2023-03-25 21:42:34,511 INFO [finetune.py:1046] (3/7) Training started 2023-03-25 21:42:34,511 INFO [finetune.py:1056] (3/7) Device: cuda:3 2023-03-25 21:42:34,514 INFO [finetune.py:1065] (3/7) {'frame_shift_ms': 10.0, 'allowed_excess_duration_ratio': 0.1, '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': '1.23.4', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '62e404dd3f3a811d73e424199b3408e309c06e1a', 'k2-git-date': 'Mon Jan 30 02:26:16 2023', 'lhotse-version': '1.12.0.dev+git.3ccfeb7.clean', 'torch-version': '1.13.0', 'torch-cuda-available': True, 'torch-cuda-version': '11.7', 'python-version': '3.8', 'icefall-git-branch': 'master', 'icefall-git-sha1': 'd74822d-dirty', 'icefall-git-date': 'Tue Mar 21 21:35:32 2023', 'icefall-path': '/home/lishaojie/icefall', 'k2-path': '/home/lishaojie/.conda/envs/env_lishaojie/lib/python3.8/site-packages/k2/__init__.py', 'lhotse-path': '/home/lishaojie/.conda/envs/env_lishaojie/lib/python3.8/site-packages/lhotse/__init__.py', 'hostname': 'cnc533', 'IP address': '127.0.1.1'}, 'world_size': 7, 'master_port': 18181, 'tensorboard': True, 'num_epochs': 30, 'start_epoch': 1, 'start_batch': 0, 'exp_dir': PosixPath('pruned_transducer_stateless7_streaming/exp1'), 'bpe_model': 'data/lang_bpe_500/bpe.model', 'base_lr': 0.004, 'lr_batches': 100000.0, 'lr_epochs': 100.0, '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': 30, 'average_period': 200, 'use_fp16': True, 'num_encoder_layers': '2,4,3,2,4', 'feedforward_dims': '1024,1024,2048,2048,1024', 'nhead': '8,8,8,8,8', 'encoder_dims': '384,384,384,384,384', 'attention_dims': '192,192,192,192,192', 'encoder_unmasked_dims': '256,256,256,256,256', 'zipformer_downsampling_factors': '1,2,4,8,2', 'cnn_module_kernels': '31,31,31,31,31', 'decoder_dim': 512, 'joiner_dim': 512, 'do_finetune': True, 'init_modules': 'encoder', 'finetune_ckpt': '/home/lishaojie/icefall/egs/commonvoice/ASR/pruned_transducer_stateless7_streaming/exp/english_pretrain/epoch-30.pt', 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 200, '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} 2023-03-25 21:42:34,514 INFO [finetune.py:1067] (3/7) About to create model 2023-03-25 21:42:34,852 INFO [zipformer.py:405] (3/7) 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. 2023-03-25 21:42:34,861 INFO [finetune.py:1071] (3/7) Number of model parameters: 70369391 2023-03-25 21:42:34,861 INFO [finetune.py:626] (3/7) Loading checkpoint from /home/lishaojie/icefall/egs/commonvoice/ASR/pruned_transducer_stateless7_streaming/exp/english_pretrain/epoch-30.pt 2023-03-25 21:42:35,456 INFO [finetune.py:647] (3/7) Loading parameters starting with prefix encoder 2023-03-25 21:42:36,906 INFO [finetune.py:1093] (3/7) Using DDP 2023-03-25 21:42:37,687 INFO [commonvoice_fr.py:392] (3/7) About to get train cuts 2023-03-25 21:42:37,690 INFO [commonvoice_fr.py:218] (3/7) Enable MUSAN 2023-03-25 21:42:37,690 INFO [commonvoice_fr.py:219] (3/7) About to get Musan cuts 2023-03-25 21:42:39,201 INFO [commonvoice_fr.py:243] (3/7) Enable SpecAugment 2023-03-25 21:42:39,202 INFO [commonvoice_fr.py:244] (3/7) Time warp factor: 80 2023-03-25 21:42:39,202 INFO [commonvoice_fr.py:254] (3/7) Num frame mask: 10 2023-03-25 21:42:39,202 INFO [commonvoice_fr.py:267] (3/7) About to create train dataset 2023-03-25 21:42:39,202 INFO [commonvoice_fr.py:294] (3/7) Using DynamicBucketingSampler. 2023-03-25 21:42:41,924 INFO [commonvoice_fr.py:309] (3/7) About to create train dataloader 2023-03-25 21:42:41,925 INFO [commonvoice_fr.py:399] (3/7) About to get dev cuts 2023-03-25 21:42:41,926 INFO [commonvoice_fr.py:340] (3/7) About to create dev dataset 2023-03-25 21:42:42,333 INFO [commonvoice_fr.py:357] (3/7) About to create dev dataloader 2023-03-25 21:42:42,334 INFO [finetune.py:1289] (3/7) Sanity check -- see if any of the batches in epoch 1 would cause OOM. 2023-03-25 21:46:46,135 INFO [finetune.py:1317] (3/7) Maximum memory allocated so far is 5236MB 2023-03-25 21:46:46,827 INFO [finetune.py:1317] (3/7) Maximum memory allocated so far is 5768MB 2023-03-25 21:46:48,913 INFO [finetune.py:1317] (3/7) Maximum memory allocated so far is 5768MB 2023-03-25 21:46:49,576 INFO [finetune.py:1317] (3/7) Maximum memory allocated so far is 5768MB 2023-03-25 21:46:50,267 INFO [finetune.py:1317] (3/7) Maximum memory allocated so far is 5768MB 2023-03-25 21:46:50,962 INFO [finetune.py:1317] (3/7) Maximum memory allocated so far is 5768MB 2023-03-25 21:46:59,843 INFO [finetune.py:976] (3/7) Epoch 1, batch 0, loss[loss=7.5, simple_loss=6.804, pruned_loss=6.95, over 4703.00 frames. ], tot_loss[loss=7.5, simple_loss=6.804, pruned_loss=6.95, over 4703.00 frames. ], batch size: 23, lr: 2.00e-03, grad_scale: 2.0 2023-03-25 21:46:59,843 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-25 21:47:14,953 INFO [finetune.py:1010] (3/7) Epoch 1, validation: loss=7.294, simple_loss=6.606, pruned_loss=6.863, over 2265189.00 frames. 2023-03-25 21:47:14,954 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 5768MB 2023-03-25 21:47:19,869 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5.0, num_to_drop=2, layers_to_drop={0, 1} 2023-03-25 21:47:27,259 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=2.48 vs. limit=2.0 2023-03-25 21:47:30,298 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=23.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 21:48:00,213 INFO [finetune.py:976] (3/7) Epoch 1, batch 50, loss[loss=2.655, simple_loss=2.5, pruned_loss=1.522, over 4927.00 frames. ], tot_loss[loss=4.173, simple_loss=3.743, pruned_loss=4.119, over 213004.45 frames. ], batch size: 33, lr: 2.20e-03, grad_scale: 0.000244140625 2023-03-25 21:48:22,470 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.0878, 0.9618, 1.2725, 0.7623, 1.3099, 1.3273, 1.2902, 1.2980], device='cuda:3'), covar=tensor([0.0396, 0.0355, 0.0196, 0.0319, 0.0231, 0.0220, 0.0291, 0.0354], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0176, 0.0133, 0.0143, 0.0148, 0.0145, 0.0168, 0.0183], device='cuda:3'), out_proj_covar=tensor([1.1290e-04, 1.3145e-04, 9.7238e-05, 1.0429e-04, 1.0792e-04, 1.0852e-04, 1.2638e-04, 1.3710e-04], device='cuda:3') 2023-03-25 21:48:33,001 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 21:48:53,453 WARNING [finetune.py:966] (3/7) Grad scale is small: 0.000244140625 2023-03-25 21:48:53,453 INFO [finetune.py:976] (3/7) Epoch 1, batch 100, loss[loss=2.328, simple_loss=2.194, pruned_loss=1.3, over 4887.00 frames. ], tot_loss[loss=3.378, simple_loss=3.11, pruned_loss=2.607, over 378676.73 frames. ], batch size: 32, lr: 2.40e-03, grad_scale: 0.00048828125 2023-03-25 21:49:13,206 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 7.539e+02 2.791e+03 6.484e+03 1.700e+04 1.722e+07, threshold=1.297e+04, percent-clipped=0.0 2023-03-25 21:49:28,881 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=144.0, num_to_drop=2, layers_to_drop={2, 3} 2023-03-25 21:49:35,384 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3736, 1.7394, 1.5587, 1.7218, 3.0017, 2.2546, 1.5423, 1.4416], device='cuda:3'), covar=tensor([0.0260, 0.0227, 0.0189, 0.0193, 0.0158, 0.0134, 0.0266, 0.0177], device='cuda:3'), in_proj_covar=tensor([0.0257, 0.0239, 0.0229, 0.0209, 0.0272, 0.0204, 0.0235, 0.0209], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-25 21:49:37,440 INFO [finetune.py:976] (3/7) Epoch 1, batch 150, loss[loss=1.587, simple_loss=1.439, pruned_loss=1.206, over 4869.00 frames. ], tot_loss[loss=2.783, simple_loss=2.57, pruned_loss=2.032, over 505616.32 frames. ], batch size: 31, lr: 2.60e-03, grad_scale: 0.00048828125 2023-03-25 21:50:12,948 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2269, 1.2346, 0.9980, 1.0688, 1.3302, 2.2212, 1.1926, 1.5308], device='cuda:3'), covar=tensor([0.1199, 0.1382, 0.2155, 0.1612, 0.1782, 0.1486, 0.1395, 0.1006], device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0081, 0.0071, 0.0073, 0.0090, 0.0076, 0.0084, 0.0076], 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') 2023-03-25 21:50:15,717 WARNING [finetune.py:966] (3/7) Grad scale is small: 0.00048828125 2023-03-25 21:50:15,717 INFO [finetune.py:976] (3/7) Epoch 1, batch 200, loss[loss=1.461, simple_loss=1.266, pruned_loss=1.356, over 4815.00 frames. ], tot_loss[loss=2.309, simple_loss=2.111, pruned_loss=1.76, over 607044.57 frames. ], batch size: 51, lr: 2.80e-03, grad_scale: 0.0009765625 2023-03-25 21:50:29,331 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 2.018e+02 7.406e+02 1.293e+03 3.197e+03 6.754e+04, threshold=2.586e+03, percent-clipped=12.0 2023-03-25 21:50:54,579 INFO [finetune.py:976] (3/7) Epoch 1, batch 250, loss[loss=1.532, simple_loss=1.309, pruned_loss=1.43, over 4839.00 frames. ], tot_loss[loss=2.011, simple_loss=1.817, pruned_loss=1.603, over 683643.92 frames. ], batch size: 47, lr: 3.00e-03, grad_scale: 0.0009765625 2023-03-25 21:51:43,795 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 21:51:44,337 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1014, 1.9886, 1.4053, 3.0014, 1.8777, 1.7905, 2.9018, 2.1842], device='cuda:3'), covar=tensor([0.0121, 0.0151, 0.0166, 0.0159, 0.0149, 0.0148, 0.0137, 0.0131], device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0222, 0.0262, 0.0295, 0.0247, 0.0200, 0.0211, 0.0215], 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') 2023-03-25 21:51:45,810 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300.0, num_to_drop=2, layers_to_drop={0, 1} 2023-03-25 21:51:46,257 WARNING [finetune.py:966] (3/7) Grad scale is small: 0.0009765625 2023-03-25 21:51:46,257 INFO [finetune.py:976] (3/7) Epoch 1, batch 300, loss[loss=1.57, simple_loss=1.319, pruned_loss=1.487, over 4815.00 frames. ], tot_loss[loss=1.808, simple_loss=1.613, pruned_loss=1.493, over 742631.54 frames. ], batch size: 41, lr: 3.20e-03, grad_scale: 0.001953125 2023-03-25 21:51:58,579 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 2.075e+01 5.781e+01 1.827e+02 5.788e+02 1.230e+04, threshold=3.655e+02, percent-clipped=4.0 2023-03-25 21:52:16,702 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.32 vs. limit=2.0 2023-03-25 21:52:39,148 INFO [finetune.py:976] (3/7) Epoch 1, batch 350, loss[loss=1.282, simple_loss=1.067, pruned_loss=1.199, over 4784.00 frames. ], tot_loss[loss=1.671, simple_loss=1.471, pruned_loss=1.42, over 789302.13 frames. ], batch size: 51, lr: 3.40e-03, grad_scale: 0.001953125 2023-03-25 21:52:47,118 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=357.0, num_to_drop=2, layers_to_drop={2, 3} 2023-03-25 21:52:47,167 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=2.72 vs. limit=2.0 2023-03-25 21:53:10,171 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 21:53:27,702 WARNING [finetune.py:966] (3/7) Grad scale is small: 0.001953125 2023-03-25 21:53:27,702 INFO [finetune.py:976] (3/7) Epoch 1, batch 400, loss[loss=1.236, simple_loss=1.006, pruned_loss=1.18, over 4905.00 frames. ], tot_loss[loss=1.566, simple_loss=1.36, pruned_loss=1.36, over 825904.20 frames. ], batch size: 37, lr: 3.60e-03, grad_scale: 0.00390625 2023-03-25 21:53:39,885 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.702e+01 2.277e+01 3.517e+01 1.113e+02 1.032e+03, threshold=7.035e+01, percent-clipped=3.0 2023-03-25 21:53:51,259 WARNING [optim.py:389] (3/7) Scaling gradients by 0.06621765345335007, model_norm_threshold=70.34587860107422 2023-03-25 21:53:51,343 INFO [optim.py:451] (3/7) Parameter Dominanting tot_sumsq module.encoder.encoder_embed.conv.0.weight with proportion 0.67, where dominant_sumsq=(grad_sumsq*orig_rms_sq)=7.539e+05, grad_sumsq = 2.933e+06, orig_rms_sq=2.571e-01 2023-03-25 21:53:58,816 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=10.83 vs. limit=5.0 2023-03-25 21:54:00,691 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=439.0, num_to_drop=2, layers_to_drop={0, 1} 2023-03-25 21:54:05,290 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=448.0, num_to_drop=2, layers_to_drop={1, 2} 2023-03-25 21:54:06,748 INFO [finetune.py:976] (3/7) Epoch 1, batch 450, loss[loss=1.109, simple_loss=0.8956, pruned_loss=1.037, over 4903.00 frames. ], tot_loss[loss=1.464, simple_loss=1.256, pruned_loss=1.292, over 854315.78 frames. ], batch size: 36, lr: 3.80e-03, grad_scale: 0.00390625 2023-03-25 21:54:21,095 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.85 vs. limit=2.0 2023-03-25 21:54:43,458 WARNING [finetune.py:966] (3/7) Grad scale is small: 0.00390625 2023-03-25 21:54:43,459 INFO [finetune.py:976] (3/7) Epoch 1, batch 500, loss[loss=1.099, simple_loss=0.8808, pruned_loss=1.01, over 4939.00 frames. ], tot_loss[loss=1.368, simple_loss=1.159, pruned_loss=1.219, over 874732.85 frames. ], batch size: 38, lr: 4.00e-03, grad_scale: 0.0078125 2023-03-25 21:54:57,597 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.430e+01 1.676e+01 1.950e+01 4.114e+01 1.062e+03, threshold=3.899e+01, percent-clipped=11.0 2023-03-25 21:55:19,619 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.58 vs. limit=2.0 2023-03-25 21:55:28,674 INFO [finetune.py:976] (3/7) Epoch 1, batch 550, loss[loss=1.013, simple_loss=0.7973, pruned_loss=0.9356, over 4751.00 frames. ], tot_loss[loss=1.282, simple_loss=1.074, pruned_loss=1.149, over 891101.62 frames. ], batch size: 28, lr: 4.00e-03, grad_scale: 0.0078125 2023-03-25 21:55:39,066 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5300, 1.7993, 2.1038, 2.2124, 2.0781, 2.0689, 2.8955, 2.0147], device='cuda:3'), covar=tensor([0.0326, 0.0545, 0.0554, 0.0394, 0.0455, 0.0261, 0.0475, 0.0350], device='cuda:3'), in_proj_covar=tensor([0.0279, 0.0308, 0.0297, 0.0337, 0.0329, 0.0268, 0.0365, 0.0257], 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') 2023-03-25 21:55:39,551 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=562.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 21:55:42,631 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=568.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 21:55:58,117 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=590.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 21:56:05,193 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.0063, 3.0705, 3.3398, 2.5793, 2.4890, 4.0268, 2.7795, 3.7419], device='cuda:3'), covar=tensor([0.0090, 0.0084, 0.0109, 0.0121, 0.0079, 0.0055, 0.0114, 0.0041], device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0216, 0.0210, 0.0195, 0.0173, 0.0219, 0.0219, 0.0195], 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') 2023-03-25 21:56:09,546 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600.0, num_to_drop=2, layers_to_drop={0, 2} 2023-03-25 21:56:09,985 WARNING [finetune.py:966] (3/7) Grad scale is small: 0.0078125 2023-03-25 21:56:09,986 INFO [finetune.py:976] (3/7) Epoch 1, batch 600, loss[loss=1.149, simple_loss=0.8975, pruned_loss=1.042, over 4798.00 frames. ], tot_loss[loss=1.218, simple_loss=1.009, pruned_loss=1.094, over 906430.90 frames. ], batch size: 45, lr: 4.00e-03, grad_scale: 0.015625 2023-03-25 21:56:22,611 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.472e+01 1.758e+01 2.024e+01 2.271e+01 8.528e+01, threshold=4.048e+01, percent-clipped=5.0 2023-03-25 21:56:27,288 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=623.0, num_to_drop=2, layers_to_drop={2, 3} 2023-03-25 21:56:36,103 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=629.0, num_to_drop=2, layers_to_drop={2, 3} 2023-03-25 21:56:36,281 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=9.69 vs. limit=5.0 2023-03-25 21:56:54,359 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=648.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 21:56:55,842 INFO [finetune.py:976] (3/7) Epoch 1, batch 650, loss[loss=1.095, simple_loss=0.8524, pruned_loss=0.9714, over 4854.00 frames. ], tot_loss[loss=1.176, simple_loss=0.9627, pruned_loss=1.055, over 918151.14 frames. ], batch size: 47, lr: 4.00e-03, grad_scale: 0.015625 2023-03-25 21:56:55,930 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651.0, num_to_drop=2, layers_to_drop={2, 3} 2023-03-25 21:56:56,431 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=652.0, num_to_drop=2, layers_to_drop={0, 2} 2023-03-25 21:57:20,838 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.38 vs. limit=2.0 2023-03-25 21:57:31,099 INFO [finetune.py:976] (3/7) Epoch 1, batch 700, loss[loss=1.064, simple_loss=0.8227, pruned_loss=0.9276, over 4905.00 frames. ], tot_loss[loss=1.142, simple_loss=0.9245, pruned_loss=1.021, over 926312.73 frames. ], batch size: 37, lr: 4.00e-03, grad_scale: 0.03125 2023-03-25 21:57:38,289 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.810e+01 2.037e+01 2.232e+01 2.628e+01 5.516e+01, threshold=4.463e+01, percent-clipped=4.0 2023-03-25 21:57:46,352 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.76 vs. limit=2.0 2023-03-25 21:57:53,599 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5942, 2.1891, 1.6684, 2.5660, 2.1161, 2.1437, 1.9603, 1.6884], device='cuda:3'), covar=tensor([0.0268, 0.0221, 0.0290, 0.0202, 0.0254, 0.0191, 0.0437, 0.0246], device='cuda:3'), in_proj_covar=tensor([0.0257, 0.0239, 0.0229, 0.0209, 0.0272, 0.0204, 0.0235, 0.0209], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-25 21:57:59,235 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=739.0, num_to_drop=2, layers_to_drop={0, 2} 2023-03-25 21:58:01,230 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=743.0, num_to_drop=2, layers_to_drop={1, 3} 2023-03-25 21:58:02,919 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.48 vs. limit=2.0 2023-03-25 21:58:05,818 INFO [finetune.py:976] (3/7) Epoch 1, batch 750, loss[loss=1.044, simple_loss=0.7993, pruned_loss=0.9023, over 4918.00 frames. ], tot_loss[loss=1.113, simple_loss=0.8916, pruned_loss=0.9892, over 932224.79 frames. ], batch size: 38, lr: 4.00e-03, grad_scale: 0.03125 2023-03-25 21:58:21,584 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=9.16 vs. limit=5.0 2023-03-25 21:58:29,194 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=787.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 21:58:36,507 INFO [finetune.py:976] (3/7) Epoch 1, batch 800, loss[loss=0.9657, simple_loss=0.7303, pruned_loss=0.8289, over 4888.00 frames. ], tot_loss[loss=1.086, simple_loss=0.861, pruned_loss=0.959, over 936632.91 frames. ], batch size: 32, lr: 4.00e-03, grad_scale: 0.0625 2023-03-25 21:58:45,190 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 2.059e+01 2.266e+01 2.508e+01 2.744e+01 4.199e+01, threshold=5.016e+01, percent-clipped=0.0 2023-03-25 21:59:20,561 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=847.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 21:59:22,542 INFO [finetune.py:976] (3/7) Epoch 1, batch 850, loss[loss=0.9986, simple_loss=0.7604, pruned_loss=0.8282, over 4771.00 frames. ], tot_loss[loss=1.055, simple_loss=0.8293, pruned_loss=0.9237, over 937449.00 frames. ], batch size: 51, lr: 4.00e-03, grad_scale: 0.0625 2023-03-25 22:00:12,167 INFO [finetune.py:976] (3/7) Epoch 1, batch 900, loss[loss=0.8994, simple_loss=0.6826, pruned_loss=0.7332, over 4246.00 frames. ], tot_loss[loss=1.025, simple_loss=0.7976, pruned_loss=0.8899, over 940249.23 frames. ], batch size: 65, lr: 4.00e-03, grad_scale: 0.125 2023-03-25 22:00:16,303 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=908.0, num_to_drop=2, layers_to_drop={1, 3} 2023-03-25 22:00:25,565 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 2.101e+01 2.406e+01 2.575e+01 3.027e+01 5.726e+01, threshold=5.150e+01, percent-clipped=1.0 2023-03-25 22:00:28,261 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918.0, num_to_drop=2, layers_to_drop={0, 2} 2023-03-25 22:00:32,448 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=924.0, num_to_drop=2, layers_to_drop={2, 3} 2023-03-25 22:00:35,047 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6783, 2.8829, 2.1893, 2.1822, 3.2479, 2.6657, 3.3544, 2.7406], device='cuda:3'), covar=tensor([0.0627, 0.0865, 0.1164, 0.1411, 0.0629, 0.0593, 0.0439, 0.0768], device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0222, 0.0262, 0.0295, 0.0247, 0.0200, 0.0211, 0.0215], 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') 2023-03-25 22:00:53,626 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=946.0, num_to_drop=2, layers_to_drop={0, 1} 2023-03-25 22:00:56,171 INFO [finetune.py:976] (3/7) Epoch 1, batch 950, loss[loss=0.9785, simple_loss=0.7202, pruned_loss=0.8136, over 4907.00 frames. ], tot_loss[loss=1.005, simple_loss=0.7753, pruned_loss=0.8642, over 944983.26 frames. ], batch size: 35, lr: 4.00e-03, grad_scale: 0.125 2023-03-25 22:00:56,755 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=952.0, num_to_drop=2, layers_to_drop={0, 2} 2023-03-25 22:01:43,866 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=1000.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 22:01:44,325 INFO [finetune.py:976] (3/7) Epoch 1, batch 1000, loss[loss=0.9508, simple_loss=0.6954, pruned_loss=0.7801, over 4822.00 frames. ], tot_loss[loss=1.004, simple_loss=0.7669, pruned_loss=0.8545, over 947481.05 frames. ], batch size: 25, lr: 4.00e-03, grad_scale: 0.25 2023-03-25 22:01:49,043 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=8.71 vs. limit=5.0 2023-03-25 22:01:58,292 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 2.382e+01 2.890e+01 3.153e+01 3.664e+01 7.462e+01, threshold=6.306e+01, percent-clipped=2.0 2023-03-25 22:02:21,750 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1043.0, num_to_drop=2, layers_to_drop={1, 2} 2023-03-25 22:02:31,287 INFO [finetune.py:976] (3/7) Epoch 1, batch 1050, loss[loss=1.033, simple_loss=0.7484, pruned_loss=0.8401, over 4841.00 frames. ], tot_loss[loss=1.01, simple_loss=0.764, pruned_loss=0.85, over 947955.63 frames. ], batch size: 30, lr: 4.00e-03, grad_scale: 0.25 2023-03-25 22:02:47,537 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=7.54 vs. limit=5.0 2023-03-25 22:03:03,073 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-03-25 22:03:07,640 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=1091.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 22:03:18,242 INFO [finetune.py:976] (3/7) Epoch 1, batch 1100, loss[loss=1.01, simple_loss=0.7377, pruned_loss=0.799, over 4884.00 frames. ], tot_loss[loss=1.007, simple_loss=0.7562, pruned_loss=0.8383, over 949134.25 frames. ], batch size: 43, lr: 4.00e-03, grad_scale: 0.5 2023-03-25 22:03:30,381 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.09 vs. limit=2.0 2023-03-25 22:03:30,769 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 2.698e+01 3.337e+01 3.640e+01 4.251e+01 7.174e+01, threshold=7.279e+01, percent-clipped=4.0 2023-03-25 22:03:37,427 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.99 vs. limit=5.0 2023-03-25 22:04:04,828 INFO [finetune.py:976] (3/7) Epoch 1, batch 1150, loss[loss=1.03, simple_loss=0.7496, pruned_loss=0.8031, over 4817.00 frames. ], tot_loss[loss=1.002, simple_loss=0.7472, pruned_loss=0.8237, over 951348.64 frames. ], batch size: 33, lr: 4.00e-03, grad_scale: 0.5 2023-03-25 22:04:05,999 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1153.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 22:04:19,179 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=24.10 vs. limit=5.0 2023-03-25 22:04:46,444 INFO [finetune.py:976] (3/7) Epoch 1, batch 1200, loss[loss=0.8892, simple_loss=0.65, pruned_loss=0.6781, over 4884.00 frames. ], tot_loss[loss=0.9887, simple_loss=0.7339, pruned_loss=0.8017, over 953296.56 frames. ], batch size: 32, lr: 4.00e-03, grad_scale: 1.0 2023-03-25 22:04:47,544 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1203.0, num_to_drop=2, layers_to_drop={0, 2} 2023-03-25 22:04:59,185 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 3.248e+01 4.460e+01 5.563e+01 6.854e+01 1.013e+02, threshold=1.113e+02, percent-clipped=20.0 2023-03-25 22:04:59,285 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4298, 1.3700, 1.5914, 2.0327, 1.6526, 2.8312, 1.4502, 1.5085], device='cuda:3'), covar=tensor([0.1161, 0.2395, 0.1763, 0.1149, 0.1890, 0.0512, 0.1925, 0.2120], device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0082, 0.0073, 0.0074, 0.0091, 0.0077, 0.0086, 0.0078], 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') 2023-03-25 22:04:59,298 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1214.0, num_to_drop=2, layers_to_drop={2, 3} 2023-03-25 22:05:01,653 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1218.0, num_to_drop=2, layers_to_drop={1, 2} 2023-03-25 22:05:06,581 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1224.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 22:05:09,670 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4055, 1.3301, 1.2614, 1.3279, 1.2466, 2.8259, 1.1881, 1.5905], device='cuda:3'), covar=tensor([0.7115, 0.5292, 0.4239, 0.4889, 0.2014, 0.0692, 0.3499, 0.1509], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0119, 0.0127, 0.0124, 0.0109, 0.0098, 0.0094, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-25 22:05:24,086 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1246.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 22:05:26,685 INFO [finetune.py:976] (3/7) Epoch 1, batch 1250, loss[loss=0.8894, simple_loss=0.6536, pruned_loss=0.664, over 4902.00 frames. ], tot_loss[loss=0.9669, simple_loss=0.7166, pruned_loss=0.7721, over 953676.39 frames. ], batch size: 35, lr: 4.00e-03, grad_scale: 1.0 2023-03-25 22:05:46,033 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=1266.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 22:05:49,133 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=1272.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:06:08,472 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=1294.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:06:14,928 INFO [finetune.py:976] (3/7) Epoch 1, batch 1300, loss[loss=0.8926, simple_loss=0.6649, pruned_loss=0.6477, over 4857.00 frames. ], tot_loss[loss=0.9405, simple_loss=0.6971, pruned_loss=0.7388, over 954479.09 frames. ], batch size: 44, lr: 4.00e-03, grad_scale: 1.0 2023-03-25 22:06:23,505 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 5.599e+01 8.403e+01 9.999e+01 1.262e+02 2.600e+02, threshold=2.000e+02, percent-clipped=40.0 2023-03-25 22:06:43,809 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.30 vs. limit=5.0 2023-03-25 22:06:57,098 INFO [finetune.py:976] (3/7) Epoch 1, batch 1350, loss[loss=0.8229, simple_loss=0.6262, pruned_loss=0.5766, over 4818.00 frames. ], tot_loss[loss=0.9231, simple_loss=0.6856, pruned_loss=0.7121, over 952718.90 frames. ], batch size: 30, lr: 4.00e-03, grad_scale: 1.0 2023-03-25 22:07:50,281 INFO [finetune.py:976] (3/7) Epoch 1, batch 1400, loss[loss=0.9116, simple_loss=0.698, pruned_loss=0.6271, over 4762.00 frames. ], tot_loss[loss=0.9027, simple_loss=0.6737, pruned_loss=0.6825, over 954888.26 frames. ], batch size: 54, lr: 4.00e-03, grad_scale: 1.0 2023-03-25 22:07:58,269 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.277e+01 1.400e+02 1.610e+02 1.980e+02 2.974e+02, threshold=3.221e+02, percent-clipped=23.0 2023-03-25 22:08:09,194 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1434.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 22:08:10,289 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4494, 1.4118, 1.3156, 1.5164, 1.2168, 3.0226, 1.3253, 1.8241], device='cuda:3'), covar=tensor([0.7804, 0.5630, 0.4331, 0.4557, 0.2598, 0.0758, 0.2923, 0.1502], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0115, 0.0123, 0.0120, 0.0106, 0.0095, 0.0090, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-25 22:08:20,136 INFO [finetune.py:976] (3/7) Epoch 1, batch 1450, loss[loss=0.8083, simple_loss=0.6234, pruned_loss=0.5459, over 4918.00 frames. ], tot_loss[loss=0.8811, simple_loss=0.6617, pruned_loss=0.6527, over 955124.67 frames. ], batch size: 42, lr: 4.00e-03, grad_scale: 1.0 2023-03-25 22:08:24,183 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-03-25 22:08:40,489 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.05 vs. limit=5.0 2023-03-25 22:08:47,670 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1495.0, num_to_drop=2, layers_to_drop={0, 3} 2023-03-25 22:08:51,790 INFO [finetune.py:976] (3/7) Epoch 1, batch 1500, loss[loss=0.777, simple_loss=0.6113, pruned_loss=0.5096, over 4804.00 frames. ], tot_loss[loss=0.8562, simple_loss=0.648, pruned_loss=0.6211, over 955990.40 frames. ], batch size: 45, lr: 4.00e-03, grad_scale: 1.0 2023-03-25 22:08:51,928 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.8366, 0.5530, 0.8511, 0.7387, 0.5976, 0.6652, 0.6990, 0.8182], device='cuda:3'), covar=tensor([27.2989, 49.7321, 36.8277, 52.1719, 70.4746, 35.1852, 72.1359, 22.2888], device='cuda:3'), in_proj_covar=tensor([0.0276, 0.0304, 0.0293, 0.0333, 0.0324, 0.0270, 0.0358, 0.0257], 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') 2023-03-25 22:08:52,984 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1503.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 22:09:02,829 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1509.0, num_to_drop=2, layers_to_drop={1, 3} 2023-03-25 22:09:03,359 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3959, 3.5872, 3.5471, 1.3818, 3.7831, 2.6493, 0.9026, 2.4800], device='cuda:3'), covar=tensor([0.2394, 0.0944, 0.1530, 0.3354, 0.0749, 0.0915, 0.4177, 0.1171], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0149, 0.0158, 0.0125, 0.0148, 0.0113, 0.0139, 0.0115], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-25 22:09:05,946 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.968e+01 1.844e+02 2.293e+02 2.711e+02 4.587e+02, threshold=4.586e+02, percent-clipped=13.0 2023-03-25 22:09:42,480 INFO [finetune.py:976] (3/7) Epoch 1, batch 1550, loss[loss=0.6819, simple_loss=0.5523, pruned_loss=0.4308, over 4819.00 frames. ], tot_loss[loss=0.8243, simple_loss=0.6298, pruned_loss=0.5855, over 956593.19 frames. ], batch size: 40, lr: 4.00e-03, grad_scale: 1.0 2023-03-25 22:09:42,539 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=1551.