1752, simple_loss=0.2756, pruned_loss=0.03738, over 944034.00 frames. +2023-02-06 22:57:22,671 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 22:57:36,057 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.77 vs. limit=5.0 +2023-02-06 22:57:38,595 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148516.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:57:56,920 INFO [train.py:901] (3/4) Epoch 19, batch 3050, loss[loss=0.2456, simple_loss=0.3001, pruned_loss=0.09556, over 7783.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2943, pruned_loss=0.06646, over 1611781.42 frames. ], batch size: 19, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:58:00,732 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148548.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:58:17,764 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148573.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:58:21,054 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.667e+02 2.398e+02 2.811e+02 3.727e+02 6.995e+02, threshold=5.622e+02, percent-clipped=3.0 +2023-02-06 22:58:30,234 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148590.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:58:32,157 INFO [train.py:901] (3/4) Epoch 19, batch 3100, loss[loss=0.1988, simple_loss=0.2844, pruned_loss=0.0566, over 7538.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2931, pruned_loss=0.06588, over 1612540.75 frames. ], batch size: 18, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:58:49,337 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148615.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:59:09,445 INFO [train.py:901] (3/4) Epoch 19, batch 3150, loss[loss=0.2544, simple_loss=0.3423, pruned_loss=0.08324, over 8295.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.292, pruned_loss=0.06538, over 1611269.28 frames. ], batch size: 23, lr: 4.01e-03, grad_scale: 8.0 +2023-02-06 22:59:10,320 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148644.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:59:13,428 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148649.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:59:26,357 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148669.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:59:26,420 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148669.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 22:59:32,307 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.732e+02 2.358e+02 3.073e+02 3.824e+02 9.523e+02, threshold=6.146e+02, percent-clipped=8.0 +2023-02-06 22:59:42,404 INFO [train.py:901] (3/4) Epoch 19, batch 3200, loss[loss=0.1754, simple_loss=0.2523, pruned_loss=0.04924, over 7529.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.293, pruned_loss=0.06613, over 1616454.26 frames. ], batch size: 18, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 23:00:12,970 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148734.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:00:19,660 INFO [train.py:901] (3/4) Epoch 19, batch 3250, loss[loss=0.2049, simple_loss=0.2819, pruned_loss=0.06398, over 8046.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2923, pruned_loss=0.06559, over 1617090.09 frames. ], batch size: 22, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 23:00:26,696 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7691, 1.6179, 1.7971, 1.6861, 1.0202, 1.6695, 2.1029, 1.9852], + device='cuda:3'), covar=tensor([0.0488, 0.1343, 0.1747, 0.1404, 0.0625, 0.1534, 0.0658, 0.0620], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0154, 0.0191, 0.0158, 0.0100, 0.0162, 0.0113, 0.0141], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 23:00:34,083 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148764.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:00:43,250 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.643e+02 2.443e+02 3.073e+02 4.112e+02 8.183e+02, threshold=6.146e+02, percent-clipped=4.0 +2023-02-06 23:00:52,318 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148791.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:00:53,597 INFO [train.py:901] (3/4) Epoch 19, batch 3300, loss[loss=0.2312, simple_loss=0.3086, pruned_loss=0.07689, over 7550.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2917, pruned_loss=0.06452, over 1617937.87 frames. ], batch size: 18, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 23:01:28,241 INFO [train.py:901] (3/4) Epoch 19, batch 3350, loss[loss=0.2614, simple_loss=0.3278, pruned_loss=0.09753, over 8367.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2923, pruned_loss=0.06473, over 1619260.48 frames. ], batch size: 24, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 23:01:41,953 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=148860.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:01:53,953 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 2.291e+02 2.864e+02 3.449e+02 6.722e+02, threshold=5.728e+02, percent-clipped=1.0 +2023-02-06 23:02:00,138 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7269, 1.9879, 2.2387, 1.4356, 2.2228, 1.5926, 0.6926, 1.9461], + device='cuda:3'), covar=tensor([0.0524, 0.0304, 0.0237, 0.0515, 0.0385, 0.0835, 0.0743, 0.0286], + device='cuda:3'), in_proj_covar=tensor([0.0440, 0.0378, 0.0328, 0.0435, 0.0362, 0.0528, 0.0382, 0.0405], + device='cuda:3'), out_proj_covar=tensor([1.1921e-04, 9.9853e-05, 8.6656e-05, 1.1564e-04, 9.6069e-05, 1.5083e-04, + 1.0330e-04, 1.0814e-04], device='cuda:3') +2023-02-06 23:02:04,174 INFO [train.py:901] (3/4) Epoch 19, batch 3400, loss[loss=0.2279, simple_loss=0.3122, pruned_loss=0.07184, over 8464.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2925, pruned_loss=0.06508, over 1619048.36 frames. ], batch size: 25, lr: 4.00e-03, grad_scale: 4.0 +2023-02-06 23:02:11,211 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1610, 1.8834, 2.5098, 2.0856, 2.4380, 2.1517, 1.8686, 1.3040], + device='cuda:3'), covar=tensor([0.5132, 0.4772, 0.1922, 0.3538, 0.2392, 0.2747, 0.1852, 0.5086], + device='cuda:3'), in_proj_covar=tensor([0.0925, 0.0947, 0.0781, 0.0913, 0.0976, 0.0867, 0.0726, 0.0807], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 23:02:13,160 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148906.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:02:31,055 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.61 vs. limit=5.0 +2023-02-06 23:02:38,070 INFO [train.py:901] (3/4) Epoch 19, batch 3450, loss[loss=0.2265, simple_loss=0.3062, pruned_loss=0.07343, over 8312.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2923, pruned_loss=0.06509, over 1617932.34 frames. ], batch size: 49, lr: 4.00e-03, grad_scale: 4.0 +2023-02-06 23:02:40,625 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-06 23:03:01,927 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148975.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:03:04,405 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.764e+02 2.340e+02 2.956e+02 3.727e+02 1.104e+03, threshold=5.912e+02, percent-clipped=3.0 +2023-02-06 23:03:14,141 INFO [train.py:901] (3/4) Epoch 19, batch 3500, loss[loss=0.2235, simple_loss=0.3007, pruned_loss=0.07319, over 8191.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2914, pruned_loss=0.06458, over 1616263.77 frames. ], batch size: 23, lr: 4.00e-03, grad_scale: 4.0 +2023-02-06 23:03:28,308 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=149013.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:03:33,350 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149020.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:03:35,938 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-06 23:03:48,892 INFO [train.py:901] (3/4) Epoch 19, batch 3550, loss[loss=0.181, simple_loss=0.2466, pruned_loss=0.05773, over 7431.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2915, pruned_loss=0.06466, over 1614090.89 frames. ], batch size: 17, lr: 4.00e-03, grad_scale: 4.0 +2023-02-06 23:03:50,371 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149045.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:04:13,084 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=149078.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:04:13,643 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.434e+02 2.461e+02 3.087e+02 3.824e+02 7.251e+02, threshold=6.175e+02, percent-clipped=6.0 +2023-02-06 23:04:25,648 INFO [train.py:901] (3/4) Epoch 19, batch 3600, loss[loss=0.2148, simple_loss=0.2954, pruned_loss=0.06711, over 8661.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2912, pruned_loss=0.06478, over 1610226.80 frames. ], batch size: 34, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 23:04:49,820 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149128.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:04:59,740 INFO [train.py:901] (3/4) Epoch 19, batch 3650, loss[loss=0.2399, simple_loss=0.3166, pruned_loss=0.08157, over 8539.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2907, pruned_loss=0.06433, over 1611486.81 frames. ], batch size: 31, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 23:05:13,229 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149162.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:05:24,388 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.497e+02 2.296e+02 2.731e+02 3.488e+02 6.725e+02, threshold=5.462e+02, percent-clipped=1.0 +2023-02-06 23:05:30,734 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149187.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:05:35,229 INFO [train.py:901] (3/4) Epoch 19, batch 3700, loss[loss=0.2143, simple_loss=0.2991, pruned_loss=0.06474, over 8514.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.291, pruned_loss=0.0641, over 1613303.84 frames. ], batch size: 31, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 23:05:35,415 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149193.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:05:38,069 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-06 23:06:02,793 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149231.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:06:10,467 INFO [train.py:901] (3/4) Epoch 19, batch 3750, loss[loss=0.2184, simple_loss=0.2838, pruned_loss=0.07645, over 7251.00 frames. ], tot_loss[loss=0.2111, simple_loss=0.2922, pruned_loss=0.06504, over 1614975.30 frames. ], batch size: 16, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 23:06:19,375 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149256.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:06:22,872 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.92 vs. limit=5.0 +2023-02-06 23:06:25,464 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4838, 1.9657, 2.9042, 1.3287, 2.1684, 1.8872, 1.6523, 2.0763], + device='cuda:3'), covar=tensor([0.1902, 0.2304, 0.0782, 0.4399, 0.1775, 0.3022, 0.2125, 0.2325], + device='cuda:3'), in_proj_covar=tensor([0.0517, 0.0581, 0.0552, 0.0628, 0.0638, 0.0588, 0.0521, 0.0630], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 23:06:34,566 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.258e+02 2.530e+02 3.028e+02 3.831e+02 7.632e+02, threshold=6.056e+02, percent-clipped=6.0 +2023-02-06 23:06:44,222 INFO [train.py:901] (3/4) Epoch 19, batch 3800, loss[loss=0.2258, simple_loss=0.3151, pruned_loss=0.06827, over 8469.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2928, pruned_loss=0.06513, over 1616470.48 frames. ], batch size: 25, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 23:07:20,713 INFO [train.py:901] (3/4) Epoch 19, batch 3850, loss[loss=0.2892, simple_loss=0.3506, pruned_loss=0.1139, over 6879.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2935, pruned_loss=0.0657, over 1613613.70 frames. ], batch size: 71, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 23:07:21,609 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4383, 2.6254, 1.9297, 2.2381, 2.2031, 1.6450, 1.9826, 2.0134], + device='cuda:3'), covar=tensor([0.1567, 0.0368, 0.1092, 0.0596, 0.0735, 0.1393, 0.1175, 0.1041], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0239, 0.0327, 0.0304, 0.0300, 0.0331, 0.0342, 0.0317], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 23:07:42,390 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-06 23:07:45,096 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.339e+02 2.409e+02 2.948e+02 3.728e+02 6.848e+02, threshold=5.896e+02, percent-clipped=3.0 +2023-02-06 23:07:46,036 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5838, 1.3477, 2.0644, 1.7102, 1.7308, 1.5204, 1.3506, 0.8186], + device='cuda:3'), covar=tensor([0.6602, 0.5530, 0.1980, 0.3533, 0.2934, 0.4234, 0.2907, 0.4698], + device='cuda:3'), in_proj_covar=tensor([0.0930, 0.0955, 0.0787, 0.0918, 0.0981, 0.0871, 0.0729, 0.0812], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 23:07:48,708 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149384.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:07:54,629 INFO [train.py:901] (3/4) Epoch 19, batch 3900, loss[loss=0.1759, simple_loss=0.2654, pruned_loss=0.04321, over 8578.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2938, pruned_loss=0.06587, over 1617694.96 frames. ], batch size: 31, lr: 4.00e-03, grad_scale: 8.0 +2023-02-06 23:07:58,841 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.6101, 5.6948, 5.0898, 2.3480, 5.0463, 5.4073, 5.1635, 4.9841], + device='cuda:3'), covar=tensor([0.0525, 0.0392, 0.0934, 0.4364, 0.0634, 0.0648, 0.1069, 0.0630], + device='cuda:3'), in_proj_covar=tensor([0.0518, 0.0427, 0.0429, 0.0530, 0.0419, 0.0433, 0.0411, 0.0375], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 23:07:58,986 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6571, 2.2948, 3.8832, 1.5863, 2.8974, 2.1607, 1.9444, 2.7889], + device='cuda:3'), covar=tensor([0.1885, 0.2303, 0.0807, 0.4132, 0.1821, 0.3120, 0.1991, 0.2450], + device='cuda:3'), in_proj_covar=tensor([0.0514, 0.0581, 0.0551, 0.0625, 0.0636, 0.0585, 0.0519, 0.0627], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 23:08:06,571 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149409.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:08:07,213 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6477, 1.4323, 4.8894, 1.8457, 4.2301, 3.9868, 4.3809, 4.2226], + device='cuda:3'), covar=tensor([0.0630, 0.5065, 0.0469, 0.4074, 0.1180, 0.0956, 0.0617, 0.0698], + device='cuda:3'), in_proj_covar=tensor([0.0600, 0.0635, 0.0676, 0.0608, 0.0686, 0.0593, 0.0584, 0.0649], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 23:08:27,992 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4955, 1.7287, 1.8209, 1.1951, 1.9142, 1.3211, 0.4503, 1.6795], + device='cuda:3'), covar=tensor([0.0495, 0.0368, 0.0294, 0.0543, 0.0425, 0.0925, 0.0808, 0.0264], + device='cuda:3'), in_proj_covar=tensor([0.0440, 0.0381, 0.0329, 0.0438, 0.0363, 0.0529, 0.0383, 0.0406], + device='cuda:3'), out_proj_covar=tensor([1.1911e-04, 1.0064e-04, 8.6970e-05, 1.1644e-04, 9.6128e-05, 1.5106e-04, + 1.0364e-04, 1.0857e-04], device='cuda:3') +2023-02-06 23:08:31,940 INFO [train.py:901] (3/4) Epoch 19, batch 3950, loss[loss=0.1892, simple_loss=0.2558, pruned_loss=0.06134, over 7264.00 frames. ], tot_loss[loss=0.2111, simple_loss=0.2925, pruned_loss=0.06485, over 1614744.72 frames. ], batch size: 16, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:08:36,337 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149449.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:08:53,030 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149474.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:08:56,233 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.548e+02 2.497e+02 2.881e+02 4.050e+02 6.266e+02, threshold=5.763e+02, percent-clipped=1.0 +2023-02-06 23:09:05,724 INFO [train.py:901] (3/4) Epoch 19, batch 4000, loss[loss=0.1817, simple_loss=0.2626, pruned_loss=0.05044, over 7811.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2916, pruned_loss=0.06443, over 1612431.95 frames. ], batch size: 19, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:09:32,416 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149532.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:09:40,134 INFO [train.py:901] (3/4) Epoch 19, batch 4050, loss[loss=0.2035, simple_loss=0.2906, pruned_loss=0.05818, over 8644.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2928, pruned_loss=0.06495, over 1616624.63 frames. ], batch size: 34, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:10:05,796 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.620e+02 2.470e+02 3.003e+02 4.246e+02 8.728e+02, threshold=6.007e+02, percent-clipped=8.0 +2023-02-06 23:10:08,590 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149583.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:10:15,179 INFO [train.py:901] (3/4) Epoch 19, batch 4100, loss[loss=0.2184, simple_loss=0.2763, pruned_loss=0.08024, over 7789.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.293, pruned_loss=0.06507, over 1613722.81 frames. ], batch size: 19, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:10:49,879 INFO [train.py:901] (3/4) Epoch 19, batch 4150, loss[loss=0.2007, simple_loss=0.2921, pruned_loss=0.0547, over 8615.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2916, pruned_loss=0.06387, over 1611313.49 frames. ], batch size: 31, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:10:58,400 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2780, 1.2979, 1.5416, 1.2076, 0.7075, 1.3897, 1.2327, 1.1507], + device='cuda:3'), covar=tensor([0.0553, 0.1250, 0.1669, 0.1443, 0.0573, 0.1484, 0.0698, 0.0644], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0152, 0.0189, 0.0157, 0.0099, 0.0161, 0.0113, 0.0140], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 23:11:16,655 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.652e+02 2.273e+02 2.791e+02 3.594e+02 5.057e+02, threshold=5.582e+02, percent-clipped=0.0 +2023-02-06 23:11:26,114 INFO [train.py:901] (3/4) Epoch 19, batch 4200, loss[loss=0.2608, simple_loss=0.3378, pruned_loss=0.09189, over 8514.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2908, pruned_loss=0.06342, over 1611296.62 frames. ], batch size: 28, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:11:36,588 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-06 23:11:59,494 INFO [train.py:901] (3/4) Epoch 19, batch 4250, loss[loss=0.1903, simple_loss=0.2836, pruned_loss=0.04853, over 8032.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2898, pruned_loss=0.0627, over 1610741.39 frames. ], batch size: 22, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:12:00,910 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-06 23:12:14,479 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149764.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 23:12:25,319 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.327e+02 2.444e+02 3.025e+02 3.928e+02 1.033e+03, threshold=6.050e+02, percent-clipped=5.0 +2023-02-06 23:12:35,589 INFO [train.py:901] (3/4) Epoch 19, batch 4300, loss[loss=0.2217, simple_loss=0.3215, pruned_loss=0.06099, over 8341.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2912, pruned_loss=0.06342, over 1611840.20 frames. ], batch size: 25, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:13:04,942 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3862, 1.3385, 1.7475, 1.2651, 1.0806, 1.7373, 0.2660, 1.1047], + device='cuda:3'), covar=tensor([0.1892, 0.1341, 0.0424, 0.0963, 0.2865, 0.0470, 0.2133, 0.1318], + device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0195, 0.0124, 0.0222, 0.0272, 0.0133, 0.0171, 0.0188], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 23:13:10,090 INFO [train.py:901] (3/4) Epoch 19, batch 4350, loss[loss=0.208, simple_loss=0.2836, pruned_loss=0.06622, over 8482.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2908, pruned_loss=0.06369, over 1608777.15 frames. ], batch size: 27, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:13:33,152 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-06 23:13:33,237 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=149876.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:13:35,197 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.529e+02 2.416e+02 2.972e+02 3.761e+02 1.184e+03, threshold=5.944e+02, percent-clipped=4.0 +2023-02-06 23:13:44,579 INFO [train.py:901] (3/4) Epoch 19, batch 4400, loss[loss=0.2163, simple_loss=0.3044, pruned_loss=0.06405, over 8288.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2909, pruned_loss=0.06393, over 1607348.29 frames. ], batch size: 23, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:14:09,977 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=149927.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:14:11,337 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5322, 2.3720, 1.7594, 2.1458, 2.0457, 1.6167, 1.8867, 1.8504], + device='cuda:3'), covar=tensor([0.1194, 0.0330, 0.1055, 0.0495, 0.0592, 0.1227, 0.0809, 0.0870], + device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0239, 0.0330, 0.0307, 0.0302, 0.0334, 0.0344, 0.0319], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 23:14:14,577 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-06 23:14:20,881 INFO [train.py:901] (3/4) Epoch 19, batch 4450, loss[loss=0.2344, simple_loss=0.3064, pruned_loss=0.08124, over 8031.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2928, pruned_loss=0.06493, over 1615025.87 frames. ], batch size: 22, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:14:23,726 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4217, 2.3785, 1.6780, 2.1683, 2.0495, 1.4579, 1.8511, 1.8228], + device='cuda:3'), covar=tensor([0.1437, 0.0385, 0.1286, 0.0534, 0.0671, 0.1532, 0.1009, 0.0979], + device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0239, 0.0331, 0.0307, 0.0303, 0.0334, 0.0345, 0.0319], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 23:14:39,765 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3122, 1.3908, 1.3032, 1.8059, 0.7827, 1.2124, 1.2472, 1.4300], + device='cuda:3'), covar=tensor([0.0954, 0.0833, 0.1055, 0.0521, 0.1118, 0.1469, 0.0828, 0.0745], + device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0199, 0.0250, 0.0215, 0.0207, 0.0249, 0.0255, 0.0212], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 23:14:44,878 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.620e+02 2.522e+02 2.925e+02 4.193e+02 1.036e+03, threshold=5.849e+02, percent-clipped=7.0 +2023-02-06 23:14:53,315 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149991.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:14:54,467 INFO [train.py:901] (3/4) Epoch 19, batch 4500, loss[loss=0.2162, simple_loss=0.3091, pruned_loss=0.0617, over 8327.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2915, pruned_loss=0.06451, over 1615697.28 frames. ], batch size: 25, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:15:08,402 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-06 23:15:31,621 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150042.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:15:32,129 INFO [train.py:901] (3/4) Epoch 19, batch 4550, loss[loss=0.2216, simple_loss=0.306, pruned_loss=0.06855, over 8247.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2914, pruned_loss=0.06441, over 1617317.06 frames. ], batch size: 24, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:15:35,177 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-06 23:15:56,356 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.731e+02 2.399e+02 2.811e+02 3.428e+02 5.502e+02, threshold=5.622e+02, percent-clipped=0.0 +2023-02-06 23:16:05,758 INFO [train.py:901] (3/4) Epoch 19, batch 4600, loss[loss=0.1833, simple_loss=0.2613, pruned_loss=0.05269, over 7653.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.29, pruned_loss=0.06374, over 1608469.24 frames. ], batch size: 19, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:16:08,477 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150097.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:16:15,971 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=150108.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 23:16:41,502 INFO [train.py:901] (3/4) Epoch 19, batch 4650, loss[loss=0.1675, simple_loss=0.2611, pruned_loss=0.03699, over 8027.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2896, pruned_loss=0.06352, over 1608336.90 frames. ], batch size: 22, lr: 3.99e-03, grad_scale: 8.0 +2023-02-06 23:16:47,967 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6196, 1.4860, 1.7236, 1.3370, 0.7963, 1.5379, 1.5824, 1.3919], + device='cuda:3'), covar=tensor([0.0546, 0.1242, 0.1630, 0.1446, 0.0579, 0.1446, 0.0659, 0.0647], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0153, 0.0189, 0.0159, 0.0100, 0.0162, 0.0113, 0.0141], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 23:17:06,556 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.641e+02 2.474e+02 2.856e+02 3.464e+02 8.049e+02, threshold=5.712e+02, percent-clipped=3.0 +2023-02-06 23:17:13,069 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-06 23:17:16,085 INFO [train.py:901] (3/4) Epoch 19, batch 4700, loss[loss=0.1795, simple_loss=0.2591, pruned_loss=0.04995, over 8253.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2889, pruned_loss=0.06354, over 1598375.69 frames. ], batch size: 22, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:17:22,191 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2023-02-06 23:17:36,661 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150223.0, num_to_drop=1, layers_to_drop={1} +2023-02-06 23:17:50,822 INFO [train.py:901] (3/4) Epoch 19, batch 4750, loss[loss=0.2395, simple_loss=0.3267, pruned_loss=0.07612, over 8486.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2901, pruned_loss=0.06407, over 1602992.67 frames. ], batch size: 28, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:17:53,804 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150247.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:18:12,325 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150272.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:18:13,463 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-06 23:18:15,512 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-06 23:18:16,850 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.498e+02 2.315e+02 2.829e+02 3.523e+02 6.730e+02, threshold=5.657e+02, percent-clipped=3.0 +2023-02-06 23:18:23,094 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2229, 1.8051, 3.2996, 1.3875, 2.2597, 3.5792, 3.7129, 3.0720], + device='cuda:3'), covar=tensor([0.1018, 0.1502, 0.0392, 0.2187, 0.1112, 0.0226, 0.0518, 0.0572], + device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0317, 0.0287, 0.0312, 0.0301, 0.0265, 0.0406, 0.0304], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 23:18:25,835 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150292.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:18:26,364 INFO [train.py:901] (3/4) Epoch 19, batch 4800, loss[loss=0.2208, simple_loss=0.3096, pruned_loss=0.06603, over 8461.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2892, pruned_loss=0.06295, over 1608182.58 frames. ], batch size: 25, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:18:29,958 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150298.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:18:46,558 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150323.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:18:52,217 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 +2023-02-06 23:19:00,006 INFO [train.py:901] (3/4) Epoch 19, batch 4850, loss[loss=0.2434, simple_loss=0.3139, pruned_loss=0.08641, over 8103.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2906, pruned_loss=0.06382, over 1612119.05 frames. ], batch size: 23, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:19:05,347 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-06 23:19:27,015 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 2.476e+02 2.899e+02 3.621e+02 6.951e+02, threshold=5.799e+02, percent-clipped=6.0 +2023-02-06 23:19:36,191 INFO [train.py:901] (3/4) Epoch 19, batch 4900, loss[loss=0.186, simple_loss=0.2624, pruned_loss=0.05478, over 7513.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2894, pruned_loss=0.06334, over 1612893.12 frames. ], batch size: 18, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:19:39,851 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6274, 1.6635, 1.7622, 1.4573, 1.8032, 1.4709, 0.9245, 1.6055], + device='cuda:3'), covar=tensor([0.0369, 0.0281, 0.0197, 0.0323, 0.0274, 0.0546, 0.0609, 0.0197], + device='cuda:3'), in_proj_covar=tensor([0.0439, 0.0379, 0.0328, 0.0435, 0.0361, 0.0523, 0.0384, 0.0406], + device='cuda:3'), out_proj_covar=tensor([1.1875e-04, 1.0024e-04, 8.6658e-05, 1.1544e-04, 9.5794e-05, 1.4929e-04, + 1.0392e-04, 1.0839e-04], device='cuda:3') +2023-02-06 23:20:07,723 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=150441.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:20:09,001 INFO [train.py:901] (3/4) Epoch 19, batch 4950, loss[loss=0.1912, simple_loss=0.2789, pruned_loss=0.05173, over 8461.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2911, pruned_loss=0.06502, over 1613115.57 frames. ], batch size: 25, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:20:33,700 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.602e+02 2.356e+02 2.775e+02 3.573e+02 1.033e+03, threshold=5.550e+02, percent-clipped=4.0 +2023-02-06 23:20:33,934 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150479.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 23:20:43,986 INFO [train.py:901] (3/4) Epoch 19, batch 5000, loss[loss=0.1841, simple_loss=0.2701, pruned_loss=0.04904, over 8072.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2903, pruned_loss=0.06442, over 1613514.08 frames. ], batch size: 21, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:20:52,078 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150504.0, num_to_drop=1, layers_to_drop={0} +2023-02-06 23:21:07,769 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.43 vs. limit=5.0 +2023-02-06 23:21:16,239 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.64 vs. limit=5.0 +2023-02-06 23:21:17,801 INFO [train.py:901] (3/4) Epoch 19, batch 5050, loss[loss=0.1827, simple_loss=0.2651, pruned_loss=0.05014, over 7934.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2905, pruned_loss=0.06461, over 1611177.93 frames. ], batch size: 20, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:21:25,939 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150555.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:21:26,624 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150556.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:21:40,931 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-06 23:21:41,605 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.454e+02 2.501e+02 3.000e+02 3.972e+02 7.212e+02, threshold=5.999e+02, percent-clipped=3.0 +2023-02-06 23:21:51,771 INFO [train.py:901] (3/4) Epoch 19, batch 5100, loss[loss=0.2293, simple_loss=0.3075, pruned_loss=0.07553, over 8578.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2899, pruned_loss=0.06464, over 1607700.24 frames. ], batch size: 31, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:22:18,178 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9825, 1.6936, 3.3769, 1.3127, 2.3825, 3.6323, 3.7270, 3.1182], + device='cuda:3'), covar=tensor([0.1198, 0.1603, 0.0313, 0.2263, 0.0945, 0.0259, 0.0592, 0.0582], + device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0317, 0.0287, 0.0312, 0.0301, 0.0265, 0.0406, 0.0303], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-06 23:22:23,321 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=150636.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:22:25,331 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2362, 1.4159, 3.3611, 1.0884, 2.9856, 2.8225, 3.0523, 2.9457], + device='cuda:3'), covar=tensor([0.0802, 0.4133, 0.0773, 0.4029, 0.1357, 0.1113, 0.0777, 0.0976], + device='cuda:3'), in_proj_covar=tensor([0.0598, 0.0633, 0.0672, 0.0610, 0.0685, 0.0589, 0.0591, 0.0650], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 23:22:27,867 INFO [train.py:901] (3/4) Epoch 19, batch 5150, loss[loss=0.2222, simple_loss=0.2986, pruned_loss=0.07286, over 8074.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2907, pruned_loss=0.06499, over 1611736.66 frames. ], batch size: 21, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:22:51,866 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.708e+02 2.510e+02 3.215e+02 4.688e+02 9.098e+02, threshold=6.429e+02, percent-clipped=11.0 +2023-02-06 23:23:01,327 INFO [train.py:901] (3/4) Epoch 19, batch 5200, loss[loss=0.2065, simple_loss=0.288, pruned_loss=0.06256, over 8504.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2916, pruned_loss=0.06547, over 1608084.03 frames. ], batch size: 26, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:23:38,106 INFO [train.py:901] (3/4) Epoch 19, batch 5250, loss[loss=0.2097, simple_loss=0.283, pruned_loss=0.06819, over 7785.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2913, pruned_loss=0.06527, over 1609927.38 frames. ], batch size: 19, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:23:40,643 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-06 23:23:42,646 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150750.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:23:43,387 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150751.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:24:00,396 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5871, 2.4971, 1.7690, 2.2051, 2.0994, 1.4845, 2.1252, 2.0513], + device='cuda:3'), covar=tensor([0.1348, 0.0428, 0.1198, 0.0613, 0.0724, 0.1581, 0.0889, 0.1064], + device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0236, 0.0326, 0.0305, 0.0299, 0.0330, 0.0340, 0.0315], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 23:24:01,539 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.351e+02 2.565e+02 3.080e+02 4.191e+02 1.354e+03, threshold=6.160e+02, percent-clipped=9.0 +2023-02-06 23:24:10,893 INFO [train.py:901] (3/4) Epoch 19, batch 5300, loss[loss=0.2229, simple_loss=0.3138, pruned_loss=0.06593, over 8137.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2936, pruned_loss=0.06645, over 1615366.74 frames. ], batch size: 22, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:24:11,729 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8172, 1.4750, 3.9499, 1.4206, 3.4345, 3.2646, 3.5443, 3.4128], + device='cuda:3'), covar=tensor([0.0675, 0.4661, 0.0694, 0.4299, 0.1374, 0.1092, 0.0705, 0.0864], + device='cuda:3'), in_proj_covar=tensor([0.0600, 0.0636, 0.0672, 0.0611, 0.0689, 0.0591, 0.0592, 0.0651], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 23:24:12,434 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7221, 2.1246, 4.2063, 1.5310, 3.2035, 2.2873, 1.7021, 2.9496], + device='cuda:3'), covar=tensor([0.1983, 0.2842, 0.0741, 0.4708, 0.1621, 0.3149, 0.2457, 0.2232], + device='cuda:3'), in_proj_covar=tensor([0.0517, 0.0584, 0.0554, 0.0633, 0.0642, 0.0589, 0.0523, 0.0630], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 23:24:13,044 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1415, 1.6446, 1.3489, 1.7330, 1.3320, 1.1872, 1.3971, 1.5043], + device='cuda:3'), covar=tensor([0.0908, 0.0470, 0.1423, 0.0461, 0.0778, 0.1671, 0.0863, 0.0669], + device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0236, 0.0327, 0.0305, 0.0300, 0.0330, 0.0341, 0.0316], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 23:24:13,364 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 +2023-02-06 23:24:23,656 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150812.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:24:31,992 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-06 23:24:34,793 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 +2023-02-06 23:24:41,666 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150837.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:24:46,253 INFO [train.py:901] (3/4) Epoch 19, batch 5350, loss[loss=0.2606, simple_loss=0.3372, pruned_loss=0.09198, over 8440.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2941, pruned_loss=0.06689, over 1612484.13 frames. ], batch size: 25, lr: 3.98e-03, grad_scale: 8.0 +2023-02-06 23:25:10,968 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.639e+02 2.499e+02 2.979e+02 3.723e+02 8.863e+02, threshold=5.959e+02, percent-clipped=1.0 +2023-02-06 23:25:20,511 INFO [train.py:901] (3/4) Epoch 19, batch 5400, loss[loss=0.2259, simple_loss=0.312, pruned_loss=0.06994, over 8462.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2943, pruned_loss=0.06693, over 1609748.87 frames. ], batch size: 25, lr: 3.98e-03, grad_scale: 16.0 +2023-02-06 23:25:24,749 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=150899.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:25:25,114 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 +2023-02-06 23:25:37,930 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150918.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:25:55,449 INFO [train.py:901] (3/4) Epoch 19, batch 5450, loss[loss=0.2214, simple_loss=0.3021, pruned_loss=0.07042, over 8426.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2925, pruned_loss=0.0663, over 1608512.45 frames. ], batch size: 27, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:26:02,824 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6298, 1.8750, 2.0850, 1.3774, 2.1803, 1.4566, 0.6699, 1.9042], + device='cuda:3'), covar=tensor([0.0601, 0.0333, 0.0267, 0.0612, 0.0393, 0.0929, 0.0871, 0.0304], + device='cuda:3'), in_proj_covar=tensor([0.0441, 0.0381, 0.0331, 0.0436, 0.0364, 0.0525, 0.0384, 0.0406], + device='cuda:3'), out_proj_covar=tensor([1.1939e-04, 1.0089e-04, 8.7397e-05, 1.1569e-04, 9.6622e-05, 1.4985e-04, + 1.0410e-04, 1.0853e-04], device='cuda:3') +2023-02-06 23:26:21,492 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3495, 1.5410, 2.1960, 1.2632, 1.6400, 1.6073, 1.3806, 1.6126], + device='cuda:3'), covar=tensor([0.1954, 0.2747, 0.0817, 0.4487, 0.1761, 0.3416, 0.2338, 0.2078], + device='cuda:3'), in_proj_covar=tensor([0.0515, 0.0582, 0.0552, 0.0630, 0.0639, 0.0588, 0.0522, 0.0628], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 23:26:22,613 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.573e+02 2.347e+02 2.658e+02 3.430e+02 7.604e+02, threshold=5.316e+02, percent-clipped=2.0 +2023-02-06 23:26:25,524 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.93 vs. limit=5.0 +2023-02-06 23:26:28,442 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-06 23:26:31,922 INFO [train.py:901] (3/4) Epoch 19, batch 5500, loss[loss=0.1978, simple_loss=0.2921, pruned_loss=0.05177, over 8749.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2929, pruned_loss=0.06617, over 1612352.20 frames. ], batch size: 40, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:26:33,480 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5029, 2.3387, 3.3298, 2.5584, 3.0782, 2.4826, 2.1823, 1.9220], + device='cuda:3'), covar=tensor([0.5085, 0.4842, 0.1625, 0.3465, 0.2349, 0.2863, 0.1890, 0.5207], + device='cuda:3'), in_proj_covar=tensor([0.0925, 0.0955, 0.0782, 0.0916, 0.0977, 0.0867, 0.0733, 0.0810], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 23:26:41,796 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151007.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:26:46,653 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151014.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:26:58,978 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151032.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:27:03,484 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2023-02-06 23:27:06,206 INFO [train.py:901] (3/4) Epoch 19, batch 5550, loss[loss=0.2303, simple_loss=0.3115, pruned_loss=0.07458, over 8587.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2928, pruned_loss=0.06609, over 1609315.19 frames. ], batch size: 31, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:27:16,585 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4901, 1.7577, 1.8457, 1.2219, 1.9088, 1.3217, 0.4638, 1.7428], + device='cuda:3'), covar=tensor([0.0540, 0.0351, 0.0287, 0.0510, 0.0417, 0.0911, 0.0832, 0.0238], + device='cuda:3'), in_proj_covar=tensor([0.0443, 0.0383, 0.0332, 0.0438, 0.0366, 0.0527, 0.0385, 0.0408], + device='cuda:3'), out_proj_covar=tensor([1.2002e-04, 1.0132e-04, 8.7803e-05, 1.1629e-04, 9.6974e-05, 1.5046e-04, + 1.0436e-04, 1.0897e-04], device='cuda:3') +2023-02-06 23:27:29,975 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6552, 2.2595, 1.6530, 3.6420, 1.7133, 1.4968, 2.2255, 2.5339], + device='cuda:3'), covar=tensor([0.1596, 0.1186, 0.2026, 0.0408, 0.1418, 0.2116, 0.1242, 0.0959], + device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0199, 0.0249, 0.0214, 0.0207, 0.0250, 0.0254, 0.0211], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 23:27:32,505 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.607e+02 2.454e+02 3.027e+02 4.195e+02 6.901e+02, threshold=6.054e+02, percent-clipped=7.0 +2023-02-06 23:27:42,415 INFO [train.py:901] (3/4) Epoch 19, batch 5600, loss[loss=0.2131, simple_loss=0.2934, pruned_loss=0.06642, over 8291.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2921, pruned_loss=0.06564, over 1609635.00 frames. ], batch size: 23, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:27:43,144 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=151094.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:28:15,312 INFO [train.py:901] (3/4) Epoch 19, batch 5650, loss[loss=0.2055, simple_loss=0.2732, pruned_loss=0.06888, over 7540.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.292, pruned_loss=0.06545, over 1610432.13 frames. ], batch size: 18, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:28:22,682 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 +2023-02-06 23:28:31,577 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-06 23:28:39,660 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.786e+02 2.685e+02 3.149e+02 3.866e+02 8.044e+02, threshold=6.298e+02, percent-clipped=3.0 +2023-02-06 23:28:50,395 INFO [train.py:901] (3/4) Epoch 19, batch 5700, loss[loss=0.1993, simple_loss=0.2806, pruned_loss=0.05896, over 8254.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.292, pruned_loss=0.06552, over 1614222.58 frames. ], batch size: 24, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:28:57,644 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151202.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:29:02,319 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151209.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:29:09,074 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0734, 2.4482, 2.5644, 1.6031, 2.7467, 1.8066, 1.6114, 2.0833], + device='cuda:3'), covar=tensor([0.0673, 0.0364, 0.0271, 0.0643, 0.0414, 0.0741, 0.0793, 0.0493], + device='cuda:3'), in_proj_covar=tensor([0.0444, 0.0383, 0.0332, 0.0439, 0.0368, 0.0527, 0.0385, 0.0408], + device='cuda:3'), out_proj_covar=tensor([1.2013e-04, 1.0119e-04, 8.7776e-05, 1.1647e-04, 9.7671e-05, 1.5051e-04, + 1.0412e-04, 1.0901e-04], device='cuda:3') +2023-02-06 23:29:15,623 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2880, 1.3546, 3.3717, 1.0406, 3.0103, 2.8033, 3.0875, 3.0009], + device='cuda:3'), covar=tensor([0.0754, 0.4118, 0.0776, 0.4073, 0.1300, 0.1087, 0.0726, 0.0835], + device='cuda:3'), in_proj_covar=tensor([0.0600, 0.0635, 0.0673, 0.0607, 0.0686, 0.0588, 0.0592, 0.0650], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 23:29:24,756 INFO [train.py:901] (3/4) Epoch 19, batch 5750, loss[loss=0.2081, simple_loss=0.2957, pruned_loss=0.06026, over 8105.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2916, pruned_loss=0.06517, over 1615994.65 frames. ], batch size: 23, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:29:30,822 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4839, 1.4968, 1.7963, 1.3779, 1.1574, 1.8473, 0.1957, 1.1724], + device='cuda:3'), covar=tensor([0.1950, 0.1412, 0.0430, 0.0975, 0.2825, 0.0477, 0.2080, 0.1249], + device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0194, 0.0123, 0.0221, 0.0268, 0.0133, 0.0168, 0.0188], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 23:29:36,097 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-06 23:29:37,495 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=151262.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:29:43,021 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151270.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:29:48,610 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.776e+02 2.389e+02 2.918e+02 3.727e+02 7.769e+02, threshold=5.836e+02, percent-clipped=3.0 +2023-02-06 23:29:58,863 INFO [train.py:901] (3/4) Epoch 19, batch 5800, loss[loss=0.1992, simple_loss=0.2654, pruned_loss=0.06648, over 7253.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2915, pruned_loss=0.06479, over 1614455.76 frames. ], batch size: 16, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:30:00,357 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151295.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:30:04,356 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151300.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:30:05,089 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6947, 2.2727, 4.0656, 1.4925, 3.0428, 2.2478, 1.8402, 2.8316], + device='cuda:3'), covar=tensor([0.2004, 0.2787, 0.0797, 0.4937, 0.1804, 0.3286, 0.2360, 0.2523], + device='cuda:3'), in_proj_covar=tensor([0.0513, 0.0580, 0.0552, 0.0629, 0.0637, 0.0587, 0.0521, 0.0626], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 23:30:16,599 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151316.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:30:34,588 INFO [train.py:901] (3/4) Epoch 19, batch 5850, loss[loss=0.2084, simple_loss=0.2942, pruned_loss=0.06129, over 8021.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2918, pruned_loss=0.06459, over 1613675.96 frames. ], batch size: 22, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:30:57,554 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151377.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:30:58,652 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.316e+02 2.176e+02 2.714e+02 3.221e+02 1.387e+03, threshold=5.429e+02, percent-clipped=3.0 +2023-02-06 23:31:08,064 INFO [train.py:901] (3/4) Epoch 19, batch 5900, loss[loss=0.1485, simple_loss=0.2272, pruned_loss=0.03485, over 7420.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2912, pruned_loss=0.06402, over 1612998.11 frames. ], batch size: 17, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:31:15,412 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-06 23:31:44,473 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0852, 2.4011, 4.5415, 1.7514, 3.3520, 2.5307, 2.1674, 3.2864], + device='cuda:3'), covar=tensor([0.1625, 0.2612, 0.0661, 0.4413, 0.1514, 0.2922, 0.2007, 0.2191], + device='cuda:3'), in_proj_covar=tensor([0.0516, 0.0583, 0.0555, 0.0632, 0.0640, 0.0590, 0.0523, 0.0629], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 23:31:44,905 INFO [train.py:901] (3/4) Epoch 19, batch 5950, loss[loss=0.1633, simple_loss=0.2384, pruned_loss=0.04413, over 7275.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2926, pruned_loss=0.06486, over 1619423.16 frames. ], batch size: 16, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:31:59,863 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151465.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:32:09,190 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.531e+02 2.424e+02 3.104e+02 3.851e+02 8.156e+02, threshold=6.208e+02, percent-clipped=3.0 +2023-02-06 23:32:16,814 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151490.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:32:18,557 INFO [train.py:901] (3/4) Epoch 19, batch 6000, loss[loss=0.2491, simple_loss=0.3209, pruned_loss=0.08866, over 8485.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2936, pruned_loss=0.06573, over 1619370.81 frames. ], batch size: 28, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:32:18,557 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 23:32:32,009 INFO [train.py:935] (3/4) Epoch 19, validation: loss=0.1763, simple_loss=0.2764, pruned_loss=0.03805, over 944034.00 frames. +2023-02-06 23:32:32,011 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 23:32:56,714 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5008, 1.7739, 1.8603, 1.2415, 1.9487, 1.3027, 0.5707, 1.7151], + device='cuda:3'), covar=tensor([0.0619, 0.0408, 0.0344, 0.0614, 0.0500, 0.1053, 0.0896, 0.0357], + device='cuda:3'), in_proj_covar=tensor([0.0444, 0.0380, 0.0332, 0.0439, 0.0368, 0.0527, 0.0386, 0.0406], + device='cuda:3'), out_proj_covar=tensor([1.2020e-04, 1.0041e-04, 8.7768e-05, 1.1665e-04, 9.7573e-05, 1.5037e-04, + 1.0445e-04, 1.0841e-04], device='cuda:3') +2023-02-06 23:33:00,162 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8227, 1.8076, 2.4950, 1.6195, 1.2856, 2.4314, 0.3938, 1.4154], + device='cuda:3'), covar=tensor([0.1735, 0.1215, 0.0270, 0.1443, 0.2930, 0.0406, 0.2317, 0.1361], + device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0193, 0.0123, 0.0221, 0.0268, 0.0133, 0.0168, 0.0187], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 23:33:06,942 INFO [train.py:901] (3/4) Epoch 19, batch 6050, loss[loss=0.2381, simple_loss=0.3069, pruned_loss=0.08462, over 7927.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.2947, pruned_loss=0.06641, over 1624047.43 frames. ], batch size: 20, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:33:09,101 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=151546.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:33:32,579 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.605e+02 2.535e+02 3.172e+02 3.888e+02 8.825e+02, threshold=6.343e+02, percent-clipped=4.0 +2023-02-06 23:33:42,767 INFO [train.py:901] (3/4) Epoch 19, batch 6100, loss[loss=0.1804, simple_loss=0.2768, pruned_loss=0.04205, over 8254.00 frames. ], tot_loss[loss=0.213, simple_loss=0.294, pruned_loss=0.06605, over 1622705.88 frames. ], batch size: 24, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:33:56,050 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1017, 1.2608, 4.5059, 1.8547, 2.4346, 5.1034, 5.1315, 4.4278], + device='cuda:3'), covar=tensor([0.1190, 0.1956, 0.0249, 0.1890, 0.1138, 0.0146, 0.0360, 0.0472], + device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0319, 0.0286, 0.0311, 0.0303, 0.0262, 0.0405, 0.0304], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 23:34:07,680 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-06 23:34:10,839 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151633.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:34:17,587 INFO [train.py:901] (3/4) Epoch 19, batch 6150, loss[loss=0.2103, simple_loss=0.2921, pruned_loss=0.0643, over 8502.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2924, pruned_loss=0.06545, over 1615756.65 frames. ], batch size: 26, lr: 3.97e-03, grad_scale: 16.0 +2023-02-06 23:34:18,379 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=151644.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:34:28,872 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151658.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:34:30,158 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=151660.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:34:30,981 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151661.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:34:43,579 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.691e+02 2.320e+02 2.846e+02 3.654e+02 5.745e+02, threshold=5.693e+02, percent-clipped=0.0 +2023-02-06 23:34:51,570 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 +2023-02-06 23:34:53,938 INFO [train.py:901] (3/4) Epoch 19, batch 6200, loss[loss=0.1996, simple_loss=0.2829, pruned_loss=0.05812, over 7922.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2917, pruned_loss=0.06512, over 1608167.51 frames. ], batch size: 20, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:35:02,696 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151706.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:35:28,537 INFO [train.py:901] (3/4) Epoch 19, batch 6250, loss[loss=0.2047, simple_loss=0.2882, pruned_loss=0.06062, over 8475.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2925, pruned_loss=0.06582, over 1611668.62 frames. ], batch size: 29, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:35:39,360 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151759.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:35:50,936 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151775.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:35:53,501 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.484e+02 2.555e+02 3.246e+02 4.070e+02 8.549e+02, threshold=6.492e+02, percent-clipped=6.0 +2023-02-06 23:36:03,717 INFO [train.py:901] (3/4) Epoch 19, batch 6300, loss[loss=0.2185, simple_loss=0.2986, pruned_loss=0.06921, over 8701.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2925, pruned_loss=0.06613, over 1612935.57 frames. ], batch size: 49, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:36:22,220 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151819.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:36:39,103 INFO [train.py:901] (3/4) Epoch 19, batch 6350, loss[loss=0.2329, simple_loss=0.3077, pruned_loss=0.07906, over 8034.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2915, pruned_loss=0.06549, over 1610357.65 frames. ], batch size: 22, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:37:03,141 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.433e+02 2.376e+02 2.921e+02 3.593e+02 6.855e+02, threshold=5.841e+02, percent-clipped=1.0 +2023-02-06 23:37:13,208 INFO [train.py:901] (3/4) Epoch 19, batch 6400, loss[loss=0.1769, simple_loss=0.2558, pruned_loss=0.04898, over 7642.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2912, pruned_loss=0.06497, over 1608784.87 frames. ], batch size: 19, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:37:24,581 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0174, 1.5874, 1.3011, 1.4862, 1.2773, 1.1557, 1.2316, 1.3217], + device='cuda:3'), covar=tensor([0.1088, 0.0458, 0.1304, 0.0533, 0.0755, 0.1455, 0.0891, 0.0723], + device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0234, 0.0327, 0.0303, 0.0298, 0.0330, 0.0343, 0.0314], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 23:37:30,644 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151917.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:37:30,810 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2023-02-06 23:37:38,746 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8901, 1.3243, 1.5513, 1.2633, 0.8955, 1.3404, 1.5247, 1.5225], + device='cuda:3'), covar=tensor([0.0487, 0.1338, 0.1752, 0.1507, 0.0636, 0.1583, 0.0740, 0.0660], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0189, 0.0158, 0.0100, 0.0161, 0.0112, 0.0141], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 23:37:40,462 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-06 23:37:42,194 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.17 vs. limit=2.0 +2023-02-06 23:37:48,037 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151942.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:37:48,526 INFO [train.py:901] (3/4) Epoch 19, batch 6450, loss[loss=0.1715, simple_loss=0.2456, pruned_loss=0.04869, over 7677.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2913, pruned_loss=0.06506, over 1606700.10 frames. ], batch size: 18, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:38:13,513 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.562e+02 2.409e+02 2.943e+02 3.710e+02 6.232e+02, threshold=5.887e+02, percent-clipped=1.0 +2023-02-06 23:38:15,120 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6127, 2.7310, 1.9602, 2.3459, 2.3164, 1.6397, 2.1324, 2.2429], + device='cuda:3'), covar=tensor([0.1603, 0.0456, 0.1239, 0.0632, 0.0731, 0.1593, 0.1038, 0.1043], + device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0235, 0.0329, 0.0305, 0.0300, 0.0333, 0.0345, 0.0316], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 23:38:23,080 INFO [train.py:901] (3/4) Epoch 19, batch 6500, loss[loss=0.1925, simple_loss=0.2827, pruned_loss=0.05113, over 8358.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2914, pruned_loss=0.06505, over 1608082.54 frames. ], batch size: 24, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:38:40,001 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152015.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:38:46,042 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8054, 3.7936, 3.4708, 1.7158, 3.3841, 3.4838, 3.4558, 3.2887], + device='cuda:3'), covar=tensor([0.0988, 0.0665, 0.1209, 0.5085, 0.1005, 0.1020, 0.1476, 0.0994], + device='cuda:3'), in_proj_covar=tensor([0.0518, 0.0429, 0.0429, 0.0532, 0.0419, 0.0432, 0.0412, 0.0375], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 23:38:51,604 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152031.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:38:58,598 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152040.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:39:00,457 INFO [train.py:901] (3/4) Epoch 19, batch 6550, loss[loss=0.2015, simple_loss=0.2785, pruned_loss=0.06224, over 8138.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2892, pruned_loss=0.06407, over 1607029.88 frames. ], batch size: 22, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:39:04,937 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=152050.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:39:09,220 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152056.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:39:21,598 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-06 23:39:24,906 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.678e+02 2.379e+02 2.761e+02 3.695e+02 7.678e+02, threshold=5.522e+02, percent-clipped=3.0 +2023-02-06 23:39:34,313 INFO [train.py:901] (3/4) Epoch 19, batch 6600, loss[loss=0.219, simple_loss=0.3036, pruned_loss=0.06724, over 8346.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2896, pruned_loss=0.06404, over 1608779.11 frames. ], batch size: 26, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:39:39,641 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-06 23:40:09,007 INFO [train.py:901] (3/4) Epoch 19, batch 6650, loss[loss=0.1891, simple_loss=0.2685, pruned_loss=0.05487, over 7807.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2913, pruned_loss=0.06511, over 1606713.48 frames. ], batch size: 20, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:40:23,462 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=152163.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:40:24,962 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152165.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:40:34,183 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.451e+02 2.686e+02 3.265e+02 3.895e+02 8.931e+02, threshold=6.531e+02, percent-clipped=7.0 +2023-02-06 23:40:44,526 INFO [train.py:901] (3/4) Epoch 19, batch 6700, loss[loss=0.1931, simple_loss=0.2804, pruned_loss=0.0529, over 8508.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2911, pruned_loss=0.06459, over 1612017.04 frames. ], batch size: 26, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:41:05,142 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6649, 2.2871, 4.1516, 1.4591, 3.0901, 2.1730, 1.8292, 2.9585], + device='cuda:3'), covar=tensor([0.1947, 0.2477, 0.0839, 0.4414, 0.1746, 0.3256, 0.2165, 0.2392], + device='cuda:3'), in_proj_covar=tensor([0.0513, 0.0581, 0.0548, 0.0627, 0.0637, 0.0586, 0.0521, 0.0627], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 23:41:06,355 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3877, 1.3513, 4.5270, 1.6582, 4.0043, 3.7586, 4.0803, 3.9779], + device='cuda:3'), covar=tensor([0.0570, 0.4981, 0.0579, 0.4145, 0.1180, 0.1035, 0.0607, 0.0679], + device='cuda:3'), in_proj_covar=tensor([0.0593, 0.0624, 0.0666, 0.0601, 0.0676, 0.0582, 0.0583, 0.0643], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 23:41:19,458 INFO [train.py:901] (3/4) Epoch 19, batch 6750, loss[loss=0.1926, simple_loss=0.2699, pruned_loss=0.05761, over 8080.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2919, pruned_loss=0.06493, over 1614567.35 frames. ], batch size: 21, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:41:19,600 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152243.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:41:44,696 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152278.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:41:45,124 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.467e+02 2.378e+02 2.909e+02 3.491e+02 6.752e+02, threshold=5.817e+02, percent-clipped=2.0 +2023-02-06 23:41:51,504 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5678, 1.4854, 1.9700, 1.3229, 1.1507, 1.9581, 0.4107, 1.3209], + device='cuda:3'), covar=tensor([0.1796, 0.1144, 0.0372, 0.1208, 0.2658, 0.0438, 0.2186, 0.1363], + device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0195, 0.0125, 0.0223, 0.0270, 0.0134, 0.0169, 0.0187], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 23:41:53,513 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152291.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:41:54,070 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-06 23:41:54,749 INFO [train.py:901] (3/4) Epoch 19, batch 6800, loss[loss=0.1883, simple_loss=0.281, pruned_loss=0.04776, over 8320.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2918, pruned_loss=0.06501, over 1612526.67 frames. ], batch size: 25, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:42:29,090 INFO [train.py:901] (3/4) Epoch 19, batch 6850, loss[loss=0.1896, simple_loss=0.2717, pruned_loss=0.05377, over 7817.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2912, pruned_loss=0.06478, over 1604740.66 frames. ], batch size: 20, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:42:43,967 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-06 23:42:54,742 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.676e+02 2.344e+02 3.012e+02 3.839e+02 8.073e+02, threshold=6.025e+02, percent-clipped=5.0 +2023-02-06 23:43:05,120 INFO [train.py:901] (3/4) Epoch 19, batch 6900, loss[loss=0.2211, simple_loss=0.2982, pruned_loss=0.07199, over 8125.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2906, pruned_loss=0.06436, over 1607382.38 frames. ], batch size: 22, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:43:25,517 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152421.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:43:40,387 INFO [train.py:901] (3/4) Epoch 19, batch 6950, loss[loss=0.1893, simple_loss=0.2719, pruned_loss=0.05334, over 8232.00 frames. ], tot_loss[loss=0.209, simple_loss=0.29, pruned_loss=0.06397, over 1608707.43 frames. ], batch size: 22, lr: 3.96e-03, grad_scale: 16.0 +2023-02-06 23:43:42,613 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152446.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:43:53,790 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-06 23:43:53,944 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152463.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:44:05,254 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.639e+02 2.443e+02 3.132e+02 3.706e+02 6.613e+02, threshold=6.264e+02, percent-clipped=2.0 +2023-02-06 23:44:14,624 INFO [train.py:901] (3/4) Epoch 19, batch 7000, loss[loss=0.2056, simple_loss=0.293, pruned_loss=0.05914, over 8504.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.289, pruned_loss=0.06369, over 1605634.99 frames. ], batch size: 26, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:44:28,407 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9092, 1.5322, 1.8257, 1.5153, 1.0233, 1.5049, 2.0638, 1.7146], + device='cuda:3'), covar=tensor([0.0441, 0.1262, 0.1669, 0.1423, 0.0594, 0.1513, 0.0635, 0.0655], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0191, 0.0158, 0.0100, 0.0162, 0.0113, 0.0142], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 23:44:44,333 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152534.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:44:51,112 INFO [train.py:901] (3/4) Epoch 19, batch 7050, loss[loss=0.2243, simple_loss=0.304, pruned_loss=0.07224, over 8190.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2891, pruned_loss=0.06386, over 1608327.72 frames. ], batch size: 23, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:45:02,290 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152559.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:45:15,723 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.757e+02 2.420e+02 2.800e+02 3.429e+02 5.549e+02, threshold=5.599e+02, percent-clipped=0.0 +2023-02-06 23:45:21,283 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=152587.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:45:25,396 INFO [train.py:901] (3/4) Epoch 19, batch 7100, loss[loss=0.1903, simple_loss=0.2681, pruned_loss=0.05628, over 7916.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2875, pruned_loss=0.0629, over 1604124.05 frames. ], batch size: 20, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:45:30,679 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152600.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:45:35,449 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152607.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:45:56,367 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=152635.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:46:01,876 INFO [train.py:901] (3/4) Epoch 19, batch 7150, loss[loss=0.2231, simple_loss=0.299, pruned_loss=0.07363, over 8626.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.289, pruned_loss=0.0641, over 1604817.43 frames. ], batch size: 39, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:46:27,178 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.485e+02 2.441e+02 2.885e+02 3.630e+02 1.043e+03, threshold=5.770e+02, percent-clipped=5.0 +2023-02-06 23:46:36,620 INFO [train.py:901] (3/4) Epoch 19, batch 7200, loss[loss=0.1841, simple_loss=0.261, pruned_loss=0.05356, over 7533.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2895, pruned_loss=0.06434, over 1604625.17 frames. ], batch size: 18, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:46:42,732 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152702.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:47:09,163 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6712, 1.5737, 2.8550, 1.3457, 2.1168, 3.0425, 3.1525, 2.5893], + device='cuda:3'), covar=tensor([0.1183, 0.1445, 0.0402, 0.2116, 0.0955, 0.0305, 0.0679, 0.0639], + device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0320, 0.0285, 0.0313, 0.0301, 0.0262, 0.0408, 0.0306], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 23:47:12,599 INFO [train.py:901] (3/4) Epoch 19, batch 7250, loss[loss=0.2, simple_loss=0.2794, pruned_loss=0.06032, over 8089.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2911, pruned_loss=0.06492, over 1610322.06 frames. ], batch size: 21, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:47:13,461 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152744.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:47:15,662 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3127, 1.9815, 2.7338, 2.2167, 2.7024, 2.2445, 2.0573, 1.3745], + device='cuda:3'), covar=tensor([0.5213, 0.4690, 0.1799, 0.3520, 0.2372, 0.3154, 0.1942, 0.5250], + device='cuda:3'), in_proj_covar=tensor([0.0929, 0.0957, 0.0783, 0.0923, 0.0981, 0.0872, 0.0740, 0.0817], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 23:47:17,701 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152750.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:47:37,386 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.468e+02 2.392e+02 2.877e+02 3.488e+02 7.359e+02, threshold=5.753e+02, percent-clipped=2.0 +2023-02-06 23:47:47,610 INFO [train.py:901] (3/4) Epoch 19, batch 7300, loss[loss=0.1788, simple_loss=0.2584, pruned_loss=0.04956, over 7974.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2906, pruned_loss=0.06475, over 1613693.90 frames. ], batch size: 21, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:47:57,329 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=152807.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:48:04,701 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0741, 2.3895, 1.9881, 2.8882, 1.4799, 1.6284, 2.0449, 2.3649], + device='cuda:3'), covar=tensor([0.0741, 0.0748, 0.0846, 0.0341, 0.1119, 0.1328, 0.0903, 0.0733], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0196, 0.0245, 0.0210, 0.0203, 0.0246, 0.0249, 0.0207], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 23:48:21,875 INFO [train.py:901] (3/4) Epoch 19, batch 7350, loss[loss=0.188, simple_loss=0.276, pruned_loss=0.05003, over 8247.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2902, pruned_loss=0.06435, over 1616101.88 frames. ], batch size: 22, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:48:46,701 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-06 23:48:48,161 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.657e+02 2.571e+02 3.070e+02 4.184e+02 8.940e+02, threshold=6.140e+02, percent-clipped=8.0 +2023-02-06 23:48:58,049 INFO [train.py:901] (3/4) Epoch 19, batch 7400, loss[loss=0.2123, simple_loss=0.2834, pruned_loss=0.07056, over 7795.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2887, pruned_loss=0.06385, over 1611507.93 frames. ], batch size: 19, lr: 3.95e-03, grad_scale: 32.0 +2023-02-06 23:49:07,696 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-06 23:49:18,789 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152922.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:49:32,918 INFO [train.py:901] (3/4) Epoch 19, batch 7450, loss[loss=0.199, simple_loss=0.281, pruned_loss=0.05848, over 7799.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2881, pruned_loss=0.06384, over 1607973.23 frames. ], batch size: 19, lr: 3.95e-03, grad_scale: 32.0 +2023-02-06 23:49:33,722 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=152944.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:49:38,522 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=152951.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:49:44,011 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152958.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:49:46,568 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-06 23:49:47,483 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4104, 2.1987, 3.2063, 2.4499, 2.9260, 2.4812, 2.1511, 1.6732], + device='cuda:3'), covar=tensor([0.5327, 0.5009, 0.1788, 0.3706, 0.2677, 0.2840, 0.1897, 0.5341], + device='cuda:3'), in_proj_covar=tensor([0.0924, 0.0953, 0.0779, 0.0917, 0.0979, 0.0867, 0.0736, 0.0811], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 23:49:58,925 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.655e+02 2.506e+02 3.079e+02 4.075e+02 8.166e+02, threshold=6.159e+02, percent-clipped=5.0 +2023-02-06 23:50:01,984 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152983.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:50:08,330 INFO [train.py:901] (3/4) Epoch 19, batch 7500, loss[loss=0.2245, simple_loss=0.3171, pruned_loss=0.06597, over 8311.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2891, pruned_loss=0.06381, over 1610714.45 frames. ], batch size: 25, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:50:11,888 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7684, 1.5907, 3.1615, 1.4241, 2.1959, 3.3799, 3.5115, 2.9237], + device='cuda:3'), covar=tensor([0.1188, 0.1565, 0.0386, 0.2056, 0.1030, 0.0279, 0.0651, 0.0601], + device='cuda:3'), in_proj_covar=tensor([0.0293, 0.0320, 0.0287, 0.0313, 0.0304, 0.0263, 0.0409, 0.0306], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-06 23:50:17,498 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153006.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:50:21,635 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6316, 1.3778, 1.5702, 1.2564, 0.8745, 1.3735, 1.5042, 1.2349], + device='cuda:3'), covar=tensor([0.0570, 0.1236, 0.1703, 0.1479, 0.0649, 0.1445, 0.0778, 0.0701], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0189, 0.0158, 0.0099, 0.0160, 0.0112, 0.0142], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 23:50:34,853 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153031.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:50:42,701 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 +2023-02-06 23:50:42,933 INFO [train.py:901] (3/4) Epoch 19, batch 7550, loss[loss=0.1879, simple_loss=0.2794, pruned_loss=0.04818, over 7973.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2899, pruned_loss=0.06415, over 1611207.27 frames. ], batch size: 21, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:50:48,594 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4107, 1.4808, 1.3583, 1.8849, 0.8122, 1.2555, 1.3554, 1.4377], + device='cuda:3'), covar=tensor([0.0874, 0.0742, 0.1006, 0.0458, 0.1021, 0.1308, 0.0657, 0.0757], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0196, 0.0246, 0.0211, 0.0203, 0.0246, 0.0250, 0.0207], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-06 23:50:49,461 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.25 vs. limit=5.0 +2023-02-06 23:50:53,914 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153059.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:50:58,700 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153066.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:51:08,483 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.670e+02 2.431e+02 2.980e+02 3.688e+02 7.634e+02, threshold=5.960e+02, percent-clipped=2.0 +2023-02-06 23:51:14,777 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=153088.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:51:15,125 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 +2023-02-06 23:51:18,065 INFO [train.py:901] (3/4) Epoch 19, batch 7600, loss[loss=0.2309, simple_loss=0.3173, pruned_loss=0.07226, over 8105.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2907, pruned_loss=0.06479, over 1610690.60 frames. ], batch size: 23, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:51:28,505 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.9987, 1.6570, 6.0708, 2.1418, 5.5534, 5.0436, 5.5855, 5.5511], + device='cuda:3'), covar=tensor([0.0511, 0.4754, 0.0400, 0.4062, 0.0969, 0.0888, 0.0476, 0.0505], + device='cuda:3'), in_proj_covar=tensor([0.0594, 0.0624, 0.0665, 0.0599, 0.0679, 0.0584, 0.0581, 0.0642], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002], + device='cuda:3') +2023-02-06 23:51:47,774 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153135.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:51:53,092 INFO [train.py:901] (3/4) Epoch 19, batch 7650, loss[loss=0.2186, simple_loss=0.3062, pruned_loss=0.06548, over 8334.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.291, pruned_loss=0.06473, over 1615626.83 frames. ], batch size: 48, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:52:08,632 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2023-02-06 23:52:17,641 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153178.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:52:18,733 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.613e+02 2.290e+02 2.780e+02 3.362e+02 7.829e+02, threshold=5.561e+02, percent-clipped=2.0 +2023-02-06 23:52:27,555 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-06 23:52:28,394 INFO [train.py:901] (3/4) Epoch 19, batch 7700, loss[loss=0.1753, simple_loss=0.2677, pruned_loss=0.04145, over 8290.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2897, pruned_loss=0.06413, over 1609928.12 frames. ], batch size: 23, lr: 3.95e-03, grad_scale: 16.0 +2023-02-06 23:52:35,489 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153203.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:52:35,511 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153203.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:52:35,889 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-06 23:52:57,462 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-06 23:53:03,345 INFO [train.py:901] (3/4) Epoch 19, batch 7750, loss[loss=0.2112, simple_loss=0.2989, pruned_loss=0.06179, over 8325.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2913, pruned_loss=0.06517, over 1613185.06 frames. ], batch size: 25, lr: 3.94e-03, grad_scale: 16.0 +2023-02-06 23:53:28,922 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.451e+02 2.456e+02 3.001e+02 3.725e+02 8.940e+02, threshold=6.003e+02, percent-clipped=11.0 +2023-02-06 23:53:37,741 INFO [train.py:901] (3/4) Epoch 19, batch 7800, loss[loss=0.2008, simple_loss=0.2906, pruned_loss=0.05545, over 8136.00 frames. ], tot_loss[loss=0.2111, simple_loss=0.2917, pruned_loss=0.0653, over 1613937.43 frames. ], batch size: 22, lr: 3.94e-03, grad_scale: 16.0 +2023-02-06 23:53:39,918 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153296.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:53:53,366 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153315.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:53:58,001 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153322.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:54:09,679 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153340.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:54:11,469 INFO [train.py:901] (3/4) Epoch 19, batch 7850, loss[loss=0.2025, simple_loss=0.2879, pruned_loss=0.0585, over 8700.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2907, pruned_loss=0.06463, over 1609239.98 frames. ], batch size: 34, lr: 3.94e-03, grad_scale: 8.0 +2023-02-06 23:54:14,341 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6786, 1.3891, 1.5600, 1.3347, 0.8897, 1.3394, 1.5158, 1.2030], + device='cuda:3'), covar=tensor([0.0587, 0.1248, 0.1698, 0.1486, 0.0624, 0.1480, 0.0739, 0.0728], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0153, 0.0191, 0.0159, 0.0100, 0.0162, 0.0113, 0.0143], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-06 23:54:14,372 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153347.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:54:16,434 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1569, 1.2967, 1.3224, 0.9677, 1.3537, 1.0379, 0.3955, 1.2780], + device='cuda:3'), covar=tensor([0.0357, 0.0250, 0.0219, 0.0356, 0.0288, 0.0622, 0.0623, 0.0208], + device='cuda:3'), in_proj_covar=tensor([0.0439, 0.0379, 0.0335, 0.0440, 0.0366, 0.0527, 0.0386, 0.0407], + device='cuda:3'), out_proj_covar=tensor([1.1879e-04, 9.9666e-05, 8.8673e-05, 1.1697e-04, 9.6864e-05, 1.5016e-04, + 1.0446e-04, 1.0865e-04], device='cuda:3') +2023-02-06 23:54:23,027 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7282, 2.0569, 2.2579, 1.4490, 2.3580, 1.5666, 0.6758, 1.8817], + device='cuda:3'), covar=tensor([0.0581, 0.0316, 0.0216, 0.0518, 0.0335, 0.0824, 0.0814, 0.0279], + device='cuda:3'), in_proj_covar=tensor([0.0440, 0.0379, 0.0335, 0.0440, 0.0366, 0.0527, 0.0386, 0.0407], + device='cuda:3'), out_proj_covar=tensor([1.1887e-04, 9.9728e-05, 8.8735e-05, 1.1695e-04, 9.6969e-05, 1.5010e-04, + 1.0448e-04, 1.0863e-04], device='cuda:3') +2023-02-06 23:54:36,624 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.498e+02 2.457e+02 2.874e+02 3.581e+02 1.670e+03, threshold=5.749e+02, percent-clipped=9.0 +2023-02-06 23:54:44,306 INFO [train.py:901] (3/4) Epoch 19, batch 7900, loss[loss=0.2463, simple_loss=0.3218, pruned_loss=0.08538, over 8504.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2902, pruned_loss=0.06455, over 1608476.92 frames. ], batch size: 26, lr: 3.94e-03, grad_scale: 8.0 +2023-02-06 23:55:15,520 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153439.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:55:17,981 INFO [train.py:901] (3/4) Epoch 19, batch 7950, loss[loss=0.2276, simple_loss=0.3058, pruned_loss=0.07472, over 7184.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2911, pruned_loss=0.06487, over 1613113.28 frames. ], batch size: 71, lr: 3.94e-03, grad_scale: 8.0 +2023-02-06 23:55:28,823 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153459.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:55:41,876 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=153479.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:55:43,176 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.397e+02 2.434e+02 3.034e+02 3.983e+02 8.510e+02, threshold=6.068e+02, percent-clipped=6.0 +2023-02-06 23:55:45,345 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153484.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:55:50,998 INFO [train.py:901] (3/4) Epoch 19, batch 8000, loss[loss=0.2261, simple_loss=0.3008, pruned_loss=0.0757, over 7536.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.2919, pruned_loss=0.06545, over 1611815.00 frames. ], batch size: 18, lr: 3.94e-03, grad_scale: 8.0 +2023-02-06 23:56:10,562 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 +2023-02-06 23:56:25,153 INFO [train.py:901] (3/4) Epoch 19, batch 8050, loss[loss=0.223, simple_loss=0.2874, pruned_loss=0.07934, over 7423.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2903, pruned_loss=0.06468, over 1603997.26 frames. ], batch size: 17, lr: 3.94e-03, grad_scale: 8.0 +2023-02-06 23:56:58,871 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-06 23:57:04,940 INFO [train.py:901] (3/4) Epoch 20, batch 0, loss[loss=0.2154, simple_loss=0.2806, pruned_loss=0.0751, over 7446.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2806, pruned_loss=0.0751, over 7446.00 frames. ], batch size: 17, lr: 3.84e-03, grad_scale: 8.0 +2023-02-06 23:57:04,940 INFO [train.py:926] (3/4) Computing validation loss +2023-02-06 23:57:16,940 INFO [train.py:935] (3/4) Epoch 20, validation: loss=0.1757, simple_loss=0.276, pruned_loss=0.03766, over 944034.00 frames. +2023-02-06 23:57:16,940 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-06 23:57:20,455 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.751e+02 2.577e+02 3.496e+02 4.495e+02 1.164e+03, threshold=6.992e+02, percent-clipped=12.0 +2023-02-06 23:57:29,438 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153594.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:57:31,325 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-06 23:57:51,322 INFO [train.py:901] (3/4) Epoch 20, batch 50, loss[loss=0.1987, simple_loss=0.2669, pruned_loss=0.06527, over 7789.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2985, pruned_loss=0.06843, over 366435.38 frames. ], batch size: 19, lr: 3.84e-03, grad_scale: 8.0 +2023-02-06 23:58:01,107 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=153640.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:58:06,579 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-06 23:58:27,829 INFO [train.py:901] (3/4) Epoch 20, batch 100, loss[loss=0.225, simple_loss=0.311, pruned_loss=0.06955, over 8254.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2962, pruned_loss=0.06632, over 649833.87 frames. ], batch size: 24, lr: 3.84e-03, grad_scale: 8.0 +2023-02-06 23:58:29,262 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-06 23:58:31,356 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.631e+02 2.446e+02 2.844e+02 3.351e+02 7.473e+02, threshold=5.688e+02, percent-clipped=2.0 +2023-02-06 23:58:58,603 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3320, 1.3825, 4.5247, 1.8005, 3.9787, 3.8102, 4.0744, 3.9831], + device='cuda:3'), covar=tensor([0.0594, 0.5304, 0.0517, 0.4069, 0.1120, 0.0916, 0.0576, 0.0619], + device='cuda:3'), in_proj_covar=tensor([0.0607, 0.0639, 0.0679, 0.0613, 0.0691, 0.0595, 0.0593, 0.0655], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-06 23:59:03,142 INFO [train.py:901] (3/4) Epoch 20, batch 150, loss[loss=0.2342, simple_loss=0.3215, pruned_loss=0.07342, over 8256.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2955, pruned_loss=0.06541, over 868139.31 frames. ], batch size: 24, lr: 3.84e-03, grad_scale: 8.0 +2023-02-06 23:59:06,136 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9614, 2.4661, 3.7082, 2.1013, 1.8262, 3.6384, 0.5502, 2.1793], + device='cuda:3'), covar=tensor([0.1285, 0.1245, 0.0222, 0.1611, 0.2881, 0.0220, 0.2658, 0.1365], + device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0194, 0.0124, 0.0221, 0.0270, 0.0133, 0.0169, 0.0186], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-06 23:59:23,320 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153755.0, num_to_drop=0, layers_to_drop=set() +2023-02-06 23:59:39,296 INFO [train.py:901] (3/4) Epoch 20, batch 200, loss[loss=0.2259, simple_loss=0.3048, pruned_loss=0.07345, over 8493.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2956, pruned_loss=0.06509, over 1039247.45 frames. ], batch size: 29, lr: 3.84e-03, grad_scale: 8.0 +2023-02-06 23:59:42,529 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.435e+02 2.177e+02 2.784e+02 3.416e+02 8.818e+02, threshold=5.569e+02, percent-clipped=1.0 +2023-02-06 23:59:43,938 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=153783.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:00:15,037 INFO [train.py:901] (3/4) Epoch 20, batch 250, loss[loss=0.2223, simple_loss=0.3001, pruned_loss=0.07227, over 8082.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2958, pruned_loss=0.06487, over 1172136.61 frames. ], batch size: 21, lr: 3.84e-03, grad_scale: 8.0 +2023-02-07 00:00:26,530 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-07 00:00:31,631 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153850.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:00:34,735 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-07 00:00:48,275 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153875.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:00:48,775 INFO [train.py:901] (3/4) Epoch 20, batch 300, loss[loss=0.2158, simple_loss=0.3042, pruned_loss=0.0637, over 8246.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2942, pruned_loss=0.06468, over 1273559.96 frames. ], batch size: 24, lr: 3.84e-03, grad_scale: 8.0 +2023-02-07 00:00:52,004 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.429e+02 2.425e+02 2.846e+02 3.739e+02 1.062e+03, threshold=5.691e+02, percent-clipped=2.0 +2023-02-07 00:01:05,174 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153898.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:01:24,550 INFO [train.py:901] (3/4) Epoch 20, batch 350, loss[loss=0.2022, simple_loss=0.2781, pruned_loss=0.06311, over 8082.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.2932, pruned_loss=0.06478, over 1349212.59 frames. ], batch size: 21, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:01:35,761 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153941.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:01:36,640 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-07 00:01:59,290 INFO [train.py:901] (3/4) Epoch 20, batch 400, loss[loss=0.2044, simple_loss=0.291, pruned_loss=0.05886, over 8182.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2933, pruned_loss=0.06433, over 1408883.80 frames. ], batch size: 23, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:02:02,798 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.367e+02 2.483e+02 2.937e+02 3.652e+02 9.410e+02, threshold=5.874e+02, percent-clipped=4.0 +2023-02-07 00:02:22,963 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-07 00:02:25,627 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154011.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:02:36,305 INFO [train.py:901] (3/4) Epoch 20, batch 450, loss[loss=0.2553, simple_loss=0.3213, pruned_loss=0.09462, over 8680.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2926, pruned_loss=0.06407, over 1455860.17 frames. ], batch size: 34, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:02:44,052 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154036.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:03:05,786 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154067.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:03:11,785 INFO [train.py:901] (3/4) Epoch 20, batch 500, loss[loss=0.1947, simple_loss=0.2773, pruned_loss=0.05602, over 7811.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2888, pruned_loss=0.06239, over 1487931.49 frames. ], batch size: 20, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:03:15,235 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.123e+02 2.274e+02 2.685e+02 3.204e+02 7.760e+02, threshold=5.371e+02, percent-clipped=3.0 +2023-02-07 00:03:24,423 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154094.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:03:29,962 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154102.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:03:46,377 INFO [train.py:901] (3/4) Epoch 20, batch 550, loss[loss=0.2271, simple_loss=0.3155, pruned_loss=0.06932, over 8461.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2877, pruned_loss=0.06244, over 1509784.18 frames. ], batch size: 29, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:03:49,724 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 +2023-02-07 00:03:59,235 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.57 vs. limit=2.0 +2023-02-07 00:04:07,821 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154154.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:04:23,278 INFO [train.py:901] (3/4) Epoch 20, batch 600, loss[loss=0.2084, simple_loss=0.2933, pruned_loss=0.0617, over 8400.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2898, pruned_loss=0.06414, over 1536219.00 frames. ], batch size: 29, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:04:25,538 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154179.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:04:26,655 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.715e+02 2.477e+02 2.962e+02 3.836e+02 8.919e+02, threshold=5.925e+02, percent-clipped=6.0 +2023-02-07 00:04:45,284 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-07 00:04:57,543 INFO [train.py:901] (3/4) Epoch 20, batch 650, loss[loss=0.1856, simple_loss=0.2621, pruned_loss=0.0545, over 7799.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2896, pruned_loss=0.06396, over 1551230.52 frames. ], batch size: 20, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:04:58,453 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4708, 2.6731, 1.8885, 2.3777, 2.2222, 1.6098, 2.0473, 2.3061], + device='cuda:3'), covar=tensor([0.1608, 0.0399, 0.1130, 0.0706, 0.0759, 0.1456, 0.1112, 0.0977], + device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0237, 0.0332, 0.0307, 0.0302, 0.0335, 0.0346, 0.0319], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 00:05:06,566 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154239.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:05:14,622 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 +2023-02-07 00:05:34,074 INFO [train.py:901] (3/4) Epoch 20, batch 700, loss[loss=0.2528, simple_loss=0.3245, pruned_loss=0.0906, over 8732.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2895, pruned_loss=0.06372, over 1566244.48 frames. ], batch size: 30, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:05:37,465 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.449e+02 2.357e+02 2.958e+02 3.586e+02 6.466e+02, threshold=5.915e+02, percent-clipped=2.0 +2023-02-07 00:05:40,241 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154285.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:05:54,040 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154304.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:05:57,548 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154309.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:06:08,843 INFO [train.py:901] (3/4) Epoch 20, batch 750, loss[loss=0.2017, simple_loss=0.2888, pruned_loss=0.05726, over 8458.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2898, pruned_loss=0.06382, over 1575207.54 frames. ], batch size: 27, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:06:11,930 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-07 00:06:28,503 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154355.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:06:33,651 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-07 00:06:43,001 INFO [train.py:901] (3/4) Epoch 20, batch 800, loss[loss=0.2018, simple_loss=0.2816, pruned_loss=0.06095, over 8088.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2897, pruned_loss=0.06385, over 1582884.73 frames. ], batch size: 21, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:06:43,010 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-07 00:06:47,166 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.337e+02 2.441e+02 3.052e+02 3.711e+02 8.675e+02, threshold=6.104e+02, percent-clipped=3.0 +2023-02-07 00:07:01,166 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154400.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:07:08,302 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154411.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:07:19,181 INFO [train.py:901] (3/4) Epoch 20, batch 850, loss[loss=0.199, simple_loss=0.2773, pruned_loss=0.06029, over 7787.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2891, pruned_loss=0.06345, over 1586278.73 frames. ], batch size: 19, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:07:27,239 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154438.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:07:32,660 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154446.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:07:52,873 INFO [train.py:901] (3/4) Epoch 20, batch 900, loss[loss=0.192, simple_loss=0.2794, pruned_loss=0.0523, over 8343.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.29, pruned_loss=0.06393, over 1594485.43 frames. ], batch size: 25, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:07:56,205 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.768e+02 2.439e+02 2.923e+02 3.686e+02 1.072e+03, threshold=5.846e+02, percent-clipped=2.0 +2023-02-07 00:07:58,423 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3951, 1.5458, 1.4165, 1.8346, 0.7521, 1.2239, 1.2722, 1.5190], + device='cuda:3'), covar=tensor([0.0878, 0.0710, 0.0987, 0.0494, 0.1106, 0.1407, 0.0773, 0.0668], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0195, 0.0246, 0.0209, 0.0205, 0.0248, 0.0249, 0.0208], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 00:07:59,847 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5339, 1.7275, 2.7442, 1.3907, 2.0994, 1.9140, 1.5445, 2.0028], + device='cuda:3'), covar=tensor([0.1951, 0.2584, 0.0841, 0.4663, 0.1731, 0.3158, 0.2455, 0.2186], + device='cuda:3'), in_proj_covar=tensor([0.0519, 0.0586, 0.0553, 0.0631, 0.0639, 0.0587, 0.0524, 0.0630], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 00:08:04,402 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154492.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:08:29,153 INFO [train.py:901] (3/4) Epoch 20, batch 950, loss[loss=0.2368, simple_loss=0.3057, pruned_loss=0.08393, over 7203.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2903, pruned_loss=0.06447, over 1596793.48 frames. ], batch size: 71, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:08:29,366 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154526.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:08:48,746 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154553.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:08:54,161 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154561.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:09:00,292 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2569, 1.6401, 4.1327, 1.9265, 2.3267, 4.7160, 4.8614, 4.0594], + device='cuda:3'), covar=tensor([0.1171, 0.1962, 0.0301, 0.2030, 0.1389, 0.0192, 0.0339, 0.0557], + device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0318, 0.0286, 0.0310, 0.0302, 0.0259, 0.0403, 0.0303], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 00:09:01,512 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-07 00:09:04,234 INFO [train.py:901] (3/4) Epoch 20, batch 1000, loss[loss=0.2647, simple_loss=0.3293, pruned_loss=0.1, over 7463.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2904, pruned_loss=0.06453, over 1597772.70 frames. ], batch size: 73, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:09:07,495 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.592e+02 2.511e+02 3.044e+02 3.807e+02 8.767e+02, threshold=6.087e+02, percent-clipped=2.0 +2023-02-07 00:09:08,984 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154583.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:09:23,942 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3446, 4.3802, 3.9111, 2.0228, 3.7587, 3.9291, 3.9340, 3.7252], + device='cuda:3'), covar=tensor([0.0819, 0.0637, 0.1137, 0.4818, 0.0963, 0.1265, 0.1458, 0.1007], + device='cuda:3'), in_proj_covar=tensor([0.0522, 0.0429, 0.0433, 0.0537, 0.0423, 0.0437, 0.0417, 0.0374], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 00:09:35,153 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-07 00:09:38,960 INFO [train.py:901] (3/4) Epoch 20, batch 1050, loss[loss=0.1984, simple_loss=0.2912, pruned_loss=0.05282, over 8439.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2915, pruned_loss=0.06528, over 1602402.06 frames. ], batch size: 27, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:09:49,447 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-07 00:09:53,678 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154646.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:09:54,882 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154648.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:09:59,106 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154653.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:10:01,446 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154656.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:10:14,711 INFO [train.py:901] (3/4) Epoch 20, batch 1100, loss[loss=0.2099, simple_loss=0.2997, pruned_loss=0.06003, over 8737.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2913, pruned_loss=0.06529, over 1606832.81 frames. ], batch size: 30, lr: 3.83e-03, grad_scale: 8.0 +2023-02-07 00:10:18,092 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.678e+02 2.486e+02 3.103e+02 3.988e+02 8.246e+02, threshold=6.206e+02, percent-clipped=6.0 +2023-02-07 00:10:18,345 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154681.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:10:29,751 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154698.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:10:30,331 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154699.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 00:10:48,868 INFO [train.py:901] (3/4) Epoch 20, batch 1150, loss[loss=0.224, simple_loss=0.3167, pruned_loss=0.06562, over 8193.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2917, pruned_loss=0.06603, over 1608305.58 frames. ], batch size: 23, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:10:53,708 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0483, 1.5382, 1.7031, 1.3993, 1.0009, 1.5040, 1.8686, 1.6838], + device='cuda:3'), covar=tensor([0.0518, 0.1188, 0.1570, 0.1401, 0.0594, 0.1353, 0.0667, 0.0598], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0151, 0.0189, 0.0157, 0.0099, 0.0160, 0.0112, 0.0141], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 00:10:57,844 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154738.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:10:59,082 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-07 00:11:16,274 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154763.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:11:19,807 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154768.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:11:25,038 INFO [train.py:901] (3/4) Epoch 20, batch 1200, loss[loss=0.2369, simple_loss=0.3181, pruned_loss=0.07781, over 8485.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2912, pruned_loss=0.06531, over 1610265.62 frames. ], batch size: 28, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:11:28,377 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.662e+02 2.412e+02 2.746e+02 3.577e+02 9.067e+02, threshold=5.492e+02, percent-clipped=2.0 +2023-02-07 00:11:29,299 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154782.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 00:11:46,348 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154807.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:11:47,770 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154809.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:11:51,162 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154814.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:11:53,284 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154817.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:11:59,126 INFO [train.py:901] (3/4) Epoch 20, batch 1250, loss[loss=0.1698, simple_loss=0.248, pruned_loss=0.04577, over 7696.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2908, pruned_loss=0.06496, over 1610108.88 frames. ], batch size: 18, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:12:05,337 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154834.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:12:06,467 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154836.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:12:11,371 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154842.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:12:22,329 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9034, 1.8693, 2.9553, 2.3275, 2.6816, 1.9321, 1.6404, 1.3753], + device='cuda:3'), covar=tensor([0.6783, 0.5662, 0.1839, 0.3646, 0.2719, 0.4138, 0.2959, 0.5465], + device='cuda:3'), in_proj_covar=tensor([0.0939, 0.0966, 0.0791, 0.0928, 0.0984, 0.0877, 0.0740, 0.0819], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 00:12:23,576 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1498, 1.5871, 4.3534, 1.6900, 3.8801, 3.6620, 3.9387, 3.8138], + device='cuda:3'), covar=tensor([0.0602, 0.4242, 0.0507, 0.3760, 0.1069, 0.0895, 0.0564, 0.0686], + device='cuda:3'), in_proj_covar=tensor([0.0599, 0.0621, 0.0673, 0.0599, 0.0679, 0.0587, 0.0586, 0.0651], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 00:12:34,985 INFO [train.py:901] (3/4) Epoch 20, batch 1300, loss[loss=0.1782, simple_loss=0.2615, pruned_loss=0.04744, over 7426.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2908, pruned_loss=0.06461, over 1612282.52 frames. ], batch size: 17, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:12:38,324 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.654e+02 2.433e+02 3.191e+02 3.995e+02 7.235e+02, threshold=6.381e+02, percent-clipped=6.0 +2023-02-07 00:13:09,379 INFO [train.py:901] (3/4) Epoch 20, batch 1350, loss[loss=0.1991, simple_loss=0.2939, pruned_loss=0.0521, over 8504.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2896, pruned_loss=0.0639, over 1612831.96 frames. ], batch size: 26, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:13:27,087 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154951.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:13:29,755 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154954.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:13:44,748 INFO [train.py:901] (3/4) Epoch 20, batch 1400, loss[loss=0.1903, simple_loss=0.2735, pruned_loss=0.05351, over 8028.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2884, pruned_loss=0.06281, over 1611927.68 frames. ], batch size: 22, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:13:47,803 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154979.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:13:48,960 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.658e+02 2.419e+02 2.969e+02 3.620e+02 8.609e+02, threshold=5.938e+02, percent-clipped=3.0 +2023-02-07 00:13:55,293 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=154990.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:13:58,160 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154994.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:14:16,373 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155019.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:14:19,579 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155024.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:14:20,734 INFO [train.py:901] (3/4) Epoch 20, batch 1450, loss[loss=0.1916, simple_loss=0.2787, pruned_loss=0.05223, over 8250.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2903, pruned_loss=0.0643, over 1611719.35 frames. ], batch size: 24, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:14:29,167 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-07 00:14:33,296 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155044.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:14:36,500 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155049.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:14:50,892 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155070.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:14:55,250 INFO [train.py:901] (3/4) Epoch 20, batch 1500, loss[loss=0.182, simple_loss=0.2705, pruned_loss=0.04677, over 8289.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2908, pruned_loss=0.06436, over 1614771.51 frames. ], batch size: 23, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:14:58,584 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.504e+02 2.482e+02 3.072e+02 3.822e+02 6.990e+02, threshold=6.143e+02, percent-clipped=2.0 +2023-02-07 00:14:59,297 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=155082.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:15:02,793 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155087.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:15:08,997 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155095.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 00:15:09,714 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.31 vs. limit=5.0 +2023-02-07 00:15:15,567 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155105.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:15:30,486 INFO [train.py:901] (3/4) Epoch 20, batch 1550, loss[loss=0.2331, simple_loss=0.3131, pruned_loss=0.07655, over 8335.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2906, pruned_loss=0.06392, over 1616293.98 frames. ], batch size: 25, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:15:36,915 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2490, 1.4299, 1.5798, 1.2905, 1.0394, 1.3399, 1.8369, 1.5547], + device='cuda:3'), covar=tensor([0.0497, 0.1295, 0.1734, 0.1495, 0.0624, 0.1583, 0.0680, 0.0667], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0151, 0.0189, 0.0157, 0.0100, 0.0160, 0.0111, 0.0141], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 00:16:04,708 INFO [train.py:901] (3/4) Epoch 20, batch 1600, loss[loss=0.1901, simple_loss=0.278, pruned_loss=0.05105, over 8506.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2898, pruned_loss=0.0635, over 1616026.64 frames. ], batch size: 39, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:16:08,765 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.617e+02 2.295e+02 2.863e+02 3.431e+02 6.352e+02, threshold=5.726e+02, percent-clipped=1.0 +2023-02-07 00:16:20,605 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155197.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:16:27,178 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155207.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:16:40,714 INFO [train.py:901] (3/4) Epoch 20, batch 1650, loss[loss=0.2151, simple_loss=0.2917, pruned_loss=0.06928, over 8675.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2902, pruned_loss=0.06373, over 1616674.64 frames. ], batch size: 39, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:16:45,136 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155232.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:17:15,948 INFO [train.py:901] (3/4) Epoch 20, batch 1700, loss[loss=0.1957, simple_loss=0.2761, pruned_loss=0.05761, over 8330.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2907, pruned_loss=0.06359, over 1616742.53 frames. ], batch size: 26, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:17:19,376 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.479e+02 2.383e+02 2.759e+02 3.259e+02 7.427e+02, threshold=5.517e+02, percent-clipped=3.0 +2023-02-07 00:17:20,956 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5713, 1.9262, 2.1474, 1.2181, 2.1857, 1.4510, 0.6486, 1.8653], + device='cuda:3'), covar=tensor([0.0662, 0.0381, 0.0281, 0.0597, 0.0398, 0.0847, 0.0901, 0.0289], + device='cuda:3'), in_proj_covar=tensor([0.0443, 0.0384, 0.0338, 0.0442, 0.0370, 0.0529, 0.0388, 0.0409], + device='cuda:3'), out_proj_covar=tensor([1.1964e-04, 1.0103e-04, 8.9166e-05, 1.1711e-04, 9.7986e-05, 1.5067e-04, + 1.0512e-04, 1.0900e-04], device='cuda:3') +2023-02-07 00:17:44,038 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.0257, 1.9727, 1.9945, 1.9546, 1.0684, 1.7095, 2.2004, 1.9617], + device='cuda:3'), covar=tensor([0.0405, 0.1118, 0.1544, 0.1229, 0.0600, 0.1396, 0.0596, 0.0579], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0151, 0.0190, 0.0158, 0.0100, 0.0161, 0.0112, 0.0141], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 00:17:51,288 INFO [train.py:901] (3/4) Epoch 20, batch 1750, loss[loss=0.2074, simple_loss=0.3027, pruned_loss=0.056, over 8464.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2909, pruned_loss=0.06401, over 1616907.34 frames. ], batch size: 27, lr: 3.82e-03, grad_scale: 16.0 +2023-02-07 00:18:00,382 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=155338.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:18:11,548 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0371, 3.5327, 2.0791, 2.6974, 2.6826, 1.9655, 2.5395, 2.9122], + device='cuda:3'), covar=tensor([0.1627, 0.0335, 0.1196, 0.0746, 0.0676, 0.1396, 0.1095, 0.1047], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0235, 0.0330, 0.0303, 0.0300, 0.0335, 0.0343, 0.0316], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 00:18:17,126 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155361.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:18:27,007 INFO [train.py:901] (3/4) Epoch 20, batch 1800, loss[loss=0.1991, simple_loss=0.2812, pruned_loss=0.0585, over 7810.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2911, pruned_loss=0.06433, over 1617261.51 frames. ], batch size: 20, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:18:31,096 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.706e+02 2.586e+02 2.965e+02 3.772e+02 7.314e+02, threshold=5.929e+02, percent-clipped=8.0 +2023-02-07 00:18:34,027 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155386.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:19:01,130 INFO [train.py:901] (3/4) Epoch 20, batch 1850, loss[loss=0.21, simple_loss=0.2924, pruned_loss=0.06381, over 7260.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2908, pruned_loss=0.06402, over 1617627.48 frames. ], batch size: 16, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:19:04,533 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=155431.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:19:20,142 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155453.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:19:20,184 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155453.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:19:36,702 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155475.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:19:37,259 INFO [train.py:901] (3/4) Epoch 20, batch 1900, loss[loss=0.2273, simple_loss=0.3123, pruned_loss=0.07113, over 8294.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2908, pruned_loss=0.06408, over 1615302.96 frames. ], batch size: 23, lr: 3.82e-03, grad_scale: 8.0 +2023-02-07 00:19:38,773 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155478.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:19:41,336 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.499e+02 2.441e+02 2.899e+02 3.473e+02 6.405e+02, threshold=5.799e+02, percent-clipped=1.0 +2023-02-07 00:20:11,844 INFO [train.py:901] (3/4) Epoch 20, batch 1950, loss[loss=0.2122, simple_loss=0.2953, pruned_loss=0.06452, over 8253.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2907, pruned_loss=0.06428, over 1616257.79 frames. ], batch size: 24, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:20:13,302 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-07 00:20:26,415 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155546.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:20:26,928 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-07 00:20:46,968 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-07 00:20:47,674 INFO [train.py:901] (3/4) Epoch 20, batch 2000, loss[loss=0.1807, simple_loss=0.2709, pruned_loss=0.0452, over 7963.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2901, pruned_loss=0.06395, over 1616395.20 frames. ], batch size: 21, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:20:51,754 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.521e+02 2.363e+02 2.911e+02 3.881e+02 1.027e+03, threshold=5.822e+02, percent-clipped=2.0 +2023-02-07 00:21:09,815 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-07 00:21:20,952 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155623.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:21:22,871 INFO [train.py:901] (3/4) Epoch 20, batch 2050, loss[loss=0.1977, simple_loss=0.2637, pruned_loss=0.06585, over 7430.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.2905, pruned_loss=0.06406, over 1618081.87 frames. ], batch size: 17, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:21:34,546 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1219, 2.3541, 1.9056, 2.8044, 1.3529, 1.6127, 2.0006, 2.2957], + device='cuda:3'), covar=tensor([0.0689, 0.0694, 0.0841, 0.0384, 0.1136, 0.1348, 0.0949, 0.0718], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0198, 0.0248, 0.0213, 0.0205, 0.0250, 0.0253, 0.0208], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 00:21:57,538 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155675.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:21:58,080 INFO [train.py:901] (3/4) Epoch 20, batch 2100, loss[loss=0.1628, simple_loss=0.2493, pruned_loss=0.03814, over 7651.00 frames. ], tot_loss[loss=0.208, simple_loss=0.289, pruned_loss=0.06349, over 1614199.16 frames. ], batch size: 19, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:22:02,102 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.545e+02 2.564e+02 2.968e+02 3.686e+02 8.256e+02, threshold=5.935e+02, percent-clipped=7.0 +2023-02-07 00:22:15,400 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7790, 1.7764, 2.4174, 1.6730, 1.3151, 2.3663, 0.4631, 1.4305], + device='cuda:3'), covar=tensor([0.1809, 0.1176, 0.0354, 0.1234, 0.3010, 0.0415, 0.2440, 0.1695], + device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0193, 0.0124, 0.0220, 0.0268, 0.0134, 0.0170, 0.0188], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 00:22:22,176 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155709.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:22:31,067 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0941, 1.6142, 1.4054, 1.6292, 1.3860, 1.2565, 1.3158, 1.3494], + device='cuda:3'), covar=tensor([0.1017, 0.0444, 0.1226, 0.0497, 0.0676, 0.1415, 0.0761, 0.0744], + device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0235, 0.0328, 0.0302, 0.0296, 0.0331, 0.0338, 0.0312], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 00:22:33,482 INFO [train.py:901] (3/4) Epoch 20, batch 2150, loss[loss=0.2345, simple_loss=0.3208, pruned_loss=0.07406, over 8460.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2887, pruned_loss=0.06372, over 1610075.75 frames. ], batch size: 25, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:22:39,034 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155734.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:22:54,736 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8294, 3.7972, 3.4506, 1.8218, 3.3369, 3.4577, 3.3846, 3.3260], + device='cuda:3'), covar=tensor([0.0871, 0.0625, 0.1126, 0.4980, 0.0950, 0.0998, 0.1386, 0.0972], + device='cuda:3'), in_proj_covar=tensor([0.0516, 0.0426, 0.0431, 0.0533, 0.0420, 0.0433, 0.0415, 0.0373], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 00:23:07,321 INFO [train.py:901] (3/4) Epoch 20, batch 2200, loss[loss=0.1752, simple_loss=0.253, pruned_loss=0.04867, over 7786.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2889, pruned_loss=0.06368, over 1608435.00 frames. ], batch size: 19, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:23:08,431 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.51 vs. limit=2.0 +2023-02-07 00:23:12,092 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.799e+02 2.519e+02 2.939e+02 3.787e+02 7.175e+02, threshold=5.878e+02, percent-clipped=4.0 +2023-02-07 00:23:26,037 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155802.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:23:26,115 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155802.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:23:38,414 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=155819.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:23:43,082 INFO [train.py:901] (3/4) Epoch 20, batch 2250, loss[loss=0.2341, simple_loss=0.3051, pruned_loss=0.08156, over 8035.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2887, pruned_loss=0.06361, over 1606766.22 frames. ], batch size: 22, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:23:44,831 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155827.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:24:17,866 INFO [train.py:901] (3/4) Epoch 20, batch 2300, loss[loss=0.2341, simple_loss=0.3176, pruned_loss=0.07529, over 8592.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2884, pruned_loss=0.0637, over 1602645.69 frames. ], batch size: 39, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:24:21,977 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.467e+02 2.500e+02 2.966e+02 3.753e+02 6.656e+02, threshold=5.933e+02, percent-clipped=3.0 +2023-02-07 00:24:54,617 INFO [train.py:901] (3/4) Epoch 20, batch 2350, loss[loss=0.2506, simple_loss=0.3197, pruned_loss=0.09072, over 7044.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2886, pruned_loss=0.06343, over 1606251.86 frames. ], batch size: 71, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:25:00,029 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155934.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:25:17,382 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5636, 2.5427, 1.8854, 2.4057, 2.2568, 1.5240, 2.1762, 2.3038], + device='cuda:3'), covar=tensor([0.1421, 0.0404, 0.1231, 0.0565, 0.0660, 0.1615, 0.0864, 0.0823], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0235, 0.0331, 0.0305, 0.0298, 0.0334, 0.0341, 0.0315], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 00:25:23,234 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=155967.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:25:23,382 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2970, 1.6389, 1.6868, 0.9690, 1.7155, 1.3678, 0.2351, 1.5323], + device='cuda:3'), covar=tensor([0.0513, 0.0336, 0.0289, 0.0511, 0.0396, 0.0860, 0.0883, 0.0268], + device='cuda:3'), in_proj_covar=tensor([0.0444, 0.0382, 0.0336, 0.0441, 0.0368, 0.0528, 0.0389, 0.0410], + device='cuda:3'), out_proj_covar=tensor([1.1986e-04, 1.0052e-04, 8.8717e-05, 1.1676e-04, 9.7577e-05, 1.5040e-04, + 1.0548e-04, 1.0922e-04], device='cuda:3') +2023-02-07 00:25:29,310 INFO [train.py:901] (3/4) Epoch 20, batch 2400, loss[loss=0.1765, simple_loss=0.2468, pruned_loss=0.05311, over 7918.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2888, pruned_loss=0.06317, over 1612511.16 frames. ], batch size: 20, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:25:33,221 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.514e+02 2.377e+02 2.729e+02 3.502e+02 6.388e+02, threshold=5.458e+02, percent-clipped=1.0 +2023-02-07 00:26:01,013 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=156019.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:26:05,601 INFO [train.py:901] (3/4) Epoch 20, batch 2450, loss[loss=0.2114, simple_loss=0.2957, pruned_loss=0.0636, over 8088.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2885, pruned_loss=0.06302, over 1613662.18 frames. ], batch size: 21, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:26:40,968 INFO [train.py:901] (3/4) Epoch 20, batch 2500, loss[loss=0.1811, simple_loss=0.2633, pruned_loss=0.04949, over 8098.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2891, pruned_loss=0.06367, over 1617080.58 frames. ], batch size: 21, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:26:45,015 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.754e+02 2.463e+02 3.105e+02 3.826e+02 1.382e+03, threshold=6.210e+02, percent-clipped=11.0 +2023-02-07 00:26:45,217 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156082.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:26:49,178 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156088.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:27:15,803 INFO [train.py:901] (3/4) Epoch 20, batch 2550, loss[loss=0.2142, simple_loss=0.3033, pruned_loss=0.06254, over 8501.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2899, pruned_loss=0.06423, over 1615815.24 frames. ], batch size: 26, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:27:21,367 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156134.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:27:23,384 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2322, 3.1725, 2.9060, 1.6139, 2.8403, 2.8928, 2.8449, 2.7534], + device='cuda:3'), covar=tensor([0.1202, 0.0878, 0.1439, 0.4785, 0.1092, 0.1206, 0.1668, 0.1120], + device='cuda:3'), in_proj_covar=tensor([0.0517, 0.0428, 0.0431, 0.0531, 0.0421, 0.0434, 0.0419, 0.0375], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 00:27:29,848 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=156146.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:27:45,120 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 +2023-02-07 00:27:50,757 INFO [train.py:901] (3/4) Epoch 20, batch 2600, loss[loss=0.2378, simple_loss=0.3189, pruned_loss=0.07832, over 8326.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2903, pruned_loss=0.06481, over 1610096.87 frames. ], batch size: 25, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:27:54,657 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.376e+02 3.118e+02 3.808e+02 9.704e+02, threshold=6.236e+02, percent-clipped=5.0 +2023-02-07 00:28:00,399 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156190.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:28:17,469 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156215.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:28:24,592 INFO [train.py:901] (3/4) Epoch 20, batch 2650, loss[loss=0.178, simple_loss=0.2591, pruned_loss=0.04845, over 8338.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2894, pruned_loss=0.06405, over 1609655.28 frames. ], batch size: 25, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:28:30,661 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156234.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:28:49,206 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156261.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:29:00,108 INFO [train.py:901] (3/4) Epoch 20, batch 2700, loss[loss=0.1973, simple_loss=0.2843, pruned_loss=0.0551, over 8471.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2903, pruned_loss=0.06458, over 1609528.57 frames. ], batch size: 25, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:29:04,083 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.694e+02 2.401e+02 3.078e+02 3.829e+02 8.557e+02, threshold=6.156e+02, percent-clipped=4.0 +2023-02-07 00:29:23,254 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156308.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:29:35,128 INFO [train.py:901] (3/4) Epoch 20, batch 2750, loss[loss=0.2187, simple_loss=0.3033, pruned_loss=0.06706, over 8515.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2906, pruned_loss=0.06443, over 1616576.50 frames. ], batch size: 28, lr: 3.81e-03, grad_scale: 8.0 +2023-02-07 00:29:38,039 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9900, 1.6253, 1.8019, 1.3587, 1.0116, 1.5171, 1.7884, 1.7623], + device='cuda:3'), covar=tensor([0.0507, 0.1245, 0.1625, 0.1461, 0.0589, 0.1535, 0.0671, 0.0579], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0190, 0.0158, 0.0100, 0.0161, 0.0112, 0.0141], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 00:29:43,525 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156338.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:30:01,814 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156363.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:30:10,361 INFO [train.py:901] (3/4) Epoch 20, batch 2800, loss[loss=0.2191, simple_loss=0.298, pruned_loss=0.07013, over 8292.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2905, pruned_loss=0.06442, over 1617776.46 frames. ], batch size: 23, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:30:15,855 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.451e+02 2.534e+02 2.983e+02 3.648e+02 6.974e+02, threshold=5.966e+02, percent-clipped=1.0 +2023-02-07 00:30:20,860 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156390.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:30:23,642 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8141, 1.7938, 2.4446, 1.6072, 1.3837, 2.4012, 0.4825, 1.5218], + device='cuda:3'), covar=tensor([0.1554, 0.1260, 0.0339, 0.1244, 0.2727, 0.0404, 0.2332, 0.1358], + device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0193, 0.0124, 0.0221, 0.0268, 0.0134, 0.0168, 0.0188], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 00:30:38,827 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156415.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:30:46,250 INFO [train.py:901] (3/4) Epoch 20, batch 2850, loss[loss=0.1554, simple_loss=0.2391, pruned_loss=0.03584, over 7553.00 frames. ], tot_loss[loss=0.21, simple_loss=0.291, pruned_loss=0.06445, over 1619468.26 frames. ], batch size: 18, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:30:50,403 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=156432.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:31:20,791 INFO [train.py:901] (3/4) Epoch 20, batch 2900, loss[loss=0.1967, simple_loss=0.285, pruned_loss=0.05425, over 8368.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2911, pruned_loss=0.06445, over 1619990.58 frames. ], batch size: 24, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:31:26,323 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.692e+02 2.409e+02 2.783e+02 3.401e+02 8.568e+02, threshold=5.566e+02, percent-clipped=1.0 +2023-02-07 00:31:50,390 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156517.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:31:53,705 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-07 00:31:57,132 INFO [train.py:901] (3/4) Epoch 20, batch 2950, loss[loss=0.2675, simple_loss=0.3424, pruned_loss=0.09631, over 8586.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2917, pruned_loss=0.06489, over 1616955.36 frames. ], batch size: 31, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:32:08,284 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156542.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:32:11,752 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6990, 1.3449, 1.5073, 1.2558, 0.9342, 1.3273, 1.4783, 1.3448], + device='cuda:3'), covar=tensor([0.0588, 0.1243, 0.1682, 0.1443, 0.0640, 0.1511, 0.0742, 0.0681], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0153, 0.0191, 0.0159, 0.0100, 0.0162, 0.0112, 0.0142], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 00:32:11,760 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156547.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:32:25,906 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0666, 1.6080, 1.3096, 1.5607, 1.3810, 1.2094, 1.2475, 1.3766], + device='cuda:3'), covar=tensor([0.1095, 0.0444, 0.1272, 0.0577, 0.0702, 0.1471, 0.0942, 0.0798], + device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0237, 0.0331, 0.0307, 0.0299, 0.0335, 0.0344, 0.0317], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 00:32:31,023 INFO [train.py:901] (3/4) Epoch 20, batch 3000, loss[loss=0.2126, simple_loss=0.2991, pruned_loss=0.06303, over 8463.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2913, pruned_loss=0.06476, over 1612132.87 frames. ], batch size: 29, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:32:31,024 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 00:32:43,174 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7724, 1.5753, 2.7012, 1.4726, 2.1976, 2.8561, 3.0427, 2.4964], + device='cuda:3'), covar=tensor([0.1179, 0.1687, 0.0402, 0.2144, 0.0932, 0.0325, 0.0551, 0.0556], + device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0323, 0.0287, 0.0315, 0.0305, 0.0262, 0.0410, 0.0306], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 00:32:46,833 INFO [train.py:935] (3/4) Epoch 20, validation: loss=0.1756, simple_loss=0.2756, pruned_loss=0.03779, over 944034.00 frames. +2023-02-07 00:32:46,834 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-07 00:32:48,394 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=156578.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:32:48,562 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5466, 1.9525, 3.0746, 1.4183, 2.2971, 2.0697, 1.7626, 2.2232], + device='cuda:3'), covar=tensor([0.1831, 0.2590, 0.0844, 0.4434, 0.1880, 0.3113, 0.2168, 0.2316], + device='cuda:3'), in_proj_covar=tensor([0.0523, 0.0595, 0.0558, 0.0638, 0.0647, 0.0598, 0.0531, 0.0635], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 00:32:51,796 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.571e+02 2.420e+02 3.007e+02 3.801e+02 6.408e+02, threshold=6.014e+02, percent-clipped=4.0 +2023-02-07 00:33:22,164 INFO [train.py:901] (3/4) Epoch 20, batch 3050, loss[loss=0.2406, simple_loss=0.3141, pruned_loss=0.08349, over 7249.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2909, pruned_loss=0.06463, over 1606973.75 frames. ], batch size: 16, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:33:40,595 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=156652.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:33:57,463 INFO [train.py:901] (3/4) Epoch 20, batch 3100, loss[loss=0.2324, simple_loss=0.3181, pruned_loss=0.07342, over 8505.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2909, pruned_loss=0.0646, over 1607669.49 frames. ], batch size: 26, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:34:02,290 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.497e+02 2.428e+02 2.992e+02 3.732e+02 8.006e+02, threshold=5.985e+02, percent-clipped=5.0 +2023-02-07 00:34:09,238 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156693.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:34:31,696 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-07 00:34:31,974 INFO [train.py:901] (3/4) Epoch 20, batch 3150, loss[loss=0.2411, simple_loss=0.3181, pruned_loss=0.08204, over 8342.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2925, pruned_loss=0.06546, over 1614405.16 frames. ], batch size: 26, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:35:01,272 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156767.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:35:01,309 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156767.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:35:07,244 INFO [train.py:901] (3/4) Epoch 20, batch 3200, loss[loss=0.2257, simple_loss=0.3208, pruned_loss=0.06534, over 8320.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2922, pruned_loss=0.06456, over 1612810.97 frames. ], batch size: 25, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:35:11,880 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.830e+02 2.338e+02 2.875e+02 3.612e+02 1.133e+03, threshold=5.749e+02, percent-clipped=4.0 +2023-02-07 00:35:24,621 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.53 vs. limit=2.0 +2023-02-07 00:35:26,504 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156803.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:35:41,814 INFO [train.py:901] (3/4) Epoch 20, batch 3250, loss[loss=0.2377, simple_loss=0.3199, pruned_loss=0.07773, over 8460.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2908, pruned_loss=0.06402, over 1612999.40 frames. ], batch size: 29, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:35:43,290 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156828.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:36:17,056 INFO [train.py:901] (3/4) Epoch 20, batch 3300, loss[loss=0.209, simple_loss=0.2803, pruned_loss=0.06883, over 7657.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2912, pruned_loss=0.06388, over 1615427.55 frames. ], batch size: 19, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:36:21,772 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.702e+02 2.341e+02 2.967e+02 3.887e+02 7.432e+02, threshold=5.934e+02, percent-clipped=7.0 +2023-02-07 00:36:32,803 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0921, 1.5702, 1.7268, 1.4091, 0.9467, 1.5121, 1.8032, 1.5386], + device='cuda:3'), covar=tensor([0.0483, 0.1243, 0.1649, 0.1418, 0.0613, 0.1470, 0.0675, 0.0629], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0153, 0.0190, 0.0158, 0.0100, 0.0161, 0.0112, 0.0141], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 00:36:35,550 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156903.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:36:51,529 INFO [train.py:901] (3/4) Epoch 20, batch 3350, loss[loss=0.2061, simple_loss=0.292, pruned_loss=0.06012, over 8181.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2915, pruned_loss=0.06405, over 1612454.97 frames. ], batch size: 23, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:37:05,904 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-07 00:37:07,172 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156949.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:37:25,819 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156974.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:37:27,098 INFO [train.py:901] (3/4) Epoch 20, batch 3400, loss[loss=0.1954, simple_loss=0.2877, pruned_loss=0.05152, over 8290.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.292, pruned_loss=0.06447, over 1609640.52 frames. ], batch size: 23, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:37:31,899 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.531e+02 2.508e+02 3.011e+02 3.882e+02 8.239e+02, threshold=6.022e+02, percent-clipped=6.0 +2023-02-07 00:37:49,723 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2447, 1.2982, 1.6492, 1.1829, 0.7287, 1.3727, 1.2482, 1.1455], + device='cuda:3'), covar=tensor([0.0596, 0.1320, 0.1699, 0.1485, 0.0558, 0.1511, 0.0690, 0.0665], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0153, 0.0191, 0.0158, 0.0100, 0.0161, 0.0112, 0.0142], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 00:37:59,125 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9489, 1.4965, 1.7376, 1.3212, 0.9098, 1.5065, 1.7388, 1.4381], + device='cuda:3'), covar=tensor([0.0523, 0.1222, 0.1658, 0.1446, 0.0610, 0.1445, 0.0672, 0.0645], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0153, 0.0191, 0.0159, 0.0100, 0.0161, 0.0112, 0.0142], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 00:38:01,290 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157023.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:38:03,918 INFO [train.py:901] (3/4) Epoch 20, batch 3450, loss[loss=0.1738, simple_loss=0.2508, pruned_loss=0.04837, over 7790.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2905, pruned_loss=0.06394, over 1609947.24 frames. ], batch size: 19, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:38:05,354 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157028.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:38:11,405 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6557, 1.6563, 2.3460, 1.5952, 1.2321, 2.2521, 0.3963, 1.3237], + device='cuda:3'), covar=tensor([0.1967, 0.1326, 0.0337, 0.1216, 0.2932, 0.0441, 0.2239, 0.1363], + device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0191, 0.0123, 0.0216, 0.0265, 0.0132, 0.0165, 0.0186], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 00:38:18,902 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157048.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:38:38,200 INFO [train.py:901] (3/4) Epoch 20, batch 3500, loss[loss=0.207, simple_loss=0.3021, pruned_loss=0.05592, over 8250.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2915, pruned_loss=0.06458, over 1608178.26 frames. ], batch size: 24, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:38:43,592 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.575e+02 2.548e+02 3.004e+02 3.939e+02 7.448e+02, threshold=6.007e+02, percent-clipped=9.0 +2023-02-07 00:39:02,219 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-07 00:39:03,072 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157111.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:39:13,060 INFO [train.py:901] (3/4) Epoch 20, batch 3550, loss[loss=0.2454, simple_loss=0.3247, pruned_loss=0.08306, over 8547.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2922, pruned_loss=0.06457, over 1613904.56 frames. ], batch size: 49, lr: 3.80e-03, grad_scale: 8.0 +2023-02-07 00:39:37,686 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157160.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:39:48,298 INFO [train.py:901] (3/4) Epoch 20, batch 3600, loss[loss=0.2453, simple_loss=0.3197, pruned_loss=0.0854, over 8195.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.292, pruned_loss=0.06429, over 1619354.85 frames. ], batch size: 23, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:39:53,036 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.655e+02 2.446e+02 2.923e+02 3.668e+02 9.434e+02, threshold=5.847e+02, percent-clipped=4.0 +2023-02-07 00:40:10,601 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-07 00:40:15,595 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1907, 1.3203, 1.5385, 1.2874, 0.7802, 1.3400, 1.2250, 1.0758], + device='cuda:3'), covar=tensor([0.0566, 0.1314, 0.1660, 0.1405, 0.0540, 0.1519, 0.0705, 0.0677], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0190, 0.0159, 0.0100, 0.0161, 0.0112, 0.0141], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 00:40:24,180 INFO [train.py:901] (3/4) Epoch 20, batch 3650, loss[loss=0.2597, simple_loss=0.3476, pruned_loss=0.08594, over 8257.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.293, pruned_loss=0.065, over 1621699.37 frames. ], batch size: 24, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:40:24,362 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157226.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:40:38,564 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157247.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:40:42,194 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2023-02-07 00:40:58,619 INFO [train.py:901] (3/4) Epoch 20, batch 3700, loss[loss=0.2159, simple_loss=0.2986, pruned_loss=0.06663, over 8143.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2918, pruned_loss=0.06493, over 1610632.10 frames. ], batch size: 22, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:41:03,188 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.510e+02 2.545e+02 3.038e+02 3.849e+02 9.039e+02, threshold=6.076e+02, percent-clipped=6.0 +2023-02-07 00:41:05,242 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-07 00:41:33,573 INFO [train.py:901] (3/4) Epoch 20, batch 3750, loss[loss=0.202, simple_loss=0.2903, pruned_loss=0.05683, over 8256.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2913, pruned_loss=0.06448, over 1611541.04 frames. ], batch size: 24, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:41:58,956 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157362.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:42:05,447 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157372.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:42:06,809 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157374.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:42:07,964 INFO [train.py:901] (3/4) Epoch 20, batch 3800, loss[loss=0.2442, simple_loss=0.3294, pruned_loss=0.07953, over 8186.00 frames. ], tot_loss[loss=0.21, simple_loss=0.291, pruned_loss=0.0645, over 1609880.54 frames. ], batch size: 23, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:42:12,517 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.709e+02 2.302e+02 2.981e+02 3.884e+02 7.104e+02, threshold=5.962e+02, percent-clipped=4.0 +2023-02-07 00:42:25,022 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157400.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:42:29,223 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.90 vs. limit=5.0 +2023-02-07 00:42:39,605 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157422.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:42:42,787 INFO [train.py:901] (3/4) Epoch 20, batch 3850, loss[loss=0.2284, simple_loss=0.3108, pruned_loss=0.07306, over 8546.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.291, pruned_loss=0.06494, over 1609511.42 frames. ], batch size: 34, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:43:05,834 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.1790, 3.0395, 2.8570, 1.5678, 2.7983, 2.9365, 2.7447, 2.7424], + device='cuda:3'), covar=tensor([0.1115, 0.0866, 0.1368, 0.4432, 0.1103, 0.1276, 0.1699, 0.1021], + device='cuda:3'), in_proj_covar=tensor([0.0511, 0.0421, 0.0426, 0.0526, 0.0415, 0.0428, 0.0413, 0.0371], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 00:43:09,760 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-07 00:43:17,668 INFO [train.py:901] (3/4) Epoch 20, batch 3900, loss[loss=0.2085, simple_loss=0.296, pruned_loss=0.06045, over 8199.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2917, pruned_loss=0.0654, over 1614012.37 frames. ], batch size: 23, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:43:21,792 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157482.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:43:22,210 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.662e+02 2.513e+02 3.153e+02 3.900e+02 7.255e+02, threshold=6.305e+02, percent-clipped=5.0 +2023-02-07 00:43:24,996 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157487.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:43:37,148 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157504.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:43:39,410 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157507.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:43:43,558 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2229, 1.1142, 3.3580, 1.0407, 2.9400, 2.8141, 3.0934, 2.9717], + device='cuda:3'), covar=tensor([0.0808, 0.4651, 0.0900, 0.4512, 0.1480, 0.1195, 0.0824, 0.0940], + device='cuda:3'), in_proj_covar=tensor([0.0612, 0.0635, 0.0687, 0.0616, 0.0697, 0.0603, 0.0599, 0.0665], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 00:43:51,268 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157524.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 00:43:52,538 INFO [train.py:901] (3/4) Epoch 20, batch 3950, loss[loss=0.1727, simple_loss=0.2519, pruned_loss=0.04675, over 7792.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2914, pruned_loss=0.06546, over 1609085.45 frames. ], batch size: 19, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:44:28,451 INFO [train.py:901] (3/4) Epoch 20, batch 4000, loss[loss=0.2244, simple_loss=0.3019, pruned_loss=0.07342, over 8291.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2911, pruned_loss=0.06472, over 1609460.75 frames. ], batch size: 49, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:44:33,891 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.914e+02 2.441e+02 3.259e+02 3.960e+02 7.383e+02, threshold=6.518e+02, percent-clipped=3.0 +2023-02-07 00:44:36,132 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157586.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:44:57,788 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157618.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:44:58,434 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157619.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:45:03,493 INFO [train.py:901] (3/4) Epoch 20, batch 4050, loss[loss=0.186, simple_loss=0.2752, pruned_loss=0.04844, over 7970.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2922, pruned_loss=0.06536, over 1611348.92 frames. ], batch size: 21, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:45:15,102 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157643.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:45:38,035 INFO [train.py:901] (3/4) Epoch 20, batch 4100, loss[loss=0.2032, simple_loss=0.2883, pruned_loss=0.05906, over 8076.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2926, pruned_loss=0.06513, over 1616362.25 frames. ], batch size: 21, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:45:42,588 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.673e+02 2.468e+02 3.178e+02 4.268e+02 8.149e+02, threshold=6.355e+02, percent-clipped=4.0 +2023-02-07 00:46:07,478 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157718.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:46:12,790 INFO [train.py:901] (3/4) Epoch 20, batch 4150, loss[loss=0.2125, simple_loss=0.3046, pruned_loss=0.06014, over 8319.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2928, pruned_loss=0.06506, over 1619811.31 frames. ], batch size: 25, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:46:25,108 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157743.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:46:25,658 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157744.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:46:40,586 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157766.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 00:46:41,946 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157768.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:46:45,980 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1163, 1.8762, 2.5280, 2.0870, 2.4467, 2.1887, 1.8969, 1.2868], + device='cuda:3'), covar=tensor([0.5228, 0.4801, 0.1786, 0.3284, 0.2350, 0.2933, 0.1926, 0.4992], + device='cuda:3'), in_proj_covar=tensor([0.0934, 0.0966, 0.0788, 0.0931, 0.0984, 0.0882, 0.0737, 0.0815], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 00:46:46,942 INFO [train.py:901] (3/4) Epoch 20, batch 4200, loss[loss=0.2576, simple_loss=0.3293, pruned_loss=0.09294, over 8470.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2921, pruned_loss=0.06475, over 1616044.84 frames. ], batch size: 25, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:46:52,350 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.549e+02 2.385e+02 2.811e+02 3.577e+02 7.269e+02, threshold=5.621e+02, percent-clipped=2.0 +2023-02-07 00:47:08,774 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-07 00:47:10,323 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5561, 1.3610, 4.7029, 1.8389, 4.1433, 3.8901, 4.2643, 4.1023], + device='cuda:3'), covar=tensor([0.0579, 0.5293, 0.0521, 0.3960, 0.1099, 0.0973, 0.0625, 0.0690], + device='cuda:3'), in_proj_covar=tensor([0.0619, 0.0640, 0.0694, 0.0622, 0.0703, 0.0610, 0.0605, 0.0670], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 00:47:23,327 INFO [train.py:901] (3/4) Epoch 20, batch 4250, loss[loss=0.1922, simple_loss=0.2815, pruned_loss=0.05145, over 8548.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2912, pruned_loss=0.06402, over 1618362.59 frames. ], batch size: 31, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:47:26,207 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8085, 1.3568, 1.6003, 1.2590, 0.8786, 1.3764, 1.6529, 1.5484], + device='cuda:3'), covar=tensor([0.0560, 0.1299, 0.1790, 0.1532, 0.0616, 0.1535, 0.0714, 0.0649], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0153, 0.0191, 0.0159, 0.0100, 0.0162, 0.0113, 0.0142], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 00:47:28,293 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157833.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:47:32,306 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-07 00:47:46,380 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157859.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:47:53,374 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157868.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 00:47:55,481 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157871.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:47:58,177 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157875.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:47:58,647 INFO [train.py:901] (3/4) Epoch 20, batch 4300, loss[loss=0.2211, simple_loss=0.2913, pruned_loss=0.07542, over 8282.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2904, pruned_loss=0.06398, over 1613494.11 frames. ], batch size: 23, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:48:02,059 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157881.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:48:03,194 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.270e+02 2.745e+02 3.400e+02 8.203e+02, threshold=5.491e+02, percent-clipped=7.0 +2023-02-07 00:48:15,326 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157900.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:48:17,997 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1262, 1.9398, 2.8973, 1.8821, 2.5056, 3.1667, 3.1194, 2.8964], + device='cuda:3'), covar=tensor([0.0888, 0.1349, 0.0580, 0.1551, 0.1299, 0.0237, 0.0697, 0.0412], + device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0318, 0.0285, 0.0311, 0.0300, 0.0262, 0.0407, 0.0303], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 00:48:24,971 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-07 00:48:33,499 INFO [train.py:901] (3/4) Epoch 20, batch 4350, loss[loss=0.2174, simple_loss=0.2986, pruned_loss=0.0681, over 8329.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2904, pruned_loss=0.06434, over 1610990.38 frames. ], batch size: 25, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:48:36,292 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=157930.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:48:48,941 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.92 vs. limit=5.0 +2023-02-07 00:49:04,096 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-07 00:49:08,270 INFO [train.py:901] (3/4) Epoch 20, batch 4400, loss[loss=0.1821, simple_loss=0.2633, pruned_loss=0.05039, over 8132.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.2903, pruned_loss=0.06416, over 1615787.47 frames. ], batch size: 22, lr: 3.79e-03, grad_scale: 8.0 +2023-02-07 00:49:13,805 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.515e+02 2.443e+02 2.894e+02 3.714e+02 1.238e+03, threshold=5.788e+02, percent-clipped=6.0 +2023-02-07 00:49:14,000 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157983.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 00:49:44,325 INFO [train.py:901] (3/4) Epoch 20, batch 4450, loss[loss=0.2091, simple_loss=0.2808, pruned_loss=0.06866, over 7912.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2909, pruned_loss=0.06453, over 1616119.83 frames. ], batch size: 20, lr: 3.78e-03, grad_scale: 8.0 +2023-02-07 00:49:45,693 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-07 00:49:57,299 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158045.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:50:18,894 INFO [train.py:901] (3/4) Epoch 20, batch 4500, loss[loss=0.3174, simple_loss=0.3671, pruned_loss=0.1339, over 6626.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2912, pruned_loss=0.06464, over 1616918.59 frames. ], batch size: 72, lr: 3.78e-03, grad_scale: 8.0 +2023-02-07 00:50:23,590 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.668e+02 2.382e+02 2.908e+02 3.384e+02 7.082e+02, threshold=5.816e+02, percent-clipped=5.0 +2023-02-07 00:50:27,967 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158089.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:50:39,145 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-07 00:50:45,180 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158114.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:50:45,855 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158115.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:50:53,563 INFO [train.py:901] (3/4) Epoch 20, batch 4550, loss[loss=0.1717, simple_loss=0.259, pruned_loss=0.04227, over 7930.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2911, pruned_loss=0.0648, over 1614350.26 frames. ], batch size: 20, lr: 3.78e-03, grad_scale: 8.0 +2023-02-07 00:51:01,051 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158137.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 00:51:02,913 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158140.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:51:02,963 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158140.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:51:18,409 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158162.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 00:51:28,236 INFO [train.py:901] (3/4) Epoch 20, batch 4600, loss[loss=0.214, simple_loss=0.2977, pruned_loss=0.06513, over 8555.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2912, pruned_loss=0.06464, over 1614902.78 frames. ], batch size: 31, lr: 3.78e-03, grad_scale: 8.0 +2023-02-07 00:51:32,831 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.472e+02 2.506e+02 3.217e+02 3.763e+02 8.986e+02, threshold=6.435e+02, percent-clipped=3.0 +2023-02-07 00:51:54,921 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=158215.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:52:01,862 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4623, 1.4911, 1.8236, 1.2713, 1.1478, 1.8160, 0.1664, 1.1738], + device='cuda:3'), covar=tensor([0.1653, 0.1301, 0.0356, 0.1086, 0.2979, 0.0460, 0.2207, 0.1273], + device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0195, 0.0124, 0.0219, 0.0268, 0.0134, 0.0167, 0.0189], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 00:52:03,088 INFO [train.py:901] (3/4) Epoch 20, batch 4650, loss[loss=0.1901, simple_loss=0.2783, pruned_loss=0.05096, over 8334.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2898, pruned_loss=0.06377, over 1610278.27 frames. ], batch size: 26, lr: 3.78e-03, grad_scale: 8.0 +2023-02-07 00:52:12,064 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158239.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 00:52:30,111 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158264.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 00:52:37,869 INFO [train.py:901] (3/4) Epoch 20, batch 4700, loss[loss=0.2405, simple_loss=0.3142, pruned_loss=0.0834, over 8110.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2902, pruned_loss=0.06446, over 1613344.51 frames. ], batch size: 23, lr: 3.78e-03, grad_scale: 8.0 +2023-02-07 00:52:42,603 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.717e+02 2.408e+02 3.012e+02 4.119e+02 1.091e+03, threshold=6.025e+02, percent-clipped=3.0 +2023-02-07 00:52:55,812 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158301.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:53:12,689 INFO [train.py:901] (3/4) Epoch 20, batch 4750, loss[loss=0.1748, simple_loss=0.2577, pruned_loss=0.04595, over 8105.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2895, pruned_loss=0.06381, over 1611602.06 frames. ], batch size: 21, lr: 3.78e-03, grad_scale: 8.0 +2023-02-07 00:53:12,915 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158326.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:53:15,564 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158330.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:53:40,915 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-07 00:53:43,656 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-07 00:53:48,267 INFO [train.py:901] (3/4) Epoch 20, batch 4800, loss[loss=0.1974, simple_loss=0.2686, pruned_loss=0.06316, over 7659.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2902, pruned_loss=0.06409, over 1613103.97 frames. ], batch size: 19, lr: 3.78e-03, grad_scale: 16.0 +2023-02-07 00:53:52,896 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.466e+02 2.386e+02 2.729e+02 3.445e+02 7.258e+02, threshold=5.458e+02, percent-clipped=2.0 +2023-02-07 00:54:06,935 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158402.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:54:22,904 INFO [train.py:901] (3/4) Epoch 20, batch 4850, loss[loss=0.2045, simple_loss=0.2991, pruned_loss=0.05493, over 8361.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2913, pruned_loss=0.06446, over 1611225.42 frames. ], batch size: 24, lr: 3.78e-03, grad_scale: 16.0 +2023-02-07 00:54:23,848 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9798, 1.7012, 2.0451, 1.7808, 1.9490, 2.0244, 1.8039, 0.7428], + device='cuda:3'), covar=tensor([0.5151, 0.4448, 0.1810, 0.3343, 0.2358, 0.2655, 0.1722, 0.4795], + device='cuda:3'), in_proj_covar=tensor([0.0934, 0.0966, 0.0788, 0.0933, 0.0990, 0.0882, 0.0740, 0.0820], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 00:54:33,540 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-07 00:54:35,079 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5113, 1.6842, 4.5182, 1.8792, 2.5420, 5.0670, 5.1462, 4.3605], + device='cuda:3'), covar=tensor([0.1120, 0.1896, 0.0239, 0.2023, 0.1124, 0.0189, 0.0495, 0.0586], + device='cuda:3'), in_proj_covar=tensor([0.0293, 0.0320, 0.0287, 0.0313, 0.0303, 0.0263, 0.0410, 0.0305], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 00:54:51,919 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0850, 1.6846, 3.2848, 1.4883, 2.2847, 3.5651, 3.7064, 3.0124], + device='cuda:3'), covar=tensor([0.1145, 0.1751, 0.0416, 0.2192, 0.1141, 0.0261, 0.0619, 0.0596], + device='cuda:3'), in_proj_covar=tensor([0.0293, 0.0319, 0.0286, 0.0312, 0.0303, 0.0262, 0.0409, 0.0305], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 00:54:57,240 INFO [train.py:901] (3/4) Epoch 20, batch 4900, loss[loss=0.2256, simple_loss=0.3042, pruned_loss=0.07346, over 8298.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.2924, pruned_loss=0.06539, over 1615735.30 frames. ], batch size: 23, lr: 3.78e-03, grad_scale: 16.0 +2023-02-07 00:55:02,468 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.824e+02 2.481e+02 3.123e+02 4.208e+02 8.958e+02, threshold=6.246e+02, percent-clipped=7.0 +2023-02-07 00:55:03,227 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=158484.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:55:32,883 INFO [train.py:901] (3/4) Epoch 20, batch 4950, loss[loss=0.2199, simple_loss=0.3019, pruned_loss=0.06899, over 8208.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2918, pruned_loss=0.06496, over 1613264.68 frames. ], batch size: 23, lr: 3.78e-03, grad_scale: 16.0 +2023-02-07 00:56:07,753 INFO [train.py:901] (3/4) Epoch 20, batch 5000, loss[loss=0.2094, simple_loss=0.2904, pruned_loss=0.06416, over 8024.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2913, pruned_loss=0.06491, over 1612527.14 frames. ], batch size: 22, lr: 3.78e-03, grad_scale: 16.0 +2023-02-07 00:56:12,217 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.598e+02 2.361e+02 2.881e+02 3.667e+02 7.563e+02, threshold=5.761e+02, percent-clipped=2.0 +2023-02-07 00:56:12,594 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.14 vs. limit=5.0 +2023-02-07 00:56:14,507 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158586.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:56:23,828 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158599.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:56:32,813 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158611.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:56:42,908 INFO [train.py:901] (3/4) Epoch 20, batch 5050, loss[loss=0.1956, simple_loss=0.2688, pruned_loss=0.06117, over 7954.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.292, pruned_loss=0.06532, over 1612590.42 frames. ], batch size: 21, lr: 3.78e-03, grad_scale: 16.0 +2023-02-07 00:56:54,150 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1521, 1.9316, 2.5161, 1.6399, 1.5107, 2.4995, 1.1737, 2.0180], + device='cuda:3'), covar=tensor([0.2023, 0.1460, 0.0416, 0.1553, 0.2630, 0.0431, 0.2071, 0.1289], + device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0197, 0.0127, 0.0222, 0.0273, 0.0135, 0.0171, 0.0192], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 00:57:10,225 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-07 00:57:17,762 INFO [train.py:901] (3/4) Epoch 20, batch 5100, loss[loss=0.1931, simple_loss=0.2721, pruned_loss=0.05707, over 7806.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2914, pruned_loss=0.06513, over 1604788.31 frames. ], batch size: 20, lr: 3.78e-03, grad_scale: 16.0 +2023-02-07 00:57:23,331 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.884e+02 2.670e+02 3.233e+02 3.910e+02 8.185e+02, threshold=6.466e+02, percent-clipped=7.0 +2023-02-07 00:57:23,581 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1761, 1.0210, 1.2324, 1.0791, 0.9017, 1.2587, 0.0815, 0.9393], + device='cuda:3'), covar=tensor([0.1557, 0.1609, 0.0564, 0.0813, 0.2969, 0.0614, 0.2219, 0.1411], + device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0195, 0.0125, 0.0219, 0.0270, 0.0134, 0.0168, 0.0190], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 00:57:28,316 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7847, 2.6272, 1.9942, 2.3344, 2.2484, 1.7431, 2.1691, 2.3452], + device='cuda:3'), covar=tensor([0.1501, 0.0394, 0.1034, 0.0643, 0.0707, 0.1429, 0.0975, 0.0979], + device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0239, 0.0336, 0.0312, 0.0304, 0.0341, 0.0346, 0.0321], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 00:57:53,851 INFO [train.py:901] (3/4) Epoch 20, batch 5150, loss[loss=0.2328, simple_loss=0.3173, pruned_loss=0.07412, over 8486.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2908, pruned_loss=0.06476, over 1608762.02 frames. ], batch size: 28, lr: 3.78e-03, grad_scale: 16.0 +2023-02-07 00:58:08,321 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=158746.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:58:28,394 INFO [train.py:901] (3/4) Epoch 20, batch 5200, loss[loss=0.1867, simple_loss=0.2516, pruned_loss=0.06091, over 7697.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2896, pruned_loss=0.0643, over 1610493.25 frames. ], batch size: 18, lr: 3.78e-03, grad_scale: 16.0 +2023-02-07 00:58:33,213 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.548e+02 2.433e+02 2.837e+02 3.461e+02 7.505e+02, threshold=5.673e+02, percent-clipped=2.0 +2023-02-07 00:58:41,636 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158795.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:58:46,634 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158801.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:59:03,969 INFO [train.py:901] (3/4) Epoch 20, batch 5250, loss[loss=0.1952, simple_loss=0.2813, pruned_loss=0.05459, over 8280.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2897, pruned_loss=0.06434, over 1613770.89 frames. ], batch size: 23, lr: 3.78e-03, grad_scale: 16.0 +2023-02-07 00:59:11,274 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-07 00:59:22,876 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158853.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 00:59:24,328 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158855.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:59:28,230 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158861.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:59:38,579 INFO [train.py:901] (3/4) Epoch 20, batch 5300, loss[loss=0.2043, simple_loss=0.2794, pruned_loss=0.06464, over 8086.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2892, pruned_loss=0.06393, over 1613877.76 frames. ], batch size: 21, lr: 3.77e-03, grad_scale: 16.0 +2023-02-07 00:59:41,443 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158880.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 00:59:43,352 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.528e+02 2.336e+02 2.792e+02 3.296e+02 7.091e+02, threshold=5.585e+02, percent-clipped=2.0 +2023-02-07 01:00:13,211 INFO [train.py:901] (3/4) Epoch 20, batch 5350, loss[loss=0.168, simple_loss=0.2567, pruned_loss=0.03968, over 7802.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2878, pruned_loss=0.06336, over 1612988.05 frames. ], batch size: 20, lr: 3.77e-03, grad_scale: 16.0 +2023-02-07 01:00:27,517 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7090, 1.9772, 2.0225, 1.4617, 2.1574, 1.4425, 0.5529, 1.8156], + device='cuda:3'), covar=tensor([0.0516, 0.0331, 0.0323, 0.0464, 0.0356, 0.0869, 0.0821, 0.0265], + device='cuda:3'), in_proj_covar=tensor([0.0453, 0.0388, 0.0339, 0.0441, 0.0370, 0.0533, 0.0395, 0.0415], + device='cuda:3'), out_proj_covar=tensor([1.2208e-04, 1.0214e-04, 8.9355e-05, 1.1664e-04, 9.7860e-05, 1.5158e-04, + 1.0671e-04, 1.1064e-04], device='cuda:3') +2023-02-07 01:00:48,001 INFO [train.py:901] (3/4) Epoch 20, batch 5400, loss[loss=0.1857, simple_loss=0.2676, pruned_loss=0.05196, over 8084.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2877, pruned_loss=0.0635, over 1608445.01 frames. ], batch size: 21, lr: 3.77e-03, grad_scale: 16.0 +2023-02-07 01:00:52,647 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.645e+02 2.394e+02 2.966e+02 3.887e+02 6.953e+02, threshold=5.932e+02, percent-clipped=4.0 +2023-02-07 01:01:09,403 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.7914, 1.4513, 4.1242, 1.8916, 3.2393, 3.2417, 3.7197, 3.6344], + device='cuda:3'), covar=tensor([0.1422, 0.6595, 0.1114, 0.4678, 0.2378, 0.1772, 0.1153, 0.1157], + device='cuda:3'), in_proj_covar=tensor([0.0609, 0.0629, 0.0674, 0.0610, 0.0689, 0.0595, 0.0595, 0.0663], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 01:01:22,917 INFO [train.py:901] (3/4) Epoch 20, batch 5450, loss[loss=0.1822, simple_loss=0.2599, pruned_loss=0.05227, over 8235.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2877, pruned_loss=0.06328, over 1608796.19 frames. ], batch size: 22, lr: 3.77e-03, grad_scale: 16.0 +2023-02-07 01:01:34,635 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 +2023-02-07 01:01:57,423 INFO [train.py:901] (3/4) Epoch 20, batch 5500, loss[loss=0.194, simple_loss=0.2717, pruned_loss=0.05818, over 7967.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2887, pruned_loss=0.06357, over 1608928.03 frames. ], batch size: 21, lr: 3.77e-03, grad_scale: 16.0 +2023-02-07 01:02:00,064 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-07 01:02:02,832 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.672e+02 2.298e+02 2.656e+02 3.222e+02 6.486e+02, threshold=5.312e+02, percent-clipped=1.0 +2023-02-07 01:02:05,795 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159087.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:02:27,308 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159117.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:02:31,195 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7198, 2.0110, 2.1424, 1.3616, 2.2204, 1.5909, 0.6841, 1.8299], + device='cuda:3'), covar=tensor([0.0566, 0.0333, 0.0250, 0.0551, 0.0331, 0.0760, 0.0847, 0.0285], + device='cuda:3'), in_proj_covar=tensor([0.0451, 0.0385, 0.0337, 0.0441, 0.0367, 0.0530, 0.0393, 0.0414], + device='cuda:3'), out_proj_covar=tensor([1.2158e-04, 1.0114e-04, 8.8887e-05, 1.1675e-04, 9.7214e-05, 1.5071e-04, + 1.0623e-04, 1.1032e-04], device='cuda:3') +2023-02-07 01:02:32,989 INFO [train.py:901] (3/4) Epoch 20, batch 5550, loss[loss=0.1942, simple_loss=0.2802, pruned_loss=0.05405, over 8316.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2877, pruned_loss=0.06244, over 1612931.40 frames. ], batch size: 26, lr: 3.77e-03, grad_scale: 16.0 +2023-02-07 01:02:41,928 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=159139.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:02:44,193 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159142.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:02:46,023 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=159145.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:02:53,805 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-07 01:03:08,170 INFO [train.py:901] (3/4) Epoch 20, batch 5600, loss[loss=0.2296, simple_loss=0.3129, pruned_loss=0.07317, over 8243.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2884, pruned_loss=0.06277, over 1612501.60 frames. ], batch size: 22, lr: 3.77e-03, grad_scale: 16.0 +2023-02-07 01:03:09,025 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159177.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:03:12,919 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.638e+02 2.419e+02 2.780e+02 3.445e+02 7.739e+02, threshold=5.561e+02, percent-clipped=2.0 +2023-02-07 01:03:23,293 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=159197.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 01:03:43,999 INFO [train.py:901] (3/4) Epoch 20, batch 5650, loss[loss=0.2096, simple_loss=0.2974, pruned_loss=0.06093, over 8304.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2896, pruned_loss=0.06339, over 1611093.63 frames. ], batch size: 48, lr: 3.77e-03, grad_scale: 16.0 +2023-02-07 01:04:03,412 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159254.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:04:04,621 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-07 01:04:07,504 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159260.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:04:19,056 INFO [train.py:901] (3/4) Epoch 20, batch 5700, loss[loss=0.2238, simple_loss=0.3054, pruned_loss=0.0711, over 8456.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2916, pruned_loss=0.06477, over 1612241.80 frames. ], batch size: 25, lr: 3.77e-03, grad_scale: 8.0 +2023-02-07 01:04:25,333 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.626e+02 2.576e+02 3.260e+02 4.013e+02 6.441e+02, threshold=6.520e+02, percent-clipped=4.0 +2023-02-07 01:04:42,295 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159308.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:04:45,043 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159312.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 01:04:49,923 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2740, 1.4182, 1.2491, 1.8256, 0.7150, 1.1469, 1.3129, 1.4090], + device='cuda:3'), covar=tensor([0.1041, 0.0844, 0.1285, 0.0566, 0.1153, 0.1556, 0.0744, 0.0790], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0196, 0.0246, 0.0212, 0.0205, 0.0247, 0.0249, 0.0207], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 01:04:53,104 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 +2023-02-07 01:04:54,516 INFO [train.py:901] (3/4) Epoch 20, batch 5750, loss[loss=0.2164, simple_loss=0.2977, pruned_loss=0.06752, over 8506.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2918, pruned_loss=0.06463, over 1615127.41 frames. ], batch size: 28, lr: 3.77e-03, grad_scale: 8.0 +2023-02-07 01:05:03,867 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-07 01:05:06,395 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.44 vs. limit=5.0 +2023-02-07 01:05:09,318 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-07 01:05:29,351 INFO [train.py:901] (3/4) Epoch 20, batch 5800, loss[loss=0.2635, simple_loss=0.3355, pruned_loss=0.09577, over 8515.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2916, pruned_loss=0.06447, over 1610076.81 frames. ], batch size: 28, lr: 3.77e-03, grad_scale: 8.0 +2023-02-07 01:05:35,572 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.629e+02 2.438e+02 2.992e+02 3.849e+02 1.447e+03, threshold=5.984e+02, percent-clipped=4.0 +2023-02-07 01:06:04,888 INFO [train.py:901] (3/4) Epoch 20, batch 5850, loss[loss=0.1792, simple_loss=0.2707, pruned_loss=0.04387, over 8290.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2925, pruned_loss=0.06472, over 1613521.75 frames. ], batch size: 23, lr: 3.77e-03, grad_scale: 8.0 +2023-02-07 01:06:08,469 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=159431.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:06:39,988 INFO [train.py:901] (3/4) Epoch 20, batch 5900, loss[loss=0.1785, simple_loss=0.2788, pruned_loss=0.03911, over 8299.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2903, pruned_loss=0.06349, over 1611385.29 frames. ], batch size: 23, lr: 3.77e-03, grad_scale: 8.0 +2023-02-07 01:06:45,387 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-07 01:06:45,624 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.574e+02 2.454e+02 2.951e+02 3.822e+02 7.063e+02, threshold=5.901e+02, percent-clipped=2.0 +2023-02-07 01:07:04,113 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159510.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:07:08,830 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159516.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:07:12,109 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=159521.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:07:15,456 INFO [train.py:901] (3/4) Epoch 20, batch 5950, loss[loss=0.255, simple_loss=0.3169, pruned_loss=0.09657, over 6897.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2913, pruned_loss=0.06387, over 1613999.34 frames. ], batch size: 71, lr: 3.77e-03, grad_scale: 8.0 +2023-02-07 01:07:15,837 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 +2023-02-07 01:07:22,429 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159535.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:07:26,381 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159541.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:07:29,669 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159546.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:07:42,011 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7727, 2.0561, 2.2235, 1.3510, 2.3251, 1.5725, 0.7114, 1.9475], + device='cuda:3'), covar=tensor([0.0675, 0.0350, 0.0291, 0.0625, 0.0405, 0.0858, 0.0896, 0.0327], + device='cuda:3'), in_proj_covar=tensor([0.0455, 0.0386, 0.0337, 0.0444, 0.0369, 0.0533, 0.0392, 0.0415], + device='cuda:3'), out_proj_covar=tensor([1.2254e-04, 1.0147e-04, 8.8862e-05, 1.1734e-04, 9.7634e-05, 1.5151e-04, + 1.0602e-04, 1.1062e-04], device='cuda:3') +2023-02-07 01:07:45,565 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159568.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 01:07:50,935 INFO [train.py:901] (3/4) Epoch 20, batch 6000, loss[loss=0.219, simple_loss=0.2866, pruned_loss=0.07571, over 8092.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2916, pruned_loss=0.06433, over 1611527.95 frames. ], batch size: 21, lr: 3.77e-03, grad_scale: 8.0 +2023-02-07 01:07:50,936 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 01:08:04,187 INFO [train.py:935] (3/4) Epoch 20, validation: loss=0.175, simple_loss=0.275, pruned_loss=0.03755, over 944034.00 frames. +2023-02-07 01:08:04,188 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-07 01:08:09,549 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.653e+02 2.504e+02 2.869e+02 3.482e+02 8.370e+02, threshold=5.739e+02, percent-clipped=5.0 +2023-02-07 01:08:15,897 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159593.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 01:08:38,922 INFO [train.py:901] (3/4) Epoch 20, batch 6050, loss[loss=0.1711, simple_loss=0.2632, pruned_loss=0.0395, over 8029.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2907, pruned_loss=0.06342, over 1613933.18 frames. ], batch size: 22, lr: 3.77e-03, grad_scale: 8.0 +2023-02-07 01:08:45,955 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159636.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:08:51,467 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3978, 1.4691, 1.3962, 1.8067, 0.7336, 1.2183, 1.2456, 1.4873], + device='cuda:3'), covar=tensor([0.0823, 0.0814, 0.0918, 0.0509, 0.1115, 0.1441, 0.0779, 0.0671], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0196, 0.0244, 0.0212, 0.0204, 0.0245, 0.0248, 0.0206], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 01:08:55,602 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159649.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:08:57,636 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=159652.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:09:14,771 INFO [train.py:901] (3/4) Epoch 20, batch 6100, loss[loss=0.2241, simple_loss=0.3225, pruned_loss=0.06289, over 8321.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2918, pruned_loss=0.06443, over 1610201.73 frames. ], batch size: 25, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:09:20,998 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.599e+02 2.453e+02 2.842e+02 3.745e+02 1.322e+03, threshold=5.684e+02, percent-clipped=4.0 +2023-02-07 01:09:35,012 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4970, 2.3204, 3.2205, 2.6036, 3.0529, 2.4772, 2.2345, 1.8241], + device='cuda:3'), covar=tensor([0.4819, 0.5039, 0.1835, 0.3382, 0.2340, 0.2962, 0.1757, 0.5284], + device='cuda:3'), in_proj_covar=tensor([0.0937, 0.0971, 0.0795, 0.0937, 0.0987, 0.0884, 0.0741, 0.0820], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 01:09:41,568 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-07 01:09:50,003 INFO [train.py:901] (3/4) Epoch 20, batch 6150, loss[loss=0.1814, simple_loss=0.2718, pruned_loss=0.04554, over 8457.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2925, pruned_loss=0.06442, over 1616904.35 frames. ], batch size: 27, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:10:18,367 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159767.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:10:24,871 INFO [train.py:901] (3/4) Epoch 20, batch 6200, loss[loss=0.2052, simple_loss=0.2915, pruned_loss=0.05938, over 7919.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2916, pruned_loss=0.0639, over 1617185.77 frames. ], batch size: 20, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:10:30,208 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.733e+02 2.429e+02 3.094e+02 3.753e+02 7.329e+02, threshold=6.188e+02, percent-clipped=3.0 +2023-02-07 01:10:31,129 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8423, 1.6766, 2.4408, 1.5460, 1.2641, 2.3390, 0.5263, 1.4287], + device='cuda:3'), covar=tensor([0.1675, 0.1472, 0.0368, 0.1396, 0.3138, 0.0458, 0.2456, 0.1557], + device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0195, 0.0125, 0.0221, 0.0270, 0.0133, 0.0169, 0.0189], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 01:10:43,484 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159802.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:10:55,864 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1881, 1.9367, 2.6791, 2.1503, 2.4954, 2.2223, 1.9782, 1.3866], + device='cuda:3'), covar=tensor([0.5637, 0.5021, 0.1856, 0.3720, 0.2595, 0.3017, 0.1996, 0.5489], + device='cuda:3'), in_proj_covar=tensor([0.0934, 0.0969, 0.0794, 0.0934, 0.0988, 0.0882, 0.0739, 0.0820], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 01:11:00,337 INFO [train.py:901] (3/4) Epoch 20, batch 6250, loss[loss=0.1992, simple_loss=0.2695, pruned_loss=0.06444, over 7227.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2907, pruned_loss=0.06344, over 1617189.81 frames. ], batch size: 16, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:11:01,219 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159827.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:11:32,486 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159873.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:11:34,374 INFO [train.py:901] (3/4) Epoch 20, batch 6300, loss[loss=0.1776, simple_loss=0.2606, pruned_loss=0.04736, over 8089.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2911, pruned_loss=0.06379, over 1615824.89 frames. ], batch size: 21, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:11:40,342 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.621e+02 2.354e+02 2.951e+02 3.644e+02 9.166e+02, threshold=5.902e+02, percent-clipped=5.0 +2023-02-07 01:11:45,865 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159892.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:12:00,084 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4528, 2.3262, 1.6734, 2.0722, 1.9840, 1.4067, 1.8940, 1.9700], + device='cuda:3'), covar=tensor([0.1435, 0.0408, 0.1246, 0.0638, 0.0724, 0.1625, 0.0951, 0.0945], + device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0238, 0.0332, 0.0310, 0.0301, 0.0338, 0.0345, 0.0316], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 01:12:03,495 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159917.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:12:09,073 INFO [train.py:901] (3/4) Epoch 20, batch 6350, loss[loss=0.2083, simple_loss=0.2813, pruned_loss=0.06763, over 7539.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2916, pruned_loss=0.06441, over 1616855.91 frames. ], batch size: 18, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:12:10,588 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159928.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:12:43,479 INFO [train.py:901] (3/4) Epoch 20, batch 6400, loss[loss=0.2429, simple_loss=0.3184, pruned_loss=0.08377, over 7093.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2918, pruned_loss=0.06425, over 1615964.83 frames. ], batch size: 71, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:12:45,737 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5813, 2.3479, 3.3039, 2.6157, 3.2374, 2.6456, 2.3332, 1.9996], + device='cuda:3'), covar=tensor([0.5236, 0.4911, 0.1919, 0.3544, 0.2312, 0.2778, 0.1776, 0.5171], + device='cuda:3'), in_proj_covar=tensor([0.0939, 0.0976, 0.0800, 0.0939, 0.0993, 0.0886, 0.0744, 0.0824], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 01:12:48,765 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.397e+02 2.436e+02 2.995e+02 3.881e+02 8.346e+02, threshold=5.989e+02, percent-clipped=6.0 +2023-02-07 01:12:55,755 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=159993.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:13:16,788 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160023.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:13:18,630 INFO [train.py:901] (3/4) Epoch 20, batch 6450, loss[loss=0.2244, simple_loss=0.3053, pruned_loss=0.07172, over 7524.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2914, pruned_loss=0.06421, over 1610682.46 frames. ], batch size: 18, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:13:34,496 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160048.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:13:49,644 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 +2023-02-07 01:13:53,995 INFO [train.py:901] (3/4) Epoch 20, batch 6500, loss[loss=0.2268, simple_loss=0.3116, pruned_loss=0.07101, over 8464.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2914, pruned_loss=0.06433, over 1613759.08 frames. ], batch size: 27, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:13:54,365 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-07 01:13:55,562 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2763, 2.0354, 1.5470, 1.9357, 1.7188, 1.3227, 1.6108, 1.6712], + device='cuda:3'), covar=tensor([0.1406, 0.0529, 0.1491, 0.0493, 0.0719, 0.1689, 0.0946, 0.0905], + device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0237, 0.0334, 0.0310, 0.0302, 0.0338, 0.0345, 0.0317], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 01:13:59,464 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.603e+02 2.613e+02 3.061e+02 4.120e+02 1.100e+03, threshold=6.122e+02, percent-clipped=8.0 +2023-02-07 01:14:16,523 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160108.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:14:20,060 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7410, 1.7848, 2.2852, 1.5349, 1.2548, 2.2231, 0.4200, 1.3971], + device='cuda:3'), covar=tensor([0.1961, 0.1181, 0.0380, 0.1218, 0.2842, 0.0436, 0.2339, 0.1347], + device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0195, 0.0126, 0.0221, 0.0271, 0.0133, 0.0169, 0.0189], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 01:14:24,659 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0363, 2.2239, 1.8593, 2.8301, 1.4002, 1.5671, 1.9867, 2.2142], + device='cuda:3'), covar=tensor([0.0737, 0.0764, 0.0890, 0.0389, 0.1168, 0.1391, 0.0971, 0.0777], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0196, 0.0245, 0.0212, 0.0203, 0.0246, 0.0248, 0.0206], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 01:14:29,720 INFO [train.py:901] (3/4) Epoch 20, batch 6550, loss[loss=0.2354, simple_loss=0.3227, pruned_loss=0.07407, over 8647.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2911, pruned_loss=0.06449, over 1609996.03 frames. ], batch size: 34, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:14:53,086 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-07 01:15:05,564 INFO [train.py:901] (3/4) Epoch 20, batch 6600, loss[loss=0.2241, simple_loss=0.3102, pruned_loss=0.06905, over 8502.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2911, pruned_loss=0.06401, over 1615460.09 frames. ], batch size: 29, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:15:10,796 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.865e+02 2.488e+02 3.067e+02 3.982e+02 8.719e+02, threshold=6.134e+02, percent-clipped=3.0 +2023-02-07 01:15:12,115 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-07 01:15:33,423 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=160217.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:15:39,331 INFO [train.py:901] (3/4) Epoch 20, batch 6650, loss[loss=0.217, simple_loss=0.2975, pruned_loss=0.06823, over 8459.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2913, pruned_loss=0.06377, over 1617188.84 frames. ], batch size: 27, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:16:12,499 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=160272.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:16:15,124 INFO [train.py:901] (3/4) Epoch 20, batch 6700, loss[loss=0.1903, simple_loss=0.2739, pruned_loss=0.0533, over 8085.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2909, pruned_loss=0.06379, over 1617883.41 frames. ], batch size: 21, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:16:20,313 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 +2023-02-07 01:16:20,502 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.602e+02 2.299e+02 2.819e+02 3.357e+02 8.975e+02, threshold=5.638e+02, percent-clipped=4.0 +2023-02-07 01:16:48,911 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160325.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 01:16:49,340 INFO [train.py:901] (3/4) Epoch 20, batch 6750, loss[loss=0.2956, simple_loss=0.3606, pruned_loss=0.1153, over 8491.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2918, pruned_loss=0.06447, over 1612291.03 frames. ], batch size: 26, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:16:53,604 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160332.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:17:09,921 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-07 01:17:16,340 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160364.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:17:22,340 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-07 01:17:23,988 INFO [train.py:901] (3/4) Epoch 20, batch 6800, loss[loss=0.2077, simple_loss=0.295, pruned_loss=0.06022, over 8032.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2911, pruned_loss=0.0639, over 1613178.50 frames. ], batch size: 22, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:17:28,103 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-07 01:17:29,319 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.714e+02 2.510e+02 3.096e+02 3.947e+02 9.727e+02, threshold=6.192e+02, percent-clipped=5.0 +2023-02-07 01:17:32,259 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160387.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:17:33,656 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160389.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:17:59,210 INFO [train.py:901] (3/4) Epoch 20, batch 6850, loss[loss=0.1855, simple_loss=0.2719, pruned_loss=0.04951, over 7968.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2899, pruned_loss=0.06367, over 1607803.69 frames. ], batch size: 21, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:18:05,667 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 +2023-02-07 01:18:18,941 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4148, 2.6893, 2.2661, 3.7807, 1.7159, 2.2513, 2.4446, 2.9410], + device='cuda:3'), covar=tensor([0.0726, 0.0845, 0.0835, 0.0345, 0.1128, 0.1192, 0.1005, 0.0681], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0197, 0.0245, 0.0213, 0.0205, 0.0247, 0.0250, 0.0207], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 01:18:19,449 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-07 01:18:34,189 INFO [train.py:901] (3/4) Epoch 20, batch 6900, loss[loss=0.1939, simple_loss=0.2758, pruned_loss=0.056, over 7924.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2893, pruned_loss=0.06317, over 1608288.04 frames. ], batch size: 20, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:18:39,573 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.365e+02 2.333e+02 2.912e+02 3.495e+02 9.213e+02, threshold=5.824e+02, percent-clipped=3.0 +2023-02-07 01:19:04,551 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-07 01:19:08,580 INFO [train.py:901] (3/4) Epoch 20, batch 6950, loss[loss=0.3009, simple_loss=0.3551, pruned_loss=0.1234, over 8475.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2893, pruned_loss=0.0629, over 1613306.01 frames. ], batch size: 27, lr: 3.76e-03, grad_scale: 8.0 +2023-02-07 01:19:12,204 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1683, 1.0430, 1.2888, 0.9986, 0.9417, 1.3321, 0.0496, 0.8824], + device='cuda:3'), covar=tensor([0.1594, 0.1367, 0.0464, 0.0844, 0.2756, 0.0510, 0.2220, 0.1254], + device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0195, 0.0126, 0.0222, 0.0271, 0.0134, 0.0169, 0.0189], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 01:19:16,603 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-07 01:19:30,214 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-07 01:19:31,076 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2496, 2.1253, 1.5735, 1.8192, 1.6724, 1.3683, 1.6236, 1.6217], + device='cuda:3'), covar=tensor([0.1429, 0.0447, 0.1329, 0.0637, 0.0813, 0.1616, 0.0990, 0.0915], + device='cuda:3'), in_proj_covar=tensor([0.0356, 0.0238, 0.0335, 0.0311, 0.0303, 0.0340, 0.0347, 0.0319], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 01:19:42,958 INFO [train.py:901] (3/4) Epoch 20, batch 7000, loss[loss=0.2164, simple_loss=0.2859, pruned_loss=0.07347, over 7810.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2881, pruned_loss=0.06249, over 1613928.43 frames. ], batch size: 20, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:19:48,348 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 2.497e+02 2.987e+02 3.377e+02 5.985e+02, threshold=5.974e+02, percent-clipped=1.0 +2023-02-07 01:19:49,239 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3772, 2.1368, 2.7895, 2.2984, 2.7846, 2.3427, 2.1362, 1.5378], + device='cuda:3'), covar=tensor([0.5047, 0.4964, 0.1971, 0.3723, 0.2379, 0.3042, 0.1865, 0.5677], + device='cuda:3'), in_proj_covar=tensor([0.0927, 0.0966, 0.0792, 0.0929, 0.0982, 0.0879, 0.0738, 0.0817], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 01:19:52,055 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160588.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:20:09,568 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160613.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:20:18,198 INFO [train.py:901] (3/4) Epoch 20, batch 7050, loss[loss=0.21, simple_loss=0.2794, pruned_loss=0.07033, over 5113.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2874, pruned_loss=0.06247, over 1607737.82 frames. ], batch size: 11, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:20:24,706 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-07 01:20:30,580 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160643.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:20:48,809 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160668.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:20:49,342 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=160669.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 01:20:53,995 INFO [train.py:901] (3/4) Epoch 20, batch 7100, loss[loss=0.2, simple_loss=0.2855, pruned_loss=0.05723, over 8087.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2875, pruned_loss=0.06274, over 1601602.75 frames. ], batch size: 21, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:20:59,623 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.602e+02 2.520e+02 2.814e+02 3.523e+02 7.232e+02, threshold=5.628e+02, percent-clipped=2.0 +2023-02-07 01:21:02,177 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-07 01:21:29,432 INFO [train.py:901] (3/4) Epoch 20, batch 7150, loss[loss=0.2079, simple_loss=0.2843, pruned_loss=0.06575, over 7923.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2879, pruned_loss=0.06303, over 1600879.09 frames. ], batch size: 20, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:21:30,763 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5781, 4.6074, 4.1382, 2.1091, 4.0878, 4.1785, 4.1740, 3.9947], + device='cuda:3'), covar=tensor([0.0653, 0.0485, 0.0939, 0.4349, 0.0818, 0.1064, 0.1178, 0.0975], + device='cuda:3'), in_proj_covar=tensor([0.0520, 0.0432, 0.0435, 0.0536, 0.0424, 0.0440, 0.0423, 0.0381], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 01:21:49,542 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2023-02-07 01:22:04,615 INFO [train.py:901] (3/4) Epoch 20, batch 7200, loss[loss=0.1924, simple_loss=0.2639, pruned_loss=0.06049, over 7804.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2895, pruned_loss=0.06388, over 1606682.71 frames. ], batch size: 19, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:22:09,775 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.612e+02 2.432e+02 3.066e+02 3.972e+02 8.502e+02, threshold=6.132e+02, percent-clipped=3.0 +2023-02-07 01:22:09,998 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160784.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 01:22:18,950 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 +2023-02-07 01:22:22,742 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0595, 1.6378, 1.3954, 1.5661, 1.3161, 1.2293, 1.3741, 1.2863], + device='cuda:3'), covar=tensor([0.0998, 0.0442, 0.1285, 0.0477, 0.0741, 0.1516, 0.0774, 0.0755], + device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0235, 0.0332, 0.0306, 0.0301, 0.0336, 0.0344, 0.0317], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 01:22:39,217 INFO [train.py:901] (3/4) Epoch 20, batch 7250, loss[loss=0.2004, simple_loss=0.2833, pruned_loss=0.05871, over 8128.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2889, pruned_loss=0.0639, over 1605378.43 frames. ], batch size: 22, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:23:13,935 INFO [train.py:901] (3/4) Epoch 20, batch 7300, loss[loss=0.1817, simple_loss=0.2642, pruned_loss=0.04964, over 8075.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2896, pruned_loss=0.06369, over 1607464.61 frames. ], batch size: 21, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:23:19,310 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.706e+02 2.519e+02 2.885e+02 3.982e+02 8.183e+02, threshold=5.771e+02, percent-clipped=5.0 +2023-02-07 01:23:28,337 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4664, 1.8355, 1.8923, 1.1722, 1.9826, 1.4182, 0.4108, 1.7493], + device='cuda:3'), covar=tensor([0.0556, 0.0338, 0.0295, 0.0556, 0.0358, 0.0891, 0.0872, 0.0264], + device='cuda:3'), in_proj_covar=tensor([0.0450, 0.0384, 0.0339, 0.0440, 0.0369, 0.0534, 0.0392, 0.0412], + device='cuda:3'), out_proj_covar=tensor([1.2107e-04, 1.0064e-04, 8.9304e-05, 1.1617e-04, 9.7327e-05, 1.5183e-04, + 1.0604e-04, 1.0947e-04], device='cuda:3') +2023-02-07 01:23:44,248 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160919.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:23:48,778 INFO [train.py:901] (3/4) Epoch 20, batch 7350, loss[loss=0.23, simple_loss=0.313, pruned_loss=0.07347, over 8611.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2887, pruned_loss=0.0631, over 1606002.62 frames. ], batch size: 34, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:23:51,525 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8358, 1.3972, 3.9957, 1.4551, 3.4797, 3.3004, 3.6361, 3.4871], + device='cuda:3'), covar=tensor([0.0610, 0.4677, 0.0558, 0.4132, 0.1181, 0.1024, 0.0623, 0.0745], + device='cuda:3'), in_proj_covar=tensor([0.0612, 0.0630, 0.0679, 0.0615, 0.0696, 0.0594, 0.0595, 0.0665], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 01:24:16,151 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-07 01:24:20,003 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 +2023-02-07 01:24:24,317 INFO [train.py:901] (3/4) Epoch 20, batch 7400, loss[loss=0.1921, simple_loss=0.2621, pruned_loss=0.06102, over 6355.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2878, pruned_loss=0.06219, over 1606324.42 frames. ], batch size: 14, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:24:29,887 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.7141, 5.8760, 5.1575, 2.4566, 5.2466, 5.5449, 5.3966, 5.3247], + device='cuda:3'), covar=tensor([0.0524, 0.0421, 0.0968, 0.4388, 0.0677, 0.0747, 0.1170, 0.0633], + device='cuda:3'), in_proj_covar=tensor([0.0520, 0.0432, 0.0433, 0.0536, 0.0427, 0.0438, 0.0421, 0.0380], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 01:24:29,932 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160983.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:24:30,419 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.621e+02 2.344e+02 3.002e+02 3.673e+02 6.079e+02, threshold=6.004e+02, percent-clipped=1.0 +2023-02-07 01:24:37,303 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-07 01:24:59,976 INFO [train.py:901] (3/4) Epoch 20, batch 7450, loss[loss=0.2, simple_loss=0.2852, pruned_loss=0.05745, over 8461.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2868, pruned_loss=0.06112, over 1608752.71 frames. ], batch size: 25, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:25:10,081 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161040.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 01:25:10,606 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7739, 1.6177, 2.8946, 1.3397, 2.1906, 3.0555, 3.2397, 2.5702], + device='cuda:3'), covar=tensor([0.1175, 0.1542, 0.0373, 0.2196, 0.0906, 0.0338, 0.0693, 0.0619], + device='cuda:3'), in_proj_covar=tensor([0.0293, 0.0319, 0.0284, 0.0313, 0.0303, 0.0261, 0.0410, 0.0301], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 01:25:16,129 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-07 01:25:28,627 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161065.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 01:25:35,936 INFO [train.py:901] (3/4) Epoch 20, batch 7500, loss[loss=0.2437, simple_loss=0.3177, pruned_loss=0.08487, over 7154.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2874, pruned_loss=0.06198, over 1606682.97 frames. ], batch size: 72, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:25:41,428 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.792e+02 2.441e+02 3.010e+02 3.756e+02 8.900e+02, threshold=6.020e+02, percent-clipped=5.0 +2023-02-07 01:26:11,145 INFO [train.py:901] (3/4) Epoch 20, batch 7550, loss[loss=0.2251, simple_loss=0.294, pruned_loss=0.07813, over 7426.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2883, pruned_loss=0.06234, over 1609891.26 frames. ], batch size: 17, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:26:23,187 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-07 01:26:46,339 INFO [train.py:901] (3/4) Epoch 20, batch 7600, loss[loss=0.1977, simple_loss=0.2811, pruned_loss=0.05719, over 8107.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.289, pruned_loss=0.06265, over 1608808.85 frames. ], batch size: 23, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:26:51,651 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.489e+02 2.460e+02 3.037e+02 4.113e+02 9.859e+02, threshold=6.074e+02, percent-clipped=9.0 +2023-02-07 01:26:57,672 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5348, 2.1618, 3.1496, 1.6927, 1.6050, 3.0425, 0.8274, 2.0432], + device='cuda:3'), covar=tensor([0.1672, 0.1256, 0.0317, 0.1873, 0.2918, 0.0611, 0.2334, 0.1505], + device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0194, 0.0125, 0.0220, 0.0269, 0.0134, 0.0168, 0.0188], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 01:27:20,295 INFO [train.py:901] (3/4) Epoch 20, batch 7650, loss[loss=0.1647, simple_loss=0.2499, pruned_loss=0.03974, over 7697.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.289, pruned_loss=0.06266, over 1609526.01 frames. ], batch size: 18, lr: 3.75e-03, grad_scale: 8.0 +2023-02-07 01:27:25,754 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161234.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:27:36,306 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161249.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:27:45,609 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=161263.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:27:54,270 INFO [train.py:901] (3/4) Epoch 20, batch 7700, loss[loss=0.1871, simple_loss=0.2726, pruned_loss=0.05075, over 8333.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2909, pruned_loss=0.0638, over 1612187.20 frames. ], batch size: 26, lr: 3.75e-03, grad_scale: 16.0 +2023-02-07 01:27:59,457 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.538e+02 2.411e+02 2.987e+02 3.572e+02 6.786e+02, threshold=5.975e+02, percent-clipped=3.0 +2023-02-07 01:28:25,738 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-07 01:28:29,788 INFO [train.py:901] (3/4) Epoch 20, batch 7750, loss[loss=0.2096, simple_loss=0.2999, pruned_loss=0.05961, over 8296.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.29, pruned_loss=0.06385, over 1607927.98 frames. ], batch size: 23, lr: 3.75e-03, grad_scale: 16.0 +2023-02-07 01:28:30,572 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=161327.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:29:05,271 INFO [train.py:901] (3/4) Epoch 20, batch 7800, loss[loss=0.2269, simple_loss=0.3137, pruned_loss=0.07002, over 8699.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2904, pruned_loss=0.0636, over 1613308.87 frames. ], batch size: 49, lr: 3.75e-03, grad_scale: 16.0 +2023-02-07 01:29:06,855 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161378.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:29:10,616 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.672e+02 2.429e+02 2.909e+02 3.732e+02 6.331e+02, threshold=5.818e+02, percent-clipped=2.0 +2023-02-07 01:29:30,832 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.98 vs. limit=5.0 +2023-02-07 01:29:34,443 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161419.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:29:39,043 INFO [train.py:901] (3/4) Epoch 20, batch 7850, loss[loss=0.1939, simple_loss=0.2773, pruned_loss=0.05524, over 7793.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2898, pruned_loss=0.06359, over 1613007.03 frames. ], batch size: 19, lr: 3.74e-03, grad_scale: 16.0 +2023-02-07 01:29:49,660 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161442.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:30:12,438 INFO [train.py:901] (3/4) Epoch 20, batch 7900, loss[loss=0.2114, simple_loss=0.2987, pruned_loss=0.06203, over 8516.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2899, pruned_loss=0.06381, over 1610859.12 frames. ], batch size: 26, lr: 3.74e-03, grad_scale: 8.0 +2023-02-07 01:30:13,739 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161478.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:30:18,875 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.320e+02 2.352e+02 2.923e+02 4.060e+02 8.940e+02, threshold=5.846e+02, percent-clipped=3.0 +2023-02-07 01:30:45,510 INFO [train.py:901] (3/4) Epoch 20, batch 7950, loss[loss=0.2114, simple_loss=0.2942, pruned_loss=0.06436, over 8641.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2908, pruned_loss=0.06369, over 1613967.45 frames. ], batch size: 39, lr: 3.74e-03, grad_scale: 8.0 +2023-02-07 01:31:18,030 INFO [train.py:901] (3/4) Epoch 20, batch 8000, loss[loss=0.2115, simple_loss=0.2983, pruned_loss=0.06236, over 8100.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2911, pruned_loss=0.06412, over 1612185.59 frames. ], batch size: 23, lr: 3.74e-03, grad_scale: 8.0 +2023-02-07 01:31:19,439 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=161578.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:31:23,859 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.454e+02 2.449e+02 3.108e+02 3.740e+02 8.675e+02, threshold=6.215e+02, percent-clipped=6.0 +2023-02-07 01:31:29,415 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=161593.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:31:51,325 INFO [train.py:901] (3/4) Epoch 20, batch 8050, loss[loss=0.267, simple_loss=0.338, pruned_loss=0.09802, over 7078.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2896, pruned_loss=0.06327, over 1604803.51 frames. ], batch size: 71, lr: 3.74e-03, grad_scale: 8.0 +2023-02-07 01:31:57,140 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161634.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:32:25,034 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-07 01:32:32,190 INFO [train.py:901] (3/4) Epoch 21, batch 0, loss[loss=0.1906, simple_loss=0.2642, pruned_loss=0.05849, over 7795.00 frames. ], tot_loss[loss=0.1906, simple_loss=0.2642, pruned_loss=0.05849, over 7795.00 frames. ], batch size: 19, lr: 3.65e-03, grad_scale: 8.0 +2023-02-07 01:32:32,191 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 01:32:44,208 INFO [train.py:935] (3/4) Epoch 21, validation: loss=0.1763, simple_loss=0.2762, pruned_loss=0.03818, over 944034.00 frames. +2023-02-07 01:32:44,209 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-07 01:32:44,415 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161659.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:32:59,367 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-07 01:33:02,227 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.382e+02 2.415e+02 2.918e+02 3.924e+02 7.413e+02, threshold=5.835e+02, percent-clipped=4.0 +2023-02-07 01:33:07,973 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161693.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:33:11,610 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161698.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:33:18,541 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9778, 1.4283, 3.1166, 1.5126, 2.7504, 2.6726, 2.8854, 2.8178], + device='cuda:3'), covar=tensor([0.0787, 0.3572, 0.0970, 0.3741, 0.1163, 0.0939, 0.0649, 0.0724], + device='cuda:3'), in_proj_covar=tensor([0.0610, 0.0623, 0.0673, 0.0608, 0.0690, 0.0590, 0.0591, 0.0656], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 01:33:18,588 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161708.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:33:19,108 INFO [train.py:901] (3/4) Epoch 21, batch 50, loss[loss=0.2562, simple_loss=0.3316, pruned_loss=0.09043, over 6692.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2923, pruned_loss=0.06673, over 364698.48 frames. ], batch size: 71, lr: 3.65e-03, grad_scale: 8.0 +2023-02-07 01:33:29,249 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161723.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:33:32,475 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-07 01:33:56,008 INFO [train.py:901] (3/4) Epoch 21, batch 100, loss[loss=0.2384, simple_loss=0.3166, pruned_loss=0.08012, over 8188.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2934, pruned_loss=0.06598, over 642771.43 frames. ], batch size: 23, lr: 3.65e-03, grad_scale: 8.0 +2023-02-07 01:33:57,255 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-07 01:33:58,658 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=161763.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:34:14,160 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.548e+02 2.511e+02 2.964e+02 4.065e+02 7.207e+02, threshold=5.927e+02, percent-clipped=4.0 +2023-02-07 01:34:18,655 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-07 01:34:30,768 INFO [train.py:901] (3/4) Epoch 21, batch 150, loss[loss=0.2042, simple_loss=0.2892, pruned_loss=0.05959, over 8686.00 frames. ], tot_loss[loss=0.2111, simple_loss=0.2925, pruned_loss=0.06481, over 859084.86 frames. ], batch size: 39, lr: 3.65e-03, grad_scale: 8.0 +2023-02-07 01:34:33,277 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-07 01:34:39,725 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=161822.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:34:47,267 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161833.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:35:06,325 INFO [train.py:901] (3/4) Epoch 21, batch 200, loss[loss=0.2115, simple_loss=0.3026, pruned_loss=0.06019, over 8228.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.293, pruned_loss=0.06558, over 1022462.16 frames. ], batch size: 22, lr: 3.65e-03, grad_scale: 8.0 +2023-02-07 01:35:19,121 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161878.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:35:23,716 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.740e+02 2.494e+02 2.791e+02 3.613e+02 7.338e+02, threshold=5.582e+02, percent-clipped=1.0 +2023-02-07 01:35:41,056 INFO [train.py:901] (3/4) Epoch 21, batch 250, loss[loss=0.2105, simple_loss=0.3029, pruned_loss=0.05904, over 8343.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.2921, pruned_loss=0.06532, over 1152197.25 frames. ], batch size: 25, lr: 3.65e-03, grad_scale: 8.0 +2023-02-07 01:35:47,951 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-07 01:35:57,070 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-07 01:36:00,707 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161937.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:36:08,760 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161949.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:36:15,273 INFO [train.py:901] (3/4) Epoch 21, batch 300, loss[loss=0.2137, simple_loss=0.286, pruned_loss=0.07071, over 8315.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2924, pruned_loss=0.06507, over 1257244.23 frames. ], batch size: 25, lr: 3.65e-03, grad_scale: 8.0 +2023-02-07 01:36:19,027 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161964.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:36:20,642 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-07 01:36:26,620 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161974.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:36:33,765 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.651e+02 2.407e+02 2.839e+02 3.558e+02 8.067e+02, threshold=5.678e+02, percent-clipped=5.0 +2023-02-07 01:36:36,647 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161989.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:36:51,875 INFO [train.py:901] (3/4) Epoch 21, batch 350, loss[loss=0.2583, simple_loss=0.3322, pruned_loss=0.0922, over 8495.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.291, pruned_loss=0.06379, over 1338754.49 frames. ], batch size: 26, lr: 3.65e-03, grad_scale: 8.0 +2023-02-07 01:36:57,892 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-07 01:37:25,807 INFO [train.py:901] (3/4) Epoch 21, batch 400, loss[loss=0.2118, simple_loss=0.3056, pruned_loss=0.05895, over 8441.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2906, pruned_loss=0.06407, over 1400170.25 frames. ], batch size: 25, lr: 3.65e-03, grad_scale: 8.0 +2023-02-07 01:37:44,475 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.641e+02 2.323e+02 2.796e+02 3.394e+02 5.024e+02, threshold=5.592e+02, percent-clipped=0.0 +2023-02-07 01:37:52,930 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162095.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:38:02,159 INFO [train.py:901] (3/4) Epoch 21, batch 450, loss[loss=0.2069, simple_loss=0.2936, pruned_loss=0.06015, over 8026.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2905, pruned_loss=0.06415, over 1449267.45 frames. ], batch size: 22, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:38:15,016 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-07 01:38:20,160 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162134.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:38:37,315 INFO [train.py:901] (3/4) Epoch 21, batch 500, loss[loss=0.2097, simple_loss=0.2901, pruned_loss=0.06465, over 8438.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2903, pruned_loss=0.0635, over 1491404.17 frames. ], batch size: 49, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:38:37,566 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162159.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:38:50,067 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=162177.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:38:55,581 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.556e+02 2.501e+02 2.975e+02 3.750e+02 9.376e+02, threshold=5.950e+02, percent-clipped=8.0 +2023-02-07 01:39:01,442 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162193.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:39:13,726 INFO [train.py:901] (3/4) Epoch 21, batch 550, loss[loss=0.1825, simple_loss=0.2634, pruned_loss=0.05083, over 7651.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2907, pruned_loss=0.06345, over 1523568.19 frames. ], batch size: 19, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:39:20,161 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162218.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:39:36,019 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8667, 3.7768, 3.4590, 1.7037, 3.3814, 3.4512, 3.4129, 3.2589], + device='cuda:3'), covar=tensor([0.0920, 0.0713, 0.1216, 0.5146, 0.1067, 0.1265, 0.1389, 0.0972], + device='cuda:3'), in_proj_covar=tensor([0.0515, 0.0432, 0.0431, 0.0532, 0.0421, 0.0435, 0.0414, 0.0377], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 01:39:42,296 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162249.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:39:48,783 INFO [train.py:901] (3/4) Epoch 21, batch 600, loss[loss=0.2355, simple_loss=0.3142, pruned_loss=0.07842, over 8287.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2906, pruned_loss=0.06393, over 1539165.04 frames. ], batch size: 23, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:39:49,653 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5275, 1.4593, 4.7526, 1.8027, 4.2468, 3.9677, 4.3158, 4.1645], + device='cuda:3'), covar=tensor([0.0566, 0.4475, 0.0435, 0.3690, 0.0936, 0.0920, 0.0541, 0.0608], + device='cuda:3'), in_proj_covar=tensor([0.0618, 0.0635, 0.0684, 0.0616, 0.0700, 0.0601, 0.0600, 0.0667], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 01:40:02,423 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-07 01:40:06,587 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.684e+02 2.365e+02 2.932e+02 3.412e+02 7.385e+02, threshold=5.863e+02, percent-clipped=2.0 +2023-02-07 01:40:08,827 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2584, 2.5785, 3.0501, 1.5906, 3.3683, 2.0633, 1.4933, 2.0758], + device='cuda:3'), covar=tensor([0.0836, 0.0409, 0.0263, 0.0813, 0.0381, 0.0796, 0.0969, 0.0636], + device='cuda:3'), in_proj_covar=tensor([0.0447, 0.0383, 0.0335, 0.0437, 0.0367, 0.0530, 0.0387, 0.0411], + device='cuda:3'), out_proj_covar=tensor([1.2021e-04, 1.0044e-04, 8.8259e-05, 1.1543e-04, 9.6905e-05, 1.5057e-04, + 1.0474e-04, 1.0932e-04], device='cuda:3') +2023-02-07 01:40:11,435 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162292.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:40:22,966 INFO [train.py:901] (3/4) Epoch 21, batch 650, loss[loss=0.1969, simple_loss=0.2806, pruned_loss=0.0566, over 8463.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2904, pruned_loss=0.06374, over 1557183.85 frames. ], batch size: 29, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:40:59,244 INFO [train.py:901] (3/4) Epoch 21, batch 700, loss[loss=0.2106, simple_loss=0.2954, pruned_loss=0.06295, over 8095.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.289, pruned_loss=0.06308, over 1571540.48 frames. ], batch size: 21, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:41:17,771 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.477e+02 2.443e+02 3.111e+02 4.032e+02 8.821e+02, threshold=6.222e+02, percent-clipped=5.0 +2023-02-07 01:41:34,560 INFO [train.py:901] (3/4) Epoch 21, batch 750, loss[loss=0.1918, simple_loss=0.2823, pruned_loss=0.05067, over 8331.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2891, pruned_loss=0.06297, over 1584371.75 frames. ], batch size: 26, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:41:40,438 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-07 01:41:45,476 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-07 01:41:54,312 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-07 01:41:56,381 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=162439.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:42:01,036 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 +2023-02-07 01:42:11,060 INFO [train.py:901] (3/4) Epoch 21, batch 800, loss[loss=0.1825, simple_loss=0.2763, pruned_loss=0.04433, over 8371.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2907, pruned_loss=0.06343, over 1594838.13 frames. ], batch size: 24, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:42:29,948 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.648e+02 2.455e+02 2.861e+02 3.570e+02 7.084e+02, threshold=5.721e+02, percent-clipped=3.0 +2023-02-07 01:42:47,158 INFO [train.py:901] (3/4) Epoch 21, batch 850, loss[loss=0.2286, simple_loss=0.3085, pruned_loss=0.0744, over 8539.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.291, pruned_loss=0.06393, over 1599913.53 frames. ], batch size: 49, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:43:01,483 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162529.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:43:07,213 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2786, 2.1616, 2.8466, 2.2845, 2.7500, 2.3698, 2.0895, 1.5047], + device='cuda:3'), covar=tensor([0.5343, 0.4872, 0.1853, 0.3660, 0.2228, 0.2913, 0.1925, 0.5147], + device='cuda:3'), in_proj_covar=tensor([0.0936, 0.0967, 0.0791, 0.0929, 0.0983, 0.0878, 0.0738, 0.0815], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 01:43:16,274 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162548.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:43:21,103 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162554.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:43:24,455 INFO [train.py:901] (3/4) Epoch 21, batch 900, loss[loss=0.2246, simple_loss=0.3087, pruned_loss=0.07026, over 8037.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2907, pruned_loss=0.0631, over 1609091.54 frames. ], batch size: 22, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:43:34,388 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162573.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:43:42,639 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.507e+02 2.319e+02 2.838e+02 3.637e+02 1.203e+03, threshold=5.677e+02, percent-clipped=5.0 +2023-02-07 01:43:49,236 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=162593.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:44:00,581 INFO [train.py:901] (3/4) Epoch 21, batch 950, loss[loss=0.2005, simple_loss=0.2821, pruned_loss=0.05943, over 8124.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2891, pruned_loss=0.06222, over 1612486.00 frames. ], batch size: 22, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:44:07,073 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6620, 1.9450, 2.8628, 1.4383, 2.1450, 1.9242, 1.7456, 2.0041], + device='cuda:3'), covar=tensor([0.1868, 0.2528, 0.0948, 0.4577, 0.1800, 0.3332, 0.2245, 0.2441], + device='cuda:3'), in_proj_covar=tensor([0.0520, 0.0592, 0.0550, 0.0632, 0.0639, 0.0591, 0.0528, 0.0629], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 01:44:14,218 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-07 01:44:18,637 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5349, 2.3121, 3.3298, 2.5882, 3.0993, 2.4987, 2.3143, 1.9527], + device='cuda:3'), covar=tensor([0.5109, 0.4952, 0.1691, 0.3676, 0.2437, 0.2838, 0.1776, 0.5224], + device='cuda:3'), in_proj_covar=tensor([0.0939, 0.0970, 0.0793, 0.0931, 0.0985, 0.0879, 0.0740, 0.0818], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 01:44:35,839 INFO [train.py:901] (3/4) Epoch 21, batch 1000, loss[loss=0.1965, simple_loss=0.2745, pruned_loss=0.05925, over 8138.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2899, pruned_loss=0.06249, over 1615832.25 frames. ], batch size: 22, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:44:48,951 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-07 01:44:55,205 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.590e+02 2.438e+02 2.954e+02 4.014e+02 9.557e+02, threshold=5.908e+02, percent-clipped=3.0 +2023-02-07 01:45:01,391 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-07 01:45:11,707 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162708.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:45:12,207 INFO [train.py:901] (3/4) Epoch 21, batch 1050, loss[loss=0.2211, simple_loss=0.3045, pruned_loss=0.06881, over 8483.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.289, pruned_loss=0.06214, over 1613855.12 frames. ], batch size: 29, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:45:46,489 INFO [train.py:901] (3/4) Epoch 21, batch 1100, loss[loss=0.1888, simple_loss=0.2745, pruned_loss=0.05149, over 5909.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2886, pruned_loss=0.06221, over 1611510.90 frames. ], batch size: 13, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:46:06,018 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.737e+02 2.501e+02 3.059e+02 3.494e+02 1.150e+03, threshold=6.119e+02, percent-clipped=4.0 +2023-02-07 01:46:14,530 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-07 01:46:23,789 INFO [train.py:901] (3/4) Epoch 21, batch 1150, loss[loss=0.2178, simple_loss=0.2809, pruned_loss=0.07742, over 7656.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2879, pruned_loss=0.06186, over 1612070.70 frames. ], batch size: 19, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:46:24,672 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162810.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:46:29,077 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-07 01:46:29,779 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-07 01:46:42,983 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162835.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:46:50,071 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162845.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:46:59,719 INFO [train.py:901] (3/4) Epoch 21, batch 1200, loss[loss=0.1699, simple_loss=0.2442, pruned_loss=0.04779, over 7813.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2877, pruned_loss=0.06177, over 1612201.00 frames. ], batch size: 19, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:47:09,539 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=162873.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:47:17,459 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.784e+02 2.368e+02 3.051e+02 3.779e+02 6.869e+02, threshold=6.103e+02, percent-clipped=3.0 +2023-02-07 01:47:36,404 INFO [train.py:901] (3/4) Epoch 21, batch 1250, loss[loss=0.2251, simple_loss=0.3124, pruned_loss=0.06891, over 8281.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2889, pruned_loss=0.062, over 1615124.45 frames. ], batch size: 23, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:48:11,285 INFO [train.py:901] (3/4) Epoch 21, batch 1300, loss[loss=0.1577, simple_loss=0.2302, pruned_loss=0.04263, over 7219.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2883, pruned_loss=0.06133, over 1614843.71 frames. ], batch size: 16, lr: 3.64e-03, grad_scale: 8.0 +2023-02-07 01:48:14,770 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162964.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:48:27,747 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 +2023-02-07 01:48:28,485 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.689e+02 2.260e+02 2.727e+02 3.317e+02 5.773e+02, threshold=5.453e+02, percent-clipped=0.0 +2023-02-07 01:48:29,597 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.82 vs. limit=5.0 +2023-02-07 01:48:30,748 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162988.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:48:31,454 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162989.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:48:44,803 INFO [train.py:901] (3/4) Epoch 21, batch 1350, loss[loss=0.2348, simple_loss=0.3093, pruned_loss=0.08016, over 8506.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2892, pruned_loss=0.06204, over 1617829.04 frames. ], batch size: 26, lr: 3.63e-03, grad_scale: 8.0 +2023-02-07 01:48:45,179 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-07 01:49:09,957 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 +2023-02-07 01:49:21,732 INFO [train.py:901] (3/4) Epoch 21, batch 1400, loss[loss=0.1944, simple_loss=0.2655, pruned_loss=0.06162, over 6801.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2888, pruned_loss=0.06224, over 1617057.38 frames. ], batch size: 15, lr: 3.63e-03, grad_scale: 8.0 +2023-02-07 01:49:39,406 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.676e+02 2.469e+02 3.010e+02 4.050e+02 1.060e+03, threshold=6.020e+02, percent-clipped=5.0 +2023-02-07 01:49:43,071 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3689, 2.7826, 2.2230, 3.9031, 1.7324, 2.1861, 2.4217, 2.8234], + device='cuda:3'), covar=tensor([0.0705, 0.0820, 0.0847, 0.0257, 0.1131, 0.1196, 0.1011, 0.0808], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0196, 0.0245, 0.0213, 0.0205, 0.0246, 0.0250, 0.0208], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 01:49:46,305 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-07 01:49:55,443 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=163108.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:49:55,931 INFO [train.py:901] (3/4) Epoch 21, batch 1450, loss[loss=0.2436, simple_loss=0.3111, pruned_loss=0.08803, over 8454.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2884, pruned_loss=0.06221, over 1615121.02 frames. ], batch size: 27, lr: 3.63e-03, grad_scale: 8.0 +2023-02-07 01:50:28,463 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-07 01:50:32,151 INFO [train.py:901] (3/4) Epoch 21, batch 1500, loss[loss=0.2076, simple_loss=0.2911, pruned_loss=0.06205, over 8317.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2893, pruned_loss=0.06226, over 1617778.84 frames. ], batch size: 25, lr: 3.63e-03, grad_scale: 8.0 +2023-02-07 01:50:50,508 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.589e+02 2.250e+02 2.722e+02 3.392e+02 6.898e+02, threshold=5.444e+02, percent-clipped=4.0 +2023-02-07 01:50:53,288 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=163189.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:51:06,788 INFO [train.py:901] (3/4) Epoch 21, batch 1550, loss[loss=0.1988, simple_loss=0.2837, pruned_loss=0.057, over 8342.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.288, pruned_loss=0.06186, over 1617830.41 frames. ], batch size: 25, lr: 3.63e-03, grad_scale: 8.0 +2023-02-07 01:51:31,323 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163244.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:51:42,268 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=163258.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:51:42,810 INFO [train.py:901] (3/4) Epoch 21, batch 1600, loss[loss=0.24, simple_loss=0.3272, pruned_loss=0.07635, over 8209.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2893, pruned_loss=0.06224, over 1618154.69 frames. ], batch size: 23, lr: 3.63e-03, grad_scale: 8.0 +2023-02-07 01:51:50,625 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163269.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:52:00,874 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.434e+02 2.380e+02 3.009e+02 4.081e+02 9.131e+02, threshold=6.018e+02, percent-clipped=6.0 +2023-02-07 01:52:14,547 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=163304.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:52:16,022 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.50 vs. limit=2.0 +2023-02-07 01:52:17,726 INFO [train.py:901] (3/4) Epoch 21, batch 1650, loss[loss=0.2282, simple_loss=0.3015, pruned_loss=0.07739, over 8095.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2905, pruned_loss=0.06316, over 1622375.46 frames. ], batch size: 21, lr: 3.63e-03, grad_scale: 8.0 +2023-02-07 01:52:33,941 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 +2023-02-07 01:52:51,343 INFO [train.py:901] (3/4) Epoch 21, batch 1700, loss[loss=0.2124, simple_loss=0.2976, pruned_loss=0.06358, over 8567.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2907, pruned_loss=0.06362, over 1621539.07 frames. ], batch size: 39, lr: 3.63e-03, grad_scale: 8.0 +2023-02-07 01:53:09,964 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.537e+02 2.344e+02 2.897e+02 3.678e+02 1.033e+03, threshold=5.793e+02, percent-clipped=5.0 +2023-02-07 01:53:27,424 INFO [train.py:901] (3/4) Epoch 21, batch 1750, loss[loss=0.2166, simple_loss=0.2991, pruned_loss=0.06699, over 8078.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2913, pruned_loss=0.06434, over 1624457.61 frames. ], batch size: 21, lr: 3.63e-03, grad_scale: 8.0 +2023-02-07 01:53:56,335 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=163452.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:54:01,076 INFO [train.py:901] (3/4) Epoch 21, batch 1800, loss[loss=0.1859, simple_loss=0.2822, pruned_loss=0.04478, over 8192.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2897, pruned_loss=0.06326, over 1620676.09 frames. ], batch size: 23, lr: 3.63e-03, grad_scale: 16.0 +2023-02-07 01:54:18,722 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.615e+02 2.661e+02 3.025e+02 4.067e+02 7.408e+02, threshold=6.049e+02, percent-clipped=6.0 +2023-02-07 01:54:18,997 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6878, 2.0932, 3.3439, 1.5222, 2.4391, 2.1809, 1.7899, 2.5225], + device='cuda:3'), covar=tensor([0.1790, 0.2662, 0.0777, 0.4449, 0.1795, 0.3083, 0.2252, 0.2209], + device='cuda:3'), in_proj_covar=tensor([0.0520, 0.0592, 0.0550, 0.0633, 0.0640, 0.0589, 0.0528, 0.0628], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 01:54:37,342 INFO [train.py:901] (3/4) Epoch 21, batch 1850, loss[loss=0.1847, simple_loss=0.2716, pruned_loss=0.0489, over 8063.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2883, pruned_loss=0.06281, over 1620716.99 frames. ], batch size: 21, lr: 3.63e-03, grad_scale: 16.0 +2023-02-07 01:54:53,885 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4132, 1.7849, 1.4498, 2.8162, 1.2101, 1.2730, 1.9618, 1.9108], + device='cuda:3'), covar=tensor([0.1633, 0.1231, 0.1864, 0.0386, 0.1338, 0.2104, 0.1018, 0.1099], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0196, 0.0244, 0.0212, 0.0204, 0.0245, 0.0250, 0.0207], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 01:55:11,697 INFO [train.py:901] (3/4) Epoch 21, batch 1900, loss[loss=0.1984, simple_loss=0.2895, pruned_loss=0.05369, over 8508.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2865, pruned_loss=0.06165, over 1614572.30 frames. ], batch size: 29, lr: 3.63e-03, grad_scale: 16.0 +2023-02-07 01:55:12,593 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163560.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:55:17,239 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=163567.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:55:21,422 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.29 vs. limit=5.0 +2023-02-07 01:55:26,434 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-07 01:55:29,012 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.695e+02 2.410e+02 2.798e+02 3.588e+02 7.290e+02, threshold=5.595e+02, percent-clipped=1.0 +2023-02-07 01:55:29,219 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163585.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:55:35,269 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7422, 2.0969, 3.4214, 1.8240, 1.6916, 3.3481, 0.6509, 2.0454], + device='cuda:3'), covar=tensor([0.1530, 0.1536, 0.0242, 0.1830, 0.2774, 0.0320, 0.2481, 0.1548], + device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0194, 0.0126, 0.0219, 0.0270, 0.0132, 0.0168, 0.0188], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 01:55:35,900 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4143, 1.6078, 2.1277, 1.2753, 1.4619, 1.7172, 1.3971, 1.4630], + device='cuda:3'), covar=tensor([0.1960, 0.2466, 0.0931, 0.4536, 0.2011, 0.3286, 0.2445, 0.2256], + device='cuda:3'), in_proj_covar=tensor([0.0520, 0.0592, 0.0550, 0.0634, 0.0639, 0.0590, 0.0528, 0.0629], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 01:55:37,702 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-07 01:55:40,619 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=163602.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:55:42,636 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8860, 3.3892, 1.7379, 2.5819, 2.5200, 1.6189, 2.4514, 2.8796], + device='cuda:3'), covar=tensor([0.1656, 0.0397, 0.1589, 0.0829, 0.0890, 0.1850, 0.1218, 0.0922], + device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0234, 0.0332, 0.0306, 0.0296, 0.0333, 0.0344, 0.0317], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 01:55:45,160 INFO [train.py:901] (3/4) Epoch 21, batch 1950, loss[loss=0.2075, simple_loss=0.2879, pruned_loss=0.06357, over 8278.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2868, pruned_loss=0.06176, over 1615827.50 frames. ], batch size: 23, lr: 3.63e-03, grad_scale: 16.0 +2023-02-07 01:55:58,525 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-07 01:56:04,067 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.7090, 4.6653, 4.1730, 2.1589, 4.1310, 4.3424, 4.3134, 4.2361], + device='cuda:3'), covar=tensor([0.0755, 0.0542, 0.1101, 0.4672, 0.0838, 0.0849, 0.1218, 0.0743], + device='cuda:3'), in_proj_covar=tensor([0.0513, 0.0428, 0.0429, 0.0534, 0.0420, 0.0435, 0.0416, 0.0378], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 01:56:05,942 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 +2023-02-07 01:56:14,480 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.7881, 5.8473, 5.0382, 2.6875, 5.1213, 5.6627, 5.3839, 5.4402], + device='cuda:3'), covar=tensor([0.0483, 0.0428, 0.0841, 0.3804, 0.0678, 0.0835, 0.1012, 0.0552], + device='cuda:3'), in_proj_covar=tensor([0.0513, 0.0428, 0.0429, 0.0534, 0.0421, 0.0435, 0.0416, 0.0379], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 01:56:21,641 INFO [train.py:901] (3/4) Epoch 21, batch 2000, loss[loss=0.2383, simple_loss=0.3025, pruned_loss=0.08702, over 8250.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2878, pruned_loss=0.06218, over 1619690.86 frames. ], batch size: 22, lr: 3.63e-03, grad_scale: 16.0 +2023-02-07 01:56:39,059 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.408e+02 2.546e+02 3.013e+02 3.975e+02 6.874e+02, threshold=6.025e+02, percent-clipped=4.0 +2023-02-07 01:56:55,192 INFO [train.py:901] (3/4) Epoch 21, batch 2050, loss[loss=0.2284, simple_loss=0.3082, pruned_loss=0.07433, over 8292.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2898, pruned_loss=0.06366, over 1618380.60 frames. ], batch size: 23, lr: 3.63e-03, grad_scale: 16.0 +2023-02-07 01:57:00,616 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=163717.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:57:30,311 INFO [train.py:901] (3/4) Epoch 21, batch 2100, loss[loss=0.1886, simple_loss=0.2674, pruned_loss=0.05488, over 8080.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2899, pruned_loss=0.06343, over 1619069.87 frames. ], batch size: 21, lr: 3.63e-03, grad_scale: 16.0 +2023-02-07 01:57:48,636 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.816e+02 2.575e+02 2.946e+02 3.630e+02 8.805e+02, threshold=5.893e+02, percent-clipped=3.0 +2023-02-07 01:58:04,871 INFO [train.py:901] (3/4) Epoch 21, batch 2150, loss[loss=0.1999, simple_loss=0.2627, pruned_loss=0.06853, over 7175.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2894, pruned_loss=0.06346, over 1613645.30 frames. ], batch size: 16, lr: 3.63e-03, grad_scale: 16.0 +2023-02-07 01:58:14,693 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163823.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:58:25,291 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0771, 1.3082, 1.6345, 1.3825, 1.0083, 1.4775, 1.7709, 1.4668], + device='cuda:3'), covar=tensor([0.0525, 0.1401, 0.1778, 0.1508, 0.0640, 0.1576, 0.0736, 0.0736], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0153, 0.0190, 0.0158, 0.0099, 0.0163, 0.0113, 0.0143], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 01:58:31,303 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163848.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 01:58:39,335 INFO [train.py:901] (3/4) Epoch 21, batch 2200, loss[loss=0.2309, simple_loss=0.2982, pruned_loss=0.08178, over 8468.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.29, pruned_loss=0.06413, over 1613143.51 frames. ], batch size: 25, lr: 3.63e-03, grad_scale: 16.0 +2023-02-07 01:58:58,248 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.717e+02 2.475e+02 2.987e+02 3.670e+02 7.762e+02, threshold=5.973e+02, percent-clipped=3.0 +2023-02-07 01:59:15,141 INFO [train.py:901] (3/4) Epoch 21, batch 2250, loss[loss=0.1731, simple_loss=0.2601, pruned_loss=0.04303, over 8320.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2905, pruned_loss=0.06396, over 1616006.88 frames. ], batch size: 25, lr: 3.62e-03, grad_scale: 16.0 +2023-02-07 01:59:49,152 INFO [train.py:901] (3/4) Epoch 21, batch 2300, loss[loss=0.2106, simple_loss=0.2832, pruned_loss=0.06901, over 8094.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2897, pruned_loss=0.06367, over 1617054.26 frames. ], batch size: 21, lr: 3.62e-03, grad_scale: 16.0 +2023-02-07 01:59:58,906 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163973.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:00:08,071 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.368e+02 2.361e+02 2.889e+02 3.736e+02 8.411e+02, threshold=5.778e+02, percent-clipped=4.0 +2023-02-07 02:00:17,858 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163998.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:00:26,188 INFO [train.py:901] (3/4) Epoch 21, batch 2350, loss[loss=0.2047, simple_loss=0.2817, pruned_loss=0.0638, over 7933.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2877, pruned_loss=0.06261, over 1615316.10 frames. ], batch size: 20, lr: 3.62e-03, grad_scale: 16.0 +2023-02-07 02:00:56,717 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7338, 1.9938, 2.1500, 1.4585, 2.2873, 1.6256, 0.8384, 1.9879], + device='cuda:3'), covar=tensor([0.0564, 0.0344, 0.0253, 0.0539, 0.0358, 0.0781, 0.0777, 0.0288], + device='cuda:3'), in_proj_covar=tensor([0.0446, 0.0382, 0.0333, 0.0436, 0.0368, 0.0524, 0.0385, 0.0411], + device='cuda:3'), out_proj_covar=tensor([1.2005e-04, 1.0031e-04, 8.7860e-05, 1.1536e-04, 9.6964e-05, 1.4837e-04, + 1.0417e-04, 1.0914e-04], device='cuda:3') +2023-02-07 02:01:01,227 INFO [train.py:901] (3/4) Epoch 21, batch 2400, loss[loss=0.2326, simple_loss=0.307, pruned_loss=0.07909, over 8541.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2888, pruned_loss=0.06355, over 1611147.32 frames. ], batch size: 39, lr: 3.62e-03, grad_scale: 16.0 +2023-02-07 02:01:19,280 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.352e+02 2.419e+02 2.926e+02 3.800e+02 6.132e+02, threshold=5.852e+02, percent-clipped=4.0 +2023-02-07 02:01:37,286 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164108.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:01:37,736 INFO [train.py:901] (3/4) Epoch 21, batch 2450, loss[loss=0.2523, simple_loss=0.3155, pruned_loss=0.09448, over 8505.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2898, pruned_loss=0.06419, over 1612400.07 frames. ], batch size: 26, lr: 3.62e-03, grad_scale: 16.0 +2023-02-07 02:02:12,735 INFO [train.py:901] (3/4) Epoch 21, batch 2500, loss[loss=0.1854, simple_loss=0.2756, pruned_loss=0.0476, over 8473.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2891, pruned_loss=0.06355, over 1613054.73 frames. ], batch size: 25, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:02:22,148 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164173.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:02:30,792 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.694e+02 2.421e+02 3.174e+02 4.025e+02 1.090e+03, threshold=6.349e+02, percent-clipped=9.0 +2023-02-07 02:02:46,233 INFO [train.py:901] (3/4) Epoch 21, batch 2550, loss[loss=0.1718, simple_loss=0.2608, pruned_loss=0.04139, over 7249.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2876, pruned_loss=0.06295, over 1609191.10 frames. ], batch size: 16, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:03:22,644 INFO [train.py:901] (3/4) Epoch 21, batch 2600, loss[loss=0.1903, simple_loss=0.2828, pruned_loss=0.04892, over 8489.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2875, pruned_loss=0.06257, over 1609317.56 frames. ], batch size: 28, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:03:40,891 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.516e+02 2.272e+02 2.670e+02 3.622e+02 6.852e+02, threshold=5.341e+02, percent-clipped=1.0 +2023-02-07 02:03:56,826 INFO [train.py:901] (3/4) Epoch 21, batch 2650, loss[loss=0.1731, simple_loss=0.2651, pruned_loss=0.04057, over 8073.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2881, pruned_loss=0.06267, over 1610519.81 frames. ], batch size: 21, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:04:33,157 INFO [train.py:901] (3/4) Epoch 21, batch 2700, loss[loss=0.2075, simple_loss=0.289, pruned_loss=0.06299, over 8249.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2876, pruned_loss=0.06228, over 1610431.51 frames. ], batch size: 24, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:04:46,948 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164378.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:04:52,076 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.622e+02 2.228e+02 2.697e+02 3.361e+02 7.045e+02, threshold=5.394e+02, percent-clipped=4.0 +2023-02-07 02:05:07,797 INFO [train.py:901] (3/4) Epoch 21, batch 2750, loss[loss=0.206, simple_loss=0.2983, pruned_loss=0.05686, over 8245.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.288, pruned_loss=0.06246, over 1610398.67 frames. ], batch size: 22, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:05:36,819 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=164452.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:05:41,734 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9170, 1.7711, 2.0226, 1.7124, 0.9962, 1.7679, 2.4432, 2.3696], + device='cuda:3'), covar=tensor([0.0426, 0.1151, 0.1550, 0.1320, 0.0596, 0.1405, 0.0549, 0.0523], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0151, 0.0188, 0.0158, 0.0098, 0.0162, 0.0112, 0.0142], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 02:05:42,233 INFO [train.py:901] (3/4) Epoch 21, batch 2800, loss[loss=0.1437, simple_loss=0.231, pruned_loss=0.02818, over 7807.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2865, pruned_loss=0.06175, over 1607950.01 frames. ], batch size: 20, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:06:02,588 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.521e+02 2.305e+02 2.813e+02 3.760e+02 7.507e+02, threshold=5.625e+02, percent-clipped=3.0 +2023-02-07 02:06:18,053 INFO [train.py:901] (3/4) Epoch 21, batch 2850, loss[loss=0.2143, simple_loss=0.2918, pruned_loss=0.06839, over 7796.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2885, pruned_loss=0.06272, over 1615882.36 frames. ], batch size: 19, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:06:21,042 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6798, 2.4324, 3.3583, 2.5613, 3.1754, 2.6153, 2.5100, 1.8962], + device='cuda:3'), covar=tensor([0.5207, 0.5133, 0.1995, 0.4022, 0.2635, 0.3068, 0.1750, 0.5872], + device='cuda:3'), in_proj_covar=tensor([0.0937, 0.0972, 0.0798, 0.0931, 0.0989, 0.0884, 0.0742, 0.0821], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 02:06:23,426 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=164517.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:06:51,365 INFO [train.py:901] (3/4) Epoch 21, batch 2900, loss[loss=0.2584, simple_loss=0.3415, pruned_loss=0.08771, over 8735.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2902, pruned_loss=0.06393, over 1611802.48 frames. ], batch size: 30, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:06:56,989 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=164567.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:07:09,769 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-07 02:07:11,685 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.734e+02 2.599e+02 3.265e+02 4.069e+02 1.074e+03, threshold=6.531e+02, percent-clipped=8.0 +2023-02-07 02:07:11,869 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164586.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:07:19,683 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.7367, 5.8870, 5.2058, 2.3808, 5.1924, 5.6106, 5.4515, 5.3385], + device='cuda:3'), covar=tensor([0.0493, 0.0351, 0.0799, 0.4101, 0.0650, 0.0614, 0.0917, 0.0550], + device='cuda:3'), in_proj_covar=tensor([0.0512, 0.0429, 0.0430, 0.0529, 0.0420, 0.0434, 0.0414, 0.0379], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 02:07:19,723 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164596.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:07:28,728 INFO [train.py:901] (3/4) Epoch 21, batch 2950, loss[loss=0.1888, simple_loss=0.264, pruned_loss=0.05681, over 7433.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2912, pruned_loss=0.06419, over 1610203.66 frames. ], batch size: 17, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:07:44,453 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=164632.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:08:02,297 INFO [train.py:901] (3/4) Epoch 21, batch 3000, loss[loss=0.2091, simple_loss=0.2929, pruned_loss=0.0627, over 8360.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2913, pruned_loss=0.06413, over 1615483.73 frames. ], batch size: 24, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:08:02,298 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 02:08:15,064 INFO [train.py:935] (3/4) Epoch 21, validation: loss=0.1742, simple_loss=0.2744, pruned_loss=0.03706, over 944034.00 frames. +2023-02-07 02:08:15,066 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-07 02:08:24,718 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.3909, 1.6651, 5.7723, 2.4351, 4.6937, 4.7345, 5.3509, 5.2561], + device='cuda:3'), covar=tensor([0.1188, 0.7288, 0.1016, 0.4635, 0.2117, 0.1824, 0.0929, 0.0820], + device='cuda:3'), in_proj_covar=tensor([0.0632, 0.0645, 0.0687, 0.0626, 0.0708, 0.0609, 0.0605, 0.0674], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 02:08:26,763 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164676.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:08:33,566 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.728e+02 2.380e+02 2.886e+02 3.399e+02 6.002e+02, threshold=5.772e+02, percent-clipped=0.0 +2023-02-07 02:08:49,851 INFO [train.py:901] (3/4) Epoch 21, batch 3050, loss[loss=0.1774, simple_loss=0.256, pruned_loss=0.04935, over 8034.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2888, pruned_loss=0.06282, over 1611977.11 frames. ], batch size: 22, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:08:59,359 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=164722.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:09:08,430 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4812, 1.7970, 1.8879, 1.2758, 1.9414, 1.4471, 0.4544, 1.7705], + device='cuda:3'), covar=tensor([0.0535, 0.0330, 0.0240, 0.0508, 0.0367, 0.0837, 0.0828, 0.0244], + device='cuda:3'), in_proj_covar=tensor([0.0452, 0.0385, 0.0340, 0.0443, 0.0373, 0.0531, 0.0389, 0.0416], + device='cuda:3'), out_proj_covar=tensor([1.2149e-04, 1.0118e-04, 8.9656e-05, 1.1709e-04, 9.8351e-05, 1.5044e-04, + 1.0530e-04, 1.1056e-04], device='cuda:3') +2023-02-07 02:09:25,495 INFO [train.py:901] (3/4) Epoch 21, batch 3100, loss[loss=0.1903, simple_loss=0.2717, pruned_loss=0.05449, over 8030.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2891, pruned_loss=0.06287, over 1614678.83 frames. ], batch size: 22, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:09:29,023 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164764.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:09:35,995 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-07 02:09:43,646 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.590e+02 2.375e+02 2.980e+02 3.572e+02 8.800e+02, threshold=5.960e+02, percent-clipped=5.0 +2023-02-07 02:09:59,122 INFO [train.py:901] (3/4) Epoch 21, batch 3150, loss[loss=0.2208, simple_loss=0.3107, pruned_loss=0.06545, over 8349.00 frames. ], tot_loss[loss=0.2087, simple_loss=0.2904, pruned_loss=0.06345, over 1616596.84 frames. ], batch size: 24, lr: 3.62e-03, grad_scale: 8.0 +2023-02-07 02:10:08,649 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164823.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:10:19,448 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=164837.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:10:26,874 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164848.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:10:34,945 INFO [train.py:901] (3/4) Epoch 21, batch 3200, loss[loss=0.2106, simple_loss=0.2921, pruned_loss=0.06456, over 8472.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2909, pruned_loss=0.06348, over 1615684.21 frames. ], batch size: 27, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:10:54,101 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.482e+02 2.324e+02 2.650e+02 3.384e+02 7.808e+02, threshold=5.299e+02, percent-clipped=1.0 +2023-02-07 02:10:54,541 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.95 vs. limit=5.0 +2023-02-07 02:10:55,683 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164888.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:11:09,469 INFO [train.py:901] (3/4) Epoch 21, batch 3250, loss[loss=0.2419, simple_loss=0.3229, pruned_loss=0.08051, over 7081.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2932, pruned_loss=0.06433, over 1620340.31 frames. ], batch size: 71, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:11:12,439 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164913.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:11:23,958 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=164930.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:11:30,775 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=164940.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:11:44,789 INFO [train.py:901] (3/4) Epoch 21, batch 3300, loss[loss=0.2284, simple_loss=0.3066, pruned_loss=0.07509, over 8359.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2918, pruned_loss=0.06384, over 1617095.29 frames. ], batch size: 24, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:12:05,208 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.558e+02 2.295e+02 2.742e+02 3.217e+02 7.829e+02, threshold=5.483e+02, percent-clipped=4.0 +2023-02-07 02:12:20,617 INFO [train.py:901] (3/4) Epoch 21, batch 3350, loss[loss=0.1725, simple_loss=0.2585, pruned_loss=0.04329, over 7786.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2917, pruned_loss=0.06382, over 1621616.48 frames. ], batch size: 19, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:12:28,066 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=165020.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:12:30,882 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8629, 3.4097, 2.1966, 2.7023, 2.5683, 1.8109, 2.4194, 3.0346], + device='cuda:3'), covar=tensor([0.1939, 0.0453, 0.1416, 0.0914, 0.0882, 0.1845, 0.1426, 0.1026], + device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0235, 0.0333, 0.0307, 0.0297, 0.0333, 0.0342, 0.0316], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 02:12:45,358 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165045.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:12:46,097 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5434, 1.5020, 1.8276, 1.2703, 1.3126, 1.8273, 0.2509, 1.2717], + device='cuda:3'), covar=tensor([0.1519, 0.1341, 0.0449, 0.0909, 0.2497, 0.0402, 0.2103, 0.1284], + device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0194, 0.0127, 0.0220, 0.0269, 0.0133, 0.0169, 0.0189], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 02:12:52,357 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165055.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:12:54,944 INFO [train.py:901] (3/4) Epoch 21, batch 3400, loss[loss=0.2291, simple_loss=0.3077, pruned_loss=0.07527, over 8575.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2915, pruned_loss=0.06389, over 1619182.18 frames. ], batch size: 49, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:13:15,619 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.386e+02 2.300e+02 2.821e+02 3.884e+02 1.046e+03, threshold=5.643e+02, percent-clipped=8.0 +2023-02-07 02:13:20,662 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165093.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:13:31,321 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=165108.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:13:31,893 INFO [train.py:901] (3/4) Epoch 21, batch 3450, loss[loss=0.1971, simple_loss=0.2828, pruned_loss=0.05566, over 7654.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2918, pruned_loss=0.06384, over 1617521.00 frames. ], batch size: 19, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:13:38,107 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165118.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:13:49,291 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165135.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:13:53,300 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165141.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:14:05,178 INFO [train.py:901] (3/4) Epoch 21, batch 3500, loss[loss=0.2068, simple_loss=0.2915, pruned_loss=0.06105, over 7808.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2903, pruned_loss=0.06321, over 1618841.15 frames. ], batch size: 20, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:14:06,056 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9328, 1.6607, 2.0503, 1.8088, 1.9263, 1.9657, 1.7829, 0.7899], + device='cuda:3'), covar=tensor([0.5253, 0.4394, 0.1969, 0.3267, 0.2409, 0.2869, 0.1852, 0.4921], + device='cuda:3'), in_proj_covar=tensor([0.0940, 0.0974, 0.0797, 0.0933, 0.0994, 0.0888, 0.0745, 0.0823], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 02:14:10,643 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-07 02:14:24,648 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.546e+02 2.436e+02 2.745e+02 3.695e+02 8.606e+02, threshold=5.490e+02, percent-clipped=3.0 +2023-02-07 02:14:41,285 INFO [train.py:901] (3/4) Epoch 21, batch 3550, loss[loss=0.1815, simple_loss=0.2576, pruned_loss=0.05272, over 7425.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2893, pruned_loss=0.06243, over 1617186.65 frames. ], batch size: 17, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:14:51,647 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165223.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:15:15,592 INFO [train.py:901] (3/4) Epoch 21, batch 3600, loss[loss=0.198, simple_loss=0.2756, pruned_loss=0.06017, over 7272.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2881, pruned_loss=0.06225, over 1614002.41 frames. ], batch size: 16, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:15:34,163 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.595e+02 2.347e+02 2.942e+02 3.699e+02 7.087e+02, threshold=5.884e+02, percent-clipped=2.0 +2023-02-07 02:15:35,260 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 +2023-02-07 02:15:44,481 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165301.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:15:51,141 INFO [train.py:901] (3/4) Epoch 21, batch 3650, loss[loss=0.2296, simple_loss=0.3195, pruned_loss=0.06989, over 8326.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2872, pruned_loss=0.0618, over 1612536.58 frames. ], batch size: 25, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:15:52,693 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165311.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:16:03,312 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165326.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:16:06,796 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3351, 2.0477, 2.8301, 2.2739, 2.6546, 2.2897, 2.0330, 1.5175], + device='cuda:3'), covar=tensor([0.5073, 0.4825, 0.1881, 0.3623, 0.2648, 0.3192, 0.2031, 0.5417], + device='cuda:3'), in_proj_covar=tensor([0.0936, 0.0970, 0.0793, 0.0932, 0.0993, 0.0885, 0.0744, 0.0821], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 02:16:10,820 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165336.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:16:16,764 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-07 02:16:23,307 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-07 02:16:25,987 INFO [train.py:901] (3/4) Epoch 21, batch 3700, loss[loss=0.2045, simple_loss=0.2825, pruned_loss=0.06326, over 7785.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2872, pruned_loss=0.06194, over 1612944.39 frames. ], batch size: 19, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:16:27,417 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.1559, 2.0950, 2.1285, 1.9795, 1.2456, 1.8663, 2.5104, 2.6234], + device='cuda:3'), covar=tensor([0.0417, 0.1085, 0.1584, 0.1277, 0.0545, 0.1372, 0.0550, 0.0493], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0151, 0.0189, 0.0159, 0.0099, 0.0162, 0.0113, 0.0142], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 02:16:44,014 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.608e+02 2.402e+02 2.885e+02 3.854e+02 8.848e+02, threshold=5.771e+02, percent-clipped=5.0 +2023-02-07 02:16:47,619 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165391.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:16:57,948 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0 +2023-02-07 02:16:59,571 INFO [train.py:901] (3/4) Epoch 21, batch 3750, loss[loss=0.226, simple_loss=0.317, pruned_loss=0.06751, over 8443.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2883, pruned_loss=0.06252, over 1614723.09 frames. ], batch size: 48, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:17:03,226 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2729, 2.1777, 1.7220, 1.9363, 1.8515, 1.4309, 1.7483, 1.6870], + device='cuda:3'), covar=tensor([0.1402, 0.0432, 0.1289, 0.0631, 0.0732, 0.1639, 0.0940, 0.0944], + device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0237, 0.0337, 0.0312, 0.0300, 0.0336, 0.0347, 0.0319], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 02:17:04,491 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165416.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:17:23,266 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7428, 2.4885, 3.2674, 2.6840, 3.1129, 2.5732, 2.4856, 2.3465], + device='cuda:3'), covar=tensor([0.3564, 0.3998, 0.1558, 0.3005, 0.1789, 0.2552, 0.1445, 0.4012], + device='cuda:3'), in_proj_covar=tensor([0.0933, 0.0967, 0.0791, 0.0930, 0.0988, 0.0883, 0.0740, 0.0816], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 02:17:36,098 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2025, 3.0883, 2.8901, 1.6524, 2.8111, 2.8961, 2.8229, 2.7846], + device='cuda:3'), covar=tensor([0.1399, 0.0950, 0.1476, 0.5083, 0.1239, 0.1626, 0.1771, 0.1286], + device='cuda:3'), in_proj_covar=tensor([0.0516, 0.0429, 0.0430, 0.0533, 0.0422, 0.0436, 0.0415, 0.0379], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 02:17:36,650 INFO [train.py:901] (3/4) Epoch 21, batch 3800, loss[loss=0.264, simple_loss=0.3367, pruned_loss=0.0957, over 8442.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2879, pruned_loss=0.06248, over 1610320.21 frames. ], batch size: 27, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:17:50,402 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165479.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:17:54,326 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=165485.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:17:54,908 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.536e+02 2.401e+02 2.925e+02 3.673e+02 6.793e+02, threshold=5.851e+02, percent-clipped=2.0 +2023-02-07 02:18:01,776 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-07 02:18:07,162 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165504.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:18:10,359 INFO [train.py:901] (3/4) Epoch 21, batch 3850, loss[loss=0.1872, simple_loss=0.2756, pruned_loss=0.04934, over 8284.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2877, pruned_loss=0.06275, over 1605475.39 frames. ], batch size: 23, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:18:18,548 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-07 02:18:46,113 INFO [train.py:901] (3/4) Epoch 21, batch 3900, loss[loss=0.1698, simple_loss=0.2639, pruned_loss=0.03784, over 8080.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2871, pruned_loss=0.06202, over 1603984.02 frames. ], batch size: 21, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:18:53,056 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165569.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:19:05,131 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.458e+02 2.182e+02 2.809e+02 3.459e+02 6.713e+02, threshold=5.619e+02, percent-clipped=4.0 +2023-02-07 02:19:12,372 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-07 02:19:14,925 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165600.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:19:20,727 INFO [train.py:901] (3/4) Epoch 21, batch 3950, loss[loss=0.286, simple_loss=0.3383, pruned_loss=0.1169, over 7186.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2885, pruned_loss=0.06261, over 1607276.34 frames. ], batch size: 71, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:19:27,201 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.13 vs. limit=5.0 +2023-02-07 02:19:29,802 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8213, 1.5789, 1.9388, 1.6752, 1.7911, 1.8316, 1.6289, 0.8019], + device='cuda:3'), covar=tensor([0.5108, 0.4551, 0.1849, 0.3153, 0.2323, 0.2829, 0.1862, 0.4633], + device='cuda:3'), in_proj_covar=tensor([0.0939, 0.0973, 0.0796, 0.0938, 0.0993, 0.0888, 0.0745, 0.0822], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 02:19:55,187 INFO [train.py:901] (3/4) Epoch 21, batch 4000, loss[loss=0.215, simple_loss=0.3077, pruned_loss=0.06119, over 8333.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2893, pruned_loss=0.06283, over 1610165.88 frames. ], batch size: 26, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:20:00,882 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2049, 1.8242, 4.2005, 1.5290, 2.5612, 4.6656, 5.0755, 3.5167], + device='cuda:3'), covar=tensor([0.1586, 0.2039, 0.0449, 0.2957, 0.1226, 0.0325, 0.0405, 0.1085], + device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0320, 0.0287, 0.0315, 0.0306, 0.0263, 0.0413, 0.0307], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 02:20:15,749 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.700e+02 2.370e+02 2.936e+02 3.785e+02 6.204e+02, threshold=5.872e+02, percent-clipped=2.0 +2023-02-07 02:20:31,251 INFO [train.py:901] (3/4) Epoch 21, batch 4050, loss[loss=0.2038, simple_loss=0.2935, pruned_loss=0.05705, over 8327.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2894, pruned_loss=0.06278, over 1611393.85 frames. ], batch size: 26, lr: 3.61e-03, grad_scale: 8.0 +2023-02-07 02:21:04,772 INFO [train.py:901] (3/4) Epoch 21, batch 4100, loss[loss=0.195, simple_loss=0.2885, pruned_loss=0.05076, over 8330.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2896, pruned_loss=0.06276, over 1614736.04 frames. ], batch size: 25, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:21:11,132 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165768.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:21:24,836 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.760e+02 2.510e+02 3.105e+02 3.860e+02 6.931e+02, threshold=6.209e+02, percent-clipped=6.0 +2023-02-07 02:21:41,907 INFO [train.py:901] (3/4) Epoch 21, batch 4150, loss[loss=0.174, simple_loss=0.2696, pruned_loss=0.03918, over 8194.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2907, pruned_loss=0.06312, over 1620060.53 frames. ], batch size: 23, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:21:50,140 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165821.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:22:13,904 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165856.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:22:15,083 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-07 02:22:15,769 INFO [train.py:901] (3/4) Epoch 21, batch 4200, loss[loss=0.218, simple_loss=0.2998, pruned_loss=0.06806, over 8514.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2906, pruned_loss=0.06301, over 1621252.95 frames. ], batch size: 26, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:22:30,517 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165881.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:22:33,700 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.532e+02 2.320e+02 2.907e+02 3.705e+02 7.802e+02, threshold=5.814e+02, percent-clipped=2.0 +2023-02-07 02:22:37,059 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-07 02:22:50,804 INFO [train.py:901] (3/4) Epoch 21, batch 4250, loss[loss=0.1803, simple_loss=0.2706, pruned_loss=0.04495, over 8364.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2906, pruned_loss=0.06291, over 1617674.61 frames. ], batch size: 24, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:22:53,703 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=165913.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:23:26,545 INFO [train.py:901] (3/4) Epoch 21, batch 4300, loss[loss=0.2079, simple_loss=0.2933, pruned_loss=0.06123, over 7928.