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:09:52,839 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1566.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 22:10:33,765 INFO [finetune.py:976] (3/7) Epoch 1, batch 1600, loss[loss=0.6528, simple_loss=0.5204, pruned_loss=0.4153, over 4898.00 frames. ], tot_loss[loss=0.787, simple_loss=0.6072, pruned_loss=0.5477, over 957472.13 frames. ], batch size: 43, lr: 4.00e-03, grad_scale: 2.0 2023-03-25 22:10:40,891 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1611.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 22:10:42,945 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.367e+02 1.965e+02 2.441e+02 2.819e+02 5.041e+02, threshold=4.882e+02, percent-clipped=1.0 2023-03-25 22:10:55,942 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1627.0, num_to_drop=2, layers_to_drop={0, 2} 2023-03-25 22:11:08,296 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.5142, 3.8620, 3.9259, 4.3168, 4.1370, 4.0135, 4.6481, 1.5071], device='cuda:3'), covar=tensor([0.0991, 0.1319, 0.1021, 0.1282, 0.1785, 0.1269, 0.0777, 0.6808], device='cuda:3'), in_proj_covar=tensor([0.0365, 0.0241, 0.0256, 0.0290, 0.0343, 0.0282, 0.0305, 0.0301], 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') 2023-03-25 22:11:18,722 INFO [finetune.py:976] (3/7) Epoch 1, batch 1650, loss[loss=0.6344, simple_loss=0.5119, pruned_loss=0.3959, over 4933.00 frames. ], tot_loss[loss=0.7532, simple_loss=0.5868, pruned_loss=0.514, over 955942.19 frames. ], batch size: 38, lr: 4.00e-03, grad_scale: 2.0 2023-03-25 22:11:24,240 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-25 22:11:41,965 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1672.0, num_to_drop=2, layers_to_drop={1, 3} 2023-03-25 22:12:02,517 INFO [finetune.py:976] (3/7) Epoch 1, batch 1700, loss[loss=0.6524, simple_loss=0.5395, pruned_loss=0.3955, over 4813.00 frames. ], tot_loss[loss=0.7224, simple_loss=0.569, pruned_loss=0.4831, over 955506.91 frames. ], batch size: 41, lr: 4.00e-03, grad_scale: 2.0 2023-03-25 22:12:14,553 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.227e+02 2.187e+02 2.736e+02 3.197e+02 8.210e+02, threshold=5.471e+02, percent-clipped=2.0 2023-03-25 22:12:53,680 INFO [finetune.py:976] (3/7) Epoch 1, batch 1750, loss[loss=0.5557, simple_loss=0.4799, pruned_loss=0.3219, over 4822.00 frames. ], tot_loss[loss=0.7037, simple_loss=0.5602, pruned_loss=0.4612, over 952998.11 frames. ], batch size: 30, lr: 4.00e-03, grad_scale: 2.0 2023-03-25 22:13:14,240 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.02 vs. limit=5.0 2023-03-25 22:13:29,979 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1790.0, num_to_drop=1, layers_to_drop={2} 2023-03-25 22:13:36,054 INFO [finetune.py:976] (3/7) Epoch 1, batch 1800, loss[loss=0.5036, simple_loss=0.4367, pruned_loss=0.2895, over 4773.00 frames. ], tot_loss[loss=0.6867, simple_loss=0.5524, pruned_loss=0.4417, over 952314.68 frames. ], batch size: 26, lr: 4.00e-03, grad_scale: 2.0 2023-03-25 22:13:40,476 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1809.0, num_to_drop=1, layers_to_drop={2} 2023-03-25 22:13:43,026 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.190e+02 2.215e+02 2.629e+02 3.291e+02 5.990e+02, threshold=5.258e+02, percent-clipped=1.0 2023-03-25 22:13:59,220 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1838.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 22:14:03,355 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.14 vs. limit=5.0 2023-03-25 22:14:06,687 INFO [finetune.py:976] (3/7) Epoch 1, batch 1850, loss[loss=0.55, simple_loss=0.4759, pruned_loss=0.3155, over 4839.00 frames. ], tot_loss[loss=0.6678, simple_loss=0.5429, pruned_loss=0.4218, over 953182.32 frames. ], batch size: 30, lr: 4.00e-03, grad_scale: 2.0 2023-03-25 22:14:10,079 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=1857.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:14:10,125 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1857.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 22:14:17,832 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1870.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 22:14:18,943 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.5157, 3.8932, 4.0216, 4.4003, 4.2475, 3.9973, 4.5958, 1.4781], device='cuda:3'), covar=tensor([0.0785, 0.0875, 0.0683, 0.0843, 0.1316, 0.1111, 0.0664, 0.4995], device='cuda:3'), in_proj_covar=tensor([0.0364, 0.0240, 0.0255, 0.0289, 0.0343, 0.0282, 0.0304, 0.0298], 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') 2023-03-25 22:14:51,506 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1899.0, num_to_drop=2, layers_to_drop={2, 3} 2023-03-25 22:14:52,522 INFO [finetune.py:976] (3/7) Epoch 1, batch 1900, loss[loss=0.5649, simple_loss=0.494, pruned_loss=0.32, over 4743.00 frames. ], tot_loss[loss=0.6508, simple_loss=0.5351, pruned_loss=0.4038, over 953722.59 frames. ], batch size: 27, lr: 4.00e-03, grad_scale: 2.0 2023-03-25 22:15:03,948 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.381e+02 2.208e+02 2.560e+02 3.227e+02 6.450e+02, threshold=5.121e+02, percent-clipped=1.0 2023-03-25 22:15:11,549 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1918.0, num_to_drop=2, layers_to_drop={2, 3} 2023-03-25 22:15:14,227 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1922.0, num_to_drop=1, layers_to_drop={3} 2023-03-25 22:15:20,196 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1931.0, num_to_drop=2, layers_to_drop={0, 1} 2023-03-25 22:15:37,065 INFO [finetune.py:976] (3/7) Epoch 1, batch 1950, loss[loss=0.6049, simple_loss=0.5151, pruned_loss=0.3487, over 4797.00 frames. ], tot_loss[loss=0.6298, simple_loss=0.5231, pruned_loss=0.3845, over 955214.24 frames. ], batch size: 26, lr: 4.00e-03, grad_scale: 2.0 2023-03-25 22:15:46,027 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1967.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:16:12,901 INFO [finetune.py:976] (3/7) Epoch 1, batch 2000, loss[loss=0.4871, simple_loss=0.4338, pruned_loss=0.2702, over 4794.00 frames. ], tot_loss[loss=0.6071, simple_loss=0.5093, pruned_loss=0.3652, over 956954.03 frames. ], batch size: 25, lr: 4.00e-03, grad_scale: 4.0 2023-03-25 22:16:22,847 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.403e+02 2.183e+02 2.758e+02 3.285e+02 7.843e+02, threshold=5.515e+02, percent-clipped=1.0 2023-03-25 22:16:57,268 INFO [finetune.py:976] (3/7) Epoch 1, batch 2050, loss[loss=0.4097, simple_loss=0.373, pruned_loss=0.2232, over 4343.00 frames. ], tot_loss[loss=0.5829, simple_loss=0.4947, pruned_loss=0.3455, over 957675.53 frames. ], batch size: 19, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:17:05,974 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0376, 1.8041, 1.4209, 1.5618, 2.1242, 2.1462, 1.7933, 1.4959], device='cuda:3'), covar=tensor([0.0576, 0.0724, 0.0929, 0.0712, 0.0521, 0.0598, 0.0766, 0.1445], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0135, 0.0132, 0.0120, 0.0109, 0.0129, 0.0135, 0.0171], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-25 22:17:17,894 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7528, 1.4370, 1.5397, 1.6233, 2.5251, 1.6054, 1.4405, 1.2648], device='cuda:3'), covar=tensor([0.5140, 0.6475, 0.5424, 0.5790, 0.4016, 0.3409, 0.7108, 0.4469], device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0214, 0.0203, 0.0188, 0.0240, 0.0187, 0.0211, 0.0189], 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') 2023-03-25 22:17:31,388 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2090.0, num_to_drop=2, layers_to_drop={1, 2} 2023-03-25 22:17:41,736 INFO [finetune.py:976] (3/7) Epoch 1, batch 2100, loss[loss=0.4817, simple_loss=0.4307, pruned_loss=0.2664, over 4759.00 frames. ], tot_loss[loss=0.5649, simple_loss=0.4845, pruned_loss=0.3304, over 955111.39 frames. ], batch size: 28, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:17:44,289 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.29 vs. limit=5.0 2023-03-25 22:17:55,062 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.336e+02 2.022e+02 2.484e+02 2.961e+02 6.695e+02, threshold=4.968e+02, percent-clipped=1.0 2023-03-25 22:18:12,541 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=2138.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:18:29,607 INFO [finetune.py:976] (3/7) Epoch 1, batch 2150, loss[loss=0.4299, simple_loss=0.4046, pruned_loss=0.2276, over 4750.00 frames. ], tot_loss[loss=0.5561, simple_loss=0.4824, pruned_loss=0.321, over 952425.67 frames. ], batch size: 26, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:18:47,854 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.31 vs. limit=5.0 2023-03-25 22:19:03,492 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2194.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:19:08,497 INFO [finetune.py:976] (3/7) Epoch 1, batch 2200, loss[loss=0.5113, simple_loss=0.4804, pruned_loss=0.2711, over 4814.00 frames. ], tot_loss[loss=0.5473, simple_loss=0.4804, pruned_loss=0.3118, over 954304.23 frames. ], batch size: 38, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:19:17,017 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2213.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:19:17,478 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.568e+02 2.355e+02 2.819e+02 3.325e+02 5.172e+02, threshold=5.637e+02, percent-clipped=1.0 2023-03-25 22:19:22,645 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2222.0, num_to_drop=2, layers_to_drop={0, 2} 2023-03-25 22:19:28,126 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2226.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:19:41,134 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3641, 1.6866, 1.5024, 1.7870, 1.6978, 3.1539, 1.3116, 1.6131], device='cuda:3'), covar=tensor([0.1147, 0.1616, 0.1424, 0.1203, 0.1688, 0.0260, 0.1681, 0.2028], device='cuda:3'), in_proj_covar=tensor([0.0068, 0.0074, 0.0067, 0.0070, 0.0085, 0.0071, 0.0080, 0.0073], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-25 22:19:57,059 INFO [finetune.py:976] (3/7) Epoch 1, batch 2250, loss[loss=0.5397, simple_loss=0.4974, pruned_loss=0.291, over 4845.00 frames. ], tot_loss[loss=0.5374, simple_loss=0.4768, pruned_loss=0.3027, over 954811.96 frames. ], batch size: 49, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:20:02,309 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-25 22:20:18,371 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2267.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 22:20:20,122 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=2270.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:20:31,787 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=9.24 vs. limit=5.0 2023-03-25 22:21:00,808 INFO [finetune.py:976] (3/7) Epoch 1, batch 2300, loss[loss=0.5128, simple_loss=0.4551, pruned_loss=0.2853, over 4120.00 frames. ], tot_loss[loss=0.526, simple_loss=0.4716, pruned_loss=0.293, over 954588.75 frames. ], batch size: 65, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:21:15,882 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.417e+02 2.050e+02 2.425e+02 2.921e+02 4.362e+02, threshold=4.850e+02, percent-clipped=0.0 2023-03-25 22:21:21,996 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=2315.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:21:22,049 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5894, 2.0382, 1.5963, 1.8698, 1.1515, 3.6568, 1.3323, 1.9024], device='cuda:3'), covar=tensor([0.3910, 0.2400, 0.2277, 0.2289, 0.2101, 0.0179, 0.2712, 0.1495], device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0099, 0.0107, 0.0104, 0.0094, 0.0083, 0.0080, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-25 22:21:37,390 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2340.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 22:21:54,802 INFO [finetune.py:976] (3/7) Epoch 1, batch 2350, loss[loss=0.4967, simple_loss=0.473, pruned_loss=0.2602, over 4894.00 frames. ], tot_loss[loss=0.5081, simple_loss=0.4609, pruned_loss=0.2799, over 955471.15 frames. ], batch size: 36, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:21:54,942 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8297, 1.9268, 1.5292, 1.3886, 2.3618, 2.2533, 1.7838, 1.7271], device='cuda:3'), covar=tensor([0.0710, 0.0624, 0.0925, 0.0917, 0.0442, 0.0563, 0.0870, 0.1299], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0134, 0.0133, 0.0121, 0.0109, 0.0130, 0.0136, 0.0170], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-25 22:22:06,924 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-03-25 22:22:57,438 INFO [finetune.py:976] (3/7) Epoch 1, batch 2400, loss[loss=0.4613, simple_loss=0.4269, pruned_loss=0.2478, over 4909.00 frames. ], tot_loss[loss=0.4939, simple_loss=0.4512, pruned_loss=0.27, over 953687.52 frames. ], batch size: 36, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:22:57,559 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2401.0, num_to_drop=2, layers_to_drop={1, 2} 2023-03-25 22:23:05,864 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.408e+02 1.953e+02 2.427e+02 2.971e+02 6.309e+02, threshold=4.853e+02, percent-clipped=1.0 2023-03-25 22:23:14,091 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-25 22:23:32,011 INFO [finetune.py:976] (3/7) Epoch 1, batch 2450, loss[loss=0.449, simple_loss=0.4005, pruned_loss=0.2488, over 4829.00 frames. ], tot_loss[loss=0.4811, simple_loss=0.4427, pruned_loss=0.2611, over 955222.95 frames. ], batch size: 25, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:24:21,669 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2494.0, num_to_drop=1, layers_to_drop={2} 2023-03-25 22:24:25,638 INFO [finetune.py:976] (3/7) Epoch 1, batch 2500, loss[loss=0.4728, simple_loss=0.4444, pruned_loss=0.2506, over 4862.00 frames. ], tot_loss[loss=0.4772, simple_loss=0.4416, pruned_loss=0.2575, over 954173.99 frames. ], batch size: 31, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:24:34,880 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2513.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:24:35,368 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.353e+02 2.241e+02 2.593e+02 3.079e+02 4.323e+02, threshold=5.185e+02, percent-clipped=0.0 2023-03-25 22:24:44,454 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2526.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:24:54,545 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=2542.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:24:56,264 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.48 vs. limit=5.0 2023-03-25 22:25:00,192 INFO [finetune.py:976] (3/7) Epoch 1, batch 2550, loss[loss=0.4915, simple_loss=0.4782, pruned_loss=0.2524, over 4819.00 frames. ], tot_loss[loss=0.4742, simple_loss=0.4428, pruned_loss=0.2536, over 954013.07 frames. ], batch size: 30, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:25:09,685 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=2561.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:25:18,673 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=2574.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:25:48,340 INFO [finetune.py:976] (3/7) Epoch 1, batch 2600, loss[loss=0.4732, simple_loss=0.4545, pruned_loss=0.2459, over 4805.00 frames. ], tot_loss[loss=0.4697, simple_loss=0.4414, pruned_loss=0.2497, over 953118.46 frames. ], batch size: 51, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:25:55,824 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.651e+02 2.205e+02 2.587e+02 2.996e+02 4.228e+02, threshold=5.174e+02, percent-clipped=0.0 2023-03-25 22:26:14,932 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.88 vs. limit=5.0 2023-03-25 22:26:19,957 INFO [finetune.py:976] (3/7) Epoch 1, batch 2650, loss[loss=0.421, simple_loss=0.4175, pruned_loss=0.2122, over 4913.00 frames. ], tot_loss[loss=0.4641, simple_loss=0.4394, pruned_loss=0.2449, over 953717.65 frames. ], batch size: 36, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:26:54,479 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2685.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 22:27:06,879 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2696.0, num_to_drop=1, layers_to_drop={3} 2023-03-25 22:27:15,135 INFO [finetune.py:976] (3/7) Epoch 1, batch 2700, loss[loss=0.4079, simple_loss=0.3948, pruned_loss=0.2105, over 4781.00 frames. ], tot_loss[loss=0.4564, simple_loss=0.4348, pruned_loss=0.2394, over 953442.70 frames. ], batch size: 26, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:27:28,252 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.346e+02 2.127e+02 2.493e+02 3.058e+02 5.200e+02, threshold=4.985e+02, percent-clipped=1.0 2023-03-25 22:28:14,554 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2746.0, num_to_drop=2, layers_to_drop={2, 3} 2023-03-25 22:28:16,312 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.06 vs. limit=5.0 2023-03-25 22:28:17,274 INFO [finetune.py:976] (3/7) Epoch 1, batch 2750, loss[loss=0.4274, simple_loss=0.4044, pruned_loss=0.2253, over 4819.00 frames. ], tot_loss[loss=0.4446, simple_loss=0.426, pruned_loss=0.2319, over 953402.73 frames. ], batch size: 39, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:28:58,716 INFO [finetune.py:976] (3/7) Epoch 1, batch 2800, loss[loss=0.3788, simple_loss=0.3779, pruned_loss=0.1899, over 4729.00 frames. ], tot_loss[loss=0.4343, simple_loss=0.4186, pruned_loss=0.2252, over 953674.90 frames. ], batch size: 23, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:29:06,641 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.489e+02 2.264e+02 2.537e+02 3.001e+02 5.007e+02, threshold=5.073e+02, percent-clipped=1.0 2023-03-25 22:29:12,542 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2824.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 22:29:40,281 INFO [finetune.py:976] (3/7) Epoch 1, batch 2850, loss[loss=0.409, simple_loss=0.3927, pruned_loss=0.2126, over 4898.00 frames. ], tot_loss[loss=0.4288, simple_loss=0.4152, pruned_loss=0.2213, over 955454.64 frames. ], batch size: 35, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:30:03,460 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2885.0, num_to_drop=2, layers_to_drop={0, 3} 2023-03-25 22:30:08,511 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.40 vs. limit=5.0 2023-03-25 22:30:15,603 INFO [finetune.py:976] (3/7) Epoch 1, batch 2900, loss[loss=0.4846, simple_loss=0.4694, pruned_loss=0.2499, over 4819.00 frames. ], tot_loss[loss=0.4312, simple_loss=0.4185, pruned_loss=0.2221, over 954938.80 frames. ], batch size: 38, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:30:23,164 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.426e+02 2.100e+02 2.461e+02 2.914e+02 6.574e+02, threshold=4.923e+02, percent-clipped=3.0 2023-03-25 22:30:26,084 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5886, 1.8094, 1.6654, 1.7108, 1.1700, 3.2475, 1.5077, 2.1001], device='cuda:3'), covar=tensor([0.3461, 0.2321, 0.1902, 0.2031, 0.1927, 0.0192, 0.2625, 0.1348], device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0100, 0.0108, 0.0105, 0.0097, 0.0085, 0.0083, 0.0081], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-25 22:30:29,762 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-25 22:30:32,554 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5428, 1.5441, 1.5191, 1.0635, 1.9054, 1.6192, 1.5764, 1.4531], device='cuda:3'), covar=tensor([0.0781, 0.0712, 0.0796, 0.1022, 0.0501, 0.0803, 0.0833, 0.1299], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0132, 0.0132, 0.0121, 0.0108, 0.0131, 0.0137, 0.0165], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-25 22:30:37,151 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2937.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 22:30:50,912 INFO [finetune.py:976] (3/7) Epoch 1, batch 2950, loss[loss=0.4716, simple_loss=0.4617, pruned_loss=0.2407, over 4923.00 frames. ], tot_loss[loss=0.431, simple_loss=0.4205, pruned_loss=0.2209, over 955803.43 frames. ], batch size: 42, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:31:01,114 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.89 vs. limit=5.0 2023-03-25 22:31:31,736 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2996.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 22:31:33,399 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2998.0, num_to_drop=2, layers_to_drop={0, 2} 2023-03-25 22:31:35,183 INFO [finetune.py:976] (3/7) Epoch 1, batch 3000, loss[loss=0.4275, simple_loss=0.4157, pruned_loss=0.2196, over 4731.00 frames. ], tot_loss[loss=0.4297, simple_loss=0.4203, pruned_loss=0.2197, over 953089.82 frames. ], batch size: 59, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:31:35,183 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-25 22:31:56,384 INFO [finetune.py:1010] (3/7) Epoch 1, validation: loss=0.4228, simple_loss=0.4589, pruned_loss=0.1933, over 2265189.00 frames. 2023-03-25 22:31:56,384 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 5768MB 2023-03-25 22:32:17,070 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.364e+02 2.092e+02 2.490e+02 2.940e+02 5.162e+02, threshold=4.980e+02, percent-clipped=2.0 2023-03-25 22:32:25,838 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3019.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 22:32:39,246 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3041.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:32:40,985 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=3044.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 22:32:44,999 INFO [finetune.py:976] (3/7) Epoch 1, batch 3050, loss[loss=0.3682, simple_loss=0.3675, pruned_loss=0.1844, over 4735.00 frames. ], tot_loss[loss=0.4255, simple_loss=0.4183, pruned_loss=0.2164, over 953441.62 frames. ], batch size: 23, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:33:09,823 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3080.0, num_to_drop=2, layers_to_drop={1, 2} 2023-03-25 22:33:38,269 INFO [finetune.py:976] (3/7) Epoch 1, batch 3100, loss[loss=0.4982, simple_loss=0.4527, pruned_loss=0.2719, over 4929.00 frames. ], tot_loss[loss=0.4188, simple_loss=0.4137, pruned_loss=0.212, over 955841.36 frames. ], batch size: 38, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:33:51,987 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.371e+02 2.009e+02 2.458e+02 3.052e+02 4.298e+02, threshold=4.916e+02, percent-clipped=0.0 2023-03-25 22:34:31,866 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2820, 1.8641, 1.8874, 0.8193, 1.7961, 1.8570, 1.4989, 2.0225], device='cuda:3'), covar=tensor([0.0693, 0.0978, 0.1273, 0.2432, 0.1016, 0.1894, 0.1940, 0.0766], device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0162, 0.0176, 0.0162, 0.0178, 0.0177, 0.0186, 0.0173], 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') 2023-03-25 22:34:39,800 INFO [finetune.py:976] (3/7) Epoch 1, batch 3150, loss[loss=0.3689, simple_loss=0.3794, pruned_loss=0.1792, over 4860.00 frames. ], tot_loss[loss=0.411, simple_loss=0.4076, pruned_loss=0.2072, over 956384.79 frames. ], batch size: 31, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:34:49,130 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.29 vs. limit=2.0 2023-03-25 22:35:16,505 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3180.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:35:39,621 INFO [finetune.py:976] (3/7) Epoch 1, batch 3200, loss[loss=0.3528, simple_loss=0.3637, pruned_loss=0.1709, over 4778.00 frames. ], tot_loss[loss=0.4005, simple_loss=0.3991, pruned_loss=0.201, over 953987.81 frames. ], batch size: 28, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:35:52,727 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.131e+02 1.973e+02 2.320e+02 2.787e+02 5.091e+02, threshold=4.641e+02, percent-clipped=1.0 2023-03-25 22:36:29,138 INFO [finetune.py:976] (3/7) Epoch 1, batch 3250, loss[loss=0.352, simple_loss=0.3619, pruned_loss=0.1711, over 4164.00 frames. ], tot_loss[loss=0.3965, simple_loss=0.3965, pruned_loss=0.1983, over 953642.64 frames. ], batch size: 18, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:37:08,539 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-25 22:37:12,276 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3288.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 22:37:19,386 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1895, 2.7728, 2.7766, 1.1849, 2.9596, 2.1518, 0.9432, 1.7671], device='cuda:3'), covar=tensor([0.2162, 0.1605, 0.1451, 0.2961, 0.1148, 0.0910, 0.3130, 0.1317], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0150, 0.0156, 0.0122, 0.0147, 0.0112, 0.0138, 0.0114], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-25 22:37:19,998 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3293.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:37:24,557 INFO [finetune.py:976] (3/7) Epoch 1, batch 3300, loss[loss=0.366, simple_loss=0.3887, pruned_loss=0.1716, over 4753.00 frames. ], tot_loss[loss=0.4003, simple_loss=0.401, pruned_loss=0.1998, over 952508.85 frames. ], batch size: 28, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:37:32,702 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.553e+02 2.210e+02 2.512e+02 3.057e+02 4.555e+02, threshold=5.024e+02, percent-clipped=0.0 2023-03-25 22:38:03,272 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3341.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:38:13,337 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3349.0, num_to_drop=2, layers_to_drop={1, 3} 2023-03-25 22:38:14,397 INFO [finetune.py:976] (3/7) Epoch 1, batch 3350, loss[loss=0.3501, simple_loss=0.3593, pruned_loss=0.1704, over 4746.00 frames. ], tot_loss[loss=0.4004, simple_loss=0.4023, pruned_loss=0.1992, over 950105.00 frames. ], batch size: 27, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:38:44,439 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3375.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:38:59,534 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=3389.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:39:06,452 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.74 vs. limit=5.0 2023-03-25 22:39:06,748 INFO [finetune.py:976] (3/7) Epoch 1, batch 3400, loss[loss=0.3951, simple_loss=0.4081, pruned_loss=0.191, over 4821.00 frames. ], tot_loss[loss=0.3978, simple_loss=0.402, pruned_loss=0.1968, over 951571.93 frames. ], batch size: 39, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:39:20,796 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.307e+02 1.988e+02 2.392e+02 2.720e+02 4.202e+02, threshold=4.784e+02, percent-clipped=0.0 2023-03-25 22:40:08,842 INFO [finetune.py:976] (3/7) Epoch 1, batch 3450, loss[loss=0.3364, simple_loss=0.3639, pruned_loss=0.1545, over 4803.00 frames. ], tot_loss[loss=0.3964, simple_loss=0.4012, pruned_loss=0.1958, over 953015.42 frames. ], batch size: 51, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:40:19,946 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6714, 4.0738, 4.0410, 2.1869, 4.3135, 3.2554, 1.0925, 3.0286], device='cuda:3'), covar=tensor([0.2867, 0.1361, 0.1412, 0.2893, 0.0760, 0.0832, 0.4110, 0.1177], device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0151, 0.0157, 0.0122, 0.0147, 0.0112, 0.0138, 0.0114], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-25 22:40:21,190 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.0143, 0.8747, 0.9168, 0.7560, 1.2369, 0.9544, 1.3411, 0.8965], device='cuda:3'), covar=tensor([2.4818, 5.0616, 3.7425, 4.5649, 2.5751, 1.7842, 2.4373, 3.6043], device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0188, 0.0225, 0.0241, 0.0201, 0.0172, 0.0178, 0.0182], 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') 2023-03-25 22:40:37,625 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4370, 1.2661, 1.3618, 1.4407, 1.9568, 1.3542, 1.2140, 1.1178], device='cuda:3'), covar=tensor([0.3462, 0.3911, 0.2903, 0.3172, 0.2769, 0.2272, 0.4657, 0.2841], device='cuda:3'), in_proj_covar=tensor([0.0214, 0.0199, 0.0186, 0.0173, 0.0220, 0.0171, 0.0195, 0.0174], 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') 2023-03-25 22:40:40,384 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3480.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:40:42,675 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2354, 1.0187, 0.8255, 0.9565, 1.0639, 0.9057, 0.9201, 1.5879], device='cuda:3'), covar=tensor([10.9156, 12.2717, 10.2608, 17.8763, 9.3680, 6.4338, 12.5770, 3.4167], device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0226, 0.0206, 0.0264, 0.0220, 0.0190, 0.0226, 0.0169], 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') 2023-03-25 22:41:03,280 INFO [finetune.py:976] (3/7) Epoch 1, batch 3500, loss[loss=0.4506, simple_loss=0.4287, pruned_loss=0.2363, over 4912.00 frames. ], tot_loss[loss=0.3912, simple_loss=0.3968, pruned_loss=0.1928, over 953512.73 frames. ], batch size: 37, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:41:14,914 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.301e+02 2.271e+02 2.832e+02 3.824e+02 1.123e+03, threshold=5.664e+02, percent-clipped=12.0 2023-03-25 22:41:30,605 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=3528.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:41:48,800 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3546.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 22:41:57,306 INFO [finetune.py:976] (3/7) Epoch 1, batch 3550, loss[loss=0.3535, simple_loss=0.365, pruned_loss=0.171, over 4695.00 frames. ], tot_loss[loss=0.3879, simple_loss=0.3928, pruned_loss=0.1915, over 952324.19 frames. ], batch size: 23, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:42:45,346 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3593.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:42:50,746 INFO [finetune.py:976] (3/7) Epoch 1, batch 3600, loss[loss=0.3594, simple_loss=0.3777, pruned_loss=0.1706, over 4822.00 frames. ], tot_loss[loss=0.3804, simple_loss=0.3873, pruned_loss=0.1867, over 953095.92 frames. ], batch size: 33, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:42:59,658 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3607.0, num_to_drop=2, layers_to_drop={1, 2} 2023-03-25 22:43:09,365 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.842e+02 2.557e+02 2.866e+02 3.769e+02 9.044e+02, threshold=5.732e+02, percent-clipped=5.0 2023-03-25 22:43:44,103 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=3641.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:43:46,916 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3644.0, num_to_drop=1, layers_to_drop={2} 2023-03-25 22:43:52,547 INFO [finetune.py:976] (3/7) Epoch 1, batch 3650, loss[loss=0.4262, simple_loss=0.4364, pruned_loss=0.208, over 4819.00 frames. ], tot_loss[loss=0.3824, simple_loss=0.3901, pruned_loss=0.1874, over 953777.17 frames. ], batch size: 38, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:44:20,322 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3675.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:44:21,470 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3676.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 22:44:40,055 INFO [finetune.py:976] (3/7) Epoch 1, batch 3700, loss[loss=0.4056, simple_loss=0.4176, pruned_loss=0.1968, over 4933.00 frames. ], tot_loss[loss=0.3836, simple_loss=0.3928, pruned_loss=0.1873, over 953426.40 frames. ], batch size: 33, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:44:52,810 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.548e+02 2.567e+02 2.980e+02 3.536e+02 5.905e+02, threshold=5.959e+02, percent-clipped=1.0 2023-03-25 22:45:09,752 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=3723.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:45:10,443 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.7213, 0.5088, 0.6695, 0.5918, 0.5171, 0.5251, 0.5956, 0.6072], device='cuda:3'), covar=tensor([ 6.7394, 11.9208, 7.0560, 10.1428, 11.0452, 6.8892, 13.6234, 6.2747], device='cuda:3'), in_proj_covar=tensor([0.0217, 0.0243, 0.0229, 0.0259, 0.0244, 0.0210, 0.0272, 0.0206], 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') 2023-03-25 22:45:23,631 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3737.0, num_to_drop=2, layers_to_drop={2, 3} 2023-03-25 22:45:43,731 INFO [finetune.py:976] (3/7) Epoch 1, batch 3750, loss[loss=0.3647, simple_loss=0.3834, pruned_loss=0.173, over 4820.00 frames. ], tot_loss[loss=0.3824, simple_loss=0.3933, pruned_loss=0.1858, over 953622.07 frames. ], batch size: 33, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:46:00,212 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3543, 1.4887, 1.3731, 1.4840, 0.8201, 2.7534, 0.8890, 1.5185], device='cuda:3'), covar=tensor([0.4153, 0.2747, 0.2561, 0.2696, 0.2685, 0.0342, 0.3361, 0.1722], device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0102, 0.0110, 0.0108, 0.0100, 0.0088, 0.0088, 0.0085], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0005, 0.0004, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-25 22:46:25,739 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5774, 1.7538, 1.6757, 2.0286, 1.8076, 3.6004, 1.5800, 1.8815], device='cuda:3'), covar=tensor([0.1166, 0.1667, 0.1191, 0.1131, 0.1618, 0.0232, 0.1431, 0.1835], device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0076, 0.0071, 0.0075, 0.0088, 0.0075, 0.0082, 0.0076], 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') 2023-03-25 22:46:33,630 INFO [finetune.py:976] (3/7) Epoch 1, batch 3800, loss[loss=0.3921, simple_loss=0.4134, pruned_loss=0.1854, over 4900.00 frames. ], tot_loss[loss=0.3837, simple_loss=0.3945, pruned_loss=0.1865, over 953784.76 frames. ], batch size: 36, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:46:47,077 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.580e+02 2.141e+02 2.900e+02 3.620e+02 1.043e+03, threshold=5.800e+02, percent-clipped=4.0 2023-03-25 22:47:02,693 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.77 vs. limit=5.0 2023-03-25 22:47:05,879 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.96 vs. limit=5.0 2023-03-25 22:47:22,184 INFO [finetune.py:976] (3/7) Epoch 1, batch 3850, loss[loss=0.343, simple_loss=0.3666, pruned_loss=0.1597, over 4896.00 frames. ], tot_loss[loss=0.3774, simple_loss=0.3898, pruned_loss=0.1825, over 954630.63 frames. ], batch size: 43, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:48:09,989 INFO [finetune.py:976] (3/7) Epoch 1, batch 3900, loss[loss=0.3293, simple_loss=0.3578, pruned_loss=0.1505, over 4936.00 frames. ], tot_loss[loss=0.373, simple_loss=0.3855, pruned_loss=0.1803, over 953439.23 frames. ], batch size: 38, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:48:10,645 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3902.0, num_to_drop=1, layers_to_drop={2} 2023-03-25 22:48:20,243 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.580e+02 2.264e+02 2.673e+02 3.196e+02 5.181e+02, threshold=5.346e+02, percent-clipped=0.0 2023-03-25 22:48:30,353 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3926.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:48:50,575 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3944.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 22:48:54,666 INFO [finetune.py:976] (3/7) Epoch 1, batch 3950, loss[loss=0.2967, simple_loss=0.314, pruned_loss=0.1397, over 4827.00 frames. ], tot_loss[loss=0.365, simple_loss=0.3785, pruned_loss=0.1757, over 953825.68 frames. ], batch size: 25, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:49:35,694 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6664, 1.1133, 0.8106, 1.4226, 1.9553, 0.6081, 1.2728, 1.4942], device='cuda:3'), covar=tensor([0.1601, 0.2124, 0.1969, 0.1284, 0.2215, 0.1993, 0.1418, 0.1884], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0094, 0.0113, 0.0090, 0.0122, 0.0093, 0.0096, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-25 22:49:45,861 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3987.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:49:53,766 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=3992.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 22:50:05,834 INFO [finetune.py:976] (3/7) Epoch 1, batch 4000, loss[loss=0.3177, simple_loss=0.3581, pruned_loss=0.1386, over 4816.00 frames. ], tot_loss[loss=0.3624, simple_loss=0.3765, pruned_loss=0.1741, over 955038.97 frames. ], batch size: 41, lr: 4.00e-03, grad_scale: 8.0 2023-03-25 22:50:18,330 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.331e+02 2.072e+02 2.562e+02 2.941e+02 5.028e+02, threshold=5.123e+02, percent-clipped=0.0 2023-03-25 22:50:31,331 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4032.0, num_to_drop=1, layers_to_drop={3} 2023-03-25 22:50:32,331 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.04 vs. limit=5.0 2023-03-25 22:50:50,289 INFO [finetune.py:976] (3/7) Epoch 1, batch 4050, loss[loss=0.3428, simple_loss=0.3838, pruned_loss=0.1509, over 4902.00 frames. ], tot_loss[loss=0.3615, simple_loss=0.3763, pruned_loss=0.1733, over 952809.47 frames. ], batch size: 43, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 22:51:46,312 INFO [finetune.py:976] (3/7) Epoch 1, batch 4100, loss[loss=0.3422, simple_loss=0.3728, pruned_loss=0.1558, over 4858.00 frames. ], tot_loss[loss=0.3625, simple_loss=0.3782, pruned_loss=0.1734, over 953120.69 frames. ], batch size: 31, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 22:52:00,028 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.481e+02 2.010e+02 2.495e+02 2.957e+02 5.246e+02, threshold=4.990e+02, percent-clipped=1.0 2023-03-25 22:52:23,675 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7278, 1.7419, 1.6317, 1.0856, 2.1447, 1.8564, 1.7214, 1.6414], device='cuda:3'), covar=tensor([0.0771, 0.0716, 0.0901, 0.1191, 0.0398, 0.0898, 0.0922, 0.1229], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0130, 0.0132, 0.0121, 0.0106, 0.0131, 0.0137, 0.0159], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-25 22:52:34,227 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6846, 1.3025, 2.0668, 3.2397, 2.2934, 2.4186, 0.9329, 2.6443], device='cuda:3'), covar=tensor([0.2096, 0.2171, 0.1505, 0.0770, 0.1052, 0.1616, 0.2218, 0.0851], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0115, 0.0133, 0.0154, 0.0102, 0.0140, 0.0124, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-25 22:52:34,845 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4141.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:52:43,038 INFO [finetune.py:976] (3/7) Epoch 1, batch 4150, loss[loss=0.3624, simple_loss=0.3857, pruned_loss=0.1695, over 4893.00 frames. ], tot_loss[loss=0.3622, simple_loss=0.3791, pruned_loss=0.1726, over 954160.93 frames. ], batch size: 43, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 22:52:43,742 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5023, 3.9608, 3.7291, 1.9980, 4.0956, 2.9765, 1.0062, 2.7061], device='cuda:3'), covar=tensor([0.2359, 0.1067, 0.1520, 0.3200, 0.0694, 0.0922, 0.4321, 0.1417], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0155, 0.0160, 0.0124, 0.0150, 0.0115, 0.0142, 0.0117], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-25 22:52:49,269 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2023-03-25 22:53:43,097 INFO [finetune.py:976] (3/7) Epoch 1, batch 4200, loss[loss=0.3073, simple_loss=0.3314, pruned_loss=0.1416, over 4704.00 frames. ], tot_loss[loss=0.3597, simple_loss=0.3781, pruned_loss=0.1706, over 952834.63 frames. ], batch size: 59, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 22:53:43,819 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4202.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 22:53:43,842 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4202.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:54:02,772 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.292e+02 2.089e+02 2.432e+02 2.936e+02 5.530e+02, threshold=4.864e+02, percent-clipped=1.0 2023-03-25 22:54:44,474 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=4250.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 22:54:45,012 INFO [finetune.py:976] (3/7) Epoch 1, batch 4250, loss[loss=0.3521, simple_loss=0.3615, pruned_loss=0.1713, over 4825.00 frames. ], tot_loss[loss=0.3549, simple_loss=0.3736, pruned_loss=0.1681, over 951946.43 frames. ], batch size: 33, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 22:55:23,141 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4282.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:55:24,439 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.96 vs. limit=5.0 2023-03-25 22:55:27,287 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4288.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:55:35,149 INFO [finetune.py:976] (3/7) Epoch 1, batch 4300, loss[loss=0.329, simple_loss=0.3514, pruned_loss=0.1533, over 4719.00 frames. ], tot_loss[loss=0.35, simple_loss=0.3694, pruned_loss=0.1653, over 951198.95 frames. ], batch size: 23, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 22:55:44,372 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1159, 0.9113, 0.8045, 0.5985, 0.8949, 0.8923, 0.8745, 1.5161], device='cuda:3'), covar=tensor([4.4432, 4.1883, 3.8629, 5.9735, 3.3187, 2.6020, 4.1769, 1.3760], device='cuda:3'), in_proj_covar=tensor([0.0212, 0.0204, 0.0188, 0.0237, 0.0199, 0.0172, 0.0203, 0.0152], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:3') 2023-03-25 22:55:45,446 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.425e+02 1.950e+02 2.267e+02 2.860e+02 4.056e+02, threshold=4.534e+02, percent-clipped=0.0 2023-03-25 22:56:13,430 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4332.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 22:56:27,586 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1361, 2.3830, 2.5081, 2.3556, 2.4821, 4.6656, 2.0186, 2.5960], device='cuda:3'), covar=tensor([0.0953, 0.1240, 0.0874, 0.0955, 0.1138, 0.0127, 0.1220, 0.1245], device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0076, 0.0072, 0.0075, 0.0088, 0.0076, 0.0082, 0.0075], 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') 2023-03-25 22:56:35,735 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4349.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:56:36,850 INFO [finetune.py:976] (3/7) Epoch 1, batch 4350, loss[loss=0.2979, simple_loss=0.3252, pruned_loss=0.1353, over 4820.00 frames. ], tot_loss[loss=0.3453, simple_loss=0.3648, pruned_loss=0.1629, over 952155.97 frames. ], batch size: 30, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 22:57:17,024 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=4380.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 22:57:44,938 INFO [finetune.py:976] (3/7) Epoch 1, batch 4400, loss[loss=0.392, simple_loss=0.4103, pruned_loss=0.1869, over 4818.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.3657, pruned_loss=0.1632, over 953243.74 frames. ], batch size: 38, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 22:57:48,140 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9641, 1.5734, 1.4076, 1.0154, 1.7205, 2.5671, 2.0532, 1.6084], device='cuda:3'), covar=tensor([0.0275, 0.0536, 0.0552, 0.0621, 0.0420, 0.0204, 0.0337, 0.0450], device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0113, 0.0132, 0.0111, 0.0104, 0.0100, 0.0087, 0.0111], device='cuda:3'), out_proj_covar=tensor([6.5906e-05, 8.9473e-05, 1.0703e-04, 8.8126e-05, 8.2889e-05, 7.4688e-05, 6.7449e-05, 8.7011e-05], device='cuda:3') 2023-03-25 22:57:53,570 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-25 22:57:57,029 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.139e+02 1.978e+02 2.430e+02 2.895e+02 4.966e+02, threshold=4.860e+02, percent-clipped=1.0 2023-03-25 22:58:07,700 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.9829, 0.6990, 0.8633, 0.7332, 0.6513, 0.6842, 0.7507, 0.8236], device='cuda:3'), covar=tensor([ 4.8527, 9.2496, 5.6129, 7.9642, 8.8965, 5.5405, 10.8715, 5.5442], device='cuda:3'), in_proj_covar=tensor([0.0201, 0.0228, 0.0214, 0.0242, 0.0226, 0.0197, 0.0251, 0.0191], 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') 2023-03-25 22:58:19,953 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3592, 1.3465, 1.4001, 0.8389, 1.6579, 1.4487, 1.3995, 1.3112], device='cuda:3'), covar=tensor([0.0735, 0.0806, 0.0735, 0.1143, 0.0621, 0.0827, 0.0873, 0.1418], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0130, 0.0133, 0.0121, 0.0106, 0.0131, 0.0137, 0.0158], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-25 22:58:22,867 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3403, 1.3804, 1.0127, 1.4780, 1.6253, 1.1085, 1.9749, 1.2448], device='cuda:3'), covar=tensor([0.4068, 0.8081, 0.7685, 0.8114, 0.4579, 0.3707, 0.5449, 0.5375], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0176, 0.0215, 0.0227, 0.0191, 0.0163, 0.0174, 0.0171], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-25 22:58:26,408 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6073, 1.7452, 1.5609, 1.8852, 1.6726, 3.3018, 1.4255, 1.7521], device='cuda:3'), covar=tensor([0.1038, 0.1603, 0.1324, 0.1075, 0.1529, 0.0249, 0.1496, 0.1689], device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0076, 0.0072, 0.0075, 0.0088, 0.0076, 0.0082, 0.0075], 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') 2023-03-25 22:58:28,110 INFO [finetune.py:976] (3/7) Epoch 1, batch 4450, loss[loss=0.3437, simple_loss=0.38, pruned_loss=0.1537, over 4831.00 frames. ], tot_loss[loss=0.3486, simple_loss=0.3692, pruned_loss=0.164, over 953595.06 frames. ], batch size: 47, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 22:59:09,843 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4497.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 22:59:12,226 INFO [finetune.py:976] (3/7) Epoch 1, batch 4500, loss[loss=0.2675, simple_loss=0.3052, pruned_loss=0.115, over 4735.00 frames. ], tot_loss[loss=0.35, simple_loss=0.3712, pruned_loss=0.1644, over 953828.72 frames. ], batch size: 23, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 22:59:29,322 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.227e+02 2.111e+02 2.516e+02 2.889e+02 5.762e+02, threshold=5.032e+02, percent-clipped=1.0 2023-03-25 23:00:15,077 INFO [finetune.py:976] (3/7) Epoch 1, batch 4550, loss[loss=0.3235, simple_loss=0.3773, pruned_loss=0.1348, over 4811.00 frames. ], tot_loss[loss=0.3494, simple_loss=0.3715, pruned_loss=0.1636, over 953591.04 frames. ], batch size: 38, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:00:55,783 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4582.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:01:14,554 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3551, 1.3835, 0.9868, 1.5022, 1.6172, 1.1722, 1.9796, 1.3096], device='cuda:3'), covar=tensor([0.4647, 0.9242, 0.8743, 0.9335, 0.5475, 0.4323, 0.6710, 0.6795], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0178, 0.0218, 0.0230, 0.0192, 0.0165, 0.0176, 0.0173], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-25 23:01:25,593 INFO [finetune.py:976] (3/7) Epoch 1, batch 4600, loss[loss=0.2869, simple_loss=0.3079, pruned_loss=0.1329, over 4738.00 frames. ], tot_loss[loss=0.3453, simple_loss=0.3682, pruned_loss=0.1611, over 953460.56 frames. ], batch size: 26, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:01:38,112 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.209e+02 2.082e+02 2.456e+02 3.064e+02 5.977e+02, threshold=4.911e+02, percent-clipped=1.0 2023-03-25 23:01:59,353 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=4630.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:02:07,940 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.4142, 3.7797, 3.9382, 4.2765, 4.1109, 3.8910, 4.5273, 1.4888], device='cuda:3'), covar=tensor([0.0683, 0.0825, 0.0767, 0.0912, 0.1218, 0.1346, 0.0587, 0.5017], device='cuda:3'), in_proj_covar=tensor([0.0372, 0.0247, 0.0269, 0.0297, 0.0350, 0.0292, 0.0313, 0.0304], 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') 2023-03-25 23:02:18,010 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4644.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:02:28,981 INFO [finetune.py:976] (3/7) Epoch 1, batch 4650, loss[loss=0.2587, simple_loss=0.3006, pruned_loss=0.1084, over 4807.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.3617, pruned_loss=0.1568, over 953104.45 frames. ], batch size: 25, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:03:06,338 INFO [finetune.py:976] (3/7) Epoch 1, batch 4700, loss[loss=0.2369, simple_loss=0.2886, pruned_loss=0.09265, over 4824.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3569, pruned_loss=0.1534, over 955588.13 frames. ], batch size: 39, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:03:20,653 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.265e+02 1.870e+02 2.224e+02 2.796e+02 5.273e+02, threshold=4.448e+02, percent-clipped=2.0 2023-03-25 23:03:29,906 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-25 23:03:59,444 INFO [finetune.py:976] (3/7) Epoch 1, batch 4750, loss[loss=0.2776, simple_loss=0.3106, pruned_loss=0.1223, over 4749.00 frames. ], tot_loss[loss=0.3307, simple_loss=0.3554, pruned_loss=0.153, over 957956.53 frames. ], batch size: 27, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:04:16,160 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.04 vs. limit=2.0 2023-03-25 23:04:35,226 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3380, 1.1114, 0.8546, 1.0713, 1.0838, 0.9708, 1.0291, 1.8414], device='cuda:3'), covar=tensor([5.1083, 5.5190, 4.4145, 7.4420, 4.3337, 3.3228, 5.4909, 1.5475], device='cuda:3'), in_proj_covar=tensor([0.0216, 0.0207, 0.0191, 0.0240, 0.0202, 0.0174, 0.0206, 0.0154], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:3') 2023-03-25 23:04:36,389 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4797.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:04:39,215 INFO [finetune.py:976] (3/7) Epoch 1, batch 4800, loss[loss=0.3536, simple_loss=0.3778, pruned_loss=0.1647, over 4933.00 frames. ], tot_loss[loss=0.3344, simple_loss=0.3588, pruned_loss=0.155, over 955424.11 frames. ], batch size: 33, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:04:56,837 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.406e+02 2.096e+02 2.556e+02 3.186e+02 5.883e+02, threshold=5.111e+02, percent-clipped=4.0 2023-03-25 23:05:32,163 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=4845.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:05:41,981 INFO [finetune.py:976] (3/7) Epoch 1, batch 4850, loss[loss=0.2641, simple_loss=0.2849, pruned_loss=0.1217, over 4419.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.3632, pruned_loss=0.1563, over 956708.65 frames. ], batch size: 19, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:06:06,807 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-03-25 23:06:28,852 INFO [finetune.py:976] (3/7) Epoch 1, batch 4900, loss[loss=0.3351, simple_loss=0.3569, pruned_loss=0.1566, over 4816.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3637, pruned_loss=0.1563, over 954252.56 frames. ], batch size: 40, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:06:45,501 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.521e+02 2.056e+02 2.408e+02 2.893e+02 5.886e+02, threshold=4.817e+02, percent-clipped=2.0 2023-03-25 23:06:47,347 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7808, 1.5799, 2.0587, 1.4704, 1.8326, 2.0410, 1.5612, 2.1826], device='cuda:3'), covar=tensor([0.2022, 0.2189, 0.1491, 0.1834, 0.1098, 0.1602, 0.2542, 0.1045], device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0200, 0.0198, 0.0188, 0.0170, 0.0214, 0.0207, 0.0190], 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') 2023-03-25 23:07:15,535 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4944.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:07:19,811 INFO [finetune.py:976] (3/7) Epoch 1, batch 4950, loss[loss=0.3217, simple_loss=0.3429, pruned_loss=0.1502, over 4763.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3639, pruned_loss=0.1562, over 954353.52 frames. ], batch size: 23, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:07:48,973 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4786, 1.5445, 1.0738, 1.6263, 1.7896, 1.2155, 2.1786, 1.3725], device='cuda:3'), covar=tensor([0.4281, 0.8683, 0.8165, 0.8938, 0.5078, 0.3983, 0.6057, 0.6160], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0181, 0.0221, 0.0234, 0.0195, 0.0168, 0.0180, 0.0175], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:3') 2023-03-25 23:08:01,149 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 2023-03-25 23:08:11,782 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=4992.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:08:20,320 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7269, 1.5196, 2.1514, 3.1152, 2.3231, 2.2975, 1.1142, 2.5853], device='cuda:3'), covar=tensor([0.1812, 0.1555, 0.1225, 0.0528, 0.0806, 0.1843, 0.1882, 0.0626], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0116, 0.0134, 0.0156, 0.0103, 0.0142, 0.0126, 0.0106], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-25 23:08:22,745 INFO [finetune.py:976] (3/7) Epoch 1, batch 5000, loss[loss=0.3032, simple_loss=0.3238, pruned_loss=0.1413, over 4142.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3604, pruned_loss=0.1534, over 953236.65 frames. ], batch size: 65, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:08:32,727 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.012e+02 2.131e+02 2.461e+02 3.038e+02 5.796e+02, threshold=4.923e+02, percent-clipped=4.0 2023-03-25 23:08:49,825 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0748, 1.6874, 1.5765, 1.1180, 1.9520, 2.7337, 2.2005, 1.6969], device='cuda:3'), covar=tensor([0.0318, 0.0527, 0.0591, 0.0636, 0.0398, 0.0165, 0.0317, 0.0439], device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0111, 0.0131, 0.0110, 0.0103, 0.0098, 0.0087, 0.0109], device='cuda:3'), out_proj_covar=tensor([6.5078e-05, 8.8045e-05, 1.0606e-04, 8.7581e-05, 8.2110e-05, 7.3767e-05, 6.7213e-05, 8.5788e-05], device='cuda:3') 2023-03-25 23:08:59,969 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9772, 1.7097, 2.1705, 3.4491, 2.5680, 2.5860, 0.9488, 2.8502], device='cuda:3'), covar=tensor([0.1665, 0.1464, 0.1274, 0.0551, 0.0784, 0.1911, 0.1946, 0.0629], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0116, 0.0135, 0.0157, 0.0103, 0.0142, 0.0127, 0.0106], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-25 23:09:20,249 INFO [finetune.py:976] (3/7) Epoch 1, batch 5050, loss[loss=0.2999, simple_loss=0.3342, pruned_loss=0.1327, over 4937.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3552, pruned_loss=0.1504, over 954010.79 frames. ], batch size: 33, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:09:25,932 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-03-25 23:09:54,624 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.0872, 4.4373, 4.5879, 4.8975, 4.7436, 4.5919, 5.1877, 1.4747], device='cuda:3'), covar=tensor([0.0616, 0.0685, 0.0621, 0.0784, 0.1147, 0.1129, 0.0531, 0.5122], device='cuda:3'), in_proj_covar=tensor([0.0370, 0.0245, 0.0268, 0.0296, 0.0347, 0.0290, 0.0314, 0.0302], 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') 2023-03-25 23:10:01,168 INFO [finetune.py:976] (3/7) Epoch 1, batch 5100, loss[loss=0.3277, simple_loss=0.3419, pruned_loss=0.1568, over 4032.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3499, pruned_loss=0.1481, over 952215.73 frames. ], batch size: 17, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:10:09,454 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 1.918e+02 2.379e+02 2.966e+02 8.444e+02, threshold=4.758e+02, percent-clipped=2.0 2023-03-25 23:10:34,835 INFO [finetune.py:976] (3/7) Epoch 1, batch 5150, loss[loss=0.2468, simple_loss=0.2985, pruned_loss=0.09757, over 4762.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3488, pruned_loss=0.1471, over 953844.84 frames. ], batch size: 28, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:10:48,148 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-25 23:11:01,700 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3680, 1.4987, 1.5176, 1.7362, 1.6325, 3.4471, 1.1810, 1.7183], device='cuda:3'), covar=tensor([0.1210, 0.1862, 0.1487, 0.1256, 0.1751, 0.0301, 0.1848, 0.2006], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0077, 0.0074, 0.0076, 0.0089, 0.0078, 0.0083, 0.0076], 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') 2023-03-25 23:11:14,842 INFO [finetune.py:976] (3/7) Epoch 1, batch 5200, loss[loss=0.3305, simple_loss=0.3653, pruned_loss=0.1479, over 4828.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3547, pruned_loss=0.1492, over 955011.31 frames. ], batch size: 39, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:11:24,773 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.601e+02 2.261e+02 2.570e+02 3.078e+02 5.221e+02, threshold=5.140e+02, percent-clipped=2.0 2023-03-25 23:12:03,342 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8987, 2.2894, 2.3229, 1.2196, 2.3322, 2.0660, 1.6823, 2.1213], device='cuda:3'), covar=tensor([0.0933, 0.1065, 0.1847, 0.2668, 0.1741, 0.2274, 0.2228, 0.1481], device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0177, 0.0190, 0.0174, 0.0197, 0.0196, 0.0200, 0.0187], 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') 2023-03-25 23:12:06,970 INFO [finetune.py:976] (3/7) Epoch 1, batch 5250, loss[loss=0.3211, simple_loss=0.3532, pruned_loss=0.1445, over 4889.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3568, pruned_loss=0.1497, over 955154.27 frames. ], batch size: 32, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:12:40,882 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6860, 1.7646, 1.7060, 1.8721, 1.1046, 4.2104, 1.5889, 2.0715], device='cuda:3'), covar=tensor([0.3531, 0.2330, 0.1990, 0.2126, 0.2184, 0.0107, 0.2752, 0.1550], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0105, 0.0111, 0.0112, 0.0105, 0.0090, 0.0092, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-25 23:12:55,131 INFO [finetune.py:976] (3/7) Epoch 1, batch 5300, loss[loss=0.2772, simple_loss=0.3218, pruned_loss=0.1163, over 4866.