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2896, pruned_loss=0.0627, over 1612701.35 frames. ], batch size: 20, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:23:44,465 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.579e+02 2.284e+02 2.728e+02 3.396e+02 7.954e+02, threshold=5.457e+02, percent-clipped=4.0 +2023-02-07 02:23:56,036 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4940, 1.9286, 3.1428, 1.2444, 2.3686, 1.7217, 1.6871, 2.2980], + device='cuda:3'), covar=tensor([0.2317, 0.2865, 0.0907, 0.5361, 0.2167, 0.4018, 0.2717, 0.2730], + device='cuda:3'), in_proj_covar=tensor([0.0521, 0.0595, 0.0552, 0.0637, 0.0643, 0.0591, 0.0534, 0.0633], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 02:23:57,382 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9081, 1.7524, 2.8155, 2.0906, 2.3867, 1.8958, 1.6060, 1.1960], + device='cuda:3'), covar=tensor([0.7262, 0.6342, 0.2004, 0.4073, 0.3351, 0.4422, 0.2992, 0.5938], + device='cuda:3'), in_proj_covar=tensor([0.0940, 0.0976, 0.0797, 0.0937, 0.0993, 0.0888, 0.0746, 0.0823], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 02:24:01,652 INFO [train.py:901] (3/4) Epoch 21, batch 4350, loss[loss=0.2056, simple_loss=0.2876, pruned_loss=0.06184, over 8635.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2891, pruned_loss=0.06222, over 1611756.13 frames. ], batch size: 34, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:24:11,728 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-07 02:24:15,248 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166028.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:24:16,541 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5857, 1.8985, 2.9844, 1.4429, 2.2746, 1.9743, 1.6596, 2.2500], + device='cuda:3'), covar=tensor([0.1888, 0.2581, 0.0870, 0.4398, 0.1802, 0.3126, 0.2311, 0.2178], + device='cuda:3'), in_proj_covar=tensor([0.0520, 0.0593, 0.0551, 0.0635, 0.0641, 0.0590, 0.0533, 0.0631], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 02:24:36,834 INFO [train.py:901] (3/4) Epoch 21, batch 4400, loss[loss=0.2288, simple_loss=0.3104, pruned_loss=0.07358, over 8647.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2883, pruned_loss=0.06174, over 1612918.96 frames. ], batch size: 39, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:24:45,737 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166072.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 02:24:54,405 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-07 02:24:55,056 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.718e+02 2.482e+02 3.095e+02 3.863e+02 7.424e+02, threshold=6.191e+02, percent-clipped=10.0 +2023-02-07 02:25:10,656 INFO [train.py:901] (3/4) Epoch 21, batch 4450, loss[loss=0.1632, simple_loss=0.2504, pruned_loss=0.03802, over 7924.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2873, pruned_loss=0.06108, over 1612396.14 frames. ], batch size: 20, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:25:12,801 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166112.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:25:22,692 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7575, 1.9169, 2.0498, 1.3692, 2.1261, 1.5489, 0.5650, 1.8501], + device='cuda:3'), covar=tensor([0.0543, 0.0352, 0.0267, 0.0529, 0.0414, 0.0782, 0.0903, 0.0283], + device='cuda:3'), in_proj_covar=tensor([0.0454, 0.0387, 0.0341, 0.0441, 0.0374, 0.0533, 0.0391, 0.0415], + device='cuda:3'), out_proj_covar=tensor([1.2196e-04, 1.0144e-04, 8.9743e-05, 1.1657e-04, 9.8624e-05, 1.5080e-04, + 1.0559e-04, 1.1017e-04], device='cuda:3') +2023-02-07 02:25:27,716 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-07 02:25:45,433 INFO [train.py:901] (3/4) Epoch 21, batch 4500, loss[loss=0.1808, simple_loss=0.2648, pruned_loss=0.04845, over 8760.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2884, pruned_loss=0.06215, over 1613541.81 frames. ], batch size: 30, lr: 3.60e-03, grad_scale: 16.0 +2023-02-07 02:25:50,227 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-07 02:25:50,295 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166165.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:25:50,362 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166165.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:26:05,003 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.806e+02 2.688e+02 3.191e+02 4.415e+02 1.086e+03, threshold=6.382e+02, percent-clipped=9.0 +2023-02-07 02:26:20,597 INFO [train.py:901] (3/4) Epoch 21, batch 4550, loss[loss=0.1706, simple_loss=0.2505, pruned_loss=0.04536, over 7806.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2892, pruned_loss=0.06278, over 1613049.38 frames. ], batch size: 19, lr: 3.60e-03, grad_scale: 16.0 +2023-02-07 02:26:27,576 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5290, 2.3374, 3.2954, 2.6069, 2.9950, 2.5334, 2.2165, 1.9468], + device='cuda:3'), covar=tensor([0.4965, 0.5113, 0.1743, 0.3536, 0.2496, 0.2925, 0.1892, 0.5111], + device='cuda:3'), in_proj_covar=tensor([0.0935, 0.0972, 0.0794, 0.0932, 0.0989, 0.0885, 0.0741, 0.0820], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 02:26:33,042 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166227.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:26:54,695 INFO [train.py:901] (3/4) Epoch 21, batch 4600, loss[loss=0.1955, simple_loss=0.2859, pruned_loss=0.05256, over 8371.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2888, pruned_loss=0.06273, over 1610044.92 frames. ], batch size: 24, lr: 3.60e-03, grad_scale: 16.0 +2023-02-07 02:27:10,560 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166280.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:27:14,137 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166284.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:27:15,243 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.492e+02 2.551e+02 3.310e+02 4.080e+02 7.820e+02, threshold=6.621e+02, percent-clipped=4.0 +2023-02-07 02:27:30,334 INFO [train.py:901] (3/4) Epoch 21, batch 4650, loss[loss=0.187, simple_loss=0.2621, pruned_loss=0.05596, over 7701.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2892, pruned_loss=0.06315, over 1612534.98 frames. ], batch size: 18, lr: 3.60e-03, grad_scale: 16.0 +2023-02-07 02:27:30,540 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166309.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:27:41,202 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166325.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:28:03,638 INFO [train.py:901] (3/4) Epoch 21, batch 4700, loss[loss=0.1824, simple_loss=0.2663, pruned_loss=0.0493, over 7804.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2882, pruned_loss=0.06277, over 1610743.48 frames. ], batch size: 19, lr: 3.60e-03, grad_scale: 16.0 +2023-02-07 02:28:23,900 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.420e+02 2.373e+02 2.801e+02 3.877e+02 1.145e+03, threshold=5.601e+02, percent-clipped=4.0 +2023-02-07 02:28:40,136 INFO [train.py:901] (3/4) Epoch 21, batch 4750, loss[loss=0.2231, simple_loss=0.2834, pruned_loss=0.0814, over 7272.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2877, pruned_loss=0.0627, over 1608999.83 frames. ], batch size: 16, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:28:45,009 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166416.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 02:28:52,942 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-07 02:28:55,082 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-07 02:29:14,177 INFO [train.py:901] (3/4) Epoch 21, batch 4800, loss[loss=0.2473, simple_loss=0.3243, pruned_loss=0.08518, over 8522.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2876, pruned_loss=0.06224, over 1607204.72 frames. ], batch size: 28, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:29:30,706 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166483.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:29:33,193 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.480e+02 2.407e+02 2.819e+02 3.849e+02 8.316e+02, threshold=5.639e+02, percent-clipped=5.0 +2023-02-07 02:29:39,661 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.96 vs. limit=5.0 +2023-02-07 02:29:42,737 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166499.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:29:43,962 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-07 02:29:48,834 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166508.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:29:49,314 INFO [train.py:901] (3/4) Epoch 21, batch 4850, loss[loss=0.1858, simple_loss=0.2742, pruned_loss=0.04869, over 8196.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2869, pruned_loss=0.06131, over 1610209.08 frames. ], batch size: 23, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:29:49,390 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166509.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:30:03,258 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166527.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:30:05,914 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166531.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 02:30:09,308 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166536.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:30:17,942 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166549.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:30:21,057 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.49 vs. limit=5.0 +2023-02-07 02:30:24,635 INFO [train.py:901] (3/4) Epoch 21, batch 4900, loss[loss=0.2063, simple_loss=0.2679, pruned_loss=0.07238, over 7800.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2862, pruned_loss=0.06081, over 1608964.18 frames. ], batch size: 19, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:30:26,219 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166561.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:30:43,068 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.635e+02 2.429e+02 3.059e+02 4.014e+02 7.599e+02, threshold=6.119e+02, percent-clipped=4.0 +2023-02-07 02:30:54,281 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5980, 1.5983, 2.1266, 1.4428, 1.2600, 2.0788, 0.3388, 1.3022], + device='cuda:3'), covar=tensor([0.1841, 0.1338, 0.0397, 0.1217, 0.2891, 0.0465, 0.2358, 0.1354], + device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0196, 0.0128, 0.0223, 0.0272, 0.0136, 0.0172, 0.0192], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 02:30:58,618 INFO [train.py:901] (3/4) Epoch 21, batch 4950, loss[loss=0.2027, simple_loss=0.2766, pruned_loss=0.06439, over 8334.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.287, pruned_loss=0.06168, over 1610523.17 frames. ], batch size: 26, lr: 3.60e-03, grad_scale: 8.0 +2023-02-07 02:31:09,545 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166624.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:31:34,524 INFO [train.py:901] (3/4) Epoch 21, batch 5000, loss[loss=0.1821, simple_loss=0.2677, pruned_loss=0.04827, over 8086.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2868, pruned_loss=0.0615, over 1607220.05 frames. ], batch size: 21, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:31:41,051 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166669.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:31:52,864 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.664e+02 2.262e+02 2.770e+02 3.475e+02 7.586e+02, threshold=5.540e+02, percent-clipped=2.0 +2023-02-07 02:32:07,640 INFO [train.py:901] (3/4) Epoch 21, batch 5050, loss[loss=0.248, simple_loss=0.3237, pruned_loss=0.08612, over 8476.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2867, pruned_loss=0.06145, over 1609245.11 frames. ], batch size: 28, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:32:23,023 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-07 02:32:38,361 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166753.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:32:42,947 INFO [train.py:901] (3/4) Epoch 21, batch 5100, loss[loss=0.1796, simple_loss=0.2444, pruned_loss=0.05739, over 7703.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2872, pruned_loss=0.06202, over 1611638.31 frames. ], batch size: 18, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:32:59,528 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166782.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:33:00,920 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166784.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:33:02,731 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.724e+02 2.499e+02 3.045e+02 3.729e+02 1.083e+03, threshold=6.090e+02, percent-clipped=5.0 +2023-02-07 02:33:02,981 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166787.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 02:33:17,653 INFO [train.py:901] (3/4) Epoch 21, batch 5150, loss[loss=0.2367, simple_loss=0.3117, pruned_loss=0.08081, over 8352.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2883, pruned_loss=0.06269, over 1615350.85 frames. ], batch size: 49, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:33:19,982 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166812.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 02:33:35,605 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9209, 2.2532, 1.8007, 2.8571, 1.3543, 1.4973, 1.9218, 2.2096], + device='cuda:3'), covar=tensor([0.0848, 0.0707, 0.0934, 0.0348, 0.1092, 0.1362, 0.0936, 0.0837], + device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0194, 0.0241, 0.0210, 0.0203, 0.0241, 0.0249, 0.0206], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 02:33:40,945 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166843.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:33:52,921 INFO [train.py:901] (3/4) Epoch 21, batch 5200, loss[loss=0.1845, simple_loss=0.2731, pruned_loss=0.04798, over 8242.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2876, pruned_loss=0.06208, over 1612888.65 frames. ], batch size: 24, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:33:54,692 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.05 vs. limit=5.0 +2023-02-07 02:34:01,101 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166871.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:34:08,926 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166880.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:34:13,418 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.573e+02 2.417e+02 2.893e+02 3.464e+02 9.071e+02, threshold=5.787e+02, percent-clipped=3.0 +2023-02-07 02:34:17,473 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=166893.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:34:22,854 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-07 02:34:25,858 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166905.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:34:28,364 INFO [train.py:901] (3/4) Epoch 21, batch 5250, loss[loss=0.1893, simple_loss=0.2819, pruned_loss=0.04837, over 8250.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2878, pruned_loss=0.06195, over 1613001.76 frames. ], batch size: 22, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:34:53,832 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166947.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:35:00,988 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166958.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:35:01,477 INFO [train.py:901] (3/4) Epoch 21, batch 5300, loss[loss=0.2517, simple_loss=0.32, pruned_loss=0.09168, over 6884.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.291, pruned_loss=0.06365, over 1616084.96 frames. ], batch size: 72, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:35:21,119 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166986.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:35:21,588 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.438e+02 2.349e+02 2.996e+02 3.802e+02 6.845e+02, threshold=5.992e+02, percent-clipped=3.0 +2023-02-07 02:35:22,372 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6293, 2.5336, 1.8554, 2.3821, 2.2130, 1.5784, 2.2421, 2.2866], + device='cuda:3'), covar=tensor([0.1491, 0.0426, 0.1226, 0.0634, 0.0733, 0.1535, 0.0879, 0.0929], + device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0235, 0.0335, 0.0308, 0.0298, 0.0334, 0.0344, 0.0316], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 02:35:37,482 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167008.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:35:37,983 INFO [train.py:901] (3/4) Epoch 21, batch 5350, loss[loss=0.1806, simple_loss=0.2669, pruned_loss=0.04718, over 8090.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2888, pruned_loss=0.06275, over 1612722.18 frames. ], batch size: 21, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:35:58,870 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167040.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:36:12,094 INFO [train.py:901] (3/4) Epoch 21, batch 5400, loss[loss=0.2113, simple_loss=0.2951, pruned_loss=0.06372, over 8336.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2885, pruned_loss=0.06276, over 1606652.74 frames. ], batch size: 25, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:36:16,470 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167065.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:36:32,162 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.695e+02 2.459e+02 3.013e+02 3.547e+02 6.118e+02, threshold=6.026e+02, percent-clipped=1.0 +2023-02-07 02:36:39,933 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=167097.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:36:47,942 INFO [train.py:901] (3/4) Epoch 21, batch 5450, loss[loss=0.2064, simple_loss=0.2941, pruned_loss=0.05935, over 8199.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2879, pruned_loss=0.06244, over 1609019.92 frames. ], batch size: 23, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:36:55,795 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167118.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:36:58,551 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167122.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:37:01,216 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=167126.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:37:12,850 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-07 02:37:19,305 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8170, 1.3315, 4.0005, 1.5078, 3.5507, 3.3960, 3.6615, 3.5390], + device='cuda:3'), covar=tensor([0.0741, 0.4533, 0.0607, 0.4129, 0.1234, 0.0954, 0.0677, 0.0777], + device='cuda:3'), in_proj_covar=tensor([0.0630, 0.0643, 0.0695, 0.0629, 0.0707, 0.0605, 0.0605, 0.0677], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 02:37:24,075 INFO [train.py:901] (3/4) Epoch 21, batch 5500, loss[loss=0.2381, simple_loss=0.3183, pruned_loss=0.07895, over 8645.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2876, pruned_loss=0.06215, over 1613946.02 frames. ], batch size: 34, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:37:43,657 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.511e+02 2.538e+02 3.099e+02 3.967e+02 8.838e+02, threshold=6.197e+02, percent-clipped=3.0 +2023-02-07 02:37:58,448 INFO [train.py:901] (3/4) Epoch 21, batch 5550, loss[loss=0.2313, simple_loss=0.3143, pruned_loss=0.07416, over 8667.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2876, pruned_loss=0.06139, over 1617678.72 frames. ], batch size: 34, lr: 3.59e-03, grad_scale: 4.0 +2023-02-07 02:38:01,331 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167212.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:38:02,595 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167214.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:38:21,134 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167239.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:38:22,453 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167241.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:38:23,135 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167242.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:38:34,480 INFO [train.py:901] (3/4) Epoch 21, batch 5600, loss[loss=0.1688, simple_loss=0.2568, pruned_loss=0.04036, over 8130.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2865, pruned_loss=0.06107, over 1616363.39 frames. ], batch size: 22, lr: 3.59e-03, grad_scale: 8.0 +2023-02-07 02:38:34,957 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.48 vs. limit=5.0 +2023-02-07 02:38:36,021 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6893, 1.3350, 4.8641, 1.7589, 4.3081, 4.0161, 4.3971, 4.2417], + device='cuda:3'), covar=tensor([0.0574, 0.5153, 0.0452, 0.4298, 0.1050, 0.0896, 0.0602, 0.0635], + device='cuda:3'), in_proj_covar=tensor([0.0629, 0.0637, 0.0691, 0.0626, 0.0701, 0.0602, 0.0601, 0.0672], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 02:38:38,123 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167264.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:38:40,083 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167267.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:38:54,594 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.626e+02 2.492e+02 3.097e+02 3.838e+02 7.086e+02, threshold=6.194e+02, percent-clipped=1.0 +2023-02-07 02:38:54,819 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167289.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:38:56,077 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=167291.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:39:08,036 INFO [train.py:901] (3/4) Epoch 21, batch 5650, loss[loss=0.2134, simple_loss=0.2985, pruned_loss=0.06411, over 8083.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2868, pruned_loss=0.06131, over 1612568.76 frames. ], batch size: 21, lr: 3.59e-03, grad_scale: 4.0 +2023-02-07 02:39:18,774 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-07 02:39:27,191 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.24 vs. limit=5.0 +2023-02-07 02:39:30,412 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8836, 2.3655, 3.8202, 1.8124, 2.9781, 2.4208, 1.9759, 2.8981], + device='cuda:3'), covar=tensor([0.1823, 0.2633, 0.0836, 0.4267, 0.1633, 0.2926, 0.2228, 0.2256], + device='cuda:3'), in_proj_covar=tensor([0.0521, 0.0594, 0.0552, 0.0637, 0.0637, 0.0588, 0.0530, 0.0627], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 02:39:44,505 INFO [train.py:901] (3/4) Epoch 21, batch 5700, loss[loss=0.2071, simple_loss=0.2898, pruned_loss=0.06219, over 8241.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2869, pruned_loss=0.06158, over 1609349.47 frames. ], batch size: 22, lr: 3.59e-03, grad_scale: 4.0 +2023-02-07 02:40:04,660 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.459e+02 2.585e+02 3.206e+02 3.925e+02 8.506e+02, threshold=6.412e+02, percent-clipped=6.0 +2023-02-07 02:40:07,027 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6410, 2.3186, 3.0175, 2.5130, 2.9381, 2.5443, 2.3884, 2.2308], + device='cuda:3'), covar=tensor([0.3514, 0.3721, 0.1536, 0.2723, 0.1853, 0.2417, 0.1489, 0.3546], + device='cuda:3'), in_proj_covar=tensor([0.0938, 0.0972, 0.0796, 0.0939, 0.0990, 0.0888, 0.0743, 0.0823], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 02:40:10,473 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.17 vs. limit=5.0 +2023-02-07 02:40:16,545 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167406.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:40:18,408 INFO [train.py:901] (3/4) Epoch 21, batch 5750, loss[loss=0.2013, simple_loss=0.2871, pruned_loss=0.05779, over 8247.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.285, pruned_loss=0.0604, over 1609203.55 frames. ], batch size: 22, lr: 3.59e-03, grad_scale: 4.0 +2023-02-07 02:40:19,352 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-07 02:40:21,795 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9976, 1.7838, 3.2683, 1.2540, 2.4778, 3.6055, 3.8871, 2.6330], + device='cuda:3'), covar=tensor([0.1439, 0.1891, 0.0470, 0.2785, 0.1079, 0.0371, 0.0596, 0.1045], + device='cuda:3'), in_proj_covar=tensor([0.0294, 0.0320, 0.0287, 0.0314, 0.0307, 0.0263, 0.0415, 0.0306], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 02:40:24,259 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-07 02:40:47,703 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167450.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:40:53,629 INFO [train.py:901] (3/4) Epoch 21, batch 5800, loss[loss=0.1773, simple_loss=0.259, pruned_loss=0.04774, over 8074.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2852, pruned_loss=0.06034, over 1606791.69 frames. ], batch size: 21, lr: 3.59e-03, grad_scale: 4.0 +2023-02-07 02:40:55,798 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=167462.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:40:59,205 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=167466.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:41:00,731 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167468.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:41:05,556 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.33 vs. limit=5.0 +2023-02-07 02:41:11,158 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4963, 1.7867, 2.6676, 1.3680, 1.8819, 1.8364, 1.5067, 1.9185], + device='cuda:3'), covar=tensor([0.1926, 0.2623, 0.0845, 0.4664, 0.2068, 0.3326, 0.2507, 0.2367], + device='cuda:3'), in_proj_covar=tensor([0.0520, 0.0593, 0.0552, 0.0635, 0.0637, 0.0587, 0.0528, 0.0627], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 02:41:15,042 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.585e+02 2.466e+02 2.953e+02 3.603e+02 7.254e+02, threshold=5.907e+02, percent-clipped=1.0 +2023-02-07 02:41:17,996 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167493.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:41:20,771 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167497.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:41:28,620 INFO [train.py:901] (3/4) Epoch 21, batch 5850, loss[loss=0.193, simple_loss=0.2833, pruned_loss=0.05136, over 8464.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2855, pruned_loss=0.06061, over 1608687.81 frames. ], batch size: 39, lr: 3.59e-03, grad_scale: 4.0 +2023-02-07 02:41:37,415 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-07 02:41:37,824 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167522.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:42:03,761 INFO [train.py:901] (3/4) Epoch 21, batch 5900, loss[loss=0.2047, simple_loss=0.2919, pruned_loss=0.05878, over 7921.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2857, pruned_loss=0.06113, over 1604280.84 frames. ], batch size: 20, lr: 3.59e-03, grad_scale: 4.0 +2023-02-07 02:42:14,756 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.9690, 1.8479, 6.1123, 2.2116, 5.4462, 5.1614, 5.6579, 5.5538], + device='cuda:3'), covar=tensor([0.0465, 0.4594, 0.0369, 0.3896, 0.1089, 0.0905, 0.0486, 0.0480], + device='cuda:3'), in_proj_covar=tensor([0.0630, 0.0640, 0.0695, 0.0632, 0.0704, 0.0606, 0.0607, 0.0680], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 02:42:16,866 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167577.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:42:19,700 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167581.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:42:25,696 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.618e+02 2.257e+02 2.859e+02 3.440e+02 7.059e+02, threshold=5.718e+02, percent-clipped=2.0 +2023-02-07 02:42:40,379 INFO [train.py:901] (3/4) Epoch 21, batch 5950, loss[loss=0.2071, simple_loss=0.275, pruned_loss=0.06956, over 7700.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2858, pruned_loss=0.0612, over 1606834.32 frames. ], batch size: 18, lr: 3.58e-03, grad_scale: 4.0 +2023-02-07 02:43:14,053 INFO [train.py:901] (3/4) Epoch 21, batch 6000, loss[loss=0.1759, simple_loss=0.2595, pruned_loss=0.04612, over 7700.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2868, pruned_loss=0.06144, over 1610556.47 frames. ], batch size: 18, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:43:14,054 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 02:43:26,394 INFO [train.py:935] (3/4) Epoch 21, validation: loss=0.174, simple_loss=0.2741, pruned_loss=0.03692, over 944034.00 frames. +2023-02-07 02:43:26,395 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-07 02:43:28,719 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167662.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:43:45,658 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167687.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:43:47,404 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.749e+02 2.382e+02 2.918e+02 3.609e+02 5.587e+02, threshold=5.837e+02, percent-clipped=0.0 +2023-02-07 02:44:01,969 INFO [train.py:901] (3/4) Epoch 21, batch 6050, loss[loss=0.1734, simple_loss=0.2598, pruned_loss=0.0435, over 7536.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2872, pruned_loss=0.06127, over 1612769.39 frames. ], batch size: 18, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:44:06,718 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 +2023-02-07 02:44:22,063 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167737.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:44:38,028 INFO [train.py:901] (3/4) Epoch 21, batch 6100, loss[loss=0.1603, simple_loss=0.2451, pruned_loss=0.03774, over 7539.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2863, pruned_loss=0.06122, over 1612603.78 frames. ], batch size: 18, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:44:56,057 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167785.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:44:57,194 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-07 02:44:58,540 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.708e+02 2.390e+02 3.045e+02 3.849e+02 6.701e+02, threshold=6.089e+02, percent-clipped=2.0 +2023-02-07 02:45:02,123 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=167794.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:45:13,116 INFO [train.py:901] (3/4) Epoch 21, batch 6150, loss[loss=0.2096, simple_loss=0.2916, pruned_loss=0.06379, over 8106.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2871, pruned_loss=0.06164, over 1613779.36 frames. ], batch size: 23, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:45:30,424 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167833.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:45:33,104 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167837.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:45:48,358 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167858.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:45:48,812 INFO [train.py:901] (3/4) Epoch 21, batch 6200, loss[loss=0.1892, simple_loss=0.2689, pruned_loss=0.05471, over 7946.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2877, pruned_loss=0.06199, over 1617824.04 frames. ], batch size: 20, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:45:51,057 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167862.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:46:09,470 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.755e+02 2.266e+02 2.776e+02 3.727e+02 8.167e+02, threshold=5.552e+02, percent-clipped=4.0 +2023-02-07 02:46:16,722 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-07 02:46:23,391 INFO [train.py:901] (3/4) Epoch 21, batch 6250, loss[loss=0.1896, simple_loss=0.2884, pruned_loss=0.04537, over 8363.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2876, pruned_loss=0.06203, over 1614113.09 frames. ], batch size: 24, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:46:23,597 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167909.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:46:58,909 INFO [train.py:901] (3/4) Epoch 21, batch 6300, loss[loss=0.2199, simple_loss=0.3115, pruned_loss=0.06419, over 8337.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2875, pruned_loss=0.06227, over 1610644.23 frames. ], batch size: 26, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:47:16,164 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.31 vs. limit=5.0 +2023-02-07 02:47:20,723 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.676e+02 2.475e+02 2.869e+02 3.545e+02 9.430e+02, threshold=5.737e+02, percent-clipped=7.0 +2023-02-07 02:47:35,237 INFO [train.py:901] (3/4) Epoch 21, batch 6350, loss[loss=0.1907, simple_loss=0.2659, pruned_loss=0.05781, over 7688.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2884, pruned_loss=0.06232, over 1615117.33 frames. ], batch size: 18, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:48:06,026 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168053.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:48:09,823 INFO [train.py:901] (3/4) Epoch 21, batch 6400, loss[loss=0.1943, simple_loss=0.2812, pruned_loss=0.05374, over 7981.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2887, pruned_loss=0.06213, over 1618830.80 frames. ], batch size: 21, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:48:25,125 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=168081.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:48:30,484 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 2.236e+02 2.639e+02 3.603e+02 6.999e+02, threshold=5.279e+02, percent-clipped=2.0 +2023-02-07 02:48:45,447 INFO [train.py:901] (3/4) Epoch 21, batch 6450, loss[loss=0.2383, simple_loss=0.3132, pruned_loss=0.08166, over 8637.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2883, pruned_loss=0.0623, over 1614866.99 frames. ], batch size: 34, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:48:59,389 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=168129.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:49:19,828 INFO [train.py:901] (3/4) Epoch 21, batch 6500, loss[loss=0.2413, simple_loss=0.312, pruned_loss=0.08529, over 8519.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2885, pruned_loss=0.06342, over 1610833.68 frames. ], batch size: 26, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:49:24,789 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168165.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:49:41,355 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.960e+02 2.483e+02 3.129e+02 4.081e+02 1.148e+03, threshold=6.258e+02, percent-clipped=13.0 +2023-02-07 02:49:42,167 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168190.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:49:46,106 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168196.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:49:54,743 INFO [train.py:901] (3/4) Epoch 21, batch 6550, loss[loss=0.1627, simple_loss=0.232, pruned_loss=0.04667, over 7436.00 frames. ], tot_loss[loss=0.207, simple_loss=0.288, pruned_loss=0.063, over 1610337.20 frames. ], batch size: 17, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:50:19,982 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168244.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:50:20,482 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-07 02:50:24,649 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168251.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:50:29,933 INFO [train.py:901] (3/4) Epoch 21, batch 6600, loss[loss=0.2149, simple_loss=0.2858, pruned_loss=0.07196, over 7538.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2873, pruned_loss=0.06267, over 1607534.94 frames. ], batch size: 18, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:50:32,313 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-07 02:50:38,745 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-07 02:50:50,802 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.640e+02 2.414e+02 2.830e+02 3.481e+02 7.637e+02, threshold=5.659e+02, percent-clipped=3.0 +2023-02-07 02:51:05,095 INFO [train.py:901] (3/4) Epoch 21, batch 6650, loss[loss=0.2553, simple_loss=0.3203, pruned_loss=0.09515, over 8543.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2885, pruned_loss=0.06325, over 1608696.86 frames. ], batch size: 49, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:51:40,101 INFO [train.py:901] (3/4) Epoch 21, batch 6700, loss[loss=0.2027, simple_loss=0.2871, pruned_loss=0.05913, over 8291.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2884, pruned_loss=0.0635, over 1606075.83 frames. ], batch size: 23, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:52:00,449 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.499e+02 2.306e+02 2.933e+02 3.476e+02 6.537e+02, threshold=5.866e+02, percent-clipped=2.0 +2023-02-07 02:52:04,737 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7272, 2.0911, 3.2437, 1.6001, 2.4926, 2.1022, 1.9036, 2.4404], + device='cuda:3'), covar=tensor([0.1827, 0.2507, 0.0859, 0.4193, 0.1870, 0.3016, 0.2045, 0.2461], + device='cuda:3'), in_proj_covar=tensor([0.0524, 0.0600, 0.0556, 0.0641, 0.0644, 0.0591, 0.0532, 0.0632], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 02:52:05,989 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=168397.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:52:14,914 INFO [train.py:901] (3/4) Epoch 21, batch 6750, loss[loss=0.2147, simple_loss=0.3026, pruned_loss=0.06347, over 8019.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2897, pruned_loss=0.0639, over 1610582.54 frames. ], batch size: 22, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:52:19,505 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 +2023-02-07 02:52:45,403 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168452.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:52:50,752 INFO [train.py:901] (3/4) Epoch 21, batch 6800, loss[loss=0.175, simple_loss=0.2516, pruned_loss=0.04914, over 8078.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2895, pruned_loss=0.06316, over 1615466.72 frames. ], batch size: 21, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:52:58,523 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-07 02:53:04,349 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168477.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:53:12,373 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.744e+02 2.333e+02 2.834e+02 3.373e+02 7.883e+02, threshold=5.669e+02, percent-clipped=5.0 +2023-02-07 02:53:20,354 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168500.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:53:26,321 INFO [train.py:901] (3/4) Epoch 21, batch 6850, loss[loss=0.1626, simple_loss=0.2394, pruned_loss=0.04288, over 7807.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2896, pruned_loss=0.06278, over 1616050.19 frames. ], batch size: 19, lr: 3.58e-03, grad_scale: 8.0 +2023-02-07 02:53:28,543 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168512.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:53:37,578 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168525.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:53:45,921 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-07 02:54:00,742 INFO [train.py:901] (3/4) Epoch 21, batch 6900, loss[loss=0.1881, simple_loss=0.2751, pruned_loss=0.05053, over 7279.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2904, pruned_loss=0.06309, over 1620248.26 frames. ], batch size: 16, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:54:22,253 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.632e+02 2.460e+02 2.867e+02 3.613e+02 6.820e+02, threshold=5.733e+02, percent-clipped=1.0 +2023-02-07 02:54:23,301 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.48 vs. limit=5.0 +2023-02-07 02:54:26,393 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=168595.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:54:26,687 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 +2023-02-07 02:54:35,737 INFO [train.py:901] (3/4) Epoch 21, batch 6950, loss[loss=0.181, simple_loss=0.2603, pruned_loss=0.05083, over 8240.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2891, pruned_loss=0.06263, over 1613504.33 frames. ], batch size: 22, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:54:45,634 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.02 vs. limit=2.0 +2023-02-07 02:54:53,393 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-07 02:55:10,656 INFO [train.py:901] (3/4) Epoch 21, batch 7000, loss[loss=0.1671, simple_loss=0.2422, pruned_loss=0.04598, over 7256.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2894, pruned_loss=0.06268, over 1613967.29 frames. ], batch size: 16, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:55:12,903 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4774, 1.4158, 1.8645, 1.3017, 1.1004, 1.7971, 0.2100, 1.1668], + device='cuda:3'), covar=tensor([0.1469, 0.1265, 0.0333, 0.0963, 0.2689, 0.0415, 0.2134, 0.1214], + device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0195, 0.0126, 0.0221, 0.0270, 0.0135, 0.0170, 0.0191], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 02:55:31,356 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.503e+02 2.444e+02 3.041e+02 3.968e+02 8.528e+02, threshold=6.083e+02, percent-clipped=8.0 +2023-02-07 02:55:35,109 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4995, 1.8808, 2.9407, 1.3826, 2.1508, 1.8856, 1.5736, 2.1275], + device='cuda:3'), covar=tensor([0.1961, 0.2452, 0.0742, 0.4451, 0.1820, 0.3127, 0.2296, 0.2236], + device='cuda:3'), in_proj_covar=tensor([0.0522, 0.0597, 0.0552, 0.0638, 0.0640, 0.0590, 0.0530, 0.0627], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 02:55:41,825 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1708, 1.4892, 1.7930, 1.4435, 0.9929, 1.5785, 1.8869, 1.6132], + device='cuda:3'), covar=tensor([0.0476, 0.1295, 0.1671, 0.1444, 0.0582, 0.1475, 0.0650, 0.0651], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0152, 0.0189, 0.0158, 0.0098, 0.0163, 0.0112, 0.0143], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 02:55:45,700 INFO [train.py:901] (3/4) Epoch 21, batch 7050, loss[loss=0.2096, simple_loss=0.2994, pruned_loss=0.05992, over 8298.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2891, pruned_loss=0.06269, over 1616187.76 frames. ], batch size: 23, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:55:46,104 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.84 vs. limit=5.0 +2023-02-07 02:55:46,596 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168710.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:55:50,178 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.96 vs. limit=5.0 +2023-02-07 02:56:19,948 INFO [train.py:901] (3/4) Epoch 21, batch 7100, loss[loss=0.197, simple_loss=0.281, pruned_loss=0.0565, over 8193.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2888, pruned_loss=0.06219, over 1618470.09 frames. ], batch size: 23, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:56:26,887 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168768.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:56:40,683 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.770e+02 2.605e+02 3.011e+02 3.811e+02 1.077e+03, threshold=6.022e+02, percent-clipped=4.0 +2023-02-07 02:56:41,625 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4212, 1.4140, 1.8104, 1.2249, 1.0933, 1.7578, 0.2441, 1.1032], + device='cuda:3'), covar=tensor([0.1709, 0.1369, 0.0402, 0.1119, 0.2956, 0.0498, 0.2198, 0.1324], + device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0196, 0.0127, 0.0222, 0.0271, 0.0136, 0.0172, 0.0192], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 02:56:43,715 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168793.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:56:48,506 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2769, 2.6096, 2.9525, 1.5425, 3.1219, 1.9544, 1.4909, 2.0759], + device='cuda:3'), covar=tensor([0.0827, 0.0407, 0.0298, 0.0870, 0.0492, 0.0897, 0.0974, 0.0601], + device='cuda:3'), in_proj_covar=tensor([0.0453, 0.0391, 0.0342, 0.0443, 0.0375, 0.0534, 0.0389, 0.0417], + device='cuda:3'), out_proj_covar=tensor([1.2173e-04, 1.0258e-04, 9.0082e-05, 1.1684e-04, 9.8664e-05, 1.5105e-04, + 1.0518e-04, 1.1060e-04], device='cuda:3') +2023-02-07 02:56:55,260 INFO [train.py:901] (3/4) Epoch 21, batch 7150, loss[loss=0.2161, simple_loss=0.2969, pruned_loss=0.06761, over 8087.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2893, pruned_loss=0.06273, over 1614225.65 frames. ], batch size: 21, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:57:07,096 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.35 vs. limit=5.0 +2023-02-07 02:57:07,568 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9979, 2.3669, 3.4640, 1.9082, 2.9001, 2.3573, 2.1804, 2.7613], + device='cuda:3'), covar=tensor([0.1577, 0.2448, 0.0779, 0.3651, 0.1536, 0.2710, 0.1855, 0.2244], + device='cuda:3'), in_proj_covar=tensor([0.0525, 0.0603, 0.0558, 0.0643, 0.0645, 0.0596, 0.0534, 0.0633], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 02:57:29,819 INFO [train.py:901] (3/4) Epoch 21, batch 7200, loss[loss=0.2011, simple_loss=0.2826, pruned_loss=0.0598, over 8028.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2891, pruned_loss=0.06256, over 1614029.38 frames. ], batch size: 22, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:57:32,851 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.04 vs. limit=5.0 +2023-02-07 02:57:44,676 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0992, 2.3359, 1.8821, 2.8873, 1.4203, 1.6255, 1.9061, 2.2501], + device='cuda:3'), covar=tensor([0.0664, 0.0703, 0.0904, 0.0303, 0.1011, 0.1322, 0.0874, 0.0714], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0197, 0.0244, 0.0212, 0.0206, 0.0247, 0.0249, 0.0209], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 02:57:51,137 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.788e+02 2.343e+02 3.196e+02 4.097e+02 7.456e+02, threshold=6.392e+02, percent-clipped=6.0 +2023-02-07 02:57:56,895 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 +2023-02-07 02:58:04,698 INFO [train.py:901] (3/4) Epoch 21, batch 7250, loss[loss=0.185, simple_loss=0.2777, pruned_loss=0.04621, over 8248.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2899, pruned_loss=0.06344, over 1610352.03 frames. ], batch size: 24, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:58:18,030 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5887, 2.4669, 1.7769, 2.1420, 2.1718, 1.5487, 2.0238, 2.0143], + device='cuda:3'), covar=tensor([0.1487, 0.0463, 0.1311, 0.0644, 0.0685, 0.1588, 0.0898, 0.1092], + device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0237, 0.0337, 0.0309, 0.0302, 0.0338, 0.0347, 0.0319], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 02:58:40,072 INFO [train.py:901] (3/4) Epoch 21, batch 7300, loss[loss=0.1923, simple_loss=0.2782, pruned_loss=0.05314, over 8592.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2903, pruned_loss=0.06399, over 1609290.35 frames. ], batch size: 34, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:58:44,986 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168966.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:58:58,205 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168985.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:58:59,441 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168987.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 02:58:59,470 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0509, 1.3011, 1.0745, 1.9265, 0.8540, 1.0271, 1.3629, 1.3856], + device='cuda:3'), covar=tensor([0.1701, 0.1115, 0.2175, 0.0472, 0.1309, 0.2043, 0.0852, 0.1050], + device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0198, 0.0245, 0.0213, 0.0207, 0.0248, 0.0250, 0.0211], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 02:59:00,600 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.470e+02 2.375e+02 2.880e+02 4.111e+02 9.346e+02, threshold=5.760e+02, percent-clipped=6.0 +2023-02-07 02:59:02,085 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168991.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 02:59:14,684 INFO [train.py:901] (3/4) Epoch 21, batch 7350, loss[loss=0.174, simple_loss=0.2661, pruned_loss=0.04097, over 8232.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2889, pruned_loss=0.06308, over 1608493.80 frames. ], batch size: 22, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:59:35,045 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-07 02:59:36,091 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.50 vs. limit=2.0 +2023-02-07 02:59:44,846 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7974, 2.0935, 3.2542, 1.5820, 2.6012, 2.1485, 1.8645, 2.4972], + device='cuda:3'), covar=tensor([0.1774, 0.2492, 0.0833, 0.4362, 0.1737, 0.2910, 0.2207, 0.2256], + device='cuda:3'), in_proj_covar=tensor([0.0526, 0.0603, 0.0558, 0.0644, 0.0645, 0.0596, 0.0534, 0.0632], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 02:59:49,830 INFO [train.py:901] (3/4) Epoch 21, batch 7400, loss[loss=0.199, simple_loss=0.2789, pruned_loss=0.05956, over 8134.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2892, pruned_loss=0.06316, over 1610457.48 frames. ], batch size: 22, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 02:59:53,390 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-07 03:00:10,721 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.483e+02 2.322e+02 3.020e+02 4.298e+02 1.187e+03, threshold=6.039e+02, percent-clipped=6.0 +2023-02-07 03:00:19,215 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169100.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:00:25,151 INFO [train.py:901] (3/4) Epoch 21, batch 7450, loss[loss=0.2151, simple_loss=0.2988, pruned_loss=0.06569, over 8611.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2885, pruned_loss=0.06236, over 1611839.54 frames. ], batch size: 39, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 03:00:33,881 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-07 03:00:42,579 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169134.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:01:01,173 INFO [train.py:901] (3/4) Epoch 21, batch 7500, loss[loss=0.1996, simple_loss=0.2832, pruned_loss=0.05797, over 8187.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2889, pruned_loss=0.06281, over 1608460.52 frames. ], batch size: 23, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 03:01:13,526 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169177.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:01:21,454 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.677e+02 2.287e+02 2.739e+02 3.438e+02 5.948e+02, threshold=5.478e+02, percent-clipped=0.0 +2023-02-07 03:01:35,748 INFO [train.py:901] (3/4) Epoch 21, batch 7550, loss[loss=0.1902, simple_loss=0.2764, pruned_loss=0.05201, over 8039.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2887, pruned_loss=0.06254, over 1611080.94 frames. ], batch size: 22, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 03:02:09,758 INFO [train.py:901] (3/4) Epoch 21, batch 7600, loss[loss=0.241, simple_loss=0.3022, pruned_loss=0.08991, over 6396.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2877, pruned_loss=0.06219, over 1607188.87 frames. ], batch size: 14, lr: 3.57e-03, grad_scale: 8.0 +2023-02-07 03:02:32,180 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.566e+02 2.243e+02 2.742e+02 3.349e+02 1.012e+03, threshold=5.485e+02, percent-clipped=5.0 +2023-02-07 03:02:45,868 INFO [train.py:901] (3/4) Epoch 21, batch 7650, loss[loss=0.2106, simple_loss=0.2954, pruned_loss=0.06287, over 8512.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2895, pruned_loss=0.0632, over 1612743.42 frames. ], batch size: 28, lr: 3.57e-03, grad_scale: 16.0 +2023-02-07 03:03:00,452 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=169329.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:03:01,835 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=169331.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 03:03:21,300 INFO [train.py:901] (3/4) Epoch 21, batch 7700, loss[loss=0.1853, simple_loss=0.2733, pruned_loss=0.04861, over 8087.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.289, pruned_loss=0.06259, over 1613180.61 frames. ], batch size: 21, lr: 3.57e-03, grad_scale: 16.0 +2023-02-07 03:03:37,726 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.47 vs. limit=5.0 +2023-02-07 03:03:42,196 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.635e+02 2.349e+02 2.901e+02 3.736e+02 6.675e+02, threshold=5.802e+02, percent-clipped=6.0 +2023-02-07 03:03:44,229 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-07 03:03:45,051 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2742, 2.1070, 1.5777, 1.9207, 1.7727, 1.3987, 1.6431, 1.6573], + device='cuda:3'), covar=tensor([0.1427, 0.0401, 0.1287, 0.0551, 0.0703, 0.1499, 0.1024, 0.0932], + device='cuda:3'), in_proj_covar=tensor([0.0356, 0.0236, 0.0335, 0.0307, 0.0301, 0.0337, 0.0345, 0.0318], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 03:03:57,027 INFO [train.py:901] (3/4) Epoch 21, batch 7750, loss[loss=0.1727, simple_loss=0.2491, pruned_loss=0.04818, over 7411.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2905, pruned_loss=0.06381, over 1616967.35 frames. ], batch size: 17, lr: 3.57e-03, grad_scale: 16.0 +2023-02-07 03:04:21,907 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=169444.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:04:22,016 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169444.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:04:23,342 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169446.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 03:04:32,649 INFO [train.py:901] (3/4) Epoch 21, batch 7800, loss[loss=0.2418, simple_loss=0.3169, pruned_loss=0.08339, over 8523.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.289, pruned_loss=0.06272, over 1617294.63 frames. ], batch size: 31, lr: 3.57e-03, grad_scale: 16.0 +2023-02-07 03:04:45,398 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=169478.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:04:52,668 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.648e+02 2.145e+02 2.738e+02 3.428e+02 8.790e+02, threshold=5.476e+02, percent-clipped=3.0 +2023-02-07 03:05:06,018 INFO [train.py:901] (3/4) Epoch 21, batch 7850, loss[loss=0.1703, simple_loss=0.2541, pruned_loss=0.04328, over 7428.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2888, pruned_loss=0.06286, over 1619892.02 frames. ], batch size: 17, lr: 3.56e-03, grad_scale: 16.0 +2023-02-07 03:05:14,138 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=169521.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:05:39,268 INFO [train.py:901] (3/4) Epoch 21, batch 7900, loss[loss=0.2304, simple_loss=0.3083, pruned_loss=0.07629, over 8496.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2888, pruned_loss=0.06263, over 1621239.05 frames. ], batch size: 26, lr: 3.56e-03, grad_scale: 16.0 +2023-02-07 03:05:39,438 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169559.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:05:52,127 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5171, 1.3855, 4.6836, 1.7636, 4.1344, 3.8780, 4.2235, 4.1009], + device='cuda:3'), covar=tensor([0.0584, 0.5072, 0.0539, 0.4299, 0.1164, 0.1009, 0.0677, 0.0703], + device='cuda:3'), in_proj_covar=tensor([0.0631, 0.0641, 0.0695, 0.0627, 0.0711, 0.0612, 0.0612, 0.0676], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 03:05:59,284 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.374e+02 2.405e+02 2.884e+02 3.520e+02 8.387e+02, threshold=5.767e+02, percent-clipped=5.0 +2023-02-07 03:06:02,041 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169593.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:06:11,879 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6381, 2.0740, 3.3006, 1.5385, 2.3757, 2.0570, 1.7803, 2.4972], + device='cuda:3'), covar=tensor([0.1909, 0.2513, 0.0733, 0.4470, 0.1882, 0.3176, 0.2262, 0.2151], + device='cuda:3'), in_proj_covar=tensor([0.0523, 0.0601, 0.0556, 0.0637, 0.0642, 0.0591, 0.0531, 0.0632], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 03:06:12,853 INFO [train.py:901] (3/4) Epoch 21, batch 7950, loss[loss=0.2026, simple_loss=0.2917, pruned_loss=0.05672, over 8289.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2892, pruned_loss=0.0624, over 1623976.50 frames. ], batch size: 23, lr: 3.56e-03, grad_scale: 16.0 +2023-02-07 03:06:31,348 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169636.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:06:33,347 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169639.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:06:46,590 INFO [train.py:901] (3/4) Epoch 21, batch 8000, loss[loss=0.2094, simple_loss=0.2966, pruned_loss=0.06105, over 8442.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2886, pruned_loss=0.06188, over 1626375.54 frames. ], batch size: 27, lr: 3.56e-03, grad_scale: 16.0 +2023-02-07 03:07:06,440 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.455e+02 2.194e+02 2.844e+02 3.383e+02 6.688e+02, threshold=5.687e+02, percent-clipped=2.0 +2023-02-07 03:07:07,193 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.9645, 6.2066, 5.3653, 2.8261, 5.5033, 5.8398, 5.7468, 5.6420], + device='cuda:3'), covar=tensor([0.0601, 0.0359, 0.0901, 0.4022, 0.0670, 0.0798, 0.1017, 0.0555], + device='cuda:3'), in_proj_covar=tensor([0.0521, 0.0428, 0.0431, 0.0531, 0.0422, 0.0440, 0.0421, 0.0380], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 03:07:12,044 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9254, 1.6912, 2.0071, 1.7820, 1.9455, 1.9477, 1.8138, 0.8431], + device='cuda:3'), covar=tensor([0.5879, 0.4993, 0.2137, 0.3783, 0.2636, 0.3234, 0.2034, 0.5197], + device='cuda:3'), in_proj_covar=tensor([0.0945, 0.0977, 0.0803, 0.0946, 0.0998, 0.0892, 0.0746, 0.0823], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 03:07:12,147 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.60 vs. limit=2.0 +2023-02-07 03:07:14,031 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169700.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:07:15,422 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169702.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 03:07:19,705 INFO [train.py:901] (3/4) Epoch 21, batch 8050, loss[loss=0.2136, simple_loss=0.2913, pruned_loss=0.06796, over 7548.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2872, pruned_loss=0.06259, over 1596067.97 frames. ], batch size: 18, lr: 3.56e-03, grad_scale: 16.0 +2023-02-07 03:07:30,627 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169725.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:07:32,001 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169727.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 03:07:52,999 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-07 03:07:58,222 INFO [train.py:901] (3/4) Epoch 22, batch 0, loss[loss=0.215, simple_loss=0.282, pruned_loss=0.07399, over 8360.00 frames. ], tot_loss[loss=0.215, simple_loss=0.282, pruned_loss=0.07399, over 8360.00 frames. ], batch size: 24, lr: 3.48e-03, grad_scale: 16.0 +2023-02-07 03:07:58,222 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 03:08:05,067 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6925, 1.8454, 1.6380, 2.2675, 1.2054, 1.5084, 1.7158, 1.7848], + device='cuda:3'), covar=tensor([0.0781, 0.0772, 0.0893, 0.0447, 0.1150, 0.1266, 0.0797, 0.0854], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0196, 0.0245, 0.0213, 0.0207, 0.0246, 0.0250, 0.0209], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 03:08:09,347 INFO [train.py:935] (3/4) Epoch 22, validation: loss=0.1743, simple_loss=0.2746, pruned_loss=0.03702, over 944034.00 frames. +2023-02-07 03:08:09,348 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6736MB +2023-02-07 03:08:12,912 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169747.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:08:17,065 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7565, 1.7369, 2.0514, 2.0126, 1.0072, 1.7397, 2.1643, 2.2175], + device='cuda:3'), covar=tensor([0.0443, 0.1195, 0.1585, 0.1243, 0.0557, 0.1389, 0.0602, 0.0594], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0153, 0.0191, 0.0160, 0.0100, 0.0164, 0.0113, 0.0144], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 03:08:24,251 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-07 03:08:25,072 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169765.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:08:36,896 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1764, 2.1214, 1.7692, 1.9052, 1.7247, 1.5180, 1.6511, 1.6754], + device='cuda:3'), covar=tensor([0.1354, 0.0425, 0.1150, 0.0585, 0.0772, 0.1419, 0.0973, 0.0962], + device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0234, 0.0331, 0.0305, 0.0298, 0.0332, 0.0340, 0.0315], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 03:08:42,193 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.425e+02 2.482e+02 2.980e+02 3.558e+02 1.069e+03, threshold=5.959e+02, percent-clipped=8.0 +2023-02-07 03:08:44,174 INFO [train.py:901] (3/4) Epoch 22, batch 50, loss[loss=0.2316, simple_loss=0.3177, pruned_loss=0.07269, over 8578.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2881, pruned_loss=0.06026, over 365480.65 frames. ], batch size: 31, lr: 3.48e-03, grad_scale: 16.0 +2023-02-07 03:08:54,109 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169804.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:09:01,061 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-07 03:09:02,047 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169815.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:09:19,169 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169840.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:09:20,325 INFO [train.py:901] (3/4) Epoch 22, batch 100, loss[loss=0.1882, simple_loss=0.2827, pruned_loss=0.04691, over 8103.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2851, pruned_loss=0.05918, over 645952.47 frames. ], batch size: 23, lr: 3.48e-03, grad_scale: 16.0 +2023-02-07 03:09:23,123 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-07 03:09:25,371 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169849.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:09:42,071 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169874.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:09:52,438 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.2508, 4.2211, 3.8602, 1.8648, 3.7830, 3.8197, 3.7984, 3.5986], + device='cuda:3'), covar=tensor([0.0765, 0.0511, 0.1008, 0.4722, 0.0853, 0.1003, 0.1224, 0.0820], + device='cuda:3'), in_proj_covar=tensor([0.0519, 0.0428, 0.0429, 0.0531, 0.0423, 0.0440, 0.0421, 0.0381], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 03:09:52,903 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.649e+02 2.356e+02 3.069e+02 3.800e+02 7.981e+02, threshold=6.138e+02, percent-clipped=3.0 +2023-02-07 03:09:55,643 INFO [train.py:901] (3/4) Epoch 22, batch 150, loss[loss=0.1824, simple_loss=0.2696, pruned_loss=0.04766, over 8519.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.287, pruned_loss=0.06005, over 866423.93 frames. ], batch size: 26, lr: 3.48e-03, grad_scale: 16.0 +2023-02-07 03:09:55,887 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169892.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:10:12,771 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169917.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:10:30,755 INFO [train.py:901] (3/4) Epoch 22, batch 200, loss[loss=0.2586, simple_loss=0.3201, pruned_loss=0.09856, over 7382.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2881, pruned_loss=0.06144, over 1030858.94 frames. ], batch size: 71, lr: 3.48e-03, grad_scale: 16.0 +2023-02-07 03:10:53,648 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.89 vs. limit=5.0 +2023-02-07 03:10:58,713 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=169983.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:11:02,622 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.705e+02 2.362e+02 2.871e+02 3.395e+02 8.094e+02, threshold=5.742e+02, percent-clipped=2.0 +2023-02-07 03:11:04,635 INFO [train.py:901] (3/4) Epoch 22, batch 250, loss[loss=0.213, simple_loss=0.2969, pruned_loss=0.06458, over 8352.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.289, pruned_loss=0.06218, over 1162269.03 frames. ], batch size: 24, lr: 3.48e-03, grad_scale: 16.0 +2023-02-07 03:11:17,877 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-07 03:11:26,137 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-07 03:11:41,665 INFO [train.py:901] (3/4) Epoch 22, batch 300, loss[loss=0.2123, simple_loss=0.3004, pruned_loss=0.06214, over 8468.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2874, pruned_loss=0.06155, over 1259569.14 frames. ], batch size: 25, lr: 3.48e-03, grad_scale: 16.0 +2023-02-07 03:11:56,572 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170063.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:12:13,699 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.365e+02 2.486e+02 2.821e+02 3.492e+02 6.452e+02, threshold=5.641e+02, percent-clipped=3.0 +2023-02-07 03:12:15,180 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=170091.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:12:15,770 INFO [train.py:901] (3/4) Epoch 22, batch 350, loss[loss=0.2059, simple_loss=0.2794, pruned_loss=0.06625, over 7810.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2878, pruned_loss=0.06135, over 1342937.55 frames. ], batch size: 19, lr: 3.48e-03, grad_scale: 16.0 +2023-02-07 03:12:19,957 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170098.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:12:27,058 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=170109.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:12:49,705 INFO [train.py:901] (3/4) Epoch 22, batch 400, loss[loss=0.1984, simple_loss=0.2847, pruned_loss=0.05609, over 8581.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2888, pruned_loss=0.06203, over 1406129.78 frames. ], batch size: 49, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:12:53,701 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=170148.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:13:20,105 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2170, 1.9318, 2.6405, 2.2161, 2.5810, 2.2298, 2.0411, 1.3851], + device='cuda:3'), covar=tensor([0.5788, 0.5278, 0.1984, 0.3609, 0.2494, 0.3158, 0.1948, 0.5553], + device='cuda:3'), in_proj_covar=tensor([0.0942, 0.0975, 0.0800, 0.0942, 0.0992, 0.0890, 0.0744, 0.0823], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 03:13:22,570 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.722e+02 2.277e+02 2.821e+02 3.460e+02 6.418e+02, threshold=5.643e+02, percent-clipped=3.0 +2023-02-07 03:13:24,660 INFO [train.py:901] (3/4) Epoch 22, batch 450, loss[loss=0.1941, simple_loss=0.2815, pruned_loss=0.05338, over 7809.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2868, pruned_loss=0.06164, over 1448129.53 frames. ], batch size: 20, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:13:34,394 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170206.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:13:46,329 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170224.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:13:58,274 INFO [train.py:901] (3/4) Epoch 22, batch 500, loss[loss=0.2448, simple_loss=0.3289, pruned_loss=0.08033, over 8321.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2877, pruned_loss=0.06172, over 1486760.35 frames. ], batch size: 26, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:14:13,729 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170263.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:14:31,697 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.523e+02 2.263e+02 2.770e+02 3.716e+02 6.957e+02, threshold=5.540e+02, percent-clipped=5.0 +2023-02-07 03:14:34,528 INFO [train.py:901] (3/4) Epoch 22, batch 550, loss[loss=0.1748, simple_loss=0.2567, pruned_loss=0.04645, over 7918.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2868, pruned_loss=0.06126, over 1517596.41 frames. ], batch size: 20, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:15:00,839 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3279, 2.7033, 2.0968, 3.6997, 1.7358, 1.9750, 2.3456, 2.7546], + device='cuda:3'), covar=tensor([0.0765, 0.0912, 0.0930, 0.0324, 0.1149, 0.1325, 0.0942, 0.0807], + device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0196, 0.0243, 0.0213, 0.0206, 0.0245, 0.0248, 0.0208], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 03:15:08,205 INFO [train.py:901] (3/4) Epoch 22, batch 600, loss[loss=0.2165, simple_loss=0.2979, pruned_loss=0.06759, over 8454.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2872, pruned_loss=0.06127, over 1542065.52 frames. ], batch size: 27, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:15:16,556 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170354.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:15:27,512 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-07 03:15:34,302 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170379.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:15:40,802 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.766e+02 2.463e+02 3.010e+02 3.561e+02 9.437e+02, threshold=6.021e+02, percent-clipped=1.0 +2023-02-07 03:15:42,758 INFO [train.py:901] (3/4) Epoch 22, batch 650, loss[loss=0.2187, simple_loss=0.2879, pruned_loss=0.07481, over 5183.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2881, pruned_loss=0.06206, over 1551342.08 frames. ], batch size: 11, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:15:52,785 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.6699, 5.7583, 5.0811, 2.5760, 5.1022, 5.4673, 5.3841, 5.2596], + device='cuda:3'), covar=tensor([0.0547, 0.0352, 0.0855, 0.4245, 0.0718, 0.0780, 0.1078, 0.0578], + device='cuda:3'), in_proj_covar=tensor([0.0524, 0.0430, 0.0428, 0.0530, 0.0422, 0.0441, 0.0423, 0.0381], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 03:15:53,430 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=170407.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:16:17,646 INFO [train.py:901] (3/4) Epoch 22, batch 700, loss[loss=0.193, simple_loss=0.2673, pruned_loss=0.05939, over 7781.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2891, pruned_loss=0.06262, over 1567534.74 frames. ], batch size: 19, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:16:31,459 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170462.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:16:42,960 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170479.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:16:43,713 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170480.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:16:49,771 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170487.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:16:50,917 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.519e+02 2.347e+02 2.936e+02 3.672e+02 5.936e+02, threshold=5.871e+02, percent-clipped=0.0 +2023-02-07 03:16:52,905 INFO [train.py:901] (3/4) Epoch 22, batch 750, loss[loss=0.2054, simple_loss=0.2961, pruned_loss=0.05733, over 8289.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2885, pruned_loss=0.06195, over 1581851.71 frames. ], batch size: 23, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:17:01,712 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170505.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:17:11,627 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170519.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:17:13,547 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170522.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:17:14,719 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-07 03:17:23,956 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-07 03:17:27,404 INFO [train.py:901] (3/4) Epoch 22, batch 800, loss[loss=0.1813, simple_loss=0.2589, pruned_loss=0.05186, over 7662.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2882, pruned_loss=0.06196, over 1590501.88 frames. ], batch size: 19, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:17:28,960 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170544.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:17:30,549 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 +2023-02-07 03:17:57,595 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170587.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:17:58,762 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.388e+02 2.234e+02 2.598e+02 3.180e+02 6.753e+02, threshold=5.195e+02, percent-clipped=1.0 +2023-02-07 03:18:00,808 INFO [train.py:901] (3/4) Epoch 22, batch 850, loss[loss=0.2129, simple_loss=0.2898, pruned_loss=0.06801, over 8244.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2876, pruned_loss=0.0615, over 1598379.19 frames. ], batch size: 22, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:18:14,746 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6712, 1.6587, 2.4391, 1.6166, 1.2823, 2.4578, 0.5210, 1.4898], + device='cuda:3'), covar=tensor([0.1741, 0.1376, 0.0312, 0.1450, 0.2977, 0.0438, 0.2278, 0.1441], + device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0196, 0.0127, 0.0222, 0.0270, 0.0135, 0.0171, 0.0193], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 03:18:31,335 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-07 03:18:36,995 INFO [train.py:901] (3/4) Epoch 22, batch 900, loss[loss=0.1942, simple_loss=0.2586, pruned_loss=0.06493, over 7270.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2878, pruned_loss=0.06175, over 1601764.96 frames. ], batch size: 16, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:18:53,611 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5174, 1.8155, 1.9423, 1.2605, 2.0883, 1.4233, 0.6339, 1.6745], + device='cuda:3'), covar=tensor([0.0616, 0.0366, 0.0288, 0.0579, 0.0394, 0.0864, 0.0830, 0.0340], + device='cuda:3'), in_proj_covar=tensor([0.0454, 0.0394, 0.0345, 0.0444, 0.0375, 0.0534, 0.0390, 0.0419], + device='cuda:3'), out_proj_covar=tensor([1.2187e-04, 1.0338e-04, 9.0653e-05, 1.1683e-04, 9.8477e-05, 1.5078e-04, + 1.0542e-04, 1.1114e-04], device='cuda:3') +2023-02-07 03:19:09,382 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.739e+02 2.381e+02 2.827e+02 3.296e+02 7.509e+02, threshold=5.655e+02, percent-clipped=4.0 +2023-02-07 03:19:11,447 INFO [train.py:901] (3/4) Epoch 22, batch 950, loss[loss=0.2083, simple_loss=0.2882, pruned_loss=0.06422, over 7800.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2879, pruned_loss=0.06174, over 1607672.02 frames. ], batch size: 19, lr: 3.47e-03, grad_scale: 16.0 +2023-02-07 03:19:13,664 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7215, 1.4668, 2.8841, 1.3732, 2.2215, 3.1148, 3.2219, 2.6404], + device='cuda:3'), covar=tensor([0.1158, 0.1565, 0.0350, 0.2000, 0.0788, 0.0275, 0.0546, 0.0571], + device='cuda:3'), in_proj_covar=tensor([0.0296, 0.0323, 0.0286, 0.0317, 0.0310, 0.0265, 0.0420, 0.0306], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 03:19:43,585 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-07 03:19:46,359 INFO [train.py:901] (3/4) Epoch 22, batch 1000, loss[loss=0.1645, simple_loss=0.2549, pruned_loss=0.03707, over 8081.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2874, pruned_loss=0.06135, over 1612892.67 frames. ], batch size: 21, lr: 3.47e-03, grad_scale: 8.0 +2023-02-07 03:19:49,225 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5163, 2.4509, 1.8182, 2.1670, 2.0686, 1.5913, 1.9695, 2.0503], + device='cuda:3'), covar=tensor([0.1383, 0.0424, 0.1181, 0.0610, 0.0794, 0.1460, 0.1010, 0.0979], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0235, 0.0331, 0.0308, 0.0299, 0.0334, 0.0341, 0.0318], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 03:20:12,107 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170778.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:20:17,096 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-07 03:20:19,771 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.250e+02 2.326e+02 2.890e+02 3.504e+02 6.405e+02, threshold=5.779e+02, percent-clipped=4.0 +2023-02-07 03:20:21,023 INFO [train.py:901] (3/4) Epoch 22, batch 1050, loss[loss=0.1845, simple_loss=0.2611, pruned_loss=0.05394, over 7796.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2884, pruned_loss=0.06193, over 1614595.34 frames. ], batch size: 19, lr: 3.47e-03, grad_scale: 8.0 +2023-02-07 03:20:28,489 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-07 03:20:28,711 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170803.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:20:41,786 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=170823.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:20:55,953 INFO [train.py:901] (3/4) Epoch 22, batch 1100, loss[loss=0.2107, simple_loss=0.2933, pruned_loss=0.06406, over 8353.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2882, pruned_loss=0.0617, over 1615122.71 frames. ], batch size: 24, lr: 3.47e-03, grad_scale: 8.0 +2023-02-07 03:21:27,510 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6784, 2.4404, 1.8264, 2.3011, 2.2409, 1.6214, 2.1149, 2.1278], + device='cuda:3'), covar=tensor([0.1283, 0.0422, 0.1163, 0.0516, 0.0697, 0.1446, 0.0873, 0.0861], + device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0237, 0.0334, 0.0310, 0.0300, 0.0336, 0.0344, 0.0319], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 03:21:29,315 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.559e+02 2.576e+02 3.127e+02 3.706e+02 1.049e+03, threshold=6.255e+02, percent-clipped=5.0 +2023-02-07 03:21:30,686 INFO [train.py:901] (3/4) Epoch 22, batch 1150, loss[loss=0.2526, simple_loss=0.3328, pruned_loss=0.08623, over 8355.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2894, pruned_loss=0.06289, over 1617274.20 frames. ], batch size: 26, lr: 3.47e-03, grad_scale: 8.0 +2023-02-07 03:21:37,435 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-07 03:21:45,387 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3695, 1.3658, 2.3097, 1.2390, 2.2321, 2.5123, 2.6165, 1.9612], + device='cuda:3'), covar=tensor([0.1215, 0.1433, 0.0521, 0.2110, 0.0750, 0.0437, 0.0705, 0.0924], + device='cuda:3'), in_proj_covar=tensor([0.0293, 0.0320, 0.0283, 0.0314, 0.0306, 0.0262, 0.0415, 0.0304], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 03:21:52,885 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4102, 1.4726, 1.3943, 1.7769, 0.7351, 1.2438, 1.2318, 1.4726], + device='cuda:3'), covar=tensor([0.0823, 0.0774, 0.0981, 0.0509, 0.1104, 0.1295, 0.0804, 0.0712], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0198, 0.0247, 0.0215, 0.0209, 0.0249, 0.0251, 0.0210], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 03:21:56,818 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=170931.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:22:01,709 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170938.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:22:04,216 INFO [train.py:901] (3/4) Epoch 22, batch 1200, loss[loss=0.2472, simple_loss=0.3268, pruned_loss=0.08379, over 8448.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2889, pruned_loss=0.06309, over 1614756.28 frames. ], batch size: 27, lr: 3.47e-03, grad_scale: 8.0 +2023-02-07 03:22:07,062 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170946.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:22:38,803 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.411e+02 2.384e+02 2.807e+02 3.549e+02 5.873e+02, threshold=5.615e+02, percent-clipped=0.0 +2023-02-07 03:22:40,087 INFO [train.py:901] (3/4) Epoch 22, batch 1250, loss[loss=0.205, simple_loss=0.2885, pruned_loss=0.06075, over 8462.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2881, pruned_loss=0.06227, over 1615130.43 frames. ], batch size: 25, lr: 3.47e-03, grad_scale: 8.0 +2023-02-07 03:22:57,673 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.7468, 4.7555, 4.2517, 2.2510, 4.1379, 4.3561, 4.3417, 4.1118], + device='cuda:3'), covar=tensor([0.0684, 0.0490, 0.1035, 0.4284, 0.0878, 0.0809, 0.1113, 0.0666], + device='cuda:3'), in_proj_covar=tensor([0.0522, 0.0430, 0.0430, 0.0530, 0.0422, 0.0441, 0.0420, 0.0382], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 03:22:58,785 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2023-02-07 03:23:14,553 INFO [train.py:901] (3/4) Epoch 22, batch 1300, loss[loss=0.2009, simple_loss=0.2806, pruned_loss=0.06063, over 7963.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2872, pruned_loss=0.06187, over 1617509.09 frames. ], batch size: 21, lr: 3.47e-03, grad_scale: 8.0 +2023-02-07 03:23:17,512 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171046.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:23:47,495 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.677e+02 2.382e+02 2.988e+02 3.753e+02 7.309e+02, threshold=5.975e+02, percent-clipped=5.0 +2023-02-07 03:23:48,840 INFO [train.py:901] (3/4) Epoch 22, batch 1350, loss[loss=0.1719, simple_loss=0.2646, pruned_loss=0.03957, over 8240.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2872, pruned_loss=0.06154, over 1614500.83 frames. ], batch size: 22, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:24:01,690 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171110.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:24:02,363 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3512, 1.6471, 4.5607, 1.6400, 4.0059, 3.7720, 4.0796, 3.9376], + device='cuda:3'), covar=tensor([0.0651, 0.4645, 0.0596, 0.4459, 0.1309, 0.1122, 0.0675, 0.0763], + device='cuda:3'), in_proj_covar=tensor([0.0625, 0.0637, 0.0688, 0.0620, 0.0704, 0.0604, 0.0606, 0.0672], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 03:24:23,450 INFO [train.py:901] (3/4) Epoch 22, batch 1400, loss[loss=0.2233, simple_loss=0.308, pruned_loss=0.06932, over 8623.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2877, pruned_loss=0.06184, over 1615656.03 frames. ], batch size: 31, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:24:23,822 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.90 vs. limit=5.0 +2023-02-07 03:24:55,486 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.434e+02 3.047e+02 3.835e+02 9.203e+02, threshold=6.094e+02, percent-clipped=3.0 +2023-02-07 03:24:57,483 INFO [train.py:901] (3/4) Epoch 22, batch 1450, loss[loss=0.2573, simple_loss=0.3349, pruned_loss=0.08979, over 8498.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.289, pruned_loss=0.06237, over 1617172.71 frames. ], batch size: 28, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:24:58,893 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171194.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:25:06,227 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-07 03:25:12,520 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171214.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:25:16,688 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171219.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:25:32,454 INFO [train.py:901] (3/4) Epoch 22, batch 1500, loss[loss=0.2323, simple_loss=0.31, pruned_loss=0.07729, over 8727.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2896, pruned_loss=0.06287, over 1619756.51 frames. ], batch size: 34, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:26:04,593 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.623e+02 2.477e+02 2.962e+02 3.885e+02 1.079e+03, threshold=5.924e+02, percent-clipped=2.0 +2023-02-07 03:26:04,686 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=171290.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:26:05,961 INFO [train.py:901] (3/4) Epoch 22, batch 1550, loss[loss=0.1948, simple_loss=0.2793, pruned_loss=0.05509, over 7972.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2899, pruned_loss=0.06317, over 1619643.00 frames. ], batch size: 21, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:26:12,947 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171302.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:26:30,106 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171327.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:26:31,051 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 +2023-02-07 03:26:40,691 INFO [train.py:901] (3/4) Epoch 22, batch 1600, loss[loss=0.1831, simple_loss=0.2572, pruned_loss=0.05448, over 7178.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2895, pruned_loss=0.06323, over 1614838.74 frames. ], batch size: 16, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:26:48,987 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1590, 1.0538, 1.2495, 1.0879, 0.9889, 1.2727, 0.1259, 0.9964], + device='cuda:3'), covar=tensor([0.1608, 0.1262, 0.0546, 0.0766, 0.2451, 0.0603, 0.2121, 0.1134], + device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0196, 0.0127, 0.0222, 0.0269, 0.0135, 0.0170, 0.0192], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 03:26:55,769 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171363.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:27:13,639 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.350e+02 2.510e+02 3.045e+02 3.987e+02 6.104e+02, threshold=6.090e+02, percent-clipped=2.0 +2023-02-07 03:27:15,007 INFO [train.py:901] (3/4) Epoch 22, batch 1650, loss[loss=0.2131, simple_loss=0.2973, pruned_loss=0.06445, over 8337.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2895, pruned_loss=0.06289, over 1613669.78 frames. ], batch size: 26, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:27:24,112 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171405.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:27:28,874 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8924, 3.5096, 2.1912, 2.6366, 2.6612, 1.9164, 2.6313, 2.9003], + device='cuda:3'), covar=tensor([0.1976, 0.0415, 0.1314, 0.0869, 0.0826, 0.1615, 0.1187, 0.1137], + device='cuda:3'), in_proj_covar=tensor([0.0356, 0.0238, 0.0337, 0.0311, 0.0302, 0.0340, 0.0346, 0.0321], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 03:27:51,031 INFO [train.py:901] (3/4) Epoch 22, batch 1700, loss[loss=0.1741, simple_loss=0.2466, pruned_loss=0.05078, over 7238.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2897, pruned_loss=0.06332, over 1608333.94 frames. ], batch size: 16, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:27:59,282 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=171454.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:28:01,709 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.89 vs. limit=5.0 +2023-02-07 03:28:24,568 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.494e+02 2.429e+02 3.050e+02 3.629e+02 7.357e+02, threshold=6.100e+02, percent-clipped=3.0 +2023-02-07 03:28:25,936 INFO [train.py:901] (3/4) Epoch 22, batch 1750, loss[loss=0.2356, simple_loss=0.3121, pruned_loss=0.07951, over 7069.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2891, pruned_loss=0.06304, over 1610298.66 frames. ], batch size: 72, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:28:42,139 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171516.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:28:58,898 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 +2023-02-07 03:29:00,470 INFO [train.py:901] (3/4) Epoch 22, batch 1800, loss[loss=0.1881, simple_loss=0.2709, pruned_loss=0.05264, over 8356.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2902, pruned_loss=0.06371, over 1614674.92 frames. ], batch size: 24, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:29:11,511 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=171558.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:29:19,688 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171569.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:29:34,591 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.630e+02 2.408e+02 2.801e+02 3.784e+02 7.831e+02, threshold=5.602e+02, percent-clipped=2.0 +2023-02-07 03:29:35,958 INFO [train.py:901] (3/4) Epoch 22, batch 1850, loss[loss=0.1682, simple_loss=0.2393, pruned_loss=0.04852, over 7701.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2899, pruned_loss=0.0634, over 1618674.94 frames. ], batch size: 18, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:30:10,047 INFO [train.py:901] (3/4) Epoch 22, batch 1900, loss[loss=0.2315, simple_loss=0.3113, pruned_loss=0.07578, over 8633.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2895, pruned_loss=0.06288, over 1617146.21 frames. ], batch size: 39, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:30:24,228 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171661.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:30:32,191 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171673.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:30:36,776 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-07 03:30:41,592 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171686.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:30:44,064 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.710e+02 2.518e+02 3.035e+02 3.649e+02 9.576e+02, threshold=6.070e+02, percent-clipped=4.0 +2023-02-07 03:30:45,465 INFO [train.py:901] (3/4) Epoch 22, batch 1950, loss[loss=0.199, simple_loss=0.2865, pruned_loss=0.05575, over 8286.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2894, pruned_loss=0.06289, over 1620290.84 frames. ], batch size: 23, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:30:48,018 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-07 03:30:56,320 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=171707.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:31:07,793 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-07 03:31:20,039 INFO [train.py:901] (3/4) Epoch 22, batch 2000, loss[loss=0.267, simple_loss=0.3318, pruned_loss=0.1011, over 8336.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2893, pruned_loss=0.06296, over 1617561.33 frames. ], batch size: 25, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:31:43,122 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9660, 1.5355, 4.3191, 1.9760, 2.4276, 4.9398, 4.9795, 4.2037], + device='cuda:3'), covar=tensor([0.1422, 0.1984, 0.0348, 0.2042, 0.1369, 0.0202, 0.0476, 0.0606], + device='cuda:3'), in_proj_covar=tensor([0.0295, 0.0323, 0.0285, 0.0318, 0.0307, 0.0264, 0.0419, 0.0305], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 03:31:53,990 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.593e+02 2.301e+02 2.928e+02 3.706e+02 6.798e+02, threshold=5.855e+02, percent-clipped=1.0 +2023-02-07 03:31:55,404 INFO [train.py:901] (3/4) Epoch 22, batch 2050, loss[loss=0.2072, simple_loss=0.2953, pruned_loss=0.0595, over 8493.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2883, pruned_loss=0.06215, over 1617886.40 frames. ], batch size: 28, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:32:17,606 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171822.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:32:19,747 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171825.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:32:31,149 INFO [train.py:901] (3/4) Epoch 22, batch 2100, loss[loss=0.1759, simple_loss=0.2594, pruned_loss=0.04618, over 8083.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.289, pruned_loss=0.06215, over 1622566.32 frames. ], batch size: 21, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:32:36,872 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171850.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:32:43,493 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=171860.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:32:45,785 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6816, 2.3374, 3.5365, 1.9011, 1.6855, 3.3766, 0.5116, 2.0398], + device='cuda:3'), covar=tensor([0.1710, 0.1300, 0.0243, 0.1986, 0.3054, 0.0359, 0.2739, 0.1631], + device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0198, 0.0127, 0.0223, 0.0272, 0.0137, 0.0171, 0.0194], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 03:33:05,591 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.927e+02 2.505e+02 2.999e+02 3.749e+02 9.868e+02, threshold=5.998e+02, percent-clipped=7.0 +2023-02-07 03:33:06,894 INFO [train.py:901] (3/4) Epoch 22, batch 2150, loss[loss=0.1906, simple_loss=0.2842, pruned_loss=0.04846, over 8193.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2886, pruned_loss=0.06193, over 1627096.93 frames. ], batch size: 23, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:33:33,131 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171929.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:33:42,593 INFO [train.py:901] (3/4) Epoch 22, batch 2200, loss[loss=0.216, simple_loss=0.292, pruned_loss=0.06995, over 7810.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.289, pruned_loss=0.0624, over 1621799.24 frames. ], batch size: 20, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:33:51,040 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171954.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:33:51,697 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171955.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:34:04,374 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0 +2023-02-07 03:34:05,585 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171975.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:34:15,550 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.459e+02 2.362e+02 2.812e+02 3.623e+02 6.076e+02, threshold=5.624e+02, percent-clipped=1.0 +2023-02-07 03:34:16,940 INFO [train.py:901] (3/4) Epoch 22, batch 2250, loss[loss=0.2031, simple_loss=0.2867, pruned_loss=0.05976, over 8091.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2873, pruned_loss=0.06153, over 1620536.72 frames. ], batch size: 21, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:34:17,294 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-07 03:34:54,269 INFO [train.py:901] (3/4) Epoch 22, batch 2300, loss[loss=0.1998, simple_loss=0.2752, pruned_loss=0.06218, over 7537.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2875, pruned_loss=0.06136, over 1621040.42 frames. ], batch size: 18, lr: 3.46e-03, grad_scale: 8.0 +2023-02-07 03:35:19,363 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0919, 3.6844, 2.3623, 2.8504, 2.8510, 2.1988, 2.7833, 2.9858], + device='cuda:3'), covar=tensor([0.1607, 0.0330, 0.1066, 0.0729, 0.0705, 0.1337, 0.1086, 0.1135], + device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0233, 0.0331, 0.0307, 0.0299, 0.0336, 0.0343, 0.0316], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 03:35:20,116 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172078.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:35:28,297 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.806e+02 2.401e+02 3.005e+02 3.667e+02 7.010e+02, threshold=6.010e+02, percent-clipped=1.0 +2023-02-07 03:35:29,621 INFO [train.py:901] (3/4) Epoch 22, batch 2350, loss[loss=0.2447, simple_loss=0.3231, pruned_loss=0.0831, over 8464.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2883, pruned_loss=0.06139, over 1623780.09 frames. ], batch size: 25, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:35:37,311 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172103.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:36:01,305 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172136.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:36:05,214 INFO [train.py:901] (3/4) Epoch 22, batch 2400, loss[loss=0.2246, simple_loss=0.3037, pruned_loss=0.07277, over 8025.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2896, pruned_loss=0.06244, over 1621581.84 frames. ], batch size: 22, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:36:39,691 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.809e+02 2.658e+02 3.455e+02 4.348e+02 7.809e+02, threshold=6.910e+02, percent-clipped=6.0 +2023-02-07 03:36:41,123 INFO [train.py:901] (3/4) Epoch 22, batch 2450, loss[loss=0.1764, simple_loss=0.2619, pruned_loss=0.04547, over 7814.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2889, pruned_loss=0.06226, over 1622260.38 frames. ], batch size: 20, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:36:45,732 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5043, 2.3823, 3.2297, 2.5999, 3.1115, 2.5333, 2.2662, 1.8451], + device='cuda:3'), covar=tensor([0.5078, 0.4854, 0.1900, 0.3322, 0.2331, 0.2848, 0.1817, 0.5222], + device='cuda:3'), in_proj_covar=tensor([0.0944, 0.0978, 0.0805, 0.0942, 0.0997, 0.0895, 0.0748, 0.0828], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 03:37:08,138 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172231.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:37:16,955 INFO [train.py:901] (3/4) Epoch 22, batch 2500, loss[loss=0.2279, simple_loss=0.3122, pruned_loss=0.07181, over 8286.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.289, pruned_loss=0.06219, over 1623976.69 frames. ], batch size: 23, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:37:26,701 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172256.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:37:37,000 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9841, 2.1893, 1.8616, 2.7790, 1.3247, 1.6252, 1.9022, 2.1884], + device='cuda:3'), covar=tensor([0.0775, 0.0815, 0.0842, 0.0399, 0.1049, 0.1263, 0.0911, 0.0781], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0197, 0.0243, 0.0215, 0.0206, 0.0248, 0.0251, 0.0210], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 03:37:50,850 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.717e+02 2.288e+02 2.722e+02 3.540e+02 9.975e+02, threshold=5.443e+02, percent-clipped=1.0 +2023-02-07 03:37:52,252 INFO [train.py:901] (3/4) Epoch 22, batch 2550, loss[loss=0.162, simple_loss=0.2386, pruned_loss=0.0427, over 7918.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2882, pruned_loss=0.06205, over 1621788.82 frames. ], batch size: 20, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:37:56,722 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=172299.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:38:25,419 INFO [train.py:901] (3/4) Epoch 22, batch 2600, loss[loss=0.1994, simple_loss=0.2864, pruned_loss=0.05624, over 8360.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.288, pruned_loss=0.0621, over 1618987.54 frames. ], batch size: 24, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:38:58,398 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.528e+02 2.484e+02 3.096e+02 3.957e+02 1.134e+03, threshold=6.191e+02, percent-clipped=6.0 +2023-02-07 03:39:00,474 INFO [train.py:901] (3/4) Epoch 22, batch 2650, loss[loss=0.2063, simple_loss=0.2944, pruned_loss=0.05913, over 8497.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2878, pruned_loss=0.06177, over 1620902.46 frames. ], batch size: 28, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:39:10,806 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3537, 2.8790, 2.0315, 3.9142, 1.7669, 2.1339, 2.4177, 2.9585], + device='cuda:3'), covar=tensor([0.0691, 0.0716, 0.0941, 0.0268, 0.1023, 0.1160, 0.0903, 0.0736], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0197, 0.0244, 0.0216, 0.0206, 0.0248, 0.0251, 0.0210], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 03:39:16,288 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=172414.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:39:35,347 INFO [train.py:901] (3/4) Epoch 22, batch 2700, loss[loss=0.2064, simple_loss=0.2961, pruned_loss=0.05839, over 8258.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2891, pruned_loss=0.06281, over 1619778.11 frames. ], batch size: 24, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:40:02,839 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=172480.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:40:09,203 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 +2023-02-07 03:40:09,437 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.361e+02 2.345e+02 2.798e+02 3.767e+02 1.133e+03, threshold=5.596e+02, percent-clipped=4.0 +2023-02-07 03:40:10,842 INFO [train.py:901] (3/4) Epoch 22, batch 2750, loss[loss=0.1774, simple_loss=0.2512, pruned_loss=0.05181, over 7531.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2892, pruned_loss=0.06285, over 1620899.61 frames. ], batch size: 18, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:40:45,667 INFO [train.py:901] (3/4) Epoch 22, batch 2800, loss[loss=0.2041, simple_loss=0.2951, pruned_loss=0.05653, over 8350.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2897, pruned_loss=0.06327, over 1621690.54 frames. ], batch size: 24, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:41:18,219 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.794e+02 2.395e+02 2.840e+02 3.614e+02 7.820e+02, threshold=5.680e+02, percent-clipped=6.0 +2023-02-07 03:41:20,376 INFO [train.py:901] (3/4) Epoch 22, batch 2850, loss[loss=0.1908, simple_loss=0.2719, pruned_loss=0.05482, over 7965.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2894, pruned_loss=0.06303, over 1619815.58 frames. ], batch size: 21, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:41:23,221 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=172595.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:41:25,863 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.2406, 4.1922, 3.8434, 1.9447, 3.7611, 3.7824, 3.7957, 3.5669], + device='cuda:3'), covar=tensor([0.0804, 0.0610, 0.1111, 0.4603, 0.1004, 0.1104, 0.1428, 0.0850], + device='cuda:3'), in_proj_covar=tensor([0.0520, 0.0431, 0.0427, 0.0532, 0.0423, 0.0443, 0.0424, 0.0383], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 03:41:37,778 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172616.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:41:44,849 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-07 03:41:56,011 INFO [train.py:901] (3/4) Epoch 22, batch 2900, loss[loss=0.1826, simple_loss=0.2626, pruned_loss=0.05127, over 7775.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2881, pruned_loss=0.06224, over 1615767.11 frames. ], batch size: 19, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:41:57,554 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172643.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:42:15,797 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172670.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:42:24,163 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-07 03:42:28,901 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.717e+02 2.482e+02 2.975e+02 3.907e+02 6.756e+02, threshold=5.949e+02, percent-clipped=4.0 +2023-02-07 03:42:30,283 INFO [train.py:901] (3/4) Epoch 22, batch 2950, loss[loss=0.2197, simple_loss=0.2958, pruned_loss=0.07175, over 8607.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2875, pruned_loss=0.06192, over 1611220.43 frames. ], batch size: 31, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:42:32,552 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172695.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:42:56,537 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4763, 2.3384, 3.1924, 2.5508, 3.0085, 2.4604, 2.2771, 1.7491], + device='cuda:3'), covar=tensor([0.5451, 0.4849, 0.1890, 0.3648, 0.2383, 0.2935, 0.1773, 0.5670], + device='cuda:3'), in_proj_covar=tensor([0.0944, 0.0978, 0.0805, 0.0942, 0.0995, 0.0896, 0.0745, 0.0827], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 03:43:05,661 INFO [train.py:901] (3/4) Epoch 22, batch 3000, loss[loss=0.1642, simple_loss=0.2373, pruned_loss=0.04551, over 7696.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2874, pruned_loss=0.06205, over 1611076.19 frames. ], batch size: 18, lr: 3.45e-03, grad_scale: 16.0 +2023-02-07 03:43:05,662 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 03:43:17,969 INFO [train.py:935] (3/4) Epoch 22, validation: loss=0.1735, simple_loss=0.2739, pruned_loss=0.03659, over 944034.00 frames. +2023-02-07 03:43:17,970 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6747MB +2023-02-07 03:43:25,648 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172752.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:43:51,448 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.429e+02 2.191e+02 2.765e+02 3.574e+02 6.067e+02, threshold=5.530e+02, percent-clipped=1.0 +2023-02-07 03:43:52,760 INFO [train.py:901] (3/4) Epoch 22, batch 3050, loss[loss=0.2046, simple_loss=0.2869, pruned_loss=0.06109, over 7804.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2868, pruned_loss=0.06168, over 1608552.45 frames. ], batch size: 19, lr: 3.45e-03, grad_scale: 16.0 +2023-02-07 03:44:23,922 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3330, 1.4385, 1.3104, 1.7636, 0.7131, 1.2455, 1.2589, 1.4297], + device='cuda:3'), covar=tensor([0.0913, 0.0812, 0.1065, 0.0509, 0.1180, 0.1360, 0.0787, 0.0828], + device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0198, 0.0246, 0.0217, 0.0208, 0.0249, 0.0252, 0.0211], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 03:44:26,388 INFO [train.py:901] (3/4) Epoch 22, batch 3100, loss[loss=0.1997, simple_loss=0.2838, pruned_loss=0.0578, over 8343.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2868, pruned_loss=0.06147, over 1605652.84 frames. ], batch size: 26, lr: 3.45e-03, grad_scale: 16.0 +2023-02-07 03:44:32,707 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172851.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:44:40,606 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172863.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:44:50,780 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172876.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:44:59,807 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.536e+02 2.451e+02 3.163e+02 4.463e+02 7.617e+02, threshold=6.327e+02, percent-clipped=7.0 +2023-02-07 03:45:01,202 INFO [train.py:901] (3/4) Epoch 22, batch 3150, loss[loss=0.2208, simple_loss=0.2947, pruned_loss=0.07346, over 8086.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2864, pruned_loss=0.06136, over 1607632.46 frames. ], batch size: 21, lr: 3.45e-03, grad_scale: 16.0 +2023-02-07 03:45:33,032 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5012, 1.9916, 3.0941, 1.3646, 2.3603, 1.8802, 1.5760, 2.3887], + device='cuda:3'), covar=tensor([0.1904, 0.2446, 0.0886, 0.4448, 0.1713, 0.3194, 0.2283, 0.2099], + device='cuda:3'), in_proj_covar=tensor([0.0530, 0.0605, 0.0558, 0.0642, 0.0645, 0.0591, 0.0534, 0.0630], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 03:45:35,482 INFO [train.py:901] (3/4) Epoch 22, batch 3200, loss[loss=0.1864, simple_loss=0.2674, pruned_loss=0.05272, over 7684.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2863, pruned_loss=0.06114, over 1608660.47 frames. ], batch size: 18, lr: 3.45e-03, grad_scale: 16.0 +2023-02-07 03:45:40,969 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1164, 2.3779, 1.9318, 2.8809, 1.3371, 1.7087, 2.0368, 2.3168], + device='cuda:3'), covar=tensor([0.0702, 0.0708, 0.0863, 0.0342, 0.1108, 0.1196, 0.0803, 0.0750], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0198, 0.0246, 0.0217, 0.0207, 0.0249, 0.0252, 0.0210], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 03:45:47,744 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=172960.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:45:53,317 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1481, 1.4690, 1.7247, 1.3757, 1.0114, 1.4572, 1.9118, 1.6447], + device='cuda:3'), covar=tensor([0.0490, 0.1238, 0.1611, 0.1461, 0.0584, 0.1482, 0.0637, 0.0622], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0189, 0.0159, 0.0099, 0.0163, 0.0111, 0.0142], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 03:46:06,753 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=172987.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:46:09,263 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.402e+02 2.554e+02 2.964e+02 3.773e+02 6.891e+02, threshold=5.928e+02, percent-clipped=2.0 +2023-02-07 03:46:10,583 INFO [train.py:901] (3/4) Epoch 22, batch 3250, loss[loss=0.1968, simple_loss=0.2951, pruned_loss=0.04924, over 8107.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2866, pruned_loss=0.06117, over 1611530.68 frames. ], batch size: 23, lr: 3.45e-03, grad_scale: 16.0 +2023-02-07 03:46:28,200 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1481, 3.7593, 2.4460, 3.0307, 2.7491, 1.9777, 2.8994, 3.2702], + device='cuda:3'), covar=tensor([0.1492, 0.0308, 0.1059, 0.0719, 0.0747, 0.1550, 0.0993, 0.0804], + device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0235, 0.0336, 0.0312, 0.0302, 0.0341, 0.0348, 0.0321], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 03:46:45,375 INFO [train.py:901] (3/4) Epoch 22, batch 3300, loss[loss=0.223, simple_loss=0.295, pruned_loss=0.07552, over 8576.00 frames. ], tot_loss[loss=0.205, simple_loss=0.287, pruned_loss=0.06153, over 1612227.59 frames. ], batch size: 31, lr: 3.45e-03, grad_scale: 8.0 +2023-02-07 03:47:07,565 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173075.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:47:17,930 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.491e+02 2.506e+02 2.831e+02 3.669e+02 6.075e+02, threshold=5.662e+02, percent-clipped=1.0 +2023-02-07 03:47:18,594 INFO [train.py:901] (3/4) Epoch 22, batch 3350, loss[loss=0.23, simple_loss=0.2974, pruned_loss=0.08131, over 8442.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2869, pruned_loss=0.06215, over 1607838.65 frames. ], batch size: 27, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:47:22,044 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=173096.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:47:26,752 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173102.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:47:54,975 INFO [train.py:901] (3/4) Epoch 22, batch 3400, loss[loss=0.2365, simple_loss=0.3179, pruned_loss=0.07753, over 8467.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2878, pruned_loss=0.06286, over 1608496.60 frames. ], batch size: 25, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:48:12,781 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173168.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:48:21,518 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4243, 1.3941, 4.6049, 1.7235, 4.1181, 3.8267, 4.1927, 4.0490], + device='cuda:3'), covar=tensor([0.0520, 0.4527, 0.0420, 0.3773, 0.0965, 0.0903, 0.0511, 0.0567], + device='cuda:3'), in_proj_covar=tensor([0.0625, 0.0636, 0.0685, 0.0618, 0.0700, 0.0604, 0.0603, 0.0669], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 03:48:28,294 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.505e+02 2.494e+02 3.128e+02 3.771e+02 6.972e+02, threshold=6.255e+02, percent-clipped=4.0 +2023-02-07 03:48:28,961 INFO [train.py:901] (3/4) Epoch 22, batch 3450, loss[loss=0.2302, simple_loss=0.3101, pruned_loss=0.07516, over 8187.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2875, pruned_loss=0.06236, over 1612225.76 frames. ], batch size: 23, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:48:39,550 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=173207.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:48:42,430 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173211.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:49:05,730 INFO [train.py:901] (3/4) Epoch 22, batch 3500, loss[loss=0.2278, simple_loss=0.294, pruned_loss=0.0808, over 7795.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.288, pruned_loss=0.0619, over 1615885.92 frames. ], batch size: 19, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:49:24,789 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-07 03:49:38,931 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.470e+02 2.642e+02 3.082e+02 3.788e+02 9.506e+02, threshold=6.164e+02, percent-clipped=4.0 +2023-02-07 03:49:39,650 INFO [train.py:901] (3/4) Epoch 22, batch 3550, loss[loss=0.187, simple_loss=0.2635, pruned_loss=0.05523, over 7789.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2877, pruned_loss=0.0617, over 1610769.45 frames. ], batch size: 19, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:49:50,462 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6431, 4.6264, 4.1446, 2.1806, 4.0783, 4.0308, 4.2194, 4.0107], + device='cuda:3'), covar=tensor([0.0686, 0.0482, 0.1055, 0.4241, 0.0921, 0.1018, 0.1150, 0.0712], + device='cuda:3'), in_proj_covar=tensor([0.0523, 0.0434, 0.0432, 0.0536, 0.0427, 0.0447, 0.0426, 0.0386], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 03:50:00,033 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173322.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:50:03,372 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173327.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:50:06,898 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173331.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:50:10,302 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8483, 1.6707, 1.9815, 1.6103, 1.3421, 1.6810, 2.3483, 1.9483], + device='cuda:3'), covar=tensor([0.0477, 0.1236, 0.1607, 0.1423, 0.0571, 0.1424, 0.0625, 0.0640], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0189, 0.0158, 0.0099, 0.0163, 0.0111, 0.0143], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 03:50:14,773 INFO [train.py:901] (3/4) Epoch 22, batch 3600, loss[loss=0.2201, simple_loss=0.3104, pruned_loss=0.06491, over 8517.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2877, pruned_loss=0.06192, over 1609835.17 frames. ], batch size: 28, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:50:24,994 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173356.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:50:26,362 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173358.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:50:43,490 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173383.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:50:48,591 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.625e+02 2.398e+02 3.034e+02 4.459e+02 8.281e+02, threshold=6.068e+02, percent-clipped=7.0 +2023-02-07 03:50:49,310 INFO [train.py:901] (3/4) Epoch 22, batch 3650, loss[loss=0.1933, simple_loss=0.266, pruned_loss=0.06032, over 6861.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.289, pruned_loss=0.06289, over 1612675.91 frames. ], batch size: 15, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:51:23,355 INFO [train.py:901] (3/4) Epoch 22, batch 3700, loss[loss=0.177, simple_loss=0.2692, pruned_loss=0.04238, over 8595.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2868, pruned_loss=0.06166, over 1607664.39 frames. ], batch size: 34, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:51:24,744 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-07 03:51:42,310 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173467.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:51:51,708 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 +2023-02-07 03:51:57,922 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.704e+02 2.519e+02 2.931e+02 3.909e+02 7.363e+02, threshold=5.861e+02, percent-clipped=2.0 +2023-02-07 03:51:58,522 INFO [train.py:901] (3/4) Epoch 22, batch 3750, loss[loss=0.1933, simple_loss=0.2669, pruned_loss=0.05987, over 7511.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2876, pruned_loss=0.06179, over 1611239.57 frames. ], batch size: 18, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:51:58,723 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173492.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:52:11,078 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173509.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 03:52:12,783 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=173512.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:52:18,145 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5054, 1.5620, 4.7165, 1.8994, 4.1906, 3.9105, 4.2916, 4.1480], + device='cuda:3'), covar=tensor([0.0556, 0.4484, 0.0473, 0.3633, 0.0990, 0.0915, 0.0560, 0.0583], + device='cuda:3'), in_proj_covar=tensor([0.0628, 0.0640, 0.0689, 0.0621, 0.0702, 0.0607, 0.0605, 0.0674], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 03:52:22,066 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2405, 1.3183, 1.6391, 1.2736, 0.7019, 1.4088, 1.2238, 1.0522], + device='cuda:3'), covar=tensor([0.0535, 0.1197, 0.1561, 0.1385, 0.0531, 0.1448, 0.0654, 0.0686], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0151, 0.0188, 0.0157, 0.0099, 0.0162, 0.0111, 0.0142], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 03:52:23,509 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-07 03:52:32,208 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0920, 1.4729, 1.7602, 1.4275, 0.9137, 1.5233, 1.8301, 1.6102], + device='cuda:3'), covar=tensor([0.0508, 0.1292, 0.1656, 0.1430, 0.0601, 0.1483, 0.0655, 0.0650], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0152, 0.0189, 0.0157, 0.0099, 0.0162, 0.0111, 0.0142], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 03:52:32,730 INFO [train.py:901] (3/4) Epoch 22, batch 3800, loss[loss=0.1698, simple_loss=0.2567, pruned_loss=0.04148, over 8093.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.288, pruned_loss=0.06172, over 1610556.59 frames. ], batch size: 21, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:52:35,956 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.98 vs. limit=5.0 +2023-02-07 03:52:41,041 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0921, 1.0143, 1.1949, 0.9358, 0.9093, 1.2178, 0.0721, 0.8570], + device='cuda:3'), covar=tensor([0.1490, 0.1330, 0.0509, 0.0861, 0.2467, 0.0548, 0.2233, 0.1294], + device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0198, 0.0127, 0.0220, 0.0267, 0.0135, 0.0170, 0.0192], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 03:52:58,747 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173578.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:53:07,973 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.750e+02 2.467e+02 3.140e+02 3.842e+02 8.904e+02, threshold=6.281e+02, percent-clipped=2.0 +2023-02-07 03:53:08,696 INFO [train.py:901] (3/4) Epoch 22, batch 3850, loss[loss=0.1878, simple_loss=0.2788, pruned_loss=0.04846, over 8114.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2876, pruned_loss=0.06139, over 1613228.76 frames. ], batch size: 23, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:53:16,530 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173603.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:53:30,772 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-07 03:53:33,550 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173627.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:53:36,894 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173632.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:53:43,568 INFO [train.py:901] (3/4) Epoch 22, batch 3900, loss[loss=0.1873, simple_loss=0.2825, pruned_loss=0.04601, over 8094.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.288, pruned_loss=0.0616, over 1614986.04 frames. ], batch size: 23, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:54:02,782 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7865, 1.7292, 2.4458, 1.4401, 1.2670, 2.4007, 0.4499, 1.3951], + device='cuda:3'), covar=tensor([0.2080, 0.1255, 0.0382, 0.1505, 0.2967, 0.0391, 0.2337, 0.1512], + device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0198, 0.0128, 0.0221, 0.0269, 0.0136, 0.0171, 0.0193], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 03:54:03,324 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=173671.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:54:17,229 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.938e+02 2.507e+02 2.945e+02 3.654e+02 8.206e+02, threshold=5.890e+02, percent-clipped=3.0 +2023-02-07 03:54:17,888 INFO [train.py:901] (3/4) Epoch 22, batch 3950, loss[loss=0.2131, simple_loss=0.293, pruned_loss=0.06663, over 8030.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2867, pruned_loss=0.06141, over 1614328.83 frames. ], batch size: 22, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:54:53,286 INFO [train.py:901] (3/4) Epoch 22, batch 4000, loss[loss=0.1952, simple_loss=0.2835, pruned_loss=0.05348, over 8114.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2873, pruned_loss=0.06144, over 1614441.93 frames. ], batch size: 23, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:55:23,150 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173786.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:55:26,124 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.569e+02 2.329e+02 2.821e+02 3.599e+02 1.045e+03, threshold=5.642e+02, percent-clipped=6.0 +2023-02-07 03:55:26,793 INFO [train.py:901] (3/4) Epoch 22, batch 4050, loss[loss=0.2227, simple_loss=0.3027, pruned_loss=0.07139, over 8322.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2871, pruned_loss=0.06108, over 1613100.93 frames. ], batch size: 26, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:56:02,761 INFO [train.py:901] (3/4) Epoch 22, batch 4100, loss[loss=0.22, simple_loss=0.3075, pruned_loss=0.06628, over 8203.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2858, pruned_loss=0.06101, over 1609116.37 frames. ], batch size: 23, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:56:10,366 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=173853.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 03:56:31,252 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173883.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:56:36,362 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.656e+02 2.329e+02 2.764e+02 3.605e+02 7.317e+02, threshold=5.528e+02, percent-clipped=2.0 +2023-02-07 03:56:37,019 INFO [train.py:901] (3/4) Epoch 22, batch 4150, loss[loss=0.1564, simple_loss=0.2343, pruned_loss=0.03923, over 7420.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2858, pruned_loss=0.0611, over 1606909.92 frames. ], batch size: 17, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:56:45,820 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6209, 1.6527, 2.4174, 1.3777, 1.1182, 2.3485, 0.4211, 1.4023], + device='cuda:3'), covar=tensor([0.1770, 0.1053, 0.0271, 0.1586, 0.2720, 0.0335, 0.2105, 0.1293], + device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0196, 0.0127, 0.0220, 0.0267, 0.0135, 0.0171, 0.0191], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 03:56:47,601 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173908.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:57:12,185 INFO [train.py:901] (3/4) Epoch 22, batch 4200, loss[loss=0.1649, simple_loss=0.2457, pruned_loss=0.04202, over 7414.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2863, pruned_loss=0.06102, over 1611749.32 frames. ], batch size: 17, lr: 3.44e-03, grad_scale: 8.0 +2023-02-07 03:57:28,914 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-07 03:57:29,777 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173968.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 03:57:35,170 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=173976.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:57:46,762 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.784e+02 2.391e+02 3.056e+02 3.931e+02 9.713e+02, threshold=6.111e+02, percent-clipped=5.0 +2023-02-07 03:57:46,783 INFO [train.py:901] (3/4) Epoch 22, batch 4250, loss[loss=0.2062, simple_loss=0.2885, pruned_loss=0.06197, over 8590.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2866, pruned_loss=0.06137, over 1611954.26 frames. ], batch size: 34, lr: 3.44e-03, grad_scale: 4.0 +2023-02-07 03:57:55,806 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-07 03:58:15,721 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174033.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:58:17,063 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174035.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:58:17,880 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-07 03:58:21,636 INFO [train.py:901] (3/4) Epoch 22, batch 4300, loss[loss=0.222, simple_loss=0.3152, pruned_loss=0.06438, over 8501.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2863, pruned_loss=0.06096, over 1612992.98 frames. ], batch size: 28, lr: 3.44e-03, grad_scale: 4.0 +2023-02-07 03:58:21,849 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174042.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:58:25,741 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9363, 1.3705, 1.6095, 1.3357, 0.8370, 1.4241, 1.7579, 1.4498], + device='cuda:3'), covar=tensor([0.0533, 0.1260, 0.1667, 0.1441, 0.0621, 0.1459, 0.0684, 0.0690], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0189, 0.0158, 0.0099, 0.0163, 0.0111, 0.0143], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 03:58:40,410 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174067.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:58:56,914 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174091.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 03:58:57,380 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.513e+02 2.427e+02 2.775e+02 3.458e+02 5.995e+02, threshold=5.550e+02, percent-clipped=0.0 +2023-02-07 03:58:57,401 INFO [train.py:901] (3/4) Epoch 22, batch 4350, loss[loss=0.2108, simple_loss=0.2991, pruned_loss=0.06127, over 8099.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2866, pruned_loss=0.06126, over 1611612.81 frames. ], batch size: 23, lr: 3.43e-03, grad_scale: 4.0 +2023-02-07 03:59:25,036 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-07 03:59:32,547 INFO [train.py:901] (3/4) Epoch 22, batch 4400, loss[loss=0.2206, simple_loss=0.3016, pruned_loss=0.06977, over 8438.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2872, pruned_loss=0.06187, over 1608976.37 frames. ], batch size: 49, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 03:59:42,831 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174157.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:00:06,456 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-07 04:00:07,758 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.716e+02 2.623e+02 3.065e+02 3.902e+02 1.119e+03, threshold=6.129e+02, percent-clipped=5.0 +2023-02-07 04:00:07,777 INFO [train.py:901] (3/4) Epoch 22, batch 4450, loss[loss=0.2424, simple_loss=0.3242, pruned_loss=0.08031, over 8590.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2884, pruned_loss=0.06229, over 1612549.34 frames. ], batch size: 49, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:00:30,152 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174224.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 04:00:41,754 INFO [train.py:901] (3/4) Epoch 22, batch 4500, loss[loss=0.263, simple_loss=0.3216, pruned_loss=0.1022, over 8646.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2885, pruned_loss=0.06307, over 1610581.93 frames. ], batch size: 39, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:00:46,574 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174249.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 04:00:52,509 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4210, 2.8131, 2.2737, 3.7890, 1.9254, 1.9821, 2.6038, 2.9455], + device='cuda:3'), covar=tensor([0.0692, 0.0762, 0.0796, 0.0334, 0.0998, 0.1225, 0.0853, 0.0786], + device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0199, 0.0246, 0.0217, 0.0209, 0.0248, 0.0251, 0.0210], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 04:00:56,994 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-07 04:00:58,613 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-07 04:00:58,684 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 +2023-02-07 04:01:03,189 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1907, 1.0428, 1.2902, 0.9929, 0.9666, 1.3259, 0.0783, 0.9708], + device='cuda:3'), covar=tensor([0.1543, 0.1422, 0.0479, 0.0907, 0.2632, 0.0525, 0.2123, 0.1276], + device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0197, 0.0128, 0.0220, 0.0268, 0.0136, 0.0170, 0.0192], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 04:01:17,050 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.456e+02 2.524e+02 3.306e+02 4.354e+02 7.569e+02, threshold=6.612e+02, percent-clipped=6.0 +2023-02-07 04:01:17,071 INFO [train.py:901] (3/4) Epoch 22, batch 4550, loss[loss=0.2222, simple_loss=0.2889, pruned_loss=0.07779, over 7690.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.2905, pruned_loss=0.06402, over 1612247.31 frames. ], batch size: 18, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:01:23,308 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174301.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:01:28,575 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174309.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:01:51,056 INFO [train.py:901] (3/4) Epoch 22, batch 4600, loss[loss=0.2232, simple_loss=0.3059, pruned_loss=0.07026, over 8465.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.2898, pruned_loss=0.06352, over 1611664.00 frames. ], batch size: 25, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:01:53,058 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-07 04:01:54,748 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174347.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:02:12,026 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174372.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:02:15,442 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=174377.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:02:16,890 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=174379.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:02:25,982 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.714e+02 2.375e+02 2.973e+02 3.873e+02 1.031e+03, threshold=5.946e+02, percent-clipped=3.0 +2023-02-07 04:02:26,002 INFO [train.py:901] (3/4) Epoch 22, batch 4650, loss[loss=0.2157, simple_loss=0.3037, pruned_loss=0.06387, over 8476.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2893, pruned_loss=0.06304, over 1607222.17 frames. ], batch size: 25, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:02:34,565 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4056, 1.5959, 2.1269, 1.3112, 1.5282, 1.6107, 1.4700, 1.4714], + device='cuda:3'), covar=tensor([0.2043, 0.2808, 0.1123, 0.4675, 0.2025, 0.3608, 0.2441, 0.2261], + device='cuda:3'), in_proj_covar=tensor([0.0526, 0.0604, 0.0557, 0.0642, 0.0645, 0.0590, 0.0534, 0.0629], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 04:02:37,944 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174406.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:03:02,511 INFO [train.py:901] (3/4) Epoch 22, batch 4700, loss[loss=0.1922, simple_loss=0.2668, pruned_loss=0.0588, over 7782.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2885, pruned_loss=0.06228, over 1606623.82 frames. ], batch size: 19, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:03:37,053 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.532e+02 2.374e+02 2.923e+02 3.899e+02 9.329e+02, threshold=5.846e+02, percent-clipped=2.0 +2023-02-07 04:03:37,074 INFO [train.py:901] (3/4) Epoch 22, batch 4750, loss[loss=0.2283, simple_loss=0.3134, pruned_loss=0.07158, over 8466.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2881, pruned_loss=0.06183, over 1612208.01 frames. ], batch size: 25, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:03:37,264 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174492.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:03:38,611 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174494.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:03:43,328 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=174501.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:04:04,462 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-07 04:04:06,509 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-07 04:04:12,377 INFO [train.py:901] (3/4) Epoch 22, batch 4800, loss[loss=0.2093, simple_loss=0.288, pruned_loss=0.06527, over 8018.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.287, pruned_loss=0.06133, over 1609067.74 frames. ], batch size: 22, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:04:21,847 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 +2023-02-07 04:04:46,102 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.675e+02 2.398e+02 2.995e+02 3.860e+02 8.125e+02, threshold=5.990e+02, percent-clipped=3.0 +2023-02-07 04:04:46,123 INFO [train.py:901] (3/4) Epoch 22, batch 4850, loss[loss=0.252, simple_loss=0.3197, pruned_loss=0.09211, over 6827.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2869, pruned_loss=0.06178, over 1603864.28 frames. ], batch size: 71, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:04:55,425 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-07 04:05:02,314 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174616.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:05:20,319 INFO [train.py:901] (3/4) Epoch 22, batch 4900, loss[loss=0.1753, simple_loss=0.2555, pruned_loss=0.04757, over 8256.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2867, pruned_loss=0.06168, over 1602642.20 frames. ], batch size: 22, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:05:23,130 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=174645.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:05:29,306 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=174653.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:05:50,583 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1145, 1.8760, 2.3938, 2.0375, 2.3121, 2.0518, 1.9565, 1.5840], + device='cuda:3'), covar=tensor([0.4105, 0.4066, 0.1608, 0.2972, 0.2100, 0.2563, 0.1530, 0.4040], + device='cuda:3'), in_proj_covar=tensor([0.0936, 0.0977, 0.0802, 0.0942, 0.0993, 0.0891, 0.0746, 0.0823], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 04:05:56,489 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.619e+02 2.582e+02 3.121e+02 3.821e+02 7.682e+02, threshold=6.242e+02, percent-clipped=2.0 +2023-02-07 04:05:56,510 INFO [train.py:901] (3/4) Epoch 22, batch 4950, loss[loss=0.2257, simple_loss=0.3038, pruned_loss=0.0738, over 7812.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2878, pruned_loss=0.06194, over 1606093.79 frames. ], batch size: 20, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:06:17,791 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174723.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:06:30,646 INFO [train.py:901] (3/4) Epoch 22, batch 5000, loss[loss=0.2043, simple_loss=0.2844, pruned_loss=0.06209, over 8561.