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3575, pruned_loss=0.1497, over 955054.29 frames. ], batch size: 31, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:13:08,326 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.321e+02 2.032e+02 2.465e+02 2.907e+02 4.480e+02, threshold=4.930e+02, percent-clipped=0.0 2023-03-25 23:13:39,633 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8330, 1.1525, 0.9294, 1.5727, 2.1449, 1.2254, 1.3397, 1.7183], device='cuda:3'), covar=tensor([0.1621, 0.2286, 0.2249, 0.1285, 0.2014, 0.2035, 0.1502, 0.2000], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0096, 0.0115, 0.0092, 0.0123, 0.0095, 0.0098, 0.0093], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-25 23:13:48,367 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-25 23:13:49,133 INFO [finetune.py:976] (3/7) Epoch 1, batch 5350, loss[loss=0.3049, simple_loss=0.3419, pruned_loss=0.134, over 4753.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3556, pruned_loss=0.1477, over 953647.65 frames. ], batch size: 26, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:14:48,106 INFO [finetune.py:976] (3/7) Epoch 1, batch 5400, loss[loss=0.3213, simple_loss=0.3478, pruned_loss=0.1474, over 4883.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3509, pruned_loss=0.1455, over 953358.91 frames. ], batch size: 35, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:14:55,971 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.226e+02 1.939e+02 2.339e+02 2.729e+02 4.650e+02, threshold=4.678e+02, percent-clipped=0.0 2023-03-25 23:15:03,501 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2822, 2.8201, 2.7684, 1.2378, 2.9897, 2.1533, 0.6026, 1.8212], device='cuda:3'), covar=tensor([0.2565, 0.2037, 0.1834, 0.3708, 0.1266, 0.1144, 0.4297, 0.1729], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0157, 0.0161, 0.0125, 0.0151, 0.0115, 0.0143, 0.0117], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-25 23:15:12,221 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-25 23:15:36,516 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6565, 1.7515, 1.6357, 1.0341, 2.0710, 1.7986, 1.6815, 1.6448], device='cuda:3'), covar=tensor([0.0771, 0.0737, 0.0837, 0.1157, 0.0452, 0.0861, 0.0903, 0.1174], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0131, 0.0136, 0.0125, 0.0107, 0.0135, 0.0140, 0.0160], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-25 23:15:39,350 INFO [finetune.py:976] (3/7) Epoch 1, batch 5450, loss[loss=0.2721, simple_loss=0.3196, pruned_loss=0.1123, over 4906.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3451, pruned_loss=0.1415, over 954390.84 frames. ], batch size: 35, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:16:00,924 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-03-25 23:16:05,962 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.0038, 4.3688, 4.5179, 4.8078, 4.6715, 4.5363, 5.1217, 1.4853], device='cuda:3'), covar=tensor([0.0669, 0.0665, 0.0631, 0.0718, 0.1133, 0.1116, 0.0472, 0.5240], device='cuda:3'), in_proj_covar=tensor([0.0369, 0.0244, 0.0269, 0.0296, 0.0347, 0.0289, 0.0312, 0.0301], 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') 2023-03-25 23:16:19,389 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3203, 2.8504, 2.0899, 1.7111, 3.1200, 2.9349, 2.5474, 2.4718], device='cuda:3'), covar=tensor([0.0917, 0.0576, 0.1038, 0.1222, 0.0366, 0.0786, 0.0892, 0.1010], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0132, 0.0137, 0.0126, 0.0107, 0.0136, 0.0141, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-25 23:16:31,592 INFO [finetune.py:976] (3/7) Epoch 1, batch 5500, loss[loss=0.3372, simple_loss=0.3582, pruned_loss=0.1581, over 4765.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3407, pruned_loss=0.1394, over 952626.10 frames. ], batch size: 54, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:16:38,571 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.90 vs. limit=5.0 2023-03-25 23:16:45,961 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.302e+02 2.026e+02 2.277e+02 2.875e+02 1.009e+03, threshold=4.553e+02, percent-clipped=5.0 2023-03-25 23:16:49,222 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-25 23:17:20,557 INFO [finetune.py:976] (3/7) Epoch 1, batch 5550, loss[loss=0.2926, simple_loss=0.3225, pruned_loss=0.1314, over 4774.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3435, pruned_loss=0.141, over 953164.03 frames. ], batch size: 26, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:17:57,720 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2023-03-25 23:18:01,091 INFO [finetune.py:976] (3/7) Epoch 1, batch 5600, loss[loss=0.3334, simple_loss=0.3629, pruned_loss=0.1519, over 4767.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3483, pruned_loss=0.1424, over 954829.62 frames. ], batch size: 27, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:18:19,452 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.468e+02 1.839e+02 2.287e+02 2.793e+02 4.099e+02, threshold=4.573e+02, percent-clipped=0.0 2023-03-25 23:18:59,405 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5650.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 23:18:59,647 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-25 23:18:59,874 INFO [finetune.py:976] (3/7) Epoch 1, batch 5650, loss[loss=0.3513, simple_loss=0.3917, pruned_loss=0.1555, over 4792.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3512, pruned_loss=0.1431, over 953921.81 frames. ], batch size: 45, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:19:16,085 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.88 vs. limit=5.0 2023-03-25 23:19:21,424 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6858, 2.9620, 2.1702, 2.0562, 3.4396, 3.3351, 2.7713, 2.6849], device='cuda:3'), covar=tensor([0.0798, 0.0556, 0.1031, 0.1104, 0.0404, 0.0640, 0.0850, 0.0918], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0132, 0.0137, 0.0126, 0.0107, 0.0135, 0.0142, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-25 23:19:35,520 INFO [finetune.py:976] (3/7) Epoch 1, batch 5700, loss[loss=0.2795, simple_loss=0.3006, pruned_loss=0.1292, over 4319.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3464, pruned_loss=0.1423, over 935921.18 frames. ], batch size: 18, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:19:35,991 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.01 vs. limit=5.0 2023-03-25 23:19:41,526 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5711.0, num_to_drop=1, layers_to_drop={3} 2023-03-25 23:19:43,180 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.079e+02 1.818e+02 2.245e+02 2.685e+02 4.321e+02, threshold=4.489e+02, percent-clipped=0.0 2023-03-25 23:20:08,412 INFO [finetune.py:976] (3/7) Epoch 2, batch 0, loss[loss=0.267, simple_loss=0.3077, pruned_loss=0.1131, over 4753.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3077, pruned_loss=0.1131, over 4753.00 frames. ], batch size: 27, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:20:08,413 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-25 23:20:27,694 INFO [finetune.py:1010] (3/7) Epoch 2, validation: loss=0.2224, simple_loss=0.2847, pruned_loss=0.08, over 2265189.00 frames. 2023-03-25 23:20:27,695 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 5939MB 2023-03-25 23:20:47,818 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.84 vs. limit=5.0 2023-03-25 23:20:56,834 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5755.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:21:22,829 INFO [finetune.py:976] (3/7) Epoch 2, batch 50, loss[loss=0.3372, simple_loss=0.3587, pruned_loss=0.1579, over 4763.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3444, pruned_loss=0.1389, over 216813.80 frames. ], batch size: 51, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:21:50,519 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-25 23:21:54,356 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.238e+02 1.870e+02 2.317e+02 2.912e+02 7.564e+02, threshold=4.633e+02, percent-clipped=3.0 2023-03-25 23:21:55,701 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5816.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:22:11,072 INFO [finetune.py:976] (3/7) Epoch 2, batch 100, loss[loss=0.2984, simple_loss=0.3375, pruned_loss=0.1297, over 4896.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3434, pruned_loss=0.1398, over 380540.01 frames. ], batch size: 43, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:22:36,193 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5858.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:22:37,706 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-25 23:22:41,060 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4101, 2.8910, 1.8500, 1.6840, 3.2042, 2.8468, 2.5201, 2.4223], device='cuda:3'), covar=tensor([0.0874, 0.0529, 0.1079, 0.1193, 0.0310, 0.0778, 0.0919, 0.1008], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0132, 0.0138, 0.0126, 0.0107, 0.0136, 0.0142, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-25 23:22:49,648 INFO [finetune.py:976] (3/7) Epoch 2, batch 150, loss[loss=0.315, simple_loss=0.3382, pruned_loss=0.1459, over 4935.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3345, pruned_loss=0.134, over 507345.02 frames. ], batch size: 38, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:23:18,195 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.293e+02 1.883e+02 2.329e+02 2.858e+02 5.160e+02, threshold=4.657e+02, percent-clipped=2.0 2023-03-25 23:23:21,786 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6782, 1.5087, 2.0632, 3.2602, 2.3710, 2.4463, 0.9756, 2.5344], device='cuda:3'), covar=tensor([0.2096, 0.1808, 0.1536, 0.0641, 0.0938, 0.1504, 0.2207, 0.0797], device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0119, 0.0137, 0.0160, 0.0105, 0.0145, 0.0129, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-25 23:23:21,822 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5919.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:23:28,108 INFO [finetune.py:976] (3/7) Epoch 2, batch 200, loss[loss=0.2576, simple_loss=0.3142, pruned_loss=0.1005, over 4840.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3309, pruned_loss=0.1317, over 609618.80 frames. ], batch size: 47, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:23:30,306 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-03-25 23:24:01,222 INFO [finetune.py:976] (3/7) Epoch 2, batch 250, loss[loss=0.3302, simple_loss=0.3718, pruned_loss=0.1443, over 4920.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.335, pruned_loss=0.1332, over 685767.59 frames. ], batch size: 42, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:24:09,406 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.2227, 3.6760, 3.8147, 4.0546, 3.9661, 3.7475, 4.2982, 1.3644], device='cuda:3'), covar=tensor([0.0729, 0.0724, 0.0697, 0.0904, 0.1157, 0.1431, 0.0664, 0.5202], device='cuda:3'), in_proj_covar=tensor([0.0371, 0.0247, 0.0271, 0.0299, 0.0350, 0.0292, 0.0313, 0.0305], 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') 2023-03-25 23:24:11,259 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.0027, 3.5203, 3.6646, 3.8448, 3.7672, 3.6004, 4.0539, 1.7876], device='cuda:3'), covar=tensor([0.0677, 0.0689, 0.0660, 0.0875, 0.1029, 0.1230, 0.0652, 0.4259], device='cuda:3'), in_proj_covar=tensor([0.0371, 0.0247, 0.0271, 0.0299, 0.0350, 0.0292, 0.0313, 0.0305], 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') 2023-03-25 23:24:20,398 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5990.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:24:42,066 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6006.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 23:24:48,585 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.078e+02 1.968e+02 2.365e+02 2.842e+02 7.361e+02, threshold=4.731e+02, percent-clipped=2.0 2023-03-25 23:24:49,314 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5750, 1.5510, 1.4894, 1.5974, 0.8908, 2.9235, 0.9765, 1.5155], device='cuda:3'), covar=tensor([0.3673, 0.2574, 0.2274, 0.2446, 0.2405, 0.0279, 0.3155, 0.1699], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0107, 0.0113, 0.0113, 0.0108, 0.0092, 0.0094, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0003, 0.0004, 0.0004], device='cuda:3') 2023-03-25 23:25:01,946 INFO [finetune.py:976] (3/7) Epoch 2, batch 300, loss[loss=0.4122, simple_loss=0.4172, pruned_loss=0.2036, over 4919.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.342, pruned_loss=0.1372, over 744844.62 frames. ], batch size: 36, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:25:29,065 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6043.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:25:33,849 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6051.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:26:11,334 INFO [finetune.py:976] (3/7) Epoch 2, batch 350, loss[loss=0.2609, simple_loss=0.3164, pruned_loss=0.1027, over 4831.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3453, pruned_loss=0.1385, over 790153.85 frames. ], batch size: 49, lr: 4.00e-03, grad_scale: 32.0 2023-03-25 23:26:35,000 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6104.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:26:39,146 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6111.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:26:41,481 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.275e+02 2.097e+02 2.536e+02 2.955e+02 5.135e+02, threshold=5.071e+02, percent-clipped=1.0 2023-03-25 23:26:59,637 INFO [finetune.py:976] (3/7) Epoch 2, batch 400, loss[loss=0.2689, simple_loss=0.3271, pruned_loss=0.1054, over 4808.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3453, pruned_loss=0.1375, over 826791.15 frames. ], batch size: 40, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:27:09,952 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6135.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:27:49,736 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6170.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:27:59,431 INFO [finetune.py:976] (3/7) Epoch 2, batch 450, loss[loss=0.2746, simple_loss=0.3181, pruned_loss=0.1155, over 4848.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3436, pruned_loss=0.1359, over 856440.08 frames. ], batch size: 44, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:28:02,511 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4107, 1.3784, 1.4117, 0.8930, 1.6624, 1.4553, 1.4185, 1.3405], device='cuda:3'), covar=tensor([0.0734, 0.0909, 0.0795, 0.1133, 0.0668, 0.0858, 0.0894, 0.1438], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0133, 0.0138, 0.0127, 0.0108, 0.0137, 0.0143, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-25 23:28:14,488 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6196.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:28:15,031 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7660, 3.8264, 3.7519, 1.8449, 3.9533, 2.8664, 0.7921, 2.6773], device='cuda:3'), covar=tensor([0.2144, 0.1622, 0.1514, 0.3205, 0.0967, 0.1010, 0.4368, 0.1490], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0160, 0.0163, 0.0127, 0.0153, 0.0117, 0.0145, 0.0119], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-25 23:28:18,022 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8462, 1.2957, 0.8906, 1.5974, 2.1437, 1.2033, 1.3097, 1.6069], device='cuda:3'), covar=tensor([0.1678, 0.2234, 0.2302, 0.1320, 0.1981, 0.1978, 0.1639, 0.2219], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0097, 0.0115, 0.0092, 0.0123, 0.0096, 0.0098, 0.0093], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-25 23:28:25,687 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6203.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:28:33,026 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6211.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:28:34,828 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6214.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:28:35,342 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.140e+02 1.932e+02 2.244e+02 2.718e+02 3.817e+02, threshold=4.487e+02, percent-clipped=0.0 2023-03-25 23:28:42,433 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-25 23:28:45,039 INFO [finetune.py:976] (3/7) Epoch 2, batch 500, loss[loss=0.2259, simple_loss=0.2777, pruned_loss=0.08711, over 4780.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3386, pruned_loss=0.1326, over 877956.46 frames. ], batch size: 29, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:28:52,705 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6231.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:28:53,095 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-25 23:28:55,253 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.76 vs. limit=5.0 2023-03-25 23:29:20,582 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3066, 0.8820, 1.0488, 1.0405, 0.9583, 0.9452, 1.0095, 1.0320], device='cuda:3'), covar=tensor([2.7607, 5.6443, 3.6981, 4.7558, 5.1595, 3.3612, 6.7844, 3.6221], device='cuda:3'), in_proj_covar=tensor([0.0215, 0.0247, 0.0231, 0.0260, 0.0239, 0.0211, 0.0269, 0.0205], 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') 2023-03-25 23:29:25,577 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6264.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:29:26,783 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6266.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:29:35,130 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6272.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:29:38,740 INFO [finetune.py:976] (3/7) Epoch 2, batch 550, loss[loss=0.3325, simple_loss=0.3416, pruned_loss=0.1617, over 4708.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3343, pruned_loss=0.1308, over 895451.06 frames. ], batch size: 59, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:30:17,693 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6306.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 23:30:29,437 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.346e+02 1.947e+02 2.353e+02 2.718e+02 5.175e+02, threshold=4.705e+02, percent-clipped=1.0 2023-03-25 23:30:40,828 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1512, 0.5512, 1.0416, 0.9055, 0.8646, 0.8878, 0.8304, 1.0643], device='cuda:3'), covar=tensor([2.8274, 5.5727, 3.8085, 4.4336, 4.6777, 3.1783, 6.0173, 3.2648], device='cuda:3'), in_proj_covar=tensor([0.0215, 0.0248, 0.0233, 0.0261, 0.0240, 0.0211, 0.0270, 0.0206], 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') 2023-03-25 23:30:47,519 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6327.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:30:48,005 INFO [finetune.py:976] (3/7) Epoch 2, batch 600, loss[loss=0.3051, simple_loss=0.3441, pruned_loss=0.133, over 4833.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3341, pruned_loss=0.1308, over 910887.30 frames. ], batch size: 49, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:31:07,296 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6346.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:31:13,144 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=6354.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 23:31:28,047 INFO [finetune.py:976] (3/7) Epoch 2, batch 650, loss[loss=0.3399, simple_loss=0.3744, pruned_loss=0.1527, over 4902.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3372, pruned_loss=0.132, over 921934.39 frames. ], batch size: 37, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:31:42,402 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6399.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:31:51,311 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6411.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:31:53,613 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.435e+02 2.011e+02 2.373e+02 2.999e+02 4.783e+02, threshold=4.746e+02, percent-clipped=1.0 2023-03-25 23:31:59,785 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0808, 1.7799, 2.7316, 4.0657, 3.0125, 2.6923, 1.0843, 3.3105], device='cuda:3'), covar=tensor([0.2076, 0.1733, 0.1450, 0.0501, 0.0846, 0.1565, 0.2208, 0.0643], device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0119, 0.0137, 0.0160, 0.0104, 0.0144, 0.0129, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-25 23:32:01,499 INFO [finetune.py:976] (3/7) Epoch 2, batch 700, loss[loss=0.3746, simple_loss=0.3959, pruned_loss=0.1767, over 4815.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3402, pruned_loss=0.1329, over 931092.06 frames. ], batch size: 40, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:32:22,200 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=6459.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:32:22,886 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5074, 1.4251, 1.1783, 1.5180, 1.7410, 1.3251, 2.0416, 1.4055], device='cuda:3'), covar=tensor([0.3790, 0.7231, 0.7016, 0.6933, 0.4379, 0.3393, 0.5188, 0.5069], device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0188, 0.0229, 0.0243, 0.0202, 0.0174, 0.0191, 0.0182], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-25 23:32:24,548 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6462.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:32:27,329 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8837, 1.6333, 2.4083, 3.5744, 2.6709, 2.5797, 0.7546, 2.9309], device='cuda:3'), covar=tensor([0.1856, 0.1593, 0.1371, 0.0563, 0.0837, 0.1710, 0.2226, 0.0659], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0119, 0.0138, 0.0161, 0.0105, 0.0145, 0.0130, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-25 23:32:27,955 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6467.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:32:34,494 INFO [finetune.py:976] (3/7) Epoch 2, batch 750, loss[loss=0.3148, simple_loss=0.3576, pruned_loss=0.1359, over 4920.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3421, pruned_loss=0.1336, over 937013.73 frames. ], batch size: 42, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:32:42,503 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6491.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:32:58,345 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6514.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:32:58,815 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.464e+02 1.998e+02 2.267e+02 2.688e+02 5.596e+02, threshold=4.534e+02, percent-clipped=2.0 2023-03-25 23:33:03,635 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1507, 3.6369, 3.7107, 4.0607, 3.8910, 3.6417, 4.2503, 1.3383], device='cuda:3'), covar=tensor([0.0754, 0.0785, 0.0831, 0.0868, 0.1243, 0.1406, 0.0712, 0.4908], device='cuda:3'), in_proj_covar=tensor([0.0370, 0.0247, 0.0270, 0.0298, 0.0348, 0.0291, 0.0313, 0.0304], 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') 2023-03-25 23:33:04,282 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6523.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 23:33:06,034 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6526.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:33:07,179 INFO [finetune.py:976] (3/7) Epoch 2, batch 800, loss[loss=0.3476, simple_loss=0.3638, pruned_loss=0.1657, over 4856.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3411, pruned_loss=0.1331, over 940831.51 frames. ], batch size: 44, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:33:07,355 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6528.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:33:15,899 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-25 23:33:40,455 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6559.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:33:42,337 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=6562.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:33:46,517 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6567.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:33:58,060 INFO [finetune.py:976] (3/7) Epoch 2, batch 850, loss[loss=0.303, simple_loss=0.3352, pruned_loss=0.1354, over 4798.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.338, pruned_loss=0.1314, over 944208.67 frames. ], batch size: 29, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:34:37,100 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.216e+02 1.789e+02 2.218e+02 2.697e+02 5.451e+02, threshold=4.436e+02, percent-clipped=1.0 2023-03-25 23:34:40,795 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1851, 1.6252, 1.2368, 2.0961, 2.6071, 1.7870, 1.8640, 2.1364], device='cuda:3'), covar=tensor([0.1318, 0.1814, 0.1838, 0.1002, 0.1548, 0.1581, 0.1252, 0.1559], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0096, 0.0114, 0.0092, 0.0123, 0.0095, 0.0098, 0.0093], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-25 23:34:42,626 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6622.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:34:46,286 INFO [finetune.py:976] (3/7) Epoch 2, batch 900, loss[loss=0.2712, simple_loss=0.3117, pruned_loss=0.1154, over 4869.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3321, pruned_loss=0.1289, over 944254.51 frames. ], batch size: 34, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:34:47,118 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6629.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:34:48,392 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4771, 1.1676, 1.1979, 1.0679, 1.5915, 1.6577, 1.4019, 1.1490], device='cuda:3'), covar=tensor([0.0302, 0.0400, 0.0522, 0.0400, 0.0249, 0.0303, 0.0309, 0.0418], device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0111, 0.0130, 0.0111, 0.0103, 0.0098, 0.0087, 0.0108], device='cuda:3'), out_proj_covar=tensor([6.4389e-05, 8.7854e-05, 1.0551e-04, 8.7862e-05, 8.1637e-05, 7.3000e-05, 6.7356e-05, 8.4602e-05], device='cuda:3') 2023-03-25 23:35:02,804 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6646.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:35:25,037 INFO [finetune.py:976] (3/7) Epoch 2, batch 950, loss[loss=0.293, simple_loss=0.3319, pruned_loss=0.127, over 4805.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3304, pruned_loss=0.1281, over 947508.71 frames. ], batch size: 25, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:35:32,472 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6690.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:35:33,094 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5877, 1.4804, 1.1669, 1.5779, 1.7126, 1.3418, 2.0374, 1.5079], device='cuda:3'), covar=tensor([0.3051, 0.6988, 0.6209, 0.6419, 0.4332, 0.3093, 0.5841, 0.4532], device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0189, 0.0230, 0.0243, 0.0203, 0.0175, 0.0192, 0.0182], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:3') 2023-03-25 23:35:34,864 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=6694.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:35:37,912 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6699.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:35:38,555 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8504, 1.4554, 2.1582, 1.4720, 1.8620, 2.0138, 1.4326, 2.1604], device='cuda:3'), covar=tensor([0.1672, 0.2284, 0.1445, 0.2203, 0.0997, 0.1659, 0.2829, 0.1035], device='cuda:3'), in_proj_covar=tensor([0.0202, 0.0204, 0.0202, 0.0193, 0.0175, 0.0221, 0.0212, 0.0195], 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') 2023-03-25 23:35:48,888 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.136e+02 1.729e+02 2.077e+02 2.706e+02 4.933e+02, threshold=4.155e+02, percent-clipped=1.0 2023-03-25 23:36:03,624 INFO [finetune.py:976] (3/7) Epoch 2, batch 1000, loss[loss=0.3, simple_loss=0.3486, pruned_loss=0.1257, over 4819.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3348, pruned_loss=0.1307, over 950295.52 frames. ], batch size: 39, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:36:03,977 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-03-25 23:36:22,930 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=6747.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:36:57,204 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3240, 1.4278, 1.1288, 1.9883, 2.2974, 1.9010, 1.6923, 2.0106], device='cuda:3'), covar=tensor([0.1578, 0.2335, 0.2121, 0.1250, 0.2192, 0.1923, 0.1485, 0.2131], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0096, 0.0115, 0.0092, 0.0123, 0.0096, 0.0098, 0.0093], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-25 23:37:01,237 INFO [finetune.py:976] (3/7) Epoch 2, batch 1050, loss[loss=0.2955, simple_loss=0.3429, pruned_loss=0.124, over 4798.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3403, pruned_loss=0.133, over 952543.23 frames. ], batch size: 45, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:37:09,229 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6791.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:37:19,969 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2595, 0.7042, 1.1015, 0.9530, 0.9362, 0.9215, 0.9015, 1.0814], device='cuda:3'), covar=tensor([2.4620, 4.5443, 3.1472, 3.8900, 4.1012, 2.8187, 5.1757, 2.9696], device='cuda:3'), in_proj_covar=tensor([0.0217, 0.0250, 0.0236, 0.0263, 0.0241, 0.0213, 0.0271, 0.0209], 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') 2023-03-25 23:37:29,664 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.307e+02 2.050e+02 2.547e+02 2.958e+02 5.414e+02, threshold=5.095e+02, percent-clipped=8.0 2023-03-25 23:37:37,340 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6818.0, num_to_drop=1, layers_to_drop={3} 2023-03-25 23:37:40,270 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6823.