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2875, pruned_loss=0.06226, over 1608016.60 frames. ], batch size: 31, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:06:35,014 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174748.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:06:36,251 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=174750.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:06:36,400 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174750.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:06:44,698 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174760.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:06:51,019 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174768.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:06:54,585 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174773.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:06:55,972 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174775.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:07:07,673 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.731e+02 2.477e+02 2.928e+02 3.612e+02 7.754e+02, threshold=5.856e+02, percent-clipped=3.0 +2023-02-07 04:07:07,694 INFO [train.py:901] (3/4) Epoch 22, batch 5050, loss[loss=0.2195, simple_loss=0.2955, pruned_loss=0.07171, over 8199.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.287, pruned_loss=0.06224, over 1604803.22 frames. ], batch size: 23, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:07:17,140 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-07 04:07:30,596 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6605, 2.0383, 3.2507, 1.4223, 2.3745, 1.9276, 1.7818, 2.2862], + device='cuda:3'), covar=tensor([0.1842, 0.2564, 0.0803, 0.4480, 0.1943, 0.3399, 0.2197, 0.2487], + device='cuda:3'), in_proj_covar=tensor([0.0528, 0.0605, 0.0556, 0.0643, 0.0646, 0.0591, 0.0536, 0.0632], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 04:07:34,490 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-07 04:07:41,819 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-07 04:07:42,662 INFO [train.py:901] (3/4) Epoch 22, batch 5100, loss[loss=0.2279, simple_loss=0.2926, pruned_loss=0.08167, over 8032.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2874, pruned_loss=0.0624, over 1608385.31 frames. ], batch size: 22, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:07:52,852 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8038, 1.5677, 1.7191, 1.4907, 0.9152, 1.5200, 1.6233, 1.5659], + device='cuda:3'), covar=tensor([0.0521, 0.1201, 0.1556, 0.1322, 0.0599, 0.1402, 0.0694, 0.0661], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0153, 0.0191, 0.0160, 0.0100, 0.0164, 0.0113, 0.0144], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 04:07:58,154 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174865.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:08:03,642 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174872.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:08:15,021 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 +2023-02-07 04:08:17,819 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.816e+02 2.498e+02 3.130e+02 3.757e+02 7.363e+02, threshold=6.259e+02, percent-clipped=3.0 +2023-02-07 04:08:17,839 INFO [train.py:901] (3/4) Epoch 22, batch 5150, loss[loss=0.187, simple_loss=0.2723, pruned_loss=0.05083, over 8239.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2878, pruned_loss=0.06273, over 1609076.40 frames. ], batch size: 22, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:08:20,514 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1690, 1.0793, 1.2917, 1.0251, 0.9527, 1.3181, 0.0509, 0.8618], + device='cuda:3'), covar=tensor([0.1978, 0.1401, 0.0514, 0.0892, 0.2804, 0.0552, 0.2152, 0.1384], + device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0197, 0.0128, 0.0220, 0.0268, 0.0136, 0.0170, 0.0194], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 04:08:21,151 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174897.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:08:41,298 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1983, 1.0853, 1.3280, 1.0475, 1.0233, 1.3230, 0.0629, 0.8866], + device='cuda:3'), covar=tensor([0.1425, 0.1441, 0.0453, 0.0752, 0.2515, 0.0527, 0.1937, 0.1269], + device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0197, 0.0128, 0.0219, 0.0268, 0.0136, 0.0169, 0.0193], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 04:08:52,066 INFO [train.py:901] (3/4) Epoch 22, batch 5200, loss[loss=0.2347, simple_loss=0.328, pruned_loss=0.07069, over 8345.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2873, pruned_loss=0.06224, over 1611394.80 frames. ], batch size: 26, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:09:26,430 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2194, 3.0928, 2.8304, 1.5781, 2.8262, 2.9383, 2.7986, 2.8076], + device='cuda:3'), covar=tensor([0.1173, 0.0808, 0.1426, 0.4536, 0.1143, 0.1194, 0.1559, 0.1097], + device='cuda:3'), in_proj_covar=tensor([0.0523, 0.0433, 0.0432, 0.0535, 0.0425, 0.0444, 0.0424, 0.0387], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 04:09:26,992 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.782e+02 2.432e+02 2.854e+02 3.739e+02 7.258e+02, threshold=5.708e+02, percent-clipped=1.0 +2023-02-07 04:09:27,013 INFO [train.py:901] (3/4) Epoch 22, batch 5250, loss[loss=0.1623, simple_loss=0.2499, pruned_loss=0.03738, over 7242.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2885, pruned_loss=0.0624, over 1609738.89 frames. ], batch size: 16, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:09:31,810 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-07 04:09:44,064 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175016.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:09:49,561 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175024.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:10:01,590 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175041.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:10:02,001 INFO [train.py:901] (3/4) Epoch 22, batch 5300, loss[loss=0.2801, simple_loss=0.3438, pruned_loss=0.1082, over 8139.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2878, pruned_loss=0.06233, over 1609470.06 frames. ], batch size: 22, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:10:07,101 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175049.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:10:19,116 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=175067.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:10:35,758 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.567e+02 2.519e+02 3.149e+02 3.909e+02 1.075e+03, threshold=6.297e+02, percent-clipped=6.0 +2023-02-07 04:10:35,778 INFO [train.py:901] (3/4) Epoch 22, batch 5350, loss[loss=0.1785, simple_loss=0.259, pruned_loss=0.04899, over 7811.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2875, pruned_loss=0.06147, over 1610359.78 frames. ], batch size: 20, lr: 3.43e-03, grad_scale: 8.0 +2023-02-07 04:10:57,903 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175121.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:11:08,811 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9398, 1.7539, 3.1665, 1.3903, 2.4349, 3.4613, 3.6308, 2.6468], + device='cuda:3'), covar=tensor([0.1340, 0.1794, 0.0439, 0.2584, 0.1075, 0.0358, 0.0666, 0.0953], + device='cuda:3'), in_proj_covar=tensor([0.0293, 0.0321, 0.0285, 0.0313, 0.0308, 0.0263, 0.0418, 0.0304], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 04:11:11,826 INFO [train.py:901] (3/4) Epoch 22, batch 5400, loss[loss=0.2178, simple_loss=0.3052, pruned_loss=0.06522, over 8141.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2881, pruned_loss=0.06183, over 1602576.61 frames. ], batch size: 22, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:11:14,643 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175146.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:11:22,726 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175157.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:11:39,707 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175182.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:11:46,276 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.680e+02 2.426e+02 2.833e+02 4.034e+02 1.686e+03, threshold=5.665e+02, percent-clipped=5.0 +2023-02-07 04:11:46,296 INFO [train.py:901] (3/4) Epoch 22, batch 5450, loss[loss=0.2216, simple_loss=0.3005, pruned_loss=0.07135, over 8502.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2873, pruned_loss=0.06181, over 1607190.19 frames. ], batch size: 26, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:11:52,130 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4512, 2.3507, 3.1439, 2.4694, 2.9878, 2.4873, 2.2845, 1.7961], + device='cuda:3'), covar=tensor([0.5247, 0.4828, 0.1986, 0.3790, 0.2807, 0.2786, 0.1796, 0.5478], + device='cuda:3'), in_proj_covar=tensor([0.0937, 0.0977, 0.0801, 0.0943, 0.0991, 0.0888, 0.0745, 0.0822], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 04:12:02,450 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7477, 1.6605, 2.4890, 2.0177, 2.1925, 1.7734, 1.5220, 1.0403], + device='cuda:3'), covar=tensor([0.7060, 0.5768, 0.1975, 0.3873, 0.3146, 0.4303, 0.2896, 0.5578], + device='cuda:3'), in_proj_covar=tensor([0.0937, 0.0977, 0.0801, 0.0944, 0.0992, 0.0889, 0.0746, 0.0822], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 04:12:04,558 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-07 04:12:09,775 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175225.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:12:20,269 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-07 04:12:22,329 INFO [train.py:901] (3/4) Epoch 22, batch 5500, loss[loss=0.1883, simple_loss=0.2823, pruned_loss=0.04714, over 8506.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2876, pruned_loss=0.06149, over 1613341.71 frames. ], batch size: 28, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:12:56,614 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.512e+02 2.376e+02 2.848e+02 3.508e+02 8.289e+02, threshold=5.697e+02, percent-clipped=6.0 +2023-02-07 04:12:56,635 INFO [train.py:901] (3/4) Epoch 22, batch 5550, loss[loss=0.2287, simple_loss=0.3058, pruned_loss=0.07582, over 8462.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2886, pruned_loss=0.06178, over 1617540.41 frames. ], batch size: 25, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:13:31,796 INFO [train.py:901] (3/4) Epoch 22, batch 5600, loss[loss=0.1612, simple_loss=0.2424, pruned_loss=0.04002, over 7281.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2881, pruned_loss=0.06149, over 1614216.72 frames. ], batch size: 16, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:14:03,823 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175388.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:14:06,408 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.734e+02 2.532e+02 3.141e+02 4.135e+02 1.836e+03, threshold=6.283e+02, percent-clipped=10.0 +2023-02-07 04:14:06,429 INFO [train.py:901] (3/4) Epoch 22, batch 5650, loss[loss=0.2137, simple_loss=0.3065, pruned_loss=0.06043, over 8495.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2883, pruned_loss=0.0617, over 1614132.89 frames. ], batch size: 28, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:14:20,103 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175412.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:14:22,663 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-07 04:14:23,537 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5216, 1.4471, 1.8216, 1.2275, 1.1766, 1.7797, 0.1659, 1.1349], + device='cuda:3'), covar=tensor([0.1557, 0.1218, 0.0399, 0.0980, 0.2549, 0.0486, 0.2042, 0.1368], + device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0196, 0.0128, 0.0219, 0.0265, 0.0136, 0.0167, 0.0193], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 04:14:37,562 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175438.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:14:40,743 INFO [train.py:901] (3/4) Epoch 22, batch 5700, loss[loss=0.2107, simple_loss=0.2862, pruned_loss=0.06762, over 8140.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.289, pruned_loss=0.06209, over 1616326.69 frames. ], batch size: 22, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:14:56,165 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175463.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:14:59,072 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-07 04:15:00,499 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-07 04:15:15,518 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 2.452e+02 2.878e+02 3.661e+02 5.836e+02, threshold=5.755e+02, percent-clipped=0.0 +2023-02-07 04:15:15,539 INFO [train.py:901] (3/4) Epoch 22, batch 5750, loss[loss=0.1979, simple_loss=0.2827, pruned_loss=0.05652, over 7807.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2881, pruned_loss=0.06164, over 1616672.67 frames. ], batch size: 20, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:15:20,437 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175499.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:15:22,434 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=175501.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:15:27,207 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-07 04:15:33,338 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8835, 2.0550, 1.7579, 2.4809, 1.3674, 1.6010, 1.9864, 2.0917], + device='cuda:3'), covar=tensor([0.0749, 0.0741, 0.0890, 0.0488, 0.0998, 0.1212, 0.0694, 0.0760], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0197, 0.0246, 0.0215, 0.0207, 0.0247, 0.0251, 0.0208], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 04:15:36,028 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.0708, 3.9656, 3.6404, 2.0417, 3.5276, 3.6880, 3.6524, 3.4998], + device='cuda:3'), covar=tensor([0.0801, 0.0663, 0.1081, 0.4561, 0.0990, 0.1075, 0.1252, 0.0892], + device='cuda:3'), in_proj_covar=tensor([0.0524, 0.0437, 0.0433, 0.0539, 0.0427, 0.0445, 0.0425, 0.0388], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 04:15:50,255 INFO [train.py:901] (3/4) Epoch 22, batch 5800, loss[loss=0.2171, simple_loss=0.3083, pruned_loss=0.06295, over 8513.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.288, pruned_loss=0.06176, over 1618016.78 frames. ], batch size: 28, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:16:09,024 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=175569.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:16:25,902 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.268e+02 2.280e+02 2.739e+02 3.457e+02 6.413e+02, threshold=5.479e+02, percent-clipped=3.0 +2023-02-07 04:16:25,923 INFO [train.py:901] (3/4) Epoch 22, batch 5850, loss[loss=0.2446, simple_loss=0.3378, pruned_loss=0.07574, over 8615.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2879, pruned_loss=0.06112, over 1622957.00 frames. ], batch size: 39, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:16:42,237 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175616.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:17:00,357 INFO [train.py:901] (3/4) Epoch 22, batch 5900, loss[loss=0.2074, simple_loss=0.283, pruned_loss=0.06596, over 7928.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2886, pruned_loss=0.06173, over 1621766.08 frames. ], batch size: 20, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:17:12,165 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8002, 2.1839, 3.4400, 1.7517, 1.7762, 3.4661, 0.7301, 2.0025], + device='cuda:3'), covar=tensor([0.1547, 0.1361, 0.0279, 0.2011, 0.2785, 0.0317, 0.2381, 0.1643], + device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0195, 0.0128, 0.0219, 0.0264, 0.0136, 0.0167, 0.0192], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 04:17:29,301 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175684.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:17:35,254 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.666e+02 2.520e+02 3.043e+02 3.699e+02 9.671e+02, threshold=6.086e+02, percent-clipped=7.0 +2023-02-07 04:17:35,275 INFO [train.py:901] (3/4) Epoch 22, batch 5950, loss[loss=0.1526, simple_loss=0.2337, pruned_loss=0.03575, over 7526.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2885, pruned_loss=0.06186, over 1619003.06 frames. ], batch size: 18, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:17:45,508 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5019, 1.3378, 2.3898, 1.2324, 2.2262, 2.5660, 2.7056, 2.1914], + device='cuda:3'), covar=tensor([0.1040, 0.1362, 0.0401, 0.2097, 0.0699, 0.0357, 0.0585, 0.0690], + device='cuda:3'), in_proj_covar=tensor([0.0294, 0.0321, 0.0285, 0.0313, 0.0309, 0.0263, 0.0419, 0.0305], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 04:18:00,065 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175728.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:18:02,750 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=175732.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:18:08,432 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4415, 1.3784, 1.7511, 1.1841, 1.1201, 1.7255, 0.2186, 1.1135], + device='cuda:3'), covar=tensor([0.1883, 0.1264, 0.0414, 0.0918, 0.2502, 0.0526, 0.2040, 0.1329], + device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0195, 0.0128, 0.0219, 0.0264, 0.0136, 0.0167, 0.0191], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 04:18:09,571 INFO [train.py:901] (3/4) Epoch 22, batch 6000, loss[loss=0.2158, simple_loss=0.3007, pruned_loss=0.06545, over 8133.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2869, pruned_loss=0.06121, over 1607740.87 frames. ], batch size: 22, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:18:09,571 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 04:18:14,236 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.7187, 1.5688, 3.9134, 1.5948, 3.4397, 3.2110, 3.5029, 3.3989], + device='cuda:3'), covar=tensor([0.0724, 0.4690, 0.0555, 0.4512, 0.1294, 0.1104, 0.0713, 0.0865], + device='cuda:3'), in_proj_covar=tensor([0.0629, 0.0644, 0.0697, 0.0626, 0.0709, 0.0606, 0.0608, 0.0678], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 04:18:21,633 INFO [train.py:935] (3/4) Epoch 22, validation: loss=0.1729, simple_loss=0.2732, pruned_loss=0.03632, over 944034.00 frames. +2023-02-07 04:18:21,634 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6747MB +2023-02-07 04:18:31,485 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=175756.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:18:56,217 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.465e+02 2.478e+02 2.934e+02 3.623e+02 7.032e+02, threshold=5.869e+02, percent-clipped=2.0 +2023-02-07 04:18:56,237 INFO [train.py:901] (3/4) Epoch 22, batch 6050, loss[loss=0.2408, simple_loss=0.3287, pruned_loss=0.0765, over 8583.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2862, pruned_loss=0.06126, over 1602312.15 frames. ], batch size: 31, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:19:18,113 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3617, 2.1588, 2.7533, 2.2827, 2.7420, 2.3907, 2.1675, 1.4991], + device='cuda:3'), covar=tensor([0.5201, 0.4721, 0.1968, 0.3692, 0.2538, 0.2982, 0.1901, 0.5324], + device='cuda:3'), in_proj_covar=tensor([0.0943, 0.0981, 0.0808, 0.0947, 0.0999, 0.0898, 0.0750, 0.0828], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 04:19:31,869 INFO [train.py:901] (3/4) Epoch 22, batch 6100, loss[loss=0.1758, simple_loss=0.2686, pruned_loss=0.04154, over 8329.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.287, pruned_loss=0.06162, over 1610070.92 frames. ], batch size: 25, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:19:32,676 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=175843.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:19:35,614 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175847.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:19:51,984 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175871.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:19:52,703 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175872.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:19:56,546 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-07 04:20:07,208 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.604e+02 2.365e+02 2.974e+02 3.880e+02 6.577e+02, threshold=5.949e+02, percent-clipped=2.0 +2023-02-07 04:20:07,229 INFO [train.py:901] (3/4) Epoch 22, batch 6150, loss[loss=0.2095, simple_loss=0.2837, pruned_loss=0.06764, over 8585.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2859, pruned_loss=0.06107, over 1611044.59 frames. ], batch size: 39, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:20:10,659 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175897.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:20:11,996 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175899.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:20:40,378 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175940.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:20:41,486 INFO [train.py:901] (3/4) Epoch 22, batch 6200, loss[loss=0.247, simple_loss=0.3228, pruned_loss=0.0856, over 8512.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2858, pruned_loss=0.06121, over 1609831.21 frames. ], batch size: 26, lr: 3.42e-03, grad_scale: 8.0 +2023-02-07 04:20:53,607 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175958.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:20:56,487 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.66 vs. limit=2.0 +2023-02-07 04:20:56,918 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175963.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:20:58,221 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175965.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:21:15,649 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.525e+02 2.329e+02 2.882e+02 3.634e+02 1.217e+03, threshold=5.765e+02, percent-clipped=6.0 +2023-02-07 04:21:15,670 INFO [train.py:901] (3/4) Epoch 22, batch 6250, loss[loss=0.1975, simple_loss=0.2878, pruned_loss=0.05355, over 8441.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2864, pruned_loss=0.06173, over 1610315.13 frames. ], batch size: 27, lr: 3.42e-03, grad_scale: 16.0 +2023-02-07 04:21:51,349 INFO [train.py:901] (3/4) Epoch 22, batch 6300, loss[loss=0.1889, simple_loss=0.2669, pruned_loss=0.05544, over 7214.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.285, pruned_loss=0.06099, over 1610783.30 frames. ], batch size: 16, lr: 3.42e-03, grad_scale: 16.0 +2023-02-07 04:22:12,731 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=176072.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:22:20,800 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5398, 4.5156, 4.1030, 2.3634, 4.0397, 4.1127, 4.0921, 3.9813], + device='cuda:3'), covar=tensor([0.0614, 0.0524, 0.0975, 0.3840, 0.0749, 0.0892, 0.1189, 0.0753], + device='cuda:3'), in_proj_covar=tensor([0.0531, 0.0437, 0.0435, 0.0542, 0.0427, 0.0446, 0.0427, 0.0388], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 04:22:26,687 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.688e+02 2.297e+02 2.795e+02 3.577e+02 6.374e+02, threshold=5.590e+02, percent-clipped=1.0 +2023-02-07 04:22:26,707 INFO [train.py:901] (3/4) Epoch 22, batch 6350, loss[loss=0.226, simple_loss=0.3151, pruned_loss=0.06845, over 8327.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2854, pruned_loss=0.06078, over 1613066.50 frames. ], batch size: 25, lr: 3.42e-03, grad_scale: 16.0 +2023-02-07 04:22:34,429 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176103.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:22:51,208 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176127.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:22:51,828 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176128.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:23:01,398 INFO [train.py:901] (3/4) Epoch 22, batch 6400, loss[loss=0.1817, simple_loss=0.2664, pruned_loss=0.04851, over 8087.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2855, pruned_loss=0.06034, over 1616605.57 frames. ], batch size: 21, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:23:08,435 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176152.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:23:33,464 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176187.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:23:36,651 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.668e+02 2.375e+02 2.869e+02 3.334e+02 7.002e+02, threshold=5.738e+02, percent-clipped=1.0 +2023-02-07 04:23:36,672 INFO [train.py:901] (3/4) Epoch 22, batch 6450, loss[loss=0.1789, simple_loss=0.2756, pruned_loss=0.04117, over 8472.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2875, pruned_loss=0.06139, over 1615673.92 frames. ], batch size: 25, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:23:52,613 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176214.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:24:09,921 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176239.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:24:11,685 INFO [train.py:901] (3/4) Epoch 22, batch 6500, loss[loss=0.2382, simple_loss=0.3126, pruned_loss=0.08188, over 7972.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2879, pruned_loss=0.06157, over 1617047.76 frames. ], batch size: 21, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:24:12,464 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=176243.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:24:22,288 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-07 04:24:23,884 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176260.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:24:34,086 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176275.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:24:45,630 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.704e+02 2.325e+02 2.725e+02 3.404e+02 5.159e+02, threshold=5.450e+02, percent-clipped=0.0 +2023-02-07 04:24:45,651 INFO [train.py:901] (3/4) Epoch 22, batch 6550, loss[loss=0.1945, simple_loss=0.2908, pruned_loss=0.0491, over 8100.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2889, pruned_loss=0.06231, over 1620426.94 frames. ], batch size: 23, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:24:57,260 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=176307.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:25:09,352 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-07 04:25:21,087 INFO [train.py:901] (3/4) Epoch 22, batch 6600, loss[loss=0.1685, simple_loss=0.2535, pruned_loss=0.04176, over 7920.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2887, pruned_loss=0.06214, over 1619297.42 frames. ], batch size: 20, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:25:29,283 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-07 04:25:30,400 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 +2023-02-07 04:25:32,747 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176358.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:25:55,380 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.481e+02 2.510e+02 3.110e+02 4.060e+02 7.968e+02, threshold=6.221e+02, percent-clipped=4.0 +2023-02-07 04:25:55,401 INFO [train.py:901] (3/4) Epoch 22, batch 6650, loss[loss=0.2199, simple_loss=0.2943, pruned_loss=0.07278, over 8083.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2885, pruned_loss=0.0624, over 1616513.79 frames. ], batch size: 21, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:26:17,170 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176422.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:26:31,229 INFO [train.py:901] (3/4) Epoch 22, batch 6700, loss[loss=0.2249, simple_loss=0.3126, pruned_loss=0.06861, over 8326.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2887, pruned_loss=0.06238, over 1619638.56 frames. ], batch size: 26, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:26:32,112 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176443.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:26:41,199 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2023-02-07 04:26:49,604 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176468.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:27:00,366 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176484.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:27:05,583 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.611e+02 2.672e+02 3.290e+02 4.002e+02 8.131e+02, threshold=6.579e+02, percent-clipped=6.0 +2023-02-07 04:27:05,603 INFO [train.py:901] (3/4) Epoch 22, batch 6750, loss[loss=0.1962, simple_loss=0.2634, pruned_loss=0.06453, over 7795.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2888, pruned_loss=0.06249, over 1621132.63 frames. ], batch size: 19, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:27:41,470 INFO [train.py:901] (3/4) Epoch 22, batch 6800, loss[loss=0.2024, simple_loss=0.2956, pruned_loss=0.05464, over 8470.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2893, pruned_loss=0.06295, over 1621734.80 frames. ], batch size: 25, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:27:44,982 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-07 04:28:16,786 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.434e+02 2.317e+02 3.026e+02 3.783e+02 8.757e+02, threshold=6.052e+02, percent-clipped=1.0 +2023-02-07 04:28:16,807 INFO [train.py:901] (3/4) Epoch 22, batch 6850, loss[loss=0.1829, simple_loss=0.2613, pruned_loss=0.05226, over 7690.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2879, pruned_loss=0.06231, over 1616772.77 frames. ], batch size: 18, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:28:24,732 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=176604.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:28:31,628 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176614.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:28:34,702 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-07 04:28:34,771 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=176619.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:28:48,420 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176639.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:28:50,268 INFO [train.py:901] (3/4) Epoch 22, batch 6900, loss[loss=0.2295, simple_loss=0.307, pruned_loss=0.07603, over 8202.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2869, pruned_loss=0.06171, over 1612699.92 frames. ], batch size: 23, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:28:51,038 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176643.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:29:03,491 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4498, 1.9328, 4.5930, 2.2573, 4.1446, 3.8751, 4.1985, 4.0874], + device='cuda:3'), covar=tensor([0.0544, 0.3956, 0.0542, 0.3413, 0.0989, 0.0900, 0.0543, 0.0575], + device='cuda:3'), in_proj_covar=tensor([0.0631, 0.0637, 0.0694, 0.0625, 0.0709, 0.0605, 0.0603, 0.0678], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 04:29:17,268 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176678.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:29:20,592 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176683.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:29:26,753 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.293e+02 2.439e+02 3.078e+02 3.806e+02 5.995e+02, threshold=6.157e+02, percent-clipped=0.0 +2023-02-07 04:29:26,773 INFO [train.py:901] (3/4) Epoch 22, batch 6950, loss[loss=0.2007, simple_loss=0.2721, pruned_loss=0.06461, over 7423.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2871, pruned_loss=0.06158, over 1614621.69 frames. ], batch size: 17, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:29:35,484 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176703.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:29:44,272 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-07 04:29:46,549 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176719.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:29:56,857 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176734.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:30:02,055 INFO [train.py:901] (3/4) Epoch 22, batch 7000, loss[loss=0.211, simple_loss=0.2974, pruned_loss=0.06229, over 8471.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2874, pruned_loss=0.06181, over 1609596.07 frames. ], batch size: 25, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:30:36,087 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4425, 1.6295, 2.1095, 1.3386, 1.5632, 1.6929, 1.4927, 1.4970], + device='cuda:3'), covar=tensor([0.1979, 0.2694, 0.1007, 0.4462, 0.1952, 0.3400, 0.2409, 0.2172], + device='cuda:3'), in_proj_covar=tensor([0.0527, 0.0606, 0.0558, 0.0646, 0.0650, 0.0595, 0.0539, 0.0634], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 04:30:37,819 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.601e+02 2.400e+02 2.923e+02 3.703e+02 8.900e+02, threshold=5.847e+02, percent-clipped=5.0 +2023-02-07 04:30:37,839 INFO [train.py:901] (3/4) Epoch 22, batch 7050, loss[loss=0.2068, simple_loss=0.2933, pruned_loss=0.06015, over 8245.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2867, pruned_loss=0.06158, over 1608771.43 frames. ], batch size: 22, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:31:02,298 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 +2023-02-07 04:31:03,287 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=176828.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:31:12,380 INFO [train.py:901] (3/4) Epoch 22, batch 7100, loss[loss=0.1967, simple_loss=0.287, pruned_loss=0.05317, over 8247.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2874, pruned_loss=0.06208, over 1609304.17 frames. ], batch size: 24, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:31:31,801 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176871.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:31:46,099 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.664e+02 2.316e+02 2.836e+02 3.633e+02 7.093e+02, threshold=5.673e+02, percent-clipped=3.0 +2023-02-07 04:31:46,120 INFO [train.py:901] (3/4) Epoch 22, batch 7150, loss[loss=0.2455, simple_loss=0.3164, pruned_loss=0.08731, over 8618.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2879, pruned_loss=0.06194, over 1614423.61 frames. ], batch size: 34, lr: 3.41e-03, grad_scale: 16.0 +2023-02-07 04:32:16,359 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8422, 2.0764, 2.1663, 1.3314, 2.3197, 1.5603, 0.7302, 2.0001], + device='cuda:3'), covar=tensor([0.0522, 0.0319, 0.0273, 0.0602, 0.0368, 0.0807, 0.0828, 0.0321], + device='cuda:3'), in_proj_covar=tensor([0.0455, 0.0392, 0.0346, 0.0447, 0.0381, 0.0536, 0.0392, 0.0422], + device='cuda:3'), out_proj_covar=tensor([1.2187e-04, 1.0283e-04, 9.0824e-05, 1.1753e-04, 1.0017e-04, 1.5111e-04, + 1.0546e-04, 1.1169e-04], device='cuda:3') +2023-02-07 04:32:22,258 INFO [train.py:901] (3/4) Epoch 22, batch 7200, loss[loss=0.2203, simple_loss=0.2965, pruned_loss=0.07207, over 7815.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.288, pruned_loss=0.06254, over 1608265.42 frames. ], batch size: 20, lr: 3.41e-03, grad_scale: 8.0 +2023-02-07 04:32:23,124 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176943.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:32:35,679 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8893, 2.1566, 2.2664, 1.4002, 2.3973, 1.6648, 0.7988, 1.9788], + device='cuda:3'), covar=tensor([0.0574, 0.0348, 0.0279, 0.0616, 0.0417, 0.0894, 0.0845, 0.0351], + device='cuda:3'), in_proj_covar=tensor([0.0455, 0.0392, 0.0346, 0.0447, 0.0380, 0.0536, 0.0392, 0.0422], + device='cuda:3'), out_proj_covar=tensor([1.2193e-04, 1.0279e-04, 9.0864e-05, 1.1763e-04, 1.0010e-04, 1.5109e-04, + 1.0549e-04, 1.1176e-04], device='cuda:3') +2023-02-07 04:32:44,917 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176975.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:32:52,859 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=176987.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:32:54,982 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176990.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:32:56,138 INFO [train.py:901] (3/4) Epoch 22, batch 7250, loss[loss=0.1627, simple_loss=0.2455, pruned_loss=0.03998, over 7544.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2885, pruned_loss=0.06253, over 1614138.07 frames. ], batch size: 18, lr: 3.41e-03, grad_scale: 8.0 +2023-02-07 04:32:56,794 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.387e+02 2.385e+02 2.852e+02 3.441e+02 7.839e+02, threshold=5.703e+02, percent-clipped=2.0 +2023-02-07 04:33:02,409 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177000.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:33:14,113 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177015.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:33:21,897 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=177027.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:33:22,934 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-07 04:33:28,004 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.9767, 6.0051, 5.1647, 2.7313, 5.2391, 5.6318, 5.3811, 5.5174], + device='cuda:3'), covar=tensor([0.0416, 0.0337, 0.0810, 0.4057, 0.0707, 0.0911, 0.0971, 0.0598], + device='cuda:3'), in_proj_covar=tensor([0.0528, 0.0438, 0.0432, 0.0541, 0.0427, 0.0448, 0.0428, 0.0388], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 04:33:31,983 INFO [train.py:901] (3/4) Epoch 22, batch 7300, loss[loss=0.1762, simple_loss=0.2653, pruned_loss=0.04353, over 8506.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2885, pruned_loss=0.06216, over 1617217.71 frames. ], batch size: 28, lr: 3.41e-03, grad_scale: 8.0 +2023-02-07 04:34:06,492 INFO [train.py:901] (3/4) Epoch 22, batch 7350, loss[loss=0.2419, simple_loss=0.3157, pruned_loss=0.084, over 8339.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2881, pruned_loss=0.06189, over 1620712.47 frames. ], batch size: 26, lr: 3.41e-03, grad_scale: 8.0 +2023-02-07 04:34:07,163 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.722e+02 2.532e+02 3.310e+02 4.342e+02 9.656e+02, threshold=6.621e+02, percent-clipped=7.0 +2023-02-07 04:34:13,493 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177102.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:34:26,059 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-07 04:34:31,739 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177127.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:34:33,086 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177129.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:34:36,854 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-07 04:34:42,503 INFO [train.py:901] (3/4) Epoch 22, batch 7400, loss[loss=0.1501, simple_loss=0.232, pruned_loss=0.03411, over 7788.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2881, pruned_loss=0.06179, over 1620258.94 frames. ], batch size: 19, lr: 3.41e-03, grad_scale: 8.0 +2023-02-07 04:34:42,675 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177142.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:34:48,000 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-07 04:34:59,914 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2992, 2.0291, 2.7801, 2.2402, 2.7465, 2.3239, 2.0969, 1.4641], + device='cuda:3'), covar=tensor([0.5447, 0.5114, 0.1908, 0.3691, 0.2445, 0.2858, 0.1853, 0.5333], + device='cuda:3'), in_proj_covar=tensor([0.0941, 0.0974, 0.0805, 0.0942, 0.0992, 0.0891, 0.0748, 0.0823], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 04:35:16,501 INFO [train.py:901] (3/4) Epoch 22, batch 7450, loss[loss=0.1867, simple_loss=0.2772, pruned_loss=0.04808, over 8246.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2883, pruned_loss=0.06211, over 1619666.32 frames. ], batch size: 24, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:35:17,191 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.489e+02 2.327e+02 2.972e+02 3.761e+02 7.589e+02, threshold=5.944e+02, percent-clipped=3.0 +2023-02-07 04:35:21,647 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177199.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:35:27,651 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-07 04:35:32,259 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=177215.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:35:38,465 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177224.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:35:51,500 INFO [train.py:901] (3/4) Epoch 22, batch 7500, loss[loss=0.1937, simple_loss=0.2773, pruned_loss=0.05505, over 7926.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2881, pruned_loss=0.06253, over 1617749.83 frames. ], batch size: 20, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:36:25,359 INFO [train.py:901] (3/4) Epoch 22, batch 7550, loss[loss=0.1573, simple_loss=0.248, pruned_loss=0.03327, over 7799.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2883, pruned_loss=0.06272, over 1619365.43 frames. ], batch size: 19, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:36:26,052 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.650e+02 2.508e+02 3.019e+02 3.781e+02 7.904e+02, threshold=6.039e+02, percent-clipped=4.0 +2023-02-07 04:36:51,643 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177330.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:36:59,617 INFO [train.py:901] (3/4) Epoch 22, batch 7600, loss[loss=0.1796, simple_loss=0.2503, pruned_loss=0.05446, over 7700.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2874, pruned_loss=0.06201, over 1615027.67 frames. ], batch size: 18, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:37:03,240 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7553, 1.5616, 1.9757, 1.6903, 1.8748, 1.7211, 1.6163, 1.1679], + device='cuda:3'), covar=tensor([0.3888, 0.3664, 0.1562, 0.2628, 0.1920, 0.2505, 0.1563, 0.3770], + device='cuda:3'), in_proj_covar=tensor([0.0943, 0.0978, 0.0810, 0.0947, 0.0997, 0.0896, 0.0752, 0.0826], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 04:37:11,458 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177358.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:37:29,860 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177383.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:37:35,784 INFO [train.py:901] (3/4) Epoch 22, batch 7650, loss[loss=0.1839, simple_loss=0.2574, pruned_loss=0.05521, over 7702.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2878, pruned_loss=0.0626, over 1613593.24 frames. ], batch size: 18, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:37:36,440 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.921e+02 2.559e+02 3.074e+02 4.315e+02 1.263e+03, threshold=6.148e+02, percent-clipped=10.0 +2023-02-07 04:37:39,972 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177398.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:37:57,400 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177423.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:38:08,849 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2427, 2.5310, 2.8434, 1.4309, 3.1564, 1.7635, 1.5620, 2.1291], + device='cuda:3'), covar=tensor([0.0760, 0.0435, 0.0265, 0.0849, 0.0336, 0.0812, 0.0972, 0.0532], + device='cuda:3'), in_proj_covar=tensor([0.0454, 0.0394, 0.0345, 0.0448, 0.0380, 0.0536, 0.0392, 0.0423], + device='cuda:3'), out_proj_covar=tensor([1.2149e-04, 1.0329e-04, 9.0714e-05, 1.1787e-04, 1.0002e-04, 1.5118e-04, + 1.0543e-04, 1.1198e-04], device='cuda:3') +2023-02-07 04:38:09,928 INFO [train.py:901] (3/4) Epoch 22, batch 7700, loss[loss=0.1788, simple_loss=0.2566, pruned_loss=0.05048, over 7789.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2875, pruned_loss=0.06253, over 1616027.62 frames. ], batch size: 19, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:38:30,448 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=177471.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:38:31,729 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=177473.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:38:38,602 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-07 04:38:45,983 INFO [train.py:901] (3/4) Epoch 22, batch 7750, loss[loss=0.2818, simple_loss=0.3428, pruned_loss=0.1104, over 8275.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2864, pruned_loss=0.062, over 1612209.84 frames. ], batch size: 48, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:38:46,654 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.641e+02 2.486e+02 3.125e+02 4.090e+02 1.041e+03, threshold=6.251e+02, percent-clipped=8.0 +2023-02-07 04:39:20,413 INFO [train.py:901] (3/4) Epoch 22, batch 7800, loss[loss=0.1674, simple_loss=0.245, pruned_loss=0.04485, over 7420.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2866, pruned_loss=0.06167, over 1614716.05 frames. ], batch size: 17, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:39:35,120 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8806, 2.0172, 1.6968, 2.5699, 1.2547, 1.5188, 1.9769, 1.9824], + device='cuda:3'), covar=tensor([0.0768, 0.0779, 0.0937, 0.0379, 0.1075, 0.1414, 0.0776, 0.0838], + device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0195, 0.0244, 0.0213, 0.0206, 0.0244, 0.0248, 0.0206], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 04:39:39,801 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177571.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:39:49,858 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177586.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:39:49,890 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177586.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:39:51,164 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177588.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:39:53,636 INFO [train.py:901] (3/4) Epoch 22, batch 7850, loss[loss=0.2205, simple_loss=0.3093, pruned_loss=0.06583, over 8275.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2879, pruned_loss=0.06179, over 1615820.19 frames. ], batch size: 23, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:39:54,290 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.658e+02 2.387e+02 2.753e+02 3.373e+02 6.542e+02, threshold=5.505e+02, percent-clipped=2.0 +2023-02-07 04:39:57,206 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6599, 1.9879, 3.2408, 1.5190, 2.4972, 2.0957, 1.6478, 2.4990], + device='cuda:3'), covar=tensor([0.1786, 0.2498, 0.0769, 0.4340, 0.1619, 0.2939, 0.2287, 0.2070], + device='cuda:3'), in_proj_covar=tensor([0.0529, 0.0609, 0.0558, 0.0650, 0.0651, 0.0598, 0.0541, 0.0636], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 04:40:06,550 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177611.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:40:26,677 INFO [train.py:901] (3/4) Epoch 22, batch 7900, loss[loss=0.2036, simple_loss=0.2809, pruned_loss=0.06316, over 7922.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.288, pruned_loss=0.06226, over 1614715.25 frames. ], batch size: 20, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:40:29,524 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2044, 1.9972, 2.5697, 2.1545, 2.5571, 2.2729, 2.0330, 1.3727], + device='cuda:3'), covar=tensor([0.5816, 0.4973, 0.2064, 0.3761, 0.2597, 0.3206, 0.1987, 0.5583], + device='cuda:3'), in_proj_covar=tensor([0.0950, 0.0987, 0.0813, 0.0955, 0.1005, 0.0904, 0.0757, 0.0833], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 04:40:53,554 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177682.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:41:00,017 INFO [train.py:901] (3/4) Epoch 22, batch 7950, loss[loss=0.1989, simple_loss=0.2855, pruned_loss=0.05618, over 8504.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2883, pruned_loss=0.0624, over 1615648.01 frames. ], batch size: 26, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:41:00,674 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.651e+02 2.424e+02 2.966e+02 3.766e+02 9.319e+02, threshold=5.931e+02, percent-clipped=7.0 +2023-02-07 04:41:31,920 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5778, 2.4977, 1.7084, 2.1640, 2.0916, 1.6141, 1.9838, 2.0703], + device='cuda:3'), covar=tensor([0.1472, 0.0411, 0.1386, 0.0697, 0.0769, 0.1505, 0.1079, 0.1013], + device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0230, 0.0331, 0.0307, 0.0296, 0.0336, 0.0342, 0.0312], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 04:41:33,735 INFO [train.py:901] (3/4) Epoch 22, batch 8000, loss[loss=0.223, simple_loss=0.3093, pruned_loss=0.06838, over 8459.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2877, pruned_loss=0.06185, over 1615193.84 frames. ], batch size: 25, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:41:35,290 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6827, 2.2295, 4.0108, 1.5675, 2.9712, 2.2022, 1.8348, 2.9715], + device='cuda:3'), covar=tensor([0.1907, 0.2692, 0.0708, 0.4577, 0.1797, 0.3182, 0.2281, 0.2199], + device='cuda:3'), in_proj_covar=tensor([0.0528, 0.0607, 0.0557, 0.0649, 0.0650, 0.0596, 0.0539, 0.0633], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 04:42:06,693 INFO [train.py:901] (3/4) Epoch 22, batch 8050, loss[loss=0.1506, simple_loss=0.2344, pruned_loss=0.03336, over 7560.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.288, pruned_loss=0.06286, over 1600607.65 frames. ], batch size: 18, lr: 3.40e-03, grad_scale: 8.0 +2023-02-07 04:42:07,276 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.642e+02 2.308e+02 2.923e+02 3.618e+02 1.070e+03, threshold=5.846e+02, percent-clipped=4.0 +2023-02-07 04:42:10,816 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177798.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:42:22,396 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177815.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:42:39,732 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-07 04:42:44,826 INFO [train.py:901] (3/4) Epoch 23, batch 0, loss[loss=0.2313, simple_loss=0.3042, pruned_loss=0.07921, over 7071.00 frames. ], tot_loss[loss=0.2313, simple_loss=0.3042, pruned_loss=0.07921, over 7071.00 frames. ], batch size: 71, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:42:44,827 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 04:42:56,156 INFO [train.py:935] (3/4) Epoch 23, validation: loss=0.1743, simple_loss=0.274, pruned_loss=0.0373, over 944034.00 frames. +2023-02-07 04:42:56,157 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6747MB +2023-02-07 04:43:08,361 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177842.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:43:10,566 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177844.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:43:12,396 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-07 04:43:26,747 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177867.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:43:28,099 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177869.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:43:32,006 INFO [train.py:901] (3/4) Epoch 23, batch 50, loss[loss=0.2142, simple_loss=0.3026, pruned_loss=0.06293, over 8465.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2869, pruned_loss=0.05999, over 365944.79 frames. ], batch size: 25, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:43:42,596 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6474, 1.9588, 3.0464, 1.4756, 2.2364, 1.8927, 1.7532, 2.0086], + device='cuda:3'), covar=tensor([0.1805, 0.2404, 0.0752, 0.4286, 0.1741, 0.3219, 0.2156, 0.2262], + device='cuda:3'), in_proj_covar=tensor([0.0526, 0.0605, 0.0556, 0.0648, 0.0648, 0.0594, 0.0538, 0.0633], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 04:43:45,279 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.670e+02 2.650e+02 3.149e+02 3.939e+02 1.519e+03, threshold=6.298e+02, percent-clipped=14.0 +2023-02-07 04:43:46,692 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-07 04:44:01,128 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=177915.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:44:02,081 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-07 04:44:07,963 INFO [train.py:901] (3/4) Epoch 23, batch 100, loss[loss=0.1745, simple_loss=0.2616, pruned_loss=0.04364, over 7979.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2867, pruned_loss=0.05988, over 642922.10 frames. ], batch size: 21, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:44:09,366 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-07 04:44:15,318 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7007, 1.8244, 1.6461, 2.2277, 1.0577, 1.4643, 1.6593, 1.8106], + device='cuda:3'), covar=tensor([0.0732, 0.0767, 0.0853, 0.0503, 0.1100, 0.1270, 0.0784, 0.0761], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0196, 0.0244, 0.0214, 0.0207, 0.0246, 0.0250, 0.0207], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 04:44:42,229 INFO [train.py:901] (3/4) Epoch 23, batch 150, loss[loss=0.1974, simple_loss=0.2716, pruned_loss=0.06162, over 7921.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2869, pruned_loss=0.06098, over 859526.82 frames. ], batch size: 20, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:44:49,547 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8990, 1.6798, 2.5958, 1.5699, 2.2462, 2.8278, 2.7608, 2.5573], + device='cuda:3'), covar=tensor([0.0951, 0.1372, 0.0615, 0.1695, 0.1409, 0.0251, 0.0759, 0.0439], + device='cuda:3'), in_proj_covar=tensor([0.0300, 0.0323, 0.0287, 0.0318, 0.0313, 0.0269, 0.0424, 0.0307], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 04:44:54,934 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.486e+02 2.352e+02 3.015e+02 3.767e+02 5.945e+02, threshold=6.031e+02, percent-clipped=0.0 +2023-02-07 04:44:55,351 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-07 04:45:18,299 INFO [train.py:901] (3/4) Epoch 23, batch 200, loss[loss=0.1752, simple_loss=0.2537, pruned_loss=0.04838, over 8090.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2877, pruned_loss=0.06071, over 1028673.18 frames. ], batch size: 21, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:45:19,113 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=178026.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:45:22,634 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178030.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:45:53,029 INFO [train.py:901] (3/4) Epoch 23, batch 250, loss[loss=0.2337, simple_loss=0.315, pruned_loss=0.07621, over 8334.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2874, pruned_loss=0.06, over 1162124.13 frames. ], batch size: 25, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:46:04,760 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-07 04:46:06,105 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.633e+02 2.380e+02 2.804e+02 3.484e+02 6.736e+02, threshold=5.609e+02, percent-clipped=2.0 +2023-02-07 04:46:12,815 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-07 04:46:28,460 INFO [train.py:901] (3/4) Epoch 23, batch 300, loss[loss=0.1822, simple_loss=0.2831, pruned_loss=0.04061, over 8107.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2862, pruned_loss=0.0596, over 1265007.51 frames. ], batch size: 23, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:46:40,070 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178141.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:46:40,631 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=178142.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:46:45,457 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6537, 2.6425, 1.9476, 2.3719, 2.2096, 1.6585, 2.1661, 2.2700], + device='cuda:3'), covar=tensor([0.1549, 0.0422, 0.1200, 0.0636, 0.0710, 0.1451, 0.1004, 0.0908], + device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0231, 0.0332, 0.0308, 0.0297, 0.0337, 0.0342, 0.0313], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 04:46:52,877 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=178159.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:46:53,827 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-07 04:46:54,588 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-07 04:46:59,078 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1294, 1.8806, 4.4243, 1.8897, 2.5689, 5.0586, 5.1758, 4.3939], + device='cuda:3'), covar=tensor([0.1392, 0.1741, 0.0281, 0.2140, 0.1212, 0.0176, 0.0399, 0.0564], + device='cuda:3'), in_proj_covar=tensor([0.0299, 0.0322, 0.0288, 0.0316, 0.0312, 0.0267, 0.0423, 0.0306], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 04:47:03,749 INFO [train.py:901] (3/4) Epoch 23, batch 350, loss[loss=0.2036, simple_loss=0.2809, pruned_loss=0.06316, over 7968.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2864, pruned_loss=0.06005, over 1342250.12 frames. ], batch size: 21, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:47:16,039 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.575e+02 2.418e+02 2.905e+02 3.451e+02 8.072e+02, threshold=5.809e+02, percent-clipped=5.0 +2023-02-07 04:47:23,327 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7134, 1.4207, 3.1431, 1.4717, 2.3686, 3.4041, 3.5171, 2.9318], + device='cuda:3'), covar=tensor([0.1259, 0.1752, 0.0342, 0.2077, 0.0888, 0.0269, 0.0527, 0.0564], + device='cuda:3'), in_proj_covar=tensor([0.0298, 0.0320, 0.0286, 0.0315, 0.0311, 0.0266, 0.0421, 0.0304], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 04:47:38,680 INFO [train.py:901] (3/4) Epoch 23, batch 400, loss[loss=0.1789, simple_loss=0.2464, pruned_loss=0.05574, over 7269.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2856, pruned_loss=0.06033, over 1399766.28 frames. ], batch size: 16, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:47:43,369 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-07 04:47:51,835 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6549, 2.0132, 3.1962, 1.4849, 2.3338, 2.0732, 1.7549, 2.3691], + device='cuda:3'), covar=tensor([0.1887, 0.2596, 0.0900, 0.4469, 0.1927, 0.3122, 0.2217, 0.2298], + device='cuda:3'), in_proj_covar=tensor([0.0526, 0.0605, 0.0556, 0.0645, 0.0647, 0.0592, 0.0537, 0.0633], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 04:48:02,278 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178257.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:48:15,034 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178274.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:48:15,476 INFO [train.py:901] (3/4) Epoch 23, batch 450, loss[loss=0.2059, simple_loss=0.2962, pruned_loss=0.0578, over 8109.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2865, pruned_loss=0.06033, over 1452169.20 frames. ], batch size: 23, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:48:17,125 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0536, 2.1843, 1.8563, 2.9105, 1.2826, 1.7186, 2.0798, 2.2121], + device='cuda:3'), covar=tensor([0.0664, 0.0730, 0.0832, 0.0286, 0.1141, 0.1205, 0.0810, 0.0716], + device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0197, 0.0245, 0.0215, 0.0207, 0.0246, 0.0250, 0.0207], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 04:48:23,160 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178286.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:48:27,631 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.562e+02 2.308e+02 2.812e+02 3.532e+02 1.107e+03, threshold=5.624e+02, percent-clipped=2.0 +2023-02-07 04:48:40,223 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178311.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:48:50,187 INFO [train.py:901] (3/4) Epoch 23, batch 500, loss[loss=0.2429, simple_loss=0.3144, pruned_loss=0.08564, over 8599.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.287, pruned_loss=0.06112, over 1487173.43 frames. ], batch size: 49, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:49:25,956 INFO [train.py:901] (3/4) Epoch 23, batch 550, loss[loss=0.2353, simple_loss=0.3025, pruned_loss=0.08403, over 7117.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.289, pruned_loss=0.0622, over 1514468.55 frames. ], batch size: 72, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:49:39,372 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.748e+02 2.448e+02 3.105e+02 3.761e+02 9.562e+02, threshold=6.211e+02, percent-clipped=5.0 +2023-02-07 04:49:42,455 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178397.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:49:59,317 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178422.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:50:01,203 INFO [train.py:901] (3/4) Epoch 23, batch 600, loss[loss=0.1834, simple_loss=0.2564, pruned_loss=0.0552, over 7543.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2875, pruned_loss=0.06115, over 1532315.54 frames. ], batch size: 18, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:50:14,795 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-07 04:50:33,529 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178470.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:50:36,800 INFO [train.py:901] (3/4) Epoch 23, batch 650, loss[loss=0.2029, simple_loss=0.2911, pruned_loss=0.05738, over 8034.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2864, pruned_loss=0.06067, over 1551054.57 frames. ], batch size: 22, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:50:49,802 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.449e+02 2.230e+02 2.701e+02 3.368e+02 8.641e+02, threshold=5.402e+02, percent-clipped=2.0 +2023-02-07 04:51:04,377 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178513.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:51:12,440 INFO [train.py:901] (3/4) Epoch 23, batch 700, loss[loss=0.2141, simple_loss=0.2971, pruned_loss=0.06555, over 8327.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2863, pruned_loss=0.06038, over 1568285.60 frames. ], batch size: 26, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:51:16,061 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178530.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:51:18,193 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2770, 1.9764, 2.5863, 2.1825, 2.4594, 2.3051, 2.0915, 1.3351], + device='cuda:3'), covar=tensor([0.5302, 0.4965, 0.2016, 0.3759, 0.2448, 0.3061, 0.1900, 0.5525], + device='cuda:3'), in_proj_covar=tensor([0.0941, 0.0986, 0.0813, 0.0952, 0.0997, 0.0899, 0.0754, 0.0831], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 04:51:21,529 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178538.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:51:33,978 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178555.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:51:47,512 INFO [train.py:901] (3/4) Epoch 23, batch 750, loss[loss=0.1929, simple_loss=0.2736, pruned_loss=0.05606, over 7938.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2863, pruned_loss=0.06065, over 1580725.48 frames. ], batch size: 20, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:51:49,830 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9614, 2.3510, 3.7240, 2.0576, 1.8404, 3.7245, 0.6560, 2.1897], + device='cuda:3'), covar=tensor([0.1253, 0.1200, 0.0203, 0.1656, 0.2713, 0.0207, 0.2210, 0.1454], + device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0200, 0.0130, 0.0221, 0.0270, 0.0137, 0.0171, 0.0196], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 04:51:59,472 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6867, 2.1051, 3.2292, 1.4622, 2.5117, 2.0934, 1.8505, 2.5167], + device='cuda:3'), covar=tensor([0.1944, 0.2565, 0.0828, 0.4639, 0.1898, 0.3219, 0.2246, 0.2436], + device='cuda:3'), in_proj_covar=tensor([0.0527, 0.0606, 0.0555, 0.0645, 0.0649, 0.0594, 0.0536, 0.0634], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 04:52:00,640 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.723e+02 2.529e+02 2.988e+02 3.531e+02 9.866e+02, threshold=5.976e+02, percent-clipped=5.0 +2023-02-07 04:52:03,332 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-07 04:52:12,887 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-07 04:52:24,030 INFO [train.py:901] (3/4) Epoch 23, batch 800, loss[loss=0.212, simple_loss=0.2921, pruned_loss=0.06595, over 8603.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2859, pruned_loss=0.06034, over 1592966.05 frames. ], batch size: 49, lr: 3.32e-03, grad_scale: 8.0 +2023-02-07 04:52:32,114 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178637.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:52:38,300 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6212, 1.7610, 1.5358, 2.3060, 1.0612, 1.3801, 1.6575, 1.7814], + device='cuda:3'), covar=tensor([0.0787, 0.0756, 0.0941, 0.0402, 0.1069, 0.1313, 0.0779, 0.0822], + device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0197, 0.0245, 0.0214, 0.0206, 0.0245, 0.0249, 0.0207], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 04:52:57,753 INFO [train.py:901] (3/4) Epoch 23, batch 850, loss[loss=0.1728, simple_loss=0.2448, pruned_loss=0.05041, over 7254.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2872, pruned_loss=0.06152, over 1600386.23 frames. ], batch size: 16, lr: 3.31e-03, grad_scale: 8.0 +2023-02-07 04:53:10,561 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.446e+02 2.561e+02 2.992e+02 3.918e+02 1.040e+03, threshold=5.984e+02, percent-clipped=6.0 +2023-02-07 04:53:24,478 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178712.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:53:26,492 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178715.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 04:53:31,391 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6584, 4.7361, 4.2283, 2.1496, 4.0617, 4.3015, 4.2710, 4.1646], + device='cuda:3'), covar=tensor([0.0756, 0.0524, 0.1029, 0.4865, 0.0864, 0.0970, 0.1253, 0.0727], + device='cuda:3'), in_proj_covar=tensor([0.0535, 0.0440, 0.0435, 0.0544, 0.0426, 0.0448, 0.0431, 0.0389], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 04:53:34,044 INFO [train.py:901] (3/4) Epoch 23, batch 900, loss[loss=0.1684, simple_loss=0.255, pruned_loss=0.04092, over 8478.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2864, pruned_loss=0.06109, over 1607685.01 frames. ], batch size: 25, lr: 3.31e-03, grad_scale: 8.0 +2023-02-07 04:53:55,222 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.99 vs. limit=5.0 +2023-02-07 04:54:03,477 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1028, 1.4747, 4.2817, 1.9540, 2.3092, 4.8924, 4.9831, 4.1743], + device='cuda:3'), covar=tensor([0.1299, 0.2042, 0.0290, 0.2009, 0.1434, 0.0167, 0.0404, 0.0570], + device='cuda:3'), in_proj_covar=tensor([0.0295, 0.0321, 0.0286, 0.0315, 0.0310, 0.0267, 0.0422, 0.0304], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 04:54:09,415 INFO [train.py:901] (3/4) Epoch 23, batch 950, loss[loss=0.2272, simple_loss=0.3038, pruned_loss=0.07526, over 7799.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2876, pruned_loss=0.06161, over 1611948.38 frames. ], batch size: 20, lr: 3.31e-03, grad_scale: 8.0 +2023-02-07 04:54:18,557 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178788.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:54:21,848 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.663e+02 2.330e+02 2.907e+02 3.544e+02 9.473e+02, threshold=5.814e+02, percent-clipped=4.0 +2023-02-07 04:54:22,773 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4422, 1.6929, 1.7706, 1.0702, 1.8525, 1.2123, 0.6087, 1.5417], + device='cuda:3'), covar=tensor([0.0653, 0.0387, 0.0280, 0.0620, 0.0384, 0.0983, 0.0900, 0.0340], + device='cuda:3'), in_proj_covar=tensor([0.0453, 0.0395, 0.0347, 0.0448, 0.0380, 0.0536, 0.0393, 0.0425], + device='cuda:3'), out_proj_covar=tensor([1.2096e-04, 1.0363e-04, 9.1189e-05, 1.1773e-04, 9.9773e-05, 1.5107e-04, + 1.0597e-04, 1.1251e-04], device='cuda:3') +2023-02-07 04:54:35,804 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-07 04:54:37,117 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=178814.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:54:45,365 INFO [train.py:901] (3/4) Epoch 23, batch 1000, loss[loss=0.2409, simple_loss=0.3141, pruned_loss=0.08382, over 8654.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2868, pruned_loss=0.06109, over 1613952.49 frames. ], batch size: 34, lr: 3.31e-03, grad_scale: 8.0 +2023-02-07 04:55:12,377 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-07 04:55:21,365 INFO [train.py:901] (3/4) Epoch 23, batch 1050, loss[loss=0.2036, simple_loss=0.2722, pruned_loss=0.0675, over 7286.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2875, pruned_loss=0.06113, over 1618899.20 frames. ], batch size: 16, lr: 3.31e-03, grad_scale: 8.0 +2023-02-07 04:55:25,412 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-07 04:55:33,396 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.651e+02 2.332e+02 2.695e+02 3.454e+02 6.847e+02, threshold=5.390e+02, percent-clipped=5.0 +2023-02-07 04:55:46,653 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178912.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 04:55:56,202 INFO [train.py:901] (3/4) Epoch 23, batch 1100, loss[loss=0.1926, simple_loss=0.2828, pruned_loss=0.05118, over 8348.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2873, pruned_loss=0.06112, over 1621576.03 frames. ], batch size: 24, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 04:55:59,151 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178929.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:56:00,122 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-07 04:56:32,129 INFO [train.py:901] (3/4) Epoch 23, batch 1150, loss[loss=0.1782, simple_loss=0.2574, pruned_loss=0.04953, over 8078.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2859, pruned_loss=0.06017, over 1621181.64 frames. ], batch size: 21, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 04:56:36,267 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-07 04:56:36,342 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=178981.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:56:45,237 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.550e+02 2.628e+02 3.162e+02 4.177e+02 1.087e+03, threshold=6.324e+02, percent-clipped=6.0 +2023-02-07 04:57:07,131 INFO [train.py:901] (3/4) Epoch 23, batch 1200, loss[loss=0.2351, simple_loss=0.3114, pruned_loss=0.07935, over 7263.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2868, pruned_loss=0.06098, over 1619474.23 frames. ], batch size: 71, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 04:57:29,093 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=179056.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:57:31,038 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=179059.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 04:57:38,236 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5501, 1.5777, 2.1062, 1.2279, 1.1525, 2.0821, 0.2572, 1.2184], + device='cuda:3'), covar=tensor([0.1822, 0.1141, 0.0367, 0.1375, 0.2529, 0.0368, 0.1983, 0.1317], + device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0199, 0.0129, 0.0219, 0.0267, 0.0136, 0.0169, 0.0193], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 04:57:42,801 INFO [train.py:901] (3/4) Epoch 23, batch 1250, loss[loss=0.2324, simple_loss=0.3101, pruned_loss=0.07738, over 8106.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2877, pruned_loss=0.06164, over 1617535.74 frames. ], batch size: 23, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 04:57:55,990 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.427e+02 2.289e+02 2.896e+02 3.686e+02 5.954e+02, threshold=5.791e+02, percent-clipped=0.0 +2023-02-07 04:57:58,279 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179096.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:58:19,010 INFO [train.py:901] (3/4) Epoch 23, batch 1300, loss[loss=0.1997, simple_loss=0.282, pruned_loss=0.05872, over 8491.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2867, pruned_loss=0.06123, over 1612033.47 frames. ], batch size: 26, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 04:58:20,053 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2991, 1.9387, 2.5406, 2.1068, 2.4828, 2.3089, 2.1299, 1.2964], + device='cuda:3'), covar=tensor([0.5150, 0.4990, 0.1881, 0.3784, 0.2452, 0.2901, 0.1811, 0.5381], + device='cuda:3'), in_proj_covar=tensor([0.0938, 0.0985, 0.0810, 0.0949, 0.0994, 0.0899, 0.0754, 0.0828], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 04:58:24,099 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=179132.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:58:51,902 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179171.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:58:53,997 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179174.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 04:58:54,481 INFO [train.py:901] (3/4) Epoch 23, batch 1350, loss[loss=0.1864, simple_loss=0.2688, pruned_loss=0.05197, over 8076.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2856, pruned_loss=0.06064, over 1610713.36 frames. ], batch size: 21, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 04:59:01,717 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179185.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:59:07,800 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.361e+02 2.184e+02 2.635e+02 3.098e+02 5.270e+02, threshold=5.271e+02, percent-clipped=0.0 +2023-02-07 04:59:20,409 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179210.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:59:30,640 INFO [train.py:901] (3/4) Epoch 23, batch 1400, loss[loss=0.1725, simple_loss=0.2469, pruned_loss=0.04906, over 7628.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2859, pruned_loss=0.06066, over 1612554.97 frames. ], batch size: 19, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 04:59:47,155 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179247.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 04:59:53,440 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=179256.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 05:00:06,540 INFO [train.py:901] (3/4) Epoch 23, batch 1450, loss[loss=0.2627, simple_loss=0.3379, pruned_loss=0.09373, over 8305.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2866, pruned_loss=0.06118, over 1614532.69 frames. ], batch size: 23, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 05:00:16,936 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-07 05:00:19,769 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.434e+02 2.293e+02 2.971e+02 3.774e+02 8.745e+02, threshold=5.941e+02, percent-clipped=9.0 +2023-02-07 05:00:43,616 INFO [train.py:901] (3/4) Epoch 23, batch 1500, loss[loss=0.2363, simple_loss=0.3221, pruned_loss=0.07521, over 7978.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2865, pruned_loss=0.0608, over 1615819.45 frames. ], batch size: 21, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 05:00:49,416 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1295, 4.1130, 3.7293, 2.0829, 3.6297, 3.7424, 3.7397, 3.6195], + device='cuda:3'), covar=tensor([0.0858, 0.0615, 0.1022, 0.4406, 0.0960, 0.1073, 0.1287, 0.0907], + device='cuda:3'), in_proj_covar=tensor([0.0537, 0.0443, 0.0438, 0.0548, 0.0431, 0.0453, 0.0435, 0.0391], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 05:01:03,395 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179352.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:01:08,426 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-07 05:01:16,374 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179371.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 05:01:18,847 INFO [train.py:901] (3/4) Epoch 23, batch 1550, loss[loss=0.2054, simple_loss=0.2882, pruned_loss=0.06132, over 8122.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2867, pruned_loss=0.06137, over 1613254.84 frames. ], batch size: 22, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 05:01:20,472 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179377.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:01:21,734 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179379.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:01:31,101 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.616e+02 2.349e+02 2.958e+02 3.969e+02 7.808e+02, threshold=5.916e+02, percent-clipped=1.0 +2023-02-07 05:01:54,019 INFO [train.py:901] (3/4) Epoch 23, batch 1600, loss[loss=0.2673, simple_loss=0.337, pruned_loss=0.09879, over 8331.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2867, pruned_loss=0.06121, over 1609795.23 frames. ], batch size: 26, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 05:01:56,481 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179427.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:01:58,505 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179430.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 05:02:14,535 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179452.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:02:16,570 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179455.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 05:02:21,383 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179462.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 05:02:31,191 INFO [train.py:901] (3/4) Epoch 23, batch 1650, loss[loss=0.2174, simple_loss=0.3102, pruned_loss=0.06229, over 8512.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.286, pruned_loss=0.0607, over 1609311.48 frames. ], batch size: 26, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 05:02:41,801 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8727, 2.2088, 1.7158, 2.7724, 1.3058, 1.4818, 2.1740, 2.2391], + device='cuda:3'), covar=tensor([0.0946, 0.0898, 0.1154, 0.0452, 0.1171, 0.1560, 0.0845, 0.0761], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0199, 0.0245, 0.0215, 0.0207, 0.0248, 0.0251, 0.0209], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 05:02:43,579 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.515e+02 2.367e+02 2.783e+02 3.381e+02 8.055e+02, threshold=5.566e+02, percent-clipped=4.0 +2023-02-07 05:02:50,738 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179503.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:03:06,276 INFO [train.py:901] (3/4) Epoch 23, batch 1700, loss[loss=0.1711, simple_loss=0.2539, pruned_loss=0.04411, over 7246.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2873, pruned_loss=0.06171, over 1608505.59 frames. ], batch size: 16, lr: 3.31e-03, grad_scale: 16.0 +2023-02-07 05:03:08,705 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179528.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:03:42,331 INFO [train.py:901] (3/4) Epoch 23, batch 1750, loss[loss=0.171, simple_loss=0.2546, pruned_loss=0.04367, over 8127.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2872, pruned_loss=0.06141, over 1612441.30 frames. ], batch size: 22, lr: 3.31e-03, grad_scale: 8.0 +2023-02-07 05:03:56,238 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.659e+02 2.481e+02 2.857e+02 3.517e+02 8.396e+02, threshold=5.713e+02, percent-clipped=3.0 +2023-02-07 05:04:17,978 INFO [train.py:901] (3/4) Epoch 23, batch 1800, loss[loss=0.1702, simple_loss=0.2478, pruned_loss=0.04633, over 7266.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2863, pruned_loss=0.06055, over 1611851.88 frames. ], batch size: 16, lr: 3.31e-03, grad_scale: 8.0 +2023-02-07 05:04:19,572 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179627.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 05:04:33,217 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-07 05:04:37,268 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179652.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 05:04:38,535 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179654.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:04:54,476 INFO [train.py:901] (3/4) Epoch 23, batch 1850, loss[loss=0.2285, simple_loss=0.3034, pruned_loss=0.07683, over 8434.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2874, pruned_loss=0.0612, over 1610422.71 frames. ], batch size: 27, lr: 3.31e-03, grad_scale: 8.0 +2023-02-07 05:05:07,486 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.599e+02 2.311e+02 2.831e+02 3.615e+02 8.108e+02, threshold=5.663e+02, percent-clipped=6.0 +2023-02-07 05:05:28,522 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=179723.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:05:29,813 INFO [train.py:901] (3/4) Epoch 23, batch 1900, loss[loss=0.2262, simple_loss=0.2998, pruned_loss=0.07633, over 8352.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2877, pruned_loss=0.06158, over 1611457.30 frames. ], batch size: 49, lr: 3.31e-03, grad_scale: 8.0 +2023-02-07 05:05:59,899 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-07 05:06:05,578 INFO [train.py:901] (3/4) Epoch 23, batch 1950, loss[loss=0.1734, simple_loss=0.2701, pruned_loss=0.03839, over 8501.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2882, pruned_loss=0.06205, over 1611283.83 frames. ], batch size: 26, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:06:12,655 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-07 05:06:12,867 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4005, 1.4243, 1.4002, 1.7798, 0.7188, 1.2752, 1.3035, 1.5013], + device='cuda:3'), covar=tensor([0.0946, 0.0966, 0.1041, 0.0522, 0.1237, 0.1497, 0.0822, 0.0769], + device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0198, 0.0242, 0.0213, 0.0206, 0.0245, 0.0250, 0.0206], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 05:06:19,480 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.621e+02 2.457e+02 2.986e+02 3.643e+02 8.972e+02, threshold=5.972e+02, percent-clipped=4.0 +2023-02-07 05:06:28,061 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=179806.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 05:06:31,257 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-07 05:06:34,179 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179814.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:06:41,610 INFO [train.py:901] (3/4) Epoch 23, batch 2000, loss[loss=0.183, simple_loss=0.2566, pruned_loss=0.05477, over 7424.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2866, pruned_loss=0.0612, over 1611535.16 frames. ], batch size: 17, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:06:50,603 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179838.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:06:56,181 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8963, 2.4399, 4.1873, 1.6844, 3.1493, 2.5766, 2.0343, 3.1101], + device='cuda:3'), covar=tensor([0.2024, 0.2965, 0.0897, 0.5163, 0.1882, 0.3199, 0.2573, 0.2384], + device='cuda:3'), in_proj_covar=tensor([0.0526, 0.0607, 0.0556, 0.0645, 0.0649, 0.0594, 0.0537, 0.0631], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 05:07:16,416 INFO [train.py:901] (3/4) Epoch 23, batch 2050, loss[loss=0.2086, simple_loss=0.3022, pruned_loss=0.05753, over 8104.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2871, pruned_loss=0.06145, over 1610705.39 frames. ], batch size: 23, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:07:30,035 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.466e+02 2.444e+02 2.856e+02 3.794e+02 1.051e+03, threshold=5.713e+02, percent-clipped=7.0 +2023-02-07 05:07:49,561 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179921.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 05:07:50,922 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.9226, 3.8452, 3.5500, 2.4685, 3.4613, 3.4750, 3.5698, 3.2952], + device='cuda:3'), covar=tensor([0.0801, 0.0705, 0.1082, 0.3713, 0.0893, 0.1257, 0.1213, 0.1091], + device='cuda:3'), in_proj_covar=tensor([0.0530, 0.0440, 0.0435, 0.0542, 0.0429, 0.0448, 0.0432, 0.0389], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 05:07:51,616 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179924.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 05:07:52,161 INFO [train.py:901] (3/4) Epoch 23, batch 2100, loss[loss=0.1797, simple_loss=0.2781, pruned_loss=0.04062, over 8500.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2871, pruned_loss=0.06141, over 1613077.24 frames. ], batch size: 26, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:08:04,852 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179942.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:08:25,020 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5442, 2.8685, 2.4621, 4.0888, 1.7569, 2.1304, 2.7892, 3.0580], + device='cuda:3'), covar=tensor([0.0662, 0.0800, 0.0758, 0.0244, 0.1078, 0.1142, 0.0787, 0.0733], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0198, 0.0243, 0.0214, 0.0206, 0.0246, 0.0249, 0.0207], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 05:08:27,544 INFO [train.py:901] (3/4) Epoch 23, batch 2150, loss[loss=0.2009, simple_loss=0.2677, pruned_loss=0.06701, over 7706.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2875, pruned_loss=0.06201, over 1611954.65 frames. ], batch size: 18, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:08:41,580 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.681e+02 2.311e+02 2.940e+02 3.642e+02 8.826e+02, threshold=5.880e+02, percent-clipped=6.0 +2023-02-07 05:08:44,589 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=179998.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:09:05,728 INFO [train.py:901] (3/4) Epoch 23, batch 2200, loss[loss=0.2057, simple_loss=0.3005, pruned_loss=0.05549, over 8194.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2875, pruned_loss=0.06191, over 1611537.22 frames. ], batch size: 23, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:09:18,806 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180044.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:09:23,721 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5726, 1.8327, 2.6756, 1.4196, 1.9283, 1.9675, 1.6087, 1.9086], + device='cuda:3'), covar=tensor([0.1841, 0.2639, 0.0808, 0.4665, 0.1953, 0.3135, 0.2424, 0.2181], + device='cuda:3'), in_proj_covar=tensor([0.0526, 0.0610, 0.0557, 0.0648, 0.0650, 0.0595, 0.0539, 0.0633], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 05:09:40,654 INFO [train.py:901] (3/4) Epoch 23, batch 2250, loss[loss=0.2184, simple_loss=0.2999, pruned_loss=0.06847, over 6461.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2883, pruned_loss=0.06212, over 1614136.93 frames. ], batch size: 71, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:09:53,773 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.578e+02 2.377e+02 2.815e+02 3.570e+02 6.536e+02, threshold=5.630e+02, percent-clipped=1.0 +2023-02-07 05:09:54,026 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180094.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:10:07,872 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180113.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:10:12,118 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180119.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:10:16,774 INFO [train.py:901] (3/4) Epoch 23, batch 2300, loss[loss=0.2142, simple_loss=0.2989, pruned_loss=0.06477, over 8500.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2879, pruned_loss=0.06181, over 1613744.98 frames. ], batch size: 26, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:10:40,104 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=180158.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:10:52,678 INFO [train.py:901] (3/4) Epoch 23, batch 2350, loss[loss=0.2017, simple_loss=0.2964, pruned_loss=0.05347, over 8192.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2885, pruned_loss=0.06206, over 1617121.72 frames. ], batch size: 23, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:10:54,329 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180177.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 05:11:05,884 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.381e+02 2.451e+02 2.928e+02 3.544e+02 9.883e+02, threshold=5.856e+02, percent-clipped=4.0 +2023-02-07 05:11:11,597 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180202.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 05:11:25,820 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180223.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:11:27,097 INFO [train.py:901] (3/4) Epoch 23, batch 2400, loss[loss=0.1934, simple_loss=0.2856, pruned_loss=0.05063, over 8241.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2875, pruned_loss=0.06139, over 1618038.02 frames. ], batch size: 24, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:11:59,473 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=180268.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 05:12:02,920 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180273.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:12:04,126 INFO [train.py:901] (3/4) Epoch 23, batch 2450, loss[loss=0.2316, simple_loss=0.307, pruned_loss=0.07814, over 8235.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2874, pruned_loss=0.0616, over 1613591.05 frames. ], batch size: 22, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:12:12,627 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=180286.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:12:18,007 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 2.501e+02 2.918e+02 3.866e+02 1.157e+03, threshold=5.835e+02, percent-clipped=6.0 +2023-02-07 05:12:39,632 INFO [train.py:901] (3/4) Epoch 23, batch 2500, loss[loss=0.1936, simple_loss=0.2814, pruned_loss=0.05293, over 8129.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2888, pruned_loss=0.06188, over 1615652.38 frames. ], batch size: 22, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:13:00,480 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180354.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:13:12,934 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180369.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:13:16,784 INFO [train.py:901] (3/4) Epoch 23, batch 2550, loss[loss=0.2036, simple_loss=0.2824, pruned_loss=0.06238, over 7974.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2884, pruned_loss=0.06164, over 1615993.29 frames. ], batch size: 21, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:13:22,407 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180383.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 05:13:25,750 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=180388.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:13:29,886 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.741e+02 2.435e+02 3.031e+02 3.942e+02 1.076e+03, threshold=6.063e+02, percent-clipped=1.0 +2023-02-07 05:13:30,150 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180394.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:13:35,751 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180401.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:13:51,966 INFO [train.py:901] (3/4) Epoch 23, batch 2600, loss[loss=0.2001, simple_loss=0.2857, pruned_loss=0.05725, over 8471.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2877, pruned_loss=0.06156, over 1615933.06 frames. ], batch size: 25, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:14:28,406 INFO [train.py:901] (3/4) Epoch 23, batch 2650, loss[loss=0.1912, simple_loss=0.2626, pruned_loss=0.05993, over 7693.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2865, pruned_loss=0.06127, over 1612664.42 frames. ], batch size: 18, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:14:42,182 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.334e+02 2.331e+02 2.876e+02 3.734e+02 9.435e+02, threshold=5.753e+02, percent-clipped=4.0 +2023-02-07 05:14:48,432 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180503.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:14:51,945 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5148, 1.5015, 2.1046, 1.4162, 1.1048, 2.0273, 0.2731, 1.2973], + device='cuda:3'), covar=tensor([0.1796, 0.1385, 0.0434, 0.1133, 0.2809, 0.0429, 0.2066, 0.1195], + device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0199, 0.0129, 0.0222, 0.0269, 0.0136, 0.0170, 0.0194], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 05:15:04,091 INFO [train.py:901] (3/4) Epoch 23, batch 2700, loss[loss=0.1867, simple_loss=0.2708, pruned_loss=0.05134, over 7797.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2853, pruned_loss=0.06074, over 1613783.73 frames. ], batch size: 19, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:15:07,059 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180529.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:15:24,142 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180554.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:15:33,269 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=180567.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:15:39,689 INFO [train.py:901] (3/4) Epoch 23, batch 2750, loss[loss=0.1948, simple_loss=0.284, pruned_loss=0.05284, over 8568.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2857, pruned_loss=0.06087, over 1611703.61 frames. ], batch size: 39, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:15:48,799 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180588.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:15:53,497 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.550e+02 2.355e+02 2.814e+02 3.432e+02 9.125e+02, threshold=5.629e+02, percent-clipped=4.0 +2023-02-07 05:16:15,675 INFO [train.py:901] (3/4) Epoch 23, batch 2800, loss[loss=0.2073, simple_loss=0.2829, pruned_loss=0.06588, over 7654.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2853, pruned_loss=0.06016, over 1614183.91 frames. ], batch size: 19, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:16:26,293 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180639.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 05:16:38,634 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180657.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:16:43,330 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180664.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 05:16:50,750 INFO [train.py:901] (3/4) Epoch 23, batch 2850, loss[loss=0.2146, simple_loss=0.2976, pruned_loss=0.06583, over 8496.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2864, pruned_loss=0.06028, over 1618315.96 frames. ], batch size: 26, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:16:55,707 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180682.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:16:55,730 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180682.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:17:04,507 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.470e+02 2.469e+02 3.037e+02 3.866e+02 9.714e+02, threshold=6.075e+02, percent-clipped=7.0 +2023-02-07 05:17:07,454 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=180698.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:17:07,605 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0392, 1.1507, 1.1437, 0.8680, 1.1526, 0.9835, 0.1785, 1.1501], + device='cuda:3'), covar=tensor([0.0425, 0.0407, 0.0372, 0.0504, 0.0468, 0.1046, 0.0890, 0.0343], + device='cuda:3'), in_proj_covar=tensor([0.0456, 0.0394, 0.0348, 0.0445, 0.0380, 0.0534, 0.0393, 0.0422], + device='cuda:3'), out_proj_covar=tensor([1.2181e-04, 1.0316e-04, 9.1228e-05, 1.1701e-04, 9.9895e-05, 1.5036e-04, + 1.0583e-04, 1.1171e-04], device='cuda:3') +2023-02-07 05:17:27,380 INFO [train.py:901] (3/4) Epoch 23, batch 2900, loss[loss=0.2017, simple_loss=0.2976, pruned_loss=0.05289, over 8493.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2863, pruned_loss=0.06083, over 1615881.23 frames. ], batch size: 29, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:17:45,109 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180750.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:17:52,141 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180759.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:17:59,469 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-07 05:18:02,965 INFO [train.py:901] (3/4) Epoch 23, batch 2950, loss[loss=0.2492, simple_loss=0.3206, pruned_loss=0.08894, over 7288.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2875, pruned_loss=0.0617, over 1616272.08 frames. ], batch size: 73, lr: 3.30e-03, grad_scale: 8.0 +2023-02-07 05:18:09,335 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180784.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:18:16,024 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 2.356e+02 2.925e+02 3.942e+02 6.480e+02, threshold=5.850e+02, percent-clipped=1.0 +2023-02-07 05:18:19,052 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8811, 1.3033, 4.3250, 1.8694, 2.4969, 4.9625, 5.0280, 4.2085], + device='cuda:3'), covar=tensor([0.1404, 0.2114, 0.0282, 0.2048, 0.1191, 0.0165, 0.0371, 0.0557], + device='cuda:3'), in_proj_covar=tensor([0.0295, 0.0318, 0.0285, 0.0314, 0.0308, 0.0267, 0.0420, 0.0303], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 05:18:30,304 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180813.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:18:33,373 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-07 05:18:38,135 INFO [train.py:901] (3/4) Epoch 23, batch 3000, loss[loss=0.1757, simple_loss=0.2645, pruned_loss=0.04346, over 8330.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2874, pruned_loss=0.06144, over 1613730.72 frames. ], batch size: 26, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:18:38,135 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 05:18:50,536 INFO [train.py:935] (3/4) Epoch 23, validation: loss=0.1735, simple_loss=0.2731, pruned_loss=0.03696, over 944034.00 frames. +2023-02-07 05:18:50,537 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6747MB +2023-02-07 05:18:57,234 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 +2023-02-07 05:19:03,699 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180843.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:19:26,991 INFO [train.py:901] (3/4) Epoch 23, batch 3050, loss[loss=0.2724, simple_loss=0.331, pruned_loss=0.1069, over 7057.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2865, pruned_loss=0.06148, over 1608892.50 frames. ], batch size: 71, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:19:40,679 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.441e+02 2.515e+02 3.107e+02 3.968e+02 1.139e+03, threshold=6.214e+02, percent-clipped=7.0 +2023-02-07 05:20:02,332 INFO [train.py:901] (3/4) Epoch 23, batch 3100, loss[loss=0.2211, simple_loss=0.3054, pruned_loss=0.06838, over 8250.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2872, pruned_loss=0.06159, over 1609134.92 frames. ], batch size: 48, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:20:07,184 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=180932.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:20:11,450 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180938.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:20:29,340 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180963.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:20:38,167 INFO [train.py:901] (3/4) Epoch 23, batch 3150, loss[loss=0.19, simple_loss=0.2927, pruned_loss=0.04364, over 8113.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2867, pruned_loss=0.06119, over 1612941.81 frames. ], batch size: 23, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:20:51,969 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.754e+02 2.467e+02 3.042e+02 3.660e+02 1.036e+03, threshold=6.084e+02, percent-clipped=2.0 +2023-02-07 05:21:14,471 INFO [train.py:901] (3/4) Epoch 23, batch 3200, loss[loss=0.1875, simple_loss=0.2794, pruned_loss=0.04784, over 8543.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2872, pruned_loss=0.06159, over 1611693.26 frames. ], batch size: 39, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:21:20,225 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8827, 1.8372, 1.8910, 1.7179, 1.0108, 1.7628, 2.2356, 1.8370], + device='cuda:3'), covar=tensor([0.0434, 0.1135, 0.1570, 0.1330, 0.0588, 0.1319, 0.0618, 0.0638], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0152, 0.0188, 0.0159, 0.0099, 0.0161, 0.0111, 0.0142], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 05:21:29,945 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181047.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:21:45,872 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181069.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:21:49,883 INFO [train.py:901] (3/4) Epoch 23, batch 3250, loss[loss=0.2459, simple_loss=0.3171, pruned_loss=0.08738, over 8467.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2884, pruned_loss=0.06223, over 1620816.62 frames. ], batch size: 27, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:21:57,726 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5924, 1.5747, 4.7790, 1.8290, 4.2585, 3.9546, 4.3450, 4.1704], + device='cuda:3'), covar=tensor([0.0537, 0.4413, 0.0450, 0.3841, 0.1133, 0.0917, 0.0535, 0.0621], + device='cuda:3'), in_proj_covar=tensor([0.0642, 0.0648, 0.0707, 0.0640, 0.0717, 0.0613, 0.0612, 0.0689], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 05:22:03,753 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.537e+02 2.376e+02 2.917e+02 3.369e+02 6.745e+02, threshold=5.834e+02, percent-clipped=1.0 +2023-02-07 05:22:04,708 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=181094.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:22:04,822 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181094.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:22:26,226 INFO [train.py:901] (3/4) Epoch 23, batch 3300, loss[loss=0.2232, simple_loss=0.2969, pruned_loss=0.07472, over 8085.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2866, pruned_loss=0.06127, over 1616526.13 frames. ], batch size: 21, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:23:01,365 INFO [train.py:901] (3/4) Epoch 23, batch 3350, loss[loss=0.1918, simple_loss=0.264, pruned_loss=0.05975, over 7795.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2871, pruned_loss=0.06158, over 1616470.06 frames. ], batch size: 19, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:23:10,457 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=181187.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:23:14,976 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.531e+02 2.358e+02 3.053e+02 3.666e+02 9.674e+02, threshold=6.107e+02, percent-clipped=1.0 +2023-02-07 05:23:26,432 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181209.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:23:38,223 INFO [train.py:901] (3/4) Epoch 23, batch 3400, loss[loss=0.1876, simple_loss=0.2677, pruned_loss=0.05377, over 7444.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2875, pruned_loss=0.06143, over 1613884.15 frames. ], batch size: 17, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:23:39,864 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6662, 1.6096, 2.1634, 1.4629, 1.1946, 2.0774, 0.2959, 1.2983], + device='cuda:3'), covar=tensor([0.1573, 0.1256, 0.0322, 0.0962, 0.2542, 0.0396, 0.1941, 0.1210], + device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0198, 0.0129, 0.0220, 0.0270, 0.0137, 0.0170, 0.0193], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 05:23:46,806 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.49 vs. limit=2.0 +2023-02-07 05:24:13,227 INFO [train.py:901] (3/4) Epoch 23, batch 3450, loss[loss=0.2414, simple_loss=0.3071, pruned_loss=0.08786, over 8497.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2886, pruned_loss=0.0623, over 1615006.25 frames. ], batch size: 29, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:24:22,689 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7388, 1.9743, 1.7221, 2.5963, 1.2695, 1.5654, 2.0428, 2.1066], + device='cuda:3'), covar=tensor([0.0993, 0.0851, 0.1096, 0.0442, 0.1056, 0.1418, 0.0784, 0.0814], + device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0197, 0.0243, 0.0213, 0.0204, 0.0244, 0.0249, 0.0207], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 05:24:27,414 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.916e+02 2.466e+02 2.960e+02 3.783e+02 8.296e+02, threshold=5.920e+02, percent-clipped=4.0 +2023-02-07 05:24:32,998 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181302.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:24:33,736 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181303.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:24:42,799 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181315.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:24:49,554 INFO [train.py:901] (3/4) Epoch 23, batch 3500, loss[loss=0.1775, simple_loss=0.2611, pruned_loss=0.047, over 8033.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2881, pruned_loss=0.06174, over 1619133.15 frames. ], batch size: 22, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:24:52,757 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181328.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:25:05,780 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6559, 1.9624, 2.0307, 1.2931, 2.0741, 1.4667, 0.9316, 1.8508], + device='cuda:3'), covar=tensor([0.0765, 0.0448, 0.0337, 0.0770, 0.0500, 0.1036, 0.0927, 0.0373], + device='cuda:3'), in_proj_covar=tensor([0.0456, 0.0394, 0.0348, 0.0451, 0.0381, 0.0537, 0.0394, 0.0424], + device='cuda:3'), out_proj_covar=tensor([1.2173e-04, 1.0319e-04, 9.1233e-05, 1.1864e-04, 1.0006e-04, 1.5137e-04, + 1.0606e-04, 1.1202e-04], device='cuda:3') +2023-02-07 05:25:07,629 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-07 05:25:25,791 INFO [train.py:901] (3/4) Epoch 23, batch 3550, loss[loss=0.18, simple_loss=0.266, pruned_loss=0.04693, over 8333.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2894, pruned_loss=0.06267, over 1618425.52 frames. ], batch size: 25, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:25:32,319 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2339, 2.0479, 2.6686, 2.2215, 2.6604, 2.2999, 2.0970, 1.5134], + device='cuda:3'), covar=tensor([0.5503, 0.5071, 0.1998, 0.3645, 0.2560, 0.3097, 0.1923, 0.5582], + device='cuda:3'), in_proj_covar=tensor([0.0949, 0.0992, 0.0815, 0.0957, 0.1003, 0.0905, 0.0757, 0.0834], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 05:25:39,015 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.671e+02 2.351e+02 2.882e+02 3.469e+02 9.271e+02, threshold=5.765e+02, percent-clipped=2.0 +2023-02-07 05:26:00,043 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0516, 2.2069, 1.8437, 2.8538, 1.3900, 1.6985, 2.0821, 2.3571], + device='cuda:3'), covar=tensor([0.0697, 0.0798, 0.0888, 0.0336, 0.1103, 0.1285, 0.0847, 0.0720], + device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0197, 0.0243, 0.0213, 0.0204, 0.0244, 0.0250, 0.0206], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 05:26:01,197 INFO [train.py:901] (3/4) Epoch 23, batch 3600, loss[loss=0.2359, simple_loss=0.3158, pruned_loss=0.07793, over 8291.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2891, pruned_loss=0.06264, over 1616753.94 frames. ], batch size: 23, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:26:16,016 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5006, 1.4114, 1.8664, 1.3338, 1.1471, 1.7855, 0.2383, 1.2063], + device='cuda:3'), covar=tensor([0.1794, 0.1586, 0.0390, 0.0943, 0.2658, 0.0520, 0.2132, 0.1319], + device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0198, 0.0129, 0.0218, 0.0268, 0.0136, 0.0169, 0.0192], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 05:26:30,562 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181465.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:26:37,776 INFO [train.py:901] (3/4) Epoch 23, batch 3650, loss[loss=0.1877, simple_loss=0.2761, pruned_loss=0.04961, over 8071.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2892, pruned_loss=0.06252, over 1619035.10 frames. ], batch size: 21, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:26:48,304 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181490.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:26:50,917 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.577e+02 2.424e+02 2.919e+02 3.720e+02 6.119e+02, threshold=5.839e+02, percent-clipped=1.0 +2023-02-07 05:27:11,122 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-07 05:27:12,461 INFO [train.py:901] (3/4) Epoch 23, batch 3700, loss[loss=0.1802, simple_loss=0.2624, pruned_loss=0.04905, over 7920.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2885, pruned_loss=0.06263, over 1613718.20 frames. ], batch size: 20, lr: 3.29e-03, grad_scale: 8.0 +2023-02-07 05:27:17,712 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0 +2023-02-07 05:27:30,291 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181548.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:27:37,401 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181558.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:27:49,553 INFO [train.py:901] (3/4) Epoch 23, batch 3750, loss[loss=0.1803, simple_loss=0.2608, pruned_loss=0.04992, over 7713.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2883, pruned_loss=0.06256, over 1613927.00 frames. ], batch size: 18, lr: 3.29e-03, grad_scale: 16.0 +2023-02-07 05:27:55,393 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181583.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:28:02,817 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.693e+02 2.354e+02 2.844e+02 3.677e+02 7.170e+02, threshold=5.688e+02, percent-clipped=4.0 +2023-02-07 05:28:24,873 INFO [train.py:901] (3/4) Epoch 23, batch 3800, loss[loss=0.2009, simple_loss=0.2889, pruned_loss=0.05643, over 7796.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2866, pruned_loss=0.06126, over 1613307.73 frames. ], batch size: 20, lr: 3.29e-03, grad_scale: 16.0 +2023-02-07 05:28:44,349 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4866, 1.8369, 1.4145, 2.9105, 1.4827, 1.3632, 2.1997, 2.0931], + device='cuda:3'), covar=tensor([0.1612, 0.1382, 0.2032, 0.0410, 0.1313, 0.2047, 0.0961, 0.1044], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0198, 0.0246, 0.0215, 0.0207, 0.0247, 0.0252, 0.0209], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 05:28:47,153 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([6.0775, 1.7351, 6.2023, 2.1984, 5.5424, 5.1447, 5.7177, 5.6536], + device='cuda:3'), covar=tensor([0.0416, 0.4584, 0.0289, 0.3733, 0.0870, 0.0784, 0.0410, 0.0485], + device='cuda:3'), in_proj_covar=tensor([0.0642, 0.0650, 0.0706, 0.0641, 0.0721, 0.0617, 0.0616, 0.0688], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 05:28:49,221 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=181659.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:29:00,794 INFO [train.py:901] (3/4) Epoch 23, batch 3850, loss[loss=0.2142, simple_loss=0.2982, pruned_loss=0.06504, over 8577.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2879, pruned_loss=0.0616, over 1617256.45 frames. ], batch size: 31, lr: 3.29e-03, grad_scale: 16.0 +2023-02-07 05:29:14,859 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.406e+02 2.361e+02 2.900e+02 3.650e+02 9.007e+02, threshold=5.800e+02, percent-clipped=7.0 +2023-02-07 05:29:22,393 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-07 05:29:36,633 INFO [train.py:901] (3/4) Epoch 23, batch 3900, loss[loss=0.196, simple_loss=0.2773, pruned_loss=0.05735, over 8329.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2869, pruned_loss=0.06125, over 1619240.95 frames. ], batch size: 26, lr: 3.29e-03, grad_scale: 16.0 +2023-02-07 05:29:42,913 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8312, 1.5425, 1.9153, 1.5912, 1.0442, 1.6605, 2.1678, 2.0195], + device='cuda:3'), covar=tensor([0.0423, 0.1295, 0.1617, 0.1424, 0.0613, 0.1430, 0.0645, 0.0588], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0151, 0.0188, 0.0159, 0.0100, 0.0161, 0.0111, 0.0142], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 05:30:10,472 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181774.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:30:11,006 INFO [train.py:901] (3/4) Epoch 23, batch 3950, loss[loss=0.1942, simple_loss=0.2798, pruned_loss=0.05427, over 7980.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2873, pruned_loss=0.06163, over 1616886.69 frames. ], batch size: 21, lr: 3.29e-03, grad_scale: 16.0 +2023-02-07 05:30:26,257 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.486e+02 2.359e+02 2.788e+02 3.393e+02 6.824e+02, threshold=5.575e+02, percent-clipped=4.0 +2023-02-07 05:30:47,704 INFO [train.py:901] (3/4) Epoch 23, batch 4000, loss[loss=0.1858, simple_loss=0.2642, pruned_loss=0.0537, over 7689.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2877, pruned_loss=0.06166, over 1618662.43 frames. ], batch size: 18, lr: 3.29e-03, grad_scale: 16.0 +2023-02-07 05:31:04,380 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2615, 1.2734, 3.3976, 1.0601, 3.0094, 2.8397, 3.0940, 3.0165], + device='cuda:3'), covar=tensor([0.0827, 0.4081, 0.0842, 0.4250, 0.1478, 0.1170, 0.0804, 0.0921], + device='cuda:3'), in_proj_covar=tensor([0.0640, 0.0650, 0.0706, 0.0641, 0.0717, 0.0615, 0.0613, 0.0686], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 05:31:22,586 INFO [train.py:901] (3/4) Epoch 23, batch 4050, loss[loss=0.1868, simple_loss=0.2683, pruned_loss=0.05266, over 7531.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2868, pruned_loss=0.06101, over 1614972.58 frames. ], batch size: 18, lr: 3.29e-03, grad_scale: 16.0 +2023-02-07 05:31:34,384 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=181892.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:31:35,721 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.450e+02 2.508e+02 2.885e+02 3.954e+02 8.020e+02, threshold=5.770e+02, percent-clipped=6.0 +2023-02-07 05:31:59,838 INFO [train.py:901] (3/4) Epoch 23, batch 4100, loss[loss=0.1779, simple_loss=0.2549, pruned_loss=0.05052, over 7652.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2873, pruned_loss=0.06109, over 1612857.54 frames. ], batch size: 19, lr: 3.29e-03, grad_scale: 16.0 +2023-02-07 05:32:10,663 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1384, 3.7161, 2.4093, 2.8959, 2.7128, 1.9738, 2.8905, 2.9192], + device='cuda:3'), covar=tensor([0.1602, 0.0375, 0.1160, 0.0710, 0.0677, 0.1521, 0.0996, 0.1156], + device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0235, 0.0340, 0.0312, 0.0303, 0.0340, 0.0347, 0.0320], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 05:32:13,920 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1673, 1.2292, 1.4830, 1.1560, 0.7217, 1.2935, 1.2607, 1.0388], + device='cuda:3'), covar=tensor([0.0663, 0.1309, 0.1720, 0.1525, 0.0598, 0.1480, 0.0732, 0.0727], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0190, 0.0159, 0.0100, 0.0163, 0.0112, 0.0143], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 05:32:34,924 INFO [train.py:901] (3/4) Epoch 23, batch 4150, loss[loss=0.1614, simple_loss=0.2415, pruned_loss=0.04061, over 7798.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.287, pruned_loss=0.06093, over 1611842.95 frames. ], batch size: 19, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:32:48,428 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.466e+02 2.356e+02 2.929e+02 3.956e+02 6.697e+02, threshold=5.858e+02, percent-clipped=3.0 +2023-02-07 05:32:49,332 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181996.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:32:58,730 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=182007.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:33:11,753 INFO [train.py:901] (3/4) Epoch 23, batch 4200, loss[loss=0.2106, simple_loss=0.2893, pruned_loss=0.06598, over 8664.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2864, pruned_loss=0.06072, over 1608999.27 frames. ], batch size: 49, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:33:16,205 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=182030.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:33:17,893 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.37 vs. limit=5.0 +2023-02-07 05:33:25,771 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-07 05:33:33,433 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=182055.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:33:47,611 INFO [train.py:901] (3/4) Epoch 23, batch 4250, loss[loss=0.2069, simple_loss=0.2988, pruned_loss=0.05753, over 8480.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.287, pruned_loss=0.06085, over 1614165.83 frames. ], batch size: 27, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:33:48,141 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.10 vs. limit=5.0 +2023-02-07 05:33:49,026 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-07 05:34:01,341 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.582e+02 2.416e+02 2.989e+02 3.588e+02 6.339e+02, threshold=5.979e+02, percent-clipped=2.0 +2023-02-07 05:34:06,984 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5044, 1.6369, 2.1783, 1.3887, 1.4892, 1.7300, 1.5019, 1.5009], + device='cuda:3'), covar=tensor([0.1883, 0.2516, 0.0984, 0.4291, 0.1959, 0.3318, 0.2380, 0.2132], + device='cuda:3'), in_proj_covar=tensor([0.0523, 0.0608, 0.0552, 0.0647, 0.0644, 0.0596, 0.0539, 0.0628], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 05:34:22,741 INFO [train.py:901] (3/4) Epoch 23, batch 4300, loss[loss=0.209, simple_loss=0.2937, pruned_loss=0.06219, over 8323.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.287, pruned_loss=0.06072, over 1615595.65 frames. ], batch size: 26, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:34:58,709 INFO [train.py:901] (3/4) Epoch 23, batch 4350, loss[loss=0.213, simple_loss=0.2988, pruned_loss=0.06357, over 8605.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2858, pruned_loss=0.0605, over 1609800.21 frames. ], batch size: 31, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:35:12,957 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5072, 1.4514, 4.7371, 1.8121, 4.1204, 3.9452, 4.2582, 4.1202], + device='cuda:3'), covar=tensor([0.0604, 0.4887, 0.0484, 0.4108, 0.1167, 0.0951, 0.0609, 0.0692], + device='cuda:3'), in_proj_covar=tensor([0.0642, 0.0653, 0.0709, 0.0642, 0.0720, 0.0617, 0.0616, 0.0689], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 05:35:13,490 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.474e+02 2.346e+02 2.960e+02 3.931e+02 9.702e+02, threshold=5.919e+02, percent-clipped=9.0 +2023-02-07 05:35:21,958 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-07 05:35:34,697 INFO [train.py:901] (3/4) Epoch 23, batch 4400, loss[loss=0.2298, simple_loss=0.3054, pruned_loss=0.07709, over 8734.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2857, pruned_loss=0.06027, over 1609514.58 frames. ], batch size: 30, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:35:50,851 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6036, 1.3312, 1.5518, 1.2396, 0.9240, 1.3904, 1.4619, 1.2343], + device='cuda:3'), covar=tensor([0.0540, 0.1311, 0.1721, 0.1468, 0.0597, 0.1495, 0.0705, 0.0718], + device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0152, 0.0189, 0.0160, 0.0100, 0.0163, 0.0111, 0.0143], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 05:35:55,268 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 +2023-02-07 05:36:03,255 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=182263.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:36:05,077 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-07 05:36:11,442 INFO [train.py:901] (3/4) Epoch 23, batch 4450, loss[loss=0.2352, simple_loss=0.3133, pruned_loss=0.07851, over 8238.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.286, pruned_loss=0.06053, over 1612467.19 frames. ], batch size: 24, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:36:20,488 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=182288.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:36:26,042 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.640e+02 2.606e+02 3.225e+02 4.349e+02 9.132e+02, threshold=6.449e+02, percent-clipped=7.0 +2023-02-07 05:36:29,003 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=182299.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:36:47,067 INFO [train.py:901] (3/4) Epoch 23, batch 4500, loss[loss=0.2092, simple_loss=0.2993, pruned_loss=0.05958, over 8346.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2877, pruned_loss=0.06164, over 1612787.09 frames. ], batch size: 24, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:36:56,825 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-07 05:36:57,584 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=182340.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:37:23,621 INFO [train.py:901] (3/4) Epoch 23, batch 4550, loss[loss=0.2331, simple_loss=0.304, pruned_loss=0.08112, over 8089.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2878, pruned_loss=0.06192, over 1614385.35 frames. ], batch size: 21, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:37:37,490 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.886e+02 2.347e+02 2.810e+02 3.651e+02 9.685e+02, threshold=5.619e+02, percent-clipped=2.0 +2023-02-07 05:37:38,401 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0520, 3.7536, 2.1400, 2.8572, 2.7662, 1.6881, 2.8486, 3.1920], + device='cuda:3'), covar=tensor([0.1837, 0.0373, 0.1368, 0.0793, 0.0817, 0.1922, 0.1222, 0.0995], + device='cuda:3'), in_proj_covar=tensor([0.0357, 0.0234, 0.0338, 0.0311, 0.0299, 0.0340, 0.0346, 0.0320], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 05:37:59,265 INFO [train.py:901] (3/4) Epoch 23, batch 4600, loss[loss=0.1837, simple_loss=0.2727, pruned_loss=0.04734, over 8094.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2877, pruned_loss=0.06179, over 1610881.52 frames. ], batch size: 21, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:38:03,826 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2241, 1.8924, 2.5916, 1.5918, 1.6029, 2.5342, 1.2400, 2.0375], + device='cuda:3'), covar=tensor([0.1586, 0.1107, 0.0278, 0.1238, 0.2016, 0.0359, 0.1446, 0.1069], + device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0200, 0.0129, 0.0220, 0.0270, 0.0137, 0.0169, 0.0192], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 05:38:20,367 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=182455.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:38:34,855 INFO [train.py:901] (3/4) Epoch 23, batch 4650, loss[loss=0.2014, simple_loss=0.2923, pruned_loss=0.05524, over 8548.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2873, pruned_loss=0.06153, over 1614246.17 frames. ], batch size: 39, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:38:44,332 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7305, 1.5792, 1.8083, 1.5883, 1.3127, 1.5363, 2.3795, 1.7230], + device='cuda:3'), covar=tensor([0.0447, 0.1249, 0.1681, 0.1411, 0.0558, 0.1500, 0.0561, 0.0667], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0153, 0.0190, 0.0160, 0.0101, 0.0163, 0.0112, 0.0144], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 05:38:45,815 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8022, 1.7299, 3.9640, 1.4635, 3.5091, 3.2962, 3.6103, 3.4973], + device='cuda:3'), covar=tensor([0.0699, 0.3961, 0.0605, 0.4261, 0.1137, 0.0985, 0.0642, 0.0754], + device='cuda:3'), in_proj_covar=tensor([0.0639, 0.0646, 0.0702, 0.0634, 0.0712, 0.0609, 0.0609, 0.0682], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 05:38:50,619 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.204e+02 2.647e+02 3.638e+02 6.712e+02, threshold=5.294e+02, percent-clipped=7.0 +2023-02-07 05:39:05,479 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7979, 2.0061, 1.8014, 2.6283, 1.0558, 1.6061, 1.8085, 2.0599], + device='cuda:3'), covar=tensor([0.0777, 0.0776, 0.0910, 0.0366, 0.1169, 0.1308, 0.0877, 0.0843], + device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0199, 0.0247, 0.0216, 0.0208, 0.0249, 0.0252, 0.0209], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 05:39:12,442 INFO [train.py:901] (3/4) Epoch 23, batch 4700, loss[loss=0.1803, simple_loss=0.2769, pruned_loss=0.0419, over 8127.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2866, pruned_loss=0.06125, over 1614370.46 frames. ], batch size: 22, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:39:17,394 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=182532.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:39:47,034 INFO [train.py:901] (3/4) Epoch 23, batch 4750, loss[loss=0.1982, simple_loss=0.2904, pruned_loss=0.05299, over 8145.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2851, pruned_loss=0.05999, over 1617802.11 frames. ], batch size: 22, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:40:01,607 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.297e+02 2.902e+02 3.418e+02 7.225e+02, threshold=5.805e+02, percent-clipped=3.0 +2023-02-07 05:40:05,943 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-07 05:40:08,827 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-07 05:40:11,797 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.7276, 1.4960, 4.9266, 1.8532, 4.4532, 4.0633, 4.4198, 4.3639], + device='cuda:3'), covar=tensor([0.0551, 0.4627, 0.0390, 0.3969, 0.0861, 0.0856, 0.0554, 0.0551], + device='cuda:3'), in_proj_covar=tensor([0.0642, 0.0648, 0.0704, 0.0635, 0.0714, 0.0611, 0.0612, 0.0684], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 05:40:24,101 INFO [train.py:901] (3/4) Epoch 23, batch 4800, loss[loss=0.2509, simple_loss=0.3202, pruned_loss=0.09082, over 8459.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2857, pruned_loss=0.06047, over 1610826.98 frames. ], batch size: 29, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:40:36,436 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=182643.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:40:59,177 INFO [train.py:901] (3/4) Epoch 23, batch 4850, loss[loss=0.1801, simple_loss=0.2556, pruned_loss=0.05231, over 7691.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2849, pruned_loss=0.06017, over 1610846.35 frames. ], batch size: 18, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:41:00,611 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-07 05:41:13,254 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.421e+02 2.288e+02 2.781e+02 3.814e+02 7.165e+02, threshold=5.562e+02, percent-clipped=4.0 +2023-02-07 05:41:25,712 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=182711.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:41:36,258 INFO [train.py:901] (3/4) Epoch 23, batch 4900, loss[loss=0.1923, simple_loss=0.2714, pruned_loss=0.05663, over 8092.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2848, pruned_loss=0.06009, over 1610071.30 frames. ], batch size: 21, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:41:44,994 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=182736.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:42:00,363 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=182758.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:42:12,966 INFO [train.py:901] (3/4) Epoch 23, batch 4950, loss[loss=0.1942, simple_loss=0.2902, pruned_loss=0.04907, over 8575.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2847, pruned_loss=0.05999, over 1611985.83 frames. ], batch size: 39, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:42:27,042 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.635e+02 2.423e+02 2.989e+02 3.745e+02 1.524e+03, threshold=5.977e+02, percent-clipped=7.0 +2023-02-07 05:42:48,225 INFO [train.py:901] (3/4) Epoch 23, batch 5000, loss[loss=0.214, simple_loss=0.3011, pruned_loss=0.06341, over 8438.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2856, pruned_loss=0.06009, over 1614341.99 frames. ], batch size: 29, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:43:25,232 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=182874.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:43:25,719 INFO [train.py:901] (3/4) Epoch 23, batch 5050, loss[loss=0.214, simple_loss=0.2854, pruned_loss=0.07129, over 7972.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2852, pruned_loss=0.05989, over 1610355.09 frames. ], batch size: 21, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:43:26,556 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=182876.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:43:29,545 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=182880.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:43:40,723 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.596e+02 2.376e+02 2.932e+02 3.646e+02 6.966e+02, threshold=5.864e+02, percent-clipped=3.0 +2023-02-07 05:43:46,344 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-07 05:44:01,707 INFO [train.py:901] (3/4) Epoch 23, batch 5100, loss[loss=0.1904, simple_loss=0.2606, pruned_loss=0.06008, over 7521.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2851, pruned_loss=0.05972, over 1612374.74 frames. ], batch size: 18, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:44:03,884 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0574, 1.4890, 1.6605, 1.4264, 0.9404, 1.4381, 1.7609, 1.5765], + device='cuda:3'), covar=tensor([0.0572, 0.1267, 0.1696, 0.1429, 0.0622, 0.1483, 0.0708, 0.0651], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0153, 0.0190, 0.0160, 0.0101, 0.0163, 0.0112, 0.0144], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 05:44:38,860 INFO [train.py:901] (3/4) Epoch 23, batch 5150, loss[loss=0.2174, simple_loss=0.3056, pruned_loss=0.06457, over 8518.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2862, pruned_loss=0.06031, over 1612059.81 frames. ], batch size: 26, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:44:50,218 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=182991.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:44:53,605 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.308e+02 2.409e+02 2.843e+02 3.449e+02 6.604e+02, threshold=5.686e+02, percent-clipped=1.0 +2023-02-07 05:45:07,250 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183014.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:45:09,905 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7970, 1.6668, 1.9717, 1.7603, 1.0982, 1.7096, 2.3219, 2.2980], + device='cuda:3'), covar=tensor([0.0466, 0.1221, 0.1633, 0.1349, 0.0600, 0.1393, 0.0576, 0.0544], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0153, 0.0189, 0.0160, 0.0101, 0.0162, 0.0112, 0.0143], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0007], + device='cuda:3') +2023-02-07 05:45:14,591 INFO [train.py:901] (3/4) Epoch 23, batch 5200, loss[loss=0.1758, simple_loss=0.2464, pruned_loss=0.05263, over 7708.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2864, pruned_loss=0.06052, over 1612350.14 frames. ], batch size: 18, lr: 3.28e-03, grad_scale: 8.0 +2023-02-07 05:45:24,446 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183039.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:45:46,578 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-07 05:45:50,597 INFO [train.py:901] (3/4) Epoch 23, batch 5250, loss[loss=0.233, simple_loss=0.3154, pruned_loss=0.07533, over 8639.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2861, pruned_loss=0.06076, over 1610956.15 frames. ], batch size: 31, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:45:55,149 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-07 05:46:05,166 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.705e+02 2.492e+02 2.942e+02 3.798e+02 7.403e+02, threshold=5.885e+02, percent-clipped=3.0 +2023-02-07 05:46:27,067 INFO [train.py:901] (3/4) Epoch 23, batch 5300, loss[loss=0.2056, simple_loss=0.2745, pruned_loss=0.06834, over 7529.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2859, pruned_loss=0.06136, over 1610093.88 frames. ], batch size: 18, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:46:29,212 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3579, 1.4414, 1.3981, 1.7547, 0.7789, 1.2575, 1.3240, 1.4655], + device='cuda:3'), covar=tensor([0.0923, 0.0880, 0.0997, 0.0510, 0.1039, 0.1383, 0.0744, 0.0763], + device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0196, 0.0243, 0.0213, 0.0205, 0.0244, 0.0249, 0.0205], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 05:46:32,528 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-07 05:46:59,015 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.55 vs. limit=5.0 +2023-02-07 05:47:02,893 INFO [train.py:901] (3/4) Epoch 23, batch 5350, loss[loss=0.2, simple_loss=0.2911, pruned_loss=0.05449, over 8656.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2861, pruned_loss=0.06146, over 1607801.34 frames. ], batch size: 39, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:47:17,714 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.584e+02 2.502e+02 3.193e+02 3.793e+02 7.809e+02, threshold=6.385e+02, percent-clipped=1.0 +2023-02-07 05:47:34,672 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=183218.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:47:38,864 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=183224.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:47:39,488 INFO [train.py:901] (3/4) Epoch 23, batch 5400, loss[loss=0.1857, simple_loss=0.2693, pruned_loss=0.05107, over 7966.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2867, pruned_loss=0.06138, over 1612742.10 frames. ], batch size: 21, lr: 3.27e-03, grad_scale: 4.0 +2023-02-07 05:47:55,900 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183247.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:48:13,031 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183272.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:48:14,972 INFO [train.py:901] (3/4) Epoch 23, batch 5450, loss[loss=0.2312, simple_loss=0.3091, pruned_loss=0.07662, over 8670.