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:37:47,818 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6826.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:37:48,961 INFO [finetune.py:976] (3/7) Epoch 2, batch 1100, loss[loss=0.3161, simple_loss=0.3553, pruned_loss=0.1385, over 4823.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3412, pruned_loss=0.1326, over 954974.34 frames. ], batch size: 47, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:37:56,677 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=6839.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:38:01,548 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3391, 1.2230, 1.5530, 2.3596, 1.6216, 2.0583, 0.9657, 1.9309], device='cuda:3'), covar=tensor([0.2140, 0.1851, 0.1354, 0.0775, 0.1134, 0.1298, 0.1726, 0.0884], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0120, 0.0139, 0.0161, 0.0105, 0.0145, 0.0130, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-25 23:38:18,881 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6859.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:38:23,790 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6867.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:38:28,922 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=6874.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:38:31,313 INFO [finetune.py:976] (3/7) Epoch 2, batch 1150, loss[loss=0.33, simple_loss=0.3676, pruned_loss=0.1461, over 4884.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3424, pruned_loss=0.1333, over 955340.76 frames. ], batch size: 35, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:38:31,431 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6878.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 23:38:52,178 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=6907.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:38:53,405 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6909.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:38:56,971 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.223e+02 2.014e+02 2.348e+02 2.865e+02 5.036e+02, threshold=4.696e+02, percent-clipped=0.0 2023-03-25 23:38:57,046 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=6915.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:39:07,706 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6922.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:39:17,348 INFO [finetune.py:976] (3/7) Epoch 2, batch 1200, loss[loss=0.3103, simple_loss=0.3389, pruned_loss=0.1409, over 4878.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3391, pruned_loss=0.1316, over 955766.22 frames. ], batch size: 31, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:39:31,058 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6939.0, num_to_drop=1, layers_to_drop={3} 2023-03-25 23:39:49,900 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=6970.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:39:49,963 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6970.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:39:56,178 INFO [finetune.py:976] (3/7) Epoch 2, batch 1250, loss[loss=0.2634, simple_loss=0.3074, pruned_loss=0.1097, over 4932.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3336, pruned_loss=0.1283, over 956275.13 frames. ], batch size: 33, lr: 4.00e-03, grad_scale: 16.0 2023-03-25 23:40:00,603 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6985.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:40:13,232 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6775, 3.9229, 3.8014, 2.0819, 4.0327, 3.0779, 1.1353, 2.8495], device='cuda:3'), covar=tensor([0.2576, 0.1632, 0.1504, 0.3490, 0.1080, 0.0960, 0.4763, 0.1556], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0163, 0.0165, 0.0128, 0.0155, 0.0119, 0.0147, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-25 23:40:20,306 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8863, 2.0174, 1.7776, 2.0119, 1.1183, 4.5679, 1.7698, 2.5792], device='cuda:3'), covar=tensor([0.3147, 0.2120, 0.1952, 0.2041, 0.2001, 0.0100, 0.2471, 0.1249], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0107, 0.0112, 0.0114, 0.0109, 0.0092, 0.0095, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-25 23:40:23,990 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.288e+02 1.918e+02 2.412e+02 2.798e+02 4.765e+02, threshold=4.825e+02, percent-clipped=1.0 2023-03-25 23:40:39,102 INFO [finetune.py:976] (3/7) Epoch 2, batch 1300, loss[loss=0.2694, simple_loss=0.286, pruned_loss=0.1264, over 4185.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3279, pruned_loss=0.1253, over 953845.90 frames. ], batch size: 17, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:40:52,830 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7048.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 23:41:17,563 INFO [finetune.py:976] (3/7) Epoch 2, batch 1350, loss[loss=0.2824, simple_loss=0.3188, pruned_loss=0.1231, over 4755.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.327, pruned_loss=0.1249, over 953623.63 frames. ], batch size: 26, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:41:35,192 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-25 23:41:41,174 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7109.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 23:41:47,302 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.456e+02 1.876e+02 2.208e+02 2.586e+02 5.614e+02, threshold=4.416e+02, percent-clipped=5.0 2023-03-25 23:41:49,182 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7118.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 23:41:52,304 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7123.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:41:55,445 INFO [finetune.py:976] (3/7) Epoch 2, batch 1400, loss[loss=0.2502, simple_loss=0.3089, pruned_loss=0.09581, over 4905.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3306, pruned_loss=0.1257, over 955133.47 frames. ], batch size: 36, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:42:31,285 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7163.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:42:33,029 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=7166.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:42:36,032 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=7171.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:42:40,184 INFO [finetune.py:976] (3/7) Epoch 2, batch 1450, loss[loss=0.3118, simple_loss=0.3509, pruned_loss=0.1363, over 4850.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3318, pruned_loss=0.1262, over 954591.41 frames. ], batch size: 44, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:43:29,960 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.045e+02 1.933e+02 2.198e+02 2.731e+02 4.077e+02, threshold=4.395e+02, percent-clipped=0.0 2023-03-25 23:43:40,431 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7224.0, num_to_drop=1, layers_to_drop={3} 2023-03-25 23:43:42,761 INFO [finetune.py:976] (3/7) Epoch 2, batch 1500, loss[loss=0.317, simple_loss=0.3591, pruned_loss=0.1374, over 4812.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3346, pruned_loss=0.1274, over 956149.67 frames. ], batch size: 38, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:43:51,989 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7234.0, num_to_drop=1, layers_to_drop={3} 2023-03-25 23:44:36,526 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7265.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:44:47,908 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7275.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:44:49,672 INFO [finetune.py:976] (3/7) Epoch 2, batch 1550, loss[loss=0.3492, simple_loss=0.3619, pruned_loss=0.1683, over 4248.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3345, pruned_loss=0.127, over 956137.81 frames. ], batch size: 65, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:44:57,965 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7285.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:45:31,202 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.380e+02 1.989e+02 2.300e+02 2.720e+02 4.709e+02, threshold=4.600e+02, percent-clipped=4.0 2023-03-25 23:45:32,006 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-25 23:45:39,141 INFO [finetune.py:976] (3/7) Epoch 2, batch 1600, loss[loss=0.2505, simple_loss=0.3026, pruned_loss=0.09919, over 4753.00 frames. ], tot_loss[loss=0.2905, simple_loss=0.3307, pruned_loss=0.1252, over 957477.93 frames. ], batch size: 27, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:45:42,236 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=7333.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:45:44,175 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7336.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:46:35,389 INFO [finetune.py:976] (3/7) Epoch 2, batch 1650, loss[loss=0.2156, simple_loss=0.2682, pruned_loss=0.08149, over 4815.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3262, pruned_loss=0.1227, over 957511.19 frames. ], batch size: 25, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:46:37,941 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.4139, 3.8492, 3.9130, 4.2876, 4.1391, 3.9231, 4.5143, 1.4204], device='cuda:3'), covar=tensor([0.0734, 0.0788, 0.0771, 0.0898, 0.1192, 0.1306, 0.0615, 0.4934], device='cuda:3'), in_proj_covar=tensor([0.0370, 0.0248, 0.0272, 0.0297, 0.0348, 0.0291, 0.0313, 0.0305], 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') 2023-03-25 23:46:59,572 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7404.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 23:47:01,922 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7407.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:47:07,155 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.344e+02 1.965e+02 2.306e+02 2.769e+02 7.650e+02, threshold=4.611e+02, percent-clipped=3.0 2023-03-25 23:47:09,179 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.89 vs. limit=5.0 2023-03-25 23:47:15,137 INFO [finetune.py:976] (3/7) Epoch 2, batch 1700, loss[loss=0.2444, simple_loss=0.2965, pruned_loss=0.09616, over 4809.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3238, pruned_loss=0.1214, over 958250.82 frames. ], batch size: 45, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:47:48,734 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7468.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:48:00,184 INFO [finetune.py:976] (3/7) Epoch 2, batch 1750, loss[loss=0.3435, simple_loss=0.3848, pruned_loss=0.1511, over 4835.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3279, pruned_loss=0.1238, over 958489.54 frames. ], batch size: 47, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:48:02,402 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-03-25 23:48:42,063 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4682, 1.3720, 1.4906, 1.6374, 2.0010, 1.5369, 1.1822, 1.2974], device='cuda:3'), covar=tensor([0.2998, 0.2853, 0.2215, 0.2207, 0.2739, 0.1662, 0.3775, 0.2234], device='cuda:3'), in_proj_covar=tensor([0.0220, 0.0202, 0.0188, 0.0174, 0.0224, 0.0169, 0.0205, 0.0178], 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') 2023-03-25 23:48:46,572 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.280e+02 2.019e+02 2.351e+02 2.794e+02 5.482e+02, threshold=4.701e+02, percent-clipped=1.0 2023-03-25 23:48:51,970 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3313, 1.1922, 1.2535, 1.4211, 1.9105, 1.3269, 1.0385, 1.1344], device='cuda:3'), covar=tensor([0.3510, 0.3382, 0.2793, 0.2530, 0.2712, 0.2002, 0.4147, 0.2772], device='cuda:3'), in_proj_covar=tensor([0.0220, 0.0202, 0.0188, 0.0174, 0.0224, 0.0169, 0.0205, 0.0178], 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') 2023-03-25 23:48:53,182 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7519.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 23:49:03,495 INFO [finetune.py:976] (3/7) Epoch 2, batch 1800, loss[loss=0.2208, simple_loss=0.2637, pruned_loss=0.08897, over 3578.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.331, pruned_loss=0.1243, over 957018.54 frames. ], batch size: 15, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:49:12,369 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7534.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 23:49:36,161 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-03-25 23:49:44,268 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7565.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:49:57,758 INFO [finetune.py:976] (3/7) Epoch 2, batch 1850, loss[loss=0.3391, simple_loss=0.3589, pruned_loss=0.1596, over 4824.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3329, pruned_loss=0.1255, over 956785.07 frames. ], batch size: 47, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:50:05,099 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=7582.0, num_to_drop=1, layers_to_drop={0} 2023-03-25 23:50:41,138 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5835, 1.3784, 1.4830, 1.6982, 2.1427, 1.5948, 1.1607, 1.3122], device='cuda:3'), covar=tensor([0.3098, 0.3096, 0.2448, 0.2329, 0.2479, 0.1676, 0.3875, 0.2448], device='cuda:3'), in_proj_covar=tensor([0.0219, 0.0202, 0.0187, 0.0173, 0.0223, 0.0168, 0.0204, 0.0177], 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') 2023-03-25 23:50:47,093 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=7613.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:50:47,751 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5783, 2.2280, 3.2167, 4.4379, 3.2597, 3.2177, 1.6004, 3.6778], device='cuda:3'), covar=tensor([0.1750, 0.1454, 0.1186, 0.0464, 0.0750, 0.1182, 0.1929, 0.0482], device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0121, 0.0140, 0.0163, 0.0106, 0.0147, 0.0131, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-25 23:50:48,710 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.368e+02 2.140e+02 2.552e+02 3.133e+02 4.516e+02, threshold=5.105e+02, percent-clipped=0.0 2023-03-25 23:51:00,589 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7624.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:51:02,925 INFO [finetune.py:976] (3/7) Epoch 2, batch 1900, loss[loss=0.2752, simple_loss=0.3281, pruned_loss=0.1112, over 4918.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3341, pruned_loss=0.1256, over 955757.39 frames. ], batch size: 38, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:51:09,871 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7631.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:51:41,823 INFO [finetune.py:976] (3/7) Epoch 2, batch 1950, loss[loss=0.2799, simple_loss=0.3183, pruned_loss=0.1207, over 4834.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3309, pruned_loss=0.1236, over 954815.78 frames. ], batch size: 47, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:51:47,806 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7685.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:51:53,264 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6850, 1.9708, 1.7454, 1.9378, 1.7919, 3.5406, 1.4796, 1.8561], device='cuda:3'), covar=tensor([0.0993, 0.1517, 0.1126, 0.1071, 0.1528, 0.0241, 0.1428, 0.1581], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0078, 0.0075, 0.0078, 0.0091, 0.0080, 0.0084, 0.0077], 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') 2023-03-25 23:51:59,346 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7704.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 23:52:07,431 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.355e+02 1.951e+02 2.306e+02 2.786e+02 5.176e+02, threshold=4.611e+02, percent-clipped=1.0 2023-03-25 23:52:17,328 INFO [finetune.py:976] (3/7) Epoch 2, batch 2000, loss[loss=0.3094, simple_loss=0.3174, pruned_loss=0.1507, over 4152.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3275, pruned_loss=0.1221, over 951993.05 frames. ], batch size: 65, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:52:26,492 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7743.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:52:31,890 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=7752.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 23:52:39,938 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7763.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:52:45,005 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-25 23:52:49,967 INFO [finetune.py:976] (3/7) Epoch 2, batch 2050, loss[loss=0.2007, simple_loss=0.2549, pruned_loss=0.07325, over 4807.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3236, pruned_loss=0.1206, over 952348.38 frames. ], batch size: 25, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:53:05,256 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7802.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:53:06,484 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7804.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:53:14,629 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.310e+02 1.893e+02 2.255e+02 2.795e+02 6.508e+02, threshold=4.510e+02, percent-clipped=3.0 2023-03-25 23:53:17,182 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7819.0, num_to_drop=1, layers_to_drop={1} 2023-03-25 23:53:26,209 INFO [finetune.py:976] (3/7) Epoch 2, batch 2100, loss[loss=0.3013, simple_loss=0.3369, pruned_loss=0.1328, over 4785.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3223, pruned_loss=0.1203, over 949777.71 frames. ], batch size: 29, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:53:27,664 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-25 23:53:56,789 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7863.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:53:58,459 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4527, 1.4226, 1.9309, 3.0078, 2.0494, 2.1882, 0.7833, 2.3706], device='cuda:3'), covar=tensor([0.2139, 0.1825, 0.1549, 0.0643, 0.0992, 0.1488, 0.2295, 0.0784], device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0120, 0.0139, 0.0162, 0.0105, 0.0146, 0.0131, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-25 23:54:00,169 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=7867.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:54:08,357 INFO [finetune.py:976] (3/7) Epoch 2, batch 2150, loss[loss=0.2646, simple_loss=0.3297, pruned_loss=0.09972, over 4755.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3267, pruned_loss=0.1221, over 949822.81 frames. ], batch size: 59, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:54:29,571 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8030, 1.6095, 1.9944, 1.3824, 1.9090, 1.9281, 1.6385, 2.1125], device='cuda:3'), covar=tensor([0.1716, 0.2215, 0.1580, 0.2133, 0.1085, 0.1679, 0.2545, 0.1142], device='cuda:3'), in_proj_covar=tensor([0.0202, 0.0204, 0.0201, 0.0193, 0.0175, 0.0220, 0.0211, 0.0196], 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') 2023-03-25 23:54:49,701 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.034e+02 1.983e+02 2.392e+02 2.856e+02 4.131e+02, threshold=4.785e+02, percent-clipped=0.0 2023-03-25 23:55:10,156 INFO [finetune.py:976] (3/7) Epoch 2, batch 2200, loss[loss=0.2839, simple_loss=0.3368, pruned_loss=0.1155, over 4891.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3282, pruned_loss=0.1225, over 952058.98 frames. ], batch size: 32, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:55:16,976 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7931.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:56:12,550 INFO [finetune.py:976] (3/7) Epoch 2, batch 2250, loss[loss=0.2549, simple_loss=0.316, pruned_loss=0.09691, over 4923.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3301, pruned_loss=0.1233, over 950756.79 frames. ], batch size: 42, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:56:13,681 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=7979.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:56:14,297 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7980.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:57:05,130 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.321e+02 1.899e+02 2.269e+02 2.653e+02 4.132e+02, threshold=4.538e+02, percent-clipped=0.0 2023-03-25 23:57:24,777 INFO [finetune.py:976] (3/7) Epoch 2, batch 2300, loss[loss=0.328, simple_loss=0.3592, pruned_loss=0.1484, over 4716.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3302, pruned_loss=0.1224, over 952537.91 frames. ], batch size: 59, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:57:31,728 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1478, 2.1029, 2.1886, 2.3398, 2.6577, 2.2612, 1.8741, 1.9774], device='cuda:3'), covar=tensor([0.2327, 0.2545, 0.1913, 0.1789, 0.2118, 0.1268, 0.2872, 0.1907], device='cuda:3'), in_proj_covar=tensor([0.0221, 0.0204, 0.0189, 0.0175, 0.0225, 0.0170, 0.0206, 0.0179], 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') 2023-03-25 23:57:51,954 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.89 vs. limit=5.0 2023-03-25 23:57:53,416 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8063.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:57:57,694 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3252, 2.1057, 1.9295, 0.9955, 2.0778, 1.8887, 1.5203, 1.9810], device='cuda:3'), covar=tensor([0.0883, 0.0978, 0.1578, 0.2423, 0.1397, 0.2105, 0.2128, 0.1206], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0188, 0.0199, 0.0183, 0.0209, 0.0205, 0.0209, 0.0196], 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') 2023-03-25 23:58:05,649 INFO [finetune.py:976] (3/7) Epoch 2, batch 2350, loss[loss=0.2716, simple_loss=0.3188, pruned_loss=0.1122, over 4893.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3275, pruned_loss=0.1212, over 953633.60 frames. ], batch size: 32, lr: 3.99e-03, grad_scale: 16.0 2023-03-25 23:58:17,661 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8648, 1.2659, 0.9391, 1.7159, 2.0976, 1.3822, 1.4634, 1.7691], device='cuda:3'), covar=tensor([0.1538, 0.2141, 0.2293, 0.1264, 0.2103, 0.1997, 0.1398, 0.1892], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0098, 0.0117, 0.0093, 0.0125, 0.0097, 0.0099, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-25 23:58:20,621 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8099.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:58:39,058 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=8111.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:58:41,411 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.369e+02 1.949e+02 2.211e+02 2.649e+02 5.737e+02, threshold=4.422e+02, percent-clipped=2.0 2023-03-25 23:59:01,109 INFO [finetune.py:976] (3/7) Epoch 2, batch 2400, loss[loss=0.2872, simple_loss=0.3252, pruned_loss=0.1246, over 4750.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3238, pruned_loss=0.1197, over 955190.22 frames. ], batch size: 27, lr: 3.99e-03, grad_scale: 32.0 2023-03-25 23:59:27,547 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8158.0, num_to_drop=0, layers_to_drop=set() 2023-03-25 23:59:39,734 INFO [finetune.py:976] (3/7) Epoch 2, batch 2450, loss[loss=0.2733, simple_loss=0.3269, pruned_loss=0.1099, over 4858.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3191, pruned_loss=0.1174, over 954840.27 frames. ], batch size: 44, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:00:20,953 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.358e+02 1.915e+02 2.165e+02 2.652e+02 4.234e+02, threshold=4.330e+02, percent-clipped=0.0 2023-03-26 00:00:34,017 INFO [finetune.py:976] (3/7) Epoch 2, batch 2500, loss[loss=0.3323, simple_loss=0.3521, pruned_loss=0.1562, over 4868.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3205, pruned_loss=0.119, over 954894.28 frames. ], batch size: 31, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:00:42,897 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-26 00:01:11,701 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8265.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:01:30,124 INFO [finetune.py:976] (3/7) Epoch 2, batch 2550, loss[loss=0.2953, simple_loss=0.3413, pruned_loss=0.1246, over 4799.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3246, pruned_loss=0.1205, over 954344.07 frames. ], batch size: 41, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:01:31,465 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8280.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:02:02,894 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.611e+02 2.134e+02 2.551e+02 3.097e+02 6.135e+02, threshold=5.102e+02, percent-clipped=4.0 2023-03-26 00:02:09,747 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8326.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:02:10,867 INFO [finetune.py:976] (3/7) Epoch 2, batch 2600, loss[loss=0.3208, simple_loss=0.3525, pruned_loss=0.1445, over 4799.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3266, pruned_loss=0.1206, over 955457.64 frames. ], batch size: 51, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:02:10,926 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=8328.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:02:33,227 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-26 00:03:09,436 INFO [finetune.py:976] (3/7) Epoch 2, batch 2650, loss[loss=0.3279, simple_loss=0.3586, pruned_loss=0.1486, over 4815.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3283, pruned_loss=0.1216, over 954771.02 frames. ], batch size: 33, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:03:28,494 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8399.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:03:32,185 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.4188, 3.8662, 4.0122, 4.3151, 4.1535, 3.8736, 4.5125, 1.4825], device='cuda:3'), covar=tensor([0.0749, 0.0784, 0.0826, 0.0797, 0.1094, 0.1461, 0.0645, 0.5220], device='cuda:3'), in_proj_covar=tensor([0.0376, 0.0250, 0.0278, 0.0302, 0.0352, 0.0294, 0.0318, 0.0310], 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') 2023-03-26 00:03:45,675 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.253e+02 1.975e+02 2.270e+02 2.796e+02 5.066e+02, threshold=4.539e+02, percent-clipped=0.0 2023-03-26 00:03:55,793 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1585, 1.8684, 2.4912, 1.7491, 2.2271, 2.3741, 1.9164, 2.5964], device='cuda:3'), covar=tensor([0.1327, 0.1981, 0.1422, 0.1855, 0.1007, 0.1330, 0.2214, 0.0836], device='cuda:3'), in_proj_covar=tensor([0.0202, 0.0205, 0.0202, 0.0193, 0.0176, 0.0221, 0.0211, 0.0196], 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') 2023-03-26 00:03:57,444 INFO [finetune.py:976] (3/7) Epoch 2, batch 2700, loss[loss=0.3363, simple_loss=0.3586, pruned_loss=0.157, over 4858.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3278, pruned_loss=0.1207, over 955284.88 frames. ], batch size: 49, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:04:21,450 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=8447.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:04:28,885 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8458.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:04:47,922 INFO [finetune.py:976] (3/7) Epoch 2, batch 2750, loss[loss=0.2863, simple_loss=0.32, pruned_loss=0.1263, over 4781.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3236, pruned_loss=0.1189, over 954561.39 frames. ], batch size: 29, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:05:09,712 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=8506.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:05:25,226 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.170e+02 1.975e+02 2.311e+02 2.901e+02 5.278e+02, threshold=4.621e+02, percent-clipped=1.0 2023-03-26 00:05:35,461 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4654, 2.1313, 1.7795, 0.8645, 1.9441, 1.9630, 1.6658, 2.0154], device='cuda:3'), covar=tensor([0.0750, 0.0965, 0.1660, 0.2405, 0.1534, 0.2271, 0.2188, 0.1052], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0189, 0.0199, 0.0183, 0.0209, 0.0204, 0.0210, 0.0196], 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') 2023-03-26 00:05:38,377 INFO [finetune.py:976] (3/7) Epoch 2, batch 2800, loss[loss=0.2439, simple_loss=0.2868, pruned_loss=0.1005, over 4838.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3198, pruned_loss=0.1172, over 956685.80 frames. ], batch size: 47, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:06:44,213 INFO [finetune.py:976] (3/7) Epoch 2, batch 2850, loss[loss=0.2341, simple_loss=0.2828, pruned_loss=0.09273, over 4849.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3162, pruned_loss=0.115, over 957348.36 frames. ], batch size: 44, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:07:08,736 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.223e+02 1.882e+02 2.289e+02 2.777e+02 4.779e+02, threshold=4.578e+02, percent-clipped=1.0 2023-03-26 00:07:13,354 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8621.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:07:18,064 INFO [finetune.py:976] (3/7) Epoch 2, batch 2900, loss[loss=0.2759, simple_loss=0.3184, pruned_loss=0.1167, over 4753.