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2858, pruned_loss=0.0615, over 1608147.94 frames. ], batch size: 34, lr: 3.27e-03, grad_scale: 4.0 +2023-02-07 05:48:22,247 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6519, 2.3439, 3.1093, 2.5912, 3.0844, 2.5380, 2.4070, 1.9372], + device='cuda:3'), covar=tensor([0.5418, 0.5090, 0.2232, 0.3693, 0.2606, 0.3045, 0.1826, 0.5581], + device='cuda:3'), in_proj_covar=tensor([0.0941, 0.0986, 0.0806, 0.0947, 0.0995, 0.0898, 0.0751, 0.0829], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 05:48:30,414 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.518e+02 2.334e+02 2.819e+02 3.622e+02 6.725e+02, threshold=5.637e+02, percent-clipped=1.0 +2023-02-07 05:48:41,154 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-07 05:48:52,618 INFO [train.py:901] (3/4) Epoch 23, batch 5500, loss[loss=0.1833, simple_loss=0.2556, pruned_loss=0.05547, over 7541.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2857, pruned_loss=0.06106, over 1612197.04 frames. ], batch size: 18, lr: 3.27e-03, grad_scale: 4.0 +2023-02-07 05:48:58,324 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=183333.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:49:02,368 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=183339.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:49:14,868 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 +2023-02-07 05:49:27,074 INFO [train.py:901] (3/4) Epoch 23, batch 5550, loss[loss=0.1841, simple_loss=0.2594, pruned_loss=0.05439, over 7538.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.285, pruned_loss=0.06083, over 1607505.79 frames. ], batch size: 18, lr: 3.27e-03, grad_scale: 4.0 +2023-02-07 05:49:41,517 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.428e+02 3.119e+02 4.010e+02 1.058e+03, threshold=6.238e+02, percent-clipped=9.0 +2023-02-07 05:50:03,271 INFO [train.py:901] (3/4) Epoch 23, batch 5600, loss[loss=0.182, simple_loss=0.2641, pruned_loss=0.04993, over 7643.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2859, pruned_loss=0.06106, over 1609766.35 frames. ], batch size: 19, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:50:04,085 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=183426.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:50:12,352 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.7803, 1.5949, 3.9934, 1.4160, 3.4962, 3.3153, 3.6226, 3.5116], + device='cuda:3'), covar=tensor([0.0773, 0.4314, 0.0599, 0.4115, 0.1210, 0.1029, 0.0708, 0.0817], + device='cuda:3'), in_proj_covar=tensor([0.0643, 0.0650, 0.0704, 0.0636, 0.0713, 0.0613, 0.0611, 0.0686], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 05:50:39,038 INFO [train.py:901] (3/4) Epoch 23, batch 5650, loss[loss=0.213, simple_loss=0.3015, pruned_loss=0.06227, over 8282.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2868, pruned_loss=0.06159, over 1610818.60 frames. ], batch size: 23, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:50:51,426 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-07 05:50:53,308 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.655e+02 2.300e+02 3.084e+02 3.921e+02 7.530e+02, threshold=6.168e+02, percent-clipped=4.0 +2023-02-07 05:51:14,122 INFO [train.py:901] (3/4) Epoch 23, batch 5700, loss[loss=0.2198, simple_loss=0.3017, pruned_loss=0.0689, over 8581.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2868, pruned_loss=0.06154, over 1610026.82 frames. ], batch size: 34, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:51:50,332 INFO [train.py:901] (3/4) Epoch 23, batch 5750, loss[loss=0.1435, simple_loss=0.2281, pruned_loss=0.02945, over 7647.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2878, pruned_loss=0.06148, over 1612161.25 frames. ], batch size: 19, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:51:57,139 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-07 05:52:00,900 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183589.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:52:01,866 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.16 vs. limit=5.0 +2023-02-07 05:52:05,032 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183595.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:52:05,458 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.560e+02 2.334e+02 3.030e+02 3.740e+02 1.347e+03, threshold=6.060e+02, percent-clipped=7.0 +2023-02-07 05:52:18,049 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183614.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:52:22,097 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183620.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:52:25,347 INFO [train.py:901] (3/4) Epoch 23, batch 5800, loss[loss=0.2061, simple_loss=0.2953, pruned_loss=0.05843, over 8721.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.287, pruned_loss=0.06079, over 1614548.57 frames. ], batch size: 34, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:52:30,175 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=183632.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:52:51,797 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.7035, 2.3068, 4.8760, 2.8343, 4.4542, 4.1932, 4.5333, 4.4107], + device='cuda:3'), covar=tensor([0.0627, 0.3610, 0.0470, 0.3124, 0.0898, 0.0848, 0.0512, 0.0572], + device='cuda:3'), in_proj_covar=tensor([0.0642, 0.0650, 0.0704, 0.0636, 0.0713, 0.0614, 0.0610, 0.0689], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 05:53:01,082 INFO [train.py:901] (3/4) Epoch 23, batch 5850, loss[loss=0.1814, simple_loss=0.2698, pruned_loss=0.04656, over 7935.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2849, pruned_loss=0.05986, over 1611856.32 frames. ], batch size: 20, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:53:16,205 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.232e+02 2.367e+02 2.798e+02 3.640e+02 5.951e+02, threshold=5.597e+02, percent-clipped=0.0 +2023-02-07 05:53:36,781 INFO [train.py:901] (3/4) Epoch 23, batch 5900, loss[loss=0.1886, simple_loss=0.2794, pruned_loss=0.0489, over 7964.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.284, pruned_loss=0.05962, over 1602167.71 frames. ], batch size: 21, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:53:38,965 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6086, 1.3061, 2.8254, 1.3247, 2.1269, 3.0399, 3.2209, 2.5875], + device='cuda:3'), covar=tensor([0.1194, 0.1746, 0.0395, 0.2159, 0.0904, 0.0314, 0.0549, 0.0591], + device='cuda:3'), in_proj_covar=tensor([0.0294, 0.0319, 0.0285, 0.0313, 0.0311, 0.0267, 0.0422, 0.0302], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 05:53:39,003 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7191, 1.9354, 1.6105, 2.3468, 1.0734, 1.4166, 1.7628, 1.9095], + device='cuda:3'), covar=tensor([0.0767, 0.0698, 0.0884, 0.0356, 0.1054, 0.1358, 0.0709, 0.0641], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0197, 0.0245, 0.0214, 0.0206, 0.0246, 0.0250, 0.0206], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 05:53:59,045 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-07 05:54:08,430 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=183770.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:54:11,755 INFO [train.py:901] (3/4) Epoch 23, batch 5950, loss[loss=0.24, simple_loss=0.3304, pruned_loss=0.07479, over 8706.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2856, pruned_loss=0.06061, over 1602048.80 frames. ], batch size: 49, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:54:23,702 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.2112, 4.1800, 3.8054, 2.0811, 3.7159, 3.6937, 3.7109, 3.5530], + device='cuda:3'), covar=tensor([0.0724, 0.0523, 0.0890, 0.3946, 0.0843, 0.0962, 0.1229, 0.0797], + device='cuda:3'), in_proj_covar=tensor([0.0524, 0.0441, 0.0426, 0.0536, 0.0427, 0.0443, 0.0427, 0.0384], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 05:54:27,016 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.622e+02 2.423e+02 2.784e+02 3.423e+02 5.836e+02, threshold=5.567e+02, percent-clipped=2.0 +2023-02-07 05:54:38,318 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0842, 1.4015, 4.2707, 1.7780, 2.5365, 4.8507, 4.9552, 4.1259], + device='cuda:3'), covar=tensor([0.1269, 0.2048, 0.0303, 0.2142, 0.1144, 0.0190, 0.0476, 0.0545], + device='cuda:3'), in_proj_covar=tensor([0.0293, 0.0317, 0.0283, 0.0311, 0.0308, 0.0265, 0.0420, 0.0300], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 05:54:43,514 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2536, 3.2253, 3.0242, 1.5917, 2.9427, 2.8704, 2.9430, 2.7922], + device='cuda:3'), covar=tensor([0.1155, 0.0862, 0.1228, 0.4410, 0.1070, 0.1402, 0.1563, 0.1060], + device='cuda:3'), in_proj_covar=tensor([0.0526, 0.0443, 0.0428, 0.0539, 0.0429, 0.0445, 0.0428, 0.0385], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 05:54:47,617 INFO [train.py:901] (3/4) Epoch 23, batch 6000, loss[loss=0.2147, simple_loss=0.2973, pruned_loss=0.06603, over 8499.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2852, pruned_loss=0.06074, over 1602448.69 frames. ], batch size: 26, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:54:47,618 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 05:55:00,698 INFO [train.py:935] (3/4) Epoch 23, validation: loss=0.1722, simple_loss=0.2724, pruned_loss=0.03597, over 944034.00 frames. +2023-02-07 05:55:00,699 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6747MB +2023-02-07 05:55:25,814 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=183860.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:55:32,453 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-07 05:55:36,137 INFO [train.py:901] (3/4) Epoch 23, batch 6050, loss[loss=0.2108, simple_loss=0.2956, pruned_loss=0.06304, over 8131.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2873, pruned_loss=0.06202, over 1607515.96 frames. ], batch size: 22, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:55:43,230 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=183885.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:55:50,619 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.494e+02 2.465e+02 3.097e+02 3.782e+02 8.398e+02, threshold=6.194e+02, percent-clipped=6.0 +2023-02-07 05:56:02,800 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-07 05:56:11,859 INFO [train.py:901] (3/4) Epoch 23, batch 6100, loss[loss=0.1909, simple_loss=0.2815, pruned_loss=0.05018, over 8111.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2876, pruned_loss=0.06192, over 1608823.97 frames. ], batch size: 23, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:56:32,460 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-07 05:56:47,367 INFO [train.py:901] (3/4) Epoch 23, batch 6150, loss[loss=0.1977, simple_loss=0.2588, pruned_loss=0.06828, over 6813.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2869, pruned_loss=0.061, over 1609979.68 frames. ], batch size: 15, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:56:48,176 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=183976.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:56:51,963 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 +2023-02-07 05:57:01,786 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.620e+02 2.512e+02 2.876e+02 3.577e+02 6.799e+02, threshold=5.752e+02, percent-clipped=2.0 +2023-02-07 05:57:22,985 INFO [train.py:901] (3/4) Epoch 23, batch 6200, loss[loss=0.1828, simple_loss=0.264, pruned_loss=0.05076, over 7801.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2864, pruned_loss=0.06051, over 1611766.85 frames. ], batch size: 20, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:57:25,507 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1148, 1.8632, 2.3998, 2.0772, 2.2928, 2.1883, 1.9644, 1.2464], + device='cuda:3'), covar=tensor([0.5283, 0.4614, 0.1864, 0.3345, 0.2317, 0.2954, 0.1836, 0.4819], + device='cuda:3'), in_proj_covar=tensor([0.0947, 0.0992, 0.0809, 0.0953, 0.0999, 0.0900, 0.0752, 0.0830], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 05:57:54,995 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184068.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:57:59,607 INFO [train.py:901] (3/4) Epoch 23, batch 6250, loss[loss=0.1965, simple_loss=0.2632, pruned_loss=0.06489, over 7435.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2872, pruned_loss=0.06111, over 1611728.10 frames. ], batch size: 17, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:58:08,019 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7507, 1.9434, 1.6940, 2.2926, 0.9526, 1.5661, 1.7501, 1.8797], + device='cuda:3'), covar=tensor([0.0765, 0.0699, 0.0885, 0.0430, 0.1097, 0.1311, 0.0728, 0.0749], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0198, 0.0246, 0.0215, 0.0207, 0.0247, 0.0250, 0.0207], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 05:58:11,446 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184091.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:58:14,643 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.476e+02 2.340e+02 2.866e+02 3.425e+02 5.984e+02, threshold=5.731e+02, percent-clipped=3.0 +2023-02-07 05:58:34,481 INFO [train.py:901] (3/4) Epoch 23, batch 6300, loss[loss=0.2206, simple_loss=0.3023, pruned_loss=0.0695, over 8438.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2866, pruned_loss=0.06095, over 1612177.23 frames. ], batch size: 29, lr: 3.27e-03, grad_scale: 8.0 +2023-02-07 05:58:45,793 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184141.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:59:04,561 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184166.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:59:06,581 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0589, 1.5771, 1.3631, 1.7085, 1.4114, 1.2313, 1.4128, 1.3976], + device='cuda:3'), covar=tensor([0.1122, 0.0492, 0.1472, 0.0486, 0.0767, 0.1681, 0.0846, 0.0702], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0231, 0.0331, 0.0305, 0.0297, 0.0337, 0.0342, 0.0314], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 05:59:10,600 INFO [train.py:901] (3/4) Epoch 23, batch 6350, loss[loss=0.1965, simple_loss=0.2729, pruned_loss=0.06002, over 7801.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2855, pruned_loss=0.0604, over 1611780.17 frames. ], batch size: 19, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 05:59:12,233 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0849, 3.7741, 2.3834, 2.9905, 2.9019, 2.2976, 2.8346, 3.2360], + device='cuda:3'), covar=tensor([0.1575, 0.0305, 0.1094, 0.0725, 0.0665, 0.1279, 0.1004, 0.0894], + device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0232, 0.0331, 0.0305, 0.0297, 0.0337, 0.0342, 0.0315], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 05:59:25,790 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.630e+02 2.298e+02 2.703e+02 3.593e+02 9.198e+02, threshold=5.406e+02, percent-clipped=6.0 +2023-02-07 05:59:32,345 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=184204.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 05:59:46,838 INFO [train.py:901] (3/4) Epoch 23, batch 6400, loss[loss=0.1939, simple_loss=0.2553, pruned_loss=0.06623, over 7714.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2852, pruned_loss=0.06019, over 1612650.66 frames. ], batch size: 18, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:00:22,069 INFO [train.py:901] (3/4) Epoch 23, batch 6450, loss[loss=0.18, simple_loss=0.269, pruned_loss=0.04547, over 8336.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2858, pruned_loss=0.0607, over 1611877.26 frames. ], batch size: 25, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:00:37,219 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.568e+02 2.429e+02 3.055e+02 3.904e+02 7.071e+02, threshold=6.109e+02, percent-clipped=5.0 +2023-02-07 06:00:53,120 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1750, 1.6743, 4.5154, 2.0954, 2.6170, 5.1092, 5.1645, 4.4289], + device='cuda:3'), covar=tensor([0.1181, 0.1824, 0.0244, 0.1923, 0.1102, 0.0176, 0.0308, 0.0503], + device='cuda:3'), in_proj_covar=tensor([0.0295, 0.0319, 0.0285, 0.0314, 0.0311, 0.0267, 0.0423, 0.0303], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 06:00:54,551 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184319.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:00:58,465 INFO [train.py:901] (3/4) Epoch 23, batch 6500, loss[loss=0.1825, simple_loss=0.2624, pruned_loss=0.05134, over 8085.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2847, pruned_loss=0.05987, over 1609096.99 frames. ], batch size: 21, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:01:13,657 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184347.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:01:30,749 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184372.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:01:32,667 INFO [train.py:901] (3/4) Epoch 23, batch 6550, loss[loss=0.2059, simple_loss=0.2917, pruned_loss=0.06002, over 8604.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2857, pruned_loss=0.06055, over 1609253.24 frames. ], batch size: 34, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:01:34,223 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184377.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:01:48,099 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.553e+02 2.251e+02 2.720e+02 3.518e+02 7.175e+02, threshold=5.440e+02, percent-clipped=6.0 +2023-02-07 06:01:51,595 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-07 06:02:00,103 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=184412.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:02:09,823 INFO [train.py:901] (3/4) Epoch 23, batch 6600, loss[loss=0.2065, simple_loss=0.2934, pruned_loss=0.05982, over 8328.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.285, pruned_loss=0.06041, over 1606318.66 frames. ], batch size: 25, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:02:09,838 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-07 06:02:45,212 INFO [train.py:901] (3/4) Epoch 23, batch 6650, loss[loss=0.1807, simple_loss=0.2568, pruned_loss=0.05231, over 7679.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2843, pruned_loss=0.05998, over 1609751.81 frames. ], batch size: 18, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:03:00,393 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.565e+02 2.187e+02 2.636e+02 3.150e+02 7.164e+02, threshold=5.273e+02, percent-clipped=1.0 +2023-02-07 06:03:06,034 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184504.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:03:14,722 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-07 06:03:21,193 INFO [train.py:901] (3/4) Epoch 23, batch 6700, loss[loss=0.1788, simple_loss=0.2651, pruned_loss=0.04621, over 7716.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2834, pruned_loss=0.05963, over 1610759.34 frames. ], batch size: 18, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:03:22,776 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184527.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:03:32,199 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.10 vs. limit=5.0 +2023-02-07 06:03:56,976 INFO [train.py:901] (3/4) Epoch 23, batch 6750, loss[loss=0.184, simple_loss=0.264, pruned_loss=0.05195, over 8077.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2846, pruned_loss=0.06014, over 1609531.70 frames. ], batch size: 21, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:03:57,226 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184575.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:04:02,293 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 +2023-02-07 06:04:11,516 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.515e+02 2.299e+02 2.705e+02 3.689e+02 1.087e+03, threshold=5.410e+02, percent-clipped=6.0 +2023-02-07 06:04:14,465 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184600.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:04:30,945 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-07 06:04:32,276 INFO [train.py:901] (3/4) Epoch 23, batch 6800, loss[loss=0.2388, simple_loss=0.3241, pruned_loss=0.07677, over 8457.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2862, pruned_loss=0.06129, over 1610480.01 frames. ], batch size: 27, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:04:46,626 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.48 vs. limit=2.0 +2023-02-07 06:05:08,933 INFO [train.py:901] (3/4) Epoch 23, batch 6850, loss[loss=0.1583, simple_loss=0.2346, pruned_loss=0.04101, over 7545.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2872, pruned_loss=0.0617, over 1609606.39 frames. ], batch size: 18, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:05:14,193 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2901, 2.0580, 2.6517, 2.1868, 2.6346, 2.2875, 2.1304, 1.4172], + device='cuda:3'), covar=tensor([0.5292, 0.4780, 0.2038, 0.3959, 0.2541, 0.3068, 0.2039, 0.5270], + device='cuda:3'), in_proj_covar=tensor([0.0943, 0.0990, 0.0807, 0.0950, 0.0997, 0.0898, 0.0752, 0.0828], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 06:05:19,335 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-07 06:05:23,525 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.780e+02 2.645e+02 3.100e+02 4.179e+02 7.238e+02, threshold=6.201e+02, percent-clipped=8.0 +2023-02-07 06:05:40,715 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=184721.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:05:43,468 INFO [train.py:901] (3/4) Epoch 23, batch 6900, loss[loss=0.2362, simple_loss=0.3237, pruned_loss=0.07435, over 8614.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2877, pruned_loss=0.06152, over 1612955.68 frames. ], batch size: 31, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:06:21,332 INFO [train.py:901] (3/4) Epoch 23, batch 6950, loss[loss=0.1962, simple_loss=0.2745, pruned_loss=0.05893, over 7917.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2882, pruned_loss=0.06218, over 1610329.06 frames. ], batch size: 20, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:06:27,129 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184783.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:06:29,047 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-07 06:06:35,999 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.448e+02 2.334e+02 2.863e+02 3.573e+02 6.345e+02, threshold=5.727e+02, percent-clipped=1.0 +2023-02-07 06:06:38,088 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.9184, 3.8292, 3.5237, 1.8861, 3.4632, 3.4442, 3.3920, 3.3605], + device='cuda:3'), covar=tensor([0.0846, 0.0660, 0.1174, 0.4372, 0.1016, 0.1137, 0.1615, 0.0843], + device='cuda:3'), in_proj_covar=tensor([0.0529, 0.0447, 0.0431, 0.0544, 0.0434, 0.0448, 0.0430, 0.0389], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 06:06:44,106 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.24 vs. limit=5.0 +2023-02-07 06:06:44,532 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184808.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:06:56,221 INFO [train.py:901] (3/4) Epoch 23, batch 7000, loss[loss=0.2316, simple_loss=0.3174, pruned_loss=0.07293, over 8660.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2869, pruned_loss=0.06164, over 1606528.40 frames. ], batch size: 34, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:07:03,960 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184836.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:07:12,028 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=184848.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:07:32,116 INFO [train.py:901] (3/4) Epoch 23, batch 7050, loss[loss=0.2077, simple_loss=0.2942, pruned_loss=0.06061, over 8085.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2852, pruned_loss=0.06074, over 1601730.12 frames. ], batch size: 21, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:07:38,586 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184884.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:07:48,063 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.510e+02 2.247e+02 2.854e+02 3.580e+02 1.056e+03, threshold=5.709e+02, percent-clipped=4.0 +2023-02-07 06:07:53,395 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.42 vs. limit=5.0 +2023-02-07 06:08:08,157 INFO [train.py:901] (3/4) Epoch 23, batch 7100, loss[loss=0.1797, simple_loss=0.2537, pruned_loss=0.05281, over 7699.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2845, pruned_loss=0.06007, over 1601052.34 frames. ], batch size: 18, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:08:30,175 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184957.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:08:34,372 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184963.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:08:42,468 INFO [train.py:901] (3/4) Epoch 23, batch 7150, loss[loss=0.2181, simple_loss=0.3073, pruned_loss=0.0644, over 8319.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2848, pruned_loss=0.05993, over 1604542.79 frames. ], batch size: 25, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:08:53,468 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9672, 1.6897, 1.9956, 1.8309, 1.9866, 2.0483, 1.8517, 0.8516], + device='cuda:3'), covar=tensor([0.5659, 0.4597, 0.2041, 0.3398, 0.2368, 0.2837, 0.1907, 0.4907], + device='cuda:3'), in_proj_covar=tensor([0.0942, 0.0991, 0.0806, 0.0950, 0.0997, 0.0896, 0.0753, 0.0829], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 06:08:58,831 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.534e+02 2.323e+02 2.664e+02 3.243e+02 7.163e+02, threshold=5.329e+02, percent-clipped=2.0 +2023-02-07 06:09:20,383 INFO [train.py:901] (3/4) Epoch 23, batch 7200, loss[loss=0.228, simple_loss=0.3014, pruned_loss=0.0773, over 8138.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2856, pruned_loss=0.06068, over 1605169.16 frames. ], batch size: 22, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:09:29,457 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9016, 1.5262, 1.6953, 1.3935, 0.8318, 1.4837, 1.6749, 1.4970], + device='cuda:3'), covar=tensor([0.0543, 0.1232, 0.1731, 0.1469, 0.0601, 0.1488, 0.0688, 0.0673], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0153, 0.0190, 0.0161, 0.0102, 0.0164, 0.0112, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 06:09:35,034 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9658, 1.3185, 1.5239, 1.2179, 0.9491, 1.3849, 1.7648, 1.4913], + device='cuda:3'), covar=tensor([0.0556, 0.1600, 0.2321, 0.1872, 0.0679, 0.1940, 0.0732, 0.0719], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0153, 0.0190, 0.0160, 0.0102, 0.0164, 0.0112, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 06:09:35,779 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7736, 2.2158, 3.5612, 1.9484, 1.6405, 3.4408, 0.5730, 2.1481], + device='cuda:3'), covar=tensor([0.1477, 0.1291, 0.0245, 0.1589, 0.2771, 0.0342, 0.2357, 0.1363], + device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0195, 0.0129, 0.0219, 0.0268, 0.0136, 0.0168, 0.0191], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 06:09:44,873 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6926, 1.7458, 2.0414, 1.7287, 1.0271, 1.7077, 2.1028, 2.0191], + device='cuda:3'), covar=tensor([0.0449, 0.1212, 0.1572, 0.1354, 0.0615, 0.1463, 0.0661, 0.0605], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0153, 0.0190, 0.0160, 0.0101, 0.0164, 0.0112, 0.0144], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 06:09:44,913 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8353, 2.2703, 3.6969, 1.7680, 1.5870, 3.6889, 0.5202, 2.1583], + device='cuda:3'), covar=tensor([0.1346, 0.1302, 0.0192, 0.1925, 0.2904, 0.0216, 0.2464, 0.1406], + device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0196, 0.0129, 0.0220, 0.0268, 0.0136, 0.0169, 0.0191], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 06:09:54,644 INFO [train.py:901] (3/4) Epoch 23, batch 7250, loss[loss=0.2372, simple_loss=0.3158, pruned_loss=0.07933, over 7525.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2862, pruned_loss=0.06101, over 1606224.86 frames. ], batch size: 18, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:09:56,898 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185078.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:10:06,367 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185092.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:10:09,714 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.492e+02 2.341e+02 2.701e+02 3.625e+02 6.528e+02, threshold=5.401e+02, percent-clipped=8.0 +2023-02-07 06:10:25,011 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185117.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:10:30,392 INFO [train.py:901] (3/4) Epoch 23, batch 7300, loss[loss=0.1826, simple_loss=0.272, pruned_loss=0.04655, over 8532.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2857, pruned_loss=0.06082, over 1608002.78 frames. ], batch size: 49, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:10:46,959 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185147.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:11:06,505 INFO [train.py:901] (3/4) Epoch 23, batch 7350, loss[loss=0.1673, simple_loss=0.2495, pruned_loss=0.04259, over 7543.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2854, pruned_loss=0.06053, over 1609529.31 frames. ], batch size: 18, lr: 3.26e-03, grad_scale: 8.0 +2023-02-07 06:11:19,725 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-07 06:11:21,070 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.569e+02 2.380e+02 2.863e+02 3.556e+02 7.708e+02, threshold=5.726e+02, percent-clipped=6.0 +2023-02-07 06:11:38,636 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185219.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:11:41,245 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-07 06:11:42,632 INFO [train.py:901] (3/4) Epoch 23, batch 7400, loss[loss=0.1915, simple_loss=0.2852, pruned_loss=0.04893, over 8036.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2845, pruned_loss=0.06011, over 1607517.64 frames. ], batch size: 22, lr: 3.26e-03, grad_scale: 16.0 +2023-02-07 06:11:44,815 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=185228.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:11:53,922 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9974, 1.3215, 1.6215, 1.3105, 0.9467, 1.4409, 1.7741, 1.7983], + device='cuda:3'), covar=tensor([0.0609, 0.1760, 0.2348, 0.1867, 0.0703, 0.2035, 0.0777, 0.0663], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0153, 0.0190, 0.0159, 0.0101, 0.0163, 0.0112, 0.0144], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 06:11:56,742 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185244.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:12:18,683 INFO [train.py:901] (3/4) Epoch 23, batch 7450, loss[loss=0.1577, simple_loss=0.2412, pruned_loss=0.03713, over 7663.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2838, pruned_loss=0.05991, over 1604692.79 frames. ], batch size: 19, lr: 3.26e-03, grad_scale: 16.0 +2023-02-07 06:12:18,956 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6245, 1.8315, 2.6464, 1.4621, 1.9474, 2.0010, 1.6515, 1.9283], + device='cuda:3'), covar=tensor([0.1878, 0.2480, 0.0817, 0.4504, 0.1906, 0.3099, 0.2287, 0.2208], + device='cuda:3'), in_proj_covar=tensor([0.0526, 0.0611, 0.0554, 0.0646, 0.0647, 0.0595, 0.0541, 0.0632], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 06:12:21,617 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-07 06:12:33,462 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.310e+02 2.954e+02 3.827e+02 6.869e+02, threshold=5.908e+02, percent-clipped=4.0 +2023-02-07 06:12:37,079 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=185301.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:12:53,820 INFO [train.py:901] (3/4) Epoch 23, batch 7500, loss[loss=0.1903, simple_loss=0.2623, pruned_loss=0.0591, over 7451.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2837, pruned_loss=0.05953, over 1607062.74 frames. ], batch size: 17, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:13:05,840 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 +2023-02-07 06:13:08,254 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185343.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:13:22,347 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9594, 2.2554, 1.7334, 2.8490, 1.2551, 1.5424, 2.1004, 2.2161], + device='cuda:3'), covar=tensor([0.0778, 0.0850, 0.0952, 0.0343, 0.1192, 0.1397, 0.0801, 0.0754], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0199, 0.0244, 0.0214, 0.0206, 0.0247, 0.0250, 0.0207], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 06:13:31,751 INFO [train.py:901] (3/4) Epoch 23, batch 7550, loss[loss=0.2206, simple_loss=0.3073, pruned_loss=0.06694, over 8327.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2845, pruned_loss=0.05927, over 1612749.80 frames. ], batch size: 25, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:13:39,221 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.64 vs. limit=2.0 +2023-02-07 06:13:46,072 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.499e+02 2.461e+02 3.059e+02 3.860e+02 7.244e+02, threshold=6.118e+02, percent-clipped=3.0 +2023-02-07 06:14:00,489 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185416.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:14:04,590 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=185422.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:14:06,608 INFO [train.py:901] (3/4) Epoch 23, batch 7600, loss[loss=0.1765, simple_loss=0.2529, pruned_loss=0.05002, over 7706.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2843, pruned_loss=0.05901, over 1612584.02 frames. ], batch size: 18, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:14:09,581 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.84 vs. limit=5.0 +2023-02-07 06:14:37,967 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.8281, 6.0913, 5.2008, 2.6546, 5.3572, 5.6406, 5.4923, 5.5105], + device='cuda:3'), covar=tensor([0.0557, 0.0350, 0.0884, 0.4129, 0.0771, 0.0838, 0.1153, 0.0513], + device='cuda:3'), in_proj_covar=tensor([0.0535, 0.0450, 0.0439, 0.0549, 0.0439, 0.0452, 0.0433, 0.0394], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 06:14:41,963 INFO [train.py:901] (3/4) Epoch 23, batch 7650, loss[loss=0.1802, simple_loss=0.2625, pruned_loss=0.04893, over 7538.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2849, pruned_loss=0.0596, over 1611342.42 frames. ], batch size: 18, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:14:50,200 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185486.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:14:54,420 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=185491.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:14:57,817 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.760e+02 2.478e+02 3.100e+02 3.999e+02 8.387e+02, threshold=6.200e+02, percent-clipped=6.0 +2023-02-07 06:15:17,442 INFO [train.py:901] (3/4) Epoch 23, batch 7700, loss[loss=0.1789, simple_loss=0.2656, pruned_loss=0.04613, over 8141.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2857, pruned_loss=0.06001, over 1614733.12 frames. ], batch size: 22, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:15:25,727 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185537.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:15:37,341 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-07 06:15:53,099 INFO [train.py:901] (3/4) Epoch 23, batch 7750, loss[loss=0.244, simple_loss=0.3317, pruned_loss=0.07814, over 8761.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2854, pruned_loss=0.05961, over 1615249.39 frames. ], batch size: 30, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:15:56,275 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-07 06:16:08,175 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.519e+02 2.336e+02 2.905e+02 3.607e+02 6.527e+02, threshold=5.810e+02, percent-clipped=2.0 +2023-02-07 06:16:10,499 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185599.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:16:11,963 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2563, 2.0032, 2.6598, 2.1471, 2.5204, 2.2771, 2.0983, 1.4063], + device='cuda:3'), covar=tensor([0.5486, 0.5102, 0.1980, 0.3651, 0.2680, 0.3085, 0.1938, 0.5237], + device='cuda:3'), in_proj_covar=tensor([0.0940, 0.0982, 0.0803, 0.0949, 0.0991, 0.0896, 0.0750, 0.0824], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 06:16:15,294 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185606.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:16:28,342 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185624.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:16:28,831 INFO [train.py:901] (3/4) Epoch 23, batch 7800, loss[loss=0.1907, simple_loss=0.2866, pruned_loss=0.04737, over 8324.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2865, pruned_loss=0.05985, over 1620062.39 frames. ], batch size: 25, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:17:01,069 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185672.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:17:02,902 INFO [train.py:901] (3/4) Epoch 23, batch 7850, loss[loss=0.233, simple_loss=0.3051, pruned_loss=0.08045, over 6441.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2859, pruned_loss=0.05972, over 1614435.62 frames. ], batch size: 72, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:17:17,283 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.434e+02 2.983e+02 3.607e+02 9.941e+02, threshold=5.966e+02, percent-clipped=5.0 +2023-02-07 06:17:18,193 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185697.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:17:28,603 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7898, 2.3479, 4.2244, 1.6817, 3.1637, 2.3884, 1.9724, 3.1379], + device='cuda:3'), covar=tensor([0.1841, 0.2889, 0.0669, 0.4649, 0.1740, 0.3044, 0.2354, 0.2232], + device='cuda:3'), in_proj_covar=tensor([0.0523, 0.0609, 0.0551, 0.0643, 0.0646, 0.0591, 0.0539, 0.0630], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 06:17:37,217 INFO [train.py:901] (3/4) Epoch 23, batch 7900, loss[loss=0.173, simple_loss=0.2549, pruned_loss=0.0455, over 8032.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2851, pruned_loss=0.05952, over 1612913.44 frames. ], batch size: 22, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:18:11,080 INFO [train.py:901] (3/4) Epoch 23, batch 7950, loss[loss=0.1918, simple_loss=0.2827, pruned_loss=0.05046, over 8594.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2853, pruned_loss=0.05944, over 1616134.55 frames. ], batch size: 31, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:18:12,603 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185777.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:18:23,314 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185793.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:18:25,077 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.419e+02 2.254e+02 2.775e+02 3.427e+02 8.244e+02, threshold=5.550e+02, percent-clipped=2.0 +2023-02-07 06:18:39,402 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185817.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 06:18:40,110 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185818.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:18:44,665 INFO [train.py:901] (3/4) Epoch 23, batch 8000, loss[loss=0.2372, simple_loss=0.3146, pruned_loss=0.07985, over 8029.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2848, pruned_loss=0.05961, over 1611788.61 frames. ], batch size: 22, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:18:48,036 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=185830.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:19:09,886 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185862.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:19:15,915 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.03 vs. limit=5.0 +2023-02-07 06:19:17,209 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-07 06:19:18,260 INFO [train.py:901] (3/4) Epoch 23, batch 8050, loss[loss=0.2826, simple_loss=0.3309, pruned_loss=0.1172, over 6741.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2845, pruned_loss=0.06059, over 1590295.39 frames. ], batch size: 71, lr: 3.25e-03, grad_scale: 16.0 +2023-02-07 06:19:26,781 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185887.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:19:32,791 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.537e+02 2.662e+02 3.318e+02 4.159e+02 9.358e+02, threshold=6.635e+02, percent-clipped=7.0 +2023-02-07 06:19:51,668 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-07 06:19:57,763 INFO [train.py:901] (3/4) Epoch 24, batch 0, loss[loss=0.2454, simple_loss=0.3025, pruned_loss=0.09414, over 7435.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3025, pruned_loss=0.09414, over 7435.00 frames. ], batch size: 17, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:19:57,763 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 06:20:01,674 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6368, 1.4279, 1.6905, 1.4116, 0.9696, 1.4481, 1.6668, 1.3033], + device='cuda:3'), covar=tensor([0.0699, 0.1376, 0.1795, 0.1529, 0.0628, 0.1595, 0.0710, 0.0729], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0153, 0.0190, 0.0160, 0.0101, 0.0164, 0.0112, 0.0144], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 06:20:09,065 INFO [train.py:935] (3/4) Epoch 24, validation: loss=0.1731, simple_loss=0.2733, pruned_loss=0.03644, over 944034.00 frames. +2023-02-07 06:20:09,067 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6747MB +2023-02-07 06:20:23,918 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-07 06:20:35,545 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185945.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:20:44,027 INFO [train.py:901] (3/4) Epoch 24, batch 50, loss[loss=0.2284, simple_loss=0.3115, pruned_loss=0.07264, over 8031.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.297, pruned_loss=0.06633, over 367463.87 frames. ], batch size: 22, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:20:57,553 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-07 06:21:11,401 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.593e+02 2.437e+02 2.851e+02 3.663e+02 1.155e+03, threshold=5.702e+02, percent-clipped=3.0 +2023-02-07 06:21:20,557 INFO [train.py:901] (3/4) Epoch 24, batch 100, loss[loss=0.1777, simple_loss=0.2544, pruned_loss=0.05055, over 7659.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2885, pruned_loss=0.06325, over 640742.45 frames. ], batch size: 19, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:21:22,597 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-07 06:21:41,725 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3743, 1.5325, 4.5862, 1.6966, 4.0837, 3.8445, 4.1886, 4.0775], + device='cuda:3'), covar=tensor([0.0579, 0.4663, 0.0494, 0.4094, 0.1057, 0.0958, 0.0564, 0.0716], + device='cuda:3'), in_proj_covar=tensor([0.0645, 0.0655, 0.0708, 0.0641, 0.0720, 0.0620, 0.0615, 0.0690], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 06:21:56,095 INFO [train.py:901] (3/4) Epoch 24, batch 150, loss[loss=0.1811, simple_loss=0.2634, pruned_loss=0.04941, over 7536.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2854, pruned_loss=0.06035, over 858025.18 frames. ], batch size: 18, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:22:00,573 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.90 vs. limit=5.0 +2023-02-07 06:22:03,342 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 +2023-02-07 06:22:07,734 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7043, 1.9046, 2.1094, 1.3311, 2.2127, 1.5317, 0.6705, 1.8670], + device='cuda:3'), covar=tensor([0.0610, 0.0394, 0.0315, 0.0653, 0.0422, 0.0919, 0.0930, 0.0330], + device='cuda:3'), in_proj_covar=tensor([0.0462, 0.0402, 0.0354, 0.0455, 0.0385, 0.0545, 0.0398, 0.0428], + device='cuda:3'), out_proj_covar=tensor([1.2331e-04, 1.0515e-04, 9.2994e-05, 1.1942e-04, 1.0126e-04, 1.5330e-04, + 1.0710e-04, 1.1294e-04], device='cuda:3') +2023-02-07 06:22:21,896 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.557e+02 2.384e+02 2.880e+02 3.401e+02 7.597e+02, threshold=5.761e+02, percent-clipped=1.0 +2023-02-07 06:22:30,271 INFO [train.py:901] (3/4) Epoch 24, batch 200, loss[loss=0.2057, simple_loss=0.2924, pruned_loss=0.05953, over 8513.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2859, pruned_loss=0.06045, over 1027395.60 frames. ], batch size: 26, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:22:35,409 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9434, 1.5973, 3.5158, 1.6221, 2.3839, 3.8721, 3.9683, 3.3299], + device='cuda:3'), covar=tensor([0.1178, 0.1747, 0.0309, 0.2015, 0.1114, 0.0215, 0.0405, 0.0523], + device='cuda:3'), in_proj_covar=tensor([0.0297, 0.0321, 0.0286, 0.0317, 0.0314, 0.0269, 0.0426, 0.0304], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 06:22:38,155 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186118.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:22:40,060 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186121.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:23:01,369 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-07 06:23:05,572 INFO [train.py:901] (3/4) Epoch 24, batch 250, loss[loss=0.1945, simple_loss=0.2769, pruned_loss=0.05602, over 7426.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.286, pruned_loss=0.06068, over 1158341.44 frames. ], batch size: 17, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:23:07,697 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186161.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 06:23:16,549 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-07 06:23:18,806 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186176.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:23:25,623 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-07 06:23:32,348 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.668e+02 2.402e+02 3.098e+02 3.972e+02 8.418e+02, threshold=6.197e+02, percent-clipped=5.0 +2023-02-07 06:23:36,037 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186201.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:23:40,499 INFO [train.py:901] (3/4) Epoch 24, batch 300, loss[loss=0.1907, simple_loss=0.2755, pruned_loss=0.05293, over 8016.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2878, pruned_loss=0.06204, over 1258826.88 frames. ], batch size: 22, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:23:53,003 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186226.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:24:00,549 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186236.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:24:00,864 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-07 06:24:13,910 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6017, 2.7110, 1.8848, 2.2640, 2.0858, 1.7403, 2.1768, 2.1626], + device='cuda:3'), covar=tensor([0.1474, 0.0379, 0.1176, 0.0649, 0.0798, 0.1427, 0.0955, 0.1029], + device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0235, 0.0337, 0.0311, 0.0303, 0.0342, 0.0350, 0.0320], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 06:24:15,084 INFO [train.py:901] (3/4) Epoch 24, batch 350, loss[loss=0.1896, simple_loss=0.2781, pruned_loss=0.05052, over 8257.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2872, pruned_loss=0.06159, over 1340501.25 frames. ], batch size: 24, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:24:24,620 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3693, 4.3175, 3.9572, 2.1764, 3.8595, 3.9213, 3.8915, 3.7857], + device='cuda:3'), covar=tensor([0.0685, 0.0563, 0.0920, 0.3869, 0.0807, 0.0940, 0.1186, 0.0723], + device='cuda:3'), in_proj_covar=tensor([0.0526, 0.0445, 0.0431, 0.0541, 0.0431, 0.0446, 0.0426, 0.0389], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 06:24:28,140 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186276.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 06:24:28,717 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186277.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:24:42,187 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.476e+02 2.428e+02 2.971e+02 3.348e+02 5.777e+02, threshold=5.941e+02, percent-clipped=0.0 +2023-02-07 06:24:47,695 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1423, 4.0725, 3.6894, 2.1238, 3.6168, 3.7531, 3.6756, 3.5816], + device='cuda:3'), covar=tensor([0.0788, 0.0596, 0.1114, 0.4231, 0.0884, 0.1031, 0.1273, 0.0838], + device='cuda:3'), in_proj_covar=tensor([0.0527, 0.0445, 0.0431, 0.0541, 0.0431, 0.0446, 0.0426, 0.0389], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 06:24:50,334 INFO [train.py:901] (3/4) Epoch 24, batch 400, loss[loss=0.2679, simple_loss=0.3453, pruned_loss=0.09528, over 8755.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2878, pruned_loss=0.06184, over 1404560.02 frames. ], batch size: 30, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:25:26,090 INFO [train.py:901] (3/4) Epoch 24, batch 450, loss[loss=0.2033, simple_loss=0.2749, pruned_loss=0.06587, over 7704.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2881, pruned_loss=0.06187, over 1454087.77 frames. ], batch size: 18, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:25:52,936 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.720e+02 2.487e+02 2.919e+02 3.580e+02 7.824e+02, threshold=5.839e+02, percent-clipped=3.0 +2023-02-07 06:26:02,022 INFO [train.py:901] (3/4) Epoch 24, batch 500, loss[loss=0.2191, simple_loss=0.3061, pruned_loss=0.06607, over 8346.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2876, pruned_loss=0.0617, over 1488925.60 frames. ], batch size: 26, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:26:23,367 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186439.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:26:37,824 INFO [train.py:901] (3/4) Epoch 24, batch 550, loss[loss=0.2157, simple_loss=0.2974, pruned_loss=0.067, over 6709.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2874, pruned_loss=0.06128, over 1520260.34 frames. ], batch size: 71, lr: 3.18e-03, grad_scale: 16.0 +2023-02-07 06:26:40,778 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186462.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:26:55,567 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2670, 2.5659, 2.8975, 1.6380, 3.2173, 1.8870, 1.5409, 2.2326], + device='cuda:3'), covar=tensor([0.0835, 0.0396, 0.0278, 0.0851, 0.0485, 0.0981, 0.0991, 0.0519], + device='cuda:3'), in_proj_covar=tensor([0.0458, 0.0399, 0.0351, 0.0449, 0.0382, 0.0539, 0.0393, 0.0426], + device='cuda:3'), out_proj_covar=tensor([1.2231e-04, 1.0434e-04, 9.2115e-05, 1.1794e-04, 1.0032e-04, 1.5167e-04, + 1.0568e-04, 1.1243e-04], device='cuda:3') +2023-02-07 06:27:01,108 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186492.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:27:01,726 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186493.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:27:03,580 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.615e+02 2.350e+02 3.005e+02 3.846e+02 7.955e+02, threshold=6.011e+02, percent-clipped=1.0 +2023-02-07 06:27:12,590 INFO [train.py:901] (3/4) Epoch 24, batch 600, loss[loss=0.291, simple_loss=0.3382, pruned_loss=0.1219, over 6693.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2857, pruned_loss=0.06006, over 1541236.29 frames. ], batch size: 71, lr: 3.17e-03, grad_scale: 16.0 +2023-02-07 06:27:19,661 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186517.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:27:21,530 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186520.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:27:23,338 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.23 vs. limit=5.0 +2023-02-07 06:27:26,202 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-07 06:27:29,796 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186532.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 06:27:31,744 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186535.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:27:40,312 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8007, 1.4581, 3.3051, 1.4340, 2.3563, 3.6099, 3.6790, 3.0820], + device='cuda:3'), covar=tensor([0.1276, 0.1802, 0.0346, 0.2070, 0.1036, 0.0221, 0.0493, 0.0543], + device='cuda:3'), in_proj_covar=tensor([0.0297, 0.0321, 0.0286, 0.0314, 0.0312, 0.0268, 0.0424, 0.0303], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 06:27:46,476 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186557.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:27:46,540 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186557.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 06:27:47,019 INFO [train.py:901] (3/4) Epoch 24, batch 650, loss[loss=0.2213, simple_loss=0.2951, pruned_loss=0.07375, over 8515.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2875, pruned_loss=0.06115, over 1557936.32 frames. ], batch size: 26, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:27:52,098 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9608, 1.5018, 3.4251, 1.5942, 2.3993, 3.7962, 3.8836, 3.2232], + device='cuda:3'), covar=tensor([0.1228, 0.1920, 0.0340, 0.2087, 0.1118, 0.0230, 0.0447, 0.0579], + device='cuda:3'), in_proj_covar=tensor([0.0296, 0.0321, 0.0285, 0.0314, 0.0311, 0.0268, 0.0423, 0.0302], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 06:28:01,240 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186577.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:28:07,496 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186585.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:28:15,660 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.811e+02 2.377e+02 2.753e+02 3.513e+02 8.271e+02, threshold=5.505e+02, percent-clipped=2.0 +2023-02-07 06:28:20,750 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186604.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:28:23,403 INFO [train.py:901] (3/4) Epoch 24, batch 700, loss[loss=0.204, simple_loss=0.2873, pruned_loss=0.06029, over 8505.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2863, pruned_loss=0.06064, over 1572670.10 frames. ], batch size: 26, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:28:33,223 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186621.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:28:43,766 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186635.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:28:59,667 INFO [train.py:901] (3/4) Epoch 24, batch 750, loss[loss=0.2092, simple_loss=0.288, pruned_loss=0.06523, over 7641.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2862, pruned_loss=0.06053, over 1581725.20 frames. ], batch size: 19, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:29:11,913 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-07 06:29:16,211 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9031, 1.2017, 3.2556, 1.2399, 2.5441, 2.6090, 2.8868, 2.8882], + device='cuda:3'), covar=tensor([0.2066, 0.6575, 0.1659, 0.5671, 0.2907, 0.2224, 0.1708, 0.1680], + device='cuda:3'), in_proj_covar=tensor([0.0645, 0.0653, 0.0707, 0.0637, 0.0721, 0.0620, 0.0613, 0.0689], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 06:29:21,540 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-07 06:29:27,092 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.478e+02 2.532e+02 3.077e+02 4.008e+02 9.294e+02, threshold=6.153e+02, percent-clipped=8.0 +2023-02-07 06:29:31,907 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-07 06:29:35,750 INFO [train.py:901] (3/4) Epoch 24, batch 800, loss[loss=0.1945, simple_loss=0.2742, pruned_loss=0.0574, over 8025.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2854, pruned_loss=0.06045, over 1586898.24 frames. ], batch size: 22, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:29:48,937 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186727.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:29:56,174 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186736.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:30:11,912 INFO [train.py:901] (3/4) Epoch 24, batch 850, loss[loss=0.1961, simple_loss=0.2713, pruned_loss=0.06046, over 7547.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2848, pruned_loss=0.06, over 1594676.69 frames. ], batch size: 18, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:30:29,503 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186783.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:30:39,062 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.556e+02 2.330e+02 2.764e+02 3.350e+02 7.186e+02, threshold=5.528e+02, percent-clipped=2.0 +2023-02-07 06:30:47,648 INFO [train.py:901] (3/4) Epoch 24, batch 900, loss[loss=0.1822, simple_loss=0.2695, pruned_loss=0.04741, over 8654.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2862, pruned_loss=0.06026, over 1601651.19 frames. ], batch size: 49, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:31:05,956 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186833.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:31:08,596 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186837.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:31:15,807 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1957, 1.9233, 2.5277, 2.1055, 2.5675, 2.2363, 2.0337, 1.3529], + device='cuda:3'), covar=tensor([0.5532, 0.4881, 0.1990, 0.3321, 0.2150, 0.3016, 0.1950, 0.5049], + device='cuda:3'), in_proj_covar=tensor([0.0948, 0.0992, 0.0812, 0.0958, 0.0998, 0.0901, 0.0756, 0.0830], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 06:31:24,264 INFO [train.py:901] (3/4) Epoch 24, batch 950, loss[loss=0.2412, simple_loss=0.3149, pruned_loss=0.08374, over 8546.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2856, pruned_loss=0.06022, over 1608027.94 frames. ], batch size: 28, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:31:24,460 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186858.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:31:24,519 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186858.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:31:30,140 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2245, 1.6051, 1.7984, 1.4736, 1.0601, 1.6167, 1.8617, 2.0123], + device='cuda:3'), covar=tensor([0.0520, 0.1159, 0.1656, 0.1383, 0.0578, 0.1363, 0.0656, 0.0547], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0152, 0.0189, 0.0159, 0.0100, 0.0162, 0.0111, 0.0143], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0009, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 06:31:39,987 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186879.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:31:43,476 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-07 06:31:48,338 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186891.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:31:50,408 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0207, 3.6879, 2.1949, 2.7348, 2.5795, 2.0220, 2.5102, 2.9876], + device='cuda:3'), covar=tensor([0.1684, 0.0296, 0.1158, 0.0791, 0.0821, 0.1482, 0.1182, 0.1114], + device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0234, 0.0336, 0.0310, 0.0301, 0.0341, 0.0346, 0.0319], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 06:31:52,264 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.527e+02 2.324e+02 2.850e+02 3.567e+02 7.043e+02, threshold=5.700e+02, percent-clipped=2.0 +2023-02-07 06:31:53,155 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186898.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:31:55,091 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186901.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:31:59,852 INFO [train.py:901] (3/4) Epoch 24, batch 1000, loss[loss=0.2125, simple_loss=0.2814, pruned_loss=0.07184, over 7707.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2857, pruned_loss=0.0601, over 1612627.37 frames. ], batch size: 18, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:32:05,641 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186916.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:32:15,432 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186929.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:32:15,510 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186929.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:32:19,252 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0830, 3.5775, 2.2029, 2.7767, 2.6189, 1.9580, 2.6456, 2.9826], + device='cuda:3'), covar=tensor([0.1872, 0.0366, 0.1265, 0.0822, 0.0902, 0.1636, 0.1265, 0.1204], + device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0235, 0.0336, 0.0310, 0.0301, 0.0341, 0.0347, 0.0319], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 06:32:20,508 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-07 06:32:29,428 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=186948.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:32:32,169 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186952.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:32:33,326 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-07 06:32:36,112 INFO [train.py:901] (3/4) Epoch 24, batch 1050, loss[loss=0.2062, simple_loss=0.288, pruned_loss=0.06215, over 8589.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2848, pruned_loss=0.05942, over 1616517.92 frames. ], batch size: 39, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:33:00,063 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3481, 1.5486, 4.5712, 1.6472, 4.0473, 3.7871, 4.1109, 3.9750], + device='cuda:3'), covar=tensor([0.0584, 0.4634, 0.0496, 0.4178, 0.1034, 0.0983, 0.0583, 0.0651], + device='cuda:3'), in_proj_covar=tensor([0.0642, 0.0649, 0.0703, 0.0634, 0.0718, 0.0618, 0.0610, 0.0686], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 06:33:01,572 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186992.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:33:02,864 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186994.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:33:04,729 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.690e+02 2.475e+02 2.949e+02 3.829e+02 9.793e+02, threshold=5.897e+02, percent-clipped=8.0 +2023-02-07 06:33:12,439 INFO [train.py:901] (3/4) Epoch 24, batch 1100, loss[loss=0.1645, simple_loss=0.243, pruned_loss=0.04295, over 7699.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2857, pruned_loss=0.05992, over 1620118.18 frames. ], batch size: 18, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:33:12,697 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7399, 2.0176, 3.2091, 1.5495, 2.4530, 2.1300, 1.7608, 2.4307], + device='cuda:3'), covar=tensor([0.1833, 0.2679, 0.0943, 0.4682, 0.1887, 0.3204, 0.2393, 0.2271], + device='cuda:3'), in_proj_covar=tensor([0.0527, 0.0615, 0.0556, 0.0650, 0.0653, 0.0600, 0.0544, 0.0634], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 06:33:18,326 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187016.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:33:19,044 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187017.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:33:37,544 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187043.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:33:38,320 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187044.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:33:45,690 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-07 06:33:47,128 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187056.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:33:48,356 INFO [train.py:901] (3/4) Epoch 24, batch 1150, loss[loss=0.2134, simple_loss=0.3002, pruned_loss=0.06327, over 8137.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2856, pruned_loss=0.05984, over 1623626.52 frames. ], batch size: 22, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:33:49,929 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0021, 1.5835, 3.3638, 1.5941, 2.4158, 3.6386, 3.7304, 3.1861], + device='cuda:3'), covar=tensor([0.1088, 0.1734, 0.0309, 0.1912, 0.1020, 0.0220, 0.0526, 0.0473], + device='cuda:3'), in_proj_covar=tensor([0.0298, 0.0324, 0.0288, 0.0317, 0.0315, 0.0270, 0.0427, 0.0304], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 06:33:52,034 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187063.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:33:57,505 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187071.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:34:02,535 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6082, 2.4596, 1.8177, 2.3147, 2.0762, 1.5570, 2.0641, 2.1053], + device='cuda:3'), covar=tensor([0.1375, 0.0403, 0.1184, 0.0577, 0.0760, 0.1560, 0.0984, 0.0962], + device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0234, 0.0336, 0.0309, 0.0301, 0.0341, 0.0346, 0.0318], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 06:34:16,146 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.611e+02 2.205e+02 2.742e+02 3.279e+02 6.267e+02, threshold=5.485e+02, percent-clipped=2.0 +2023-02-07 06:34:24,629 INFO [train.py:901] (3/4) Epoch 24, batch 1200, loss[loss=0.1999, simple_loss=0.2917, pruned_loss=0.05408, over 8682.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2844, pruned_loss=0.05886, over 1623332.22 frames. ], batch size: 34, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:34:57,400 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187154.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:35:00,007 INFO [train.py:901] (3/4) Epoch 24, batch 1250, loss[loss=0.2264, simple_loss=0.3081, pruned_loss=0.07233, over 8452.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2843, pruned_loss=0.05909, over 1620875.35 frames. ], batch size: 27, lr: 3.17e-03, grad_scale: 4.0 +2023-02-07 06:35:15,149 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187179.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:35:19,762 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187186.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:35:27,942 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.566e+02 2.417e+02 2.916e+02 3.659e+02 9.833e+02, threshold=5.832e+02, percent-clipped=6.0 +2023-02-07 06:35:30,894 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187202.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:35:35,746 INFO [train.py:901] (3/4) Epoch 24, batch 1300, loss[loss=0.2302, simple_loss=0.3098, pruned_loss=0.07534, over 8104.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2848, pruned_loss=0.05918, over 1620923.50 frames. ], batch size: 23, lr: 3.17e-03, grad_scale: 4.0 +2023-02-07 06:35:35,979 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187208.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:35:51,429 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.01 vs. limit=5.0 +2023-02-07 06:35:53,902 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187233.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:36:05,785 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187250.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:36:10,945 INFO [train.py:901] (3/4) Epoch 24, batch 1350, loss[loss=0.1959, simple_loss=0.2914, pruned_loss=0.05018, over 8602.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2853, pruned_loss=0.0592, over 1623487.70 frames. ], batch size: 31, lr: 3.17e-03, grad_scale: 4.0 +2023-02-07 06:36:20,764 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187272.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:36:21,335 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187273.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:36:22,852 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187275.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:36:39,321 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187297.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:36:39,750 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.841e+02 2.391e+02 3.088e+02 3.702e+02 1.176e+03, threshold=6.175e+02, percent-clipped=8.0 +2023-02-07 06:36:41,428 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187300.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:36:46,713 INFO [train.py:901] (3/4) Epoch 24, batch 1400, loss[loss=0.2093, simple_loss=0.2828, pruned_loss=0.06794, over 8446.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2853, pruned_loss=0.05965, over 1620613.81 frames. ], batch size: 27, lr: 3.17e-03, grad_scale: 4.0 +2023-02-07 06:36:52,969 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187317.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:36:54,426 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187319.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:36:59,284 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187325.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:37:03,370 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4911, 1.8041, 2.5281, 1.4101, 1.8583, 1.8147, 1.5969, 1.8015], + device='cuda:3'), covar=tensor([0.2046, 0.2510, 0.1094, 0.4492, 0.2032, 0.3284, 0.2380, 0.2498], + device='cuda:3'), in_proj_covar=tensor([0.0529, 0.0617, 0.0558, 0.0652, 0.0653, 0.0600, 0.0545, 0.0636], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 06:37:05,350 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4513, 3.7062, 2.3885, 3.0134, 3.0181, 2.2089, 3.2048, 3.1182], + device='cuda:3'), covar=tensor([0.1357, 0.0359, 0.1090, 0.0657, 0.0665, 0.1366, 0.0836, 0.0963], + device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0233, 0.0335, 0.0308, 0.0300, 0.0338, 0.0346, 0.0316], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 06:37:12,871 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187344.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:37:16,708 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 +2023-02-07 06:37:21,744 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-07 06:37:22,427 INFO [train.py:901] (3/4) Epoch 24, batch 1450, loss[loss=0.1891, simple_loss=0.2861, pruned_loss=0.04609, over 8334.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.285, pruned_loss=0.06022, over 1614323.92 frames. ], batch size: 26, lr: 3.17e-03, grad_scale: 4.0 +2023-02-07 06:37:42,336 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187387.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:37:43,074 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187388.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:37:44,408 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187390.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:37:49,534 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.586e+02 2.482e+02 2.870e+02 4.012e+02 8.494e+02, threshold=5.740e+02, percent-clipped=8.0 +2023-02-07 06:37:51,822 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187400.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:37:56,987 INFO [train.py:901] (3/4) Epoch 24, batch 1500, loss[loss=0.2157, simple_loss=0.2926, pruned_loss=0.06939, over 8513.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2849, pruned_loss=0.05966, over 1614666.08 frames. ], batch size: 28, lr: 3.17e-03, grad_scale: 4.0 +2023-02-07 06:38:22,160 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187442.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:38:33,247 INFO [train.py:901] (3/4) Epoch 24, batch 1550, loss[loss=0.1796, simple_loss=0.2592, pruned_loss=0.05001, over 7928.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2844, pruned_loss=0.05951, over 1617237.93 frames. ], batch size: 20, lr: 3.17e-03, grad_scale: 4.0 +2023-02-07 06:38:39,570 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187467.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:38:59,888 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.508e+02 2.355e+02 2.764e+02 3.622e+02 7.454e+02, threshold=5.529e+02, percent-clipped=4.0 +2023-02-07 06:39:02,835 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187502.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:39:06,537 INFO [train.py:901] (3/4) Epoch 24, batch 1600, loss[loss=0.1773, simple_loss=0.2521, pruned_loss=0.05122, over 7704.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2841, pruned_loss=0.05961, over 1614667.47 frames. ], batch size: 18, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:39:11,512 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187515.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:39:22,052 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187530.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:39:27,623 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187537.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:39:30,383 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187541.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:39:41,434 INFO [train.py:901] (3/4) Epoch 24, batch 1650, loss[loss=0.2627, simple_loss=0.3359, pruned_loss=0.09481, over 8199.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2845, pruned_loss=0.05977, over 1615359.10 frames. ], batch size: 23, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:39:52,525 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187573.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:40:09,686 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.577e+02 2.451e+02 2.921e+02 3.516e+02 7.853e+02, threshold=5.842e+02, percent-clipped=7.0 +2023-02-07 06:40:09,875 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187598.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:40:10,450 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9383, 1.4648, 4.3670, 2.3224, 2.4764, 5.0112, 5.1181, 4.3954], + device='cuda:3'), covar=tensor([0.1350, 0.1917, 0.0245, 0.1810, 0.1224, 0.0179, 0.0407, 0.0539], + device='cuda:3'), in_proj_covar=tensor([0.0295, 0.0319, 0.0283, 0.0312, 0.0311, 0.0267, 0.0422, 0.0301], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 06:40:16,433 INFO [train.py:901] (3/4) Epoch 24, batch 1700, loss[loss=0.1788, simple_loss=0.256, pruned_loss=0.05083, over 7429.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2849, pruned_loss=0.05973, over 1615141.07 frames. ], batch size: 17, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:40:40,841 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187644.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:40:50,899 INFO [train.py:901] (3/4) Epoch 24, batch 1750, loss[loss=0.1748, simple_loss=0.2582, pruned_loss=0.04573, over 7651.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2849, pruned_loss=0.05948, over 1617793.19 frames. ], batch size: 19, lr: 3.17e-03, grad_scale: 8.0 +2023-02-07 06:40:58,581 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187669.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:41:18,565 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.694e+02 2.502e+02 3.000e+02 3.757e+02 9.885e+02, threshold=5.999e+02, percent-clipped=2.0 +2023-02-07 06:41:25,257 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-07 06:41:26,170 INFO [train.py:901] (3/4) Epoch 24, batch 1800, loss[loss=0.1544, simple_loss=0.2301, pruned_loss=0.03935, over 7438.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2849, pruned_loss=0.05983, over 1614622.51 frames. ], batch size: 17, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:41:43,300 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187734.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:41:59,225 INFO [train.py:901] (3/4) Epoch 24, batch 1850, loss[loss=0.2109, simple_loss=0.2802, pruned_loss=0.07084, over 7813.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2866, pruned_loss=0.06116, over 1612417.30 frames. ], batch size: 20, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:41:59,465 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187758.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:42:09,324 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187771.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:42:18,457 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187783.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:42:27,406 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187796.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:42:28,592 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.405e+02 2.492e+02 2.912e+02 4.002e+02 8.326e+02, threshold=5.824e+02, percent-clipped=6.0 +2023-02-07 06:42:36,372 INFO [train.py:901] (3/4) Epoch 24, batch 1900, loss[loss=0.2084, simple_loss=0.2923, pruned_loss=0.06226, over 8462.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.287, pruned_loss=0.06126, over 1611880.35 frames. ], batch size: 27, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:42:42,662 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1963, 4.1491, 3.7970, 2.0240, 3.7227, 3.8338, 3.6693, 3.6600], + device='cuda:3'), covar=tensor([0.0729, 0.0546, 0.0882, 0.4356, 0.0927, 0.1220, 0.1243, 0.0921], + device='cuda:3'), in_proj_covar=tensor([0.0529, 0.0446, 0.0435, 0.0543, 0.0435, 0.0448, 0.0428, 0.0391], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 06:42:49,917 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.46 vs. limit=2.0 +2023-02-07 06:43:02,033 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-07 06:43:05,614 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-07 06:43:05,823 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187849.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:43:11,796 INFO [train.py:901] (3/4) Epoch 24, batch 1950, loss[loss=0.189, simple_loss=0.2626, pruned_loss=0.05776, over 8079.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2858, pruned_loss=0.06063, over 1609393.71 frames. ], batch size: 21, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:43:18,404 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-07 06:43:22,598 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187874.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:43:28,014 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187881.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:43:30,765 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=187885.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:43:39,205 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.274e+02 2.375e+02 2.745e+02 3.412e+02 6.105e+02, threshold=5.491e+02, percent-clipped=1.0 +2023-02-07 06:43:39,237 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-07 06:43:42,660 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9279, 2.1853, 3.6833, 1.8491, 1.8683, 3.6838, 0.7303, 2.1330], + device='cuda:3'), covar=tensor([0.1268, 0.1510, 0.0293, 0.1655, 0.2533, 0.0246, 0.2120, 0.1268], + device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0200, 0.0132, 0.0222, 0.0273, 0.0137, 0.0171, 0.0194], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 06:43:46,291 INFO [train.py:901] (3/4) Epoch 24, batch 2000, loss[loss=0.2075, simple_loss=0.2973, pruned_loss=0.05881, over 8466.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2854, pruned_loss=0.06024, over 1606253.12 frames. ], batch size: 29, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:44:21,546 INFO [train.py:901] (3/4) Epoch 24, batch 2050, loss[loss=0.1985, simple_loss=0.2863, pruned_loss=0.05532, over 8634.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2848, pruned_loss=0.06004, over 1602660.45 frames. ], batch size: 34, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:44:23,780 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4771, 1.8438, 1.9038, 1.0824, 1.9728, 1.4512, 0.4341, 1.7129], + device='cuda:3'), covar=tensor([0.0549, 0.0372, 0.0298, 0.0624, 0.0452, 0.0949, 0.0944, 0.0316], + device='cuda:3'), in_proj_covar=tensor([0.0458, 0.0400, 0.0354, 0.0450, 0.0384, 0.0539, 0.0394, 0.0425], + device='cuda:3'), out_proj_covar=tensor([1.2219e-04, 1.0450e-04, 9.3072e-05, 1.1817e-04, 1.0087e-04, 1.5174e-04, + 1.0582e-04, 1.1198e-04], device='cuda:3') +2023-02-07 06:44:42,298 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187989.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:44:46,914 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187996.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:44:48,824 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.710e+02 2.387e+02 2.958e+02 3.531e+02 6.524e+02, threshold=5.915e+02, percent-clipped=3.0 +2023-02-07 06:44:51,434 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=188000.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:44:56,705 INFO [train.py:901] (3/4) Epoch 24, batch 2100, loss[loss=0.2149, simple_loss=0.306, pruned_loss=0.06188, over 8746.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2862, pruned_loss=0.06089, over 1607340.63 frames. ], batch size: 30, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:45:24,480 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3202, 2.0035, 2.5382, 2.2119, 2.4943, 2.3708, 2.1909, 1.3055], + device='cuda:3'), covar=tensor([0.5150, 0.4640, 0.1885, 0.3476, 0.2413, 0.2965, 0.1902, 0.5254], + device='cuda:3'), in_proj_covar=tensor([0.0954, 0.1003, 0.0820, 0.0967, 0.1004, 0.0911, 0.0763, 0.0837], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 06:45:32,170 INFO [train.py:901] (3/4) Epoch 24, batch 2150, loss[loss=0.1898, simple_loss=0.2679, pruned_loss=0.05591, over 8468.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2862, pruned_loss=0.06075, over 1612585.20 frames. ], batch size: 27, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:45:35,777 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2697, 2.0424, 2.7943, 2.3799, 2.7379, 2.2871, 2.0779, 1.4398], + device='cuda:3'), covar=tensor([0.5583, 0.5112, 0.1971, 0.3624, 0.2435, 0.3281, 0.2000, 0.5709], + device='cuda:3'), in_proj_covar=tensor([0.0954, 0.1003, 0.0820, 0.0967, 0.1004, 0.0911, 0.0763, 0.0837], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 06:45:58,768 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.633e+02 2.464e+02 3.048e+02 3.692e+02 7.821e+02, threshold=6.095e+02, percent-clipped=5.0 +2023-02-07 06:46:03,903 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188105.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:46:05,702 INFO [train.py:901] (3/4) Epoch 24, batch 2200, loss[loss=0.2213, simple_loss=0.298, pruned_loss=0.07231, over 8085.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2855, pruned_loss=0.06003, over 1616064.68 frames. ], batch size: 21, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:46:21,948 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188130.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:46:41,366 INFO [train.py:901] (3/4) Epoch 24, batch 2250, loss[loss=0.1691, simple_loss=0.2606, pruned_loss=0.0388, over 7821.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2838, pruned_loss=0.05911, over 1616074.75 frames. ], batch size: 20, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:47:09,411 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.377e+02 2.349e+02 3.047e+02 3.898e+02 9.680e+02, threshold=6.095e+02, percent-clipped=4.0 +2023-02-07 06:47:16,303 INFO [train.py:901] (3/4) Epoch 24, batch 2300, loss[loss=0.1664, simple_loss=0.2472, pruned_loss=0.0428, over 7557.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.283, pruned_loss=0.05873, over 1616247.48 frames. ], batch size: 18, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:47:42,370 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188245.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:47:47,131 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188252.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:47:50,573 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188256.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:47:51,707 INFO [train.py:901] (3/4) Epoch 24, batch 2350, loss[loss=0.2417, simple_loss=0.3259, pruned_loss=0.07877, over 8341.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2831, pruned_loss=0.05942, over 1611120.43 frames. ], batch size: 26, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:48:00,185 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188270.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:48:05,507 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188277.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:48:08,191 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188281.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:48:20,104 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.904e+02 2.514e+02 3.085e+02 3.939e+02 8.316e+02, threshold=6.171e+02, percent-clipped=4.0 +2023-02-07 06:48:22,398 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=188301.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:48:27,100 INFO [train.py:901] (3/4) Epoch 24, batch 2400, loss[loss=0.2053, simple_loss=0.2781, pruned_loss=0.06627, over 7646.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2837, pruned_loss=0.05989, over 1609708.28 frames. ], batch size: 19, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:48:30,046 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8272, 1.5501, 3.3976, 1.4940, 2.3495, 3.7683, 3.9518, 3.2998], + device='cuda:3'), covar=tensor([0.1400, 0.1940, 0.0388, 0.2389, 0.1260, 0.0261, 0.0543, 0.0537], + device='cuda:3'), in_proj_covar=tensor([0.0298, 0.0323, 0.0285, 0.0315, 0.0314, 0.0270, 0.0425, 0.0304], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 06:49:00,630 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.91 vs. limit=5.0 +2023-02-07 06:49:02,252 INFO [train.py:901] (3/4) Epoch 24, batch 2450, loss[loss=0.231, simple_loss=0.2995, pruned_loss=0.08127, over 7602.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2841, pruned_loss=0.0599, over 1605042.95 frames. ], batch size: 71, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:49:30,925 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.441e+02 2.263e+02 2.982e+02 3.612e+02 7.179e+02, threshold=5.965e+02, percent-clipped=1.0 +2023-02-07 06:49:38,097 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 +2023-02-07 06:49:38,407 INFO [train.py:901] (3/4) Epoch 24, batch 2500, loss[loss=0.205, simple_loss=0.2928, pruned_loss=0.05858, over 8106.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2842, pruned_loss=0.05966, over 1608367.92 frames. ], batch size: 23, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:50:11,884 INFO [train.py:901] (3/4) Epoch 24, batch 2550, loss[loss=0.2131, simple_loss=0.2905, pruned_loss=0.06782, over 8449.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2834, pruned_loss=0.05942, over 1611961.05 frames. ], batch size: 27, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:50:40,459 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.347e+02 2.538e+02 2.905e+02 3.766e+02 9.788e+02, threshold=5.809e+02, percent-clipped=4.0 +2023-02-07 06:50:41,317 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=188499.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:50:47,918 INFO [train.py:901] (3/4) Epoch 24, batch 2600, loss[loss=0.1928, simple_loss=0.2704, pruned_loss=0.05762, over 8139.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2828, pruned_loss=0.05888, over 1614221.37 frames. ], batch size: 22, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:51:21,921 INFO [train.py:901] (3/4) Epoch 24, batch 2650, loss[loss=0.2184, simple_loss=0.2999, pruned_loss=0.06847, over 8266.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2829, pruned_loss=0.05925, over 1612896.20 frames. ], batch size: 24, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:51:31,890 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 +2023-02-07 06:51:41,777 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.82 vs. limit=5.0 +2023-02-07 06:51:48,604 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.561e+02 2.338e+02 2.924e+02 3.924e+02 7.774e+02, threshold=5.847e+02, percent-clipped=4.0 +2023-02-07 06:51:55,400 INFO [train.py:901] (3/4) Epoch 24, batch 2700, loss[loss=0.1943, simple_loss=0.2823, pruned_loss=0.05313, over 8260.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2836, pruned_loss=0.05966, over 1615260.54 frames. ], batch size: 24, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:52:02,969 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.9497, 6.2311, 5.3247, 2.9968, 5.4800, 5.8345, 5.6134, 5.6454], + device='cuda:3'), covar=tensor([0.0420, 0.0314, 0.0808, 0.3839, 0.0667, 0.0739, 0.0957, 0.0540], + device='cuda:3'), in_proj_covar=tensor([0.0521, 0.0439, 0.0426, 0.0535, 0.0426, 0.0441, 0.0421, 0.0386], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 06:52:21,732 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=188645.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:52:31,269 INFO [train.py:901] (3/4) Epoch 24, batch 2750, loss[loss=0.2124, simple_loss=0.3018, pruned_loss=0.06144, over 8703.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.284, pruned_loss=0.06012, over 1614482.93 frames. ], batch size: 34, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:52:57,783 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.419e+02 2.375e+02 3.087e+02 4.139e+02 1.460e+03, threshold=6.174e+02, percent-clipped=4.0 +2023-02-07 06:53:05,388 INFO [train.py:901] (3/4) Epoch 24, batch 2800, loss[loss=0.2051, simple_loss=0.2942, pruned_loss=0.058, over 8292.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2842, pruned_loss=0.06026, over 1611135.06 frames. ], batch size: 23, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:53:23,062 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1930, 2.0390, 2.6821, 2.1691, 2.6398, 2.2743, 2.0911, 1.3987], + device='cuda:3'), covar=tensor([0.5541, 0.4838, 0.2089, 0.3992, 0.2561, 0.3396, 0.2015, 0.5702], + device='cuda:3'), in_proj_covar=tensor([0.0947, 0.0992, 0.0814, 0.0956, 0.0995, 0.0905, 0.0756, 0.0832], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 06:53:40,611 INFO [train.py:901] (3/4) Epoch 24, batch 2850, loss[loss=0.2125, simple_loss=0.2949, pruned_loss=0.06508, over 8456.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2841, pruned_loss=0.06007, over 1612650.22 frames. ], batch size: 27, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:53:42,193 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=188760.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:54:07,932 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.612e+02 2.346e+02 3.064e+02 3.754e+02 6.997e+02, threshold=6.129e+02, percent-clipped=3.0 +2023-02-07 06:54:14,851 INFO [train.py:901] (3/4) Epoch 24, batch 2900, loss[loss=0.2036, simple_loss=0.2919, pruned_loss=0.05758, over 8472.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2845, pruned_loss=0.06009, over 1613821.56 frames. ], batch size: 25, lr: 3.16e-03, grad_scale: 8.0 +2023-02-07 06:54:39,478 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=188843.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:54:48,793 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4132, 1.5009, 1.3963, 1.8309, 0.7662, 1.2634, 1.3931, 1.4876], + device='cuda:3'), covar=tensor([0.0822, 0.0703, 0.0922, 0.0490, 0.1088, 0.1326, 0.0646, 0.0732], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0197, 0.0242, 0.0214, 0.0205, 0.0246, 0.0249, 0.0208], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 06:54:51,387 INFO [train.py:901] (3/4) Epoch 24, batch 2950, loss[loss=0.2183, simple_loss=0.3059, pruned_loss=0.06532, over 8280.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2853, pruned_loss=0.06044, over 1609798.76 frames. ], batch size: 23, lr: 3.15e-03, grad_scale: 8.0 +2023-02-07 06:54:51,396 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-07 06:55:19,114 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.629e+02 2.217e+02 2.685e+02 3.700e+02 9.567e+02, threshold=5.370e+02, percent-clipped=4.0 +2023-02-07 06:55:25,889 INFO [train.py:901] (3/4) Epoch 24, batch 3000, loss[loss=0.1726, simple_loss=0.264, pruned_loss=0.04056, over 7970.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2856, pruned_loss=0.06067, over 1611555.26 frames. ], batch size: 21, lr: 3.15e-03, grad_scale: 8.0 +2023-02-07 06:55:25,889 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 06:55:39,554 INFO [train.py:935] (3/4) Epoch 24, validation: loss=0.1724, simple_loss=0.2726, pruned_loss=0.03604, over 944034.00 frames. +2023-02-07 06:55:39,555 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6747MB +2023-02-07 06:56:06,004 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=188947.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:56:07,364 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=188949.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:56:13,969 INFO [train.py:901] (3/4) Epoch 24, batch 3050, loss[loss=0.2196, simple_loss=0.3095, pruned_loss=0.06483, over 8193.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2867, pruned_loss=0.06108, over 1614933.05 frames. ], batch size: 23, lr: 3.15e-03, grad_scale: 8.0 +2023-02-07 06:56:14,159 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=188958.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:56:41,562 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.497e+02 2.422e+02 3.010e+02 3.817e+02 9.746e+02, threshold=6.020e+02, percent-clipped=4.0 +2023-02-07 06:56:49,126 INFO [train.py:901] (3/4) Epoch 24, batch 3100, loss[loss=0.2726, simple_loss=0.3531, pruned_loss=0.09601, over 8191.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2871, pruned_loss=0.06133, over 1613062.25 frames. ], batch size: 23, lr: 3.15e-03, grad_scale: 8.0 +2023-02-07 06:56:54,808 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189016.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:57:12,122 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189041.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:57:16,980 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189048.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:57:23,537 INFO [train.py:901] (3/4) Epoch 24, batch 3150, loss[loss=0.2403, simple_loss=0.3192, pruned_loss=0.0807, over 7219.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2877, pruned_loss=0.0618, over 1609102.71 frames. ], batch size: 71, lr: 3.15e-03, grad_scale: 8.0 +2023-02-07 06:57:40,666 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189082.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:57:50,853 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189097.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:57:51,369 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.687e+02 2.394e+02 2.917e+02 3.565e+02 6.979e+02, threshold=5.834e+02, percent-clipped=3.0 +2023-02-07 06:57:59,743 INFO [train.py:901] (3/4) Epoch 24, batch 3200, loss[loss=0.1944, simple_loss=0.287, pruned_loss=0.05084, over 8244.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2884, pruned_loss=0.06196, over 1613391.61 frames. ], batch size: 24, lr: 3.15e-03, grad_scale: 8.0 +2023-02-07 06:58:33,957 INFO [train.py:901] (3/4) Epoch 24, batch 3250, loss[loss=0.2789, simple_loss=0.3339, pruned_loss=0.1119, over 6921.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2883, pruned_loss=0.06184, over 1613517.18 frames. ], batch size: 71, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 06:59:01,533 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.806e+02 2.438e+02 3.003e+02 3.759e+02 6.490e+02, threshold=6.005e+02, percent-clipped=4.0 +2023-02-07 06:59:08,532 INFO [train.py:901] (3/4) Epoch 24, batch 3300, loss[loss=0.1899, simple_loss=0.278, pruned_loss=0.05088, over 7813.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.287, pruned_loss=0.06129, over 1616435.36 frames. ], batch size: 20, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 06:59:12,861 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189214.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:59:30,906 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189239.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 06:59:44,170 INFO [train.py:901] (3/4) Epoch 24, batch 3350, loss[loss=0.1849, simple_loss=0.2587, pruned_loss=0.05553, over 7796.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2863, pruned_loss=0.06093, over 1612848.20 frames. ], batch size: 19, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 06:59:49,412 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189266.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:00:05,166 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8460, 1.8539, 1.9398, 1.6534, 0.9637, 1.8121, 2.1949, 1.9790], + device='cuda:3'), covar=tensor([0.0432, 0.1111, 0.1612, 0.1348, 0.0586, 0.1342, 0.0593, 0.0591], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0153, 0.0189, 0.0160, 0.0100, 0.0163, 0.0112, 0.0144], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 07:00:05,777 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189291.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:00:07,055 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189293.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:00:10,342 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.695e+02 2.367e+02 2.967e+02 3.575e+02 9.298e+02, threshold=5.934e+02, percent-clipped=5.0 +2023-02-07 07:00:17,750 INFO [train.py:901] (3/4) Epoch 24, batch 3400, loss[loss=0.2061, simple_loss=0.2809, pruned_loss=0.0656, over 7781.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2863, pruned_loss=0.06125, over 1610982.97 frames. ], batch size: 19, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:00:19,422 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-07 07:00:50,350 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 +2023-02-07 07:00:52,451 INFO [train.py:901] (3/4) Epoch 24, batch 3450, loss[loss=0.1773, simple_loss=0.2594, pruned_loss=0.04762, over 7804.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2856, pruned_loss=0.06109, over 1608113.75 frames. ], batch size: 19, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:00:57,500 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189365.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:00:58,320 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1031, 1.8480, 2.3794, 2.0636, 2.3866, 2.1672, 1.9562, 1.1682], + device='cuda:3'), covar=tensor([0.5806, 0.5028, 0.1995, 0.3678, 0.2368, 0.3071, 0.1950, 0.5307], + device='cuda:3'), in_proj_covar=tensor([0.0944, 0.0992, 0.0811, 0.0957, 0.0995, 0.0903, 0.0755, 0.0828], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 07:01:04,474 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8268, 1.4600, 3.5408, 1.6039, 2.5422, 3.8565, 4.0365, 3.3080], + device='cuda:3'), covar=tensor([0.1314, 0.1791, 0.0301, 0.2094, 0.0951, 0.0229, 0.0551, 0.0535], + device='cuda:3'), in_proj_covar=tensor([0.0299, 0.0323, 0.0285, 0.0316, 0.0314, 0.0270, 0.0426, 0.0304], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 07:01:16,613 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189392.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:01:20,611 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 2.399e+02 2.884e+02 3.624e+02 7.571e+02, threshold=5.767e+02, percent-clipped=3.0 +2023-02-07 07:01:20,796 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3430, 1.3971, 4.5668, 1.8059, 4.0262, 3.8171, 4.1225, 3.9664], + device='cuda:3'), covar=tensor([0.0688, 0.5001, 0.0527, 0.4165, 0.1131, 0.0883, 0.0624, 0.0747], + device='cuda:3'), in_proj_covar=tensor([0.0650, 0.0657, 0.0716, 0.0647, 0.0725, 0.0622, 0.0618, 0.0697], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 07:01:26,260 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189406.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:01:27,406 INFO [train.py:901] (3/4) Epoch 24, batch 3500, loss[loss=0.1911, simple_loss=0.2841, pruned_loss=0.04903, over 8342.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2861, pruned_loss=0.06151, over 1609451.38 frames. ], batch size: 24, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:01:27,611 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189408.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:01:40,501 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-07 07:01:40,600 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189426.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:01:50,805 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189441.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:01:58,236 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5580, 1.2878, 2.8876, 1.5319, 2.1634, 3.0635, 3.2443, 2.6255], + device='cuda:3'), covar=tensor([0.1269, 0.1770, 0.0369, 0.2029, 0.0913, 0.0339, 0.0635, 0.0632], + device='cuda:3'), in_proj_covar=tensor([0.0298, 0.0322, 0.0285, 0.0316, 0.0314, 0.0270, 0.0427, 0.0304], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 07:02:03,505 INFO [train.py:901] (3/4) Epoch 24, batch 3550, loss[loss=0.2022, simple_loss=0.2889, pruned_loss=0.05777, over 8500.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2859, pruned_loss=0.06117, over 1608046.31 frames. ], batch size: 29, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:02:31,305 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.645e+02 2.509e+02 2.981e+02 3.708e+02 7.370e+02, threshold=5.962e+02, percent-clipped=4.0 +2023-02-07 07:02:33,858 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 +2023-02-07 07:02:37,645 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189507.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:02:38,137 INFO [train.py:901] (3/4) Epoch 24, batch 3600, loss[loss=0.1773, simple_loss=0.251, pruned_loss=0.05182, over 7245.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2849, pruned_loss=0.06013, over 1609383.82 frames. ], batch size: 16, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:02:45,842 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189519.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:02:58,695 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 +2023-02-07 07:02:59,867 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9342, 1.6059, 1.8304, 1.4907, 0.9266, 1.6409, 1.8388, 1.5772], + device='cuda:3'), covar=tensor([0.0519, 0.1177, 0.1557, 0.1349, 0.0588, 0.1409, 0.0615, 0.0620], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0153, 0.0190, 0.0160, 0.0100, 0.0163, 0.0112, 0.0144], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 07:03:01,354 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-07 07:03:01,842 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189541.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:03:12,159 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189556.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:03:13,414 INFO [train.py:901] (3/4) Epoch 24, batch 3650, loss[loss=0.202, simple_loss=0.2652, pruned_loss=0.06936, over 7270.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2846, pruned_loss=0.06035, over 1610084.64 frames. ], batch size: 16, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:03:41,107 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-07 07:03:41,750 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.441e+02 2.496e+02 2.930e+02 3.600e+02 6.319e+02, threshold=5.860e+02, percent-clipped=2.0 +2023-02-07 07:03:48,376 INFO [train.py:901] (3/4) Epoch 24, batch 3700, loss[loss=0.1695, simple_loss=0.2518, pruned_loss=0.04364, over 7534.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2838, pruned_loss=0.05977, over 1609967.32 frames. ], batch size: 18, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:03:49,802 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189610.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:04:23,128 INFO [train.py:901] (3/4) Epoch 24, batch 3750, loss[loss=0.1934, simple_loss=0.2659, pruned_loss=0.06051, over 7423.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2855, pruned_loss=0.06019, over 1608823.03 frames. ], batch size: 17, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:04:23,299 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189658.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:04:26,014 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189662.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:04:27,231 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189664.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:04:42,765 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189687.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:04:43,976 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189689.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:04:51,127 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.609e+02 3.129e+02 4.249e+02 7.016e+02, threshold=6.258e+02, percent-clipped=8.0 +2023-02-07 07:04:57,798 INFO [train.py:901] (3/4) Epoch 24, batch 3800, loss[loss=0.211, simple_loss=0.2983, pruned_loss=0.06185, over 8529.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2856, pruned_loss=0.0603, over 1608295.00 frames. ], batch size: 28, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:04:58,621 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189709.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:05:05,504 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2360, 2.0832, 2.7121, 2.2613, 2.7157, 2.3233, 2.0942, 1.5750], + device='cuda:3'), covar=tensor([0.5462, 0.4860, 0.2007, 0.3742, 0.2277, 0.3120, 0.1809, 0.5237], + device='cuda:3'), in_proj_covar=tensor([0.0946, 0.0994, 0.0812, 0.0961, 0.0999, 0.0905, 0.0756, 0.0831], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 07:05:07,479 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189722.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:05:09,545 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189725.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:05:24,369 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189746.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:05:32,060 INFO [train.py:901] (3/4) Epoch 24, batch 3850, loss[loss=0.1979, simple_loss=0.2861, pruned_loss=0.05485, over 8531.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2861, pruned_loss=0.06024, over 1609566.02 frames. ], batch size: 31, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:05:35,656 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189763.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:05:47,661 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-07 07:05:53,013 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189788.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:05:59,101 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189797.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:05:59,534 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.598e+02 2.407e+02 2.910e+02 3.432e+02 8.251e+02, threshold=5.819e+02, percent-clipped=1.0 +2023-02-07 07:06:06,326 INFO [train.py:901] (3/4) Epoch 24, batch 3900, loss[loss=0.1808, simple_loss=0.2744, pruned_loss=0.04357, over 8257.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2857, pruned_loss=0.05976, over 1613164.42 frames. ], batch size: 24, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:06:10,783 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189812.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:06:17,542 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189822.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:06:18,893 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189824.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:06:27,526 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189837.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:06:27,536 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9016, 1.7240, 2.5373, 1.7540, 1.3629, 2.4634, 0.5287, 1.6667], + device='cuda:3'), covar=tensor([0.1337, 0.0993, 0.0247, 0.0997, 0.2169, 0.0289, 0.1930, 0.1044], + device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0200, 0.0130, 0.0221, 0.0269, 0.0137, 0.0170, 0.0194], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 07:06:42,146 INFO [train.py:901] (3/4) Epoch 24, batch 3950, loss[loss=0.2028, simple_loss=0.269, pruned_loss=0.06828, over 7558.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2856, pruned_loss=0.05979, over 1611498.54 frames. ], batch size: 18, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:06:45,569 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=189863.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:07:07,915 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-07 07:07:09,586 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.473e+02 2.387e+02 3.217e+02 3.997e+02 8.874e+02, threshold=6.434e+02, percent-clipped=5.0 +2023-02-07 07:07:16,321 INFO [train.py:901] (3/4) Epoch 24, batch 4000, loss[loss=0.1721, simple_loss=0.2469, pruned_loss=0.04868, over 7539.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2853, pruned_loss=0.05956, over 1617681.23 frames. ], batch size: 18, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:07:51,147 INFO [train.py:901] (3/4) Epoch 24, batch 4050, loss[loss=0.195, simple_loss=0.2841, pruned_loss=0.05296, over 8241.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2856, pruned_loss=0.059, over 1622909.52 frames. ], batch size: 22, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:08:02,816 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7630, 1.9822, 2.0886, 1.5063, 2.2304, 1.5678, 0.8914, 1.9579], + device='cuda:3'), covar=tensor([0.0614, 0.0355, 0.0273, 0.0579, 0.0400, 0.0820, 0.0907, 0.0309], + device='cuda:3'), in_proj_covar=tensor([0.0455, 0.0396, 0.0353, 0.0450, 0.0381, 0.0534, 0.0394, 0.0425], + device='cuda:3'), out_proj_covar=tensor([1.2137e-04, 1.0340e-04, 9.2506e-05, 1.1828e-04, 1.0005e-04, 1.4993e-04, + 1.0600e-04, 1.1207e-04], device='cuda:3') +2023-02-07 07:08:05,479 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189978.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:08:07,504 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189981.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:08:18,671 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.498e+02 2.334e+02 2.770e+02 3.399e+02 1.124e+03, threshold=5.539e+02, percent-clipped=1.0 +2023-02-07 07:08:22,558 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190002.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:08:26,239 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190006.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:08:27,378 INFO [train.py:901] (3/4) Epoch 24, batch 4100, loss[loss=0.2115, simple_loss=0.291, pruned_loss=0.06597, over 8189.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2846, pruned_loss=0.05894, over 1618907.99 frames. ], batch size: 23, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:08:36,503 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-07 07:09:02,428 INFO [train.py:901] (3/4) Epoch 24, batch 4150, loss[loss=0.172, simple_loss=0.2513, pruned_loss=0.04636, over 7538.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2846, pruned_loss=0.0591, over 1617337.39 frames. ], batch size: 18, lr: 3.15e-03, grad_scale: 16.0 +2023-02-07 07:09:08,098 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190066.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:09:15,374 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-07 07:09:17,890 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190080.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:09:25,156 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190090.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:09:30,574 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.686e+02 2.377e+02 2.724e+02 3.400e+02 7.023e+02, threshold=5.448e+02, percent-clipped=3.0 +2023-02-07 07:09:35,503 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190105.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:09:37,386 INFO [train.py:901] (3/4) Epoch 24, batch 4200, loss[loss=0.2011, simple_loss=0.2863, pruned_loss=0.05792, over 8334.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2834, pruned_loss=0.05843, over 1615817.35 frames. ], batch size: 26, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:09:43,678 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190117.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:09:48,188 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-07 07:09:54,431 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6470, 2.3985, 3.2340, 2.5799, 3.2192, 2.7131, 2.5188, 1.9671], + device='cuda:3'), covar=tensor([0.5210, 0.5127, 0.1932, 0.3935, 0.2324, 0.2751, 0.1675, 0.5401], + device='cuda:3'), in_proj_covar=tensor([0.0943, 0.0992, 0.0813, 0.0960, 0.1000, 0.0905, 0.0757, 0.0830], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 07:10:04,794 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7101, 1.5220, 1.8000, 1.6072, 1.7612, 1.7471, 1.6042, 0.7777], + device='cuda:3'), covar=tensor([0.5234, 0.4323, 0.2016, 0.3199, 0.2343, 0.2913, 0.1824, 0.4725], + device='cuda:3'), in_proj_covar=tensor([0.0943, 0.0992, 0.0814, 0.0960, 0.1001, 0.0905, 0.0756, 0.0830], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 07:10:10,652 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-07 07:10:11,424 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190157.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:10:11,869 INFO [train.py:901] (3/4) Epoch 24, batch 4250, loss[loss=0.1859, simple_loss=0.2697, pruned_loss=0.05108, over 7200.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2837, pruned_loss=0.05865, over 1613450.95 frames. ], batch size: 16, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:10:19,365 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190169.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:10:28,691 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190181.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:10:40,140 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.586e+02 2.309e+02 2.865e+02 3.517e+02 8.092e+02, threshold=5.730e+02, percent-clipped=6.0 +2023-02-07 07:10:45,775 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190205.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:10:47,629 INFO [train.py:901] (3/4) Epoch 24, batch 4300, loss[loss=0.1791, simple_loss=0.2733, pruned_loss=0.04245, over 8338.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2843, pruned_loss=0.05891, over 1611201.49 frames. ], batch size: 25, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:10:56,762 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.70 vs. limit=5.0 +2023-02-07 07:11:05,307 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190234.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:11:21,909 INFO [train.py:901] (3/4) Epoch 24, batch 4350, loss[loss=0.1673, simple_loss=0.2455, pruned_loss=0.04457, over 7704.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2845, pruned_loss=0.05846, over 1615030.30 frames. ], batch size: 18, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:11:22,805 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190259.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:11:31,583 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0395, 1.7960, 2.3215, 1.9731, 2.2876, 2.0917, 1.9157, 1.1675], + device='cuda:3'), covar=tensor([0.5926, 0.4921, 0.2024, 0.3981, 0.2582, 0.3308, 0.2153, 0.5548], + device='cuda:3'), in_proj_covar=tensor([0.0948, 0.0994, 0.0815, 0.0963, 0.1003, 0.0906, 0.0758, 0.0832], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 07:11:40,177 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-07 07:11:50,403 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.582e+02 2.430e+02 2.823e+02 3.493e+02 1.012e+03, threshold=5.646e+02, percent-clipped=3.0 +2023-02-07 07:11:51,035 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 +2023-02-07 07:11:55,518 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4063, 1.6688, 1.6622, 1.0776, 1.6745, 1.3207, 0.3080, 1.6422], + device='cuda:3'), covar=tensor([0.0504, 0.0391, 0.0386, 0.0552, 0.0456, 0.0962, 0.0908, 0.0283], + device='cuda:3'), in_proj_covar=tensor([0.0456, 0.0397, 0.0353, 0.0450, 0.0382, 0.0535, 0.0395, 0.0426], + device='cuda:3'), out_proj_covar=tensor([1.2174e-04, 1.0369e-04, 9.2694e-05, 1.1833e-04, 1.0045e-04, 1.4998e-04, + 1.0619e-04, 1.1234e-04], device='cuda:3') +2023-02-07 07:11:57,336 INFO [train.py:901] (3/4) Epoch 24, batch 4400, loss[loss=0.2109, simple_loss=0.2909, pruned_loss=0.06546, over 7917.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2855, pruned_loss=0.05917, over 1613385.32 frames. ], batch size: 20, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:12:05,647 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9920, 3.5047, 2.2953, 2.9548, 2.8979, 2.0271, 2.8742, 2.9223], + device='cuda:3'), covar=tensor([0.1905, 0.0420, 0.1262, 0.0742, 0.0690, 0.1435, 0.1061, 0.1154], + device='cuda:3'), in_proj_covar=tensor([0.0356, 0.0235, 0.0338, 0.0312, 0.0299, 0.0343, 0.0346, 0.0318], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 07:12:15,563 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190334.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:12:16,330 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7977, 2.0574, 2.1874, 1.3855, 2.3233, 1.5745, 0.7933, 1.9770], + device='cuda:3'), covar=tensor([0.0670, 0.0421, 0.0357, 0.0718, 0.0470, 0.1019, 0.0923, 0.0361], + device='cuda:3'), in_proj_covar=tensor([0.0457, 0.0398, 0.0354, 0.0451, 0.0384, 0.0536, 0.0395, 0.0427], + device='cuda:3'), out_proj_covar=tensor([1.2195e-04, 1.0388e-04, 9.2742e-05, 1.1853e-04, 1.0082e-04, 1.5028e-04, + 1.0625e-04, 1.1252e-04], device='cuda:3') +2023-02-07 07:12:21,443 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-07 07:12:23,132 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-07 07:12:32,981 INFO [train.py:901] (3/4) Epoch 24, batch 4450, loss[loss=0.1894, simple_loss=0.2726, pruned_loss=0.05307, over 7977.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.283, pruned_loss=0.05774, over 1610081.00 frames. ], batch size: 21, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:12:43,413 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190373.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:13:00,241 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.418e+02 2.201e+02 2.691e+02 3.403e+02 6.534e+02, threshold=5.381e+02, percent-clipped=2.0 +2023-02-07 07:13:00,487 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190398.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:13:07,779 INFO [train.py:901] (3/4) Epoch 24, batch 4500, loss[loss=0.217, simple_loss=0.2973, pruned_loss=0.0683, over 8504.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2835, pruned_loss=0.05845, over 1610112.82 frames. ], batch size: 29, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:13:14,945 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190417.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:13:17,443 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-07 07:13:29,012 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190437.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:13:33,599 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8486, 1.7571, 2.4474, 1.6792, 1.3552, 2.4097, 0.4028, 1.5399], + device='cuda:3'), covar=tensor([0.1549, 0.1241, 0.0302, 0.1161, 0.2542, 0.0354, 0.1997, 0.1176], + device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0201, 0.0131, 0.0222, 0.0272, 0.0138, 0.0171, 0.0196], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 07:13:42,697 INFO [train.py:901] (3/4) Epoch 24, batch 4550, loss[loss=0.2043, simple_loss=0.2731, pruned_loss=0.0677, over 7695.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2838, pruned_loss=0.05877, over 1614715.38 frames. ], batch size: 18, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:13:44,284 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-07 07:13:44,879 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190461.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:13:45,494 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190462.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:14:02,627 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190486.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:14:10,720 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.617e+02 2.430e+02 2.973e+02 3.981e+02 9.647e+02, threshold=5.946e+02, percent-clipped=9.0 +2023-02-07 07:14:12,834 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190501.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 07:14:17,565 INFO [train.py:901] (3/4) Epoch 24, batch 4600, loss[loss=0.1895, simple_loss=0.2698, pruned_loss=0.0546, over 8117.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2841, pruned_loss=0.05949, over 1612301.93 frames. ], batch size: 23, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:14:21,251 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190513.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:14:40,894 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-07 07:14:50,269 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0697, 1.2528, 1.2248, 0.8118, 1.2308, 1.0721, 0.1380, 1.1887], + device='cuda:3'), covar=tensor([0.0444, 0.0424, 0.0380, 0.0574, 0.0525, 0.1017, 0.0901, 0.0346], + device='cuda:3'), in_proj_covar=tensor([0.0458, 0.0399, 0.0356, 0.0454, 0.0385, 0.0538, 0.0397, 0.0429], + device='cuda:3'), out_proj_covar=tensor([1.2233e-04, 1.0417e-04, 9.3240e-05, 1.1914e-04, 1.0106e-04, 1.5096e-04, + 1.0660e-04, 1.1314e-04], device='cuda:3') +2023-02-07 07:14:54,259 INFO [train.py:901] (3/4) Epoch 24, batch 4650, loss[loss=0.1769, simple_loss=0.2644, pruned_loss=0.0447, over 8511.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2835, pruned_loss=0.05893, over 1613331.32 frames. ], batch size: 26, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:15:06,809 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([0.8114, 0.7230, 0.8892, 0.7701, 0.6584, 0.8965, 0.1178, 0.7437], + device='cuda:3'), covar=tensor([0.0996, 0.0856, 0.0348, 0.0573, 0.1662, 0.0378, 0.1483, 0.0886], + device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0200, 0.0130, 0.0221, 0.0270, 0.0138, 0.0170, 0.0195], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 07:15:22,379 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.616e+02 2.469e+02 2.987e+02 3.787e+02 1.231e+03, threshold=5.974e+02, percent-clipped=5.0 +2023-02-07 07:15:29,196 INFO [train.py:901] (3/4) Epoch 24, batch 4700, loss[loss=0.2197, simple_loss=0.3057, pruned_loss=0.06686, over 8494.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2841, pruned_loss=0.05938, over 1610374.23 frames. ], batch size: 26, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:15:29,951 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190609.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 07:15:34,413 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190616.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 07:15:41,744 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190627.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:15:42,480 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190628.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:15:43,207 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6033, 1.9816, 3.2678, 1.4189, 2.5512, 2.0301, 1.7599, 2.4487], + device='cuda:3'), covar=tensor([0.2082, 0.2884, 0.0880, 0.4920, 0.1915, 0.3380, 0.2430, 0.2474], + device='cuda:3'), in_proj_covar=tensor([0.0529, 0.0616, 0.0556, 0.0654, 0.0651, 0.0600, 0.0548, 0.0637], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 07:16:04,014 INFO [train.py:901] (3/4) Epoch 24, batch 4750, loss[loss=0.2181, simple_loss=0.2958, pruned_loss=0.07014, over 8591.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2837, pruned_loss=0.059, over 1615893.97 frames. ], batch size: 31, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:16:18,718 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190678.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:16:20,754 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-07 07:16:22,896 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-07 07:16:29,412 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190693.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:16:32,691 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.630e+02 2.279e+02 2.789e+02 3.393e+02 7.815e+02, threshold=5.578e+02, percent-clipped=3.0 +2023-02-07 07:16:40,394 INFO [train.py:901] (3/4) Epoch 24, batch 4800, loss[loss=0.1842, simple_loss=0.2658, pruned_loss=0.05128, over 8252.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.283, pruned_loss=0.05894, over 1605804.14 frames. ], batch size: 22, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:17:13,016 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-07 07:17:15,032 INFO [train.py:901] (3/4) Epoch 24, batch 4850, loss[loss=0.1978, simple_loss=0.2823, pruned_loss=0.0567, over 8034.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2824, pruned_loss=0.05877, over 1604207.56 frames. ], batch size: 22, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:17:15,935 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.1631, 2.4893, 3.8186, 2.0891, 2.0770, 3.8026, 0.7764, 2.3649], + device='cuda:3'), covar=tensor([0.1353, 0.1203, 0.0194, 0.1641, 0.2292, 0.0231, 0.2048, 0.1180], + device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0199, 0.0130, 0.0220, 0.0270, 0.0138, 0.0169, 0.0195], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 07:17:17,936 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190761.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:17:40,144 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190793.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:17:43,508 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.462e+02 2.278e+02 2.775e+02 3.178e+02 7.824e+02, threshold=5.550e+02, percent-clipped=3.0 +2023-02-07 07:17:50,723 INFO [train.py:901] (3/4) Epoch 24, batch 4900, loss[loss=0.1673, simple_loss=0.2429, pruned_loss=0.04586, over 7166.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2818, pruned_loss=0.05848, over 1605917.00 frames. ], batch size: 16, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:17:53,677 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2428, 2.3705, 1.9616, 2.9693, 1.4063, 1.6434, 2.1664, 2.2649], + device='cuda:3'), covar=tensor([0.0644, 0.0761, 0.0835, 0.0329, 0.1119, 0.1312, 0.0836, 0.0851], + device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0199, 0.0245, 0.0215, 0.0206, 0.0248, 0.0252, 0.0208], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 07:17:55,387 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 +2023-02-07 07:18:09,308 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-07 07:18:25,852 INFO [train.py:901] (3/4) Epoch 24, batch 4950, loss[loss=0.221, simple_loss=0.2924, pruned_loss=0.07478, over 7547.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2826, pruned_loss=0.05861, over 1608898.48 frames. ], batch size: 72, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:18:36,136 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190872.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 07:18:39,588 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190876.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:18:45,201 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190884.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:18:53,851 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190897.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 07:18:54,307 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.730e+02 2.393e+02 2.890e+02 3.701e+02 7.772e+02, threshold=5.780e+02, percent-clipped=4.0 +2023-02-07 07:19:01,824 INFO [train.py:901] (3/4) Epoch 24, batch 5000, loss[loss=0.2179, simple_loss=0.2895, pruned_loss=0.07314, over 8036.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2822, pruned_loss=0.05862, over 1607340.76 frames. ], batch size: 22, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:19:02,720 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190909.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:19:19,904 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1329, 2.3296, 1.8118, 2.9009, 1.3216, 1.6380, 2.1025, 2.3234], + device='cuda:3'), covar=tensor([0.0715, 0.0745, 0.0904, 0.0348, 0.1134, 0.1266, 0.0811, 0.0751], + device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0199, 0.0246, 0.0216, 0.0206, 0.0248, 0.0253, 0.0208], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 07:19:33,491 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190953.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:19:36,599 INFO [train.py:901] (3/4) Epoch 24, batch 5050, loss[loss=0.1894, simple_loss=0.2725, pruned_loss=0.05312, over 7826.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2828, pruned_loss=0.05888, over 1609986.51 frames. ], batch size: 20, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:19:45,290 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=190971.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:19:51,186 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-07 07:20:04,896 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.504e+02 2.521e+02 3.221e+02 4.359e+02 8.705e+02, threshold=6.442e+02, percent-clipped=11.0 +2023-02-07 07:20:05,164 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1377, 1.9913, 2.6263, 2.2391, 2.5168, 2.2225, 2.0171, 1.4668], + device='cuda:3'), covar=tensor([0.5441, 0.4883, 0.1827, 0.3372, 0.2351, 0.3144, 0.1981, 0.5078], + device='cuda:3'), in_proj_covar=tensor([0.0947, 0.0994, 0.0812, 0.0963, 0.1001, 0.0906, 0.0755, 0.0830], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 07:20:11,616 INFO [train.py:901] (3/4) Epoch 24, batch 5100, loss[loss=0.1873, simple_loss=0.284, pruned_loss=0.04524, over 8654.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.284, pruned_loss=0.05955, over 1607991.68 frames. ], batch size: 34, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:20:31,920 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=191037.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:20:33,352 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6833, 1.6382, 2.1245, 1.6258, 1.3324, 2.1436, 0.4960, 1.4543], + device='cuda:3'), covar=tensor([0.1601, 0.1234, 0.0392, 0.0802, 0.2371, 0.0354, 0.1643, 0.1040], + device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0200, 0.0130, 0.0221, 0.0271, 0.0138, 0.0171, 0.0195], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 07:20:40,071 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191049.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:20:40,224 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 +2023-02-07 07:20:46,727 INFO [train.py:901] (3/4) Epoch 24, batch 5150, loss[loss=0.1931, simple_loss=0.2739, pruned_loss=0.05617, over 8236.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2843, pruned_loss=0.05944, over 1611398.78 frames. ], batch size: 22, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:20:53,683 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191068.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:20:57,803 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191074.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:21:00,592 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 +2023-02-07 07:21:05,749 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191086.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:21:05,786 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191086.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:21:13,667 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.533e+02 2.286e+02 2.691e+02 3.617e+02 7.196e+02, threshold=5.383e+02, percent-clipped=2.0 +2023-02-07 07:21:20,834 INFO [train.py:901] (3/4) Epoch 24, batch 5200, loss[loss=0.1957, simple_loss=0.2844, pruned_loss=0.05347, over 8195.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2849, pruned_loss=0.05948, over 1613693.11 frames. ], batch size: 23, lr: 3.14e-03, grad_scale: 16.0 +2023-02-07 07:21:38,118 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191132.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:21:50,042 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-07 07:21:50,823 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191150.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:21:52,273 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191152.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:21:55,614 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191157.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:21:56,093 INFO [train.py:901] (3/4) Epoch 24, batch 5250, loss[loss=0.2445, simple_loss=0.3053, pruned_loss=0.09183, over 6753.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2833, pruned_loss=0.05877, over 1607775.67 frames. ], batch size: 71, lr: 3.14e-03, grad_scale: 8.0 +2023-02-07 07:22:25,865 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.575e+02 2.418e+02 2.847e+02 3.981e+02 6.971e+02, threshold=5.694e+02, percent-clipped=11.0 +2023-02-07 07:22:28,316 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191203.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:22:31,590 INFO [train.py:901] (3/4) Epoch 24, batch 5300, loss[loss=0.1844, simple_loss=0.2748, pruned_loss=0.04699, over 8133.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2835, pruned_loss=0.05869, over 1611323.63 frames. ], batch size: 22, lr: 3.14e-03, grad_scale: 8.0 +2023-02-07 07:23:07,057 INFO [train.py:901] (3/4) Epoch 24, batch 5350, loss[loss=0.245, simple_loss=0.3051, pruned_loss=0.09247, over 7078.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2831, pruned_loss=0.05888, over 1608494.28 frames. ], batch size: 71, lr: 3.14e-03, grad_scale: 8.0 +2023-02-07 07:23:31,507 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 +2023-02-07 07:23:36,917 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.602e+02 2.294e+02 2.754e+02 3.342e+02 1.056e+03, threshold=5.508e+02, percent-clipped=2.0 +2023-02-07 07:23:40,179 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 +2023-02-07 07:23:42,385 INFO [train.py:901] (3/4) Epoch 24, batch 5400, loss[loss=0.2381, simple_loss=0.3168, pruned_loss=0.07968, over 8512.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2839, pruned_loss=0.05882, over 1611608.06 frames. ], batch size: 26, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:23:52,739 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-07 07:23:53,147 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191324.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:24:05,746 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191342.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:24:11,124 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191349.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:24:17,037 INFO [train.py:901] (3/4) Epoch 24, batch 5450, loss[loss=0.1787, simple_loss=0.256, pruned_loss=0.05073, over 7813.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.285, pruned_loss=0.05929, over 1611612.85 frames. ], batch size: 19, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:24:22,004 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9251, 1.7560, 2.5715, 1.6179, 1.4599, 2.5168, 0.4557, 1.6047], + device='cuda:3'), covar=tensor([0.1565, 0.1260, 0.0303, 0.1281, 0.2571, 0.0381, 0.2089, 0.1335], + device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0202, 0.0131, 0.0224, 0.0274, 0.0140, 0.0172, 0.0197], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 07:24:23,304 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191367.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:24:34,971 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191383.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:24:39,468 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-07 07:24:46,090 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.378e+02 2.294e+02 2.951e+02 3.676e+02 7.135e+02, threshold=5.902e+02, percent-clipped=5.0 +2023-02-07 07:24:48,440 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4361, 1.5481, 2.1084, 1.3406, 1.4301, 1.7001, 1.4582, 1.4715], + device='cuda:3'), covar=tensor([0.1960, 0.2633, 0.1029, 0.4475, 0.2086, 0.3262, 0.2457, 0.2355], + device='cuda:3'), in_proj_covar=tensor([0.0530, 0.0617, 0.0556, 0.0657, 0.0652, 0.0599, 0.0548, 0.0637], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 07:24:52,410 INFO [train.py:901] (3/4) Epoch 24, batch 5500, loss[loss=0.1959, simple_loss=0.2825, pruned_loss=0.05467, over 8544.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2847, pruned_loss=0.05943, over 1613145.31 frames. ], batch size: 31, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:24:52,650 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191408.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:25:07,823 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=191430.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:25:10,099 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191433.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:25:28,128 INFO [train.py:901] (3/4) Epoch 24, batch 5550, loss[loss=0.1931, simple_loss=0.2892, pruned_loss=0.04852, over 8246.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2848, pruned_loss=0.05906, over 1617623.33 frames. ], batch size: 24, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:25:36,927 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0198, 1.2471, 1.2311, 0.6799, 1.2200, 1.0328, 0.0736, 1.1949], + device='cuda:3'), covar=tensor([0.0451, 0.0415, 0.0387, 0.0614, 0.0474, 0.1025, 0.0867, 0.0372], + device='cuda:3'), in_proj_covar=tensor([0.0463, 0.0404, 0.0359, 0.0456, 0.0388, 0.0541, 0.0401, 0.0433], + device='cuda:3'), out_proj_covar=tensor([1.2354e-04, 1.0540e-04, 9.4053e-05, 1.1961e-04, 1.0190e-04, 1.5177e-04, + 1.0760e-04, 1.1426e-04], device='cuda:3') +2023-02-07 07:25:49,506 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8877, 1.7417, 2.4573, 1.6190, 1.4619, 2.3754, 0.4144, 1.5429], + device='cuda:3'), covar=tensor([0.1743, 0.1418, 0.0363, 0.1204, 0.2455, 0.0477, 0.2084, 0.1251], + device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0200, 0.0130, 0.0223, 0.0273, 0.0139, 0.0172, 0.0196], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 07:25:53,334 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=191494.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:25:57,921 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.572e+02 2.420e+02 3.039e+02 3.989e+02 7.925e+02, threshold=6.078e+02, percent-clipped=5.0 +2023-02-07 07:26:03,379 INFO [train.py:901] (3/4) Epoch 24, batch 5600, loss[loss=0.2249, simple_loss=0.3056, pruned_loss=0.07207, over 8500.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2848, pruned_loss=0.05923, over 1616243.51 frames. ], batch size: 28, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:26:23,607 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6959, 1.9403, 2.0634, 1.3387, 2.1438, 1.5895, 0.5505, 1.9201], + device='cuda:3'), covar=tensor([0.0612, 0.0385, 0.0324, 0.0614, 0.0469, 0.0856, 0.0972, 0.0313], + device='cuda:3'), in_proj_covar=tensor([0.0464, 0.0404, 0.0359, 0.0457, 0.0389, 0.0542, 0.0401, 0.0434], + device='cuda:3'), out_proj_covar=tensor([1.2366e-04, 1.0560e-04, 9.4065e-05, 1.1995e-04, 1.0212e-04, 1.5210e-04, + 1.0776e-04, 1.1433e-04], device='cuda:3') +2023-02-07 07:26:29,661 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191545.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:26:30,967 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=191547.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:26:38,449 INFO [train.py:901] (3/4) Epoch 24, batch 5650, loss[loss=0.2196, simple_loss=0.307, pruned_loss=0.06608, over 8360.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2843, pruned_loss=0.05903, over 1613841.99 frames. ], batch size: 26, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:26:45,398 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-07 07:27:08,570 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.600e+02 2.481e+02 2.901e+02 3.753e+02 9.237e+02, threshold=5.802e+02, percent-clipped=5.0 +2023-02-07 07:27:14,232 INFO [train.py:901] (3/4) Epoch 24, batch 5700, loss[loss=0.2313, simple_loss=0.3171, pruned_loss=0.07279, over 8512.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.285, pruned_loss=0.05907, over 1614943.16 frames. ], batch size: 26, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:27:15,046 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191609.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:27:49,746 INFO [train.py:901] (3/4) Epoch 24, batch 5750, loss[loss=0.1578, simple_loss=0.2497, pruned_loss=0.03293, over 8470.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2854, pruned_loss=0.05904, over 1617026.28 frames. ], batch size: 25, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:27:52,731 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191662.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:27:53,293 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-07 07:28:19,025 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.664e+02 2.394e+02 2.726e+02 3.524e+02 6.240e+02, threshold=5.452e+02, percent-clipped=4.0 +2023-02-07 07:28:25,087 INFO [train.py:901] (3/4) Epoch 24, batch 5800, loss[loss=0.2415, simple_loss=0.3142, pruned_loss=0.08442, over 8313.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2856, pruned_loss=0.05913, over 1616790.62 frames. ], batch size: 25, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:28:31,560 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5897, 2.7255, 3.2109, 1.9392, 3.2991, 2.0983, 1.8413, 2.4559], + device='cuda:3'), covar=tensor([0.0675, 0.0417, 0.0247, 0.0734, 0.0441, 0.0747, 0.0908, 0.0456], + device='cuda:3'), in_proj_covar=tensor([0.0465, 0.0405, 0.0360, 0.0459, 0.0389, 0.0543, 0.0402, 0.0434], + device='cuda:3'), out_proj_covar=tensor([1.2401e-04, 1.0586e-04, 9.4376e-05, 1.2048e-04, 1.0221e-04, 1.5225e-04, + 1.0794e-04, 1.1443e-04], device='cuda:3') +2023-02-07 07:28:38,095 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=191727.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:28:57,750 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3233, 4.3009, 3.9843, 2.0134, 3.8288, 3.9665, 3.9196, 3.8130], + device='cuda:3'), covar=tensor([0.0788, 0.0575, 0.0855, 0.4504, 0.0884, 0.1002, 0.1225, 0.0790], + device='cuda:3'), in_proj_covar=tensor([0.0539, 0.0454, 0.0439, 0.0553, 0.0436, 0.0455, 0.0435, 0.0398], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 07:28:59,277 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.11 vs. limit=5.0 +2023-02-07 07:28:59,639 INFO [train.py:901] (3/4) Epoch 24, batch 5850, loss[loss=0.1899, simple_loss=0.279, pruned_loss=0.05043, over 8195.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2847, pruned_loss=0.05859, over 1619303.39 frames. ], batch size: 23, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:29:26,690 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6964, 2.3452, 3.7300, 1.4838, 2.8350, 1.9876, 1.8540, 2.6077], + device='cuda:3'), covar=tensor([0.2201, 0.2858, 0.1243, 0.5217, 0.2085, 0.3933, 0.2628, 0.3047], + device='cuda:3'), in_proj_covar=tensor([0.0531, 0.0616, 0.0556, 0.0653, 0.0651, 0.0601, 0.0548, 0.0636], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 07:29:29,158 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.399e+02 2.223e+02 2.821e+02 3.422e+02 9.012e+02, threshold=5.641e+02, percent-clipped=8.0 +2023-02-07 07:29:30,134 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191801.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:29:34,726 INFO [train.py:901] (3/4) Epoch 24, batch 5900, loss[loss=0.2147, simple_loss=0.2894, pruned_loss=0.07, over 7229.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2852, pruned_loss=0.05878, over 1616893.26 frames. ], batch size: 72, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:29:48,215 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191826.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:29:59,347 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191842.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:30:10,944 INFO [train.py:901] (3/4) Epoch 24, batch 5950, loss[loss=0.1975, simple_loss=0.2799, pruned_loss=0.05751, over 8338.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2858, pruned_loss=0.05979, over 1615948.55 frames. ], batch size: 25, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:30:16,035 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191865.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:30:33,605 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191890.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:30:40,232 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.416e+02 2.338e+02 2.991e+02 3.628e+02 7.270e+02, threshold=5.982e+02, percent-clipped=3.0 +2023-02-07 07:30:45,681 INFO [train.py:901] (3/4) Epoch 24, batch 6000, loss[loss=0.1852, simple_loss=0.2598, pruned_loss=0.05532, over 7690.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2856, pruned_loss=0.0599, over 1611887.34 frames. ], batch size: 18, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:30:45,681 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 07:30:58,017 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9285, 1.2031, 3.0531, 1.1635, 2.6803, 2.5223, 2.8321, 2.6813], + device='cuda:3'), covar=tensor([0.0653, 0.3510, 0.0452, 0.3585, 0.1053, 0.0926, 0.0626, 0.0668], + device='cuda:3'), in_proj_covar=tensor([0.0648, 0.0653, 0.0710, 0.0643, 0.0720, 0.0617, 0.0618, 0.0690], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 07:31:01,021 INFO [train.py:935] (3/4) Epoch 24, validation: loss=0.1718, simple_loss=0.2718, pruned_loss=0.0359, over 944034.00 frames. +2023-02-07 07:31:01,022 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6747MB +2023-02-07 07:31:08,198 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191918.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:31:24,848 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191943.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:31:35,317 INFO [train.py:901] (3/4) Epoch 24, batch 6050, loss[loss=0.1935, simple_loss=0.2777, pruned_loss=0.05463, over 8080.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.285, pruned_loss=0.0597, over 1611076.27 frames. ], batch size: 21, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:32:04,601 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.623e+02 2.415e+02 2.742e+02 3.441e+02 8.508e+02, threshold=5.485e+02, percent-clipped=2.0 +2023-02-07 07:32:11,924 INFO [train.py:901] (3/4) Epoch 24, batch 6100, loss[loss=0.2027, simple_loss=0.2923, pruned_loss=0.05657, over 8283.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2845, pruned_loss=0.0593, over 1613204.38 frames. ], batch size: 23, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:32:12,376 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.77 vs. limit=5.0 +2023-02-07 07:32:17,377 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-07 07:32:34,063 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-07 07:32:46,753 INFO [train.py:901] (3/4) Epoch 24, batch 6150, loss[loss=0.1953, simple_loss=0.2671, pruned_loss=0.06176, over 8198.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2836, pruned_loss=0.0589, over 1613747.79 frames. ], batch size: 23, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:33:15,565 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192098.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:33:16,714 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.669e+02 2.367e+02 2.762e+02 3.348e+02 6.106e+02, threshold=5.524e+02, percent-clipped=2.0 +2023-02-07 07:33:22,017 INFO [train.py:901] (3/4) Epoch 24, batch 6200, loss[loss=0.2006, simple_loss=0.2836, pruned_loss=0.05877, over 8018.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2835, pruned_loss=0.05894, over 1608573.88 frames. ], batch size: 22, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:33:30,663 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192120.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:33:32,771 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192123.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:33:40,932 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 +2023-02-07 07:33:56,606 INFO [train.py:901] (3/4) Epoch 24, batch 6250, loss[loss=0.1979, simple_loss=0.2877, pruned_loss=0.05404, over 8344.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2833, pruned_loss=0.05917, over 1606811.74 frames. ], batch size: 26, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:34:05,369 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-07 07:34:07,843 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192173.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:34:26,650 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.677e+02 2.364e+02 2.949e+02 3.646e+02 8.976e+02, threshold=5.898e+02, percent-clipped=7.0 +2023-02-07 07:34:33,010 INFO [train.py:901] (3/4) Epoch 24, batch 6300, loss[loss=0.2619, simple_loss=0.3455, pruned_loss=0.08915, over 8449.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2833, pruned_loss=0.05889, over 1609791.02 frames. ], batch size: 50, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:34:52,685 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192237.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:34:54,069 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192239.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:34:56,871 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3079, 2.0290, 2.6485, 2.1906, 2.6542, 2.3285, 2.1529, 1.5260], + device='cuda:3'), covar=tensor([0.5595, 0.4920, 0.1943, 0.3710, 0.2482, 0.3188, 0.1913, 0.5480], + device='cuda:3'), in_proj_covar=tensor([0.0949, 0.0998, 0.0817, 0.0970, 0.1005, 0.0910, 0.0760, 0.0833], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 07:34:59,519 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1549, 1.2481, 1.5408, 1.1980, 0.7738, 1.3428, 1.1816, 0.9400], + device='cuda:3'), covar=tensor([0.0634, 0.1292, 0.1746, 0.1501, 0.0569, 0.1495, 0.0712, 0.0767], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0153, 0.0188, 0.0159, 0.0100, 0.0163, 0.0112, 0.0144], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 07:35:07,368 INFO [train.py:901] (3/4) Epoch 24, batch 6350, loss[loss=0.1954, simple_loss=0.2778, pruned_loss=0.05647, over 7819.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2831, pruned_loss=0.05912, over 1606251.48 frames. ], batch size: 20, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:35:36,831 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.340e+02 2.329e+02 2.896e+02 3.640e+02 5.459e+02, threshold=5.791e+02, percent-clipped=0.0 +2023-02-07 07:35:38,467 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4679, 2.4258, 1.8156, 2.2005, 2.1305, 1.4724, 1.9313, 1.9191], + device='cuda:3'), covar=tensor([0.1618, 0.0469, 0.1349, 0.0713, 0.0766, 0.1867, 0.1209, 0.1163], + device='cuda:3'), in_proj_covar=tensor([0.0356, 0.0235, 0.0337, 0.0310, 0.0301, 0.0342, 0.0346, 0.0319], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 07:35:43,003 INFO [train.py:901] (3/4) Epoch 24, batch 6400, loss[loss=0.1997, simple_loss=0.2972, pruned_loss=0.05114, over 8289.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2837, pruned_loss=0.05918, over 1608042.11 frames. ], batch size: 23, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:36:19,118 INFO [train.py:901] (3/4) Epoch 24, batch 6450, loss[loss=0.2928, simple_loss=0.3534, pruned_loss=0.1161, over 7176.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2846, pruned_loss=0.05975, over 1610720.04 frames. ], batch size: 72, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:36:49,114 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.491e+02 2.965e+02 3.858e+02 7.678e+02, threshold=5.930e+02, percent-clipped=7.0 +2023-02-07 07:36:54,692 INFO [train.py:901] (3/4) Epoch 24, batch 6500, loss[loss=0.1799, simple_loss=0.2704, pruned_loss=0.04476, over 8200.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2844, pruned_loss=0.05962, over 1611783.34 frames. ], batch size: 23, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:36:56,155 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192410.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:37:00,676 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192417.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:37:25,015 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192451.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:37:29,462 INFO [train.py:901] (3/4) Epoch 24, batch 6550, loss[loss=0.2038, simple_loss=0.2858, pruned_loss=0.06085, over 8462.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2842, pruned_loss=0.05913, over 1618901.54 frames. ], batch size: 25, lr: 3.13e-03, grad_scale: 8.0 +2023-02-07 07:37:33,493 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2023-02-07 07:37:33,655 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192464.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:37:36,747 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.70 vs. limit=5.0 +2023-02-07 07:37:48,317 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-07 07:37:58,367 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.534e+02 2.233e+02 2.734e+02 3.455e+02 6.558e+02, threshold=5.467e+02, percent-clipped=2.0 +2023-02-07 07:38:03,873 INFO [train.py:901] (3/4) Epoch 24, batch 6600, loss[loss=0.2571, simple_loss=0.3444, pruned_loss=0.08491, over 8197.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2854, pruned_loss=0.06018, over 1615975.38 frames. ], batch size: 23, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:38:08,113 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-07 07:38:10,834 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192517.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:38:17,072 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8361, 1.3759, 4.0038, 1.4337, 3.6234, 3.3529, 3.6762, 3.5119], + device='cuda:3'), covar=tensor([0.0707, 0.4596, 0.0612, 0.4270, 0.1194, 0.1038, 0.0599, 0.0764], + device='cuda:3'), in_proj_covar=tensor([0.0645, 0.0655, 0.0709, 0.0638, 0.0718, 0.0613, 0.0616, 0.0685], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 07:38:39,627 INFO [train.py:901] (3/4) Epoch 24, batch 6650, loss[loss=0.1974, simple_loss=0.2879, pruned_loss=0.05346, over 8625.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2855, pruned_loss=0.06008, over 1617258.48 frames. ], batch size: 39, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:38:53,938 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192579.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:38:55,137 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192581.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:38:55,969 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4844, 1.4302, 1.8184, 1.2018, 1.1138, 1.7954, 0.1637, 1.1660], + device='cuda:3'), covar=tensor([0.1468, 0.1282, 0.0401, 0.0876, 0.2501, 0.0434, 0.1919, 0.1161], + device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0199, 0.0130, 0.0220, 0.0271, 0.0138, 0.0171, 0.0196], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 07:38:56,512 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192583.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:39:08,650 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.608e+02 2.365e+02 2.876e+02 3.746e+02 9.522e+02, threshold=5.752e+02, percent-clipped=3.0 +2023-02-07 07:39:14,188 INFO [train.py:901] (3/4) Epoch 24, batch 6700, loss[loss=0.1736, simple_loss=0.25, pruned_loss=0.04862, over 7303.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2854, pruned_loss=0.05967, over 1619573.10 frames. ], batch size: 16, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:39:31,225 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192632.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:39:32,523 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192634.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:39:48,725 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 +2023-02-07 07:39:49,028 INFO [train.py:901] (3/4) Epoch 24, batch 6750, loss[loss=0.1708, simple_loss=0.2519, pruned_loss=0.04488, over 7712.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2849, pruned_loss=0.05907, over 1621467.74 frames. ], batch size: 18, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:40:00,543 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5585, 1.8678, 1.9225, 1.2180, 1.9951, 1.4624, 0.4139, 1.8687], + device='cuda:3'), covar=tensor([0.0476, 0.0281, 0.0236, 0.0479, 0.0411, 0.0794, 0.0855, 0.0224], + device='cuda:3'), in_proj_covar=tensor([0.0458, 0.0398, 0.0353, 0.0451, 0.0383, 0.0534, 0.0397, 0.0427], + device='cuda:3'), out_proj_covar=tensor([1.2197e-04, 1.0390e-04, 9.2514e-05, 1.1836e-04, 1.0065e-04, 1.4965e-04, + 1.0652e-04, 1.1246e-04], device='cuda:3') +2023-02-07 07:40:12,033 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192691.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:40:15,245 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192696.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:40:16,607 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192698.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:40:17,823 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.495e+02 2.601e+02 3.163e+02 3.841e+02 9.507e+02, threshold=6.325e+02, percent-clipped=3.0 +2023-02-07 07:40:23,396 INFO [train.py:901] (3/4) Epoch 24, batch 6800, loss[loss=0.2353, simple_loss=0.319, pruned_loss=0.07584, over 7931.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2853, pruned_loss=0.05955, over 1625813.67 frames. ], batch size: 20, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:40:24,828 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-07 07:40:56,187 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192753.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:40:56,813 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192754.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:40:59,526 INFO [train.py:901] (3/4) Epoch 24, batch 6850, loss[loss=0.1699, simple_loss=0.2586, pruned_loss=0.04063, over 8498.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2839, pruned_loss=0.05907, over 1621199.52 frames. ], batch size: 29, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:41:01,650 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192761.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:41:08,693 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3574, 2.3958, 1.8677, 2.2421, 2.0326, 1.5814, 1.8252, 1.9135], + device='cuda:3'), covar=tensor([0.1381, 0.0371, 0.1196, 0.0549, 0.0639, 0.1537, 0.1039, 0.1035], + device='cuda:3'), in_proj_covar=tensor([0.0356, 0.0235, 0.0337, 0.0308, 0.0300, 0.0341, 0.0348, 0.0318], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 07:41:15,298 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-07 07:41:26,177 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192795.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:41:29,430 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.645e+02 2.230e+02 2.864e+02 3.611e+02 9.090e+02, threshold=5.729e+02, percent-clipped=1.0 +2023-02-07 07:41:35,202 INFO [train.py:901] (3/4) Epoch 24, batch 6900, loss[loss=0.2352, simple_loss=0.3218, pruned_loss=0.07434, over 8243.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2842, pruned_loss=0.05906, over 1618106.70 frames. ], batch size: 24, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:41:54,069 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192835.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:42:03,105 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192849.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:42:08,876 INFO [train.py:901] (3/4) Epoch 24, batch 6950, loss[loss=0.2127, simple_loss=0.2892, pruned_loss=0.06809, over 8649.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.285, pruned_loss=0.05998, over 1615452.66 frames. ], batch size: 34, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:42:10,323 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192860.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:42:17,158 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192869.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:42:21,982 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192876.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:42:23,176 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-07 07:42:30,047 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192888.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:42:38,071 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.6397, 5.7330, 5.0501, 2.6932, 5.0999, 5.4684, 5.2999, 5.2810], + device='cuda:3'), covar=tensor([0.0551, 0.0344, 0.0811, 0.3936, 0.0719, 0.0908, 0.0994, 0.0604], + device='cuda:3'), in_proj_covar=tensor([0.0529, 0.0444, 0.0431, 0.0543, 0.0431, 0.0447, 0.0427, 0.0390], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 07:42:38,579 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.536e+02 2.384e+02 2.955e+02 3.597e+02 9.319e+02, threshold=5.910e+02, percent-clipped=1.0 +2023-02-07 07:42:44,702 INFO [train.py:901] (3/4) Epoch 24, batch 7000, loss[loss=0.2061, simple_loss=0.3009, pruned_loss=0.05565, over 8323.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2841, pruned_loss=0.05963, over 1614277.47 frames. ], batch size: 25, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:42:46,230 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192910.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:42:48,262 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192913.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:43:15,540 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192952.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:43:16,866 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192954.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:43:19,408 INFO [train.py:901] (3/4) Epoch 24, batch 7050, loss[loss=0.1888, simple_loss=0.2644, pruned_loss=0.0566, over 7699.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2847, pruned_loss=0.05972, over 1613835.34 frames. ], batch size: 18, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:43:24,405 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5893, 1.6314, 2.1446, 1.3675, 1.1976, 2.1620, 0.3459, 1.3548], + device='cuda:3'), covar=tensor([0.1410, 0.1059, 0.0299, 0.1015, 0.2234, 0.0327, 0.1763, 0.1095], + device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0199, 0.0129, 0.0221, 0.0272, 0.0139, 0.0171, 0.0196], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 07:43:32,658 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192977.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:43:33,236 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=192978.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:43:34,071 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192979.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:43:48,978 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.388e+02 2.407e+02 3.090e+02 3.925e+02 9.689e+02, threshold=6.179e+02, percent-clipped=7.0 +2023-02-07 07:43:54,440 INFO [train.py:901] (3/4) Epoch 24, batch 7100, loss[loss=0.2108, simple_loss=0.3032, pruned_loss=0.05925, over 8452.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2837, pruned_loss=0.05941, over 1609705.06 frames. ], batch size: 27, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:44:14,236 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193035.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:44:14,978 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.2031, 1.6193, 4.3434, 1.6255, 3.9152, 3.5908, 3.9264, 3.8364], + device='cuda:3'), covar=tensor([0.0587, 0.4415, 0.0582, 0.4374, 0.1129, 0.0981, 0.0622, 0.0657], + device='cuda:3'), in_proj_covar=tensor([0.0646, 0.0657, 0.0711, 0.0642, 0.0721, 0.0617, 0.0619, 0.0690], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 07:44:20,314 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6401, 2.3600, 4.0275, 1.5596, 3.0216, 2.3147, 1.8303, 2.8987], + device='cuda:3'), covar=tensor([0.2050, 0.2660, 0.0890, 0.4728, 0.1852, 0.3202, 0.2464, 0.2445], + device='cuda:3'), in_proj_covar=tensor([0.0532, 0.0619, 0.0557, 0.0653, 0.0651, 0.0601, 0.0549, 0.0639], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 07:44:29,519 INFO [train.py:901] (3/4) Epoch 24, batch 7150, loss[loss=0.1992, simple_loss=0.2819, pruned_loss=0.05823, over 8447.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.283, pruned_loss=0.05892, over 1609379.02 frames. ], batch size: 24, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:44:30,377 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193059.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:44:40,889 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 +2023-02-07 07:44:54,306 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193093.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:44:56,946 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193097.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:44:58,924 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.669e+02 2.272e+02 2.945e+02 3.915e+02 7.728e+02, threshold=5.890e+02, percent-clipped=4.0 +2023-02-07 07:45:05,031 INFO [train.py:901] (3/4) Epoch 24, batch 7200, loss[loss=0.1949, simple_loss=0.2829, pruned_loss=0.05346, over 8191.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2841, pruned_loss=0.05944, over 1609110.77 frames. ], batch size: 23, lr: 3.12e-03, grad_scale: 8.0 +2023-02-07 07:45:05,478 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-07 07:45:16,880 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193125.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:45:21,518 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193132.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:45:34,266 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193150.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:45:34,289 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193150.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:45:39,619 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193157.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:45:40,080 INFO [train.py:901] (3/4) Epoch 24, batch 7250, loss[loss=0.2287, simple_loss=0.3167, pruned_loss=0.0704, over 8305.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2834, pruned_loss=0.05907, over 1610606.58 frames. ], batch size: 25, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:45:45,721 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193166.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:46:03,209 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193191.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:46:04,351 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193193.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:46:08,899 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.435e+02 2.389e+02 2.780e+02 3.377e+02 1.311e+03, threshold=5.561e+02, percent-clipped=2.0 +2023-02-07 07:46:14,404 INFO [train.py:901] (3/4) Epoch 24, batch 7300, loss[loss=0.1842, simple_loss=0.2743, pruned_loss=0.04709, over 8246.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2836, pruned_loss=0.05923, over 1608945.39 frames. ], batch size: 22, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:46:17,214 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193212.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:46:29,134 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193229.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:46:48,076 INFO [train.py:901] (3/4) Epoch 24, batch 7350, loss[loss=0.1487, simple_loss=0.2291, pruned_loss=0.03417, over 8232.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2844, pruned_loss=0.05971, over 1608665.41 frames. ], batch size: 22, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:46:54,300 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1081, 2.3511, 1.9054, 2.9419, 1.4728, 1.7017, 2.1996, 2.2881], + device='cuda:3'), covar=tensor([0.0713, 0.0762, 0.0890, 0.0311, 0.0960, 0.1189, 0.0703, 0.0658], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0197, 0.0245, 0.0214, 0.0205, 0.0247, 0.0251, 0.0207], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 07:47:11,631 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-07 07:47:17,799 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.681e+02 2.477e+02 2.971e+02 3.853e+02 6.522e+02, threshold=5.942e+02, percent-clipped=4.0 +2023-02-07 07:47:19,303 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193302.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:47:23,212 INFO [train.py:901] (3/4) Epoch 24, batch 7400, loss[loss=0.2009, simple_loss=0.286, pruned_loss=0.05792, over 8147.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2835, pruned_loss=0.05906, over 1606748.73 frames. ], batch size: 22, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:47:24,067 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193308.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:47:31,912 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-07 07:47:51,689 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1816, 4.1672, 3.7497, 1.9809, 3.6805, 3.8286, 3.7224, 3.7002], + device='cuda:3'), covar=tensor([0.0757, 0.0565, 0.1004, 0.4499, 0.0875, 0.0988, 0.1302, 0.0775], + device='cuda:3'), in_proj_covar=tensor([0.0539, 0.0455, 0.0444, 0.0557, 0.0443, 0.0458, 0.0439, 0.0398], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 07:47:52,397 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193349.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:47:58,366 INFO [train.py:901] (3/4) Epoch 24, batch 7450, loss[loss=0.2017, simple_loss=0.295, pruned_loss=0.05415, over 8465.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2841, pruned_loss=0.05973, over 1607058.78 frames. ], batch size: 25, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:48:09,531 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-07 07:48:09,632 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.1826, 3.0924, 2.8693, 1.9365, 2.7619, 2.8524, 2.8603, 2.7818], + device='cuda:3'), covar=tensor([0.0847, 0.0779, 0.1025, 0.3561, 0.0980, 0.1202, 0.1273, 0.0931], + device='cuda:3'), in_proj_covar=tensor([0.0538, 0.0454, 0.0444, 0.0557, 0.0442, 0.0457, 0.0438, 0.0398], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 07:48:10,325 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193374.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:48:11,141 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 +2023-02-07 07:48:27,885 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1446, 4.1379, 3.7277, 1.9555, 3.6367, 3.7587, 3.7430, 3.6623], + device='cuda:3'), covar=tensor([0.0793, 0.0599, 0.1088, 0.4655, 0.0951, 0.1295, 0.1310, 0.0761], + device='cuda:3'), in_proj_covar=tensor([0.0539, 0.0454, 0.0443, 0.0556, 0.0442, 0.0456, 0.0437, 0.0397], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 07:48:28,441 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.637e+02 2.340e+02 2.930e+02 4.048e+02 8.147e+02, threshold=5.861e+02, percent-clipped=5.0 +2023-02-07 07:48:30,638 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193403.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:48:32,906 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193406.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:48:34,028 INFO [train.py:901] (3/4) Epoch 24, batch 7500, loss[loss=0.1315, simple_loss=0.2123, pruned_loss=0.02536, over 7424.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2852, pruned_loss=0.0605, over 1608101.25 frames. ], batch size: 17, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:48:34,154 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3516, 2.2789, 2.1554, 1.2615, 2.0785, 2.1203, 2.0994, 2.0415], + device='cuda:3'), covar=tensor([0.1022, 0.0865, 0.1037, 0.3598, 0.0935, 0.1244, 0.1391, 0.0961], + device='cuda:3'), in_proj_covar=tensor([0.0538, 0.0454, 0.0442, 0.0556, 0.0441, 0.0456, 0.0436, 0.0397], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 07:48:50,715 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193431.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:48:56,223 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193439.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:49:09,713 INFO [train.py:901] (3/4) Epoch 24, batch 7550, loss[loss=0.2713, simple_loss=0.3465, pruned_loss=0.09812, over 8627.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2849, pruned_loss=0.06028, over 1605137.65 frames. ], batch size: 39, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:49:16,840 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193468.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:49:19,413 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193472.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:49:34,407 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193493.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:49:35,003 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1928, 4.0957, 3.7969, 1.8801, 3.7202, 3.8205, 3.7167, 3.6955], + device='cuda:3'), covar=tensor([0.0802, 0.0598, 0.0984, 0.4773, 0.0947, 0.1036, 0.1316, 0.0779], + device='cuda:3'), in_proj_covar=tensor([0.0532, 0.0449, 0.0438, 0.0551, 0.0437, 0.0452, 0.0433, 0.0394], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 07:49:39,065 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.434e+02 2.474e+02 3.046e+02 3.751e+02 6.843e+02, threshold=6.092e+02, percent-clipped=3.0 +2023-02-07 07:49:45,274 INFO [train.py:901] (3/4) Epoch 24, batch 7600, loss[loss=0.2202, simple_loss=0.2923, pruned_loss=0.07401, over 8289.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2856, pruned_loss=0.06051, over 1610862.43 frames. ], batch size: 23, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:49:51,993 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193518.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:50:16,721 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8259, 3.7515, 3.4506, 1.8553, 3.3886, 3.5096, 3.3688, 3.3838], + device='cuda:3'), covar=tensor([0.0912, 0.0676, 0.1068, 0.4710, 0.0966, 0.1085, 0.1451, 0.0962], + device='cuda:3'), in_proj_covar=tensor([0.0533, 0.0449, 0.0437, 0.0551, 0.0436, 0.0452, 0.0433, 0.0394], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 07:50:19,357 INFO [train.py:901] (3/4) Epoch 24, batch 7650, loss[loss=0.1959, simple_loss=0.2862, pruned_loss=0.05276, over 8458.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2866, pruned_loss=0.06105, over 1612944.79 frames. ], batch size: 25, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:50:24,438 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193564.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:50:27,915 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 +2023-02-07 07:50:30,315 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193573.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:50:30,442 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9537, 3.4315, 2.2775, 2.8235, 2.7460, 1.9900, 2.6294, 2.8579], + device='cuda:3'), covar=tensor([0.1856, 0.0413, 0.1209, 0.0802, 0.0750, 0.1570, 0.1283, 0.1305], + device='cuda:3'), in_proj_covar=tensor([0.0356, 0.0235, 0.0337, 0.0308, 0.0301, 0.0342, 0.0347, 0.0320], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 07:50:41,159 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193589.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:50:48,584 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.797e+02 2.626e+02 3.196e+02 4.372e+02 7.437e+02, threshold=6.392e+02, percent-clipped=4.0 +2023-02-07 07:50:53,957 INFO [train.py:901] (3/4) Epoch 24, batch 7700, loss[loss=0.1749, simple_loss=0.2474, pruned_loss=0.05124, over 7552.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2865, pruned_loss=0.06093, over 1613238.53 frames. ], batch size: 18, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:51:11,794 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193633.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:51:14,886 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-07 07:51:20,929 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193646.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:51:29,073 INFO [train.py:901] (3/4) Epoch 24, batch 7750, loss[loss=0.1996, simple_loss=0.2774, pruned_loss=0.06086, over 7961.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2871, pruned_loss=0.06137, over 1613452.43 frames. ], batch size: 21, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:51:50,186 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193688.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:51:58,097 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.447e+02 2.679e+02 3.147e+02 3.999e+02 8.742e+02, threshold=6.294e+02, percent-clipped=3.0 +2023-02-07 07:52:03,354 INFO [train.py:901] (3/4) Epoch 24, batch 7800, loss[loss=0.1978, simple_loss=0.2928, pruned_loss=0.05135, over 8025.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2857, pruned_loss=0.06055, over 1612492.97 frames. ], batch size: 22, lr: 3.12e-03, grad_scale: 16.0 +2023-02-07 07:52:05,695 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.58 vs. limit=2.0 +2023-02-07 07:52:07,958 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1547, 1.7945, 3.4854, 1.5926, 2.3793, 3.8180, 3.9751, 3.3235], + device='cuda:3'), covar=tensor([0.1116, 0.1657, 0.0327, 0.2073, 0.1119, 0.0240, 0.0613, 0.0539], + device='cuda:3'), in_proj_covar=tensor([0.0298, 0.0322, 0.0287, 0.0316, 0.0315, 0.0272, 0.0429, 0.0302], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 07:52:37,270 INFO [train.py:901] (3/4) Epoch 24, batch 7850, loss[loss=0.2228, simple_loss=0.304, pruned_loss=0.07085, over 8511.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2865, pruned_loss=0.06124, over 1609736.37 frames. ], batch size: 26, lr: 3.11e-03, grad_scale: 16.0 +2023-02-07 07:52:39,505 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193761.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:52:41,748 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 +2023-02-07 07:52:48,222 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193774.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:52:54,180 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193783.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:53:05,050 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193799.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:53:05,510 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.568e+02 2.471e+02 2.801e+02 3.652e+02 8.352e+02, threshold=5.603e+02, percent-clipped=2.0 +2023-02-07 07:53:05,882 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7261, 1.7391, 2.5619, 2.0430, 2.2813, 1.8183, 1.5619, 1.1427], + device='cuda:3'), covar=tensor([0.8337, 0.6720, 0.2411, 0.4561, 0.3816, 0.5087, 0.3579, 0.6439], + device='cuda:3'), in_proj_covar=tensor([0.0950, 0.1001, 0.0824, 0.0969, 0.1009, 0.0911, 0.0762, 0.0836], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 07:53:10,834 INFO [train.py:901] (3/4) Epoch 24, batch 7900, loss[loss=0.2261, simple_loss=0.2908, pruned_loss=0.08072, over 7527.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2867, pruned_loss=0.0612, over 1616631.52 frames. ], batch size: 18, lr: 3.11e-03, grad_scale: 16.0 +2023-02-07 07:53:16,270 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193816.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:53:30,971 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5105, 2.3070, 4.0129, 1.5491, 2.8277, 2.0554, 1.8772, 2.7359], + device='cuda:3'), covar=tensor([0.2639, 0.3003, 0.0905, 0.5468, 0.2219, 0.4026, 0.2822, 0.2831], + device='cuda:3'), in_proj_covar=tensor([0.0528, 0.0615, 0.0554, 0.0649, 0.0650, 0.0597, 0.0546, 0.0634], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 07:53:43,829 INFO [train.py:901] (3/4) Epoch 24, batch 7950, loss[loss=0.2074, simple_loss=0.3066, pruned_loss=0.05413, over 8304.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2871, pruned_loss=0.06106, over 1619945.75 frames. ], batch size: 23, lr: 3.11e-03, grad_scale: 16.0 +2023-02-07 07:53:48,055 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2475, 3.2101, 2.9319, 1.6984, 2.8723, 2.8881, 2.8773, 2.8479], + device='cuda:3'), covar=tensor([0.1131, 0.0816, 0.1286, 0.4329, 0.1163, 0.1642, 0.1507, 0.1095], + device='cuda:3'), in_proj_covar=tensor([0.0533, 0.0451, 0.0437, 0.0552, 0.0438, 0.0454, 0.0433, 0.0396], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 07:54:11,271 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193898.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:54:12,511 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.518e+02 2.494e+02 3.061e+02 3.521e+02 6.741e+02, threshold=6.122e+02, percent-clipped=2.0 +2023-02-07 07:54:13,964 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193902.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:54:17,826 INFO [train.py:901] (3/4) Epoch 24, batch 8000, loss[loss=0.195, simple_loss=0.2679, pruned_loss=0.06103, over 7426.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2867, pruned_loss=0.06063, over 1614335.05 frames. ], batch size: 17, lr: 3.11e-03, grad_scale: 16.0 +2023-02-07 07:54:33,398 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193931.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:54:42,133 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193944.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:54:51,058 INFO [train.py:901] (3/4) Epoch 24, batch 8050, loss[loss=0.1661, simple_loss=0.2562, pruned_loss=0.03795, over 5966.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2839, pruned_loss=0.06, over 1591724.55 frames. ], batch size: 13, lr: 3.11e-03, grad_scale: 16.0 +2023-02-07 07:54:58,169 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193969.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:55:03,441 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=193977.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:55:23,316 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-07 07:55:28,454 INFO [train.py:901] (3/4) Epoch 25, batch 0, loss[loss=0.2473, simple_loss=0.3091, pruned_loss=0.09273, over 8141.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3091, pruned_loss=0.09273, over 8141.00 frames. ], batch size: 22, lr: 3.05e-03, grad_scale: 16.0 +2023-02-07 07:55:28,454 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 07:55:39,670 INFO [train.py:935] (3/4) Epoch 25, validation: loss=0.1722, simple_loss=0.2724, pruned_loss=0.03604, over 944034.00 frames. +2023-02-07 07:55:39,672 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6747MB +2023-02-07 07:55:46,488 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.408e+02 2.577e+02 3.086e+02 3.975e+02 9.885e+02, threshold=6.172e+02, percent-clipped=3.0 +2023-02-07 07:55:57,050 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-07 07:56:00,149 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194017.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:56:15,973 INFO [train.py:901] (3/4) Epoch 25, batch 50, loss[loss=0.2244, simple_loss=0.308, pruned_loss=0.07046, over 8106.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2808, pruned_loss=0.05579, over 365068.30 frames. ], batch size: 23, lr: 3.05e-03, grad_scale: 16.0 +2023-02-07 07:56:17,563 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194042.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:56:32,518 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-07 07:56:51,125 INFO [train.py:901] (3/4) Epoch 25, batch 100, loss[loss=0.2492, simple_loss=0.337, pruned_loss=0.0807, over 8281.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2852, pruned_loss=0.06028, over 640271.74 frames. ], batch size: 23, lr: 3.05e-03, grad_scale: 16.0 +2023-02-07 07:56:51,283 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194090.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:56:52,674 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194092.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:56:55,675 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-07 07:56:57,725 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.530e+02 2.655e+02 3.251e+02 4.247e+02 7.218e+02, threshold=6.502e+02, percent-clipped=2.0 +2023-02-07 07:57:19,066 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-07 07:57:22,506 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-07 07:57:25,375 INFO [train.py:901] (3/4) Epoch 25, batch 150, loss[loss=0.1934, simple_loss=0.2822, pruned_loss=0.05236, over 8293.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2832, pruned_loss=0.05917, over 855942.28 frames. ], batch size: 23, lr: 3.05e-03, grad_scale: 16.0 +2023-02-07 07:57:35,089 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194154.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:57:52,150 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194179.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:57:58,884 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194187.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 07:58:00,655 INFO [train.py:901] (3/4) Epoch 25, batch 200, loss[loss=0.1699, simple_loss=0.2425, pruned_loss=0.04861, over 6794.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2838, pruned_loss=0.05928, over 1020252.45 frames. ], batch size: 15, lr: 3.05e-03, grad_scale: 16.0 +2023-02-07 07:58:07,396 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.663e+02 2.343e+02 2.842e+02 3.543e+02 5.999e+02, threshold=5.685e+02, percent-clipped=0.0 +2023-02-07 07:58:16,692 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194212.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 07:58:35,372 INFO [train.py:901] (3/4) Epoch 25, batch 250, loss[loss=0.1622, simple_loss=0.2541, pruned_loss=0.03509, over 7964.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2844, pruned_loss=0.05963, over 1153430.35 frames. ], batch size: 21, lr: 3.05e-03, grad_scale: 16.0 +2023-02-07 07:58:39,449 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=194246.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:58:49,482 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-07 07:58:58,163 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-07 07:59:08,685 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 +2023-02-07 07:59:09,586 INFO [train.py:901] (3/4) Epoch 25, batch 300, loss[loss=0.2221, simple_loss=0.3094, pruned_loss=0.06741, over 8438.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2841, pruned_loss=0.05881, over 1257804.65 frames. ], batch size: 49, lr: 3.05e-03, grad_scale: 16.0 +2023-02-07 07:59:17,104 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.595e+02 2.353e+02 2.857e+02 3.504e+02 7.851e+02, threshold=5.715e+02, percent-clipped=2.0 +2023-02-07 07:59:20,167 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7971, 1.3972, 3.4294, 1.5654, 2.5322, 3.6718, 3.7961, 3.1588], + device='cuda:3'), covar=tensor([0.1304, 0.1864, 0.0290, 0.2045, 0.0866, 0.0237, 0.0584, 0.0539], + device='cuda:3'), in_proj_covar=tensor([0.0298, 0.0322, 0.0287, 0.0316, 0.0316, 0.0273, 0.0429, 0.0303], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 07:59:43,445 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194336.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 07:59:45,900 INFO [train.py:901] (3/4) Epoch 25, batch 350, loss[loss=0.2254, simple_loss=0.306, pruned_loss=0.07242, over 7927.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2853, pruned_loss=0.05912, over 1340854.27 frames. ], batch size: 20, lr: 3.05e-03, grad_scale: 16.0 +2023-02-07 07:59:51,529 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194348.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:00:00,427 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194361.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:00:08,062 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194371.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:00:09,455 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194373.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:00:14,173 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194380.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 08:00:20,655 INFO [train.py:901] (3/4) Epoch 25, batch 400, loss[loss=0.1886, simple_loss=0.2707, pruned_loss=0.05327, over 7801.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2859, pruned_loss=0.05989, over 1401370.25 frames. ], batch size: 20, lr: 3.05e-03, grad_scale: 16.0 +2023-02-07 08:00:27,619 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.733e+02 2.445e+02 3.013e+02 3.982e+02 8.525e+02, threshold=6.027e+02, percent-clipped=7.0 +2023-02-07 08:00:29,489 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.45 vs. limit=2.0 +2023-02-07 08:00:52,194 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=194434.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:00:56,959 INFO [train.py:901] (3/4) Epoch 25, batch 450, loss[loss=0.193, simple_loss=0.2812, pruned_loss=0.05241, over 8444.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2856, pruned_loss=0.05977, over 1446515.95 frames. ], batch size: 25, lr: 3.05e-03, grad_scale: 16.0 +2023-02-07 08:01:30,915 INFO [train.py:901] (3/4) Epoch 25, batch 500, loss[loss=0.2424, simple_loss=0.3207, pruned_loss=0.08211, over 8689.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.286, pruned_loss=0.06025, over 1486849.89 frames. ], batch size: 34, lr: 3.04e-03, grad_scale: 16.0 +2023-02-07 08:01:37,840 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.666e+02 2.459e+02 3.156e+02 4.025e+02 7.800e+02, threshold=6.312e+02, percent-clipped=3.0 +2023-02-07 08:02:06,149 INFO [train.py:901] (3/4) Epoch 25, batch 550, loss[loss=0.1834, simple_loss=0.2759, pruned_loss=0.04549, over 8327.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2849, pruned_loss=0.05942, over 1514495.81 frames. ], batch size: 25, lr: 3.04e-03, grad_scale: 16.0 +2023-02-07 08:02:13,482 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194549.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:02:30,143 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2709, 2.4915, 2.7376, 1.6652, 2.9476, 1.7354, 1.6576, 2.2254], + device='cuda:3'), covar=tensor([0.0751, 0.0419, 0.0327, 0.0737, 0.0485, 0.0923, 0.0862, 0.0503], + device='cuda:3'), in_proj_covar=tensor([0.0459, 0.0401, 0.0355, 0.0450, 0.0387, 0.0537, 0.0398, 0.0431], + device='cuda:3'), out_proj_covar=tensor([1.2215e-04, 1.0461e-04, 9.3020e-05, 1.1807e-04, 1.0144e-04, 1.5059e-04, + 1.0673e-04, 1.1351e-04], device='cuda:3') +2023-02-07 08:02:42,129 INFO [train.py:901] (3/4) Epoch 25, batch 600, loss[loss=0.222, simple_loss=0.2903, pruned_loss=0.07689, over 7095.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2845, pruned_loss=0.05931, over 1537019.48 frames. ], batch size: 72, lr: 3.04e-03, grad_scale: 16.0 +2023-02-07 08:02:48,738 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.589e+02 2.361e+02 2.970e+02 3.663e+02 1.001e+03, threshold=5.941e+02, percent-clipped=3.0 +2023-02-07 08:03:01,136 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-07 08:03:01,340 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194617.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:03:10,866 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0725, 1.3927, 1.7152, 1.4078, 1.0345, 1.4931, 1.8987, 1.6922], + device='cuda:3'), covar=tensor([0.0517, 0.1349, 0.1676, 0.1472, 0.0618, 0.1584, 0.0691, 0.0669], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0153, 0.0189, 0.0160, 0.0101, 0.0163, 0.0112, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 08:03:16,782 INFO [train.py:901] (3/4) Epoch 25, batch 650, loss[loss=0.1852, simple_loss=0.2771, pruned_loss=0.04664, over 8360.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2836, pruned_loss=0.05901, over 1553532.67 frames. ], batch size: 24, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:03:18,074 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194642.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:03:23,750 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4859, 2.2874, 2.8627, 2.5076, 2.8610, 2.4733, 2.3559, 2.1383], + device='cuda:3'), covar=tensor([0.3679, 0.3644, 0.1590, 0.2724, 0.1724, 0.2348, 0.1434, 0.3505], + device='cuda:3'), in_proj_covar=tensor([0.0942, 0.0995, 0.0814, 0.0961, 0.1001, 0.0904, 0.0754, 0.0830], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 08:03:25,638 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194652.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:03:45,808 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=194680.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:03:52,562 INFO [train.py:901] (3/4) Epoch 25, batch 700, loss[loss=0.1932, simple_loss=0.2795, pruned_loss=0.05344, over 8112.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2845, pruned_loss=0.05913, over 1570411.70 frames. ], batch size: 23, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:04:00,038 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.681e+02 2.448e+02 2.849e+02 3.638e+02 5.412e+02, threshold=5.698e+02, percent-clipped=0.0 +2023-02-07 08:04:09,652 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=194715.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:04:16,431 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=194724.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 08:04:27,405 INFO [train.py:901] (3/4) Epoch 25, batch 750, loss[loss=0.2111, simple_loss=0.2922, pruned_loss=0.06497, over 8282.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2846, pruned_loss=0.05928, over 1580083.11 frames. ], batch size: 23, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:04:49,459 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-07 08:04:58,544 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-07 08:05:03,377 INFO [train.py:901] (3/4) Epoch 25, batch 800, loss[loss=0.2021, simple_loss=0.3002, pruned_loss=0.05205, over 8105.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2851, pruned_loss=0.05958, over 1587988.27 frames. ], batch size: 23, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:05:07,606 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194795.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:05:11,591 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.325e+02 2.417e+02 2.991e+02 3.771e+02 6.788e+02, threshold=5.982e+02, percent-clipped=2.0 +2023-02-07 08:05:14,543 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194805.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:05:31,800 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194830.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:05:31,829 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194830.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:05:37,857 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194839.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 08:05:38,297 INFO [train.py:901] (3/4) Epoch 25, batch 850, loss[loss=0.2169, simple_loss=0.2954, pruned_loss=0.06919, over 8180.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2856, pruned_loss=0.06014, over 1590782.78 frames. ], batch size: 23, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:05:49,042 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7610, 2.6533, 1.9986, 2.5045, 2.3015, 1.6735, 2.1990, 2.3087], + device='cuda:3'), covar=tensor([0.1479, 0.0441, 0.1179, 0.0625, 0.0835, 0.1649, 0.1107, 0.1069], + device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0234, 0.0335, 0.0307, 0.0298, 0.0339, 0.0344, 0.0318], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 08:05:56,613 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194865.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:06:14,016 INFO [train.py:901] (3/4) Epoch 25, batch 900, loss[loss=0.191, simple_loss=0.2858, pruned_loss=0.04814, over 8464.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2849, pruned_loss=0.06001, over 1589640.56 frames. ], batch size: 27, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:06:22,179 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.549e+02 2.491e+02 2.923e+02 3.701e+02 8.623e+02, threshold=5.846e+02, percent-clipped=3.0 +2023-02-07 08:06:49,618 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 +2023-02-07 08:06:49,724 INFO [train.py:901] (3/4) Epoch 25, batch 950, loss[loss=0.2063, simple_loss=0.2929, pruned_loss=0.05984, over 8292.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2845, pruned_loss=0.05978, over 1596253.63 frames. ], batch size: 23, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:07:02,589 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7096, 1.7951, 1.9630, 1.9247, 1.2333, 1.7781, 2.4222, 2.2586], + device='cuda:3'), covar=tensor([0.0493, 0.1201, 0.1640, 0.1312, 0.0558, 0.1409, 0.0560, 0.0570], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0153, 0.0191, 0.0160, 0.0101, 0.0164, 0.0112, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 08:07:19,538 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-07 08:07:24,291 INFO [train.py:901] (3/4) Epoch 25, batch 1000, loss[loss=0.2164, simple_loss=0.3022, pruned_loss=0.06528, over 8759.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2844, pruned_loss=0.05976, over 1600291.69 frames. ], batch size: 30, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:07:29,064 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=194996.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:07:32,166 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.541e+02 2.651e+02 3.101e+02 3.894e+02 6.477e+02, threshold=6.202e+02, percent-clipped=4.0 +2023-02-07 08:07:51,786 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195029.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:07:54,473 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-07 08:07:59,733 INFO [train.py:901] (3/4) Epoch 25, batch 1050, loss[loss=0.1892, simple_loss=0.2807, pruned_loss=0.04883, over 8242.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2862, pruned_loss=0.06084, over 1608161.65 frames. ], batch size: 24, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:08:06,394 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-07 08:08:07,162 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195051.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:08:14,490 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195062.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 08:08:23,833 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195076.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:08:23,887 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9451, 1.7170, 2.0448, 1.8572, 1.9982, 1.9778, 1.7992, 0.8681], + device='cuda:3'), covar=tensor([0.5534, 0.4369, 0.2036, 0.3387, 0.2222, 0.3098, 0.2030, 0.4703], + device='cuda:3'), in_proj_covar=tensor([0.0950, 0.1003, 0.0824, 0.0968, 0.1009, 0.0914, 0.0761, 0.0836], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 08:08:31,308 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195086.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:08:33,316 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195089.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:08:33,850 INFO [train.py:901] (3/4) Epoch 25, batch 1100, loss[loss=0.1602, simple_loss=0.2339, pruned_loss=0.04323, over 7646.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.286, pruned_loss=0.0609, over 1610486.67 frames. ], batch size: 19, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:08:36,611 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8732, 1.4713, 3.5001, 1.4737, 2.4276, 3.8007, 3.9229, 3.2786], + device='cuda:3'), covar=tensor([0.1223, 0.1843, 0.0307, 0.2110, 0.1008, 0.0241, 0.0561, 0.0532], + device='cuda:3'), in_proj_covar=tensor([0.0300, 0.0324, 0.0288, 0.0317, 0.0316, 0.0275, 0.0430, 0.0304], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 08:08:37,387 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195095.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 08:08:41,229 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.797e+02 2.550e+02 3.152e+02 4.111e+02 6.650e+02, threshold=6.304e+02, percent-clipped=3.0 +2023-02-07 08:08:48,180 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195111.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:08:48,202 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195111.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:08:53,479 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.9089, 3.8774, 3.5129, 2.0384, 3.4447, 3.5263, 3.5150, 3.3789], + device='cuda:3'), covar=tensor([0.0883, 0.0645, 0.1110, 0.4146, 0.1068, 0.1255, 0.1294, 0.0957], + device='cuda:3'), in_proj_covar=tensor([0.0531, 0.0450, 0.0432, 0.0545, 0.0434, 0.0452, 0.0426, 0.0396], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 08:08:54,917 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195120.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 08:09:09,221 INFO [train.py:901] (3/4) Epoch 25, batch 1150, loss[loss=0.2549, simple_loss=0.3374, pruned_loss=0.08616, over 8483.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2853, pruned_loss=0.06042, over 1609383.51 frames. ], batch size: 29, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:09:16,851 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-07 08:09:43,696 INFO [train.py:901] (3/4) Epoch 25, batch 1200, loss[loss=0.18, simple_loss=0.271, pruned_loss=0.0445, over 8034.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2863, pruned_loss=0.06059, over 1614880.21 frames. ], batch size: 22, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:09:44,707 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-07 08:09:51,815 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.628e+02 2.229e+02 2.843e+02 3.492e+02 1.399e+03, threshold=5.685e+02, percent-clipped=2.0 +2023-02-07 08:09:57,121 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=195209.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:10:18,384 INFO [train.py:901] (3/4) Epoch 25, batch 1250, loss[loss=0.1666, simple_loss=0.2535, pruned_loss=0.03983, over 8086.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2862, pruned_loss=0.06095, over 1615127.70 frames. ], batch size: 21, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:10:34,136 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.76 vs. limit=5.0 +2023-02-07 08:10:53,103 INFO [train.py:901] (3/4) Epoch 25, batch 1300, loss[loss=0.1899, simple_loss=0.287, pruned_loss=0.04638, over 8030.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2858, pruned_loss=0.06077, over 1613352.42 frames. ], batch size: 22, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:11:00,279 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.776e+02 2.452e+02 2.850e+02 4.025e+02 1.071e+03, threshold=5.700e+02, percent-clipped=7.0 +2023-02-07 08:11:16,286 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195324.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:11:26,834 INFO [train.py:901] (3/4) Epoch 25, batch 1350, loss[loss=0.2694, simple_loss=0.3557, pruned_loss=0.09157, over 8187.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2863, pruned_loss=0.06109, over 1613402.25 frames. ], batch size: 23, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:11:45,671 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195367.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:11:50,113 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=195373.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:11:57,384 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 +2023-02-07 08:12:02,400 INFO [train.py:901] (3/4) Epoch 25, batch 1400, loss[loss=0.1881, simple_loss=0.2698, pruned_loss=0.05316, over 7650.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2867, pruned_loss=0.06096, over 1616819.97 frames. ], batch size: 19, lr: 3.04e-03, grad_scale: 4.0 +2023-02-07 08:12:03,997 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195392.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:12:10,479 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.598e+02 2.570e+02 2.915e+02 3.833e+02 8.465e+02, threshold=5.831e+02, percent-clipped=6.0 +2023-02-07 08:12:13,227 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=195406.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 08:12:15,768 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195410.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:12:31,395 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=195433.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:12:36,118 INFO [train.py:901] (3/4) Epoch 25, batch 1450, loss[loss=0.1702, simple_loss=0.2516, pruned_loss=0.04443, over 7810.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2882, pruned_loss=0.06196, over 1617204.38 frames. ], batch size: 19, lr: 3.04e-03, grad_scale: 4.0 +2023-02-07 08:12:36,363 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6792, 1.6418, 2.2517, 1.4687, 1.3452, 2.2681, 0.5001, 1.4737], + device='cuda:3'), covar=tensor([0.1901, 0.1216, 0.0386, 0.1180, 0.2273, 0.0362, 0.1930, 0.1280], + device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0201, 0.0130, 0.0221, 0.0273, 0.0139, 0.0171, 0.0196], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 08:12:44,148 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-07 08:13:10,181 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195488.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:13:11,407 INFO [train.py:901] (3/4) Epoch 25, batch 1500, loss[loss=0.1805, simple_loss=0.2655, pruned_loss=0.04781, over 8688.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2875, pruned_loss=0.06106, over 1617406.31 frames. ], batch size: 34, lr: 3.04e-03, grad_scale: 4.0 +2023-02-07 08:13:19,802 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.596e+02 2.401e+02 3.375e+02 4.255e+02 1.024e+03, threshold=6.749e+02, percent-clipped=12.0 +2023-02-07 08:13:33,938 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195521.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 08:13:46,247 INFO [train.py:901] (3/4) Epoch 25, batch 1550, loss[loss=0.1976, simple_loss=0.2909, pruned_loss=0.05218, over 8254.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2862, pruned_loss=0.06008, over 1617633.90 frames. ], batch size: 24, lr: 3.04e-03, grad_scale: 4.0 +2023-02-07 08:13:51,916 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195548.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:14:14,119 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195580.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:14:21,360 INFO [train.py:901] (3/4) Epoch 25, batch 1600, loss[loss=0.1987, simple_loss=0.2857, pruned_loss=0.05587, over 8497.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.287, pruned_loss=0.06006, over 1622883.47 frames. ], batch size: 28, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:14:29,486 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.537e+02 2.317e+02 3.105e+02 3.813e+02 7.132e+02, threshold=6.211e+02, percent-clipped=3.0 +2023-02-07 08:14:32,324 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195605.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:14:37,017 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1195, 2.0197, 2.6098, 2.1342, 2.6243, 2.2297, 2.0598, 1.5487], + device='cuda:3'), covar=tensor([0.5967, 0.5298, 0.2123, 0.4324, 0.2879, 0.3138, 0.2153, 0.5681], + device='cuda:3'), in_proj_covar=tensor([0.0954, 0.1009, 0.0826, 0.0977, 0.1019, 0.0919, 0.0765, 0.0842], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 08:14:55,975 INFO [train.py:901] (3/4) Epoch 25, batch 1650, loss[loss=0.2177, simple_loss=0.2914, pruned_loss=0.07201, over 7063.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2862, pruned_loss=0.06009, over 1615488.11 frames. ], batch size: 71, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:15:29,744 INFO [train.py:901] (3/4) Epoch 25, batch 1700, loss[loss=0.21, simple_loss=0.2775, pruned_loss=0.07127, over 7664.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2855, pruned_loss=0.05968, over 1616854.08 frames. ], batch size: 19, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:15:38,030 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.465e+02 2.590e+02 3.116e+02 3.996e+02 7.880e+02, threshold=6.232e+02, percent-clipped=2.0 +2023-02-07 08:16:05,339 INFO [train.py:901] (3/4) Epoch 25, batch 1750, loss[loss=0.2205, simple_loss=0.3073, pruned_loss=0.0668, over 8200.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.286, pruned_loss=0.06026, over 1616652.48 frames. ], batch size: 23, lr: 3.04e-03, grad_scale: 8.0 +2023-02-07 08:16:06,318 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1678, 2.4064, 1.9945, 2.9936, 1.4866, 1.8196, 2.2196, 2.3431], + device='cuda:3'), covar=tensor([0.0606, 0.0725, 0.0827, 0.0303, 0.1001, 0.1147, 0.0713, 0.0773], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0196, 0.0245, 0.0213, 0.0205, 0.0245, 0.0249, 0.0206], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 08:16:09,112 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195744.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:16:15,704 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=195754.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:16:17,276 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2385, 2.0174, 2.6215, 2.1033, 2.6571, 2.3072, 2.0839, 1.4801], + device='cuda:3'), covar=tensor([0.5607, 0.4943, 0.1983, 0.3859, 0.2469, 0.3009, 0.1920, 0.5303], + device='cuda:3'), in_proj_covar=tensor([0.0953, 0.1008, 0.0825, 0.0976, 0.1016, 0.0917, 0.0764, 0.0841], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 08:16:26,012 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195769.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:16:28,849 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0703, 1.2772, 1.2161, 0.6946, 1.2331, 1.0595, 0.0841, 1.2170], + device='cuda:3'), covar=tensor([0.0470, 0.0424, 0.0380, 0.0626, 0.0443, 0.1068, 0.0979, 0.0343], + device='cuda:3'), in_proj_covar=tensor([0.0466, 0.0406, 0.0362, 0.0458, 0.0390, 0.0546, 0.0403, 0.0435], + device='cuda:3'), out_proj_covar=tensor([1.2384e-04, 1.0602e-04, 9.4739e-05, 1.2015e-04, 1.0234e-04, 1.5297e-04, + 1.0806e-04, 1.1439e-04], device='cuda:3') +2023-02-07 08:16:31,518 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195777.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 08:16:40,107 INFO [train.py:901] (3/4) Epoch 25, batch 1800, loss[loss=0.1873, simple_loss=0.2733, pruned_loss=0.05068, over 7975.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2866, pruned_loss=0.06074, over 1616732.35 frames. ], batch size: 21, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:16:48,983 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.557e+02 2.456e+02 2.857e+02 3.484e+02 7.816e+02, threshold=5.715e+02, percent-clipped=1.0 +2023-02-07 08:16:49,212 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195802.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 08:16:50,554 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195804.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:17:07,907 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195829.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:17:15,240 INFO [train.py:901] (3/4) Epoch 25, batch 1850, loss[loss=0.2021, simple_loss=0.2862, pruned_loss=0.05896, over 8253.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2861, pruned_loss=0.05983, over 1619352.19 frames. ], batch size: 24, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:17:36,413 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195869.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:17:50,154 INFO [train.py:901] (3/4) Epoch 25, batch 1900, loss[loss=0.1745, simple_loss=0.2674, pruned_loss=0.04073, over 8138.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2853, pruned_loss=0.05965, over 1615068.47 frames. ], batch size: 22, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:17:58,370 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.976e+02 2.686e+02 3.045e+02 3.689e+02 8.196e+02, threshold=6.090e+02, percent-clipped=3.0 +2023-02-07 08:18:24,458 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-07 08:18:25,117 INFO [train.py:901] (3/4) Epoch 25, batch 1950, loss[loss=0.1883, simple_loss=0.2676, pruned_loss=0.05452, over 8134.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2847, pruned_loss=0.0587, over 1618003.36 frames. ], batch size: 22, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:18:37,846 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-07 08:18:56,450 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 +2023-02-07 08:18:57,275 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-07 08:18:58,139 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6216, 2.4902, 1.7838, 2.3610, 2.2035, 1.5570, 2.1236, 2.2218], + device='cuda:3'), covar=tensor([0.1520, 0.0431, 0.1280, 0.0634, 0.0743, 0.1622, 0.1113, 0.0929], + device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0237, 0.0341, 0.0312, 0.0302, 0.0345, 0.0349, 0.0323], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 08:19:00,585 INFO [train.py:901] (3/4) Epoch 25, batch 2000, loss[loss=0.2045, simple_loss=0.2707, pruned_loss=0.0692, over 7686.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2852, pruned_loss=0.05893, over 1620769.77 frames. ], batch size: 18, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:19:09,742 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.479e+02 2.344e+02 2.823e+02 3.287e+02 7.423e+02, threshold=5.646e+02, percent-clipped=4.0 +2023-02-07 08:19:36,107 INFO [train.py:901] (3/4) Epoch 25, batch 2050, loss[loss=0.1659, simple_loss=0.2398, pruned_loss=0.04604, over 7523.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2844, pruned_loss=0.05887, over 1616615.32 frames. ], batch size: 18, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:19:55,113 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-07 08:19:56,298 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4346, 1.5731, 2.1603, 1.3458, 1.4942, 1.7041, 1.4463, 1.5042], + device='cuda:3'), covar=tensor([0.1984, 0.2667, 0.0882, 0.4769, 0.2134, 0.3571, 0.2508, 0.2284], + device='cuda:3'), in_proj_covar=tensor([0.0532, 0.0624, 0.0557, 0.0660, 0.0656, 0.0603, 0.0551, 0.0640], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 08:20:11,106 INFO [train.py:901] (3/4) Epoch 25, batch 2100, loss[loss=0.1802, simple_loss=0.2631, pruned_loss=0.04869, over 7655.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2844, pruned_loss=0.059, over 1615949.27 frames. ], batch size: 19, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:20:20,386 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.711e+02 2.403e+02 2.946e+02 3.659e+02 8.101e+02, threshold=5.892e+02, percent-clipped=3.0 +2023-02-07 08:20:35,890 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=196125.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:20:46,048 INFO [train.py:901] (3/4) Epoch 25, batch 2150, loss[loss=0.1856, simple_loss=0.2598, pruned_loss=0.05576, over 7417.00 frames. ], tot_loss[loss=0.202, simple_loss=0.285, pruned_loss=0.05945, over 1614199.11 frames. ], batch size: 17, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:20:54,033 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=196150.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:21:22,082 INFO [train.py:901] (3/4) Epoch 25, batch 2200, loss[loss=0.1719, simple_loss=0.268, pruned_loss=0.03791, over 8358.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2848, pruned_loss=0.05971, over 1615740.37 frames. ], batch size: 24, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:21:30,646 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.548e+02 2.555e+02 3.213e+02 4.289e+02 6.887e+02, threshold=6.426e+02, percent-clipped=5.0 +2023-02-07 08:21:37,557 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.8582, 5.9972, 5.1801, 2.8007, 5.2756, 5.6252, 5.4541, 5.4553], + device='cuda:3'), covar=tensor([0.0564, 0.0376, 0.1027, 0.4090, 0.0811, 0.0708, 0.1172, 0.0532], + device='cuda:3'), in_proj_covar=tensor([0.0532, 0.0449, 0.0435, 0.0546, 0.0433, 0.0452, 0.0427, 0.0396], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 08:21:56,957 INFO [train.py:901] (3/4) Epoch 25, batch 2250, loss[loss=0.2207, simple_loss=0.2941, pruned_loss=0.07364, over 8140.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2843, pruned_loss=0.05943, over 1614899.20 frames. ], batch size: 22, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:22:07,582 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9233, 1.2666, 1.5932, 1.1689, 0.9793, 1.3648, 1.7420, 1.6937], + device='cuda:3'), covar=tensor([0.0583, 0.1784, 0.2493, 0.1985, 0.0689, 0.2091, 0.0788, 0.0737], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0152, 0.0190, 0.0160, 0.0100, 0.0164, 0.0112, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 08:22:12,519 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6221, 2.4294, 1.7827, 2.2582, 2.1681, 1.6403, 2.0791, 2.0668], + device='cuda:3'), covar=tensor([0.1335, 0.0438, 0.1319, 0.0598, 0.0700, 0.1551, 0.0968, 0.1010], + device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0238, 0.0341, 0.0313, 0.0303, 0.0345, 0.0350, 0.0323], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 08:22:32,049 INFO [train.py:901] (3/4) Epoch 25, batch 2300, loss[loss=0.2014, simple_loss=0.2913, pruned_loss=0.05572, over 8463.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2835, pruned_loss=0.05854, over 1612118.12 frames. ], batch size: 29, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:22:40,946 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.515e+02 2.420e+02 2.794e+02 3.530e+02 9.865e+02, threshold=5.587e+02, percent-clipped=2.0 +2023-02-07 08:23:04,030 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([0.9971, 1.0135, 0.9637, 1.2059, 0.6122, 0.9100, 0.9585, 1.0361], + device='cuda:3'), covar=tensor([0.0599, 0.0552, 0.0695, 0.0438, 0.0795, 0.0913, 0.0514, 0.0521], + device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0194, 0.0242, 0.0210, 0.0203, 0.0243, 0.0247, 0.0205], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 08:23:07,221 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196339.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:23:07,751 INFO [train.py:901] (3/4) Epoch 25, batch 2350, loss[loss=0.2314, simple_loss=0.3169, pruned_loss=0.07291, over 8522.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2833, pruned_loss=0.05909, over 1606760.97 frames. ], batch size: 28, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:23:34,290 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196378.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:23:42,081 INFO [train.py:901] (3/4) Epoch 25, batch 2400, loss[loss=0.175, simple_loss=0.2689, pruned_loss=0.04058, over 8096.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2822, pruned_loss=0.05844, over 1609851.21 frames. ], batch size: 23, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:23:50,280 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.515e+02 2.420e+02 2.902e+02 3.432e+02 7.434e+02, threshold=5.805e+02, percent-clipped=2.0 +2023-02-07 08:24:17,356 INFO [train.py:901] (3/4) Epoch 25, batch 2450, loss[loss=0.1568, simple_loss=0.2384, pruned_loss=0.03761, over 7958.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2829, pruned_loss=0.05854, over 1612223.92 frames. ], batch size: 21, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:24:51,934 INFO [train.py:901] (3/4) Epoch 25, batch 2500, loss[loss=0.2908, simple_loss=0.3703, pruned_loss=0.1057, over 8746.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2844, pruned_loss=0.05904, over 1614952.05 frames. ], batch size: 30, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:25:00,802 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.834e+02 2.398e+02 2.858e+02 3.242e+02 5.404e+02, threshold=5.717e+02, percent-clipped=0.0 +2023-02-07 08:25:18,288 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1270, 1.3190, 1.5836, 1.2510, 0.7935, 1.3795, 1.0707, 0.9159], + device='cuda:3'), covar=tensor([0.0609, 0.1291, 0.1648, 0.1515, 0.0566, 0.1535, 0.0758, 0.0768], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0153, 0.0190, 0.0161, 0.0101, 0.0164, 0.0112, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 08:25:27,002 INFO [train.py:901] (3/4) Epoch 25, batch 2550, loss[loss=0.2241, simple_loss=0.3011, pruned_loss=0.07351, over 8342.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2847, pruned_loss=0.05911, over 1614895.48 frames. ], batch size: 26, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:25:54,370 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3158, 2.0886, 2.6558, 2.2647, 2.6392, 2.3477, 2.1812, 1.6311], + device='cuda:3'), covar=tensor([0.5731, 0.5161, 0.2238, 0.4185, 0.2666, 0.3405, 0.1983, 0.5576], + device='cuda:3'), in_proj_covar=tensor([0.0952, 0.1004, 0.0824, 0.0974, 0.1014, 0.0915, 0.0761, 0.0837], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 08:26:02,150 INFO [train.py:901] (3/4) Epoch 25, batch 2600, loss[loss=0.2023, simple_loss=0.2882, pruned_loss=0.05817, over 8344.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2848, pruned_loss=0.0591, over 1615856.21 frames. ], batch size: 26, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:26:06,425 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196596.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:26:07,752 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.6130, 5.7218, 4.9959, 2.6510, 5.0705, 5.4191, 5.2825, 5.2059], + device='cuda:3'), covar=tensor([0.0502, 0.0385, 0.0961, 0.3869, 0.0751, 0.0775, 0.1035, 0.0622], + device='cuda:3'), in_proj_covar=tensor([0.0530, 0.0447, 0.0433, 0.0542, 0.0434, 0.0450, 0.0424, 0.0396], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 08:26:10,242 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.789e+02 2.366e+02 2.911e+02 3.287e+02 8.101e+02, threshold=5.822e+02, percent-clipped=1.0 +2023-02-07 08:26:13,505 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 +2023-02-07 08:26:37,069 INFO [train.py:901] (3/4) Epoch 25, batch 2650, loss[loss=0.2055, simple_loss=0.295, pruned_loss=0.05797, over 8660.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2849, pruned_loss=0.05919, over 1616639.56 frames. ], batch size: 34, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:27:08,210 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=196683.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:27:12,709 INFO [train.py:901] (3/4) Epoch 25, batch 2700, loss[loss=0.1855, simple_loss=0.2512, pruned_loss=0.05995, over 7545.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2849, pruned_loss=0.05968, over 1612399.96 frames. ], batch size: 18, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:27:20,578 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.512e+02 2.452e+02 2.909e+02 3.648e+02 8.771e+02, threshold=5.818e+02, percent-clipped=3.0 +2023-02-07 08:27:33,980 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=196722.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:27:46,991 INFO [train.py:901] (3/4) Epoch 25, batch 2750, loss[loss=0.1806, simple_loss=0.2627, pruned_loss=0.04925, over 5159.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2854, pruned_loss=0.05991, over 1612316.48 frames. ], batch size: 11, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:27:55,343 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3898, 2.5895, 3.0553, 1.7441, 3.1697, 1.8855, 1.6407, 2.2959], + device='cuda:3'), covar=tensor([0.0801, 0.0499, 0.0322, 0.0858, 0.0599, 0.0962, 0.0945, 0.0604], + device='cuda:3'), in_proj_covar=tensor([0.0465, 0.0404, 0.0360, 0.0456, 0.0388, 0.0543, 0.0402, 0.0432], + device='cuda:3'), out_proj_covar=tensor([1.2356e-04, 1.0535e-04, 9.4351e-05, 1.1950e-04, 1.0162e-04, 1.5227e-04, + 1.0762e-04, 1.1362e-04], device='cuda:3') +2023-02-07 08:28:11,176 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.82 vs. limit=5.0 +2023-02-07 08:28:22,167 INFO [train.py:901] (3/4) Epoch 25, batch 2800, loss[loss=0.1722, simple_loss=0.2545, pruned_loss=0.04495, over 8056.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2857, pruned_loss=0.05993, over 1616015.61 frames. ], batch size: 20, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:28:27,860 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196797.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:28:28,544 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=196798.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:28:31,148 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.590e+02 2.445e+02 2.946e+02 3.604e+02 6.151e+02, threshold=5.892e+02, percent-clipped=2.0 +2023-02-07 08:28:52,423 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 +2023-02-07 08:28:54,960 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=196837.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:28:56,676 INFO [train.py:901] (3/4) Epoch 25, batch 2850, loss[loss=0.2438, simple_loss=0.3184, pruned_loss=0.08461, over 8185.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2863, pruned_loss=0.05996, over 1618850.52 frames. ], batch size: 23, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:29:19,399 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196872.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:29:32,387 INFO [train.py:901] (3/4) Epoch 25, batch 2900, loss[loss=0.207, simple_loss=0.2814, pruned_loss=0.06629, over 7656.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2856, pruned_loss=0.05937, over 1618159.77 frames. ], batch size: 19, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:29:39,455 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196899.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:29:41,309 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.503e+02 2.504e+02 3.053e+02 3.742e+02 6.617e+02, threshold=6.106e+02, percent-clipped=2.0 +2023-02-07 08:30:08,124 INFO [train.py:901] (3/4) Epoch 25, batch 2950, loss[loss=0.2115, simple_loss=0.3071, pruned_loss=0.05789, over 8536.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2846, pruned_loss=0.05869, over 1617448.15 frames. ], batch size: 28, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:30:08,202 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=196940.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:30:08,828 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-07 08:30:19,844 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 +2023-02-07 08:30:42,487 INFO [train.py:901] (3/4) Epoch 25, batch 3000, loss[loss=0.1907, simple_loss=0.2644, pruned_loss=0.05848, over 7926.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2853, pruned_loss=0.05928, over 1617236.59 frames. ], batch size: 20, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:30:42,487 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 08:30:55,640 INFO [train.py:935] (3/4) Epoch 25, validation: loss=0.1722, simple_loss=0.2721, pruned_loss=0.03618, over 944034.00 frames. +2023-02-07 08:30:55,642 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6747MB +2023-02-07 08:31:03,953 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.269e+02 2.477e+02 2.955e+02 3.925e+02 7.788e+02, threshold=5.910e+02, percent-clipped=1.0 +2023-02-07 08:31:30,700 INFO [train.py:901] (3/4) Epoch 25, batch 3050, loss[loss=0.1827, simple_loss=0.2754, pruned_loss=0.045, over 8293.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2857, pruned_loss=0.05957, over 1619554.31 frames. ], batch size: 23, lr: 3.03e-03, grad_scale: 8.0 +2023-02-07 08:31:40,541 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197054.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:31:41,167 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197055.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:31:57,870 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197079.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:32:05,206 INFO [train.py:901] (3/4) Epoch 25, batch 3100, loss[loss=0.1822, simple_loss=0.2672, pruned_loss=0.04862, over 7649.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2854, pruned_loss=0.05935, over 1620954.34 frames. ], batch size: 19, lr: 3.02e-03, grad_scale: 8.0 +2023-02-07 08:32:07,488 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197093.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:32:13,236 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.503e+02 2.425e+02 3.089e+02 3.818e+02 7.102e+02, threshold=6.178e+02, percent-clipped=3.0 +2023-02-07 08:32:24,976 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197118.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:32:40,194 INFO [train.py:901] (3/4) Epoch 25, batch 3150, loss[loss=0.2017, simple_loss=0.297, pruned_loss=0.05314, over 8324.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2851, pruned_loss=0.05932, over 1617538.60 frames. ], batch size: 25, lr: 3.02e-03, grad_scale: 8.0 +2023-02-07 08:32:40,983 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=197141.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:32:42,603 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-07 08:33:15,303 INFO [train.py:901] (3/4) Epoch 25, batch 3200, loss[loss=0.2103, simple_loss=0.2976, pruned_loss=0.06154, over 8187.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2859, pruned_loss=0.05957, over 1618984.52 frames. ], batch size: 23, lr: 3.02e-03, grad_scale: 8.0 +2023-02-07 08:33:23,539 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.521e+02 2.400e+02 2.739e+02 3.315e+02 1.024e+03, threshold=5.479e+02, percent-clipped=5.0 +2023-02-07 08:33:33,160 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=197216.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:33:50,045 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-07 08:33:50,304 INFO [train.py:901] (3/4) Epoch 25, batch 3250, loss[loss=0.2337, simple_loss=0.3175, pruned_loss=0.07494, over 8757.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2856, pruned_loss=0.05938, over 1615570.19 frames. ], batch size: 30, lr: 3.02e-03, grad_scale: 8.0 +2023-02-07 08:33:52,476 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=197243.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:34:02,150 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197256.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:34:24,835 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4037, 1.1125, 2.5034, 0.9687, 2.2354, 2.1026, 2.2916, 2.2289], + device='cuda:3'), covar=tensor([0.0838, 0.3195, 0.0964, 0.3742, 0.1116, 0.0960, 0.0699, 0.0798], + device='cuda:3'), in_proj_covar=tensor([0.0649, 0.0653, 0.0711, 0.0643, 0.0724, 0.0615, 0.0620, 0.0693], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 08:34:25,380 INFO [train.py:901] (3/4) Epoch 25, batch 3300, loss[loss=0.2832, simple_loss=0.3396, pruned_loss=0.1134, over 7009.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2851, pruned_loss=0.05932, over 1612944.66 frames. ], batch size: 71, lr: 3.02e-03, grad_scale: 8.0 +2023-02-07 08:34:34,244 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.558e+02 2.560e+02 3.230e+02 4.212e+02 8.703e+02, threshold=6.460e+02, percent-clipped=10.0 +2023-02-07 08:34:40,504 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197311.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:34:54,231 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197331.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:34:57,730 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197336.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:35:00,243 INFO [train.py:901] (3/4) Epoch 25, batch 3350, loss[loss=0.1607, simple_loss=0.2417, pruned_loss=0.03985, over 7925.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2844, pruned_loss=0.05892, over 1612355.05 frames. ], batch size: 20, lr: 3.02e-03, grad_scale: 8.0 +2023-02-07 08:35:13,336 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197358.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:35:17,449 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2263, 1.1026, 1.2707, 0.9949, 1.0389, 1.3109, 0.0507, 0.8881], + device='cuda:3'), covar=tensor([0.1545, 0.1269, 0.0519, 0.0786, 0.2443, 0.0573, 0.2133, 0.1253], + device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0200, 0.0130, 0.0220, 0.0271, 0.0139, 0.0170, 0.0197], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 08:35:36,172 INFO [train.py:901] (3/4) Epoch 25, batch 3400, loss[loss=0.1902, simple_loss=0.2571, pruned_loss=0.06162, over 7728.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2843, pruned_loss=0.05905, over 1615528.69 frames. ], batch size: 18, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:35:39,765 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3853, 1.4058, 1.8088, 1.1061, 1.0446, 1.8178, 0.1574, 1.0577], + device='cuda:3'), covar=tensor([0.1615, 0.1234, 0.0397, 0.1164, 0.2600, 0.0378, 0.1954, 0.1344], + device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0199, 0.0130, 0.0220, 0.0271, 0.0138, 0.0170, 0.0197], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 08:35:44,272 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.558e+02 2.489e+02 3.044e+02 3.734e+02 7.163e+02, threshold=6.087e+02, percent-clipped=2.0 +2023-02-07 08:36:06,305 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197433.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 08:36:11,059 INFO [train.py:901] (3/4) Epoch 25, batch 3450, loss[loss=0.2255, simple_loss=0.3127, pruned_loss=0.06909, over 8708.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2856, pruned_loss=0.05991, over 1614284.57 frames. ], batch size: 34, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:36:16,733 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1794, 3.8087, 2.3586, 3.0273, 2.7172, 2.0071, 2.7649, 3.1128], + device='cuda:3'), covar=tensor([0.1480, 0.0265, 0.1040, 0.0673, 0.0753, 0.1417, 0.1036, 0.1036], + device='cuda:3'), in_proj_covar=tensor([0.0357, 0.0236, 0.0340, 0.0311, 0.0300, 0.0345, 0.0350, 0.0323], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 08:36:46,262 INFO [train.py:901] (3/4) Epoch 25, batch 3500, loss[loss=0.1795, simple_loss=0.2565, pruned_loss=0.05124, over 7520.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2852, pruned_loss=0.05949, over 1613849.49 frames. ], batch size: 18, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:36:54,904 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.387e+02 2.953e+02 3.537e+02 5.869e+02, threshold=5.907e+02, percent-clipped=0.0 +2023-02-07 08:37:02,043 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197512.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:37:07,264 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-07 08:37:17,735 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197534.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:37:19,778 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197537.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:37:21,593 INFO [train.py:901] (3/4) Epoch 25, batch 3550, loss[loss=0.196, simple_loss=0.2744, pruned_loss=0.05885, over 7512.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2847, pruned_loss=0.05943, over 1617707.40 frames. ], batch size: 18, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:37:22,462 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8621, 1.7177, 2.1249, 1.5178, 1.5836, 2.1729, 1.0101, 1.7083], + device='cuda:3'), covar=tensor([0.1160, 0.0915, 0.0316, 0.0808, 0.1678, 0.0294, 0.1540, 0.1089], + device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0201, 0.0131, 0.0221, 0.0273, 0.0139, 0.0172, 0.0198], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 08:37:24,416 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0914, 1.9034, 2.3828, 2.0456, 2.3190, 2.1936, 1.9890, 1.1526], + device='cuda:3'), covar=tensor([0.5669, 0.4702, 0.1981, 0.3605, 0.2389, 0.2935, 0.1904, 0.5143], + device='cuda:3'), in_proj_covar=tensor([0.0951, 0.1006, 0.0822, 0.0974, 0.1014, 0.0914, 0.0763, 0.0840], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 08:37:37,367 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197563.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:37:52,449 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1201, 2.4169, 1.9089, 2.9099, 1.4489, 1.7872, 2.2127, 2.3708], + device='cuda:3'), covar=tensor([0.0719, 0.0719, 0.0878, 0.0367, 0.1084, 0.1243, 0.0747, 0.0740], + device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0194, 0.0245, 0.0212, 0.0204, 0.0247, 0.0248, 0.0205], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 08:37:54,507 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197587.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:37:56,270 INFO [train.py:901] (3/4) Epoch 25, batch 3600, loss[loss=0.1961, simple_loss=0.2685, pruned_loss=0.06187, over 7444.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2837, pruned_loss=0.05939, over 1607838.86 frames. ], batch size: 17, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:38:05,202 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.545e+02 2.311e+02 2.881e+02 3.803e+02 6.346e+02, threshold=5.762e+02, percent-clipped=1.0 +2023-02-07 08:38:12,109 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197612.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:38:13,509 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197614.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:38:31,222 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197639.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:38:31,671 INFO [train.py:901] (3/4) Epoch 25, batch 3650, loss[loss=0.2218, simple_loss=0.301, pruned_loss=0.07128, over 8080.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.285, pruned_loss=0.06, over 1607024.51 frames. ], batch size: 21, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:38:40,962 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8605, 2.4264, 3.7933, 1.8691, 1.9298, 3.7648, 0.7353, 2.2244], + device='cuda:3'), covar=tensor([0.1372, 0.1059, 0.0163, 0.1573, 0.2292, 0.0177, 0.1956, 0.1139], + device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0201, 0.0131, 0.0221, 0.0273, 0.0139, 0.0171, 0.0197], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 08:38:54,176 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6006, 1.9974, 2.8779, 1.5056, 2.1246, 2.0172, 1.6838, 2.1962], + device='cuda:3'), covar=tensor([0.1842, 0.2450, 0.0849, 0.4354, 0.1875, 0.3034, 0.2292, 0.2172], + device='cuda:3'), in_proj_covar=tensor([0.0532, 0.0621, 0.0557, 0.0659, 0.0654, 0.0603, 0.0549, 0.0637], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 08:39:06,710 INFO [train.py:901] (3/4) Epoch 25, batch 3700, loss[loss=0.2053, simple_loss=0.2973, pruned_loss=0.05668, over 8330.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2852, pruned_loss=0.05993, over 1605851.24 frames. ], batch size: 25, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:39:09,548 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-07 08:39:15,747 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.564e+02 2.484e+02 2.942e+02 3.783e+02 7.174e+02, threshold=5.884e+02, percent-clipped=5.0 +2023-02-07 08:39:35,848 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5322, 1.2556, 4.7567, 1.8306, 4.2597, 3.9412, 4.3319, 4.1983], + device='cuda:3'), covar=tensor([0.0570, 0.5257, 0.0453, 0.4017, 0.1074, 0.0945, 0.0555, 0.0646], + device='cuda:3'), in_proj_covar=tensor([0.0655, 0.0660, 0.0721, 0.0650, 0.0730, 0.0624, 0.0626, 0.0701], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 08:39:40,719 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-07 08:39:43,096 INFO [train.py:901] (3/4) Epoch 25, batch 3750, loss[loss=0.1904, simple_loss=0.2656, pruned_loss=0.05757, over 7702.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2857, pruned_loss=0.05989, over 1607779.40 frames. ], batch size: 18, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:40:09,411 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=197777.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 08:40:18,200 INFO [train.py:901] (3/4) Epoch 25, batch 3800, loss[loss=0.1898, simple_loss=0.2677, pruned_loss=0.05593, over 7402.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2858, pruned_loss=0.06022, over 1606802.40 frames. ], batch size: 17, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:40:26,493 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.709e+02 2.549e+02 3.044e+02 3.681e+02 9.424e+02, threshold=6.087e+02, percent-clipped=5.0 +2023-02-07 08:40:34,308 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-07 08:40:43,133 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 +2023-02-07 08:40:53,471 INFO [train.py:901] (3/4) Epoch 25, batch 3850, loss[loss=0.2281, simple_loss=0.3115, pruned_loss=0.07234, over 8518.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2856, pruned_loss=0.05979, over 1611595.91 frames. ], batch size: 26, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:40:57,614 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4460, 1.4876, 1.4389, 1.8378, 0.8381, 1.2963, 1.3845, 1.4749], + device='cuda:3'), covar=tensor([0.0845, 0.0766, 0.1029, 0.0494, 0.1090, 0.1404, 0.0712, 0.0738], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0195, 0.0245, 0.0211, 0.0205, 0.0247, 0.0247, 0.0206], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 08:41:12,905 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-07 08:41:19,585 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=197878.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:41:28,515 INFO [train.py:901] (3/4) Epoch 25, batch 3900, loss[loss=0.2102, simple_loss=0.2973, pruned_loss=0.06153, over 8325.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.285, pruned_loss=0.05926, over 1612114.24 frames. ], batch size: 25, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:41:29,975 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197892.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 08:41:36,383 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.837e+02 2.445e+02 2.982e+02 3.609e+02 8.629e+02, threshold=5.963e+02, percent-clipped=3.0 +2023-02-07 08:41:39,827 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=197907.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:41:41,926 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197910.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:42:02,716 INFO [train.py:901] (3/4) Epoch 25, batch 3950, loss[loss=0.174, simple_loss=0.246, pruned_loss=0.05095, over 7521.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.285, pruned_loss=0.05916, over 1611273.54 frames. ], batch size: 18, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:42:27,075 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5809, 1.3424, 1.6094, 1.2645, 0.9811, 1.3886, 1.5305, 1.3197], + device='cuda:3'), covar=tensor([0.0591, 0.1373, 0.1755, 0.1520, 0.0615, 0.1613, 0.0740, 0.0699], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0153, 0.0189, 0.0160, 0.0100, 0.0163, 0.0112, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 08:42:37,829 INFO [train.py:901] (3/4) Epoch 25, batch 4000, loss[loss=0.2012, simple_loss=0.2701, pruned_loss=0.06613, over 7697.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2866, pruned_loss=0.06045, over 1611811.21 frames. ], batch size: 18, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:42:40,144 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197993.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:42:47,778 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.662e+02 2.312e+02 2.768e+02 3.562e+02 7.475e+02, threshold=5.536e+02, percent-clipped=2.0 +2023-02-07 08:43:01,564 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198022.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:43:14,007 INFO [train.py:901] (3/4) Epoch 25, batch 4050, loss[loss=0.1361, simple_loss=0.2186, pruned_loss=0.02682, over 7542.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2849, pruned_loss=0.05953, over 1614290.77 frames. ], batch size: 18, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:43:24,465 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0303, 2.3652, 3.6526, 1.9117, 2.9062, 2.4122, 1.9570, 2.8174], + device='cuda:3'), covar=tensor([0.1673, 0.2507, 0.0974, 0.4108, 0.1796, 0.2956, 0.2267, 0.2488], + device='cuda:3'), in_proj_covar=tensor([0.0534, 0.0622, 0.0557, 0.0659, 0.0656, 0.0604, 0.0552, 0.0640], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 08:43:32,401 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6845, 1.6236, 2.2770, 1.4612, 1.3202, 2.2557, 0.4305, 1.3795], + device='cuda:3'), covar=tensor([0.1519, 0.1309, 0.0309, 0.1103, 0.2400, 0.0418, 0.1973, 0.1430], + device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0201, 0.0131, 0.0220, 0.0272, 0.0139, 0.0170, 0.0197], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 08:43:48,797 INFO [train.py:901] (3/4) Epoch 25, batch 4100, loss[loss=0.2341, simple_loss=0.3136, pruned_loss=0.07733, over 8537.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2849, pruned_loss=0.05928, over 1618011.56 frames. ], batch size: 28, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:43:55,131 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198099.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:43:57,000 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.511e+02 2.413e+02 2.876e+02 3.434e+02 5.292e+02, threshold=5.752e+02, percent-clipped=1.0 +2023-02-07 08:44:04,779 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5797, 2.5838, 1.8720, 2.2808, 2.2070, 1.5644, 2.0243, 2.1851], + device='cuda:3'), covar=tensor([0.1511, 0.0469, 0.1312, 0.0671, 0.0775, 0.1783, 0.1070, 0.1012], + device='cuda:3'), in_proj_covar=tensor([0.0360, 0.0238, 0.0343, 0.0315, 0.0304, 0.0346, 0.0352, 0.0324], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 08:44:24,272 INFO [train.py:901] (3/4) Epoch 25, batch 4150, loss[loss=0.2328, simple_loss=0.3032, pruned_loss=0.08118, over 7809.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.284, pruned_loss=0.05875, over 1610910.76 frames. ], batch size: 20, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:44:27,219 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4768, 2.0101, 3.9344, 1.5185, 2.7634, 2.1025, 1.5787, 2.7986], + device='cuda:3'), covar=tensor([0.2350, 0.3188, 0.0917, 0.5171, 0.2171, 0.3715, 0.2858, 0.2603], + device='cuda:3'), in_proj_covar=tensor([0.0535, 0.0624, 0.0558, 0.0661, 0.0657, 0.0606, 0.0552, 0.0641], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 08:44:29,933 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198148.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 08:44:33,248 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5308, 1.5665, 4.7651, 1.7606, 4.2417, 3.9351, 4.2965, 4.1825], + device='cuda:3'), covar=tensor([0.0610, 0.4706, 0.0499, 0.4368, 0.1085, 0.0934, 0.0559, 0.0645], + device='cuda:3'), in_proj_covar=tensor([0.0652, 0.0657, 0.0718, 0.0645, 0.0726, 0.0621, 0.0625, 0.0695], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 08:44:44,867 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4743, 1.3634, 2.3572, 1.2822, 2.3750, 2.5335, 2.7102, 2.1241], + device='cuda:3'), covar=tensor([0.1112, 0.1372, 0.0384, 0.2070, 0.0614, 0.0389, 0.0722, 0.0723], + device='cuda:3'), in_proj_covar=tensor([0.0303, 0.0325, 0.0290, 0.0318, 0.0318, 0.0275, 0.0434, 0.0307], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 08:44:47,427 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198173.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 08:44:58,965 INFO [train.py:901] (3/4) Epoch 25, batch 4200, loss[loss=0.1659, simple_loss=0.2462, pruned_loss=0.04283, over 7431.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2845, pruned_loss=0.05869, over 1613415.37 frames. ], batch size: 17, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:44:59,122 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5551, 1.4399, 4.7586, 1.7232, 4.2448, 3.9220, 4.3231, 4.1756], + device='cuda:3'), covar=tensor([0.0542, 0.4869, 0.0485, 0.4397, 0.1004, 0.0958, 0.0495, 0.0611], + device='cuda:3'), in_proj_covar=tensor([0.0649, 0.0654, 0.0714, 0.0642, 0.0723, 0.0618, 0.0622, 0.0692], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 08:45:08,029 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.632e+02 2.351e+02 3.091e+02 3.845e+02 7.201e+02, threshold=6.182e+02, percent-clipped=4.0 +2023-02-07 08:45:09,394 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-07 08:45:25,548 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7463, 2.0542, 2.9456, 1.5705, 2.3219, 1.9935, 1.8332, 2.0761], + device='cuda:3'), covar=tensor([0.1802, 0.2531, 0.0852, 0.4485, 0.1784, 0.3367, 0.2210, 0.2345], + device='cuda:3'), in_proj_covar=tensor([0.0534, 0.0624, 0.0558, 0.0661, 0.0656, 0.0607, 0.0552, 0.0640], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 08:45:33,125 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-07 08:45:35,175 INFO [train.py:901] (3/4) Epoch 25, batch 4250, loss[loss=0.1653, simple_loss=0.2553, pruned_loss=0.03765, over 8286.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2836, pruned_loss=0.05779, over 1616143.73 frames. ], batch size: 23, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:45:41,587 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198249.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:45:43,798 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-07 08:45:44,708 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=198254.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:45:59,331 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198274.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:46:02,100 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198278.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:46:09,935 INFO [train.py:901] (3/4) Epoch 25, batch 4300, loss[loss=0.1759, simple_loss=0.2704, pruned_loss=0.04067, over 8468.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2835, pruned_loss=0.05783, over 1612127.78 frames. ], batch size: 25, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:46:18,866 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.684e+02 2.314e+02 2.735e+02 3.533e+02 6.805e+02, threshold=5.471e+02, percent-clipped=1.0 +2023-02-07 08:46:19,821 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198303.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:46:45,627 INFO [train.py:901] (3/4) Epoch 25, batch 4350, loss[loss=0.1845, simple_loss=0.2729, pruned_loss=0.04808, over 8498.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2816, pruned_loss=0.05656, over 1611370.51 frames. ], batch size: 26, lr: 3.02e-03, grad_scale: 16.0 +2023-02-07 08:46:56,198 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2406, 2.6063, 2.8091, 1.6638, 3.0888, 1.9090, 1.6065, 2.1963], + device='cuda:3'), covar=tensor([0.1078, 0.0531, 0.0483, 0.1034, 0.0650, 0.1112, 0.1068, 0.0632], + device='cuda:3'), in_proj_covar=tensor([0.0464, 0.0404, 0.0361, 0.0458, 0.0390, 0.0545, 0.0400, 0.0432], + device='cuda:3'), out_proj_covar=tensor([1.2334e-04, 1.0525e-04, 9.4623e-05, 1.2014e-04, 1.0228e-04, 1.5263e-04, + 1.0718e-04, 1.1366e-04], device='cuda:3') +2023-02-07 08:47:04,264 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-07 08:47:06,465 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198369.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:47:21,574 INFO [train.py:901] (3/4) Epoch 25, batch 4400, loss[loss=0.1789, simple_loss=0.2709, pruned_loss=0.04343, over 8333.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2831, pruned_loss=0.05751, over 1613332.58 frames. ], batch size: 25, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:47:27,079 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6509, 1.8575, 2.6673, 1.5103, 1.9649, 1.9965, 1.6439, 1.9616], + device='cuda:3'), covar=tensor([0.2026, 0.2799, 0.0985, 0.4653, 0.2111, 0.3433, 0.2543, 0.2379], + device='cuda:3'), in_proj_covar=tensor([0.0534, 0.0622, 0.0557, 0.0659, 0.0655, 0.0605, 0.0552, 0.0639], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 08:47:29,518 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.496e+02 2.496e+02 2.935e+02 3.768e+02 7.665e+02, threshold=5.870e+02, percent-clipped=6.0 +2023-02-07 08:47:45,273 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-07 08:47:49,446 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4368, 2.6092, 2.9324, 1.5996, 3.1990, 2.0327, 1.6012, 2.2975], + device='cuda:3'), covar=tensor([0.0889, 0.0545, 0.0321, 0.0924, 0.0582, 0.0882, 0.0987, 0.0568], + device='cuda:3'), in_proj_covar=tensor([0.0464, 0.0404, 0.0362, 0.0457, 0.0391, 0.0543, 0.0399, 0.0431], + device='cuda:3'), out_proj_covar=tensor([1.2331e-04, 1.0517e-04, 9.4670e-05, 1.1996e-04, 1.0243e-04, 1.5217e-04, + 1.0701e-04, 1.1337e-04], device='cuda:3') +2023-02-07 08:47:56,744 INFO [train.py:901] (3/4) Epoch 25, batch 4450, loss[loss=0.2252, simple_loss=0.3002, pruned_loss=0.0751, over 8510.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2831, pruned_loss=0.0579, over 1611811.12 frames. ], batch size: 26, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:47:58,933 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=198443.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:48:29,551 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8912, 1.7698, 2.7626, 2.0891, 2.4636, 1.8546, 1.6797, 1.3473], + device='cuda:3'), covar=tensor([0.7301, 0.6213, 0.2198, 0.4607, 0.3369, 0.4609, 0.3002, 0.5935], + device='cuda:3'), in_proj_covar=tensor([0.0946, 0.0999, 0.0816, 0.0969, 0.1011, 0.0911, 0.0756, 0.0836], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 08:48:31,963 INFO [train.py:901] (3/4) Epoch 25, batch 4500, loss[loss=0.2318, simple_loss=0.3035, pruned_loss=0.08011, over 7918.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2829, pruned_loss=0.05797, over 1610222.10 frames. ], batch size: 20, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:48:40,436 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.409e+02 2.274e+02 2.771e+02 3.541e+02 5.802e+02, threshold=5.543e+02, percent-clipped=0.0 +2023-02-07 08:48:40,467 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-07 08:48:46,360 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8046, 1.7560, 2.5071, 1.6519, 1.3861, 2.4974, 0.4806, 1.5122], + device='cuda:3'), covar=tensor([0.1980, 0.1372, 0.0356, 0.1389, 0.2803, 0.0423, 0.2106, 0.1544], + device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0202, 0.0132, 0.0222, 0.0275, 0.0141, 0.0172, 0.0199], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 08:48:51,919 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198517.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:48:57,525 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.9416, 6.0824, 5.2148, 2.4821, 5.3840, 5.6867, 5.4468, 5.5536], + device='cuda:3'), covar=tensor([0.0442, 0.0397, 0.0869, 0.4184, 0.0681, 0.0606, 0.1093, 0.0514], + device='cuda:3'), in_proj_covar=tensor([0.0532, 0.0450, 0.0438, 0.0547, 0.0435, 0.0454, 0.0426, 0.0398], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 08:49:08,826 INFO [train.py:901] (3/4) Epoch 25, batch 4550, loss[loss=0.2123, simple_loss=0.3017, pruned_loss=0.06146, over 8467.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2835, pruned_loss=0.05839, over 1613966.58 frames. ], batch size: 25, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:49:22,068 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198558.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:49:44,787 INFO [train.py:901] (3/4) Epoch 25, batch 4600, loss[loss=0.1786, simple_loss=0.2726, pruned_loss=0.04229, over 8475.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2849, pruned_loss=0.05929, over 1611931.57 frames. ], batch size: 25, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:49:52,974 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.563e+02 2.449e+02 2.940e+02 3.432e+02 8.422e+02, threshold=5.881e+02, percent-clipped=6.0 +2023-02-07 08:50:09,324 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198625.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:50:14,649 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198633.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:50:19,158 INFO [train.py:901] (3/4) Epoch 25, batch 4650, loss[loss=0.1979, simple_loss=0.2888, pruned_loss=0.05356, over 8353.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2851, pruned_loss=0.05984, over 1610244.92 frames. ], batch size: 24, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:50:26,833 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198650.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:50:40,270 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5866, 1.7190, 2.0912, 1.6429, 1.0400, 1.7585, 2.3028, 1.8707], + device='cuda:3'), covar=tensor([0.0471, 0.1223, 0.1527, 0.1378, 0.0589, 0.1426, 0.0603, 0.0634], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0153, 0.0189, 0.0160, 0.0100, 0.0163, 0.0112, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 08:50:54,512 INFO [train.py:901] (3/4) Epoch 25, batch 4700, loss[loss=0.1963, simple_loss=0.2866, pruned_loss=0.05302, over 8473.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2847, pruned_loss=0.0596, over 1606458.05 frames. ], batch size: 29, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:51:03,371 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.550e+02 2.164e+02 2.735e+02 3.323e+02 7.623e+02, threshold=5.470e+02, percent-clipped=2.0 +2023-02-07 08:51:29,739 INFO [train.py:901] (3/4) Epoch 25, batch 4750, loss[loss=0.2367, simple_loss=0.3268, pruned_loss=0.07331, over 8293.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2841, pruned_loss=0.0593, over 1608294.32 frames. ], batch size: 23, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:51:31,918 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3188, 2.6424, 2.9796, 1.7075, 3.2201, 1.9776, 1.5958, 2.3523], + device='cuda:3'), covar=tensor([0.0862, 0.0445, 0.0366, 0.0881, 0.0411, 0.0950, 0.0960, 0.0525], + device='cuda:3'), in_proj_covar=tensor([0.0466, 0.0404, 0.0362, 0.0457, 0.0391, 0.0544, 0.0400, 0.0431], + device='cuda:3'), out_proj_covar=tensor([1.2369e-04, 1.0541e-04, 9.4773e-05, 1.1975e-04, 1.0267e-04, 1.5243e-04, + 1.0723e-04, 1.1324e-04], device='cuda:3') +2023-02-07 08:51:42,014 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-07 08:51:45,378 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-07 08:52:05,183 INFO [train.py:901] (3/4) Epoch 25, batch 4800, loss[loss=0.2412, simple_loss=0.3246, pruned_loss=0.07893, over 8516.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2844, pruned_loss=0.05979, over 1608267.70 frames. ], batch size: 39, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:52:13,383 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.326e+02 2.392e+02 2.917e+02 3.409e+02 6.169e+02, threshold=5.835e+02, percent-clipped=3.0 +2023-02-07 08:52:22,058 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198814.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:52:36,110 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-07 08:52:39,636 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198839.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:52:40,102 INFO [train.py:901] (3/4) Epoch 25, batch 4850, loss[loss=0.2281, simple_loss=0.2869, pruned_loss=0.08465, over 7804.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2827, pruned_loss=0.05903, over 1602677.72 frames. ], batch size: 19, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:52:55,413 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=198861.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:52:56,958 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5743, 1.4835, 1.8236, 1.3124, 1.2636, 1.8296, 0.2669, 1.3046], + device='cuda:3'), covar=tensor([0.1689, 0.1124, 0.0419, 0.0764, 0.2548, 0.0450, 0.1953, 0.1197], + device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0201, 0.0131, 0.0220, 0.0273, 0.0141, 0.0171, 0.0198], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 08:53:01,733 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198870.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:53:15,870 INFO [train.py:901] (3/4) Epoch 25, batch 4900, loss[loss=0.3222, simple_loss=0.3709, pruned_loss=0.1368, over 6607.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2828, pruned_loss=0.05913, over 1601992.87 frames. ], batch size: 71, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:53:24,157 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198901.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:53:24,667 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.754e+02 2.376e+02 2.954e+02 3.660e+02 6.336e+02, threshold=5.908e+02, percent-clipped=3.0 +2023-02-07 08:53:30,960 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1411, 3.5628, 2.5001, 2.9718, 2.6661, 2.0650, 2.6660, 3.1398], + device='cuda:3'), covar=tensor([0.1790, 0.0458, 0.1082, 0.0718, 0.0817, 0.1518, 0.1213, 0.1112], + device='cuda:3'), in_proj_covar=tensor([0.0356, 0.0235, 0.0339, 0.0311, 0.0300, 0.0343, 0.0346, 0.0318], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 08:53:50,032 INFO [train.py:901] (3/4) Epoch 25, batch 4950, loss[loss=0.2183, simple_loss=0.2939, pruned_loss=0.07136, over 8018.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2824, pruned_loss=0.05859, over 1604890.11 frames. ], batch size: 22, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:53:54,448 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198945.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:53:54,544 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5383, 1.8620, 2.9300, 1.3927, 2.0556, 1.9546, 1.5759, 2.1766], + device='cuda:3'), covar=tensor([0.1996, 0.2734, 0.0888, 0.4782, 0.2146, 0.3174, 0.2486, 0.2364], + device='cuda:3'), in_proj_covar=tensor([0.0532, 0.0622, 0.0557, 0.0661, 0.0657, 0.0603, 0.0553, 0.0641], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 08:54:15,962 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198976.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:54:16,531 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=198977.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:54:25,252 INFO [train.py:901] (3/4) Epoch 25, batch 5000, loss[loss=0.1776, simple_loss=0.264, pruned_loss=0.04559, over 7981.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2823, pruned_loss=0.05803, over 1603765.16 frames. ], batch size: 21, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:54:33,917 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.471e+02 2.360e+02 2.883e+02 3.509e+02 6.136e+02, threshold=5.766e+02, percent-clipped=1.0 +2023-02-07 08:54:59,851 INFO [train.py:901] (3/4) Epoch 25, batch 5050, loss[loss=0.2051, simple_loss=0.299, pruned_loss=0.05563, over 8297.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2837, pruned_loss=0.05919, over 1606262.09 frames. ], batch size: 23, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:55:14,355 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-07 08:55:28,312 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0 +2023-02-07 08:55:35,772 INFO [train.py:901] (3/4) Epoch 25, batch 5100, loss[loss=0.203, simple_loss=0.2895, pruned_loss=0.05827, over 8894.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2841, pruned_loss=0.05929, over 1605185.84 frames. ], batch size: 40, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:55:37,432 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199092.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:55:44,130 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.454e+02 2.470e+02 3.005e+02 3.768e+02 7.063e+02, threshold=6.010e+02, percent-clipped=5.0 +2023-02-07 08:56:11,855 INFO [train.py:901] (3/4) Epoch 25, batch 5150, loss[loss=0.1647, simple_loss=0.2543, pruned_loss=0.0376, over 8076.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2843, pruned_loss=0.05939, over 1603368.89 frames. ], batch size: 21, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:56:47,046 INFO [train.py:901] (3/4) Epoch 25, batch 5200, loss[loss=0.1973, simple_loss=0.2807, pruned_loss=0.057, over 8099.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2836, pruned_loss=0.05874, over 1606095.67 frames. ], batch size: 23, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:56:49,910 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199194.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:56:50,569 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1230, 1.4630, 1.6608, 1.3598, 1.0057, 1.4716, 1.9706, 1.9906], + device='cuda:3'), covar=tensor([0.0559, 0.1698, 0.2287, 0.1864, 0.0673, 0.1951, 0.0733, 0.0641], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0153, 0.0189, 0.0161, 0.0100, 0.0164, 0.0113, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 08:56:55,021 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.800e+02 2.381e+02 2.894e+02 3.514e+02 1.206e+03, threshold=5.788e+02, percent-clipped=6.0 +2023-02-07 08:57:04,084 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199214.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:57:12,777 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-07 08:57:17,169 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199232.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:57:22,320 INFO [train.py:901] (3/4) Epoch 25, batch 5250, loss[loss=0.191, simple_loss=0.2833, pruned_loss=0.04936, over 8748.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2841, pruned_loss=0.05928, over 1605226.70 frames. ], batch size: 30, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:57:25,800 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199245.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:57:34,835 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199257.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:57:34,966 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.56 vs. limit=2.0 +2023-02-07 08:57:56,837 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199289.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:57:57,413 INFO [train.py:901] (3/4) Epoch 25, batch 5300, loss[loss=0.2179, simple_loss=0.3019, pruned_loss=0.06699, over 8193.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2842, pruned_loss=0.05898, over 1609200.52 frames. ], batch size: 23, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:58:05,707 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.533e+02 2.313e+02 2.718e+02 3.488e+02 6.386e+02, threshold=5.437e+02, percent-clipped=3.0 +2023-02-07 08:58:25,239 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199329.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:58:32,735 INFO [train.py:901] (3/4) Epoch 25, batch 5350, loss[loss=0.1886, simple_loss=0.2721, pruned_loss=0.05255, over 7815.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2843, pruned_loss=0.05887, over 1610706.96 frames. ], batch size: 20, lr: 3.01e-03, grad_scale: 16.0 +2023-02-07 08:58:38,545 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199348.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:58:47,620 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199360.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:58:57,362 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199373.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:59:02,728 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0420, 1.6887, 3.2331, 1.5889, 2.5960, 3.5232, 3.6285, 3.0459], + device='cuda:3'), covar=tensor([0.1112, 0.1703, 0.0392, 0.1941, 0.1122, 0.0236, 0.0649, 0.0488], + device='cuda:3'), in_proj_covar=tensor([0.0302, 0.0323, 0.0288, 0.0316, 0.0318, 0.0274, 0.0432, 0.0305], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 08:59:08,731 INFO [train.py:901] (3/4) Epoch 25, batch 5400, loss[loss=0.2148, simple_loss=0.2926, pruned_loss=0.06853, over 8508.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2838, pruned_loss=0.05863, over 1613009.40 frames. ], batch size: 26, lr: 3.01e-03, grad_scale: 32.0 +2023-02-07 08:59:18,150 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.323e+02 2.292e+02 2.858e+02 3.757e+02 5.815e+02, threshold=5.716e+02, percent-clipped=3.0 +2023-02-07 08:59:18,353 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199404.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:59:28,357 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199418.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 08:59:42,053 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8226, 2.0247, 2.1264, 1.4310, 2.2157, 1.5474, 0.7791, 1.9728], + device='cuda:3'), covar=tensor([0.0739, 0.0383, 0.0313, 0.0735, 0.0520, 0.1043, 0.1052, 0.0384], + device='cuda:3'), in_proj_covar=tensor([0.0462, 0.0401, 0.0359, 0.0456, 0.0389, 0.0541, 0.0399, 0.0430], + device='cuda:3'), out_proj_covar=tensor([1.2274e-04, 1.0438e-04, 9.3970e-05, 1.1961e-04, 1.0184e-04, 1.5163e-04, + 1.0682e-04, 1.1304e-04], device='cuda:3') +2023-02-07 08:59:43,202 INFO [train.py:901] (3/4) Epoch 25, batch 5450, loss[loss=0.2212, simple_loss=0.2998, pruned_loss=0.07135, over 8350.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2844, pruned_loss=0.05868, over 1610516.87 frames. ], batch size: 26, lr: 3.01e-03, grad_scale: 8.0 +2023-02-07 09:00:08,085 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-07 09:00:08,234 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199476.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:00:17,975 INFO [train.py:901] (3/4) Epoch 25, batch 5500, loss[loss=0.2289, simple_loss=0.2923, pruned_loss=0.08273, over 7652.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2844, pruned_loss=0.05849, over 1616009.48 frames. ], batch size: 19, lr: 3.01e-03, grad_scale: 8.0 +2023-02-07 09:00:28,263 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.719e+02 2.278e+02 2.767e+02 3.622e+02 8.817e+02, threshold=5.534e+02, percent-clipped=3.0 +2023-02-07 09:00:33,817 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199512.