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3196, pruned_loss=0.1165, over 956333.79 frames. ], batch size: 26, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:07:51,245 INFO [finetune.py:976] (3/7) Epoch 2, batch 2950, loss[loss=0.252, simple_loss=0.3134, pruned_loss=0.09528, over 4868.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3239, pruned_loss=0.1185, over 955934.35 frames. ], batch size: 31, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:07:51,998 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7494, 1.4359, 1.3419, 1.4250, 1.4626, 1.4148, 1.4209, 2.2428], device='cuda:3'), covar=tensor([1.7671, 1.7417, 1.4522, 1.8991, 1.2942, 0.9902, 1.7165, 0.4744], device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0230, 0.0208, 0.0265, 0.0222, 0.0188, 0.0227, 0.0170], 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') 2023-03-26 00:08:17,805 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.163e+02 1.871e+02 2.334e+02 2.881e+02 5.890e+02, threshold=4.669e+02, percent-clipped=2.0 2023-03-26 00:08:25,130 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8279, 1.6208, 1.4443, 1.3338, 1.9132, 2.1086, 1.8049, 1.3027], device='cuda:3'), covar=tensor([0.0242, 0.0468, 0.0584, 0.0462, 0.0287, 0.0303, 0.0332, 0.0497], device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0113, 0.0133, 0.0113, 0.0104, 0.0098, 0.0088, 0.0109], device='cuda:3'), out_proj_covar=tensor([6.4575e-05, 8.9618e-05, 1.0722e-04, 8.9751e-05, 8.2501e-05, 7.2956e-05, 6.7983e-05, 8.5361e-05], device='cuda:3') 2023-03-26 00:08:27,822 INFO [finetune.py:976] (3/7) Epoch 2, batch 3000, loss[loss=0.2833, simple_loss=0.3347, pruned_loss=0.1159, over 4806.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3247, pruned_loss=0.1189, over 955785.74 frames. ], batch size: 51, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:08:27,822 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 00:08:31,667 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8817, 1.2941, 1.0146, 1.6861, 2.0933, 1.2166, 1.5121, 1.6979], device='cuda:3'), covar=tensor([0.1412, 0.1975, 0.1938, 0.1136, 0.1947, 0.2055, 0.1262, 0.1963], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0097, 0.0115, 0.0092, 0.0124, 0.0096, 0.0098, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 00:08:33,593 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.8020, 3.3650, 3.4308, 3.6601, 3.4701, 3.3641, 3.8845, 1.4057], device='cuda:3'), covar=tensor([0.0970, 0.0841, 0.0918, 0.1101, 0.1595, 0.1402, 0.0706, 0.5189], device='cuda:3'), in_proj_covar=tensor([0.0372, 0.0248, 0.0276, 0.0300, 0.0349, 0.0291, 0.0316, 0.0306], 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') 2023-03-26 00:08:43,568 INFO [finetune.py:1010] (3/7) Epoch 2, validation: loss=0.1956, simple_loss=0.2636, pruned_loss=0.06384, over 2265189.00 frames. 2023-03-26 00:08:43,568 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6061MB 2023-03-26 00:08:50,164 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9717, 1.4881, 2.4012, 3.8736, 2.6407, 2.5712, 0.7279, 3.1101], device='cuda:3'), covar=tensor([0.1923, 0.1842, 0.1444, 0.0471, 0.0898, 0.1770, 0.2359, 0.0596], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0120, 0.0139, 0.0163, 0.0105, 0.0145, 0.0132, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 00:09:09,177 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8760.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:09:26,206 INFO [finetune.py:976] (3/7) Epoch 2, batch 3050, loss[loss=0.2971, simple_loss=0.3432, pruned_loss=0.1255, over 4787.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3252, pruned_loss=0.1186, over 953376.74 frames. ], batch size: 51, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:09:58,136 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6083, 1.4256, 1.3250, 1.3260, 1.7425, 1.7910, 1.5836, 1.1975], device='cuda:3'), covar=tensor([0.0308, 0.0380, 0.0576, 0.0374, 0.0302, 0.0342, 0.0336, 0.0428], device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0114, 0.0133, 0.0113, 0.0104, 0.0098, 0.0088, 0.0109], device='cuda:3'), out_proj_covar=tensor([6.4779e-05, 8.9895e-05, 1.0784e-04, 8.9934e-05, 8.2593e-05, 7.3046e-05, 6.8118e-05, 8.5312e-05], device='cuda:3') 2023-03-26 00:10:07,399 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.149e+02 1.935e+02 2.296e+02 2.584e+02 4.666e+02, threshold=4.592e+02, percent-clipped=0.0 2023-03-26 00:10:11,719 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8821.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:10:16,913 INFO [finetune.py:976] (3/7) Epoch 2, batch 3100, loss[loss=0.2599, simple_loss=0.3117, pruned_loss=0.104, over 4850.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3216, pruned_loss=0.1165, over 953165.09 frames. ], batch size: 47, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:10:48,289 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.78 vs. limit=5.0 2023-03-26 00:10:59,132 INFO [finetune.py:976] (3/7) Epoch 2, batch 3150, loss[loss=0.2179, simple_loss=0.2737, pruned_loss=0.08101, over 4827.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3177, pruned_loss=0.1153, over 954761.52 frames. ], batch size: 39, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:11:15,177 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6693, 1.5252, 1.5801, 1.7680, 2.2118, 1.6971, 1.4438, 1.4272], device='cuda:3'), covar=tensor([0.2957, 0.2884, 0.2371, 0.2202, 0.2520, 0.1496, 0.3378, 0.2279], device='cuda:3'), in_proj_covar=tensor([0.0222, 0.0204, 0.0190, 0.0176, 0.0226, 0.0169, 0.0206, 0.0179], 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') 2023-03-26 00:11:28,135 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.188e+02 1.838e+02 2.165e+02 2.835e+02 5.909e+02, threshold=4.329e+02, percent-clipped=1.0 2023-03-26 00:11:33,427 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8921.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:11:43,254 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8927.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:11:43,729 INFO [finetune.py:976] (3/7) Epoch 2, batch 3200, loss[loss=0.258, simple_loss=0.2998, pruned_loss=0.1082, over 4904.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3134, pruned_loss=0.1131, over 956089.59 frames. ], batch size: 35, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:12:02,982 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.0347, 0.9002, 1.0468, 0.2783, 0.7544, 1.1593, 1.2110, 1.0960], device='cuda:3'), covar=tensor([0.0937, 0.0565, 0.0417, 0.0690, 0.0488, 0.0507, 0.0353, 0.0554], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0152, 0.0115, 0.0130, 0.0127, 0.0116, 0.0142, 0.0141], device='cuda:3'), out_proj_covar=tensor([9.5506e-05, 1.1333e-04, 8.4622e-05, 9.5301e-05, 9.2322e-05, 8.5378e-05, 1.0648e-04, 1.0444e-04], device='cuda:3') 2023-03-26 00:12:25,338 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=8969.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:12:30,757 INFO [finetune.py:976] (3/7) Epoch 2, batch 3250, loss[loss=0.3711, simple_loss=0.3968, pruned_loss=0.1727, over 4906.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3162, pruned_loss=0.1148, over 954520.53 frames. ], batch size: 43, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:12:36,304 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8986.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:12:37,511 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8988.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:12:59,394 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5024, 1.3876, 1.2451, 1.5383, 1.5935, 1.2627, 1.8558, 1.4071], device='cuda:3'), covar=tensor([0.2930, 0.5924, 0.6130, 0.5269, 0.4310, 0.3078, 0.4660, 0.4125], device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0194, 0.0237, 0.0251, 0.0213, 0.0181, 0.0202, 0.0187], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 00:13:02,646 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9013.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:13:03,775 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.184e+02 1.894e+02 2.281e+02 2.922e+02 4.541e+02, threshold=4.561e+02, percent-clipped=3.0 2023-03-26 00:13:11,710 INFO [finetune.py:976] (3/7) Epoch 2, batch 3300, loss[loss=0.3054, simple_loss=0.3521, pruned_loss=0.1293, over 4829.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3226, pruned_loss=0.1186, over 953796.30 frames. ], batch size: 40, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:13:25,344 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9047.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:13:30,769 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-03-26 00:13:42,687 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9074.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:13:44,938 INFO [finetune.py:976] (3/7) Epoch 2, batch 3350, loss[loss=0.3502, simple_loss=0.3694, pruned_loss=0.1655, over 4832.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3257, pruned_loss=0.1195, over 954310.49 frames. ], batch size: 47, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:13:50,397 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9580, 1.6801, 2.1505, 3.0780, 2.2705, 2.3940, 1.2506, 2.4477], device='cuda:3'), covar=tensor([0.1409, 0.1368, 0.1090, 0.0503, 0.0639, 0.2235, 0.1589, 0.0639], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0120, 0.0139, 0.0163, 0.0105, 0.0146, 0.0132, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 00:13:51,673 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6809, 1.0092, 1.3147, 1.3096, 1.2207, 1.2271, 1.3003, 1.2515], device='cuda:3'), covar=tensor([1.5646, 3.2425, 2.3954, 2.6414, 2.9620, 1.9768, 3.3974, 2.1693], device='cuda:3'), in_proj_covar=tensor([0.0224, 0.0257, 0.0245, 0.0270, 0.0245, 0.0218, 0.0279, 0.0216], 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') 2023-03-26 00:14:11,345 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 2023-03-26 00:14:20,188 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.430e+02 1.942e+02 2.307e+02 2.996e+02 6.023e+02, threshold=4.614e+02, percent-clipped=2.0 2023-03-26 00:14:20,879 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9116.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:14:28,092 INFO [finetune.py:976] (3/7) Epoch 2, batch 3400, loss[loss=0.261, simple_loss=0.3236, pruned_loss=0.09918, over 4877.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3268, pruned_loss=0.1197, over 954431.28 frames. ], batch size: 32, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:14:28,245 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7184, 1.0826, 1.3900, 1.3409, 1.2283, 1.2533, 1.3023, 1.3792], device='cuda:3'), covar=tensor([1.8471, 3.5074, 2.3879, 2.9524, 3.0408, 2.0136, 3.8488, 2.1981], device='cuda:3'), in_proj_covar=tensor([0.0225, 0.0258, 0.0246, 0.0270, 0.0246, 0.0219, 0.0279, 0.0216], 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') 2023-03-26 00:15:23,469 INFO [finetune.py:976] (3/7) Epoch 2, batch 3450, loss[loss=0.3003, simple_loss=0.3266, pruned_loss=0.137, over 4765.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3247, pruned_loss=0.1178, over 954662.22 frames. ], batch size: 26, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:15:24,746 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3268, 2.8899, 2.8083, 1.4289, 3.0076, 2.2034, 0.8170, 1.8652], device='cuda:3'), covar=tensor([0.2182, 0.1883, 0.1700, 0.3444, 0.1269, 0.1206, 0.4042, 0.1713], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0165, 0.0165, 0.0128, 0.0155, 0.0119, 0.0146, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 00:15:59,478 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.266e+02 1.993e+02 2.349e+02 2.850e+02 4.291e+02, threshold=4.698e+02, percent-clipped=0.0 2023-03-26 00:16:12,594 INFO [finetune.py:976] (3/7) Epoch 2, batch 3500, loss[loss=0.2681, simple_loss=0.3138, pruned_loss=0.1113, over 4802.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3204, pruned_loss=0.1158, over 954371.97 frames. ], batch size: 25, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:17:13,577 INFO [finetune.py:976] (3/7) Epoch 2, batch 3550, loss[loss=0.2264, simple_loss=0.2728, pruned_loss=0.09001, over 4870.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3156, pruned_loss=0.1131, over 954929.44 frames. ], batch size: 31, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:17:16,743 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9283.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:17:50,671 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5530, 1.3442, 1.2911, 1.1761, 1.6783, 1.3665, 1.7011, 1.4603], device='cuda:3'), covar=tensor([0.2502, 0.4696, 0.4990, 0.4736, 0.3459, 0.2535, 0.3604, 0.3514], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0195, 0.0238, 0.0252, 0.0214, 0.0182, 0.0203, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 00:17:53,395 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.215e+02 1.769e+02 2.188e+02 2.766e+02 5.069e+02, threshold=4.376e+02, percent-clipped=2.0 2023-03-26 00:17:53,503 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6463, 1.3850, 2.0699, 3.2953, 2.2755, 2.4379, 1.0510, 2.5601], device='cuda:3'), covar=tensor([0.2020, 0.1813, 0.1533, 0.0640, 0.0954, 0.1717, 0.2170, 0.0747], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0120, 0.0138, 0.0162, 0.0105, 0.0145, 0.0131, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 00:18:09,359 INFO [finetune.py:976] (3/7) Epoch 2, batch 3600, loss[loss=0.2196, simple_loss=0.2682, pruned_loss=0.08546, over 4781.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3127, pruned_loss=0.112, over 956754.75 frames. ], batch size: 26, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:18:21,785 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5546, 1.0963, 0.8688, 1.3772, 1.8548, 0.6606, 1.2008, 1.4472], device='cuda:3'), covar=tensor([0.1588, 0.2235, 0.1906, 0.1303, 0.2356, 0.2140, 0.1503, 0.2008], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0098, 0.0118, 0.0094, 0.0126, 0.0098, 0.0100, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 00:18:22,987 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9342.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:18:30,889 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4948, 1.7242, 2.0671, 1.9613, 1.9809, 4.2896, 1.5515, 1.9391], device='cuda:3'), covar=tensor([0.1098, 0.1575, 0.1073, 0.1034, 0.1435, 0.0181, 0.1458, 0.1598], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0079, 0.0076, 0.0078, 0.0092, 0.0081, 0.0084, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 00:18:45,310 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9369.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:18:51,739 INFO [finetune.py:976] (3/7) Epoch 2, batch 3650, loss[loss=0.3458, simple_loss=0.3677, pruned_loss=0.162, over 4760.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.317, pruned_loss=0.1148, over 956011.75 frames. ], batch size: 54, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:19:06,834 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8870, 1.7167, 1.6081, 1.2285, 1.8958, 2.3530, 1.9292, 1.6287], device='cuda:3'), covar=tensor([0.0305, 0.0428, 0.0525, 0.0465, 0.0408, 0.0311, 0.0353, 0.0475], device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0112, 0.0133, 0.0112, 0.0103, 0.0097, 0.0087, 0.0108], device='cuda:3'), out_proj_covar=tensor([6.3887e-05, 8.8820e-05, 1.0736e-04, 8.8917e-05, 8.1681e-05, 7.2261e-05, 6.6978e-05, 8.4435e-05], device='cuda:3') 2023-03-26 00:19:24,500 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9414.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:19:24,982 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.270e+02 1.902e+02 2.269e+02 2.850e+02 5.426e+02, threshold=4.539e+02, percent-clipped=4.0 2023-03-26 00:19:31,925 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9416.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:19:45,283 INFO [finetune.py:976] (3/7) Epoch 2, batch 3700, loss[loss=0.2379, simple_loss=0.2968, pruned_loss=0.08955, over 4821.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3203, pruned_loss=0.1151, over 955935.53 frames. ], batch size: 38, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:20:05,753 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5407, 1.2598, 1.1499, 0.9091, 1.3073, 1.3055, 1.2146, 1.9628], device='cuda:3'), covar=tensor([1.9857, 1.8038, 1.5090, 2.1521, 1.3979, 1.1329, 1.9127, 0.5862], device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0235, 0.0212, 0.0273, 0.0226, 0.0191, 0.0232, 0.0174], 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') 2023-03-26 00:20:09,675 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-26 00:20:15,849 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=9464.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:20:23,937 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9475.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:20:26,040 INFO [finetune.py:976] (3/7) Epoch 2, batch 3750, loss[loss=0.335, simple_loss=0.378, pruned_loss=0.146, over 4930.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3228, pruned_loss=0.1168, over 955598.12 frames. ], batch size: 42, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:20:40,220 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6565, 0.9628, 1.3742, 1.3284, 1.2554, 1.2359, 1.1807, 1.3035], device='cuda:3'), covar=tensor([1.5504, 3.0857, 1.9826, 2.3969, 2.5437, 1.8295, 3.2953, 1.8913], device='cuda:3'), in_proj_covar=tensor([0.0225, 0.0258, 0.0246, 0.0270, 0.0246, 0.0219, 0.0280, 0.0217], 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') 2023-03-26 00:20:41,094 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-26 00:20:55,390 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.357e+02 1.893e+02 2.405e+02 2.686e+02 6.929e+02, threshold=4.810e+02, percent-clipped=1.0 2023-03-26 00:21:10,670 INFO [finetune.py:976] (3/7) Epoch 2, batch 3800, loss[loss=0.25, simple_loss=0.3069, pruned_loss=0.0966, over 4879.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3211, pruned_loss=0.1157, over 953331.49 frames. ], batch size: 32, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:21:54,474 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-03-26 00:22:03,070 INFO [finetune.py:976] (3/7) Epoch 2, batch 3850, loss[loss=0.2416, simple_loss=0.2952, pruned_loss=0.09404, over 4867.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.318, pruned_loss=0.1134, over 954210.91 frames. ], batch size: 34, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:22:07,250 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9583.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:22:07,867 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9584.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:22:07,890 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8657, 1.7293, 1.2284, 2.0153, 2.0982, 1.6014, 2.5837, 1.8035], device='cuda:3'), covar=tensor([0.2998, 0.6974, 0.6676, 0.6387, 0.4164, 0.2995, 0.4954, 0.4100], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0195, 0.0238, 0.0252, 0.0214, 0.0182, 0.0203, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 00:22:33,608 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.288e+02 1.890e+02 2.331e+02 2.867e+02 5.576e+02, threshold=4.662e+02, percent-clipped=3.0 2023-03-26 00:22:41,950 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-26 00:22:48,507 INFO [finetune.py:976] (3/7) Epoch 2, batch 3900, loss[loss=0.2508, simple_loss=0.2937, pruned_loss=0.104, over 4751.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3147, pruned_loss=0.1124, over 953979.08 frames. ], batch size: 27, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:22:49,817 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.2097, 1.2473, 1.3280, 0.6371, 1.0498, 1.4529, 1.4751, 1.3449], device='cuda:3'), covar=tensor([0.0861, 0.0482, 0.0448, 0.0629, 0.0484, 0.0477, 0.0319, 0.0498], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0154, 0.0116, 0.0132, 0.0130, 0.0117, 0.0143, 0.0143], device='cuda:3'), out_proj_covar=tensor([9.5643e-05, 1.1454e-04, 8.5167e-05, 9.7042e-05, 9.4134e-05, 8.6115e-05, 1.0713e-04, 1.0573e-04], device='cuda:3') 2023-03-26 00:22:55,985 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=9631.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:23:08,253 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9642.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:23:10,107 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9645.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:23:28,459 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9658.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:23:40,492 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9669.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:23:42,689 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-26 00:23:46,466 INFO [finetune.py:976] (3/7) Epoch 2, batch 3950, loss[loss=0.2861, simple_loss=0.3173, pruned_loss=0.1275, over 4896.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3096, pruned_loss=0.11, over 953483.48 frames. ], batch size: 35, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:24:00,117 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=9690.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:24:07,945 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6401, 1.6646, 1.4848, 1.7361, 1.2199, 3.6093, 1.4356, 2.1208], device='cuda:3'), covar=tensor([0.3688, 0.2462, 0.2228, 0.2363, 0.2035, 0.0196, 0.2670, 0.1381], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0108, 0.0114, 0.0116, 0.0113, 0.0095, 0.0098, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 00:24:08,542 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7105, 1.6895, 1.5556, 1.7045, 1.3635, 4.2610, 1.6873, 2.2898], device='cuda:3'), covar=tensor([0.4379, 0.3037, 0.2427, 0.2807, 0.1764, 0.0157, 0.2529, 0.1343], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0108, 0.0114, 0.0116, 0.0113, 0.0095, 0.0098, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 00:24:28,618 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.008e+02 1.967e+02 2.302e+02 2.784e+02 7.100e+02, threshold=4.604e+02, percent-clipped=2.0 2023-03-26 00:24:29,311 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=9717.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:24:30,633 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9719.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:24:32,395 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-26 00:24:36,431 INFO [finetune.py:976] (3/7) Epoch 2, batch 4000, loss[loss=0.2671, simple_loss=0.309, pruned_loss=0.1126, over 4709.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3079, pruned_loss=0.1091, over 954102.60 frames. ], batch size: 23, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:24:39,220 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-26 00:24:48,458 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9743.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:25:10,940 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9770.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:25:11,300 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.00 vs. limit=5.0 2023-03-26 00:25:16,120 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-26 00:25:16,246 INFO [finetune.py:976] (3/7) Epoch 2, batch 4050, loss[loss=0.3078, simple_loss=0.3465, pruned_loss=0.1346, over 4740.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3137, pruned_loss=0.1121, over 952867.82 frames. ], batch size: 26, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:25:37,124 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9804.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:25:38,343 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4714, 1.3313, 1.1855, 1.2795, 1.6282, 1.6724, 1.4294, 1.1362], device='cuda:3'), covar=tensor([0.0274, 0.0339, 0.0534, 0.0327, 0.0236, 0.0287, 0.0308, 0.0389], device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0112, 0.0133, 0.0112, 0.0103, 0.0098, 0.0087, 0.0108], device='cuda:3'), out_proj_covar=tensor([6.4333e-05, 8.8915e-05, 1.0746e-04, 8.9152e-05, 8.1825e-05, 7.3263e-05, 6.7069e-05, 8.4610e-05], device='cuda:3') 2023-03-26 00:25:44,425 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.258e+02 1.934e+02 2.184e+02 2.585e+02 5.351e+02, threshold=4.368e+02, percent-clipped=2.0 2023-03-26 00:25:57,343 INFO [finetune.py:976] (3/7) Epoch 2, batch 4100, loss[loss=0.2785, simple_loss=0.3315, pruned_loss=0.1128, over 4901.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.317, pruned_loss=0.1132, over 954596.08 frames. ], batch size: 37, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:26:34,084 INFO [finetune.py:976] (3/7) Epoch 2, batch 4150, loss[loss=0.2488, simple_loss=0.3184, pruned_loss=0.08956, over 4894.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3204, pruned_loss=0.1152, over 954346.96 frames. ], batch size: 43, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:27:05,051 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.284e+02 1.789e+02 2.151e+02 2.605e+02 5.306e+02, threshold=4.302e+02, percent-clipped=2.0 2023-03-26 00:27:17,333 INFO [finetune.py:976] (3/7) Epoch 2, batch 4200, loss[loss=0.2865, simple_loss=0.3235, pruned_loss=0.1247, over 4717.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3191, pruned_loss=0.1137, over 953219.25 frames. ], batch size: 54, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:27:25,691 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9940.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:27:29,170 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.29 vs. limit=5.0 2023-03-26 00:28:05,867 INFO [finetune.py:976] (3/7) Epoch 2, batch 4250, loss[loss=0.2424, simple_loss=0.2908, pruned_loss=0.09705, over 4842.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3154, pruned_loss=0.1121, over 953580.64 frames. ], batch size: 47, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:28:41,978 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10014.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:28:47,605 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.000e+02 1.798e+02 2.175e+02 2.768e+02 4.624e+02, threshold=4.351e+02, percent-clipped=3.0 2023-03-26 00:29:00,126 INFO [finetune.py:976] (3/7) Epoch 2, batch 4300, loss[loss=0.26, simple_loss=0.2996, pruned_loss=0.1102, over 4822.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3125, pruned_loss=0.1116, over 950842.69 frames. ], batch size: 30, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:29:12,678 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10039.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:29:27,794 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8211, 1.1966, 1.5374, 1.4643, 1.3489, 1.3663, 1.4127, 1.5117], device='cuda:3'), covar=tensor([1.5989, 3.0284, 1.9391, 2.4959, 2.8340, 1.7711, 3.6391, 1.8198], device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0258, 0.0248, 0.0270, 0.0247, 0.0220, 0.0280, 0.0217], 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') 2023-03-26 00:29:36,704 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10062.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 00:29:46,273 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10070.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:29:52,923 INFO [finetune.py:976] (3/7) Epoch 2, batch 4350, loss[loss=0.2473, simple_loss=0.2998, pruned_loss=0.09746, over 4792.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3093, pruned_loss=0.1107, over 949904.56 frames. ], batch size: 29, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:30:06,664 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10099.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:30:07,343 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10100.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 00:30:25,353 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.331e+02 1.930e+02 2.275e+02 2.717e+02 4.483e+02, threshold=4.550e+02, percent-clipped=1.0 2023-03-26 00:30:26,670 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7431, 1.1653, 1.4587, 1.3693, 1.2196, 1.2401, 1.3475, 1.4177], device='cuda:3'), covar=tensor([1.6764, 2.9273, 1.9607, 2.5365, 2.7038, 1.9102, 3.3233, 1.6930], device='cuda:3'), in_proj_covar=tensor([0.0225, 0.0257, 0.0247, 0.0269, 0.0246, 0.0219, 0.0279, 0.0216], 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') 2023-03-26 00:30:27,186 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=10118.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:30:34,452 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10123.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 00:30:36,344 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-26 00:30:37,374 INFO [finetune.py:976] (3/7) Epoch 2, batch 4400, loss[loss=0.2751, simple_loss=0.3239, pruned_loss=0.1131, over 4886.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3115, pruned_loss=0.1116, over 952802.93 frames. ], batch size: 32, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:31:35,877 INFO [finetune.py:976] (3/7) Epoch 2, batch 4450, loss[loss=0.3159, simple_loss=0.3501, pruned_loss=0.1409, over 4871.