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:00:49,506 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8012, 1.6129, 2.4045, 1.5510, 1.3801, 2.3483, 0.3820, 1.5050], + device='cuda:3'), covar=tensor([0.1428, 0.1460, 0.0367, 0.1182, 0.2360, 0.0455, 0.2082, 0.1383], + device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0201, 0.0132, 0.0221, 0.0276, 0.0142, 0.0173, 0.0199], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 09:00:52,119 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199538.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:00:53,322 INFO [train.py:901] (3/4) Epoch 25, batch 5550, loss[loss=0.2054, simple_loss=0.2783, pruned_loss=0.06623, over 7226.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.284, pruned_loss=0.05847, over 1614123.42 frames. ], batch size: 16, lr: 3.01e-03, grad_scale: 8.0 +2023-02-07 09:01:02,092 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199553.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:01:24,436 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199585.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:01:27,625 INFO [train.py:901] (3/4) Epoch 25, batch 5600, loss[loss=0.1913, simple_loss=0.2865, pruned_loss=0.048, over 8467.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2843, pruned_loss=0.05879, over 1613004.96 frames. ], batch size: 29, lr: 3.01e-03, grad_scale: 8.0 +2023-02-07 09:01:38,060 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.538e+02 3.116e+02 4.016e+02 1.228e+03, threshold=6.232e+02, percent-clipped=11.0 +2023-02-07 09:01:39,616 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3438, 1.6388, 4.5413, 1.8131, 4.0402, 3.7922, 4.1245, 3.9975], + device='cuda:3'), covar=tensor([0.0633, 0.4578, 0.0566, 0.4079, 0.1074, 0.0923, 0.0554, 0.0653], + device='cuda:3'), in_proj_covar=tensor([0.0654, 0.0654, 0.0722, 0.0644, 0.0727, 0.0618, 0.0621, 0.0697], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 09:01:43,188 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199610.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:01:47,365 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199616.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:02:03,339 INFO [train.py:901] (3/4) Epoch 25, batch 5650, loss[loss=0.2223, simple_loss=0.2977, pruned_loss=0.07341, over 7655.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.284, pruned_loss=0.0585, over 1616417.80 frames. ], batch size: 19, lr: 3.01e-03, grad_scale: 8.0 +2023-02-07 09:02:04,211 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199641.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:02:13,531 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199653.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:02:14,005 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-07 09:02:18,334 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199660.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:02:36,572 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199685.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:02:39,924 INFO [train.py:901] (3/4) Epoch 25, batch 5700, loss[loss=0.1929, simple_loss=0.2763, pruned_loss=0.05479, over 8105.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2818, pruned_loss=0.05739, over 1608621.59 frames. ], batch size: 23, lr: 3.01e-03, grad_scale: 8.0 +2023-02-07 09:02:49,769 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.681e+02 2.200e+02 2.647e+02 3.419e+02 7.306e+02, threshold=5.294e+02, percent-clipped=3.0 +2023-02-07 09:02:52,382 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-07 09:03:16,065 INFO [train.py:901] (3/4) Epoch 25, batch 5750, loss[loss=0.2074, simple_loss=0.2893, pruned_loss=0.06274, over 8504.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2812, pruned_loss=0.05745, over 1607598.67 frames. ], batch size: 26, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:03:21,551 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-07 09:03:29,026 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3270, 1.6845, 4.5384, 2.0032, 2.6995, 5.2323, 5.2726, 4.4949], + device='cuda:3'), covar=tensor([0.1201, 0.1925, 0.0272, 0.1978, 0.1117, 0.0193, 0.0523, 0.0573], + device='cuda:3'), in_proj_covar=tensor([0.0302, 0.0323, 0.0287, 0.0317, 0.0318, 0.0274, 0.0432, 0.0305], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 09:03:30,822 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199762.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:03:50,495 INFO [train.py:901] (3/4) Epoch 25, batch 5800, loss[loss=0.1707, simple_loss=0.2468, pruned_loss=0.04726, over 7447.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.282, pruned_loss=0.05777, over 1611493.31 frames. ], batch size: 17, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:04:00,808 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.628e+02 2.348e+02 2.869e+02 3.742e+02 6.332e+02, threshold=5.738e+02, percent-clipped=6.0 +2023-02-07 09:04:11,738 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199820.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:04:13,183 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2667, 3.6550, 2.2275, 2.8563, 2.9105, 1.9557, 2.9344, 3.0855], + device='cuda:3'), covar=tensor([0.1507, 0.0383, 0.1162, 0.0715, 0.0696, 0.1430, 0.0938, 0.0877], + device='cuda:3'), in_proj_covar=tensor([0.0357, 0.0236, 0.0340, 0.0311, 0.0301, 0.0342, 0.0346, 0.0319], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 09:04:26,580 INFO [train.py:901] (3/4) Epoch 25, batch 5850, loss[loss=0.2087, simple_loss=0.2957, pruned_loss=0.06089, over 8459.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2817, pruned_loss=0.05761, over 1611115.25 frames. ], batch size: 25, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:04:37,361 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199856.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:04:51,708 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199877.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:05:01,012 INFO [train.py:901] (3/4) Epoch 25, batch 5900, loss[loss=0.2807, simple_loss=0.3628, pruned_loss=0.09931, over 8489.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2818, pruned_loss=0.05752, over 1611229.82 frames. ], batch size: 28, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:05:05,825 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=199897.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:05:10,363 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.492e+02 2.352e+02 2.828e+02 3.481e+02 7.421e+02, threshold=5.657e+02, percent-clipped=3.0 +2023-02-07 09:05:13,988 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199909.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:05:26,460 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199927.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:05:31,354 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199934.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:05:32,031 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199935.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:05:35,262 INFO [train.py:901] (3/4) Epoch 25, batch 5950, loss[loss=0.2239, simple_loss=0.3195, pruned_loss=0.06415, over 8366.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2827, pruned_loss=0.05791, over 1616177.59 frames. ], batch size: 24, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:05:58,240 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199971.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:06:11,076 INFO [train.py:901] (3/4) Epoch 25, batch 6000, loss[loss=0.2416, simple_loss=0.3123, pruned_loss=0.08552, over 7438.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2825, pruned_loss=0.05826, over 1612337.59 frames. ], batch size: 72, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:06:11,076 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 09:06:23,699 INFO [train.py:935] (3/4) Epoch 25, validation: loss=0.1725, simple_loss=0.2721, pruned_loss=0.03643, over 944034.00 frames. +2023-02-07 09:06:23,700 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6747MB +2023-02-07 09:06:34,575 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.772e+02 2.373e+02 2.952e+02 3.581e+02 7.260e+02, threshold=5.903e+02, percent-clipped=4.0 +2023-02-07 09:06:38,841 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6335, 2.0183, 1.6808, 2.7727, 1.2849, 1.4175, 2.0246, 2.0415], + device='cuda:3'), covar=tensor([0.1049, 0.0902, 0.1192, 0.0382, 0.1186, 0.1673, 0.0917, 0.0869], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0194, 0.0244, 0.0211, 0.0204, 0.0247, 0.0248, 0.0207], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 09:06:40,192 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200012.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:06:46,328 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0136, 2.2876, 1.9495, 2.8929, 1.5722, 1.7560, 2.3108, 2.3450], + device='cuda:3'), covar=tensor([0.0723, 0.0763, 0.0823, 0.0321, 0.0942, 0.1244, 0.0630, 0.0762], + device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0194, 0.0244, 0.0211, 0.0205, 0.0247, 0.0248, 0.0207], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 09:06:53,225 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-07 09:06:59,664 INFO [train.py:901] (3/4) Epoch 25, batch 6050, loss[loss=0.1764, simple_loss=0.2533, pruned_loss=0.04974, over 7528.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2816, pruned_loss=0.05846, over 1608669.71 frames. ], batch size: 18, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:07:32,409 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4364, 2.5390, 1.8939, 2.2685, 2.1243, 1.6325, 2.1132, 2.1915], + device='cuda:3'), covar=tensor([0.1684, 0.0481, 0.1197, 0.0679, 0.0779, 0.1630, 0.0983, 0.1040], + device='cuda:3'), in_proj_covar=tensor([0.0361, 0.0239, 0.0344, 0.0315, 0.0305, 0.0346, 0.0350, 0.0323], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 09:07:35,022 INFO [train.py:901] (3/4) Epoch 25, batch 6100, loss[loss=0.2136, simple_loss=0.2977, pruned_loss=0.06475, over 8415.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2813, pruned_loss=0.05831, over 1606502.66 frames. ], batch size: 49, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:07:36,028 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6618, 1.9188, 3.0574, 1.4426, 2.3403, 2.0241, 1.7522, 2.2422], + device='cuda:3'), covar=tensor([0.2251, 0.3111, 0.1066, 0.5438, 0.2173, 0.3824, 0.2728, 0.2971], + device='cuda:3'), in_proj_covar=tensor([0.0533, 0.0624, 0.0558, 0.0660, 0.0658, 0.0606, 0.0554, 0.0640], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 09:07:45,350 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.318e+02 2.345e+02 2.959e+02 3.596e+02 7.197e+02, threshold=5.919e+02, percent-clipped=3.0 +2023-02-07 09:07:54,349 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-07 09:08:05,126 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2490, 3.1066, 2.9051, 1.6401, 2.8747, 2.8894, 2.8054, 2.8165], + device='cuda:3'), covar=tensor([0.1061, 0.0787, 0.1246, 0.4139, 0.1050, 0.1159, 0.1599, 0.1017], + device='cuda:3'), in_proj_covar=tensor([0.0533, 0.0450, 0.0436, 0.0547, 0.0434, 0.0454, 0.0430, 0.0398], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 09:08:05,860 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200133.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:08:11,250 INFO [train.py:901] (3/4) Epoch 25, batch 6150, loss[loss=0.2056, simple_loss=0.2598, pruned_loss=0.07575, over 7697.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2805, pruned_loss=0.05792, over 1607738.59 frames. ], batch size: 18, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:08:23,401 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200158.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:08:46,118 INFO [train.py:901] (3/4) Epoch 25, batch 6200, loss[loss=0.2552, simple_loss=0.3219, pruned_loss=0.0942, over 6514.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2793, pruned_loss=0.05729, over 1599686.37 frames. ], batch size: 72, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:08:47,079 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200191.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:08:48,286 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5691, 2.6180, 1.7647, 2.2466, 2.0978, 1.5022, 2.0418, 2.2214], + device='cuda:3'), covar=tensor([0.1628, 0.0427, 0.1286, 0.0691, 0.0846, 0.1745, 0.1074, 0.0966], + device='cuda:3'), in_proj_covar=tensor([0.0361, 0.0239, 0.0345, 0.0316, 0.0305, 0.0347, 0.0351, 0.0323], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 09:08:49,563 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200195.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 09:08:55,699 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.515e+02 2.314e+02 2.821e+02 3.535e+02 6.331e+02, threshold=5.643e+02, percent-clipped=2.0 +2023-02-07 09:09:01,020 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.16 vs. limit=5.0 +2023-02-07 09:09:05,004 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200216.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:09:13,111 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200227.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:09:21,664 INFO [train.py:901] (3/4) Epoch 25, batch 6250, loss[loss=0.2159, simple_loss=0.2993, pruned_loss=0.06626, over 8201.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2806, pruned_loss=0.05762, over 1606498.80 frames. ], batch size: 23, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:09:29,782 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200252.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 09:09:41,368 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200268.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:09:43,284 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=200271.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:09:56,025 INFO [train.py:901] (3/4) Epoch 25, batch 6300, loss[loss=0.1774, simple_loss=0.2669, pruned_loss=0.04392, over 8241.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2822, pruned_loss=0.05858, over 1609501.96 frames. ], batch size: 22, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:09:58,153 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200293.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:10:00,758 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200297.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:10:06,123 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.446e+02 2.537e+02 3.046e+02 4.211e+02 7.306e+02, threshold=6.092e+02, percent-clipped=6.0 +2023-02-07 09:10:31,233 INFO [train.py:901] (3/4) Epoch 25, batch 6350, loss[loss=0.2072, simple_loss=0.2897, pruned_loss=0.06228, over 8294.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2822, pruned_loss=0.05851, over 1612932.81 frames. ], batch size: 23, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:10:42,963 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.51 vs. limit=2.0 +2023-02-07 09:11:03,765 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200386.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:11:06,376 INFO [train.py:901] (3/4) Epoch 25, batch 6400, loss[loss=0.2102, simple_loss=0.2964, pruned_loss=0.06203, over 8362.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2824, pruned_loss=0.05849, over 1613782.09 frames. ], batch size: 24, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:11:15,859 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.521e+02 2.247e+02 2.600e+02 3.696e+02 8.014e+02, threshold=5.200e+02, percent-clipped=2.0 +2023-02-07 09:11:19,412 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8493, 2.4165, 3.7372, 2.0199, 2.0621, 3.7536, 0.7859, 2.1790], + device='cuda:3'), covar=tensor([0.1317, 0.1420, 0.0230, 0.1549, 0.2179, 0.0262, 0.2085, 0.1432], + device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0200, 0.0131, 0.0219, 0.0274, 0.0141, 0.0171, 0.0197], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 09:11:40,859 INFO [train.py:901] (3/4) Epoch 25, batch 6450, loss[loss=0.1849, simple_loss=0.2766, pruned_loss=0.04658, over 8366.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.282, pruned_loss=0.05816, over 1613198.54 frames. ], batch size: 24, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:11:44,528 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 +2023-02-07 09:12:16,057 INFO [train.py:901] (3/4) Epoch 25, batch 6500, loss[loss=0.1686, simple_loss=0.2551, pruned_loss=0.04104, over 7659.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2826, pruned_loss=0.05797, over 1611914.58 frames. ], batch size: 19, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:12:26,027 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.548e+02 2.181e+02 2.613e+02 3.190e+02 4.719e+02, threshold=5.226e+02, percent-clipped=0.0 +2023-02-07 09:12:49,416 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=200539.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 09:12:49,927 INFO [train.py:901] (3/4) Epoch 25, batch 6550, loss[loss=0.2662, simple_loss=0.3315, pruned_loss=0.1005, over 6894.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2842, pruned_loss=0.05864, over 1611077.74 frames. ], batch size: 71, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:13:09,855 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-07 09:13:26,032 INFO [train.py:901] (3/4) Epoch 25, batch 6600, loss[loss=0.2183, simple_loss=0.3056, pruned_loss=0.06551, over 8468.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2834, pruned_loss=0.05856, over 1610126.27 frames. ], batch size: 29, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:13:30,830 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-07 09:13:35,558 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.529e+02 2.321e+02 2.722e+02 3.541e+02 8.507e+02, threshold=5.445e+02, percent-clipped=6.0 +2023-02-07 09:14:00,770 INFO [train.py:901] (3/4) Epoch 25, batch 6650, loss[loss=0.2222, simple_loss=0.3028, pruned_loss=0.07084, over 8493.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2848, pruned_loss=0.05932, over 1614540.76 frames. ], batch size: 26, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:14:01,602 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=200641.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:14:02,424 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200642.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:14:10,430 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200654.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:14:16,377 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200663.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:14:19,893 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200667.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:14:35,348 INFO [train.py:901] (3/4) Epoch 25, batch 6700, loss[loss=0.185, simple_loss=0.2791, pruned_loss=0.04546, over 8186.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.284, pruned_loss=0.0588, over 1611937.52 frames. ], batch size: 23, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:14:45,630 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.726e+02 2.443e+02 2.859e+02 3.397e+02 5.440e+02, threshold=5.717e+02, percent-clipped=0.0 +2023-02-07 09:15:01,497 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5476, 1.7562, 1.8828, 1.1593, 1.9237, 1.3855, 0.3930, 1.7843], + device='cuda:3'), covar=tensor([0.0609, 0.0430, 0.0360, 0.0656, 0.0438, 0.1055, 0.0978, 0.0346], + device='cuda:3'), in_proj_covar=tensor([0.0466, 0.0403, 0.0363, 0.0460, 0.0393, 0.0546, 0.0402, 0.0432], + device='cuda:3'), out_proj_covar=tensor([1.2363e-04, 1.0499e-04, 9.4921e-05, 1.2055e-04, 1.0303e-04, 1.5292e-04, + 1.0762e-04, 1.1364e-04], device='cuda:3') +2023-02-07 09:15:10,910 INFO [train.py:901] (3/4) Epoch 25, batch 6750, loss[loss=0.1885, simple_loss=0.2746, pruned_loss=0.05115, over 8254.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2842, pruned_loss=0.05819, over 1616038.12 frames. ], batch size: 24, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:15:11,172 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2541, 2.0555, 2.7066, 2.2356, 2.7215, 2.3302, 2.1551, 1.5863], + device='cuda:3'), covar=tensor([0.5799, 0.5067, 0.2007, 0.3774, 0.2534, 0.3304, 0.1967, 0.5459], + device='cuda:3'), in_proj_covar=tensor([0.0956, 0.1007, 0.0823, 0.0978, 0.1019, 0.0918, 0.0766, 0.0841], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 09:15:11,517 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 +2023-02-07 09:15:22,795 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200756.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:15:47,008 INFO [train.py:901] (3/4) Epoch 25, batch 6800, loss[loss=0.2396, simple_loss=0.3206, pruned_loss=0.07935, over 8656.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2843, pruned_loss=0.0585, over 1617863.91 frames. ], batch size: 34, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:15:51,862 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-07 09:15:56,787 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.622e+02 2.322e+02 2.853e+02 3.502e+02 6.162e+02, threshold=5.706e+02, percent-clipped=1.0 +2023-02-07 09:16:21,863 INFO [train.py:901] (3/4) Epoch 25, batch 6850, loss[loss=0.1961, simple_loss=0.2904, pruned_loss=0.05092, over 8250.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2841, pruned_loss=0.05832, over 1617694.97 frames. ], batch size: 24, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:16:40,949 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-07 09:16:56,587 INFO [train.py:901] (3/4) Epoch 25, batch 6900, loss[loss=0.1917, simple_loss=0.2834, pruned_loss=0.05006, over 8138.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2845, pruned_loss=0.05863, over 1619439.86 frames. ], batch size: 22, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:17:06,815 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.582e+02 2.244e+02 2.770e+02 3.533e+02 6.127e+02, threshold=5.541e+02, percent-clipped=2.0 +2023-02-07 09:17:11,222 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200910.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:17:28,201 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200935.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:17:31,354 INFO [train.py:901] (3/4) Epoch 25, batch 6950, loss[loss=0.2407, simple_loss=0.3315, pruned_loss=0.07494, over 8099.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2848, pruned_loss=0.05906, over 1617234.51 frames. ], batch size: 23, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:17:50,969 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-07 09:17:58,881 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1109, 1.1912, 1.2533, 0.9340, 1.2491, 1.0160, 0.3468, 1.2149], + device='cuda:3'), covar=tensor([0.0422, 0.0347, 0.0259, 0.0401, 0.0376, 0.0680, 0.0796, 0.0248], + device='cuda:3'), in_proj_covar=tensor([0.0467, 0.0405, 0.0363, 0.0462, 0.0393, 0.0547, 0.0404, 0.0434], + device='cuda:3'), out_proj_covar=tensor([1.2390e-04, 1.0556e-04, 9.5089e-05, 1.2130e-04, 1.0309e-04, 1.5310e-04, + 1.0798e-04, 1.1394e-04], device='cuda:3') +2023-02-07 09:18:07,788 INFO [train.py:901] (3/4) Epoch 25, batch 7000, loss[loss=0.2063, simple_loss=0.2914, pruned_loss=0.06054, over 8477.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2853, pruned_loss=0.05956, over 1617243.85 frames. ], batch size: 39, lr: 3.00e-03, grad_scale: 8.0 +2023-02-07 09:18:12,093 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9177, 2.1177, 1.7998, 2.6768, 1.2628, 1.5737, 1.9578, 2.0656], + device='cuda:3'), covar=tensor([0.0731, 0.0775, 0.0878, 0.0357, 0.1101, 0.1391, 0.0767, 0.0764], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0195, 0.0244, 0.0211, 0.0204, 0.0247, 0.0248, 0.0205], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 09:18:17,568 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.644e+02 2.593e+02 3.026e+02 3.851e+02 8.547e+02, threshold=6.052e+02, percent-clipped=7.0 +2023-02-07 09:18:19,776 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=201007.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:18:23,201 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201012.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:18:40,488 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201037.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:18:42,298 INFO [train.py:901] (3/4) Epoch 25, batch 7050, loss[loss=0.1969, simple_loss=0.2839, pruned_loss=0.05497, over 8465.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2847, pruned_loss=0.05948, over 1614933.70 frames. ], batch size: 29, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:19:16,835 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201089.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:19:17,389 INFO [train.py:901] (3/4) Epoch 25, batch 7100, loss[loss=0.1596, simple_loss=0.2484, pruned_loss=0.03541, over 7797.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2836, pruned_loss=0.05879, over 1615251.49 frames. ], batch size: 19, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:19:26,871 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.650e+02 2.246e+02 2.728e+02 3.277e+02 5.322e+02, threshold=5.456e+02, percent-clipped=0.0 +2023-02-07 09:19:40,021 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201122.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:19:52,248 INFO [train.py:901] (3/4) Epoch 25, batch 7150, loss[loss=0.1949, simple_loss=0.2861, pruned_loss=0.05183, over 8194.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2835, pruned_loss=0.05853, over 1617387.62 frames. ], batch size: 23, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:19:53,558 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.19 vs. limit=5.0 +2023-02-07 09:20:28,441 INFO [train.py:901] (3/4) Epoch 25, batch 7200, loss[loss=0.2318, simple_loss=0.3246, pruned_loss=0.06946, over 8607.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2839, pruned_loss=0.05887, over 1618212.54 frames. ], batch size: 31, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:20:34,999 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 +2023-02-07 09:20:38,240 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.637e+02 2.475e+02 3.123e+02 4.294e+02 9.608e+02, threshold=6.246e+02, percent-clipped=8.0 +2023-02-07 09:20:56,018 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-07 09:21:03,479 INFO [train.py:901] (3/4) Epoch 25, batch 7250, loss[loss=0.1899, simple_loss=0.2754, pruned_loss=0.05218, over 8227.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2831, pruned_loss=0.05843, over 1617218.57 frames. ], batch size: 22, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:21:25,201 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201271.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:21:37,431 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4100, 1.1846, 2.3715, 1.2690, 2.1519, 2.5659, 2.7119, 2.1620], + device='cuda:3'), covar=tensor([0.1144, 0.1665, 0.0462, 0.2123, 0.0784, 0.0406, 0.0856, 0.0716], + device='cuda:3'), in_proj_covar=tensor([0.0303, 0.0325, 0.0290, 0.0319, 0.0318, 0.0276, 0.0436, 0.0307], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 09:21:37,959 INFO [train.py:901] (3/4) Epoch 25, batch 7300, loss[loss=0.1885, simple_loss=0.2772, pruned_loss=0.04991, over 7974.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2824, pruned_loss=0.05769, over 1617006.39 frames. ], batch size: 21, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:21:39,379 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201292.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:21:48,706 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.671e+02 2.341e+02 2.809e+02 3.464e+02 9.506e+02, threshold=5.617e+02, percent-clipped=4.0 +2023-02-07 09:21:50,349 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2907, 2.0939, 2.6779, 2.2630, 2.6630, 2.4165, 2.2056, 1.6202], + device='cuda:3'), covar=tensor([0.5737, 0.5416, 0.2202, 0.4071, 0.2614, 0.3355, 0.1972, 0.5719], + device='cuda:3'), in_proj_covar=tensor([0.0957, 0.1008, 0.0826, 0.0979, 0.1021, 0.0918, 0.0767, 0.0844], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 09:22:13,163 INFO [train.py:901] (3/4) Epoch 25, batch 7350, loss[loss=0.2221, simple_loss=0.3057, pruned_loss=0.0693, over 8470.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2823, pruned_loss=0.05723, over 1614189.20 frames. ], batch size: 25, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:22:27,893 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.86 vs. limit=5.0 +2023-02-07 09:22:39,016 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-07 09:22:40,539 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201378.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:22:48,513 INFO [train.py:901] (3/4) Epoch 25, batch 7400, loss[loss=0.1617, simple_loss=0.2438, pruned_loss=0.03976, over 7806.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2822, pruned_loss=0.05715, over 1616324.07 frames. ], batch size: 19, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:22:57,496 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201403.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:22:57,970 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.658e+02 2.231e+02 2.880e+02 3.857e+02 7.685e+02, threshold=5.759e+02, percent-clipped=5.0 +2023-02-07 09:22:58,695 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-07 09:23:19,303 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=201433.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:23:23,998 INFO [train.py:901] (3/4) Epoch 25, batch 7450, loss[loss=0.1866, simple_loss=0.2734, pruned_loss=0.04991, over 8246.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.282, pruned_loss=0.05756, over 1615611.35 frames. ], batch size: 22, lr: 2.99e-03, grad_scale: 16.0 +2023-02-07 09:23:37,754 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-07 09:23:59,779 INFO [train.py:901] (3/4) Epoch 25, batch 7500, loss[loss=0.1815, simple_loss=0.263, pruned_loss=0.05006, over 7642.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2824, pruned_loss=0.05769, over 1612692.07 frames. ], batch size: 19, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:24:09,788 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.564e+02 2.281e+02 2.758e+02 3.564e+02 6.593e+02, threshold=5.515e+02, percent-clipped=6.0 +2023-02-07 09:24:12,067 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5918, 2.4303, 1.7493, 2.2954, 1.9835, 1.4005, 1.9916, 2.2198], + device='cuda:3'), covar=tensor([0.1436, 0.0419, 0.1411, 0.0534, 0.0867, 0.1862, 0.1126, 0.0878], + device='cuda:3'), in_proj_covar=tensor([0.0360, 0.0239, 0.0345, 0.0316, 0.0304, 0.0348, 0.0352, 0.0324], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 09:24:34,709 INFO [train.py:901] (3/4) Epoch 25, batch 7550, loss[loss=0.1992, simple_loss=0.2894, pruned_loss=0.05454, over 8194.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2831, pruned_loss=0.05838, over 1610674.60 frames. ], batch size: 23, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:24:40,275 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201548.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:25:09,416 INFO [train.py:901] (3/4) Epoch 25, batch 7600, loss[loss=0.2022, simple_loss=0.2899, pruned_loss=0.05727, over 8401.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2827, pruned_loss=0.05838, over 1608291.83 frames. ], batch size: 48, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:25:20,539 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.458e+02 2.467e+02 2.939e+02 3.909e+02 7.265e+02, threshold=5.878e+02, percent-clipped=5.0 +2023-02-07 09:25:27,306 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=201615.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:25:41,314 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=201636.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:25:43,930 INFO [train.py:901] (3/4) Epoch 25, batch 7650, loss[loss=0.2134, simple_loss=0.2959, pruned_loss=0.06542, over 8252.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2842, pruned_loss=0.05898, over 1614650.69 frames. ], batch size: 24, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:26:00,386 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201662.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:26:12,135 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201679.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:26:19,113 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.46 vs. limit=5.0 +2023-02-07 09:26:19,457 INFO [train.py:901] (3/4) Epoch 25, batch 7700, loss[loss=0.2088, simple_loss=0.2898, pruned_loss=0.06392, over 8491.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2839, pruned_loss=0.0589, over 1613179.70 frames. ], batch size: 29, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:26:20,539 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-07 09:26:21,002 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8481, 1.4468, 4.0161, 1.5218, 3.5634, 3.3616, 3.6822, 3.5540], + device='cuda:3'), covar=tensor([0.0631, 0.4269, 0.0616, 0.4066, 0.1182, 0.0924, 0.0602, 0.0749], + device='cuda:3'), in_proj_covar=tensor([0.0658, 0.0656, 0.0722, 0.0648, 0.0736, 0.0627, 0.0626, 0.0699], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 09:26:30,382 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.665e+02 2.306e+02 2.805e+02 3.732e+02 7.115e+02, threshold=5.609e+02, percent-clipped=1.0 +2023-02-07 09:26:43,965 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201724.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:26:48,029 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201730.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:26:49,246 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-07 09:26:54,629 INFO [train.py:901] (3/4) Epoch 25, batch 7750, loss[loss=0.2136, simple_loss=0.2882, pruned_loss=0.0695, over 8094.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2828, pruned_loss=0.05851, over 1613697.33 frames. ], batch size: 21, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:26:58,887 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7403, 2.0158, 2.1406, 1.3730, 2.2457, 1.5611, 0.6780, 2.0467], + device='cuda:3'), covar=tensor([0.0766, 0.0417, 0.0331, 0.0744, 0.0517, 0.0951, 0.1125, 0.0356], + device='cuda:3'), in_proj_covar=tensor([0.0463, 0.0402, 0.0360, 0.0458, 0.0391, 0.0543, 0.0402, 0.0431], + device='cuda:3'), out_proj_covar=tensor([1.2294e-04, 1.0476e-04, 9.3949e-05, 1.1998e-04, 1.0235e-04, 1.5191e-04, + 1.0748e-04, 1.1315e-04], device='cuda:3') +2023-02-07 09:27:02,274 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201751.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:27:11,230 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.9165, 2.3803, 3.8143, 1.9296, 2.0013, 3.7819, 0.6122, 2.2191], + device='cuda:3'), covar=tensor([0.1254, 0.1101, 0.0155, 0.1454, 0.2193, 0.0182, 0.2023, 0.1157], + device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0201, 0.0130, 0.0221, 0.0274, 0.0141, 0.0171, 0.0198], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 09:27:14,639 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5742, 1.7547, 1.5719, 2.1232, 1.1308, 1.3729, 1.6726, 1.7324], + device='cuda:3'), covar=tensor([0.0797, 0.0812, 0.0954, 0.0565, 0.1056, 0.1338, 0.0719, 0.0719], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0195, 0.0245, 0.0211, 0.0204, 0.0247, 0.0248, 0.0206], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 09:27:29,980 INFO [train.py:901] (3/4) Epoch 25, batch 7800, loss[loss=0.1813, simple_loss=0.2724, pruned_loss=0.04516, over 8197.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2821, pruned_loss=0.0582, over 1612202.27 frames. ], batch size: 23, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:27:40,037 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201804.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:27:40,476 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.301e+02 2.955e+02 3.831e+02 1.047e+03, threshold=5.910e+02, percent-clipped=5.0 +2023-02-07 09:27:56,325 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201829.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:28:03,384 INFO [train.py:901] (3/4) Epoch 25, batch 7850, loss[loss=0.1533, simple_loss=0.2344, pruned_loss=0.03615, over 7702.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.284, pruned_loss=0.05875, over 1617468.26 frames. ], batch size: 18, lr: 2.99e-03, grad_scale: 4.0 +2023-02-07 09:28:36,528 INFO [train.py:901] (3/4) Epoch 25, batch 7900, loss[loss=0.2092, simple_loss=0.2923, pruned_loss=0.06302, over 8321.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2835, pruned_loss=0.05874, over 1615095.53 frames. ], batch size: 25, lr: 2.99e-03, grad_scale: 4.0 +2023-02-07 09:28:47,146 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 2.505e+02 3.187e+02 3.787e+02 7.491e+02, threshold=6.375e+02, percent-clipped=2.0 +2023-02-07 09:28:48,375 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-07 09:29:01,888 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1713, 1.5959, 1.7886, 1.5139, 0.9969, 1.5965, 1.8338, 1.7758], + device='cuda:3'), covar=tensor([0.0529, 0.1275, 0.1651, 0.1406, 0.0628, 0.1476, 0.0680, 0.0621], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0153, 0.0189, 0.0160, 0.0100, 0.0163, 0.0112, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 09:29:09,579 INFO [train.py:901] (3/4) Epoch 25, batch 7950, loss[loss=0.168, simple_loss=0.2564, pruned_loss=0.03983, over 7927.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2843, pruned_loss=0.0593, over 1616528.73 frames. ], batch size: 20, lr: 2.99e-03, grad_scale: 4.0 +2023-02-07 09:29:37,534 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1699, 1.4317, 1.6516, 1.3730, 0.9685, 1.4347, 1.8946, 1.7247], + device='cuda:3'), covar=tensor([0.0520, 0.1215, 0.1705, 0.1441, 0.0613, 0.1467, 0.0671, 0.0595], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0152, 0.0189, 0.0160, 0.0100, 0.0162, 0.0112, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 09:29:40,347 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201986.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:29:42,800 INFO [train.py:901] (3/4) Epoch 25, batch 8000, loss[loss=0.1548, simple_loss=0.2385, pruned_loss=0.03553, over 7800.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2834, pruned_loss=0.05897, over 1614659.21 frames. ], batch size: 19, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:29:54,398 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.313e+02 2.298e+02 3.131e+02 3.789e+02 6.155e+02, threshold=6.263e+02, percent-clipped=0.0 +2023-02-07 09:29:54,504 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=202006.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:29:55,326 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202007.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:29:55,946 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202008.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:29:58,064 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202011.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:30:06,543 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=202023.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:30:12,416 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202032.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:30:17,710 INFO [train.py:901] (3/4) Epoch 25, batch 8050, loss[loss=0.1916, simple_loss=0.2728, pruned_loss=0.05523, over 7942.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2833, pruned_loss=0.05944, over 1599617.34 frames. ], batch size: 20, lr: 2.99e-03, grad_scale: 8.0 +2023-02-07 09:30:36,925 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=202068.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:30:50,486 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-07 09:30:55,056 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-07 09:30:55,323 INFO [train.py:901] (3/4) Epoch 26, batch 0, loss[loss=0.1994, simple_loss=0.2694, pruned_loss=0.06474, over 7521.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2694, pruned_loss=0.06474, over 7521.00 frames. ], batch size: 18, lr: 2.93e-03, grad_scale: 8.0 +2023-02-07 09:30:55,323 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 09:31:05,242 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0304, 1.6960, 1.3927, 1.5034, 1.4471, 1.2325, 1.3616, 1.3770], + device='cuda:3'), covar=tensor([0.1092, 0.0340, 0.0957, 0.0432, 0.0618, 0.1315, 0.0860, 0.0646], + device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0239, 0.0341, 0.0314, 0.0303, 0.0346, 0.0350, 0.0322], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 09:31:06,909 INFO [train.py:935] (3/4) Epoch 26, validation: loss=0.1717, simple_loss=0.2716, pruned_loss=0.03591, over 944034.00 frames. +2023-02-07 09:31:06,910 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6747MB +2023-02-07 09:31:21,599 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-07 09:31:29,809 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.361e+02 2.411e+02 2.993e+02 3.956e+02 9.314e+02, threshold=5.987e+02, percent-clipped=4.0 +2023-02-07 09:31:40,833 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202121.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:31:41,325 INFO [train.py:901] (3/4) Epoch 26, batch 50, loss[loss=0.1995, simple_loss=0.2839, pruned_loss=0.05755, over 8606.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2834, pruned_loss=0.05896, over 367187.02 frames. ], batch size: 39, lr: 2.93e-03, grad_scale: 8.0 +2023-02-07 09:31:52,559 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202138.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:31:55,752 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-07 09:32:15,980 INFO [train.py:901] (3/4) Epoch 26, batch 100, loss[loss=0.1725, simple_loss=0.2526, pruned_loss=0.04619, over 7921.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2819, pruned_loss=0.05807, over 640503.17 frames. ], batch size: 20, lr: 2.93e-03, grad_scale: 8.0 +2023-02-07 09:32:18,612 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-07 09:32:23,607 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202183.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:32:40,583 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.631e+02 2.435e+02 2.962e+02 3.649e+02 8.375e+02, threshold=5.925e+02, percent-clipped=4.0 +2023-02-07 09:32:51,110 INFO [train.py:901] (3/4) Epoch 26, batch 150, loss[loss=0.2015, simple_loss=0.2858, pruned_loss=0.05856, over 7814.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2852, pruned_loss=0.05963, over 856463.60 frames. ], batch size: 20, lr: 2.93e-03, grad_scale: 8.0 +2023-02-07 09:33:22,416 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8294, 1.6228, 3.9887, 1.5708, 3.5536, 3.3472, 3.6626, 3.5796], + device='cuda:3'), covar=tensor([0.0697, 0.4104, 0.0628, 0.4134, 0.1161, 0.0996, 0.0633, 0.0706], + device='cuda:3'), in_proj_covar=tensor([0.0660, 0.0653, 0.0721, 0.0646, 0.0730, 0.0626, 0.0625, 0.0699], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 09:33:26,394 INFO [train.py:901] (3/4) Epoch 26, batch 200, loss[loss=0.184, simple_loss=0.2653, pruned_loss=0.05135, over 8030.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2861, pruned_loss=0.06047, over 1023931.05 frames. ], batch size: 20, lr: 2.93e-03, grad_scale: 8.0 +2023-02-07 09:33:49,943 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.393e+02 2.408e+02 2.928e+02 3.669e+02 9.390e+02, threshold=5.857e+02, percent-clipped=3.0 +2023-02-07 09:34:01,569 INFO [train.py:901] (3/4) Epoch 26, batch 250, loss[loss=0.1942, simple_loss=0.2771, pruned_loss=0.05564, over 8109.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2849, pruned_loss=0.05945, over 1157042.39 frames. ], batch size: 23, lr: 2.93e-03, grad_scale: 8.0 +2023-02-07 09:34:09,766 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-07 09:34:19,972 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-07 09:34:22,771 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=202352.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:34:35,413 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 +2023-02-07 09:34:36,401 INFO [train.py:901] (3/4) Epoch 26, batch 300, loss[loss=0.1897, simple_loss=0.278, pruned_loss=0.05065, over 8033.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2843, pruned_loss=0.05957, over 1253007.37 frames. ], batch size: 22, lr: 2.93e-03, grad_scale: 8.0 +2023-02-07 09:34:40,021 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202377.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:34:52,245 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202394.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:34:57,265 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202402.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:34:59,615 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.339e+02 2.510e+02 3.033e+02 3.572e+02 1.183e+03, threshold=6.066e+02, percent-clipped=2.0 +2023-02-07 09:35:08,715 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202419.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:35:10,555 INFO [train.py:901] (3/4) Epoch 26, batch 350, loss[loss=0.1607, simple_loss=0.2439, pruned_loss=0.03876, over 7787.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2853, pruned_loss=0.05998, over 1340892.93 frames. ], batch size: 19, lr: 2.93e-03, grad_scale: 4.0 +2023-02-07 09:35:22,062 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.6431, 2.4444, 3.2450, 2.6471, 3.3510, 2.6138, 2.4653, 2.0904], + device='cuda:3'), covar=tensor([0.5526, 0.5189, 0.2164, 0.3881, 0.2440, 0.3185, 0.1885, 0.5746], + device='cuda:3'), in_proj_covar=tensor([0.0954, 0.1006, 0.0820, 0.0977, 0.1015, 0.0916, 0.0764, 0.0840], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 09:35:23,413 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202439.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:35:40,995 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202464.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:35:43,118 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202467.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:35:46,424 INFO [train.py:901] (3/4) Epoch 26, batch 400, loss[loss=0.1808, simple_loss=0.2672, pruned_loss=0.04723, over 8085.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2837, pruned_loss=0.05913, over 1397363.71 frames. ], batch size: 21, lr: 2.93e-03, grad_scale: 8.0 +2023-02-07 09:35:46,710 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0829, 1.8538, 2.3068, 2.0006, 2.3043, 2.1213, 1.9867, 1.2644], + device='cuda:3'), covar=tensor([0.5478, 0.4875, 0.2125, 0.3768, 0.2635, 0.3294, 0.1937, 0.5152], + device='cuda:3'), in_proj_covar=tensor([0.0953, 0.1005, 0.0820, 0.0977, 0.1015, 0.0915, 0.0763, 0.0840], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 09:36:09,159 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7375, 1.8842, 1.6189, 2.0595, 1.3290, 1.5276, 1.7374, 1.8982], + device='cuda:3'), covar=tensor([0.0646, 0.0692, 0.0831, 0.0610, 0.1032, 0.1030, 0.0634, 0.0595], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0195, 0.0245, 0.0212, 0.0206, 0.0246, 0.0249, 0.0207], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 09:36:11,049 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.432e+02 2.504e+02 3.071e+02 3.633e+02 8.131e+02, threshold=6.142e+02, percent-clipped=3.0 +2023-02-07 09:36:21,079 INFO [train.py:901] (3/4) Epoch 26, batch 450, loss[loss=0.1977, simple_loss=0.2935, pruned_loss=0.05094, over 8651.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2844, pruned_loss=0.05964, over 1445618.40 frames. ], batch size: 34, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:36:40,725 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.45 vs. limit=2.0 +2023-02-07 09:36:55,498 INFO [train.py:901] (3/4) Epoch 26, batch 500, loss[loss=0.2032, simple_loss=0.2848, pruned_loss=0.06079, over 8508.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.283, pruned_loss=0.05886, over 1480018.65 frames. ], batch size: 26, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:37:19,246 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.631e+02 2.396e+02 2.962e+02 4.085e+02 8.069e+02, threshold=5.924e+02, percent-clipped=6.0 +2023-02-07 09:37:29,360 INFO [train.py:901] (3/4) Epoch 26, batch 550, loss[loss=0.1764, simple_loss=0.269, pruned_loss=0.04195, over 8338.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2847, pruned_loss=0.06007, over 1513062.66 frames. ], batch size: 26, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:37:41,179 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 +2023-02-07 09:37:52,230 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.62 vs. limit=2.0 +2023-02-07 09:38:01,539 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5422, 1.2000, 1.3634, 1.1475, 0.9549, 1.2378, 1.5266, 1.4919], + device='cuda:3'), covar=tensor([0.0637, 0.1323, 0.1742, 0.1485, 0.0684, 0.1489, 0.0786, 0.0537], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0153, 0.0190, 0.0161, 0.0100, 0.0163, 0.0112, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 09:38:05,143 INFO [train.py:901] (3/4) Epoch 26, batch 600, loss[loss=0.2263, simple_loss=0.3095, pruned_loss=0.07155, over 8555.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2855, pruned_loss=0.06066, over 1534549.89 frames. ], batch size: 31, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:38:21,735 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-07 09:38:29,016 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.819e+02 2.445e+02 2.916e+02 3.512e+02 6.749e+02, threshold=5.833e+02, percent-clipped=3.0 +2023-02-07 09:38:38,961 INFO [train.py:901] (3/4) Epoch 26, batch 650, loss[loss=0.2065, simple_loss=0.289, pruned_loss=0.06202, over 8247.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2864, pruned_loss=0.06087, over 1552896.76 frames. ], batch size: 24, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:38:39,879 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202723.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:38:57,386 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202748.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:39:01,544 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=5.04 vs. limit=5.0 +2023-02-07 09:39:14,899 INFO [train.py:901] (3/4) Epoch 26, batch 700, loss[loss=0.2253, simple_loss=0.3016, pruned_loss=0.07452, over 8467.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2852, pruned_loss=0.05999, over 1562140.66 frames. ], batch size: 49, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:39:38,616 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.871e+02 2.498e+02 3.029e+02 3.750e+02 8.351e+02, threshold=6.058e+02, percent-clipped=3.0 +2023-02-07 09:39:49,883 INFO [train.py:901] (3/4) Epoch 26, batch 750, loss[loss=0.216, simple_loss=0.2887, pruned_loss=0.07165, over 7828.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2842, pruned_loss=0.0594, over 1573746.37 frames. ], batch size: 20, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:40:05,052 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-07 09:40:08,006 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202848.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:40:13,827 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-07 09:40:24,780 INFO [train.py:901] (3/4) Epoch 26, batch 800, loss[loss=0.1656, simple_loss=0.2404, pruned_loss=0.04538, over 7804.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2826, pruned_loss=0.05812, over 1585552.73 frames. ], batch size: 19, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:40:26,981 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1400, 1.4002, 1.7577, 1.3117, 1.0176, 1.4993, 1.8443, 1.5401], + device='cuda:3'), covar=tensor([0.0521, 0.1374, 0.1748, 0.1598, 0.0617, 0.1485, 0.0716, 0.0730], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0153, 0.0191, 0.0161, 0.0100, 0.0163, 0.0112, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 09:40:49,966 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.744e+02 2.489e+02 2.818e+02 3.827e+02 7.280e+02, threshold=5.635e+02, percent-clipped=3.0 +2023-02-07 09:40:59,909 INFO [train.py:901] (3/4) Epoch 26, batch 850, loss[loss=0.2248, simple_loss=0.3019, pruned_loss=0.07386, over 8486.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2835, pruned_loss=0.05889, over 1589329.83 frames. ], batch size: 26, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:41:09,138 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1539, 3.5933, 2.3215, 2.9895, 2.9845, 1.9515, 3.0360, 3.0754], + device='cuda:3'), covar=tensor([0.1655, 0.0438, 0.1134, 0.0718, 0.0687, 0.1576, 0.0973, 0.1038], + device='cuda:3'), in_proj_covar=tensor([0.0359, 0.0239, 0.0343, 0.0315, 0.0303, 0.0348, 0.0351, 0.0322], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 09:41:33,357 INFO [train.py:901] (3/4) Epoch 26, batch 900, loss[loss=0.2067, simple_loss=0.2937, pruned_loss=0.05982, over 8485.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2839, pruned_loss=0.05897, over 1596527.07 frames. ], batch size: 49, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:41:58,979 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.525e+02 2.742e+02 3.265e+02 4.005e+02 6.934e+02, threshold=6.531e+02, percent-clipped=5.0 +2023-02-07 09:42:08,858 INFO [train.py:901] (3/4) Epoch 26, batch 950, loss[loss=0.2072, simple_loss=0.2908, pruned_loss=0.06176, over 8455.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2829, pruned_loss=0.05844, over 1599880.78 frames. ], batch size: 27, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:42:19,052 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8138, 1.6793, 1.9250, 1.7281, 1.0067, 1.7650, 2.2090, 2.0195], + device='cuda:3'), covar=tensor([0.0471, 0.1235, 0.1625, 0.1375, 0.0631, 0.1442, 0.0631, 0.0610], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0153, 0.0190, 0.0161, 0.0100, 0.0162, 0.0112, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 09:42:31,714 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-07 09:42:32,517 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203056.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:42:43,152 INFO [train.py:901] (3/4) Epoch 26, batch 1000, loss[loss=0.1825, simple_loss=0.2709, pruned_loss=0.04705, over 8499.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2833, pruned_loss=0.05897, over 1604498.85 frames. ], batch size: 26, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:42:50,158 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-07 09:43:04,725 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203103.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:43:05,257 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-07 09:43:07,153 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.540e+02 2.348e+02 2.857e+02 3.355e+02 6.976e+02, threshold=5.714e+02, percent-clipped=1.0 +2023-02-07 09:43:17,240 INFO [train.py:901] (3/4) Epoch 26, batch 1050, loss[loss=0.183, simple_loss=0.2619, pruned_loss=0.05199, over 7788.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2837, pruned_loss=0.05885, over 1604334.08 frames. ], batch size: 19, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:43:18,682 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-07 09:43:38,435 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1131, 1.0041, 1.1952, 0.9385, 0.9092, 1.1956, 0.0984, 0.8764], + device='cuda:3'), covar=tensor([0.1440, 0.1052, 0.0478, 0.0710, 0.2231, 0.0505, 0.1853, 0.1249], + device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0202, 0.0132, 0.0222, 0.0277, 0.0143, 0.0171, 0.0199], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 09:43:53,220 INFO [train.py:901] (3/4) Epoch 26, batch 1100, loss[loss=0.2036, simple_loss=0.3013, pruned_loss=0.05289, over 8247.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.283, pruned_loss=0.05833, over 1608401.22 frames. ], batch size: 24, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:44:01,157 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 +2023-02-07 09:44:06,848 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=203192.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 09:44:16,724 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.532e+02 2.511e+02 2.912e+02 3.711e+02 8.666e+02, threshold=5.824e+02, percent-clipped=4.0 +2023-02-07 09:44:24,304 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.6789, 5.8124, 4.9837, 2.6354, 5.0361, 5.4798, 5.1901, 5.3203], + device='cuda:3'), covar=tensor([0.0581, 0.0330, 0.0905, 0.4273, 0.0743, 0.0777, 0.1025, 0.0554], + device='cuda:3'), in_proj_covar=tensor([0.0530, 0.0450, 0.0437, 0.0548, 0.0435, 0.0452, 0.0427, 0.0394], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 09:44:27,579 INFO [train.py:901] (3/4) Epoch 26, batch 1150, loss[loss=0.2235, simple_loss=0.3111, pruned_loss=0.06794, over 8488.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2827, pruned_loss=0.05831, over 1610454.27 frames. ], batch size: 28, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:44:27,585 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-07 09:45:02,743 INFO [train.py:901] (3/4) Epoch 26, batch 1200, loss[loss=0.1792, simple_loss=0.2775, pruned_loss=0.04042, over 8460.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2834, pruned_loss=0.05817, over 1616516.20 frames. ], batch size: 25, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:45:27,280 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=203307.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 09:45:27,743 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.702e+02 2.402e+02 2.806e+02 3.306e+02 6.331e+02, threshold=5.612e+02, percent-clipped=2.0 +2023-02-07 09:45:35,888 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.2231, 1.5130, 1.2190, 2.6293, 1.1110, 1.1820, 1.7404, 1.7030], + device='cuda:3'), covar=tensor([0.1725, 0.1459, 0.2175, 0.0435, 0.1447, 0.2191, 0.1085, 0.1168], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0194, 0.0246, 0.0212, 0.0205, 0.0247, 0.0249, 0.0206], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 09:45:37,087 INFO [train.py:901] (3/4) Epoch 26, batch 1250, loss[loss=0.1892, simple_loss=0.2772, pruned_loss=0.05063, over 8243.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2829, pruned_loss=0.05819, over 1614427.42 frames. ], batch size: 22, lr: 2.92e-03, grad_scale: 4.0 +2023-02-07 09:46:12,727 INFO [train.py:901] (3/4) Epoch 26, batch 1300, loss[loss=0.1756, simple_loss=0.2625, pruned_loss=0.04435, over 8239.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.284, pruned_loss=0.05883, over 1617521.51 frames. ], batch size: 22, lr: 2.92e-03, grad_scale: 4.0 +2023-02-07 09:46:31,894 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=203400.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:46:37,064 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.602e+02 2.437e+02 2.919e+02 3.429e+02 9.499e+02, threshold=5.838e+02, percent-clipped=5.0 +2023-02-07 09:46:46,434 INFO [train.py:901] (3/4) Epoch 26, batch 1350, loss[loss=0.1954, simple_loss=0.2821, pruned_loss=0.05435, over 8131.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2847, pruned_loss=0.05905, over 1621421.65 frames. ], batch size: 22, lr: 2.92e-03, grad_scale: 4.0 +2023-02-07 09:47:04,180 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=203447.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:47:22,495 INFO [train.py:901] (3/4) Epoch 26, batch 1400, loss[loss=0.1866, simple_loss=0.2761, pruned_loss=0.04851, over 8744.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2836, pruned_loss=0.05844, over 1621704.54 frames. ], batch size: 39, lr: 2.92e-03, grad_scale: 4.0 +2023-02-07 09:47:24,885 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-07 09:47:31,496 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203485.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:47:47,752 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 2.419e+02 2.906e+02 3.589e+02 5.599e+02, threshold=5.812e+02, percent-clipped=0.0 +2023-02-07 09:47:52,601 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=203515.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:47:54,406 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-07 09:47:57,076 INFO [train.py:901] (3/4) Epoch 26, batch 1450, loss[loss=0.1948, simple_loss=0.2851, pruned_loss=0.05226, over 8250.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2837, pruned_loss=0.05835, over 1622426.29 frames. ], batch size: 24, lr: 2.92e-03, grad_scale: 4.0 +2023-02-07 09:48:01,590 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-07 09:48:10,648 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6799, 2.0623, 3.1423, 1.5447, 2.2281, 2.1883, 1.7631, 2.4594], + device='cuda:3'), covar=tensor([0.1866, 0.2548, 0.0846, 0.4724, 0.2003, 0.3070, 0.2351, 0.2096], + device='cuda:3'), in_proj_covar=tensor([0.0533, 0.0623, 0.0554, 0.0659, 0.0654, 0.0602, 0.0552, 0.0637], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 09:48:18,113 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0171, 1.8597, 2.3658, 1.9911, 2.2687, 2.1178, 1.9113, 1.1056], + device='cuda:3'), covar=tensor([0.5706, 0.4929, 0.2046, 0.3797, 0.2463, 0.3268, 0.1979, 0.5467], + device='cuda:3'), in_proj_covar=tensor([0.0962, 0.1018, 0.0830, 0.0988, 0.1025, 0.0926, 0.0772, 0.0849], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 09:48:24,036 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=203562.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:48:24,762 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203563.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:48:31,364 INFO [train.py:901] (3/4) Epoch 26, batch 1500, loss[loss=0.2772, simple_loss=0.3434, pruned_loss=0.1055, over 6593.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2835, pruned_loss=0.05847, over 1618024.95 frames. ], batch size: 71, lr: 2.92e-03, grad_scale: 4.0 +2023-02-07 09:48:33,569 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 +2023-02-07 09:48:42,961 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203588.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:48:56,866 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.551e+02 2.262e+02 2.668e+02 3.517e+02 8.500e+02, threshold=5.335e+02, percent-clipped=2.0 +2023-02-07 09:49:06,820 INFO [train.py:901] (3/4) Epoch 26, batch 1550, loss[loss=0.2107, simple_loss=0.2948, pruned_loss=0.06333, over 8511.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2832, pruned_loss=0.05873, over 1619165.43 frames. ], batch size: 49, lr: 2.92e-03, grad_scale: 4.0 +2023-02-07 09:49:15,916 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6470, 1.9878, 2.0458, 1.2893, 2.1271, 1.5870, 0.5533, 1.8963], + device='cuda:3'), covar=tensor([0.0628, 0.0386, 0.0316, 0.0697, 0.0384, 0.0848, 0.0913, 0.0343], + device='cuda:3'), in_proj_covar=tensor([0.0466, 0.0405, 0.0359, 0.0457, 0.0392, 0.0545, 0.0399, 0.0432], + device='cuda:3'), out_proj_covar=tensor([1.2384e-04, 1.0548e-04, 9.3703e-05, 1.1969e-04, 1.0256e-04, 1.5250e-04, + 1.0683e-04, 1.1347e-04], device='cuda:3') +2023-02-07 09:49:40,410 INFO [train.py:901] (3/4) Epoch 26, batch 1600, loss[loss=0.2415, simple_loss=0.3186, pruned_loss=0.08226, over 8452.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2843, pruned_loss=0.05913, over 1619959.68 frames. ], batch size: 27, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:50:05,089 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.410e+02 3.021e+02 3.901e+02 1.362e+03, threshold=6.042e+02, percent-clipped=8.0 +2023-02-07 09:50:15,007 INFO [train.py:901] (3/4) Epoch 26, batch 1650, loss[loss=0.21, simple_loss=0.2823, pruned_loss=0.06886, over 7777.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2843, pruned_loss=0.05907, over 1618018.95 frames. ], batch size: 19, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:50:44,787 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203765.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:50:48,721 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203771.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:50:49,155 INFO [train.py:901] (3/4) Epoch 26, batch 1700, loss[loss=0.1666, simple_loss=0.2429, pruned_loss=0.04514, over 7441.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2828, pruned_loss=0.05805, over 1618271.00 frames. ], batch size: 17, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:50:57,400 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203784.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:51:05,472 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203796.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:51:14,297 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.578e+02 2.478e+02 3.017e+02 3.791e+02 8.735e+02, threshold=6.035e+02, percent-clipped=4.0 +2023-02-07 09:51:20,893 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203818.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:51:23,317 INFO [train.py:901] (3/4) Epoch 26, batch 1750, loss[loss=0.1883, simple_loss=0.281, pruned_loss=0.04781, over 8583.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2842, pruned_loss=0.05904, over 1617409.12 frames. ], batch size: 34, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:51:28,102 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=203829.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:51:38,328 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203843.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:51:54,076 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203865.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:51:58,819 INFO [train.py:901] (3/4) Epoch 26, batch 1800, loss[loss=0.171, simple_loss=0.2641, pruned_loss=0.03895, over 8110.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.284, pruned_loss=0.05858, over 1618199.15 frames. ], batch size: 23, lr: 2.92e-03, grad_scale: 8.0 +2023-02-07 09:52:11,901 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2085, 1.4669, 1.7112, 1.3857, 1.0831, 1.5770, 1.9889, 1.7846], + device='cuda:3'), covar=tensor([0.0557, 0.1560, 0.2080, 0.1774, 0.0724, 0.1778, 0.0773, 0.0702], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0152, 0.0188, 0.0160, 0.0099, 0.0162, 0.0112, 0.0144], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 09:52:14,504 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8756, 1.4764, 3.4147, 1.5526, 2.3769, 3.8479, 3.9485, 3.3229], + device='cuda:3'), covar=tensor([0.1265, 0.1938, 0.0337, 0.2110, 0.1096, 0.0236, 0.0449, 0.0536], + device='cuda:3'), in_proj_covar=tensor([0.0301, 0.0321, 0.0288, 0.0317, 0.0316, 0.0273, 0.0431, 0.0303], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 09:52:23,002 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.497e+02 2.441e+02 2.799e+02 3.336e+02 4.977e+02, threshold=5.598e+02, percent-clipped=0.0 +2023-02-07 09:52:29,701 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203918.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 09:52:32,265 INFO [train.py:901] (3/4) Epoch 26, batch 1850, loss[loss=0.194, simple_loss=0.2781, pruned_loss=0.05498, over 7922.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2831, pruned_loss=0.05853, over 1617414.03 frames. ], batch size: 20, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:52:47,219 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-07 09:52:47,700 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=203944.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:53:07,704 INFO [train.py:901] (3/4) Epoch 26, batch 1900, loss[loss=0.1973, simple_loss=0.2785, pruned_loss=0.05801, over 8519.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2835, pruned_loss=0.05897, over 1617997.52 frames. ], batch size: 28, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:53:33,465 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.537e+02 2.507e+02 3.073e+02 4.108e+02 9.647e+02, threshold=6.146e+02, percent-clipped=9.0 +2023-02-07 09:53:36,910 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-07 09:53:42,921 INFO [train.py:901] (3/4) Epoch 26, batch 1950, loss[loss=0.1995, simple_loss=0.2805, pruned_loss=0.05922, over 7427.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2844, pruned_loss=0.05927, over 1621727.03 frames. ], batch size: 17, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:53:49,444 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-07 09:54:07,164 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-07 09:54:17,189 INFO [train.py:901] (3/4) Epoch 26, batch 2000, loss[loss=0.1977, simple_loss=0.2848, pruned_loss=0.05533, over 8322.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2844, pruned_loss=0.05886, over 1619316.57 frames. ], batch size: 25, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:54:34,347 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204095.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:54:43,733 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.375e+02 2.388e+02 3.050e+02 3.690e+02 7.171e+02, threshold=6.101e+02, percent-clipped=4.0 +2023-02-07 09:54:44,461 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=204109.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:54:53,066 INFO [train.py:901] (3/4) Epoch 26, batch 2050, loss[loss=0.2143, simple_loss=0.3025, pruned_loss=0.063, over 8103.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2834, pruned_loss=0.0579, over 1619205.81 frames. ], batch size: 23, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:54:57,142 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=204128.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:55:26,604 INFO [train.py:901] (3/4) Epoch 26, batch 2100, loss[loss=0.1594, simple_loss=0.2469, pruned_loss=0.03597, over 7915.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2833, pruned_loss=0.05772, over 1618180.82 frames. ], batch size: 20, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:55:29,666 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.7254, 1.7009, 1.7872, 1.6406, 0.8962, 1.7066, 2.1316, 1.9364], + device='cuda:3'), covar=tensor([0.0447, 0.1240, 0.1649, 0.1397, 0.0616, 0.1470, 0.0666, 0.0613], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0153, 0.0190, 0.0161, 0.0100, 0.0164, 0.0112, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 09:55:33,650 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204181.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:55:47,793 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204200.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:55:52,931 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.644e+02 2.312e+02 2.797e+02 3.552e+02 6.063e+02, threshold=5.595e+02, percent-clipped=0.0 +2023-02-07 09:55:53,728 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=204209.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:55:56,894 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-07 09:56:02,441 INFO [train.py:901] (3/4) Epoch 26, batch 2150, loss[loss=0.1812, simple_loss=0.2577, pruned_loss=0.05231, over 7808.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2832, pruned_loss=0.0575, over 1618400.78 frames. ], batch size: 19, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:56:03,969 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204224.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:56:04,624 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204225.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:56:17,182 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204243.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:56:29,894 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=204262.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 09:56:36,283 INFO [train.py:901] (3/4) Epoch 26, batch 2200, loss[loss=0.232, simple_loss=0.3086, pruned_loss=0.07769, over 8747.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2829, pruned_loss=0.05744, over 1620304.77 frames. ], batch size: 30, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:57:01,394 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.537e+02 2.362e+02 3.099e+02 4.074e+02 1.599e+03, threshold=6.197e+02, percent-clipped=8.0 +2023-02-07 09:57:11,827 INFO [train.py:901] (3/4) Epoch 26, batch 2250, loss[loss=0.1872, simple_loss=0.2793, pruned_loss=0.0476, over 8486.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2836, pruned_loss=0.05775, over 1617540.03 frames. ], batch size: 28, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:57:13,367 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204324.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:57:46,492 INFO [train.py:901] (3/4) Epoch 26, batch 2300, loss[loss=0.1518, simple_loss=0.238, pruned_loss=0.03279, over 6412.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2831, pruned_loss=0.05787, over 1609012.79 frames. ], batch size: 14, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:57:50,096 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204377.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 09:58:10,722 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.733e+02 2.304e+02 2.813e+02 3.713e+02 7.684e+02, threshold=5.626e+02, percent-clipped=3.0 +2023-02-07 09:58:21,053 INFO [train.py:901] (3/4) Epoch 26, batch 2350, loss[loss=0.2124, simple_loss=0.2842, pruned_loss=0.07031, over 7811.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2843, pruned_loss=0.0589, over 1611182.79 frames. ], batch size: 20, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:58:33,863 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=204439.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:58:57,208 INFO [train.py:901] (3/4) Epoch 26, batch 2400, loss[loss=0.1925, simple_loss=0.2894, pruned_loss=0.04776, over 8188.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2852, pruned_loss=0.0591, over 1615371.30 frames. ], batch size: 23, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:59:02,881 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204480.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:59:16,088 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204499.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:59:20,395 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204505.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:59:22,322 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.744e+02 2.551e+02 2.904e+02 3.805e+02 7.023e+02, threshold=5.807e+02, percent-clipped=3.0 +2023-02-07 09:59:32,142 INFO [train.py:901] (3/4) Epoch 26, batch 2450, loss[loss=0.2057, simple_loss=0.2666, pruned_loss=0.07235, over 8038.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2842, pruned_loss=0.05871, over 1618969.96 frames. ], batch size: 20, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 09:59:33,746 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204524.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:59:34,344 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=204525.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 09:59:56,604 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204554.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:00:08,716 INFO [train.py:901] (3/4) Epoch 26, batch 2500, loss[loss=0.212, simple_loss=0.2851, pruned_loss=0.06942, over 5582.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2828, pruned_loss=0.0582, over 1616646.55 frames. ], batch size: 12, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:00:15,148 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204580.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:00:25,202 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1299, 3.6762, 2.3664, 2.9687, 2.9299, 2.0909, 2.9921, 3.0343], + device='cuda:3'), covar=tensor([0.1767, 0.0366, 0.1120, 0.0727, 0.0747, 0.1585, 0.0977, 0.1044], + device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0240, 0.0341, 0.0312, 0.0303, 0.0346, 0.0348, 0.0321], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 10:00:31,998 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204605.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:00:33,800 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.562e+02 2.483e+02 3.074e+02 3.585e+02 8.993e+02, threshold=6.148e+02, percent-clipped=7.0 +2023-02-07 10:00:43,170 INFO [train.py:901] (3/4) Epoch 26, batch 2550, loss[loss=0.1821, simple_loss=0.2692, pruned_loss=0.0475, over 8082.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2833, pruned_loss=0.05856, over 1620476.72 frames. ], batch size: 21, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:00:50,560 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204633.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 10:00:55,141 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204640.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:01:07,860 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204658.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 10:01:18,448 INFO [train.py:901] (3/4) Epoch 26, batch 2600, loss[loss=0.1823, simple_loss=0.2598, pruned_loss=0.05243, over 7694.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2839, pruned_loss=0.05881, over 1619574.47 frames. ], batch size: 18, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:01:43,342 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.840e+02 2.465e+02 3.094e+02 3.874e+02 9.576e+02, threshold=6.187e+02, percent-clipped=4.0 +2023-02-07 10:01:52,895 INFO [train.py:901] (3/4) Epoch 26, batch 2650, loss[loss=0.2051, simple_loss=0.2929, pruned_loss=0.05862, over 8033.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2834, pruned_loss=0.0585, over 1620715.57 frames. ], batch size: 22, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:02:14,670 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-07 10:02:28,028 INFO [train.py:901] (3/4) Epoch 26, batch 2700, loss[loss=0.1833, simple_loss=0.2554, pruned_loss=0.05567, over 7542.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2843, pruned_loss=0.05912, over 1622256.26 frames. ], batch size: 18, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:02:53,800 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.711e+02 2.359e+02 2.865e+02 3.674e+02 6.992e+02, threshold=5.730e+02, percent-clipped=1.0 +2023-02-07 10:02:55,450 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204810.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:03:04,002 INFO [train.py:901] (3/4) Epoch 26, batch 2750, loss[loss=0.2118, simple_loss=0.3012, pruned_loss=0.0612, over 8663.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.284, pruned_loss=0.05907, over 1617549.62 frames. ], batch size: 39, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:03:13,287 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204835.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:03:16,024 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204839.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:03:38,123 INFO [train.py:901] (3/4) Epoch 26, batch 2800, loss[loss=0.2317, simple_loss=0.3096, pruned_loss=0.07694, over 7810.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2837, pruned_loss=0.05875, over 1615766.53 frames. ], batch size: 20, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:03:56,005 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204896.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:04:04,627 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.740e+02 2.285e+02 3.108e+02 3.828e+02 9.944e+02, threshold=6.216e+02, percent-clipped=6.0 +2023-02-07 10:04:09,689 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6721, 1.4857, 4.8639, 1.9238, 4.2599, 4.0229, 4.3458, 4.2878], + device='cuda:3'), covar=tensor([0.0615, 0.5345, 0.0461, 0.4302, 0.1187, 0.0991, 0.0620, 0.0708], + device='cuda:3'), in_proj_covar=tensor([0.0666, 0.0664, 0.0731, 0.0651, 0.0745, 0.0630, 0.0631, 0.0707], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 10:04:13,938 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204921.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:04:14,419 INFO [train.py:901] (3/4) Epoch 26, batch 2850, loss[loss=0.2071, simple_loss=0.2839, pruned_loss=0.06513, over 8356.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2837, pruned_loss=0.05856, over 1618140.95 frames. ], batch size: 24, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:04:48,525 INFO [train.py:901] (3/4) Epoch 26, batch 2900, loss[loss=0.1953, simple_loss=0.2714, pruned_loss=0.05959, over 8193.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.284, pruned_loss=0.05906, over 1616061.79 frames. ], batch size: 23, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:05:13,380 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.634e+02 2.393e+02 3.052e+02 3.991e+02 9.487e+02, threshold=6.105e+02, percent-clipped=5.0 +2023-02-07 10:05:20,487 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-07 10:05:23,875 INFO [train.py:901] (3/4) Epoch 26, batch 2950, loss[loss=0.1858, simple_loss=0.2699, pruned_loss=0.05085, over 8096.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2839, pruned_loss=0.05845, over 1617167.50 frames. ], batch size: 21, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:05:24,057 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0973, 1.6200, 3.3945, 1.6451, 2.6531, 3.7091, 3.8485, 3.1979], + device='cuda:3'), covar=tensor([0.1140, 0.1729, 0.0334, 0.1973, 0.0921, 0.0239, 0.0590, 0.0517], + device='cuda:3'), in_proj_covar=tensor([0.0302, 0.0326, 0.0289, 0.0318, 0.0321, 0.0275, 0.0434, 0.0305], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 10:05:54,158 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5980, 1.5612, 2.0818, 1.3120, 1.1887, 2.0643, 0.3153, 1.2980], + device='cuda:3'), covar=tensor([0.1406, 0.1248, 0.0349, 0.1045, 0.2412, 0.0379, 0.1896, 0.1237], + device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0203, 0.0131, 0.0222, 0.0276, 0.0143, 0.0172, 0.0199], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 10:05:58,487 INFO [train.py:901] (3/4) Epoch 26, batch 3000, loss[loss=0.1619, simple_loss=0.2429, pruned_loss=0.04047, over 7646.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2843, pruned_loss=0.05871, over 1614401.94 frames. ], batch size: 19, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:05:58,488 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 10:06:11,416 INFO [train.py:935] (3/4) Epoch 26, validation: loss=0.1716, simple_loss=0.2713, pruned_loss=0.03593, over 944034.00 frames. +2023-02-07 10:06:11,417 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6747MB +2023-02-07 10:06:20,127 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.67 vs. limit=2.0 +2023-02-07 10:06:36,708 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.328e+02 2.246e+02 2.785e+02 3.735e+02 7.523e+02, threshold=5.571e+02, percent-clipped=3.0 +2023-02-07 10:06:46,004 INFO [train.py:901] (3/4) Epoch 26, batch 3050, loss[loss=0.186, simple_loss=0.2721, pruned_loss=0.04992, over 8243.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2856, pruned_loss=0.05972, over 1611666.33 frames. ], batch size: 24, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:06:49,445 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=205127.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:07:22,789 INFO [train.py:901] (3/4) Epoch 26, batch 3100, loss[loss=0.21, simple_loss=0.3084, pruned_loss=0.05584, over 8248.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2844, pruned_loss=0.05922, over 1609008.59 frames. ], batch size: 24, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:07:26,346 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6888, 1.3161, 4.9125, 1.9676, 4.4346, 4.0520, 4.4343, 4.3356], + device='cuda:3'), covar=tensor([0.0584, 0.4987, 0.0438, 0.3991, 0.0980, 0.0866, 0.0514, 0.0622], + device='cuda:3'), in_proj_covar=tensor([0.0666, 0.0663, 0.0730, 0.0652, 0.0743, 0.0630, 0.0631, 0.0708], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 10:07:30,228 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=205183.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:07:48,156 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.609e+02 2.327e+02 2.997e+02 4.038e+02 1.256e+03, threshold=5.993e+02, percent-clipped=7.0 +2023-02-07 10:07:53,695 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=205216.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:07:53,941 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 +2023-02-07 10:07:57,549 INFO [train.py:901] (3/4) Epoch 26, batch 3150, loss[loss=0.197, simple_loss=0.2827, pruned_loss=0.0556, over 8245.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2844, pruned_loss=0.05892, over 1612641.67 frames. ], batch size: 24, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:08:33,477 INFO [train.py:901] (3/4) Epoch 26, batch 3200, loss[loss=0.2438, simple_loss=0.3143, pruned_loss=0.08664, over 7042.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2846, pruned_loss=0.05938, over 1616188.36 frames. ], batch size: 72, lr: 2.91e-03, grad_scale: 8.0 +2023-02-07 10:08:52,245 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=205298.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:08:58,818 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.683e+02 2.527e+02 3.010e+02 3.735e+02 6.895e+02, threshold=6.021e+02, percent-clipped=2.0 +2023-02-07 10:09:09,102 INFO [train.py:901] (3/4) Epoch 26, batch 3250, loss[loss=0.1821, simple_loss=0.268, pruned_loss=0.04804, over 7807.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2845, pruned_loss=0.05908, over 1612776.60 frames. ], batch size: 20, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:09:18,634 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4850, 2.0778, 2.2240, 2.0423, 1.4607, 2.0947, 2.3831, 2.1916], + device='cuda:3'), covar=tensor([0.0535, 0.0887, 0.1227, 0.1117, 0.0632, 0.1080, 0.0652, 0.0488], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0152, 0.0189, 0.0160, 0.0100, 0.0163, 0.0112, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 10:09:40,705 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5947, 2.0848, 3.2324, 1.4982, 2.4925, 2.0029, 1.7204, 2.4863], + device='cuda:3'), covar=tensor([0.1961, 0.2622, 0.0854, 0.4594, 0.1932, 0.3269, 0.2460, 0.2257], + device='cuda:3'), in_proj_covar=tensor([0.0534, 0.0622, 0.0555, 0.0657, 0.0654, 0.0602, 0.0553, 0.0635], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 10:09:43,188 INFO [train.py:901] (3/4) Epoch 26, batch 3300, loss[loss=0.1668, simple_loss=0.245, pruned_loss=0.04435, over 7198.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2836, pruned_loss=0.05896, over 1610367.17 frames. ], batch size: 16, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:10:10,314 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.669e+02 2.313e+02 2.653e+02 3.358e+02 9.214e+02, threshold=5.305e+02, percent-clipped=4.0 +2023-02-07 10:10:17,141 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 +2023-02-07 10:10:20,066 INFO [train.py:901] (3/4) Epoch 26, batch 3350, loss[loss=0.2812, simple_loss=0.349, pruned_loss=0.1067, over 8650.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2839, pruned_loss=0.05926, over 1606461.58 frames. ], batch size: 34, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:10:34,608 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3630, 1.7093, 2.6412, 1.2983, 1.9247, 1.8790, 1.4357, 1.9871], + device='cuda:3'), covar=tensor([0.2281, 0.2832, 0.1038, 0.5323, 0.2128, 0.3427, 0.2794, 0.2421], + device='cuda:3'), in_proj_covar=tensor([0.0538, 0.0627, 0.0559, 0.0663, 0.0659, 0.0606, 0.0558, 0.0641], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 10:10:54,105 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=205471.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:10:54,722 INFO [train.py:901] (3/4) Epoch 26, batch 3400, loss[loss=0.1881, simple_loss=0.2656, pruned_loss=0.05527, over 7412.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2842, pruned_loss=0.05953, over 1607466.57 frames. ], batch size: 17, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:11:20,313 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.723e+02 2.409e+02 2.883e+02 3.635e+02 7.106e+02, threshold=5.767e+02, percent-clipped=3.0 +2023-02-07 10:11:30,470 INFO [train.py:901] (3/4) Epoch 26, batch 3450, loss[loss=0.1765, simple_loss=0.2638, pruned_loss=0.04456, over 8499.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.283, pruned_loss=0.05901, over 1606453.28 frames. ], batch size: 28, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:11:53,164 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=205554.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:11:57,170 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=205560.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:12:05,236 INFO [train.py:901] (3/4) Epoch 26, batch 3500, loss[loss=0.1814, simple_loss=0.2691, pruned_loss=0.04684, over 7917.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2829, pruned_loss=0.05911, over 1608141.14 frames. ], batch size: 20, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:12:10,378 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=205579.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:12:15,150 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=205586.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:12:24,736 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-07 10:12:30,113 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.630e+02 2.208e+02 2.714e+02 3.358e+02 5.744e+02, threshold=5.428e+02, percent-clipped=0.0 +2023-02-07 10:12:35,814 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 +2023-02-07 10:12:39,511 INFO [train.py:901] (3/4) Epoch 26, batch 3550, loss[loss=0.2019, simple_loss=0.2821, pruned_loss=0.06084, over 8533.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2834, pruned_loss=0.05905, over 1610807.81 frames. ], batch size: 39, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:13:15,379 INFO [train.py:901] (3/4) Epoch 26, batch 3600, loss[loss=0.1777, simple_loss=0.265, pruned_loss=0.04525, over 8032.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2834, pruned_loss=0.05874, over 1612403.94 frames. ], batch size: 22, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:13:16,213 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=205673.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:13:17,486 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=205675.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:13:19,668 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.46 vs. limit=2.0 +2023-02-07 10:13:20,867 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7875, 1.4905, 3.1625, 1.4907, 2.3004, 3.4191, 3.5543, 2.8975], + device='cuda:3'), covar=tensor([0.1235, 0.1749, 0.0353, 0.2078, 0.0943, 0.0271, 0.0628, 0.0617], + device='cuda:3'), in_proj_covar=tensor([0.0304, 0.0328, 0.0291, 0.0320, 0.0322, 0.0277, 0.0437, 0.0308], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 10:13:39,661 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.491e+02 2.295e+02 2.882e+02 3.730e+02 8.207e+02, threshold=5.763e+02, percent-clipped=6.0 +2023-02-07 10:13:49,104 INFO [train.py:901] (3/4) Epoch 26, batch 3650, loss[loss=0.191, simple_loss=0.2826, pruned_loss=0.04967, over 8272.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2827, pruned_loss=0.05815, over 1612671.49 frames. ], batch size: 24, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:14:18,433 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=205762.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:14:24,803 INFO [train.py:901] (3/4) Epoch 26, batch 3700, loss[loss=0.2087, simple_loss=0.2977, pruned_loss=0.05984, over 8470.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2832, pruned_loss=0.05842, over 1614886.55 frames. ], batch size: 25, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:14:27,593 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-07 10:14:49,668 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.454e+02 2.377e+02 2.968e+02 3.727e+02 1.221e+03, threshold=5.937e+02, percent-clipped=5.0 +2023-02-07 10:14:59,197 INFO [train.py:901] (3/4) Epoch 26, batch 3750, loss[loss=0.1951, simple_loss=0.2785, pruned_loss=0.05585, over 7803.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.283, pruned_loss=0.05813, over 1612057.55 frames. ], batch size: 19, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:15:12,917 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=205842.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:15:31,310 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=205867.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:15:34,018 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1962, 1.0706, 1.2958, 1.0381, 0.9956, 1.3326, 0.0517, 0.8846], + device='cuda:3'), covar=tensor([0.1443, 0.1272, 0.0529, 0.0659, 0.2287, 0.0523, 0.1881, 0.1225], + device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0202, 0.0132, 0.0221, 0.0274, 0.0144, 0.0171, 0.0198], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 10:15:34,433 INFO [train.py:901] (3/4) Epoch 26, batch 3800, loss[loss=0.2272, simple_loss=0.3039, pruned_loss=0.0752, over 8094.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2825, pruned_loss=0.05793, over 1612340.34 frames. ], batch size: 21, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:15:59,218 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.674e+02 2.381e+02 2.847e+02 3.364e+02 6.986e+02, threshold=5.694e+02, percent-clipped=1.0 +2023-02-07 10:15:59,380 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=205908.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:16:08,813 INFO [train.py:901] (3/4) Epoch 26, batch 3850, loss[loss=0.2481, simple_loss=0.3187, pruned_loss=0.08878, over 6990.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2813, pruned_loss=0.05757, over 1605159.60 frames. ], batch size: 72, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:16:15,137 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=205931.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:16:20,592 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7045, 1.4683, 2.8164, 1.2651, 2.1774, 3.0153, 3.2478, 2.5609], + device='cuda:3'), covar=tensor([0.1305, 0.1783, 0.0403, 0.2411, 0.0957, 0.0343, 0.0624, 0.0657], + device='cuda:3'), in_proj_covar=tensor([0.0305, 0.0328, 0.0291, 0.0321, 0.0322, 0.0279, 0.0438, 0.0309], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 10:16:23,271 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5822, 2.1291, 3.1757, 1.4104, 2.3860, 1.9809, 1.7477, 2.4972], + device='cuda:3'), covar=tensor([0.2093, 0.2839, 0.1003, 0.4838, 0.2174, 0.3466, 0.2454, 0.2484], + device='cuda:3'), in_proj_covar=tensor([0.0537, 0.0626, 0.0558, 0.0662, 0.0657, 0.0605, 0.0557, 0.0641], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 10:16:29,172 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-07 10:16:32,102 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=205956.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:16:42,797 INFO [train.py:901] (3/4) Epoch 26, batch 3900, loss[loss=0.1909, simple_loss=0.2623, pruned_loss=0.05977, over 7420.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2817, pruned_loss=0.05764, over 1607438.50 frames. ], batch size: 17, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:17:09,995 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.608e+02 2.441e+02 2.892e+02 3.706e+02 7.796e+02, threshold=5.785e+02, percent-clipped=3.0 +2023-02-07 10:17:16,700 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=206017.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:17:18,467 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 +2023-02-07 10:17:20,038 INFO [train.py:901] (3/4) Epoch 26, batch 3950, loss[loss=0.2066, simple_loss=0.2978, pruned_loss=0.05764, over 8508.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2815, pruned_loss=0.05741, over 1608144.69 frames. ], batch size: 26, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:17:53,915 INFO [train.py:901] (3/4) Epoch 26, batch 4000, loss[loss=0.1699, simple_loss=0.2558, pruned_loss=0.04197, over 7541.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2821, pruned_loss=0.05788, over 1606292.74 frames. ], batch size: 18, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:17:55,452 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=206074.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:18:18,675 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=206106.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:18:19,947 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.865e+02 2.407e+02 2.986e+02 3.556e+02 8.558e+02, threshold=5.971e+02, percent-clipped=6.0 +2023-02-07 10:18:23,572 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8433, 1.6026, 3.4688, 1.6266, 2.4373, 3.7911, 3.9112, 3.2710], + device='cuda:3'), covar=tensor([0.1321, 0.1892, 0.0299, 0.2092, 0.1071, 0.0221, 0.0434, 0.0532], + device='cuda:3'), in_proj_covar=tensor([0.0303, 0.0326, 0.0291, 0.0320, 0.0321, 0.0278, 0.0437, 0.0308], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 10:18:26,939 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.7493, 1.4962, 4.9204, 1.9067, 4.4671, 4.1298, 4.4864, 4.3241], + device='cuda:3'), covar=tensor([0.0533, 0.4474, 0.0422, 0.3939, 0.0874, 0.0817, 0.0448, 0.0596], + device='cuda:3'), in_proj_covar=tensor([0.0659, 0.0655, 0.0721, 0.0644, 0.0734, 0.0620, 0.0621, 0.0696], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 10:18:29,517 INFO [train.py:901] (3/4) Epoch 26, batch 4050, loss[loss=0.2178, simple_loss=0.2935, pruned_loss=0.07101, over 8341.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.282, pruned_loss=0.0579, over 1605860.93 frames. ], batch size: 26, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:18:37,178 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206132.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:19:03,795 INFO [train.py:901] (3/4) Epoch 26, batch 4100, loss[loss=0.1465, simple_loss=0.2242, pruned_loss=0.0344, over 7430.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2826, pruned_loss=0.05815, over 1607606.75 frames. ], batch size: 17, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:19:28,869 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.590e+02 2.376e+02 2.755e+02 3.418e+02 9.873e+02, threshold=5.510e+02, percent-clipped=4.0 +2023-02-07 10:19:31,076 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0154, 1.4538, 1.7690, 1.3808, 1.0314, 1.5107, 1.8456, 1.5225], + device='cuda:3'), covar=tensor([0.0502, 0.1265, 0.1668, 0.1445, 0.0563, 0.1400, 0.0645, 0.0696], + device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0153, 0.0189, 0.0160, 0.0100, 0.0162, 0.0112, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 10:19:35,179 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=206215.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:19:39,253 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206221.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:19:39,709 INFO [train.py:901] (3/4) Epoch 26, batch 4150, loss[loss=0.1844, simple_loss=0.2654, pruned_loss=0.05169, over 7803.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2819, pruned_loss=0.05767, over 1605497.28 frames. ], batch size: 19, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:19:52,402 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.34 vs. limit=5.0 +2023-02-07 10:20:00,473 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-07 10:20:00,792 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=206252.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:20:10,914 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.9107, 3.8136, 3.5168, 2.4486, 3.4081, 3.5534, 3.4663, 3.4046], + device='cuda:3'), covar=tensor([0.0831, 0.0718, 0.1049, 0.3607, 0.0913, 0.1170, 0.1387, 0.0883], + device='cuda:3'), in_proj_covar=tensor([0.0538, 0.0455, 0.0441, 0.0553, 0.0436, 0.0459, 0.0436, 0.0400], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 10:20:14,202 INFO [train.py:901] (3/4) Epoch 26, batch 4200, loss[loss=0.2404, simple_loss=0.313, pruned_loss=0.08387, over 7155.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2833, pruned_loss=0.05851, over 1604348.62 frames. ], batch size: 73, lr: 2.90e-03, grad_scale: 16.0 +2023-02-07 10:20:22,970 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-07 10:20:38,388 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.475e+02 2.335e+02 2.968e+02 3.755e+02 9.805e+02, threshold=5.936e+02, percent-clipped=3.0 +2023-02-07 10:20:44,918 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-07 10:20:49,136 INFO [train.py:901] (3/4) Epoch 26, batch 4250, loss[loss=0.1875, simple_loss=0.274, pruned_loss=0.05053, over 7927.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.284, pruned_loss=0.05883, over 1607991.61 frames. ], batch size: 20, lr: 2.90e-03, grad_scale: 8.0 +2023-02-07 10:21:03,041 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0792, 1.2298, 1.1906, 0.8119, 1.2339, 1.0341, 0.0733, 1.2030], + device='cuda:3'), covar=tensor([0.0447, 0.0420, 0.0394, 0.0603, 0.0492, 0.1106, 0.0964, 0.0368], + device='cuda:3'), in_proj_covar=tensor([0.0472, 0.0409, 0.0364, 0.0461, 0.0395, 0.0555, 0.0406, 0.0440], + device='cuda:3'), out_proj_covar=tensor([1.2541e-04, 1.0671e-04, 9.5195e-05, 1.2081e-04, 1.0329e-04, 1.5543e-04, + 1.0858e-04, 1.1577e-04], device='cuda:3') +2023-02-07 10:21:21,331 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206367.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:21:24,640 INFO [train.py:901] (3/4) Epoch 26, batch 4300, loss[loss=0.1766, simple_loss=0.2674, pruned_loss=0.04293, over 8285.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2828, pruned_loss=0.05801, over 1607951.56 frames. ], batch size: 23, lr: 2.90e-03, grad_scale: 8.0 +2023-02-07 10:21:26,011 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.1860, 3.1359, 2.8919, 1.6363, 2.7418, 2.9201, 2.8417, 2.8132], + device='cuda:3'), covar=tensor([0.1197, 0.0850, 0.1366, 0.4868, 0.1352, 0.1627, 0.1587, 0.1106], + device='cuda:3'), in_proj_covar=tensor([0.0539, 0.0457, 0.0443, 0.0556, 0.0440, 0.0462, 0.0439, 0.0402], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 10:21:26,056 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2403, 1.2340, 3.3552, 1.1525, 2.9509, 2.8132, 3.0727, 2.9669], + device='cuda:3'), covar=tensor([0.0860, 0.4641, 0.0856, 0.4275, 0.1362, 0.1104, 0.0808, 0.0925], + device='cuda:3'), in_proj_covar=tensor([0.0661, 0.0655, 0.0722, 0.0646, 0.0735, 0.0622, 0.0623, 0.0696], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 10:21:35,889 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206388.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:21:50,295 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.467e+02 2.331e+02 2.890e+02 3.800e+02 6.492e+02, threshold=5.781e+02, percent-clipped=2.0 +2023-02-07 10:21:53,279 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206413.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:21:56,607 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=206418.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:21:59,346 INFO [train.py:901] (3/4) Epoch 26, batch 4350, loss[loss=0.1715, simple_loss=0.2535, pruned_loss=0.04473, over 7918.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2825, pruned_loss=0.05789, over 1607868.34 frames. ], batch size: 20, lr: 2.90e-03, grad_scale: 8.0 +2023-02-07 10:22:18,983 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-07 10:22:20,066 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.86 vs. limit=5.0 +2023-02-07 10:22:20,824 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.91 vs. limit=5.0 +2023-02-07 10:22:32,200 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.2861, 3.2084, 2.9825, 1.6508, 2.8906, 2.8746, 2.8587, 2.7472], + device='cuda:3'), covar=tensor([0.1067, 0.0799, 0.1224, 0.4465, 0.1163, 0.1258, 0.1669, 0.1131], + device='cuda:3'), in_proj_covar=tensor([0.0536, 0.0454, 0.0439, 0.0552, 0.0435, 0.0457, 0.0434, 0.0399], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 10:22:34,794 INFO [train.py:901] (3/4) Epoch 26, batch 4400, loss[loss=0.177, simple_loss=0.2539, pruned_loss=0.0501, over 8084.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2828, pruned_loss=0.05829, over 1608619.59 frames. ], batch size: 21, lr: 2.90e-03, grad_scale: 8.0 +2023-02-07 10:22:36,358 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3580, 1.6302, 1.6105, 1.0822, 1.6853, 1.3179, 0.3114, 1.6042], + device='cuda:3'), covar=tensor([0.0540, 0.0403, 0.0348, 0.0576, 0.0447, 0.0956, 0.0989, 0.0306], + device='cuda:3'), in_proj_covar=tensor([0.0468, 0.0405, 0.0361, 0.0457, 0.0392, 0.0550, 0.0401, 0.0436], + device='cuda:3'), out_proj_covar=tensor([1.2424e-04, 1.0562e-04, 9.4419e-05, 1.1984e-04, 1.0259e-04, 1.5377e-04, + 1.0723e-04, 1.1472e-04], device='cuda:3') +2023-02-07 10:22:38,337 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206477.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:22:55,679 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206502.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:23:00,162 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.634e+02 2.619e+02 3.000e+02 3.925e+02 8.429e+02, threshold=6.000e+02, percent-clipped=7.0 +2023-02-07 10:23:00,191 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-07 10:23:04,479 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.54 vs. limit=2.0 +2023-02-07 10:23:08,814 INFO [train.py:901] (3/4) Epoch 26, batch 4450, loss[loss=0.1832, simple_loss=0.2779, pruned_loss=0.04428, over 8349.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2833, pruned_loss=0.05817, over 1613244.77 frames. ], batch size: 26, lr: 2.90e-03, grad_scale: 8.0 +2023-02-07 10:23:16,260 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206533.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:23:34,156 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=206559.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:23:44,259 INFO [train.py:901] (3/4) Epoch 26, batch 4500, loss[loss=0.176, simple_loss=0.2656, pruned_loss=0.04318, over 8101.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2829, pruned_loss=0.05784, over 1616447.61 frames. ], batch size: 23, lr: 2.90e-03, grad_scale: 8.0 +2023-02-07 10:23:55,209 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-07 10:24:07,630 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8055, 1.4304, 3.9757, 1.3893, 3.5111, 3.3092, 3.5834, 3.4618], + device='cuda:3'), covar=tensor([0.0758, 0.4750, 0.0691, 0.4676, 0.1269, 0.1070, 0.0760, 0.0805], + device='cuda:3'), in_proj_covar=tensor([0.0662, 0.0655, 0.0723, 0.0648, 0.0735, 0.0623, 0.0623, 0.0697], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 10:24:10,074 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.435e+02 2.363e+02 2.961e+02 3.499e+02 6.135e+02, threshold=5.921e+02, percent-clipped=1.0 +2023-02-07 10:24:18,689 INFO [train.py:901] (3/4) Epoch 26, batch 4550, loss[loss=0.2009, simple_loss=0.2907, pruned_loss=0.05552, over 8502.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2825, pruned_loss=0.05784, over 1613517.63 frames. ], batch size: 26, lr: 2.90e-03, grad_scale: 8.0 +2023-02-07 10:24:19,502 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206623.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:24:35,787 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206648.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:24:51,534 INFO [train.py:901] (3/4) Epoch 26, batch 4600, loss[loss=0.1723, simple_loss=0.2604, pruned_loss=0.04208, over 8248.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2832, pruned_loss=0.05817, over 1611670.38 frames. ], batch size: 24, lr: 2.90e-03, grad_scale: 8.0 +2023-02-07 10:24:53,054 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206674.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:25:01,031 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1585, 4.0759, 3.7570, 1.8330, 3.6205, 3.7582, 3.7882, 3.5729], + device='cuda:3'), covar=tensor([0.0775, 0.0560, 0.1038, 0.4874, 0.0941, 0.1000, 0.1258, 0.0801], + device='cuda:3'), in_proj_covar=tensor([0.0537, 0.0452, 0.0438, 0.0550, 0.0434, 0.0456, 0.0434, 0.0399], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 10:25:18,502 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.466e+02 2.342e+02 2.811e+02 3.625e+02 9.770e+02, threshold=5.622e+02, percent-clipped=5.0 +2023-02-07 10:25:21,726 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 +2023-02-07 10:25:23,276 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 +2023-02-07 10:25:28,319 INFO [train.py:901] (3/4) Epoch 26, batch 4650, loss[loss=0.2478, simple_loss=0.3336, pruned_loss=0.08095, over 8421.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2832, pruned_loss=0.05805, over 1615313.35 frames. ], batch size: 49, lr: 2.90e-03, grad_scale: 8.0 +2023-02-07 10:25:40,975 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 +2023-02-07 10:26:02,069 INFO [train.py:901] (3/4) Epoch 26, batch 4700, loss[loss=0.1821, simple_loss=0.2732, pruned_loss=0.04547, over 8089.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2832, pruned_loss=0.05804, over 1611766.58 frames. ], batch size: 21, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:26:08,426 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3997, 1.6889, 1.6604, 1.1293, 1.6784, 1.4004, 0.2480, 1.6040], + device='cuda:3'), covar=tensor([0.0656, 0.0426, 0.0386, 0.0641, 0.0556, 0.1110, 0.1017, 0.0368], + device='cuda:3'), in_proj_covar=tensor([0.0466, 0.0402, 0.0360, 0.0455, 0.0389, 0.0547, 0.0399, 0.0434], + device='cuda:3'), out_proj_covar=tensor([1.2370e-04, 1.0477e-04, 9.4011e-05, 1.1921e-04, 1.0177e-04, 1.5321e-04, + 1.0686e-04, 1.1408e-04], device='cuda:3') +2023-02-07 10:26:13,620 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206789.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:26:28,921 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.591e+02 2.506e+02 2.890e+02 3.298e+02 6.611e+02, threshold=5.779e+02, percent-clipped=3.0 +2023-02-07 10:26:32,508 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206814.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:26:34,594 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2322, 2.0497, 2.6171, 2.2415, 2.5609, 2.3093, 2.0790, 1.4125], + device='cuda:3'), covar=tensor([0.5512, 0.4864, 0.2120, 0.3628, 0.2464, 0.3003, 0.1955, 0.5105], + device='cuda:3'), in_proj_covar=tensor([0.0952, 0.1003, 0.0824, 0.0979, 0.1014, 0.0917, 0.0764, 0.0838], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 10:26:37,699 INFO [train.py:901] (3/4) Epoch 26, batch 4750, loss[loss=0.2001, simple_loss=0.2849, pruned_loss=0.05766, over 8243.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.283, pruned_loss=0.05813, over 1614472.17 frames. ], batch size: 24, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:26:53,345 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-07 10:26:55,380 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-07 10:26:58,852 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=206852.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:27:00,991 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 +2023-02-07 10:27:12,097 INFO [train.py:901] (3/4) Epoch 26, batch 4800, loss[loss=0.1783, simple_loss=0.2695, pruned_loss=0.04355, over 8295.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2831, pruned_loss=0.05788, over 1617812.63 frames. ], batch size: 23, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:27:21,155 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-07 10:27:37,299 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.341e+02 2.410e+02 2.886e+02 3.541e+02 7.542e+02, threshold=5.772e+02, percent-clipped=6.0 +2023-02-07 10:27:46,734 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-07 10:27:47,378 INFO [train.py:901] (3/4) Epoch 26, batch 4850, loss[loss=0.2223, simple_loss=0.2979, pruned_loss=0.07333, over 6864.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2827, pruned_loss=0.05773, over 1615536.02 frames. ], batch size: 71, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:27:52,971 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206930.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:28:10,406 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206955.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:28:21,662 INFO [train.py:901] (3/4) Epoch 26, batch 4900, loss[loss=0.1457, simple_loss=0.2255, pruned_loss=0.03296, over 7446.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2821, pruned_loss=0.05807, over 1610544.66 frames. ], batch size: 17, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:28:25,971 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.31 vs. limit=5.0 +2023-02-07 10:28:46,039 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.596e+02 2.535e+02 3.142e+02 3.836e+02 8.051e+02, threshold=6.285e+02, percent-clipped=2.0 +2023-02-07 10:28:55,275 INFO [train.py:901] (3/4) Epoch 26, batch 4950, loss[loss=0.219, simple_loss=0.3079, pruned_loss=0.06507, over 8585.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2831, pruned_loss=0.05858, over 1611036.40 frames. ], batch size: 31, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:29:04,281 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9658, 1.5412, 1.7296, 1.3719, 1.1019, 1.5243, 1.8306, 1.4391], + device='cuda:3'), covar=tensor([0.0524, 0.1249, 0.1619, 0.1454, 0.0578, 0.1429, 0.0654, 0.0650], + device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0152, 0.0189, 0.0160, 0.0101, 0.0162, 0.0112, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 10:29:32,338 INFO [train.py:901] (3/4) Epoch 26, batch 5000, loss[loss=0.1562, simple_loss=0.2318, pruned_loss=0.04028, over 7558.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2824, pruned_loss=0.05837, over 1606758.85 frames. ], batch size: 18, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:29:35,482 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.51 vs. limit=2.0 +2023-02-07 10:29:57,380 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.625e+02 2.413e+02 2.985e+02 3.933e+02 1.062e+03, threshold=5.970e+02, percent-clipped=3.0 +2023-02-07 10:30:06,447 INFO [train.py:901] (3/4) Epoch 26, batch 5050, loss[loss=0.2019, simple_loss=0.2741, pruned_loss=0.06486, over 7921.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2828, pruned_loss=0.05814, over 1613572.73 frames. ], batch size: 20, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:30:13,973 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5961, 2.5471, 1.8323, 2.3127, 2.2084, 1.5878, 2.1073, 2.2548], + device='cuda:3'), covar=tensor([0.1522, 0.0434, 0.1242, 0.0667, 0.0781, 0.1572, 0.1059, 0.0901], + device='cuda:3'), in_proj_covar=tensor([0.0361, 0.0240, 0.0341, 0.0313, 0.0302, 0.0348, 0.0349, 0.0325], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 10:30:24,761 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-07 10:30:42,639 INFO [train.py:901] (3/4) Epoch 26, batch 5100, loss[loss=0.2538, simple_loss=0.3171, pruned_loss=0.09521, over 6972.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2836, pruned_loss=0.05819, over 1617499.84 frames. ], batch size: 71, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:30:58,171 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=207194.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:30:59,446 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=207196.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:31:05,086 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0394, 1.5530, 3.5985, 1.4769, 2.5073, 3.9715, 4.0467, 3.4175], + device='cuda:3'), covar=tensor([0.1138, 0.1825, 0.0329, 0.2093, 0.1010, 0.0213, 0.0538, 0.0512], + device='cuda:3'), in_proj_covar=tensor([0.0300, 0.0324, 0.0287, 0.0316, 0.0317, 0.0274, 0.0432, 0.0303], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 10:31:08,239 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.490e+02 2.087e+02 2.633e+02 3.622e+02 6.552e+02, threshold=5.265e+02, percent-clipped=1.0 +2023-02-07 10:31:16,944 INFO [train.py:901] (3/4) Epoch 26, batch 5150, loss[loss=0.1744, simple_loss=0.2495, pruned_loss=0.04966, over 7215.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2834, pruned_loss=0.05807, over 1618434.66 frames. ], batch size: 16, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:31:52,794 INFO [train.py:901] (3/4) Epoch 26, batch 5200, loss[loss=0.195, simple_loss=0.276, pruned_loss=0.057, over 8290.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2833, pruned_loss=0.05827, over 1615485.12 frames. ], batch size: 23, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:32:10,485 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-07 10:32:17,988 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.620e+02 2.585e+02 3.464e+02 4.468e+02 1.375e+03, threshold=6.928e+02, percent-clipped=16.0 +2023-02-07 10:32:19,441 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=207311.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:32:19,977 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-07 10:32:26,691 INFO [train.py:901] (3/4) Epoch 26, batch 5250, loss[loss=0.2141, simple_loss=0.3137, pruned_loss=0.05728, over 8522.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2839, pruned_loss=0.05869, over 1614165.30 frames. ], batch size: 26, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:32:52,569 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9962, 2.0334, 1.7903, 2.2936, 1.6121, 1.7455, 2.0285, 2.0891], + device='cuda:3'), covar=tensor([0.0628, 0.0710, 0.0764, 0.0645, 0.0946, 0.1065, 0.0591, 0.0645], + device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0193, 0.0244, 0.0212, 0.0202, 0.0245, 0.0248, 0.0204], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 10:33:00,341 INFO [train.py:901] (3/4) Epoch 26, batch 5300, loss[loss=0.1927, simple_loss=0.2703, pruned_loss=0.05757, over 7972.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2846, pruned_loss=0.05982, over 1604382.57 frames. ], batch size: 21, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:33:27,787 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.886e+02 2.392e+02 2.913e+02 3.782e+02 6.658e+02, threshold=5.826e+02, percent-clipped=0.0 +2023-02-07 10:33:36,853 INFO [train.py:901] (3/4) Epoch 26, batch 5350, loss[loss=0.1697, simple_loss=0.2624, pruned_loss=0.0385, over 7923.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2832, pruned_loss=0.05878, over 1606867.99 frames. ], batch size: 20, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:34:10,288 INFO [train.py:901] (3/4) Epoch 26, batch 5400, loss[loss=0.1991, simple_loss=0.2928, pruned_loss=0.05269, over 8034.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2838, pruned_loss=0.05875, over 1610054.56 frames. ], batch size: 22, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:34:37,323 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.635e+02 2.344e+02 3.061e+02 4.157e+02 9.885e+02, threshold=6.122e+02, percent-clipped=8.0 +2023-02-07 10:34:46,158 INFO [train.py:901] (3/4) Epoch 26, batch 5450, loss[loss=0.1896, simple_loss=0.2745, pruned_loss=0.05232, over 8092.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2848, pruned_loss=0.05932, over 1608752.02 frames. ], batch size: 21, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:34:57,844 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=207538.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:35:06,615 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-07 10:35:17,363 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=207567.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:35:20,581 INFO [train.py:901] (3/4) Epoch 26, batch 5500, loss[loss=0.2175, simple_loss=0.3018, pruned_loss=0.06657, over 8253.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2852, pruned_loss=0.05939, over 1614767.68 frames. ], batch size: 24, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:35:34,400 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=207592.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:35:47,181 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.641e+02 2.395e+02 2.975e+02 3.460e+02 7.775e+02, threshold=5.949e+02, percent-clipped=2.0 +2023-02-07 10:35:56,712 INFO [train.py:901] (3/4) Epoch 26, batch 5550, loss[loss=0.1747, simple_loss=0.256, pruned_loss=0.04665, over 7427.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.284, pruned_loss=0.05843, over 1615383.43 frames. ], batch size: 17, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:36:17,892 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=207653.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:36:30,415 INFO [train.py:901] (3/4) Epoch 26, batch 5600, loss[loss=0.177, simple_loss=0.2609, pruned_loss=0.04653, over 7656.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2842, pruned_loss=0.05856, over 1612670.13 frames. ], batch size: 19, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:36:55,073 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.643e+02 2.412e+02 3.079e+02 3.750e+02 8.490e+02, threshold=6.158e+02, percent-clipped=5.0 +2023-02-07 10:37:04,607 INFO [train.py:901] (3/4) Epoch 26, batch 5650, loss[loss=0.1889, simple_loss=0.281, pruned_loss=0.04844, over 8018.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2847, pruned_loss=0.05864, over 1611575.09 frames. ], batch size: 22, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:37:12,976 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-07 10:37:40,854 INFO [train.py:901] (3/4) Epoch 26, batch 5700, loss[loss=0.1655, simple_loss=0.2478, pruned_loss=0.04162, over 8136.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2834, pruned_loss=0.05766, over 1612074.04 frames. ], batch size: 22, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:38:05,994 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.232e+02 2.912e+02 3.330e+02 6.698e+02, threshold=5.824e+02, percent-clipped=1.0 +2023-02-07 10:38:14,800 INFO [train.py:901] (3/4) Epoch 26, batch 5750, loss[loss=0.1913, simple_loss=0.2739, pruned_loss=0.05429, over 7813.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.283, pruned_loss=0.05741, over 1609687.99 frames. ], batch size: 20, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:38:16,880 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-07 10:38:50,532 INFO [train.py:901] (3/4) Epoch 26, batch 5800, loss[loss=0.191, simple_loss=0.2834, pruned_loss=0.0493, over 8521.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.283, pruned_loss=0.05778, over 1611191.07 frames. ], batch size: 28, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:39:15,995 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.602e+02 2.614e+02 3.148e+02 4.020e+02 8.026e+02, threshold=6.297e+02, percent-clipped=4.0 +2023-02-07 10:39:16,244 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=207909.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:39:24,804 INFO [train.py:901] (3/4) Epoch 26, batch 5850, loss[loss=0.1993, simple_loss=0.2761, pruned_loss=0.0613, over 8240.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2816, pruned_loss=0.05687, over 1613315.53 frames. ], batch size: 22, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:39:32,992 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=207934.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:39:59,800 INFO [train.py:901] (3/4) Epoch 26, batch 5900, loss[loss=0.1855, simple_loss=0.2709, pruned_loss=0.05002, over 8135.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2828, pruned_loss=0.05749, over 1611637.15 frames. ], batch size: 22, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:40:15,814 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1025, 1.0132, 1.6575, 0.9797, 1.6203, 1.8166, 1.8783, 1.5642], + device='cuda:3'), covar=tensor([0.0939, 0.1264, 0.0614, 0.1644, 0.1294, 0.0354, 0.0696, 0.0549], + device='cuda:3'), in_proj_covar=tensor([0.0300, 0.0324, 0.0290, 0.0316, 0.0317, 0.0275, 0.0435, 0.0305], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 10:40:26,851 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.528e+02 2.451e+02 2.961e+02 3.656e+02 5.483e+02, threshold=5.923e+02, percent-clipped=0.0 +2023-02-07 10:40:35,634 INFO [train.py:901] (3/4) Epoch 26, batch 5950, loss[loss=0.2277, simple_loss=0.2935, pruned_loss=0.08098, over 8028.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2838, pruned_loss=0.05836, over 1612252.51 frames. ], batch size: 22, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:40:54,934 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=208050.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:41:09,815 INFO [train.py:901] (3/4) Epoch 26, batch 6000, loss[loss=0.2082, simple_loss=0.3003, pruned_loss=0.05799, over 8462.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2851, pruned_loss=0.05895, over 1617394.51 frames. ], batch size: 27, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:41:09,816 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 10:41:24,447 INFO [train.py:935] (3/4) Epoch 26, validation: loss=0.1721, simple_loss=0.2717, pruned_loss=0.03627, over 944034.00 frames. +2023-02-07 10:41:24,447 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6747MB +2023-02-07 10:41:32,217 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.8224, 2.1165, 3.6276, 1.9873, 1.7424, 3.5156, 0.7053, 2.2308], + device='cuda:3'), covar=tensor([0.1275, 0.1339, 0.0194, 0.1445, 0.2461, 0.0356, 0.1987, 0.1200], + device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0203, 0.0133, 0.0222, 0.0276, 0.0144, 0.0170, 0.0197], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 10:41:51,028 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.622e+02 2.308e+02 2.837e+02 3.630e+02 6.769e+02, threshold=5.675e+02, percent-clipped=2.0 +2023-02-07 10:42:00,869 INFO [train.py:901] (3/4) Epoch 26, batch 6050, loss[loss=0.1902, simple_loss=0.2751, pruned_loss=0.05266, over 8463.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.284, pruned_loss=0.05827, over 1616226.56 frames. ], batch size: 25, lr: 2.89e-03, grad_scale: 8.0 +2023-02-07 10:42:25,848 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5735, 4.6845, 4.1541, 2.0924, 3.9861, 4.2052, 4.1380, 3.9446], + device='cuda:3'), covar=tensor([0.0698, 0.0468, 0.0919, 0.4684, 0.0936, 0.0999, 0.1172, 0.0752], + device='cuda:3'), in_proj_covar=tensor([0.0529, 0.0446, 0.0435, 0.0541, 0.0427, 0.0451, 0.0427, 0.0395], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 10:42:36,037 INFO [train.py:901] (3/4) Epoch 26, batch 6100, loss[loss=0.1986, simple_loss=0.2856, pruned_loss=0.05582, over 8367.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2842, pruned_loss=0.0584, over 1615330.01 frames. ], batch size: 24, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:42:48,574 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-07 10:43:01,339 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.708e+02 2.422e+02 2.947e+02 3.994e+02 1.088e+03, threshold=5.894e+02, percent-clipped=8.0 +2023-02-07 10:43:06,118 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.7827, 5.9567, 5.1737, 2.4583, 5.1495, 5.5582, 5.5100, 5.3864], + device='cuda:3'), covar=tensor([0.0549, 0.0363, 0.0849, 0.4397, 0.0765, 0.0837, 0.0925, 0.0601], + device='cuda:3'), in_proj_covar=tensor([0.0532, 0.0450, 0.0438, 0.0545, 0.0429, 0.0454, 0.0429, 0.0397], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 10:43:10,773 INFO [train.py:901] (3/4) Epoch 26, batch 6150, loss[loss=0.2071, simple_loss=0.2929, pruned_loss=0.06061, over 8293.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2842, pruned_loss=0.05849, over 1610855.02 frames. ], batch size: 23, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:43:45,582 INFO [train.py:901] (3/4) Epoch 26, batch 6200, loss[loss=0.1993, simple_loss=0.296, pruned_loss=0.05129, over 8291.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2836, pruned_loss=0.05817, over 1610964.84 frames. ], batch size: 23, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:43:53,059 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=208283.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:44:10,220 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.423e+02 2.196e+02 2.837e+02 3.308e+02 7.178e+02, threshold=5.674e+02, percent-clipped=2.0 +2023-02-07 10:44:19,005 INFO [train.py:901] (3/4) Epoch 26, batch 6250, loss[loss=0.1684, simple_loss=0.2515, pruned_loss=0.04268, over 7786.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2822, pruned_loss=0.05769, over 1610298.14 frames. ], batch size: 19, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:44:55,642 INFO [train.py:901] (3/4) Epoch 26, batch 6300, loss[loss=0.2082, simple_loss=0.2904, pruned_loss=0.06296, over 8491.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2812, pruned_loss=0.0573, over 1607402.89 frames. ], batch size: 26, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:45:10,475 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=208394.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:45:20,546 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.628e+02 2.443e+02 2.919e+02 3.618e+02 1.192e+03, threshold=5.838e+02, percent-clipped=3.0 +2023-02-07 10:45:29,140 INFO [train.py:901] (3/4) Epoch 26, batch 6350, loss[loss=0.1941, simple_loss=0.2904, pruned_loss=0.04892, over 8531.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2819, pruned_loss=0.05735, over 1611504.66 frames. ], batch size: 28, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:46:04,931 INFO [train.py:901] (3/4) Epoch 26, batch 6400, loss[loss=0.1878, simple_loss=0.2685, pruned_loss=0.05353, over 8085.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2816, pruned_loss=0.05702, over 1611503.34 frames. ], batch size: 21, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:46:30,402 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1924, 2.0307, 2.5969, 2.2068, 2.5546, 2.2975, 2.1081, 1.4935], + device='cuda:3'), covar=tensor([0.5774, 0.4940, 0.2161, 0.3864, 0.2755, 0.3165, 0.1941, 0.5423], + device='cuda:3'), in_proj_covar=tensor([0.0962, 0.1013, 0.0831, 0.0985, 0.1022, 0.0923, 0.0770, 0.0847], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 10:46:30,800 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.490e+02 2.580e+02 3.188e+02 3.813e+02 6.849e+02, threshold=6.376e+02, percent-clipped=3.0 +2023-02-07 10:46:31,022 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=208509.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:46:39,653 INFO [train.py:901] (3/4) Epoch 26, batch 6450, loss[loss=0.226, simple_loss=0.3071, pruned_loss=0.0725, over 6989.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2823, pruned_loss=0.0576, over 1609397.28 frames. ], batch size: 71, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:47:13,749 INFO [train.py:901] (3/4) Epoch 26, batch 6500, loss[loss=0.2332, simple_loss=0.3186, pruned_loss=0.07391, over 8608.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2826, pruned_loss=0.05774, over 1607958.39 frames. ], batch size: 31, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:47:29,483 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=208594.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:47:39,234 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.676e+02 2.371e+02 2.869e+02 3.528e+02 8.936e+02, threshold=5.738e+02, percent-clipped=3.0 +2023-02-07 10:47:44,315 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3702, 2.1046, 2.7205, 2.1995, 2.7364, 2.3954, 2.2323, 1.5409], + device='cuda:3'), covar=tensor([0.5627, 0.5089, 0.2253, 0.4340, 0.2738, 0.3176, 0.2003, 0.5788], + device='cuda:3'), in_proj_covar=tensor([0.0968, 0.1019, 0.0835, 0.0991, 0.1025, 0.0928, 0.0773, 0.0852], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 10:47:48,765 INFO [train.py:901] (3/4) Epoch 26, batch 6550, loss[loss=0.2108, simple_loss=0.3029, pruned_loss=0.05928, over 8593.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2831, pruned_loss=0.05805, over 1614932.43 frames. ], batch size: 31, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:47:52,200 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=208627.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:47:56,170 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-07 10:48:12,891 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-07 10:48:22,502 INFO [train.py:901] (3/4) Epoch 26, batch 6600, loss[loss=0.1663, simple_loss=0.2433, pruned_loss=0.04463, over 7916.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2823, pruned_loss=0.0575, over 1616405.31 frames. ], batch size: 20, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:48:49,206 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.703e+02 2.581e+02 2.930e+02 3.571e+02 6.165e+02, threshold=5.859e+02, percent-clipped=2.0 +2023-02-07 10:48:58,720 INFO [train.py:901] (3/4) Epoch 26, batch 6650, loss[loss=0.1968, simple_loss=0.2884, pruned_loss=0.05256, over 8470.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2823, pruned_loss=0.05766, over 1613663.84 frames. ], batch size: 25, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:49:12,762 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=208742.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:49:28,863 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=208765.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:49:29,602 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3274, 2.0315, 2.6515, 2.1645, 2.7104, 2.3626, 2.2431, 1.5501], + device='cuda:3'), covar=tensor([0.6104, 0.5353, 0.2108, 0.4395, 0.2777, 0.3346, 0.2155, 0.5636], + device='cuda:3'), in_proj_covar=tensor([0.0967, 0.1017, 0.0834, 0.0989, 0.1024, 0.0927, 0.0771, 0.0849], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 10:49:33,535 INFO [train.py:901] (3/4) Epoch 26, batch 6700, loss[loss=0.2212, simple_loss=0.3084, pruned_loss=0.06698, over 8578.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2823, pruned_loss=0.0581, over 1609762.05 frames. ], batch size: 31, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:49:46,230 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=208790.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:49:59,642 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.693e+02 2.530e+02 3.053e+02 4.076e+02 9.744e+02, threshold=6.106e+02, percent-clipped=7.0 +2023-02-07 10:50:03,319 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5593, 2.7392, 2.3183, 4.0746, 1.5890, 2.0815, 2.3919, 2.7491], + device='cuda:3'), covar=tensor([0.0734, 0.0809, 0.0867, 0.0239, 0.1129, 0.1266, 0.0983, 0.0854], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0195, 0.0245, 0.0212, 0.0203, 0.0246, 0.0250, 0.0206], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 10:50:09,367 INFO [train.py:901] (3/4) Epoch 26, batch 6750, loss[loss=0.1827, simple_loss=0.2726, pruned_loss=0.04638, over 8202.