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3146, pruned_loss=0.1123, over 951833.73 frames. ], batch size: 31, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:32:18,678 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.270e+02 1.960e+02 2.424e+02 3.015e+02 7.276e+02, threshold=4.848e+02, percent-clipped=5.0 2023-03-26 00:32:36,845 INFO [finetune.py:976] (3/7) Epoch 2, batch 4500, loss[loss=0.2751, simple_loss=0.3289, pruned_loss=0.1106, over 4906.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3169, pruned_loss=0.113, over 952498.70 frames. ], batch size: 46, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:32:44,209 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10240.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:33:06,703 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-26 00:33:20,857 INFO [finetune.py:976] (3/7) Epoch 2, batch 4550, loss[loss=0.233, simple_loss=0.2797, pruned_loss=0.09317, over 4726.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.318, pruned_loss=0.1131, over 953212.01 frames. ], batch size: 23, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:33:32,455 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=10288.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:33:41,487 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.0708, 0.9020, 1.0488, 0.2572, 0.7323, 1.1824, 1.2058, 1.0711], device='cuda:3'), covar=tensor([0.1069, 0.0655, 0.0473, 0.0771, 0.0584, 0.0509, 0.0454, 0.0642], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0154, 0.0116, 0.0132, 0.0129, 0.0116, 0.0143, 0.0143], device='cuda:3'), out_proj_covar=tensor([9.6010e-05, 1.1488e-04, 8.4881e-05, 9.7318e-05, 9.3759e-05, 8.5836e-05, 1.0690e-04, 1.0571e-04], device='cuda:3') 2023-03-26 00:33:59,119 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10314.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:34:00,220 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.209e+02 1.803e+02 2.127e+02 2.576e+02 5.771e+02, threshold=4.255e+02, percent-clipped=1.0 2023-03-26 00:34:14,432 INFO [finetune.py:976] (3/7) Epoch 2, batch 4600, loss[loss=0.2765, simple_loss=0.3204, pruned_loss=0.1163, over 4766.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3169, pruned_loss=0.1119, over 955396.19 frames. ], batch size: 28, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:34:48,345 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-03-26 00:34:50,217 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=10362.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:34:50,882 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0040, 2.2092, 1.9590, 1.3174, 2.4582, 2.3221, 2.0966, 1.9470], device='cuda:3'), covar=tensor([0.0840, 0.0584, 0.0943, 0.1202, 0.0371, 0.0850, 0.0869, 0.1078], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0131, 0.0142, 0.0129, 0.0107, 0.0139, 0.0146, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 00:35:01,344 INFO [finetune.py:976] (3/7) Epoch 2, batch 4650, loss[loss=0.2741, simple_loss=0.3186, pruned_loss=0.1148, over 4911.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.314, pruned_loss=0.1108, over 955942.71 frames. ], batch size: 36, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:35:06,237 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4146, 1.4532, 1.3942, 1.6582, 1.5908, 3.0569, 1.2671, 1.5573], device='cuda:3'), covar=tensor([0.1139, 0.2071, 0.1192, 0.1139, 0.1732, 0.0316, 0.1861, 0.2233], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0080, 0.0077, 0.0079, 0.0092, 0.0082, 0.0084, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 00:35:12,309 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10395.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 00:35:14,763 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10399.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:35:21,410 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10409.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:35:25,532 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.369e+02 1.834e+02 2.117e+02 2.478e+02 4.313e+02, threshold=4.233e+02, percent-clipped=1.0 2023-03-26 00:35:27,257 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10418.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 00:35:34,604 INFO [finetune.py:976] (3/7) Epoch 2, batch 4700, loss[loss=0.1756, simple_loss=0.2373, pruned_loss=0.05694, over 4764.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3109, pruned_loss=0.1094, over 957087.76 frames. ], batch size: 28, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:35:46,697 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=10447.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:35:49,170 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6794, 1.5989, 1.5536, 1.7141, 1.0609, 3.4442, 1.2496, 1.7790], device='cuda:3'), covar=tensor([0.3713, 0.2541, 0.2265, 0.2342, 0.2206, 0.0219, 0.2848, 0.1567], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0109, 0.0114, 0.0117, 0.0113, 0.0095, 0.0099, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 00:35:50,976 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10454.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:36:00,542 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.99 vs. limit=5.0 2023-03-26 00:36:01,590 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10470.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:36:06,176 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3533, 2.0156, 1.7056, 0.7612, 1.9166, 1.9183, 1.5974, 1.8313], device='cuda:3'), covar=tensor([0.0742, 0.0958, 0.1498, 0.2231, 0.1278, 0.2206, 0.2256, 0.1024], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0192, 0.0200, 0.0185, 0.0211, 0.0207, 0.0212, 0.0198], 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') 2023-03-26 00:36:07,252 INFO [finetune.py:976] (3/7) Epoch 2, batch 4750, loss[loss=0.295, simple_loss=0.3255, pruned_loss=0.1323, over 4860.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3059, pruned_loss=0.1063, over 958329.89 frames. ], batch size: 49, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:36:20,344 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7577, 1.5915, 1.9468, 2.8060, 2.0711, 2.1699, 1.1810, 2.2660], device='cuda:3'), covar=tensor([0.1788, 0.1601, 0.1328, 0.0708, 0.0913, 0.1443, 0.1760, 0.0870], device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0121, 0.0140, 0.0166, 0.0107, 0.0149, 0.0132, 0.0109], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 00:36:41,912 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10515.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 00:36:42,343 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.257e+02 1.733e+02 2.215e+02 2.656e+02 7.843e+02, threshold=4.429e+02, percent-clipped=2.0 2023-03-26 00:36:51,177 INFO [finetune.py:976] (3/7) Epoch 2, batch 4800, loss[loss=0.3084, simple_loss=0.3408, pruned_loss=0.138, over 4904.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.308, pruned_loss=0.1075, over 957278.81 frames. ], batch size: 35, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:36:57,798 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5386, 1.4629, 1.2970, 1.3168, 1.8731, 1.8637, 1.5805, 1.2692], device='cuda:3'), covar=tensor([0.0273, 0.0335, 0.0510, 0.0363, 0.0220, 0.0307, 0.0288, 0.0399], device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0111, 0.0132, 0.0111, 0.0101, 0.0097, 0.0087, 0.0106], device='cuda:3'), out_proj_covar=tensor([6.3299e-05, 8.7812e-05, 1.0642e-04, 8.8319e-05, 8.0564e-05, 7.2461e-05, 6.7020e-05, 8.3227e-05], device='cuda:3') 2023-03-26 00:37:07,376 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6773, 3.8125, 3.7999, 1.9515, 4.0696, 3.1130, 0.8567, 2.7963], device='cuda:3'), covar=tensor([0.2277, 0.1705, 0.1433, 0.2982, 0.0839, 0.0755, 0.4383, 0.1195], device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0168, 0.0167, 0.0129, 0.0158, 0.0121, 0.0147, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 00:37:43,277 INFO [finetune.py:976] (3/7) Epoch 2, batch 4850, loss[loss=0.2724, simple_loss=0.3033, pruned_loss=0.1208, over 4349.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3119, pruned_loss=0.109, over 956924.03 frames. ], batch size: 19, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:38:04,466 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4121, 1.3372, 1.3317, 1.3626, 0.8293, 2.2465, 0.7541, 1.2782], device='cuda:3'), covar=tensor([0.3448, 0.2415, 0.2210, 0.2397, 0.2233, 0.0398, 0.2762, 0.1496], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0109, 0.0115, 0.0117, 0.0113, 0.0096, 0.0099, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 00:38:22,411 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.244e+02 1.945e+02 2.280e+02 2.839e+02 5.226e+02, threshold=4.560e+02, percent-clipped=1.0 2023-03-26 00:38:33,268 INFO [finetune.py:976] (3/7) Epoch 2, batch 4900, loss[loss=0.2302, simple_loss=0.2806, pruned_loss=0.08983, over 4747.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3142, pruned_loss=0.11, over 956949.09 frames. ], batch size: 27, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:38:50,193 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-26 00:38:53,015 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7119, 1.5143, 1.1818, 1.3512, 1.4171, 1.3573, 1.3522, 2.2842], device='cuda:3'), covar=tensor([1.4627, 1.3835, 1.1891, 1.5939, 1.1969, 0.8243, 1.4529, 0.4193], device='cuda:3'), in_proj_covar=tensor([0.0258, 0.0237, 0.0213, 0.0274, 0.0228, 0.0192, 0.0233, 0.0176], 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') 2023-03-26 00:39:05,954 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4023, 3.2486, 3.2503, 1.5185, 3.4486, 2.5056, 0.6732, 2.2477], device='cuda:3'), covar=tensor([0.2568, 0.1923, 0.1532, 0.3357, 0.1123, 0.1076, 0.4398, 0.1691], device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0168, 0.0167, 0.0130, 0.0157, 0.0121, 0.0148, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 00:39:15,147 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-03-26 00:39:15,446 INFO [finetune.py:976] (3/7) Epoch 2, batch 4950, loss[loss=0.2447, simple_loss=0.3035, pruned_loss=0.09296, over 4917.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3149, pruned_loss=0.11, over 954891.82 frames. ], batch size: 42, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:39:27,300 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10695.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:39:28,498 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7849, 1.8592, 1.8683, 1.2205, 2.2225, 2.0532, 1.8217, 1.5362], device='cuda:3'), covar=tensor([0.0782, 0.0681, 0.0814, 0.1154, 0.0426, 0.0740, 0.0944, 0.1415], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0132, 0.0143, 0.0129, 0.0108, 0.0140, 0.0147, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 00:39:33,979 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.53 vs. limit=5.0 2023-03-26 00:39:40,465 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.382e+02 1.832e+02 2.144e+02 2.563e+02 4.788e+02, threshold=4.289e+02, percent-clipped=1.0 2023-03-26 00:39:42,245 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10718.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 00:39:44,642 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8564, 1.5733, 2.0911, 3.4056, 2.4888, 2.3187, 0.8192, 2.7144], device='cuda:3'), covar=tensor([0.1889, 0.1626, 0.1460, 0.0580, 0.0808, 0.1635, 0.2235, 0.0681], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0120, 0.0139, 0.0164, 0.0105, 0.0146, 0.0131, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 00:39:48,192 INFO [finetune.py:976] (3/7) Epoch 2, batch 5000, loss[loss=0.2214, simple_loss=0.2805, pruned_loss=0.08116, over 4904.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3118, pruned_loss=0.1085, over 953692.17 frames. ], batch size: 46, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:39:59,861 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=10743.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:40:00,521 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10744.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:40:13,900 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10765.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:40:14,500 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=10766.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 00:40:15,768 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5273, 1.9375, 1.6819, 0.7739, 1.8846, 1.9338, 1.3828, 1.7238], device='cuda:3'), covar=tensor([0.0699, 0.1364, 0.2027, 0.2676, 0.1799, 0.2214, 0.2830, 0.1547], device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0193, 0.0202, 0.0186, 0.0212, 0.0208, 0.0214, 0.0198], 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') 2023-03-26 00:40:19,852 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10774.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:40:25,991 INFO [finetune.py:976] (3/7) Epoch 2, batch 5050, loss[loss=0.2004, simple_loss=0.2553, pruned_loss=0.0728, over 4855.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3083, pruned_loss=0.1073, over 954721.55 frames. ], batch size: 47, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:40:48,615 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10805.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:40:52,131 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10810.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 00:40:55,640 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.117e+02 1.711e+02 2.025e+02 2.497e+02 4.623e+02, threshold=4.049e+02, percent-clipped=1.0 2023-03-26 00:41:03,403 INFO [finetune.py:976] (3/7) Epoch 2, batch 5100, loss[loss=0.2252, simple_loss=0.2709, pruned_loss=0.08978, over 4905.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3051, pruned_loss=0.1054, over 954317.19 frames. ], batch size: 32, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:41:07,739 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10835.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:41:47,380 INFO [finetune.py:976] (3/7) Epoch 2, batch 5150, loss[loss=0.2855, simple_loss=0.3298, pruned_loss=0.1206, over 4742.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3054, pruned_loss=0.106, over 956519.49 frames. ], batch size: 59, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:41:54,269 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10885.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 00:42:14,384 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4201, 1.3619, 1.7140, 1.7558, 1.4951, 3.2365, 1.1615, 1.5272], device='cuda:3'), covar=tensor([0.1114, 0.1778, 0.1483, 0.1162, 0.1653, 0.0241, 0.1564, 0.1823], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0080, 0.0077, 0.0080, 0.0092, 0.0082, 0.0085, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 00:42:25,067 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.204e+02 1.747e+02 2.089e+02 2.521e+02 5.475e+02, threshold=4.178e+02, percent-clipped=1.0 2023-03-26 00:42:37,757 INFO [finetune.py:976] (3/7) Epoch 2, batch 5200, loss[loss=0.2653, simple_loss=0.317, pruned_loss=0.1068, over 4843.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3104, pruned_loss=0.1082, over 955300.93 frames. ], batch size: 44, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:43:00,795 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10946.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 00:43:19,376 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1110, 1.7355, 1.6162, 1.5227, 2.2300, 2.5168, 2.0839, 1.7640], device='cuda:3'), covar=tensor([0.0263, 0.0477, 0.0578, 0.0392, 0.0280, 0.0364, 0.0261, 0.0390], device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0112, 0.0133, 0.0112, 0.0102, 0.0097, 0.0087, 0.0106], device='cuda:3'), out_proj_covar=tensor([6.3232e-05, 8.8258e-05, 1.0735e-04, 8.8893e-05, 8.0916e-05, 7.2238e-05, 6.6794e-05, 8.3478e-05], device='cuda:3') 2023-03-26 00:43:25,013 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-26 00:43:29,722 INFO [finetune.py:976] (3/7) Epoch 2, batch 5250, loss[loss=0.2726, simple_loss=0.323, pruned_loss=0.1111, over 4836.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3124, pruned_loss=0.1082, over 955574.14 frames. ], batch size: 49, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:43:44,675 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.0074, 1.2682, 1.0119, 1.1686, 1.3202, 2.1830, 1.1250, 1.4344], device='cuda:3'), covar=tensor([0.0927, 0.1392, 0.1020, 0.0887, 0.1410, 0.0353, 0.1248, 0.1308], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0080, 0.0077, 0.0080, 0.0092, 0.0082, 0.0085, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 00:44:03,160 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.206e+02 2.014e+02 2.376e+02 2.957e+02 8.531e+02, threshold=4.753e+02, percent-clipped=3.0 2023-03-26 00:44:10,959 INFO [finetune.py:976] (3/7) Epoch 2, batch 5300, loss[loss=0.2243, simple_loss=0.2896, pruned_loss=0.07946, over 4849.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3122, pruned_loss=0.1081, over 952063.09 frames. ], batch size: 31, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:44:44,448 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11065.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:44:58,486 INFO [finetune.py:976] (3/7) Epoch 2, batch 5350, loss[loss=0.2498, simple_loss=0.3128, pruned_loss=0.09343, over 4822.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3127, pruned_loss=0.1072, over 954100.64 frames. ], batch size: 39, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:45:16,015 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11100.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:45:16,075 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6193, 1.3297, 1.3401, 1.4481, 1.7628, 1.8310, 1.4923, 1.2485], device='cuda:3'), covar=tensor([0.0286, 0.0395, 0.0568, 0.0321, 0.0250, 0.0305, 0.0337, 0.0395], device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0112, 0.0133, 0.0113, 0.0102, 0.0097, 0.0087, 0.0107], device='cuda:3'), out_proj_covar=tensor([6.3432e-05, 8.8513e-05, 1.0782e-04, 8.9162e-05, 8.1335e-05, 7.2813e-05, 6.7420e-05, 8.3770e-05], device='cuda:3') 2023-03-26 00:45:23,061 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11110.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 00:45:25,462 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=11113.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:45:27,189 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.217e+02 1.830e+02 2.178e+02 2.684e+02 7.389e+02, threshold=4.357e+02, percent-clipped=1.0 2023-03-26 00:45:39,911 INFO [finetune.py:976] (3/7) Epoch 2, batch 5400, loss[loss=0.2396, simple_loss=0.2863, pruned_loss=0.09645, over 3992.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3095, pruned_loss=0.1059, over 954393.34 frames. ], batch size: 17, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:45:41,697 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11130.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:46:05,246 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=11158.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:46:22,031 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11174.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:46:24,381 INFO [finetune.py:976] (3/7) Epoch 2, batch 5450, loss[loss=0.2739, simple_loss=0.3182, pruned_loss=0.1148, over 4894.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3054, pruned_loss=0.1042, over 953539.67 frames. ], batch size: 35, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:46:55,233 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.192e+02 1.719e+02 1.995e+02 2.396e+02 4.116e+02, threshold=3.991e+02, percent-clipped=0.0 2023-03-26 00:47:11,562 INFO [finetune.py:976] (3/7) Epoch 2, batch 5500, loss[loss=0.2108, simple_loss=0.2694, pruned_loss=0.07615, over 4895.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3022, pruned_loss=0.1029, over 954967.28 frames. ], batch size: 32, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:47:21,362 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11235.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:47:25,458 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11241.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 00:48:00,491 INFO [finetune.py:976] (3/7) Epoch 2, batch 5550, loss[loss=0.2274, simple_loss=0.26, pruned_loss=0.09746, over 3810.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3037, pruned_loss=0.1036, over 956141.69 frames. ], batch size: 16, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:48:09,293 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 2023-03-26 00:48:10,389 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 2023-03-26 00:48:30,779 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.394e+02 1.988e+02 2.263e+02 2.636e+02 5.646e+02, threshold=4.525e+02, percent-clipped=4.0 2023-03-26 00:48:38,279 INFO [finetune.py:976] (3/7) Epoch 2, batch 5600, loss[loss=0.2361, simple_loss=0.2757, pruned_loss=0.09827, over 4751.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3077, pruned_loss=0.1041, over 957914.37 frames. ], batch size: 23, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:49:19,623 INFO [finetune.py:976] (3/7) Epoch 2, batch 5650, loss[loss=0.2513, simple_loss=0.3047, pruned_loss=0.09891, over 4833.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.31, pruned_loss=0.1051, over 953834.49 frames. ], batch size: 47, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:49:44,837 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11400.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:49:48,872 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.6764, 4.9382, 5.2926, 5.3319, 5.2001, 4.7349, 5.6880, 2.5993], device='cuda:3'), covar=tensor([0.0769, 0.1504, 0.0973, 0.1451, 0.1289, 0.1612, 0.0739, 0.5776], device='cuda:3'), in_proj_covar=tensor([0.0369, 0.0245, 0.0274, 0.0295, 0.0340, 0.0287, 0.0310, 0.0299], 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') 2023-03-26 00:49:51,253 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6013, 1.1109, 0.8809, 1.4104, 2.0296, 0.7405, 1.3360, 1.5235], device='cuda:3'), covar=tensor([0.1445, 0.2095, 0.1807, 0.1123, 0.1840, 0.1893, 0.1366, 0.1810], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0098, 0.0118, 0.0094, 0.0125, 0.0097, 0.0100, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 00:49:54,699 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.185e+02 1.700e+02 2.043e+02 2.333e+02 3.537e+02, threshold=4.085e+02, percent-clipped=0.0 2023-03-26 00:50:01,883 INFO [finetune.py:976] (3/7) Epoch 2, batch 5700, loss[loss=0.2312, simple_loss=0.273, pruned_loss=0.09472, over 4357.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3064, pruned_loss=0.1049, over 938410.68 frames. ], batch size: 19, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:50:03,154 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11430.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:50:13,813 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=11448.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:50:13,913 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0339, 1.7179, 1.4728, 1.6592, 1.7413, 1.6916, 1.6208, 2.4391], device='cuda:3'), covar=tensor([1.3808, 1.2805, 1.0798, 1.4046, 1.0473, 0.7770, 1.3314, 0.4228], device='cuda:3'), in_proj_covar=tensor([0.0261, 0.0240, 0.0216, 0.0277, 0.0230, 0.0193, 0.0235, 0.0178], 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') 2023-03-26 00:50:33,841 INFO [finetune.py:976] (3/7) Epoch 3, batch 0, loss[loss=0.3528, simple_loss=0.3831, pruned_loss=0.1613, over 4737.00 frames. ], tot_loss[loss=0.3528, simple_loss=0.3831, pruned_loss=0.1613, over 4737.00 frames. ], batch size: 54, lr: 3.99e-03, grad_scale: 16.0 2023-03-26 00:50:33,841 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 00:50:55,329 INFO [finetune.py:1010] (3/7) Epoch 3, validation: loss=0.1864, simple_loss=0.2566, pruned_loss=0.05807, over 2265189.00 frames. 2023-03-26 00:50:55,330 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 00:51:20,329 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=11478.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:51:22,753 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6955, 3.9824, 3.8246, 1.7611, 3.9771, 2.9337, 0.8028, 2.7693], device='cuda:3'), covar=tensor([0.2228, 0.1137, 0.1482, 0.3166, 0.0959, 0.0910, 0.4342, 0.1291], device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0167, 0.0166, 0.0129, 0.0156, 0.0120, 0.0146, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 00:51:25,830 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6941, 3.8112, 3.6827, 1.8939, 3.8122, 2.8965, 0.6640, 2.6969], device='cuda:3'), covar=tensor([0.2351, 0.1319, 0.1500, 0.3216, 0.1017, 0.0946, 0.4686, 0.1413], device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0167, 0.0165, 0.0129, 0.0156, 0.0120, 0.0146, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 00:51:38,815 INFO [finetune.py:976] (3/7) Epoch 3, batch 50, loss[loss=0.2503, simple_loss=0.308, pruned_loss=0.09627, over 4924.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3115, pruned_loss=0.107, over 216214.25 frames. ], batch size: 33, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:51:45,831 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.161e+02 1.779e+02 2.075e+02 2.495e+02 4.593e+02, threshold=4.151e+02, percent-clipped=1.0 2023-03-26 00:51:54,921 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11530.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:51:56,193 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11532.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:52:02,018 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11541.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 00:52:09,650 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1904, 1.4395, 1.2315, 1.4812, 1.4952, 2.9336, 1.2815, 1.5518], device='cuda:3'), covar=tensor([0.1133, 0.1733, 0.1280, 0.1103, 0.1632, 0.0290, 0.1464, 0.1741], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0080, 0.0077, 0.0079, 0.0092, 0.0083, 0.0085, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 00:52:12,048 INFO [finetune.py:976] (3/7) Epoch 3, batch 100, loss[loss=0.2575, simple_loss=0.3027, pruned_loss=0.1061, over 4806.00 frames. ], tot_loss[loss=0.253, simple_loss=0.302, pruned_loss=0.102, over 380011.63 frames. ], batch size: 41, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:52:33,515 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=11589.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 00:52:36,420 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11593.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:52:47,407 INFO [finetune.py:976] (3/7) Epoch 3, batch 150, loss[loss=0.2302, simple_loss=0.2846, pruned_loss=0.08793, over 4819.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.2985, pruned_loss=0.1018, over 509874.01 frames. ], batch size: 51, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:53:00,296 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.308e+02 1.853e+02 2.230e+02 2.582e+02 4.758e+02, threshold=4.459e+02, percent-clipped=3.0 2023-03-26 00:53:24,125 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11635.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:53:46,448 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7148, 1.4217, 1.4914, 1.6638, 2.1454, 1.6743, 1.2216, 1.3835], device='cuda:3'), covar=tensor([0.2581, 0.2664, 0.2105, 0.2057, 0.2161, 0.1394, 0.3352, 0.2193], device='cuda:3'), in_proj_covar=tensor([0.0225, 0.0207, 0.0194, 0.0179, 0.0228, 0.0170, 0.0210, 0.0183], 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') 2023-03-26 00:53:48,108 INFO [finetune.py:976] (3/7) Epoch 3, batch 200, loss[loss=0.2276, simple_loss=0.2731, pruned_loss=0.09099, over 4690.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.2974, pruned_loss=0.1016, over 610103.35 frames. ], batch size: 23, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:54:35,503 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11696.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:54:37,312 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11699.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:54:44,849 INFO [finetune.py:976] (3/7) Epoch 3, batch 250, loss[loss=0.3103, simple_loss=0.3568, pruned_loss=0.1318, over 4731.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3014, pruned_loss=0.1028, over 689203.13 frames. ], batch size: 59, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:55:02,876 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.255e+02 1.783e+02 2.124e+02 2.629e+02 6.988e+02, threshold=4.248e+02, percent-clipped=1.0 2023-03-26 00:55:32,488 INFO [finetune.py:976] (3/7) Epoch 3, batch 300, loss[loss=0.2419, simple_loss=0.2952, pruned_loss=0.09427, over 4788.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3052, pruned_loss=0.1041, over 748439.11 frames. ], batch size: 26, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:55:40,982 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11760.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:56:13,665 INFO [finetune.py:976] (3/7) Epoch 3, batch 350, loss[loss=0.2482, simple_loss=0.2945, pruned_loss=0.1009, over 4218.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3054, pruned_loss=0.1036, over 791809.90 frames. ], batch size: 65, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:56:20,323 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.015e+02 1.891e+02 2.265e+02 2.576e+02 3.939e+02, threshold=4.529e+02, percent-clipped=0.0 2023-03-26 00:56:20,469 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11816.