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2823, pruned_loss=0.05793, over 1610184.41 frames. ], batch size: 23, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:50:30,942 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-07 10:50:43,658 INFO [train.py:901] (3/4) Epoch 26, batch 6800, loss[loss=0.1769, simple_loss=0.2671, pruned_loss=0.04335, over 8129.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.283, pruned_loss=0.0586, over 1611211.09 frames. ], batch size: 22, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:50:57,051 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-07 10:51:08,812 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.587e+02 2.376e+02 2.847e+02 3.449e+02 1.016e+03, threshold=5.694e+02, percent-clipped=2.0 +2023-02-07 10:51:18,808 INFO [train.py:901] (3/4) Epoch 26, batch 6850, loss[loss=0.1577, simple_loss=0.2301, pruned_loss=0.04267, over 7441.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2833, pruned_loss=0.05867, over 1613178.29 frames. ], batch size: 17, lr: 2.88e-03, grad_scale: 16.0 +2023-02-07 10:51:19,502 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-07 10:51:30,532 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=208938.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:51:33,406 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8943, 2.0770, 1.6912, 2.6685, 1.3175, 1.5487, 2.0720, 2.1566], + device='cuda:3'), covar=tensor([0.0778, 0.0847, 0.0915, 0.0371, 0.1087, 0.1360, 0.0767, 0.0737], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0194, 0.0244, 0.0212, 0.0202, 0.0245, 0.0249, 0.0205], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 10:51:37,688 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4684, 2.2740, 2.9733, 2.4319, 3.0269, 2.4590, 2.3175, 1.9407], + device='cuda:3'), covar=tensor([0.5517, 0.5193, 0.2081, 0.4106, 0.2897, 0.3435, 0.2034, 0.5665], + device='cuda:3'), in_proj_covar=tensor([0.0961, 0.1014, 0.0830, 0.0986, 0.1020, 0.0924, 0.0769, 0.0847], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 10:51:42,436 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8787, 1.6703, 1.9666, 1.7582, 1.9098, 1.9430, 1.7716, 0.8648], + device='cuda:3'), covar=tensor([0.6098, 0.4990, 0.2199, 0.3442, 0.2602, 0.3065, 0.2153, 0.5183], + device='cuda:3'), in_proj_covar=tensor([0.0961, 0.1013, 0.0829, 0.0986, 0.1019, 0.0924, 0.0769, 0.0846], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 10:51:45,311 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-07 10:51:54,560 INFO [train.py:901] (3/4) Epoch 26, batch 6900, loss[loss=0.1819, simple_loss=0.2586, pruned_loss=0.05266, over 7816.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2827, pruned_loss=0.05836, over 1606368.54 frames. ], batch size: 20, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:52:10,629 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0042, 3.5404, 2.1098, 2.7385, 2.6528, 2.0948, 2.6362, 2.9078], + device='cuda:3'), covar=tensor([0.1755, 0.0421, 0.1266, 0.0790, 0.0737, 0.1437, 0.1125, 0.1027], + device='cuda:3'), in_proj_covar=tensor([0.0359, 0.0239, 0.0339, 0.0310, 0.0302, 0.0345, 0.0347, 0.0322], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 10:52:12,016 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=208998.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:52:12,732 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9817, 1.7198, 2.0228, 1.8437, 1.9855, 2.0192, 1.9096, 0.8316], + device='cuda:3'), covar=tensor([0.5764, 0.4978, 0.2181, 0.3629, 0.2581, 0.3069, 0.1929, 0.5258], + device='cuda:3'), in_proj_covar=tensor([0.0961, 0.1013, 0.0829, 0.0986, 0.1018, 0.0923, 0.0769, 0.0846], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 10:52:20,218 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.776e+02 2.430e+02 2.933e+02 3.890e+02 9.541e+02, threshold=5.866e+02, percent-clipped=7.0 +2023-02-07 10:52:23,042 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-07 10:52:23,724 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5847, 4.5698, 4.2065, 2.3173, 4.0463, 4.1599, 4.1399, 3.9968], + device='cuda:3'), covar=tensor([0.0651, 0.0479, 0.0879, 0.3991, 0.0861, 0.0995, 0.1229, 0.0792], + device='cuda:3'), in_proj_covar=tensor([0.0537, 0.0456, 0.0442, 0.0552, 0.0436, 0.0460, 0.0436, 0.0402], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 10:52:28,984 INFO [train.py:901] (3/4) Epoch 26, batch 6950, loss[loss=0.2103, simple_loss=0.3007, pruned_loss=0.05999, over 8300.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2831, pruned_loss=0.05872, over 1606264.27 frames. ], batch size: 23, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:52:29,857 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=209023.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:52:51,125 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209053.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:52:51,888 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9073, 1.8787, 3.2125, 2.3961, 2.8499, 2.0089, 1.6951, 1.5427], + device='cuda:3'), covar=tensor([0.8159, 0.6996, 0.2301, 0.4347, 0.3360, 0.4791, 0.3151, 0.6285], + device='cuda:3'), in_proj_covar=tensor([0.0961, 0.1013, 0.0829, 0.0986, 0.1017, 0.0922, 0.0770, 0.0847], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 10:53:04,211 INFO [train.py:901] (3/4) Epoch 26, batch 7000, loss[loss=0.2254, simple_loss=0.2982, pruned_loss=0.0763, over 8337.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2822, pruned_loss=0.05844, over 1605814.57 frames. ], batch size: 25, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:53:30,453 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.766e+02 2.543e+02 3.075e+02 4.223e+02 1.225e+03, threshold=6.150e+02, percent-clipped=4.0 +2023-02-07 10:53:38,406 INFO [train.py:901] (3/4) Epoch 26, batch 7050, loss[loss=0.2047, simple_loss=0.2957, pruned_loss=0.05682, over 8512.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2828, pruned_loss=0.05837, over 1604398.28 frames. ], batch size: 26, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:53:44,277 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5548, 1.8468, 2.6219, 1.4947, 1.9314, 1.8944, 1.6492, 1.9820], + device='cuda:3'), covar=tensor([0.2002, 0.2577, 0.0972, 0.4626, 0.1955, 0.3390, 0.2456, 0.2283], + device='cuda:3'), in_proj_covar=tensor([0.0538, 0.0633, 0.0563, 0.0668, 0.0660, 0.0610, 0.0561, 0.0648], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 10:54:15,161 INFO [train.py:901] (3/4) Epoch 26, batch 7100, loss[loss=0.1998, simple_loss=0.2914, pruned_loss=0.05408, over 8483.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2822, pruned_loss=0.05778, over 1604614.03 frames. ], batch size: 49, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:54:42,204 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.719e+02 2.538e+02 3.057e+02 3.964e+02 1.199e+03, threshold=6.114e+02, percent-clipped=9.0 +2023-02-07 10:54:49,198 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209220.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:54:50,429 INFO [train.py:901] (3/4) Epoch 26, batch 7150, loss[loss=0.1759, simple_loss=0.2558, pruned_loss=0.04801, over 7784.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2823, pruned_loss=0.05788, over 1603397.23 frames. ], batch size: 19, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:55:25,278 INFO [train.py:901] (3/4) Epoch 26, batch 7200, loss[loss=0.1899, simple_loss=0.2662, pruned_loss=0.05679, over 7434.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2818, pruned_loss=0.0577, over 1601282.43 frames. ], batch size: 17, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:55:26,109 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209273.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:55:51,840 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=209309.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:55:52,281 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.622e+02 2.329e+02 2.733e+02 3.562e+02 6.414e+02, threshold=5.467e+02, percent-clipped=2.0 +2023-02-07 10:55:56,508 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209316.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:56:00,334 INFO [train.py:901] (3/4) Epoch 26, batch 7250, loss[loss=0.2088, simple_loss=0.2998, pruned_loss=0.0589, over 8544.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.282, pruned_loss=0.05792, over 1601070.50 frames. ], batch size: 31, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:56:08,455 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=209334.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:56:33,562 INFO [train.py:901] (3/4) Epoch 26, batch 7300, loss[loss=0.1929, simple_loss=0.2678, pruned_loss=0.05901, over 7417.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2833, pruned_loss=0.05864, over 1600770.17 frames. ], batch size: 17, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:56:59,843 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.697e+02 2.463e+02 3.055e+02 3.934e+02 7.151e+02, threshold=6.111e+02, percent-clipped=5.0 +2023-02-07 10:57:06,726 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-07 10:57:09,426 INFO [train.py:901] (3/4) Epoch 26, batch 7350, loss[loss=0.2331, simple_loss=0.3098, pruned_loss=0.07824, over 8460.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2835, pruned_loss=0.05882, over 1600406.83 frames. ], batch size: 25, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:57:26,326 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-07 10:57:42,840 INFO [train.py:901] (3/4) Epoch 26, batch 7400, loss[loss=0.1818, simple_loss=0.2682, pruned_loss=0.04769, over 8294.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2814, pruned_loss=0.05809, over 1596554.60 frames. ], batch size: 23, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:57:51,016 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209484.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:58:04,804 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-07 10:58:09,472 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.528e+02 2.534e+02 3.042e+02 3.812e+02 9.347e+02, threshold=6.084e+02, percent-clipped=5.0 +2023-02-07 10:58:17,131 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6119, 1.4769, 2.2261, 1.3966, 1.1435, 2.1537, 0.3913, 1.2974], + device='cuda:3'), covar=tensor([0.1693, 0.1391, 0.0370, 0.1265, 0.2707, 0.0394, 0.1919, 0.1369], + device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0205, 0.0133, 0.0224, 0.0276, 0.0144, 0.0172, 0.0198], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 10:58:17,541 INFO [train.py:901] (3/4) Epoch 26, batch 7450, loss[loss=0.2294, simple_loss=0.3255, pruned_loss=0.06668, over 8335.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2826, pruned_loss=0.05864, over 1601775.13 frames. ], batch size: 25, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:58:22,475 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209528.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:58:47,411 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=209564.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:58:52,745 INFO [train.py:901] (3/4) Epoch 26, batch 7500, loss[loss=0.1916, simple_loss=0.2884, pruned_loss=0.04744, over 8246.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2823, pruned_loss=0.05842, over 1606623.27 frames. ], batch size: 22, lr: 2.88e-03, grad_scale: 8.0 +2023-02-07 10:59:18,694 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.713e+02 2.436e+02 2.983e+02 3.503e+02 8.056e+02, threshold=5.967e+02, percent-clipped=5.0 +2023-02-07 10:59:24,167 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=209617.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:59:27,339 INFO [train.py:901] (3/4) Epoch 26, batch 7550, loss[loss=0.1912, simple_loss=0.2842, pruned_loss=0.04911, over 8104.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.282, pruned_loss=0.05821, over 1605838.22 frames. ], batch size: 23, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 10:59:28,771 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209624.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:59:54,107 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=209660.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 10:59:55,498 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8262, 3.7884, 3.4574, 1.8401, 3.3441, 3.5108, 3.3769, 3.3188], + device='cuda:3'), covar=tensor([0.0982, 0.0718, 0.1317, 0.4716, 0.1040, 0.1181, 0.1416, 0.0971], + device='cuda:3'), in_proj_covar=tensor([0.0538, 0.0456, 0.0443, 0.0554, 0.0436, 0.0462, 0.0436, 0.0403], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 11:00:01,941 INFO [train.py:901] (3/4) Epoch 26, batch 7600, loss[loss=0.187, simple_loss=0.2752, pruned_loss=0.04942, over 8469.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2823, pruned_loss=0.05795, over 1607564.03 frames. ], batch size: 25, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 11:00:07,008 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209679.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:00:25,119 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209706.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:00:27,679 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.685e+02 2.404e+02 2.880e+02 3.478e+02 6.437e+02, threshold=5.761e+02, percent-clipped=3.0 +2023-02-07 11:00:35,718 INFO [train.py:901] (3/4) Epoch 26, batch 7650, loss[loss=0.1969, simple_loss=0.2638, pruned_loss=0.06493, over 7441.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2816, pruned_loss=0.05756, over 1605908.32 frames. ], batch size: 17, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 11:00:43,088 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209732.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:01:00,269 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.09 vs. limit=5.0 +2023-02-07 11:01:10,730 INFO [train.py:901] (3/4) Epoch 26, batch 7700, loss[loss=0.2003, simple_loss=0.276, pruned_loss=0.06231, over 7820.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2816, pruned_loss=0.05769, over 1606673.86 frames. ], batch size: 20, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 11:01:12,911 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209775.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:01:14,128 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-07 11:01:23,013 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0 +2023-02-07 11:01:36,859 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.782e+02 2.688e+02 3.083e+02 3.850e+02 9.382e+02, threshold=6.167e+02, percent-clipped=8.0 +2023-02-07 11:01:39,195 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5479, 2.0327, 3.2012, 1.4252, 2.4187, 1.9873, 1.6329, 2.4396], + device='cuda:3'), covar=tensor([0.2087, 0.2739, 0.0915, 0.4888, 0.1969, 0.3283, 0.2601, 0.2346], + device='cuda:3'), in_proj_covar=tensor([0.0537, 0.0631, 0.0560, 0.0666, 0.0659, 0.0610, 0.0559, 0.0644], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 11:01:44,892 INFO [train.py:901] (3/4) Epoch 26, batch 7750, loss[loss=0.1986, simple_loss=0.2833, pruned_loss=0.05692, over 8342.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.282, pruned_loss=0.05793, over 1612198.32 frames. ], batch size: 25, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 11:01:48,919 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=209828.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:02:19,753 INFO [train.py:901] (3/4) Epoch 26, batch 7800, loss[loss=0.181, simple_loss=0.2727, pruned_loss=0.04462, over 8610.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2824, pruned_loss=0.05798, over 1613499.24 frames. ], batch size: 34, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 11:02:19,851 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=209872.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:02:44,107 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.7069, 1.5542, 4.8874, 1.8305, 4.3733, 4.1534, 4.4291, 4.3458], + device='cuda:3'), covar=tensor([0.0590, 0.5128, 0.0467, 0.4125, 0.1057, 0.0930, 0.0586, 0.0639], + device='cuda:3'), in_proj_covar=tensor([0.0672, 0.0666, 0.0734, 0.0658, 0.0742, 0.0633, 0.0633, 0.0709], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 11:02:45,312 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.499e+02 2.465e+02 2.962e+02 3.430e+02 5.705e+02, threshold=5.924e+02, percent-clipped=0.0 +2023-02-07 11:02:53,248 INFO [train.py:901] (3/4) Epoch 26, batch 7850, loss[loss=0.217, simple_loss=0.2972, pruned_loss=0.06842, over 8022.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.284, pruned_loss=0.05862, over 1615114.32 frames. ], batch size: 22, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 11:03:01,894 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=209935.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:03:07,221 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209943.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:03:18,183 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=209960.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:03:23,281 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=209968.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:03:25,898 INFO [train.py:901] (3/4) Epoch 26, batch 7900, loss[loss=0.1867, simple_loss=0.2731, pruned_loss=0.05021, over 8502.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.284, pruned_loss=0.0585, over 1619053.98 frames. ], batch size: 29, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 11:03:35,880 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209987.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:03:36,569 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=209988.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:03:51,906 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.615e+02 2.305e+02 2.795e+02 3.387e+02 5.942e+02, threshold=5.591e+02, percent-clipped=1.0 +2023-02-07 11:03:54,074 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210013.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:03:59,748 INFO [train.py:901] (3/4) Epoch 26, batch 7950, loss[loss=0.246, simple_loss=0.3255, pruned_loss=0.08321, over 8482.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2834, pruned_loss=0.05836, over 1618039.62 frames. ], batch size: 29, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 11:04:06,009 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210031.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:04:18,599 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=210050.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:04:22,777 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210056.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:04:33,087 INFO [train.py:901] (3/4) Epoch 26, batch 8000, loss[loss=0.1945, simple_loss=0.2712, pruned_loss=0.05892, over 7938.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2836, pruned_loss=0.05823, over 1614708.14 frames. ], batch size: 20, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 11:04:35,214 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210075.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:04:40,671 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210083.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:04:58,001 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.537e+02 2.316e+02 2.819e+02 3.710e+02 9.270e+02, threshold=5.638e+02, percent-clipped=7.0 +2023-02-07 11:05:05,833 INFO [train.py:901] (3/4) Epoch 26, batch 8050, loss[loss=0.173, simple_loss=0.2518, pruned_loss=0.04712, over 7931.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2816, pruned_loss=0.05767, over 1598547.63 frames. ], batch size: 20, lr: 2.87e-03, grad_scale: 8.0 +2023-02-07 11:05:38,130 WARNING [train.py:1067] (3/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-07 11:05:43,130 INFO [train.py:901] (3/4) Epoch 27, batch 0, loss[loss=0.1826, simple_loss=0.2734, pruned_loss=0.0459, over 8085.00 frames. ], tot_loss[loss=0.1826, simple_loss=0.2734, pruned_loss=0.0459, over 8085.00 frames. ], batch size: 21, lr: 2.82e-03, grad_scale: 8.0 +2023-02-07 11:05:43,131 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 11:05:54,191 INFO [train.py:935] (3/4) Epoch 27, validation: loss=0.172, simple_loss=0.2713, pruned_loss=0.03628, over 944034.00 frames. +2023-02-07 11:05:54,192 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6747MB +2023-02-07 11:06:01,185 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210165.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:06:08,374 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-07 11:06:24,774 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210199.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:06:28,655 INFO [train.py:901] (3/4) Epoch 27, batch 50, loss[loss=0.1982, simple_loss=0.2846, pruned_loss=0.0559, over 8291.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2875, pruned_loss=0.06129, over 366851.73 frames. ], batch size: 23, lr: 2.82e-03, grad_scale: 8.0 +2023-02-07 11:06:33,574 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.449e+02 2.417e+02 2.930e+02 3.516e+02 7.088e+02, threshold=5.860e+02, percent-clipped=5.0 +2023-02-07 11:06:41,910 WARNING [train.py:1067] (3/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-07 11:06:43,320 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210224.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:06:56,862 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210243.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:07:04,542 INFO [train.py:901] (3/4) Epoch 27, batch 100, loss[loss=0.2247, simple_loss=0.2944, pruned_loss=0.07755, over 8110.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2861, pruned_loss=0.0605, over 644245.89 frames. ], batch size: 23, lr: 2.82e-03, grad_scale: 8.0 +2023-02-07 11:07:05,182 WARNING [train.py:1067] (3/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-07 11:07:13,375 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210268.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:07:38,001 INFO [train.py:901] (3/4) Epoch 27, batch 150, loss[loss=0.1972, simple_loss=0.2885, pruned_loss=0.05293, over 8337.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2857, pruned_loss=0.06022, over 859768.71 frames. ], batch size: 26, lr: 2.82e-03, grad_scale: 8.0 +2023-02-07 11:07:41,159 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.399e+02 2.352e+02 2.905e+02 3.661e+02 1.089e+03, threshold=5.811e+02, percent-clipped=3.0 +2023-02-07 11:07:41,642 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 +2023-02-07 11:08:02,852 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210339.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:08:14,089 INFO [train.py:901] (3/4) Epoch 27, batch 200, loss[loss=0.1972, simple_loss=0.2861, pruned_loss=0.05413, over 8142.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.286, pruned_loss=0.06037, over 1026167.35 frames. ], batch size: 22, lr: 2.82e-03, grad_scale: 8.0 +2023-02-07 11:08:20,482 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210364.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:08:48,338 INFO [train.py:901] (3/4) Epoch 27, batch 250, loss[loss=0.2112, simple_loss=0.2806, pruned_loss=0.07084, over 8030.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2861, pruned_loss=0.06005, over 1160280.03 frames. ], batch size: 22, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:08:49,372 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-07 11:08:51,580 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.513e+02 2.304e+02 2.819e+02 3.559e+02 6.263e+02, threshold=5.638e+02, percent-clipped=1.0 +2023-02-07 11:08:57,565 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-07 11:08:57,631 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=210419.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:08:59,067 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210421.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:09:06,462 WARNING [train.py:1067] (3/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-07 11:09:10,764 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1814, 1.9584, 2.3994, 2.0757, 2.4266, 2.2602, 2.0921, 1.2885], + device='cuda:3'), covar=tensor([0.5996, 0.5028, 0.2125, 0.4177, 0.2780, 0.3143, 0.2060, 0.5525], + device='cuda:3'), in_proj_covar=tensor([0.0965, 0.1017, 0.0828, 0.0985, 0.1021, 0.0926, 0.0772, 0.0847], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 11:09:16,573 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210446.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:09:23,009 INFO [train.py:901] (3/4) Epoch 27, batch 300, loss[loss=0.2043, simple_loss=0.2899, pruned_loss=0.05938, over 7963.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2851, pruned_loss=0.05944, over 1260302.23 frames. ], batch size: 21, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:09:57,343 INFO [train.py:901] (3/4) Epoch 27, batch 350, loss[loss=0.2381, simple_loss=0.2969, pruned_loss=0.08963, over 6933.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2845, pruned_loss=0.05968, over 1336567.14 frames. ], batch size: 72, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:10:00,691 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.497e+02 2.321e+02 2.740e+02 3.479e+02 7.751e+02, threshold=5.481e+02, percent-clipped=4.0 +2023-02-07 11:10:06,930 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3261, 1.7899, 1.3037, 2.9443, 1.2926, 1.2798, 2.0209, 1.9266], + device='cuda:3'), covar=tensor([0.1617, 0.1395, 0.2040, 0.0381, 0.1481, 0.2104, 0.1069, 0.1156], + device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0194, 0.0246, 0.0211, 0.0202, 0.0245, 0.0249, 0.0205], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 11:10:16,907 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210534.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:10:29,881 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-07 11:10:30,766 INFO [train.py:901] (3/4) Epoch 27, batch 400, loss[loss=0.2203, simple_loss=0.3081, pruned_loss=0.06624, over 8806.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2843, pruned_loss=0.05939, over 1396326.47 frames. ], batch size: 39, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:11:06,884 INFO [train.py:901] (3/4) Epoch 27, batch 450, loss[loss=0.1952, simple_loss=0.2895, pruned_loss=0.05049, over 8257.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.284, pruned_loss=0.05919, over 1444575.80 frames. ], batch size: 22, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:11:10,233 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.392e+02 2.445e+02 3.096e+02 3.744e+02 6.670e+02, threshold=6.192e+02, percent-clipped=5.0 +2023-02-07 11:11:28,589 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.1553, 4.1590, 3.7924, 1.9938, 3.6365, 3.8246, 3.7037, 3.7087], + device='cuda:3'), covar=tensor([0.0831, 0.0570, 0.1064, 0.4203, 0.0908, 0.1106, 0.1334, 0.0850], + device='cuda:3'), in_proj_covar=tensor([0.0540, 0.0458, 0.0442, 0.0555, 0.0437, 0.0460, 0.0436, 0.0405], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 11:11:37,206 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210650.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:11:40,169 INFO [train.py:901] (3/4) Epoch 27, batch 500, loss[loss=0.1963, simple_loss=0.279, pruned_loss=0.0568, over 8466.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2844, pruned_loss=0.05972, over 1483367.95 frames. ], batch size: 25, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:11:55,133 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8492, 2.1155, 2.2222, 1.3768, 2.3101, 1.6862, 0.6794, 2.0625], + device='cuda:3'), covar=tensor([0.0603, 0.0364, 0.0295, 0.0711, 0.0438, 0.0903, 0.0992, 0.0292], + device='cuda:3'), in_proj_covar=tensor([0.0469, 0.0406, 0.0360, 0.0458, 0.0392, 0.0548, 0.0402, 0.0436], + device='cuda:3'), out_proj_covar=tensor([1.2454e-04, 1.0560e-04, 9.4168e-05, 1.2007e-04, 1.0262e-04, 1.5318e-04, + 1.0753e-04, 1.1439e-04], device='cuda:3') +2023-02-07 11:12:01,202 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210684.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:12:12,486 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-07 11:12:12,805 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210700.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:12:15,962 INFO [train.py:901] (3/4) Epoch 27, batch 550, loss[loss=0.2214, simple_loss=0.3045, pruned_loss=0.06912, over 8560.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2851, pruned_loss=0.05967, over 1517817.55 frames. ], batch size: 31, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:12:19,368 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.605e+02 2.336e+02 2.792e+02 3.793e+02 8.487e+02, threshold=5.584e+02, percent-clipped=3.0 +2023-02-07 11:12:50,313 INFO [train.py:901] (3/4) Epoch 27, batch 600, loss[loss=0.1633, simple_loss=0.2464, pruned_loss=0.04007, over 7787.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2847, pruned_loss=0.05909, over 1541933.06 frames. ], batch size: 19, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:13:08,463 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210782.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:13:11,593 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-07 11:13:13,655 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210790.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:13:23,578 INFO [train.py:901] (3/4) Epoch 27, batch 650, loss[loss=0.1716, simple_loss=0.2637, pruned_loss=0.03973, over 8568.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2852, pruned_loss=0.0594, over 1558847.36 frames. ], batch size: 31, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:13:28,262 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.565e+02 2.351e+02 2.894e+02 3.474e+02 6.032e+02, threshold=5.788e+02, percent-clipped=3.0 +2023-02-07 11:13:32,519 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210815.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:13:59,777 INFO [train.py:901] (3/4) Epoch 27, batch 700, loss[loss=0.2108, simple_loss=0.294, pruned_loss=0.06386, over 8507.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2844, pruned_loss=0.05909, over 1568796.05 frames. ], batch size: 29, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:14:29,003 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0935, 2.2452, 1.9591, 2.7723, 1.2612, 1.7138, 2.0821, 2.2090], + device='cuda:3'), covar=tensor([0.0722, 0.0781, 0.0797, 0.0320, 0.1069, 0.1210, 0.0712, 0.0755], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0193, 0.0245, 0.0210, 0.0202, 0.0245, 0.0248, 0.0205], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 11:14:32,721 INFO [train.py:901] (3/4) Epoch 27, batch 750, loss[loss=0.2137, simple_loss=0.2936, pruned_loss=0.06694, over 8282.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2851, pruned_loss=0.05952, over 1580678.26 frames. ], batch size: 23, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:14:35,973 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.614e+02 2.536e+02 2.996e+02 3.960e+02 1.304e+03, threshold=5.993e+02, percent-clipped=7.0 +2023-02-07 11:14:54,994 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-07 11:15:04,238 WARNING [train.py:1067] (3/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-07 11:15:08,844 INFO [train.py:901] (3/4) Epoch 27, batch 800, loss[loss=0.2242, simple_loss=0.3004, pruned_loss=0.07398, over 8528.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2854, pruned_loss=0.05969, over 1584929.96 frames. ], batch size: 26, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:15:11,731 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.76 vs. limit=5.0 +2023-02-07 11:15:34,999 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=210994.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:15:38,615 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 +2023-02-07 11:15:40,390 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211002.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:15:42,223 INFO [train.py:901] (3/4) Epoch 27, batch 850, loss[loss=0.1989, simple_loss=0.2877, pruned_loss=0.05512, over 7936.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2846, pruned_loss=0.05917, over 1594669.94 frames. ], batch size: 20, lr: 2.81e-03, grad_scale: 16.0 +2023-02-07 11:15:45,636 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.581e+02 2.265e+02 2.725e+02 3.482e+02 8.151e+02, threshold=5.450e+02, percent-clipped=2.0 +2023-02-07 11:15:46,884 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-07 11:15:57,723 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=211028.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:16:10,470 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=211044.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:16:17,677 INFO [train.py:901] (3/4) Epoch 27, batch 900, loss[loss=0.1767, simple_loss=0.2633, pruned_loss=0.0451, over 7926.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2837, pruned_loss=0.05884, over 1596256.43 frames. ], batch size: 20, lr: 2.81e-03, grad_scale: 16.0 +2023-02-07 11:16:51,937 INFO [train.py:901] (3/4) Epoch 27, batch 950, loss[loss=0.1883, simple_loss=0.2779, pruned_loss=0.04934, over 8067.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2836, pruned_loss=0.05875, over 1600012.05 frames. ], batch size: 21, lr: 2.81e-03, grad_scale: 16.0 +2023-02-07 11:16:54,834 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211109.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:16:55,251 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.525e+02 2.483e+02 2.981e+02 4.008e+02 9.530e+02, threshold=5.961e+02, percent-clipped=10.0 +2023-02-07 11:17:06,170 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=211126.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:17:11,860 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 +2023-02-07 11:17:17,582 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211143.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:17:18,072 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-07 11:17:26,813 INFO [train.py:901] (3/4) Epoch 27, batch 1000, loss[loss=0.2034, simple_loss=0.2936, pruned_loss=0.0566, over 8501.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2844, pruned_loss=0.05867, over 1610582.12 frames. ], batch size: 26, lr: 2.81e-03, grad_scale: 16.0 +2023-02-07 11:17:30,479 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211159.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:17:46,838 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.89 vs. limit=5.0 +2023-02-07 11:17:53,305 WARNING [train.py:1067] (3/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-07 11:18:03,302 INFO [train.py:901] (3/4) Epoch 27, batch 1050, loss[loss=0.1674, simple_loss=0.2578, pruned_loss=0.03854, over 8356.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2842, pruned_loss=0.05817, over 1610298.45 frames. ], batch size: 24, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:18:05,227 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-07 11:18:07,261 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.750e+02 2.494e+02 3.070e+02 3.818e+02 8.233e+02, threshold=6.140e+02, percent-clipped=4.0 +2023-02-07 11:18:27,174 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211241.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:18:36,464 INFO [train.py:901] (3/4) Epoch 27, batch 1100, loss[loss=0.2057, simple_loss=0.2874, pruned_loss=0.06197, over 8133.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2846, pruned_loss=0.0586, over 1613835.47 frames. ], batch size: 22, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:19:13,343 INFO [train.py:901] (3/4) Epoch 27, batch 1150, loss[loss=0.2068, simple_loss=0.2976, pruned_loss=0.05803, over 8494.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2851, pruned_loss=0.05865, over 1620002.22 frames. ], batch size: 28, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:19:15,967 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-07 11:19:17,152 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 +2023-02-07 11:19:17,279 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.610e+02 2.371e+02 2.782e+02 3.549e+02 6.262e+02, threshold=5.564e+02, percent-clipped=1.0 +2023-02-07 11:19:40,861 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=211346.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:19:46,770 INFO [train.py:901] (3/4) Epoch 27, batch 1200, loss[loss=0.2175, simple_loss=0.3016, pruned_loss=0.06669, over 8361.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2844, pruned_loss=0.05832, over 1618381.28 frames. ], batch size: 26, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:19:53,754 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211365.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:20:08,488 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3353, 2.3336, 1.6183, 2.1830, 1.8421, 1.3785, 1.7966, 1.9986], + device='cuda:3'), covar=tensor([0.1709, 0.0536, 0.1486, 0.0650, 0.0940, 0.2027, 0.1242, 0.1026], + device='cuda:3'), in_proj_covar=tensor([0.0363, 0.0241, 0.0343, 0.0314, 0.0305, 0.0347, 0.0349, 0.0323], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 11:20:11,160 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=211390.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:20:18,556 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211399.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:20:21,056 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8646, 3.8546, 3.5602, 1.9036, 3.3891, 3.5272, 3.4894, 3.3279], + device='cuda:3'), covar=tensor([0.0920, 0.0648, 0.1102, 0.4469, 0.1009, 0.0928, 0.1368, 0.0909], + device='cuda:3'), in_proj_covar=tensor([0.0543, 0.0459, 0.0446, 0.0557, 0.0440, 0.0463, 0.0437, 0.0407], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 11:20:21,774 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211404.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:20:22,042 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-07 11:20:22,321 INFO [train.py:901] (3/4) Epoch 27, batch 1250, loss[loss=0.1706, simple_loss=0.2454, pruned_loss=0.04794, over 7443.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2835, pruned_loss=0.05759, over 1617229.90 frames. ], batch size: 17, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:20:26,160 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.392e+02 2.344e+02 2.922e+02 3.484e+02 6.390e+02, threshold=5.843e+02, percent-clipped=2.0 +2023-02-07 11:20:29,709 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211415.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:20:35,526 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=211424.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:20:39,484 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211430.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:20:46,235 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=211440.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:20:49,935 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 +2023-02-07 11:20:56,235 INFO [train.py:901] (3/4) Epoch 27, batch 1300, loss[loss=0.2219, simple_loss=0.3109, pruned_loss=0.06645, over 8673.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2832, pruned_loss=0.0578, over 1617683.80 frames. ], batch size: 34, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:20:59,852 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5114, 1.7986, 2.6295, 1.4145, 1.8647, 1.8650, 1.5765, 1.8985], + device='cuda:3'), covar=tensor([0.2032, 0.2553, 0.0912, 0.4809, 0.2124, 0.3379, 0.2499, 0.2332], + device='cuda:3'), in_proj_covar=tensor([0.0539, 0.0632, 0.0564, 0.0669, 0.0659, 0.0610, 0.0560, 0.0644], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 11:21:00,467 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211461.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:21:23,172 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6407, 1.4740, 2.8740, 1.3556, 2.2196, 3.0789, 3.2546, 2.6306], + device='cuda:3'), covar=tensor([0.1311, 0.1692, 0.0375, 0.2235, 0.0894, 0.0309, 0.0567, 0.0565], + device='cuda:3'), in_proj_covar=tensor([0.0307, 0.0327, 0.0293, 0.0321, 0.0320, 0.0278, 0.0437, 0.0307], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 11:21:24,574 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211497.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:21:30,515 INFO [train.py:901] (3/4) Epoch 27, batch 1350, loss[loss=0.2132, simple_loss=0.2961, pruned_loss=0.06519, over 8540.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2827, pruned_loss=0.05772, over 1615348.08 frames. ], batch size: 39, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:21:34,486 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.772e+02 2.433e+02 2.859e+02 3.519e+02 6.900e+02, threshold=5.717e+02, percent-clipped=5.0 +2023-02-07 11:21:43,624 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=211522.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:22:05,945 INFO [train.py:901] (3/4) Epoch 27, batch 1400, loss[loss=0.1833, simple_loss=0.2827, pruned_loss=0.04191, over 8326.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2834, pruned_loss=0.05806, over 1614544.60 frames. ], batch size: 25, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:22:39,557 INFO [train.py:901] (3/4) Epoch 27, batch 1450, loss[loss=0.2362, simple_loss=0.3156, pruned_loss=0.07839, over 8282.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2844, pruned_loss=0.05885, over 1609846.20 frames. ], batch size: 23, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:22:43,640 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.809e+02 2.739e+02 3.417e+02 5.363e+02 1.739e+03, threshold=6.835e+02, percent-clipped=22.0 +2023-02-07 11:22:45,697 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-07 11:23:15,887 INFO [train.py:901] (3/4) Epoch 27, batch 1500, loss[loss=0.2511, simple_loss=0.3263, pruned_loss=0.08797, over 6848.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2846, pruned_loss=0.05901, over 1611023.14 frames. ], batch size: 71, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:23:49,650 INFO [train.py:901] (3/4) Epoch 27, batch 1550, loss[loss=0.2093, simple_loss=0.2898, pruned_loss=0.06441, over 8508.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2839, pruned_loss=0.05852, over 1613890.49 frames. ], batch size: 26, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:23:53,681 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.543e+02 2.372e+02 3.027e+02 3.476e+02 5.786e+02, threshold=6.054e+02, percent-clipped=0.0 +2023-02-07 11:23:57,840 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211717.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:24:15,294 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=211742.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:24:19,425 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=211748.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:24:24,774 INFO [train.py:901] (3/4) Epoch 27, batch 1600, loss[loss=0.2176, simple_loss=0.2993, pruned_loss=0.06796, over 8510.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2834, pruned_loss=0.05833, over 1612134.07 frames. ], batch size: 26, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:24:28,362 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4382, 1.3280, 1.6845, 1.1169, 1.1612, 1.6779, 0.5713, 1.2757], + device='cuda:3'), covar=tensor([0.1376, 0.0927, 0.0405, 0.0879, 0.1858, 0.0394, 0.1682, 0.1174], + device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0203, 0.0133, 0.0221, 0.0275, 0.0144, 0.0172, 0.0197], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 11:24:38,874 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=211774.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:24:59,722 INFO [train.py:901] (3/4) Epoch 27, batch 1650, loss[loss=0.1895, simple_loss=0.2774, pruned_loss=0.05078, over 8077.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2843, pruned_loss=0.05871, over 1614276.01 frames. ], batch size: 21, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:25:03,544 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-07 11:25:03,799 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.685e+02 2.547e+02 3.089e+02 3.889e+02 1.356e+03, threshold=6.177e+02, percent-clipped=3.0 +2023-02-07 11:25:34,307 INFO [train.py:901] (3/4) Epoch 27, batch 1700, loss[loss=0.1792, simple_loss=0.2711, pruned_loss=0.04366, over 8108.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2852, pruned_loss=0.05885, over 1616675.68 frames. ], batch size: 23, lr: 2.81e-03, grad_scale: 8.0 +2023-02-07 11:25:39,902 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211863.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:25:57,119 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0430, 1.3917, 3.4540, 1.6206, 2.2973, 3.8617, 3.9792, 3.3377], + device='cuda:3'), covar=tensor([0.1150, 0.1965, 0.0353, 0.2145, 0.1187, 0.0229, 0.0408, 0.0503], + device='cuda:3'), in_proj_covar=tensor([0.0307, 0.0329, 0.0295, 0.0322, 0.0323, 0.0279, 0.0440, 0.0308], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 11:25:59,112 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211889.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:26:09,566 INFO [train.py:901] (3/4) Epoch 27, batch 1750, loss[loss=0.1803, simple_loss=0.2714, pruned_loss=0.04465, over 7964.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2839, pruned_loss=0.0587, over 1611296.96 frames. ], batch size: 21, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:26:13,486 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.324e+02 2.460e+02 2.972e+02 3.773e+02 5.726e+02, threshold=5.944e+02, percent-clipped=0.0 +2023-02-07 11:26:24,664 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.82 vs. limit=5.0 +2023-02-07 11:26:43,420 INFO [train.py:901] (3/4) Epoch 27, batch 1800, loss[loss=0.2058, simple_loss=0.2947, pruned_loss=0.05846, over 8139.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2826, pruned_loss=0.05797, over 1611762.93 frames. ], batch size: 22, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:27:03,808 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 +2023-02-07 11:27:20,497 INFO [train.py:901] (3/4) Epoch 27, batch 1850, loss[loss=0.2343, simple_loss=0.3185, pruned_loss=0.07503, over 8623.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2821, pruned_loss=0.05759, over 1611915.49 frames. ], batch size: 34, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:27:24,575 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.542e+02 2.252e+02 2.767e+02 3.484e+02 5.487e+02, threshold=5.534e+02, percent-clipped=0.0 +2023-02-07 11:27:54,162 INFO [train.py:901] (3/4) Epoch 27, batch 1900, loss[loss=0.1839, simple_loss=0.2668, pruned_loss=0.05049, over 7965.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2831, pruned_loss=0.05796, over 1611523.16 frames. ], batch size: 21, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:28:23,913 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8756, 3.8071, 3.4894, 1.7032, 3.4175, 3.5218, 3.4084, 3.3921], + device='cuda:3'), covar=tensor([0.1045, 0.0739, 0.1261, 0.4810, 0.1026, 0.1131, 0.1610, 0.0837], + device='cuda:3'), in_proj_covar=tensor([0.0540, 0.0456, 0.0442, 0.0552, 0.0438, 0.0460, 0.0437, 0.0402], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 11:28:25,185 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-07 11:28:28,380 INFO [train.py:901] (3/4) Epoch 27, batch 1950, loss[loss=0.2144, simple_loss=0.2966, pruned_loss=0.06608, over 8026.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2829, pruned_loss=0.05801, over 1611866.37 frames. ], batch size: 22, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:28:33,126 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.526e+02 2.484e+02 3.059e+02 3.727e+02 7.478e+02, threshold=6.119e+02, percent-clipped=3.0 +2023-02-07 11:28:38,414 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-07 11:28:39,362 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=212119.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:28:41,243 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=212122.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 11:28:56,767 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=212144.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:28:57,282 WARNING [train.py:1067] (3/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-07 11:28:57,487 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=212145.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:29:03,980 INFO [train.py:901] (3/4) Epoch 27, batch 2000, loss[loss=0.1921, simple_loss=0.2884, pruned_loss=0.0479, over 8199.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2825, pruned_loss=0.05788, over 1609208.96 frames. ], batch size: 23, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:29:14,158 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=212170.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:29:37,643 INFO [train.py:901] (3/4) Epoch 27, batch 2050, loss[loss=0.1969, simple_loss=0.2817, pruned_loss=0.05603, over 7807.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2816, pruned_loss=0.05706, over 1611598.65 frames. ], batch size: 19, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:29:41,738 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.688e+02 2.364e+02 2.966e+02 3.655e+02 9.314e+02, threshold=5.932e+02, percent-clipped=4.0 +2023-02-07 11:29:53,497 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1772, 1.4405, 1.6567, 1.4002, 0.9701, 1.4435, 1.6569, 1.8064], + device='cuda:3'), covar=tensor([0.0531, 0.1272, 0.1702, 0.1508, 0.0635, 0.1492, 0.0743, 0.0591], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0154, 0.0190, 0.0163, 0.0102, 0.0164, 0.0113, 0.0146], + device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 11:30:13,817 INFO [train.py:901] (3/4) Epoch 27, batch 2100, loss[loss=0.2081, simple_loss=0.2928, pruned_loss=0.0617, over 8286.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2812, pruned_loss=0.05671, over 1614419.69 frames. ], batch size: 23, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:30:24,389 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.4495, 4.3515, 4.0356, 2.1940, 3.8794, 4.1054, 3.8758, 3.9300], + device='cuda:3'), covar=tensor([0.0622, 0.0497, 0.0908, 0.4095, 0.0817, 0.0761, 0.1127, 0.0684], + device='cuda:3'), in_proj_covar=tensor([0.0539, 0.0456, 0.0442, 0.0554, 0.0438, 0.0459, 0.0437, 0.0404], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 11:30:25,137 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=212271.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:30:42,130 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 +2023-02-07 11:30:47,851 INFO [train.py:901] (3/4) Epoch 27, batch 2150, loss[loss=0.1817, simple_loss=0.266, pruned_loss=0.04873, over 8294.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2807, pruned_loss=0.05648, over 1614413.21 frames. ], batch size: 23, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:30:48,031 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.0038, 1.6436, 1.3808, 1.5212, 1.3057, 1.2434, 1.2722, 1.2565], + device='cuda:3'), covar=tensor([0.1233, 0.0510, 0.1310, 0.0614, 0.0799, 0.1599, 0.1007, 0.0879], + device='cuda:3'), in_proj_covar=tensor([0.0365, 0.0240, 0.0341, 0.0316, 0.0304, 0.0349, 0.0350, 0.0323], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 11:30:51,762 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.722e+02 2.395e+02 2.764e+02 3.582e+02 6.444e+02, threshold=5.527e+02, percent-clipped=1.0 +2023-02-07 11:31:22,777 INFO [train.py:901] (3/4) Epoch 27, batch 2200, loss[loss=0.2306, simple_loss=0.3122, pruned_loss=0.07455, over 8615.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2815, pruned_loss=0.0571, over 1616560.66 frames. ], batch size: 34, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:31:57,344 INFO [train.py:901] (3/4) Epoch 27, batch 2250, loss[loss=0.2245, simple_loss=0.3138, pruned_loss=0.06765, over 8735.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2817, pruned_loss=0.05699, over 1616521.47 frames. ], batch size: 34, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:32:01,575 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.617e+02 2.301e+02 2.815e+02 3.457e+02 5.141e+02, threshold=5.631e+02, percent-clipped=0.0 +2023-02-07 11:32:31,465 INFO [train.py:901] (3/4) Epoch 27, batch 2300, loss[loss=0.2208, simple_loss=0.3148, pruned_loss=0.06337, over 8245.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2817, pruned_loss=0.05692, over 1614289.56 frames. ], batch size: 24, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:32:33,511 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=212457.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:32:39,373 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=212466.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 11:33:07,929 INFO [train.py:901] (3/4) Epoch 27, batch 2350, loss[loss=0.218, simple_loss=0.3042, pruned_loss=0.06589, over 8594.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2812, pruned_loss=0.05695, over 1614699.87 frames. ], batch size: 31, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:33:12,148 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.444e+02 2.366e+02 2.801e+02 3.492e+02 6.818e+02, threshold=5.601e+02, percent-clipped=4.0 +2023-02-07 11:33:42,852 INFO [train.py:901] (3/4) Epoch 27, batch 2400, loss[loss=0.2292, simple_loss=0.3061, pruned_loss=0.07612, over 8489.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2813, pruned_loss=0.05739, over 1616609.50 frames. ], batch size: 39, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:34:01,680 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=212581.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 11:34:04,909 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8510, 3.7569, 3.4445, 1.8894, 3.4064, 3.3910, 3.3267, 3.2490], + device='cuda:3'), covar=tensor([0.0890, 0.0703, 0.1184, 0.4564, 0.0970, 0.1243, 0.1497, 0.0927], + device='cuda:3'), in_proj_covar=tensor([0.0542, 0.0460, 0.0444, 0.0558, 0.0440, 0.0464, 0.0439, 0.0406], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 11:34:19,818 INFO [train.py:901] (3/4) Epoch 27, batch 2450, loss[loss=0.1775, simple_loss=0.2616, pruned_loss=0.04675, over 8460.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2801, pruned_loss=0.05729, over 1610576.97 frames. ], batch size: 29, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:34:23,906 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.730e+02 2.406e+02 2.879e+02 3.948e+02 9.646e+02, threshold=5.757e+02, percent-clipped=9.0 +2023-02-07 11:34:26,763 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=212615.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:34:46,276 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 +2023-02-07 11:34:54,052 INFO [train.py:901] (3/4) Epoch 27, batch 2500, loss[loss=0.1735, simple_loss=0.2529, pruned_loss=0.04705, over 7813.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2795, pruned_loss=0.05702, over 1608891.68 frames. ], batch size: 19, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:34:59,682 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3378, 1.2231, 2.3695, 1.1936, 2.0826, 2.5119, 2.7211, 2.1034], + device='cuda:3'), covar=tensor([0.1296, 0.1625, 0.0449, 0.2416, 0.0829, 0.0407, 0.0716, 0.0740], + device='cuda:3'), in_proj_covar=tensor([0.0305, 0.0327, 0.0292, 0.0321, 0.0320, 0.0277, 0.0437, 0.0308], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 11:35:28,281 INFO [train.py:901] (3/4) Epoch 27, batch 2550, loss[loss=0.1954, simple_loss=0.2839, pruned_loss=0.05343, over 8455.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2798, pruned_loss=0.05735, over 1615036.67 frames. ], batch size: 27, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:35:33,066 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.684e+02 2.333e+02 2.985e+02 3.926e+02 7.498e+02, threshold=5.971e+02, percent-clipped=4.0 +2023-02-07 11:35:37,327 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=212716.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:35:46,883 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=212729.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:35:47,587 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=212730.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:36:04,562 INFO [train.py:901] (3/4) Epoch 27, batch 2600, loss[loss=0.1743, simple_loss=0.2584, pruned_loss=0.04508, over 7654.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2793, pruned_loss=0.05678, over 1612852.13 frames. ], batch size: 19, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:36:09,618 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 +2023-02-07 11:36:18,127 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4447, 1.2635, 2.4162, 1.2838, 2.2564, 2.5630, 2.7158, 2.1771], + device='cuda:3'), covar=tensor([0.1117, 0.1486, 0.0429, 0.2142, 0.0725, 0.0366, 0.0642, 0.0657], + device='cuda:3'), in_proj_covar=tensor([0.0305, 0.0327, 0.0292, 0.0320, 0.0320, 0.0277, 0.0438, 0.0307], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 11:36:21,581 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.4126, 3.8848, 2.7120, 3.0485, 3.0885, 2.3593, 3.1560, 3.2899], + device='cuda:3'), covar=tensor([0.1631, 0.0378, 0.1009, 0.0733, 0.0698, 0.1499, 0.0996, 0.1079], + device='cuda:3'), in_proj_covar=tensor([0.0359, 0.0238, 0.0337, 0.0312, 0.0301, 0.0345, 0.0346, 0.0319], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0002, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 11:36:29,686 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=212792.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:36:33,817 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8157, 1.5838, 2.4004, 1.5992, 1.2643, 2.3237, 0.5192, 1.4915], + device='cuda:3'), covar=tensor([0.1407, 0.1335, 0.0298, 0.1019, 0.2472, 0.0327, 0.1838, 0.1270], + device='cuda:3'), in_proj_covar=tensor([0.0198, 0.0205, 0.0134, 0.0223, 0.0276, 0.0144, 0.0171, 0.0198], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 11:36:35,853 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=212801.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:36:38,506 INFO [train.py:901] (3/4) Epoch 27, batch 2650, loss[loss=0.254, simple_loss=0.3149, pruned_loss=0.0966, over 6668.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2799, pruned_loss=0.05688, over 1616460.46 frames. ], batch size: 71, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:36:43,306 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.695e+02 2.517e+02 2.957e+02 3.589e+02 7.428e+02, threshold=5.913e+02, percent-clipped=3.0 +2023-02-07 11:37:02,814 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=212837.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 11:37:04,100 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9686, 1.7922, 2.5842, 1.5869, 1.4334, 2.4932, 0.5282, 1.6161], + device='cuda:3'), covar=tensor([0.1359, 0.1309, 0.0300, 0.1147, 0.2261, 0.0358, 0.1776, 0.1240], + device='cuda:3'), in_proj_covar=tensor([0.0198, 0.0205, 0.0134, 0.0223, 0.0276, 0.0145, 0.0172, 0.0198], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 11:37:14,954 INFO [train.py:901] (3/4) Epoch 27, batch 2700, loss[loss=0.2023, simple_loss=0.2787, pruned_loss=0.06294, over 7531.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2802, pruned_loss=0.05712, over 1618091.62 frames. ], batch size: 18, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:37:19,850 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=212862.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 11:37:49,368 INFO [train.py:901] (3/4) Epoch 27, batch 2750, loss[loss=0.1937, simple_loss=0.2877, pruned_loss=0.04989, over 8287.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2799, pruned_loss=0.05725, over 1613753.52 frames. ], batch size: 23, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:37:53,337 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.521e+02 2.387e+02 2.942e+02 3.576e+02 8.277e+02, threshold=5.883e+02, percent-clipped=4.0 +2023-02-07 11:37:56,794 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=212916.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:38:25,584 INFO [train.py:901] (3/4) Epoch 27, batch 2800, loss[loss=0.1968, simple_loss=0.2869, pruned_loss=0.05331, over 8338.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2819, pruned_loss=0.05821, over 1610909.85 frames. ], batch size: 25, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:38:43,404 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5024, 1.4566, 2.1358, 1.2056, 0.9674, 2.1111, 0.3153, 1.2403], + device='cuda:3'), covar=tensor([0.1624, 0.1292, 0.0315, 0.1208, 0.2633, 0.0318, 0.1779, 0.1333], + device='cuda:3'), in_proj_covar=tensor([0.0199, 0.0205, 0.0134, 0.0224, 0.0277, 0.0145, 0.0172, 0.0199], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 11:38:46,067 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=212986.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:38:58,530 INFO [train.py:901] (3/4) Epoch 27, batch 2850, loss[loss=0.222, simple_loss=0.3085, pruned_loss=0.06779, over 8133.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2816, pruned_loss=0.05822, over 1608485.28 frames. ], batch size: 22, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:39:02,632 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.396e+02 2.420e+02 3.040e+02 3.738e+02 9.771e+02, threshold=6.080e+02, percent-clipped=4.0 +2023-02-07 11:39:02,866 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213011.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:39:03,498 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9100, 1.5536, 1.7360, 1.4490, 1.0322, 1.5554, 1.6922, 1.4316], + device='cuda:3'), covar=tensor([0.0547, 0.1228, 0.1556, 0.1431, 0.0614, 0.1421, 0.0719, 0.0682], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0153, 0.0189, 0.0162, 0.0101, 0.0164, 0.0112, 0.0146], + device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 11:39:33,429 INFO [train.py:901] (3/4) Epoch 27, batch 2900, loss[loss=0.2022, simple_loss=0.2907, pruned_loss=0.05689, over 8742.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2821, pruned_loss=0.0578, over 1614193.24 frames. ], batch size: 30, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:39:36,855 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=213060.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:39:46,829 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=213073.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:40:08,834 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=3.88 vs. limit=5.0 +2023-02-07 11:40:09,156 INFO [train.py:901] (3/4) Epoch 27, batch 2950, loss[loss=0.1654, simple_loss=0.2461, pruned_loss=0.0424, over 7804.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2822, pruned_loss=0.05766, over 1613560.11 frames. ], batch size: 19, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:40:12,510 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-07 11:40:13,189 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.490e+02 2.292e+02 2.734e+02 3.601e+02 6.803e+02, threshold=5.467e+02, percent-clipped=1.0 +2023-02-07 11:40:30,261 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=213136.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:40:42,820 INFO [train.py:901] (3/4) Epoch 27, batch 3000, loss[loss=0.2126, simple_loss=0.3076, pruned_loss=0.05885, over 8251.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2838, pruned_loss=0.05847, over 1613979.52 frames. ], batch size: 24, lr: 2.80e-03, grad_scale: 8.0 +2023-02-07 11:40:42,820 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 11:40:56,476 INFO [train.py:935] (3/4) Epoch 27, validation: loss=0.171, simple_loss=0.2706, pruned_loss=0.03572, over 944034.00 frames. +2023-02-07 11:40:56,477 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6747MB +2023-02-07 11:41:08,340 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=213172.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:41:10,350 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=213175.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:41:19,878 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=213188.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:41:25,955 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213197.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:41:30,815 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-07 11:41:31,708 INFO [train.py:901] (3/4) Epoch 27, batch 3050, loss[loss=0.23, simple_loss=0.319, pruned_loss=0.07046, over 8593.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2841, pruned_loss=0.05871, over 1613713.15 frames. ], batch size: 31, lr: 2.80e-03, grad_scale: 16.0 +2023-02-07 11:41:36,535 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.721e+02 2.283e+02 2.877e+02 3.649e+02 6.604e+02, threshold=5.754e+02, percent-clipped=7.0 +2023-02-07 11:41:40,805 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7503, 1.3720, 2.8886, 1.3796, 2.2604, 3.0692, 3.2497, 2.6188], + device='cuda:3'), covar=tensor([0.1142, 0.1667, 0.0357, 0.2208, 0.0916, 0.0299, 0.0660, 0.0520], + device='cuda:3'), in_proj_covar=tensor([0.0306, 0.0328, 0.0293, 0.0321, 0.0321, 0.0279, 0.0438, 0.0309], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 11:42:04,116 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=213251.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:42:06,436 INFO [train.py:901] (3/4) Epoch 27, batch 3100, loss[loss=0.2362, simple_loss=0.3119, pruned_loss=0.08024, over 8038.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2838, pruned_loss=0.05867, over 1612767.67 frames. ], batch size: 22, lr: 2.80e-03, grad_scale: 16.0 +2023-02-07 11:42:29,244 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-07 11:42:40,166 INFO [train.py:901] (3/4) Epoch 27, batch 3150, loss[loss=0.1946, simple_loss=0.2847, pruned_loss=0.05221, over 8503.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2837, pruned_loss=0.0585, over 1612909.52 frames. ], batch size: 26, lr: 2.80e-03, grad_scale: 16.0 +2023-02-07 11:42:44,226 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.764e+02 2.557e+02 3.186e+02 3.836e+02 1.080e+03, threshold=6.372e+02, percent-clipped=6.0 +2023-02-07 11:42:57,218 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5108, 2.2988, 3.1217, 2.4875, 3.0659, 2.5480, 2.4039, 1.9242], + device='cuda:3'), covar=tensor([0.5840, 0.5518, 0.2139, 0.4117, 0.2613, 0.3069, 0.1902, 0.5869], + device='cuda:3'), in_proj_covar=tensor([0.0965, 0.1016, 0.0828, 0.0989, 0.1022, 0.0927, 0.0770, 0.0850], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 11:43:15,308 INFO [train.py:901] (3/4) Epoch 27, batch 3200, loss[loss=0.1772, simple_loss=0.2624, pruned_loss=0.04605, over 7810.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2834, pruned_loss=0.05791, over 1617840.67 frames. ], batch size: 20, lr: 2.80e-03, grad_scale: 16.0 +2023-02-07 11:43:38,615 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.6973, 4.7742, 4.2390, 2.1557, 4.1661, 4.4166, 4.3573, 4.2565], + device='cuda:3'), covar=tensor([0.0658, 0.0453, 0.0977, 0.4103, 0.0834, 0.0815, 0.1035, 0.0603], + device='cuda:3'), in_proj_covar=tensor([0.0542, 0.0460, 0.0447, 0.0557, 0.0441, 0.0464, 0.0437, 0.0405], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 11:43:46,357 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.5489, 2.4115, 3.1255, 2.5571, 3.2423, 2.6112, 2.4838, 1.9423], + device='cuda:3'), covar=tensor([0.5867, 0.5387, 0.2285, 0.4233, 0.2635, 0.3295, 0.1883, 0.6082], + device='cuda:3'), in_proj_covar=tensor([0.0968, 0.1021, 0.0831, 0.0992, 0.1026, 0.0930, 0.0772, 0.0854], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 11:43:48,886 INFO [train.py:901] (3/4) Epoch 27, batch 3250, loss[loss=0.1982, simple_loss=0.2858, pruned_loss=0.05531, over 8454.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2833, pruned_loss=0.05813, over 1616374.56 frames. ], batch size: 25, lr: 2.80e-03, grad_scale: 16.0 +2023-02-07 11:43:52,809 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.819e+02 2.478e+02 2.885e+02 3.413e+02 5.983e+02, threshold=5.770e+02, percent-clipped=0.0 +2023-02-07 11:44:07,290 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=213431.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:44:17,687 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=213444.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:44:21,743 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 +2023-02-07 11:44:25,453 INFO [train.py:901] (3/4) Epoch 27, batch 3300, loss[loss=0.1579, simple_loss=0.2473, pruned_loss=0.03424, over 7664.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2836, pruned_loss=0.05802, over 1616268.64 frames. ], batch size: 19, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:44:26,274 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213456.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:44:35,225 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213469.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:44:58,305 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7933, 2.4912, 4.0026, 1.6170, 3.0698, 2.3095, 1.9770, 2.9168], + device='cuda:3'), covar=tensor([0.2214, 0.2831, 0.1020, 0.5150, 0.1973, 0.3553, 0.2698, 0.2703], + device='cuda:3'), in_proj_covar=tensor([0.0538, 0.0634, 0.0567, 0.0670, 0.0660, 0.0608, 0.0561, 0.0645], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 11:44:59,366 INFO [train.py:901] (3/4) Epoch 27, batch 3350, loss[loss=0.2072, simple_loss=0.2954, pruned_loss=0.05944, over 8628.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2821, pruned_loss=0.05764, over 1609778.73 frames. ], batch size: 39, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:45:00,990 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=213507.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:45:03,448 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.542e+02 2.606e+02 3.102e+02 3.998e+02 8.787e+02, threshold=6.203e+02, percent-clipped=8.0 +2023-02-07 11:45:18,441 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213532.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:45:34,511 INFO [train.py:901] (3/4) Epoch 27, batch 3400, loss[loss=0.1894, simple_loss=0.2824, pruned_loss=0.0482, over 8283.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2824, pruned_loss=0.0579, over 1609132.50 frames. ], batch size: 23, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:45:54,959 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 +2023-02-07 11:45:55,326 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=213584.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:46:09,623 INFO [train.py:901] (3/4) Epoch 27, batch 3450, loss[loss=0.1908, simple_loss=0.2775, pruned_loss=0.05202, over 8079.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.285, pruned_loss=0.05895, over 1610545.96 frames. ], batch size: 21, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:46:13,709 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.444e+02 2.297e+02 2.616e+02 3.439e+02 9.820e+02, threshold=5.232e+02, percent-clipped=1.0 +2023-02-07 11:46:13,985 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1895, 1.9211, 2.4392, 2.0793, 2.5129, 2.2477, 2.1121, 1.3486], + device='cuda:3'), covar=tensor([0.5876, 0.4998, 0.2189, 0.3641, 0.2529, 0.3058, 0.1948, 0.5447], + device='cuda:3'), in_proj_covar=tensor([0.0964, 0.1019, 0.0829, 0.0988, 0.1019, 0.0929, 0.0770, 0.0851], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 11:46:44,530 INFO [train.py:901] (3/4) Epoch 27, batch 3500, loss[loss=0.1861, simple_loss=0.2642, pruned_loss=0.05398, over 7710.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2855, pruned_loss=0.05939, over 1612165.82 frames. ], batch size: 18, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:46:57,107 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 +2023-02-07 11:47:11,033 WARNING [train.py:1067] (3/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-07 11:47:20,454 INFO [train.py:901] (3/4) Epoch 27, batch 3550, loss[loss=0.1984, simple_loss=0.2818, pruned_loss=0.0575, over 7977.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.284, pruned_loss=0.05865, over 1607025.99 frames. ], batch size: 21, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:47:24,356 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.422e+02 2.484e+02 3.157e+02 3.893e+02 8.912e+02, threshold=6.313e+02, percent-clipped=7.0 +2023-02-07 11:47:55,119 INFO [train.py:901] (3/4) Epoch 27, batch 3600, loss[loss=0.233, simple_loss=0.3224, pruned_loss=0.07185, over 8035.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2833, pruned_loss=0.05842, over 1606240.19 frames. ], batch size: 22, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:47:56,230 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.85 vs. limit=5.0 +2023-02-07 11:48:31,565 INFO [train.py:901] (3/4) Epoch 27, batch 3650, loss[loss=0.1552, simple_loss=0.2369, pruned_loss=0.03673, over 7536.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.282, pruned_loss=0.05756, over 1603847.07 frames. ], batch size: 18, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:48:35,656 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.619e+02 2.435e+02 3.005e+02 4.000e+02 1.001e+03, threshold=6.009e+02, percent-clipped=1.0 +2023-02-07 11:49:05,221 INFO [train.py:901] (3/4) Epoch 27, batch 3700, loss[loss=0.1776, simple_loss=0.2655, pruned_loss=0.04481, over 7815.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2815, pruned_loss=0.05722, over 1603352.91 frames. ], batch size: 20, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:49:11,344 WARNING [train.py:1067] (3/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-07 11:49:18,805 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3404, 1.6284, 1.6717, 1.0116, 1.7048, 1.3768, 0.2562, 1.5775], + device='cuda:3'), covar=tensor([0.0594, 0.0422, 0.0330, 0.0600, 0.0487, 0.0979, 0.1008, 0.0340], + device='cuda:3'), in_proj_covar=tensor([0.0472, 0.0408, 0.0362, 0.0458, 0.0393, 0.0547, 0.0404, 0.0438], + device='cuda:3'), out_proj_covar=tensor([1.2525e-04, 1.0608e-04, 9.4603e-05, 1.1988e-04, 1.0296e-04, 1.5253e-04, + 1.0774e-04, 1.1502e-04], device='cuda:3') +2023-02-07 11:49:40,755 INFO [train.py:901] (3/4) Epoch 27, batch 3750, loss[loss=0.2211, simple_loss=0.3015, pruned_loss=0.07035, over 8661.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2817, pruned_loss=0.05757, over 1608925.73 frames. ], batch size: 34, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:49:44,686 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.478e+02 2.240e+02 2.670e+02 3.453e+02 6.024e+02, threshold=5.340e+02, percent-clipped=1.0 +2023-02-07 11:49:57,383 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=213928.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:50:15,102 INFO [train.py:901] (3/4) Epoch 27, batch 3800, loss[loss=0.1807, simple_loss=0.2569, pruned_loss=0.05221, over 7436.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2827, pruned_loss=0.05842, over 1610746.85 frames. ], batch size: 17, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:50:51,696 INFO [train.py:901] (3/4) Epoch 27, batch 3850, loss[loss=0.1883, simple_loss=0.2721, pruned_loss=0.0523, over 7943.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2825, pruned_loss=0.058, over 1611537.43 frames. ], batch size: 20, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:50:55,678 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.477e+02 2.325e+02 2.987e+02 3.815e+02 9.366e+02, threshold=5.974e+02, percent-clipped=6.0 +2023-02-07 11:51:19,440 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=214043.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:51:21,268 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-07 11:51:27,208 INFO [train.py:901] (3/4) Epoch 27, batch 3900, loss[loss=0.1746, simple_loss=0.2674, pruned_loss=0.04092, over 7921.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2827, pruned_loss=0.05799, over 1610195.84 frames. ], batch size: 20, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:52:00,517 INFO [train.py:901] (3/4) Epoch 27, batch 3950, loss[loss=0.1803, simple_loss=0.2677, pruned_loss=0.04645, over 7924.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2833, pruned_loss=0.05794, over 1613385.87 frames. ], batch size: 20, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:52:04,375 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.530e+02 2.377e+02 2.717e+02 3.364e+02 5.097e+02, threshold=5.435e+02, percent-clipped=0.0 +2023-02-07 11:52:30,744 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=214147.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:52:36,017 INFO [train.py:901] (3/4) Epoch 27, batch 4000, loss[loss=0.184, simple_loss=0.2646, pruned_loss=0.05169, over 7924.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2825, pruned_loss=0.05775, over 1614329.84 frames. ], batch size: 20, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:53:10,790 INFO [train.py:901] (3/4) Epoch 27, batch 4050, loss[loss=0.2117, simple_loss=0.2824, pruned_loss=0.0705, over 8074.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2838, pruned_loss=0.05845, over 1617251.06 frames. ], batch size: 21, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:53:14,930 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.649e+02 2.379e+02 2.958e+02 3.648e+02 7.596e+02, threshold=5.915e+02, percent-clipped=3.0 +2023-02-07 11:53:31,866 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=214236.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:53:41,647 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-07 11:53:45,410 INFO [train.py:901] (3/4) Epoch 27, batch 4100, loss[loss=0.174, simple_loss=0.2551, pruned_loss=0.04645, over 7553.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2829, pruned_loss=0.05789, over 1617004.84 frames. ], batch size: 18, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:53:57,019 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1951, 1.9211, 2.3630, 2.0602, 2.4189, 2.2186, 2.0560, 1.2038], + device='cuda:3'), covar=tensor([0.5826, 0.4996, 0.2256, 0.3717, 0.2363, 0.3378, 0.1855, 0.5301], + device='cuda:3'), in_proj_covar=tensor([0.0966, 0.1019, 0.0831, 0.0988, 0.1024, 0.0928, 0.0768, 0.0849], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 11:54:17,258 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=214299.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:54:21,105 INFO [train.py:901] (3/4) Epoch 27, batch 4150, loss[loss=0.164, simple_loss=0.2364, pruned_loss=0.04581, over 7512.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2837, pruned_loss=0.05865, over 1618760.41 frames. ], batch size: 18, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:54:25,184 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.491e+02 2.522e+02 2.957e+02 3.518e+02 6.524e+02, threshold=5.913e+02, percent-clipped=2.0 +2023-02-07 11:54:34,232 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=214324.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:54:55,413 INFO [train.py:901] (3/4) Epoch 27, batch 4200, loss[loss=0.214, simple_loss=0.3121, pruned_loss=0.05796, over 8451.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2841, pruned_loss=0.05893, over 1618648.16 frames. ], batch size: 27, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:55:15,716 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-07 11:55:30,770 INFO [train.py:901] (3/4) Epoch 27, batch 4250, loss[loss=0.18, simple_loss=0.2723, pruned_loss=0.04389, over 8296.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2845, pruned_loss=0.05936, over 1621470.33 frames. ], batch size: 23, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:55:34,782 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.646e+02 2.351e+02 2.930e+02 3.605e+02 8.966e+02, threshold=5.860e+02, percent-clipped=4.0 +2023-02-07 11:55:40,120 WARNING [train.py:1067] (3/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-07 11:55:52,924 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.43 vs. limit=2.0 +2023-02-07 11:56:04,673 INFO [train.py:901] (3/4) Epoch 27, batch 4300, loss[loss=0.214, simple_loss=0.297, pruned_loss=0.06552, over 8360.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.283, pruned_loss=0.05871, over 1614850.34 frames. ], batch size: 24, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:56:29,280 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=214491.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:56:40,082 INFO [train.py:901] (3/4) Epoch 27, batch 4350, loss[loss=0.1906, simple_loss=0.2768, pruned_loss=0.05217, over 8027.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2845, pruned_loss=0.05937, over 1614869.31 frames. ], batch size: 22, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:56:44,889 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.529e+02 2.457e+02 2.989e+02 4.041e+02 8.697e+02, threshold=5.978e+02, percent-clipped=4.0 +2023-02-07 11:57:04,813 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.0700, 2.3348, 3.7626, 2.0193, 2.0954, 3.6865, 0.7107, 2.1703], + device='cuda:3'), covar=tensor([0.1022, 0.1203, 0.0181, 0.1544, 0.2166, 0.0243, 0.1957, 0.1308], + device='cuda:3'), in_proj_covar=tensor([0.0200, 0.0205, 0.0136, 0.0223, 0.0277, 0.0145, 0.0172, 0.0199], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:3') +2023-02-07 11:57:11,210 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.64 vs. limit=2.0 +2023-02-07 11:57:11,498 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-07 11:57:14,883 INFO [train.py:901] (3/4) Epoch 27, batch 4400, loss[loss=0.1846, simple_loss=0.2766, pruned_loss=0.04629, over 8183.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2828, pruned_loss=0.05816, over 1615562.72 frames. ], batch size: 23, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:57:26,154 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.0621, 2.9533, 2.8082, 4.2541, 1.8094, 2.5591, 2.7482, 3.2468], + device='cuda:3'), covar=tensor([0.0513, 0.0680, 0.0694, 0.0197, 0.1056, 0.1009, 0.0839, 0.0653], + device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0195, 0.0243, 0.0211, 0.0202, 0.0245, 0.0248, 0.0203], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 11:57:32,006 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=214580.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:57:46,454 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.0477, 2.7142, 4.0171, 1.7613, 3.0975, 2.2679, 2.4027, 2.7096], + device='cuda:3'), covar=tensor([0.1985, 0.2267, 0.0944, 0.4725, 0.1759, 0.3539, 0.2324, 0.2679], + device='cuda:3'), in_proj_covar=tensor([0.0536, 0.0630, 0.0562, 0.0666, 0.0656, 0.0605, 0.0560, 0.0641], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 11:57:49,420 INFO [train.py:901] (3/4) Epoch 27, batch 4450, loss[loss=0.2194, simple_loss=0.3113, pruned_loss=0.06379, over 8105.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2812, pruned_loss=0.05757, over 1612320.50 frames. ], batch size: 23, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:57:50,297 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=214606.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:57:51,446 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-07 11:57:53,336 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.553e+02 2.435e+02 2.910e+02 3.675e+02 1.096e+03, threshold=5.821e+02, percent-clipped=3.0 +2023-02-07 11:57:54,950 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.42 vs. limit=5.0 +2023-02-07 11:58:25,065 INFO [train.py:901] (3/4) Epoch 27, batch 4500, loss[loss=0.1941, simple_loss=0.2843, pruned_loss=0.05199, over 8109.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2802, pruned_loss=0.05718, over 1610561.96 frames. ], batch size: 23, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:58:49,087 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-07 11:58:51,961 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=214695.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 11:58:58,573 INFO [train.py:901] (3/4) Epoch 27, batch 4550, loss[loss=0.2045, simple_loss=0.294, pruned_loss=0.05747, over 8458.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.279, pruned_loss=0.05635, over 1607307.29 frames. ], batch size: 25, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:59:03,198 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.415e+02 2.259e+02 2.795e+02 3.667e+02 7.490e+02, threshold=5.591e+02, percent-clipped=6.0 +2023-02-07 11:59:15,587 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 +2023-02-07 11:59:34,573 INFO [train.py:901] (3/4) Epoch 27, batch 4600, loss[loss=0.1505, simple_loss=0.2334, pruned_loss=0.03382, over 7801.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2812, pruned_loss=0.05776, over 1606728.61 frames. ], batch size: 20, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 11:59:43,925 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.7349, 5.8518, 5.1430, 2.5481, 5.1298, 5.5537, 5.3479, 5.3638], + device='cuda:3'), covar=tensor([0.0515, 0.0369, 0.0765, 0.4067, 0.0713, 0.0679, 0.1019, 0.0509], + device='cuda:3'), in_proj_covar=tensor([0.0542, 0.0462, 0.0447, 0.0556, 0.0443, 0.0466, 0.0440, 0.0406], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 12:00:07,855 INFO [train.py:901] (3/4) Epoch 27, batch 4650, loss[loss=0.2577, simple_loss=0.3209, pruned_loss=0.09727, over 6557.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2825, pruned_loss=0.05849, over 1607885.81 frames. ], batch size: 71, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 12:00:11,914 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.775e+02 2.452e+02 3.083e+02 3.974e+02 1.018e+03, threshold=6.165e+02, percent-clipped=5.0 +2023-02-07 12:00:43,657 INFO [train.py:901] (3/4) Epoch 27, batch 4700, loss[loss=0.2325, simple_loss=0.3106, pruned_loss=0.07719, over 7061.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2825, pruned_loss=0.05866, over 1610728.14 frames. ], batch size: 72, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 12:00:48,569 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=214862.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:01:05,463 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=214887.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:01:17,192 INFO [train.py:901] (3/4) Epoch 27, batch 4750, loss[loss=0.2439, simple_loss=0.3333, pruned_loss=0.07725, over 8483.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2837, pruned_loss=0.05879, over 1614422.18 frames. ], batch size: 25, lr: 2.79e-03, grad_scale: 16.0 +2023-02-07 12:01:21,119 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.799e+02 2.444e+02 3.016e+02 3.790e+02 1.117e+03, threshold=6.032e+02, percent-clipped=6.0 +2023-02-07 12:01:27,694 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.80 vs. limit=2.0 +2023-02-07 12:01:42,268 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-07 12:01:45,079 WARNING [train.py:1067] (3/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-07 12:01:49,027 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=214951.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:01:51,315 INFO [train.py:901] (3/4) Epoch 27, batch 4800, loss[loss=0.2046, simple_loss=0.306, pruned_loss=0.05162, over 8329.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2826, pruned_loss=0.05814, over 1611036.02 frames. ], batch size: 25, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:02:07,903 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=214976.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:02:27,032 INFO [train.py:901] (3/4) Epoch 27, batch 4850, loss[loss=0.2303, simple_loss=0.3089, pruned_loss=0.0758, over 8251.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2817, pruned_loss=0.058, over 1609038.50 frames. ], batch size: 24, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:02:31,208 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.607e+02 2.294e+02 2.705e+02 3.274e+02 6.085e+02, threshold=5.409e+02, percent-clipped=1.0 +2023-02-07 12:02:36,649 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-07 12:03:01,830 INFO [train.py:901] (3/4) Epoch 27, batch 4900, loss[loss=0.2149, simple_loss=0.3044, pruned_loss=0.06269, over 8247.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2811, pruned_loss=0.05721, over 1611595.67 frames. ], batch size: 24, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:03:07,092 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.45 vs. limit=5.0 +2023-02-07 12:03:34,914 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5989, 1.7658, 1.8956, 1.3147, 1.9500, 1.4345, 0.4845, 1.8549], + device='cuda:3'), covar=tensor([0.0622, 0.0401, 0.0352, 0.0585, 0.0444, 0.0992, 0.1017, 0.0317], + device='cuda:3'), in_proj_covar=tensor([0.0473, 0.0410, 0.0364, 0.0460, 0.0395, 0.0552, 0.0404, 0.0440], + device='cuda:3'), out_proj_covar=tensor([1.2553e-04, 1.0663e-04, 9.4976e-05, 1.2026e-04, 1.0346e-04, 1.5417e-04, + 1.0780e-04, 1.1539e-04], device='cuda:3') +2023-02-07 12:03:37,414 INFO [train.py:901] (3/4) Epoch 27, batch 4950, loss[loss=0.1853, simple_loss=0.2741, pruned_loss=0.04819, over 8476.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2815, pruned_loss=0.05746, over 1614913.96 frames. ], batch size: 26, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:03:41,324 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.461e+02 2.323e+02 2.858e+02 3.502e+02 9.819e+02, threshold=5.716e+02, percent-clipped=5.0 +2023-02-07 12:03:46,706 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8318, 1.4101, 4.0038, 1.4990, 3.4951, 3.3499, 3.6097, 3.5358], + device='cuda:3'), covar=tensor([0.0682, 0.4799, 0.0700, 0.4394, 0.1391, 0.1048, 0.0728, 0.0803], + device='cuda:3'), in_proj_covar=tensor([0.0680, 0.0675, 0.0745, 0.0662, 0.0755, 0.0643, 0.0643, 0.0723], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 12:04:10,562 INFO [train.py:901] (3/4) Epoch 27, batch 5000, loss[loss=0.2112, simple_loss=0.2987, pruned_loss=0.06184, over 8594.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2824, pruned_loss=0.0583, over 1613046.88 frames. ], batch size: 39, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:04:40,485 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.6172, 2.5312, 1.9040, 2.3056, 2.1988, 1.6469, 2.1035, 2.1803], + device='cuda:3'), covar=tensor([0.1696, 0.0472, 0.1261, 0.0674, 0.0833, 0.1648, 0.1104, 0.1053], + device='cuda:3'), in_proj_covar=tensor([0.0361, 0.0242, 0.0342, 0.0313, 0.0302, 0.0347, 0.0350, 0.0324], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 12:04:47,122 INFO [train.py:901] (3/4) Epoch 27, batch 5050, loss[loss=0.2576, simple_loss=0.3399, pruned_loss=0.08765, over 8315.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2834, pruned_loss=0.05819, over 1616717.56 frames. ], batch size: 26, lr: 2.78e-03, grad_scale: 32.0 +2023-02-07 12:04:50,091 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-07 12:04:50,998 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.808e+02 2.415e+02 2.920e+02 3.667e+02 5.760e+02, threshold=5.840e+02, percent-clipped=1.0 +2023-02-07 12:05:10,165 WARNING [train.py:1067] (3/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-07 12:05:15,553 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215248.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 12:05:20,048 INFO [train.py:901] (3/4) Epoch 27, batch 5100, loss[loss=0.2021, simple_loss=0.2923, pruned_loss=0.05599, over 8500.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2846, pruned_loss=0.05853, over 1620674.02 frames. ], batch size: 26, lr: 2.78e-03, grad_scale: 32.0 +2023-02-07 12:05:54,330 INFO [train.py:901] (3/4) Epoch 27, batch 5150, loss[loss=0.1723, simple_loss=0.2595, pruned_loss=0.04257, over 7662.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2835, pruned_loss=0.05772, over 1623462.25 frames. ], batch size: 19, lr: 2.78e-03, grad_scale: 32.0 +2023-02-07 12:05:59,259 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.789e+02 2.449e+02 2.868e+02 3.492e+02 6.640e+02, threshold=5.736e+02, percent-clipped=1.0 +2023-02-07 12:06:22,707 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215343.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 12:06:28,166 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215351.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:06:30,701 INFO [train.py:901] (3/4) Epoch 27, batch 5200, loss[loss=0.2197, simple_loss=0.2924, pruned_loss=0.07349, over 6791.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2826, pruned_loss=0.05761, over 1612692.80 frames. ], batch size: 15, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:06:58,936 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5580, 1.8535, 2.8897, 1.4454, 2.1458, 2.0027, 1.5556, 2.2442], + device='cuda:3'), covar=tensor([0.2015, 0.2901, 0.0870, 0.5014, 0.2045, 0.3421, 0.2635, 0.2304], + device='cuda:3'), in_proj_covar=tensor([0.0541, 0.0637, 0.0568, 0.0673, 0.0663, 0.0613, 0.0567, 0.0648], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 12:07:05,198 INFO [train.py:901] (3/4) Epoch 27, batch 5250, loss[loss=0.1612, simple_loss=0.2473, pruned_loss=0.03753, over 7797.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2826, pruned_loss=0.05744, over 1611685.41 frames. ], batch size: 19, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:07:09,826 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.650e+02 2.358e+02 2.790e+02 3.638e+02 8.125e+02, threshold=5.579e+02, percent-clipped=3.0 +2023-02-07 12:07:12,595 WARNING [train.py:1067] (3/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-07 12:07:17,503 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3961, 2.3083, 1.7756, 2.1466, 1.9153, 1.5138, 1.8683, 1.9199], + device='cuda:3'), covar=tensor([0.1541, 0.0461, 0.1248, 0.0571, 0.0770, 0.1634, 0.1002, 0.0988], + device='cuda:3'), in_proj_covar=tensor([0.0362, 0.0242, 0.0344, 0.0314, 0.0304, 0.0348, 0.0351, 0.0324], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 12:07:28,890 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.9810, 1.5807, 1.7576, 1.4804, 0.9627, 1.5794, 1.7506, 1.6051], + device='cuda:3'), covar=tensor([0.0547, 0.1222, 0.1592, 0.1433, 0.0571, 0.1446, 0.0676, 0.0664], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0153, 0.0188, 0.0161, 0.0101, 0.0162, 0.0112, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 12:07:40,340 INFO [train.py:901] (3/4) Epoch 27, batch 5300, loss[loss=0.2123, simple_loss=0.3113, pruned_loss=0.05663, over 8450.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.281, pruned_loss=0.0567, over 1613388.19 frames. ], batch size: 27, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:08:13,770 INFO [train.py:901] (3/4) Epoch 27, batch 5350, loss[loss=0.2278, simple_loss=0.3125, pruned_loss=0.07159, over 8607.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.282, pruned_loss=0.05727, over 1613945.73 frames. ], batch size: 34, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:08:18,669 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.346e+02 2.455e+02 2.847e+02 3.988e+02 1.267e+03, threshold=5.693e+02, percent-clipped=12.0 +2023-02-07 12:08:25,719 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215521.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:08:48,885 INFO [train.py:901] (3/4) Epoch 27, batch 5400, loss[loss=0.1607, simple_loss=0.2346, pruned_loss=0.04338, over 7411.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2826, pruned_loss=0.05813, over 1610315.17 frames. ], batch size: 17, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:08:52,411 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.3095, 2.1414, 1.7271, 2.0144, 1.7381, 1.4515, 1.6593, 1.7078], + device='cuda:3'), covar=tensor([0.1281, 0.0419, 0.1194, 0.0499, 0.0768, 0.1561, 0.0971, 0.0851], + device='cuda:3'), in_proj_covar=tensor([0.0361, 0.0241, 0.0343, 0.0312, 0.0303, 0.0347, 0.0350, 0.0323], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 12:08:57,735 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215566.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:09:08,912 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215583.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:09:14,846 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=215592.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:09:23,335 INFO [train.py:901] (3/4) Epoch 27, batch 5450, loss[loss=0.1791, simple_loss=0.269, pruned_loss=0.0446, over 8252.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2822, pruned_loss=0.05821, over 1608047.20 frames. ], batch size: 22, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:09:27,892 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.259e+02 2.540e+02 3.136e+02 3.819e+02 8.555e+02, threshold=6.272e+02, percent-clipped=5.0 +2023-02-07 12:09:47,134 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.8185, 1.5923, 1.7617, 1.4263, 0.9923, 1.5695, 1.6813, 1.6166], + device='cuda:3'), covar=tensor([0.0535, 0.1177, 0.1513, 0.1394, 0.0587, 0.1383, 0.0688, 0.0603], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0153, 0.0189, 0.0161, 0.0101, 0.0162, 0.0112, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 12:09:56,871 INFO [train.py:901] (3/4) Epoch 27, batch 5500, loss[loss=0.2027, simple_loss=0.2923, pruned_loss=0.05648, over 8549.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.283, pruned_loss=0.05825, over 1611332.14 frames. ], batch size: 31, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:09:56,881 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-07 12:10:20,607 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=215687.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:10:25,966 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=215695.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:10:32,544 INFO [train.py:901] (3/4) Epoch 27, batch 5550, loss[loss=0.1991, simple_loss=0.2799, pruned_loss=0.05913, over 8438.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.282, pruned_loss=0.05754, over 1613202.45 frames. ], batch size: 27, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:10:34,105 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=215707.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 12:10:37,242 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.599e+02 2.445e+02 2.973e+02 3.969e+02 8.778e+02, threshold=5.947e+02, percent-clipped=4.0 +2023-02-07 12:11:06,807 INFO [train.py:901] (3/4) Epoch 27, batch 5600, loss[loss=0.1766, simple_loss=0.2472, pruned_loss=0.053, over 7521.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2805, pruned_loss=0.0568, over 1605824.73 frames. ], batch size: 18, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:11:40,025 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=215802.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:11:41,824 INFO [train.py:901] (3/4) Epoch 27, batch 5650, loss[loss=0.2136, simple_loss=0.3024, pruned_loss=0.06241, over 8486.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2813, pruned_loss=0.05732, over 1609868.07 frames. ], batch size: 27, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:11:46,131 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=215810.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:11:47,203 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.511e+02 2.361e+02 2.799e+02 3.308e+02 5.877e+02, threshold=5.598e+02, percent-clipped=0.0 +2023-02-07 12:11:50,148 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([0.9567, 1.1992, 1.1486, 0.7620, 1.1762, 0.9627, 0.1445, 1.1226], + device='cuda:3'), covar=tensor([0.0675, 0.0502, 0.0514, 0.0761, 0.0637, 0.1417, 0.1211, 0.0445], + device='cuda:3'), in_proj_covar=tensor([0.0473, 0.0410, 0.0363, 0.0459, 0.0394, 0.0552, 0.0403, 0.0439], + device='cuda:3'), out_proj_covar=tensor([1.2555e-04, 1.0640e-04, 9.4739e-05, 1.2018e-04, 1.0317e-04, 1.5401e-04, + 1.0772e-04, 1.1511e-04], device='cuda:3') +2023-02-07 12:12:04,971 WARNING [train.py:1067] (3/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-07 12:12:11,223 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215847.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:12:16,648 INFO [train.py:901] (3/4) Epoch 27, batch 5700, loss[loss=0.2119, simple_loss=0.2983, pruned_loss=0.06275, over 8387.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2814, pruned_loss=0.05745, over 1609604.12 frames. ], batch size: 49, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:12:23,694 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=215865.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:12:51,896 INFO [train.py:901] (3/4) Epoch 27, batch 5750, loss[loss=0.2191, simple_loss=0.3066, pruned_loss=0.06581, over 8496.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2812, pruned_loss=0.05722, over 1606944.65 frames. ], batch size: 26, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:12:56,022 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=215910.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 12:12:57,182 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.493e+02 2.395e+02 2.899e+02 3.864e+02 7.116e+02, threshold=5.798e+02, percent-clipped=7.0 +2023-02-07 12:13:07,811 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=215927.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:13:10,334 WARNING [train.py:1067] (3/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-07 12:13:26,385 INFO [train.py:901] (3/4) Epoch 27, batch 5800, loss[loss=0.1929, simple_loss=0.2859, pruned_loss=0.04991, over 8504.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2811, pruned_loss=0.05772, over 1602170.49 frames. ], batch size: 28, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:13:31,953 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=215963.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 12:13:43,176 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=215980.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:13:47,606 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1227, 1.4452, 1.6251, 1.2780, 0.9871, 1.4343, 1.8033, 1.5849], + device='cuda:3'), covar=tensor([0.0513, 0.1308, 0.1765, 0.1589, 0.0610, 0.1595, 0.0692, 0.0699], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0153, 0.0189, 0.0161, 0.0101, 0.0163, 0.0112, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 12:13:48,928 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=215988.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:14:01,069 INFO [train.py:901] (3/4) Epoch 27, batch 5850, loss[loss=0.1612, simple_loss=0.2588, pruned_loss=0.03175, over 7958.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2815, pruned_loss=0.05726, over 1608148.49 frames. ], batch size: 21, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:14:05,651 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.506e+02 2.428e+02 2.871e+02 3.760e+02 7.078e+02, threshold=5.742e+02, percent-clipped=9.0 +2023-02-07 12:14:16,038 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216025.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 12:14:27,445 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216042.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:14:30,146 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216046.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:14:36,738 INFO [train.py:901] (3/4) Epoch 27, batch 5900, loss[loss=0.1891, simple_loss=0.2862, pruned_loss=0.04599, over 8337.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2809, pruned_loss=0.05692, over 1609189.32 frames. ], batch size: 26, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:14:38,998 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216058.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:14:44,145 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216066.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:14:55,477 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216083.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 12:15:00,979 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216091.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:15:10,148 INFO [train.py:901] (3/4) Epoch 27, batch 5950, loss[loss=0.2225, simple_loss=0.3094, pruned_loss=0.06785, over 8470.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2808, pruned_loss=0.05665, over 1609999.19 frames. ], batch size: 25, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:15:15,772 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.609e+02 2.411e+02 2.864e+02 3.625e+02 8.908e+02, threshold=5.728e+02, percent-clipped=5.0 +2023-02-07 12:15:46,883 INFO [train.py:901] (3/4) Epoch 27, batch 6000, loss[loss=0.1789, simple_loss=0.252, pruned_loss=0.05289, over 7530.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2815, pruned_loss=0.05691, over 1613742.83 frames. ], batch size: 18, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:15:46,883 INFO [train.py:926] (3/4) Computing validation loss +2023-02-07 12:15:59,961 INFO [train.py:935] (3/4) Epoch 27, validation: loss=0.1711, simple_loss=0.2711, pruned_loss=0.03554, over 944034.00 frames. +2023-02-07 12:15:59,962 INFO [train.py:936] (3/4) Maximum memory allocated so far is 6747MB +2023-02-07 12:16:06,493 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=2.98 vs. limit=5.0 +2023-02-07 12:16:25,739 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=216191.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:16:35,300 INFO [train.py:901] (3/4) Epoch 27, batch 6050, loss[loss=0.1893, simple_loss=0.2736, pruned_loss=0.05248, over 8084.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.281, pruned_loss=0.0569, over 1610702.12 frames. ], batch size: 21, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:16:40,116 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.642e+02 2.565e+02 3.207e+02 4.227e+02 9.285e+02, threshold=6.415e+02, percent-clipped=9.0 +2023-02-07 12:16:56,676 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216236.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:17:09,872 INFO [train.py:901] (3/4) Epoch 27, batch 6100, loss[loss=0.2124, simple_loss=0.2979, pruned_loss=0.06347, over 8410.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2825, pruned_loss=0.05758, over 1614560.44 frames. ], batch size: 49, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:17:14,011 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216261.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:17:28,451 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216281.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:17:39,620 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216298.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:17:44,976 INFO [train.py:901] (3/4) Epoch 27, batch 6150, loss[loss=0.1732, simple_loss=0.259, pruned_loss=0.04373, over 8460.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2828, pruned_loss=0.0578, over 1618981.18 frames. ], batch size: 49, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:17:44,988 WARNING [train.py:1067] (3/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-07 12:17:45,830 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216306.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:17:45,853 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216306.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 12:17:49,772 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.535e+02 2.311e+02 2.985e+02 4.036e+02 8.594e+02, threshold=5.970e+02, percent-clipped=2.0 +2023-02-07 12:17:52,545 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.3880, 4.3762, 3.9122, 1.9864, 3.8636, 4.0532, 3.9214, 3.8535], + device='cuda:3'), covar=tensor([0.0770, 0.0595, 0.1119, 0.4745, 0.0946, 0.1030, 0.1358, 0.0801], + device='cuda:3'), in_proj_covar=tensor([0.0549, 0.0469, 0.0451, 0.0564, 0.0449, 0.0471, 0.0447, 0.0411], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 12:17:57,198 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216323.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:18:18,525 INFO [train.py:901] (3/4) Epoch 27, batch 6200, loss[loss=0.1697, simple_loss=0.256, pruned_loss=0.0417, over 8029.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.283, pruned_loss=0.05757, over 1621599.44 frames. ], batch size: 22, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:18:42,476 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=216390.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:18:53,898 INFO [train.py:901] (3/4) Epoch 27, batch 6250, loss[loss=0.2272, simple_loss=0.2997, pruned_loss=0.07733, over 8589.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2841, pruned_loss=0.05816, over 1623607.44 frames. ], batch size: 31, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:18:58,453 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.643e+02 2.470e+02 2.901e+02 3.405e+02 7.374e+02, threshold=5.803e+02, percent-clipped=1.0 +2023-02-07 12:19:19,291 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([3.8517, 3.7931, 3.4254, 1.8337, 3.3462, 3.4229, 3.3493, 3.3294], + device='cuda:3'), covar=tensor([0.0838, 0.0658, 0.1152, 0.4817, 0.0997, 0.1113, 0.1400, 0.0955], + device='cuda:3'), in_proj_covar=tensor([0.0545, 0.0465, 0.0448, 0.0560, 0.0446, 0.0467, 0.0444, 0.0406], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 12:19:27,772 INFO [train.py:901] (3/4) Epoch 27, batch 6300, loss[loss=0.1766, simple_loss=0.259, pruned_loss=0.04712, over 8505.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2841, pruned_loss=0.05873, over 1620049.89 frames. ], batch size: 26, lr: 2.78e-03, grad_scale: 16.0 +2023-02-07 12:20:01,955 INFO [train.py:901] (3/4) Epoch 27, batch 6350, loss[loss=0.178, simple_loss=0.2548, pruned_loss=0.05061, over 7534.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2841, pruned_loss=0.05872, over 1618537.23 frames. ], batch size: 18, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:20:02,151 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216505.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:20:07,853 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.751e+02 2.511e+02 2.994e+02 4.018e+02 7.521e+02, threshold=5.987e+02, percent-clipped=5.0 +2023-02-07 12:20:36,668 INFO [train.py:901] (3/4) Epoch 27, batch 6400, loss[loss=0.1849, simple_loss=0.2689, pruned_loss=0.05051, over 6378.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2847, pruned_loss=0.05903, over 1613794.71 frames. ], batch size: 14, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:20:41,615 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216562.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:20:53,533 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([0.9590, 1.6492, 1.3751, 1.5475, 1.2591, 1.2713, 1.2067, 1.2791], + device='cuda:3'), covar=tensor([0.1318, 0.0558, 0.1453, 0.0650, 0.0935, 0.1694, 0.1108, 0.0902], + device='cuda:3'), in_proj_covar=tensor([0.0361, 0.0241, 0.0342, 0.0312, 0.0303, 0.0347, 0.0350, 0.0323], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 12:20:58,186 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216587.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:21:10,669 INFO [train.py:901] (3/4) Epoch 27, batch 6450, loss[loss=0.1794, simple_loss=0.2585, pruned_loss=0.05017, over 7921.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.285, pruned_loss=0.05883, over 1615244.60 frames. ], batch size: 20, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:21:13,549 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216609.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:21:16,161 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.641e+02 2.469e+02 2.882e+02 3.609e+02 7.919e+02, threshold=5.765e+02, percent-clipped=2.0 +2023-02-07 12:21:46,025 INFO [train.py:901] (3/4) Epoch 27, batch 6500, loss[loss=0.1985, simple_loss=0.2867, pruned_loss=0.05509, over 8808.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2846, pruned_loss=0.05873, over 1614233.49 frames. ], batch size: 40, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:21:46,207 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.1387, 1.3504, 1.5677, 1.2887, 0.7877, 1.3331, 1.1612, 1.0580], + device='cuda:3'), covar=tensor([0.0685, 0.1257, 0.1636, 0.1448, 0.0582, 0.1482, 0.0741, 0.0710], + device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0153, 0.0190, 0.0161, 0.0101, 0.0163, 0.0112, 0.0145], + device='cuda:3'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:3') +2023-02-07 12:21:58,505 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216673.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:22:14,731 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.4293, 1.8358, 1.4141, 2.8180, 1.3120, 1.2585, 2.0727, 1.9728], + device='cuda:3'), covar=tensor([0.1540, 0.1319, 0.1895, 0.0387, 0.1329, 0.2177, 0.0907, 0.0940], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0193, 0.0244, 0.0212, 0.0203, 0.0245, 0.0249, 0.0203], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 12:22:19,974 INFO [train.py:901] (3/4) Epoch 27, batch 6550, loss[loss=0.1362, simple_loss=0.2185, pruned_loss=0.027, over 7258.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2843, pruned_loss=0.05873, over 1616225.52 frames. ], batch size: 16, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:22:25,119 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.409e+02 2.544e+02 2.876e+02 3.743e+02 6.730e+02, threshold=5.752e+02, percent-clipped=5.0 +2023-02-07 12:22:32,752 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216723.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:22:54,580 WARNING [train.py:1067] (3/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-07 12:22:55,723 INFO [train.py:901] (3/4) Epoch 27, batch 6600, loss[loss=0.2434, simple_loss=0.2955, pruned_loss=0.09563, over 7677.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2839, pruned_loss=0.05894, over 1614970.20 frames. ], batch size: 18, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:22:59,997 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216761.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:23:09,884 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([0.8831, 1.2137, 0.9669, 1.2068, 0.9797, 0.8958, 1.0413, 1.0853], + device='cuda:3'), covar=tensor([0.0848, 0.0407, 0.1079, 0.0444, 0.0656, 0.1250, 0.0671, 0.0544], + device='cuda:3'), in_proj_covar=tensor([0.0362, 0.0241, 0.0342, 0.0313, 0.0304, 0.0347, 0.0350, 0.0323], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 12:23:12,957 WARNING [train.py:1067] (3/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-07 12:23:17,022 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216786.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:23:29,549 INFO [train.py:901] (3/4) Epoch 27, batch 6650, loss[loss=0.1918, simple_loss=0.2799, pruned_loss=0.05186, over 8317.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.283, pruned_loss=0.05911, over 1609873.69 frames. ], batch size: 26, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:23:34,797 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.735e+02 2.571e+02 3.099e+02 3.859e+02 9.745e+02, threshold=6.199e+02, percent-clipped=7.0 +2023-02-07 12:24:03,769 INFO [train.py:901] (3/4) Epoch 27, batch 6700, loss[loss=0.1967, simple_loss=0.2869, pruned_loss=0.05327, over 8503.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2841, pruned_loss=0.05956, over 1607513.59 frames. ], batch size: 28, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:24:32,231 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.3137, 2.0228, 2.5799, 2.2238, 2.5376, 2.3550, 2.2214, 1.4322], + device='cuda:3'), covar=tensor([0.5587, 0.5161, 0.2101, 0.3763, 0.2420, 0.3323, 0.1956, 0.5318], + device='cuda:3'), in_proj_covar=tensor([0.0966, 0.1021, 0.0834, 0.0992, 0.1031, 0.0931, 0.0773, 0.0853], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 12:24:38,620 INFO [train.py:901] (3/4) Epoch 27, batch 6750, loss[loss=0.217, simple_loss=0.3058, pruned_loss=0.06406, over 8179.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2839, pruned_loss=0.05923, over 1609562.39 frames. ], batch size: 23, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:24:40,694 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([4.5953, 4.6121, 4.1354, 2.1224, 4.0814, 4.1831, 4.1770, 4.0567], + device='cuda:3'), covar=tensor([0.0663, 0.0459, 0.0939, 0.4524, 0.0835, 0.0963, 0.1146, 0.0736], + device='cuda:3'), in_proj_covar=tensor([0.0542, 0.0464, 0.0445, 0.0556, 0.0442, 0.0464, 0.0439, 0.0403], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 12:24:43,892 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.585e+02 2.479e+02 3.006e+02 3.687e+02 6.813e+02, threshold=6.012e+02, percent-clipped=1.0 +2023-02-07 12:24:49,475 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1488, 3.7295, 2.4318, 2.9863, 2.8886, 2.1408, 2.8964, 3.0408], + device='cuda:3'), covar=tensor([0.1689, 0.0355, 0.1099, 0.0728, 0.0777, 0.1408, 0.0961, 0.1081], + device='cuda:3'), in_proj_covar=tensor([0.0363, 0.0243, 0.0343, 0.0314, 0.0306, 0.0348, 0.0352, 0.0325], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 12:24:59,756 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 +2023-02-07 12:25:11,172 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=216953.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:25:12,358 INFO [train.py:901] (3/4) Epoch 27, batch 6800, loss[loss=0.2211, simple_loss=0.3068, pruned_loss=0.06771, over 8636.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2845, pruned_loss=0.0594, over 1607727.33 frames. ], batch size: 34, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:25:24,113 WARNING [train.py:1067] (3/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-07 12:25:32,329 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216983.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:25:47,374 INFO [train.py:901] (3/4) Epoch 27, batch 6850, loss[loss=0.2285, simple_loss=0.3151, pruned_loss=0.07097, over 8553.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2836, pruned_loss=0.05885, over 1607661.55 frames. ], batch size: 49, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:25:52,583 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.618e+02 2.408e+02 3.097e+02 3.751e+02 9.876e+02, threshold=6.193e+02, percent-clipped=4.0 +2023-02-07 12:25:55,366 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=217017.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:26:10,970 WARNING [train.py:1067] (3/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-07 12:26:21,077 INFO [train.py:901] (3/4) Epoch 27, batch 6900, loss[loss=0.1819, simple_loss=0.2628, pruned_loss=0.05047, over 7532.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2826, pruned_loss=0.05862, over 1607688.95 frames. ], batch size: 18, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:26:29,754 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=217067.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:26:30,512 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217068.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:26:56,822 INFO [train.py:901] (3/4) Epoch 27, batch 6950, loss[loss=0.2269, simple_loss=0.3005, pruned_loss=0.07669, over 8630.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2836, pruned_loss=0.05893, over 1610068.09 frames. ], batch size: 34, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:27:02,041 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.613e+02 2.303e+02 2.670e+02 3.410e+02 6.861e+02, threshold=5.340e+02, percent-clipped=1.0 +2023-02-07 12:27:12,678 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=256, metric=4.05 vs. limit=5.0 +2023-02-07 12:27:15,674 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217132.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:27:19,477 WARNING [train.py:1067] (3/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-07 12:27:30,834 INFO [train.py:901] (3/4) Epoch 27, batch 7000, loss[loss=0.2152, simple_loss=0.3005, pruned_loss=0.06501, over 8495.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2834, pruned_loss=0.05849, over 1615309.87 frames. ], batch size: 26, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:27:47,592 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.5094, 1.6136, 1.6512, 1.2173, 1.7276, 1.3524, 0.6669, 1.6291], + device='cuda:3'), covar=tensor([0.0505, 0.0344, 0.0264, 0.0503, 0.0328, 0.0717, 0.0823, 0.0274], + device='cuda:3'), in_proj_covar=tensor([0.0471, 0.0407, 0.0362, 0.0457, 0.0392, 0.0550, 0.0401, 0.0438], + device='cuda:3'), out_proj_covar=tensor([1.2482e-04, 1.0563e-04, 9.4482e-05, 1.1964e-04, 1.0279e-04, 1.5348e-04, + 1.0697e-04, 1.1495e-04], device='cuda:3') +2023-02-07 12:27:49,482 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217182.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:28:05,127 INFO [train.py:901] (3/4) Epoch 27, batch 7050, loss[loss=0.1998, simple_loss=0.2932, pruned_loss=0.0532, over 8344.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2848, pruned_loss=0.059, over 1617040.15 frames. ], batch size: 26, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:28:11,277 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.614e+02 2.356e+02 3.046e+02 3.591e+02 8.726e+02, threshold=6.092e+02, percent-clipped=6.0 +2023-02-07 12:28:20,121 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([5.9743, 1.6608, 6.1321, 2.1862, 5.5409, 5.1090, 5.6851, 5.5327], + device='cuda:3'), covar=tensor([0.0527, 0.4638, 0.0352, 0.3880, 0.0942, 0.0851, 0.0442, 0.0497], + device='cuda:3'), in_proj_covar=tensor([0.0674, 0.0662, 0.0731, 0.0652, 0.0741, 0.0627, 0.0634, 0.0714], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:3') +2023-02-07 12:28:40,055 INFO [train.py:901] (3/4) Epoch 27, batch 7100, loss[loss=0.1753, simple_loss=0.264, pruned_loss=0.04327, over 8569.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2835, pruned_loss=0.05809, over 1614682.59 frames. ], batch size: 31, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:29:14,482 INFO [train.py:901] (3/4) Epoch 27, batch 7150, loss[loss=0.1724, simple_loss=0.2558, pruned_loss=0.04455, over 8087.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2833, pruned_loss=0.05838, over 1612884.66 frames. ], batch size: 21, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:29:19,657 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.328e+02 2.444e+02 3.123e+02 4.113e+02 1.134e+03, threshold=6.246e+02, percent-clipped=7.0 +2023-02-07 12:29:27,896 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=217324.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:29:29,711 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=217327.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:29:45,935 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=217349.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:29:49,810 INFO [train.py:901] (3/4) Epoch 27, batch 7200, loss[loss=0.1976, simple_loss=0.2871, pruned_loss=0.05407, over 8463.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2839, pruned_loss=0.05814, over 1613830.14 frames. ], batch size: 25, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:30:04,008 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.2343, 2.0307, 2.6297, 2.2235, 2.6579, 2.3151, 2.1566, 1.6344], + device='cuda:3'), covar=tensor([0.5805, 0.5260, 0.2150, 0.3763, 0.2622, 0.3194, 0.1901, 0.5523], + device='cuda:3'), in_proj_covar=tensor([0.0965, 0.1022, 0.0832, 0.0992, 0.1030, 0.0928, 0.0770, 0.0851], + device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:3') +2023-02-07 12:30:11,975 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=217388.0, num_to_drop=1, layers_to_drop={0} +2023-02-07 12:30:23,037 INFO [train.py:901] (3/4) Epoch 27, batch 7250, loss[loss=0.2028, simple_loss=0.2711, pruned_loss=0.06727, over 7209.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2837, pruned_loss=0.05766, over 1614357.20 frames. ], batch size: 16, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:30:23,876 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=217406.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:30:28,402 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.659e+02 2.296e+02 2.784e+02 3.610e+02 7.832e+02, threshold=5.568e+02, percent-clipped=2.0 +2023-02-07 12:30:28,621 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=217413.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:30:45,931 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=217438.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:30:49,266 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217442.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:30:58,443 INFO [train.py:901] (3/4) Epoch 27, batch 7300, loss[loss=0.1618, simple_loss=0.2469, pruned_loss=0.0384, over 7812.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.282, pruned_loss=0.05677, over 1613788.05 frames. ], batch size: 20, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:31:04,059 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=217463.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:31:33,103 INFO [train.py:901] (3/4) Epoch 27, batch 7350, loss[loss=0.1907, simple_loss=0.269, pruned_loss=0.05624, over 8026.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2809, pruned_loss=0.05629, over 1613767.95 frames. ], batch size: 22, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:31:36,623 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=217510.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:31:38,502 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.663e+02 2.322e+02 2.888e+02 3.768e+02 6.651e+02, threshold=5.777e+02, percent-clipped=4.0 +2023-02-07 12:31:59,842 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-07 12:32:07,068 INFO [train.py:901] (3/4) Epoch 27, batch 7400, loss[loss=0.1931, simple_loss=0.2871, pruned_loss=0.04955, over 8585.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2825, pruned_loss=0.05693, over 1616582.70 frames. ], batch size: 34, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:32:19,126 WARNING [train.py:1067] (3/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-07 12:32:42,456 INFO [train.py:901] (3/4) Epoch 27, batch 7450, loss[loss=0.1823, simple_loss=0.2657, pruned_loss=0.04944, over 7808.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2832, pruned_loss=0.05756, over 1618707.06 frames. ], batch size: 20, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:32:43,318 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([2.1086, 1.8639, 3.5409, 1.8013, 2.7502, 3.8456, 3.9512, 3.3388], + device='cuda:3'), covar=tensor([0.1166, 0.1654, 0.0292, 0.1943, 0.0816, 0.0223, 0.0645, 0.0546], + device='cuda:3'), in_proj_covar=tensor([0.0306, 0.0322, 0.0291, 0.0318, 0.0320, 0.0276, 0.0437, 0.0306], + device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:3') +2023-02-07 12:32:47,778 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.771e+02 2.478e+02 3.262e+02 4.062e+02 8.102e+02, threshold=6.523e+02, percent-clipped=5.0 +2023-02-07 12:32:58,353 WARNING [train.py:1067] (3/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-07 12:33:05,996 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 +2023-02-07 12:33:16,130 INFO [train.py:901] (3/4) Epoch 27, batch 7500, loss[loss=0.215, simple_loss=0.2924, pruned_loss=0.0688, over 7809.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2827, pruned_loss=0.05742, over 1617179.72 frames. ], batch size: 20, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:33:34,977 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=217682.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:33:46,542 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=217698.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:33:51,442 INFO [train.py:901] (3/4) Epoch 27, batch 7550, loss[loss=0.2277, simple_loss=0.3017, pruned_loss=0.07688, over 8118.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2824, pruned_loss=0.05747, over 1617145.70 frames. ], batch size: 22, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:33:56,749 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.627e+02 2.428e+02 3.024e+02 3.911e+02 8.560e+02, threshold=6.047e+02, percent-clipped=1.0 +2023-02-07 12:34:03,671 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=217723.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:34:21,847 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=217750.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:34:25,190 INFO [train.py:901] (3/4) Epoch 27, batch 7600, loss[loss=0.1513, simple_loss=0.2374, pruned_loss=0.03262, over 8240.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2817, pruned_loss=0.05721, over 1617586.67 frames. ], batch size: 22, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:35:01,502 INFO [train.py:901] (3/4) Epoch 27, batch 7650, loss[loss=0.1855, simple_loss=0.265, pruned_loss=0.05298, over 7540.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2811, pruned_loss=0.05705, over 1618436.06 frames. ], batch size: 18, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:35:06,799 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.640e+02 2.541e+02 2.896e+02 3.920e+02 6.720e+02, threshold=5.793e+02, percent-clipped=4.0 +2023-02-07 12:35:35,074 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=217854.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:35:35,687 INFO [train.py:901] (3/4) Epoch 27, batch 7700, loss[loss=0.1765, simple_loss=0.274, pruned_loss=0.03948, over 8033.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2813, pruned_loss=0.05717, over 1613834.22 frames. ], batch size: 22, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:35:42,371 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217865.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:35:54,057 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.52 vs. limit=2.0 +2023-02-07 12:36:05,141 WARNING [train.py:1067] (3/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-07 12:36:10,562 INFO [train.py:901] (3/4) Epoch 27, batch 7750, loss[loss=0.1629, simple_loss=0.2342, pruned_loss=0.04581, over 7540.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2813, pruned_loss=0.057, over 1610421.82 frames. ], batch size: 18, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:36:15,959 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.659e+02 2.515e+02 3.033e+02 3.634e+02 8.452e+02, threshold=6.066e+02, percent-clipped=4.0 +2023-02-07 12:36:18,822 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=217916.0, num_to_drop=1, layers_to_drop={1} +2023-02-07 12:36:45,564 INFO [train.py:901] (3/4) Epoch 27, batch 7800, loss[loss=0.2362, simple_loss=0.3003, pruned_loss=0.08605, over 7661.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2817, pruned_loss=0.05723, over 1607444.12 frames. ], batch size: 19, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:36:55,074 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217969.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:37:08,180 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7386, 2.7129, 1.8797, 2.3569, 2.2935, 1.7560, 2.2425, 2.4249], + device='cuda:3'), covar=tensor([0.1510, 0.0367, 0.1259, 0.0645, 0.0737, 0.1487, 0.0964, 0.1028], + device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0241, 0.0341, 0.0312, 0.0303, 0.0346, 0.0349, 0.0322], + device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:3') +2023-02-07 12:37:17,817 INFO [zipformer.py:2431] (3/4) attn_weights_entropy = tensor([1.7202, 1.7929, 1.6191, 2.3440, 1.0346, 1.4660, 1.7044, 1.8479], + device='cuda:3'), covar=tensor([0.0748, 0.0865, 0.0933, 0.0389, 0.1147, 0.1317, 0.0784, 0.0707], + device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0193, 0.0245, 0.0212, 0.0202, 0.0245, 0.0249, 0.0203], + device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:3') +2023-02-07 12:37:19,652 INFO [train.py:901] (3/4) Epoch 27, batch 7850, loss[loss=0.157, simple_loss=0.2344, pruned_loss=0.03982, over 7233.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.281, pruned_loss=0.05672, over 1607754.84 frames. ], batch size: 16, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:37:24,965 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.749e+02 2.277e+02 2.828e+02 3.912e+02 8.712e+02, threshold=5.655e+02, percent-clipped=7.0 +2023-02-07 12:37:33,523 INFO [zipformer.py:1185] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=218026.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:37:52,846 INFO [train.py:901] (3/4) Epoch 27, batch 7900, loss[loss=0.206, simple_loss=0.2966, pruned_loss=0.05765, over 8513.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2822, pruned_loss=0.05795, over 1606329.14 frames. ], batch size: 28, lr: 2.77e-03, grad_scale: 8.0 +2023-02-07 12:38:23,930 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-07 12:38:26,198 INFO [train.py:901] (3/4) Epoch 27, batch 7950, loss[loss=0.2413, simple_loss=0.323, pruned_loss=0.07981, over 8511.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2833, pruned_loss=0.05848, over 1606962.93 frames. ], batch size: 28, lr: 2.76e-03, grad_scale: 8.0 +2023-02-07 12:38:28,874 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 +2023-02-07 12:38:31,693 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.382e+02 2.553e+02 3.230e+02 4.059e+02 8.354e+02, threshold=6.459e+02, percent-clipped=5.0 +2023-02-07 12:38:35,241 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218118.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:38:37,398 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218121.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:38:50,521 INFO [zipformer.py:1185] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218141.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:38:53,607 INFO [zipformer.py:1185] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218146.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:38:59,458 INFO [train.py:901] (3/4) Epoch 27, batch 8000, loss[loss=0.2205, simple_loss=0.3045, pruned_loss=0.06829, over 8188.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2828, pruned_loss=0.05848, over 1605455.68 frames. ], batch size: 23, lr: 2.76e-03, grad_scale: 8.0 +2023-02-07 12:39:29,225 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218200.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:39:32,351 INFO [train.py:901] (3/4) Epoch 27, batch 8050, loss[loss=0.1665, simple_loss=0.2599, pruned_loss=0.03658, over 7917.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2819, pruned_loss=0.05795, over 1599146.51 frames. ], batch size: 20, lr: 2.76e-03, grad_scale: 8.0 +2023-02-07 12:39:38,070 INFO [optim.py:369] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.565e+02 2.285e+02 2.948e+02 3.498e+02 7.136e+02, threshold=5.897e+02, percent-clipped=2.0 +2023-02-07 12:39:46,279 INFO [zipformer.py:1185] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218225.0, num_to_drop=0, layers_to_drop=set() +2023-02-07 12:39:48,253 INFO [zipformer.py:1185] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218228.0, num_to_drop=0, layers_to_drop=set()