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:56:30,481 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11830.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:56:42,663 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-03-26 00:56:45,070 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11846.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:56:51,650 INFO [finetune.py:976] (3/7) Epoch 3, batch 400, loss[loss=0.2808, simple_loss=0.3274, pruned_loss=0.1171, over 4901.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3085, pruned_loss=0.1045, over 829041.61 frames. ], batch size: 36, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:57:20,642 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11877.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:57:21,183 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=11878.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:57:23,046 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1610, 1.8748, 2.4697, 4.0984, 3.0144, 2.5975, 0.9901, 3.3248], device='cuda:3'), covar=tensor([0.1856, 0.1621, 0.1504, 0.0489, 0.0753, 0.1721, 0.2091, 0.0537], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0120, 0.0138, 0.0164, 0.0105, 0.0145, 0.0130, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 00:57:33,138 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11888.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:57:55,602 INFO [finetune.py:976] (3/7) Epoch 3, batch 450, loss[loss=0.2545, simple_loss=0.3131, pruned_loss=0.09788, over 4904.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3075, pruned_loss=0.1033, over 859003.49 frames. ], batch size: 32, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:58:01,501 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11907.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:58:12,545 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.239e+02 1.814e+02 2.317e+02 2.793e+02 4.030e+02, threshold=4.633e+02, percent-clipped=0.0 2023-03-26 00:58:15,057 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11920.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:58:39,237 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11948.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:58:50,439 INFO [finetune.py:976] (3/7) Epoch 3, batch 500, loss[loss=0.2286, simple_loss=0.2767, pruned_loss=0.0902, over 4831.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3043, pruned_loss=0.1021, over 882379.31 frames. ], batch size: 33, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:58:58,238 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8656, 1.6232, 1.3722, 1.6806, 1.5233, 1.5196, 1.4678, 2.4766], device='cuda:3'), covar=tensor([1.2750, 1.3036, 1.0271, 1.3853, 1.1380, 0.7475, 1.4407, 0.3729], device='cuda:3'), in_proj_covar=tensor([0.0262, 0.0241, 0.0216, 0.0277, 0.0231, 0.0193, 0.0235, 0.0179], 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') 2023-03-26 00:59:19,991 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11981.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:59:27,099 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11984.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:59:31,810 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11991.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:59:46,860 INFO [finetune.py:976] (3/7) Epoch 3, batch 550, loss[loss=0.2467, simple_loss=0.2997, pruned_loss=0.09687, over 4758.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.2995, pruned_loss=0.1004, over 899130.80 frames. ], batch size: 59, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 00:59:49,290 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12009.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 00:59:53,392 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.083e+02 1.776e+02 2.066e+02 2.700e+02 4.009e+02, threshold=4.133e+02, percent-clipped=0.0 2023-03-26 01:00:02,298 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5974, 1.6088, 1.7320, 1.9330, 1.8163, 3.6646, 1.4135, 1.8513], device='cuda:3'), covar=tensor([0.1036, 0.1731, 0.1174, 0.1042, 0.1565, 0.0234, 0.1482, 0.1667], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0080, 0.0077, 0.0079, 0.0092, 0.0082, 0.0084, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 01:00:10,590 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9307, 1.8843, 1.7724, 2.0995, 1.3391, 4.6903, 1.6991, 2.3824], device='cuda:3'), covar=tensor([0.3218, 0.2221, 0.1908, 0.2069, 0.1837, 0.0106, 0.2531, 0.1343], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0109, 0.0114, 0.0117, 0.0114, 0.0096, 0.0099, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 01:00:13,576 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12045.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:00:21,763 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12055.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:00:22,311 INFO [finetune.py:976] (3/7) Epoch 3, batch 600, loss[loss=0.2969, simple_loss=0.3546, pruned_loss=0.1196, over 4760.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3009, pruned_loss=0.1017, over 911345.12 frames. ], batch size: 54, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 01:00:41,413 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12076.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:00:47,397 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.0537, 1.0299, 1.0954, 0.4139, 0.7335, 1.1881, 1.2317, 1.1078], device='cuda:3'), covar=tensor([0.1011, 0.0537, 0.0461, 0.0688, 0.0525, 0.0570, 0.0358, 0.0604], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0156, 0.0117, 0.0135, 0.0132, 0.0118, 0.0146, 0.0143], device='cuda:3'), out_proj_covar=tensor([9.7573e-05, 1.1579e-04, 8.5631e-05, 9.9222e-05, 9.5680e-05, 8.7229e-05, 1.0896e-04, 1.0614e-04], device='cuda:3') 2023-03-26 01:00:48,059 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-03-26 01:01:08,115 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-26 01:01:09,710 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-03-26 01:01:16,812 INFO [finetune.py:976] (3/7) Epoch 3, batch 650, loss[loss=0.2693, simple_loss=0.3226, pruned_loss=0.108, over 4708.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3033, pruned_loss=0.1026, over 918767.04 frames. ], batch size: 59, lr: 3.99e-03, grad_scale: 32.0 2023-03-26 01:01:23,452 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.190e+02 1.901e+02 2.250e+02 2.651e+02 5.885e+02, threshold=4.501e+02, percent-clipped=2.0 2023-03-26 01:01:50,153 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12137.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:02:02,048 INFO [finetune.py:976] (3/7) Epoch 3, batch 700, loss[loss=0.2438, simple_loss=0.2983, pruned_loss=0.09461, over 4891.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.306, pruned_loss=0.1038, over 924077.66 frames. ], batch size: 32, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:02:02,770 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.5669, 1.5799, 1.7117, 0.9226, 1.6331, 1.8266, 1.8688, 1.5598], device='cuda:3'), covar=tensor([0.1024, 0.0695, 0.0453, 0.0762, 0.0384, 0.0666, 0.0371, 0.0613], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0157, 0.0118, 0.0136, 0.0133, 0.0119, 0.0147, 0.0144], device='cuda:3'), out_proj_covar=tensor([9.8391e-05, 1.1709e-04, 8.6563e-05, 9.9923e-05, 9.6455e-05, 8.7789e-05, 1.0979e-04, 1.0685e-04], device='cuda:3') 2023-03-26 01:02:12,356 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12172.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:02:24,516 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12188.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:02:39,281 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12202.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:02:43,054 INFO [finetune.py:976] (3/7) Epoch 3, batch 750, loss[loss=0.2676, simple_loss=0.3144, pruned_loss=0.1104, over 4761.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3064, pruned_loss=0.1036, over 928716.49 frames. ], batch size: 27, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:02:54,948 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.207e+02 1.858e+02 2.308e+02 2.738e+02 5.308e+02, threshold=4.616e+02, percent-clipped=1.0 2023-03-26 01:03:14,110 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=12236.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:03:26,059 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.5428, 3.8912, 4.0680, 4.3589, 4.2332, 4.0122, 4.6206, 1.3499], device='cuda:3'), covar=tensor([0.0678, 0.0814, 0.0692, 0.0874, 0.1225, 0.1428, 0.0623, 0.5117], device='cuda:3'), in_proj_covar=tensor([0.0368, 0.0245, 0.0274, 0.0297, 0.0341, 0.0286, 0.0310, 0.0300], 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') 2023-03-26 01:03:30,123 INFO [finetune.py:976] (3/7) Epoch 3, batch 800, loss[loss=0.2347, simple_loss=0.2951, pruned_loss=0.08716, over 4736.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3066, pruned_loss=0.1038, over 933927.58 frames. ], batch size: 59, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:03:41,322 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-26 01:03:42,858 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12276.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:03:54,053 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-26 01:04:03,324 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12291.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:04:11,114 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12304.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:04:11,250 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-26 01:04:12,240 INFO [finetune.py:976] (3/7) Epoch 3, batch 850, loss[loss=0.2547, simple_loss=0.292, pruned_loss=0.1088, over 4742.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3034, pruned_loss=0.1025, over 937439.62 frames. ], batch size: 27, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:04:19,500 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.224e+02 1.828e+02 2.109e+02 2.576e+02 5.946e+02, threshold=4.217e+02, percent-clipped=1.0 2023-03-26 01:04:46,214 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=12339.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:04:46,836 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12340.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:05:05,338 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.16 vs. limit=5.0 2023-03-26 01:05:06,400 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12355.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:05:06,917 INFO [finetune.py:976] (3/7) Epoch 3, batch 900, loss[loss=0.2118, simple_loss=0.2717, pruned_loss=0.07598, over 4816.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3009, pruned_loss=0.1016, over 941551.61 frames. ], batch size: 39, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:05:40,647 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4336, 1.2421, 1.2246, 1.4492, 1.4732, 1.3794, 0.8534, 1.2284], device='cuda:3'), covar=tensor([0.2772, 0.2676, 0.2299, 0.2125, 0.2057, 0.1432, 0.3191, 0.2222], device='cuda:3'), in_proj_covar=tensor([0.0223, 0.0205, 0.0192, 0.0178, 0.0227, 0.0169, 0.0209, 0.0182], 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') 2023-03-26 01:05:43,766 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=12403.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:05:45,595 INFO [finetune.py:976] (3/7) Epoch 3, batch 950, loss[loss=0.2465, simple_loss=0.2927, pruned_loss=0.1001, over 4827.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3, pruned_loss=0.1017, over 942598.78 frames. ], batch size: 33, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:05:49,421 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7177, 1.5109, 1.4863, 1.6719, 2.1360, 1.7161, 1.3082, 1.4000], device='cuda:3'), covar=tensor([0.2283, 0.2411, 0.2014, 0.1897, 0.2199, 0.1359, 0.2931, 0.1888], device='cuda:3'), in_proj_covar=tensor([0.0225, 0.0206, 0.0193, 0.0179, 0.0229, 0.0171, 0.0210, 0.0183], 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') 2023-03-26 01:05:57,754 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.215e+02 1.779e+02 2.123e+02 2.532e+02 4.452e+02, threshold=4.246e+02, percent-clipped=1.0 2023-03-26 01:06:12,422 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12432.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:06:22,017 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12446.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:06:28,915 INFO [finetune.py:976] (3/7) Epoch 3, batch 1000, loss[loss=0.2562, simple_loss=0.3023, pruned_loss=0.1051, over 4891.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.303, pruned_loss=0.1033, over 944003.84 frames. ], batch size: 32, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:06:39,166 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12472.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:07:17,557 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12502.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:07:18,757 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7755, 1.6070, 2.1595, 3.3694, 2.4681, 2.4277, 0.7878, 2.6545], device='cuda:3'), covar=tensor([0.1855, 0.1680, 0.1370, 0.0594, 0.0826, 0.1450, 0.2266, 0.0660], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0120, 0.0138, 0.0165, 0.0105, 0.0145, 0.0131, 0.0107], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 01:07:19,894 INFO [finetune.py:976] (3/7) Epoch 3, batch 1050, loss[loss=0.3181, simple_loss=0.3567, pruned_loss=0.1398, over 4889.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3069, pruned_loss=0.1043, over 948214.59 frames. ], batch size: 35, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:07:20,634 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12507.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:07:31,080 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.261e+02 2.008e+02 2.380e+02 2.733e+02 7.204e+02, threshold=4.759e+02, percent-clipped=3.0 2023-03-26 01:07:38,118 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=12520.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:08:10,387 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=12550.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:08:17,959 INFO [finetune.py:976] (3/7) Epoch 3, batch 1100, loss[loss=0.2366, simple_loss=0.2972, pruned_loss=0.08802, over 4820.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3089, pruned_loss=0.1046, over 951188.41 frames. ], batch size: 33, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:08:35,467 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12576.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:08:39,122 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.5409, 1.4674, 1.6583, 0.7857, 1.5703, 1.7099, 1.7910, 1.5385], device='cuda:3'), covar=tensor([0.1059, 0.0738, 0.0442, 0.0759, 0.0443, 0.0496, 0.0361, 0.0562], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0156, 0.0117, 0.0135, 0.0132, 0.0119, 0.0146, 0.0144], device='cuda:3'), out_proj_covar=tensor([9.7418e-05, 1.1617e-04, 8.5844e-05, 9.8955e-05, 9.5761e-05, 8.7563e-05, 1.0930e-04, 1.0666e-04], device='cuda:3') 2023-03-26 01:08:52,541 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.76 vs. limit=5.0 2023-03-26 01:08:58,305 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12604.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:08:59,395 INFO [finetune.py:976] (3/7) Epoch 3, batch 1150, loss[loss=0.2767, simple_loss=0.3213, pruned_loss=0.116, over 4883.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3085, pruned_loss=0.1035, over 951972.19 frames. ], batch size: 43, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:09:11,543 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.144e+02 1.712e+02 1.953e+02 2.432e+02 5.551e+02, threshold=3.906e+02, percent-clipped=1.0 2023-03-26 01:09:11,690 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6063, 1.4839, 1.4140, 1.6305, 2.0058, 1.5769, 1.2237, 1.3332], device='cuda:3'), covar=tensor([0.2419, 0.2473, 0.2106, 0.1915, 0.2169, 0.1482, 0.3202, 0.1991], device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0207, 0.0194, 0.0180, 0.0229, 0.0171, 0.0211, 0.0184], 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') 2023-03-26 01:09:21,031 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=12624.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:09:42,279 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12640.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:09:55,582 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=12652.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:10:03,871 INFO [finetune.py:976] (3/7) Epoch 3, batch 1200, loss[loss=0.2818, simple_loss=0.321, pruned_loss=0.1213, over 4169.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3066, pruned_loss=0.1028, over 951435.12 frames. ], batch size: 65, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:10:06,423 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.67 vs. limit=5.0 2023-03-26 01:10:15,207 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6307, 1.5570, 1.5523, 1.6720, 1.0041, 3.5678, 1.3386, 1.8038], device='cuda:3'), covar=tensor([0.3428, 0.2371, 0.2025, 0.2218, 0.2012, 0.0147, 0.2655, 0.1512], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0110, 0.0114, 0.0118, 0.0115, 0.0096, 0.0100, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 01:10:31,613 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=12688.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:10:43,570 INFO [finetune.py:976] (3/7) Epoch 3, batch 1250, loss[loss=0.2154, simple_loss=0.2581, pruned_loss=0.08639, over 4860.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3024, pruned_loss=0.1012, over 950629.96 frames. ], batch size: 49, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:10:51,325 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.066e+02 1.791e+02 2.152e+02 2.550e+02 3.946e+02, threshold=4.304e+02, percent-clipped=1.0 2023-03-26 01:11:06,473 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12732.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:11:25,780 INFO [finetune.py:976] (3/7) Epoch 3, batch 1300, loss[loss=0.2302, simple_loss=0.2957, pruned_loss=0.08232, over 4822.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.2998, pruned_loss=0.09998, over 953723.42 frames. ], batch size: 40, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:11:26,785 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-26 01:11:31,205 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12763.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:11:48,951 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=12780.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:12:19,503 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12802.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:12:22,421 INFO [finetune.py:976] (3/7) Epoch 3, batch 1350, loss[loss=0.2828, simple_loss=0.3135, pruned_loss=0.1261, over 4901.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.2998, pruned_loss=0.1003, over 954392.39 frames. ], batch size: 35, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:12:40,681 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.079e+02 1.832e+02 2.182e+02 2.674e+02 3.468e+02, threshold=4.364e+02, percent-clipped=0.0 2023-03-26 01:12:50,687 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12824.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 01:13:21,547 INFO [finetune.py:976] (3/7) Epoch 3, batch 1400, loss[loss=0.3145, simple_loss=0.3595, pruned_loss=0.1348, over 4817.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3051, pruned_loss=0.1023, over 956832.88 frames. ], batch size: 45, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:13:50,008 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.58 vs. limit=5.0 2023-03-26 01:14:18,028 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0187, 1.7656, 1.3896, 2.0933, 2.0286, 1.6716, 2.4031, 1.8177], device='cuda:3'), covar=tensor([0.2289, 0.5185, 0.5749, 0.4671, 0.3675, 0.2652, 0.4641, 0.3274], device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0197, 0.0239, 0.0254, 0.0220, 0.0184, 0.0208, 0.0189], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 01:14:19,107 INFO [finetune.py:976] (3/7) Epoch 3, batch 1450, loss[loss=0.2644, simple_loss=0.329, pruned_loss=0.09995, over 4815.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3055, pruned_loss=0.1017, over 955533.59 frames. ], batch size: 38, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:14:36,815 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-03-26 01:14:38,011 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.344e+02 1.849e+02 2.229e+02 2.789e+02 4.972e+02, threshold=4.459e+02, percent-clipped=1.0 2023-03-26 01:15:13,776 INFO [finetune.py:976] (3/7) Epoch 3, batch 1500, loss[loss=0.2218, simple_loss=0.2506, pruned_loss=0.09647, over 4405.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3083, pruned_loss=0.1035, over 954112.80 frames. ], batch size: 19, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:15:13,881 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.4006, 3.8754, 3.9342, 4.2390, 4.0982, 3.8926, 4.4961, 1.3615], device='cuda:3'), covar=tensor([0.0725, 0.0732, 0.0743, 0.0945, 0.1225, 0.1345, 0.0551, 0.5187], device='cuda:3'), in_proj_covar=tensor([0.0371, 0.0246, 0.0277, 0.0298, 0.0343, 0.0288, 0.0313, 0.0303], 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') 2023-03-26 01:15:29,720 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-03-26 01:15:58,820 INFO [finetune.py:976] (3/7) Epoch 3, batch 1550, loss[loss=0.2399, simple_loss=0.2787, pruned_loss=0.1006, over 4734.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3067, pruned_loss=0.1021, over 953275.17 frames. ], batch size: 23, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:16:09,621 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.917e+01 1.840e+02 2.219e+02 2.788e+02 8.539e+02, threshold=4.437e+02, percent-clipped=2.0 2023-03-26 01:16:44,395 INFO [finetune.py:976] (3/7) Epoch 3, batch 1600, loss[loss=0.1995, simple_loss=0.2626, pruned_loss=0.06823, over 4890.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3038, pruned_loss=0.1007, over 953714.88 frames. ], batch size: 35, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:17:41,255 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=13102.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:17:43,645 INFO [finetune.py:976] (3/7) Epoch 3, batch 1650, loss[loss=0.2149, simple_loss=0.2636, pruned_loss=0.08308, over 4935.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3003, pruned_loss=0.09929, over 954862.96 frames. ], batch size: 33, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:17:55,147 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.264e+02 1.841e+02 2.109e+02 2.450e+02 4.226e+02, threshold=4.217e+02, percent-clipped=0.0 2023-03-26 01:17:56,994 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=13119.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 01:18:32,151 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=13150.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:18:41,622 INFO [finetune.py:976] (3/7) Epoch 3, batch 1700, loss[loss=0.2575, simple_loss=0.306, pruned_loss=0.1045, over 4892.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.2993, pruned_loss=0.09918, over 955955.53 frames. ], batch size: 35, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:19:22,357 INFO [finetune.py:976] (3/7) Epoch 3, batch 1750, loss[loss=0.2852, simple_loss=0.3286, pruned_loss=0.1209, over 4831.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.301, pruned_loss=0.09997, over 955267.37 frames. ], batch size: 33, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:19:31,217 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.198e+02 1.973e+02 2.314e+02 2.693e+02 6.749e+02, threshold=4.629e+02, percent-clipped=3.0 2023-03-26 01:20:07,994 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3396, 1.5158, 0.9677, 2.1525, 2.4202, 1.7774, 1.6367, 2.1000], device='cuda:3'), covar=tensor([0.1427, 0.2201, 0.2242, 0.1178, 0.2065, 0.1915, 0.1492, 0.2057], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0098, 0.0116, 0.0094, 0.0123, 0.0097, 0.0100, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 01:20:09,787 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6311, 3.4760, 3.4384, 1.5469, 3.6846, 2.7544, 0.8085, 2.4254], device='cuda:3'), covar=tensor([0.3128, 0.1624, 0.1534, 0.3294, 0.0992, 0.0925, 0.4327, 0.1466], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0169, 0.0166, 0.0129, 0.0157, 0.0121, 0.0147, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 01:20:18,404 INFO [finetune.py:976] (3/7) Epoch 3, batch 1800, loss[loss=0.2422, simple_loss=0.2991, pruned_loss=0.09268, over 4848.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3032, pruned_loss=0.0998, over 954497.23 frames. ], batch size: 47, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:20:59,915 INFO [finetune.py:976] (3/7) Epoch 3, batch 1850, loss[loss=0.2715, simple_loss=0.3231, pruned_loss=0.11, over 4819.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3055, pruned_loss=0.1012, over 954819.61 frames. ], batch size: 39, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:21:08,031 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.104e+02 1.691e+02 1.934e+02 2.488e+02 4.482e+02, threshold=3.868e+02, percent-clipped=0.0 2023-03-26 01:21:50,069 INFO [finetune.py:976] (3/7) Epoch 3, batch 1900, loss[loss=0.2721, simple_loss=0.3254, pruned_loss=0.1094, over 4804.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3061, pruned_loss=0.1014, over 952897.91 frames. ], batch size: 39, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:22:08,793 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.13 vs. limit=5.0 2023-03-26 01:22:40,354 INFO [finetune.py:976] (3/7) Epoch 3, batch 1950, loss[loss=0.1943, simple_loss=0.2522, pruned_loss=0.06822, over 4792.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3044, pruned_loss=0.1008, over 954257.31 frames. ], batch size: 45, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:22:46,999 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.125e+02 1.849e+02 2.191e+02 2.472e+02 6.030e+02, threshold=4.381e+02, percent-clipped=4.0 2023-03-26 01:22:48,813 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=13419.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:23:25,633 INFO [finetune.py:976] (3/7) Epoch 3, batch 2000, loss[loss=0.1934, simple_loss=0.2516, pruned_loss=0.06761, over 4735.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3008, pruned_loss=0.09925, over 953576.48 frames. ], batch size: 23, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:23:26,755 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2023-03-26 01:23:36,602 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=13467.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:24:12,571 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1048, 1.8850, 1.5019, 2.0315, 1.7950, 1.7940, 1.7854, 2.7239], device='cuda:3'), covar=tensor([1.2803, 1.3670, 1.0028, 1.4553, 1.1236, 0.7319, 1.3349, 0.4188], device='cuda:3'), in_proj_covar=tensor([0.0267, 0.0244, 0.0217, 0.0280, 0.0232, 0.0194, 0.0236, 0.0182], 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') 2023-03-26 01:24:18,863 INFO [finetune.py:976] (3/7) Epoch 3, batch 2050, loss[loss=0.2228, simple_loss=0.2832, pruned_loss=0.08119, over 4907.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.297, pruned_loss=0.0977, over 956042.84 frames. ], batch size: 32, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:24:33,951 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.160e+02 1.806e+02 2.206e+02 2.674e+02 5.377e+02, threshold=4.412e+02, percent-clipped=2.0 2023-03-26 01:25:00,754 INFO [finetune.py:976] (3/7) Epoch 3, batch 2100, loss[loss=0.3123, simple_loss=0.3528, pruned_loss=0.1358, over 4209.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.2972, pruned_loss=0.09802, over 955455.81 frames. ], batch size: 65, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:25:23,156 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-03-26 01:25:33,711 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9047, 2.0546, 1.8656, 1.4395, 2.2020, 2.1585, 2.0737, 1.7436], device='cuda:3'), covar=tensor([0.0619, 0.0516, 0.0784, 0.0931, 0.0491, 0.0629, 0.0595, 0.0988], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0134, 0.0146, 0.0131, 0.0112, 0.0145, 0.0150, 0.0165], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 01:25:45,304 INFO [finetune.py:976] (3/7) Epoch 3, batch 2150, loss[loss=0.2524, simple_loss=0.3086, pruned_loss=0.09809, over 4860.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.302, pruned_loss=0.1003, over 955355.52 frames. ], batch size: 49, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:26:01,365 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.135e+02 1.918e+02 2.256e+02 2.684e+02 5.304e+02, threshold=4.512e+02, percent-clipped=2.0 2023-03-26 01:26:07,020 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8400, 1.4509, 1.6372, 1.6396, 1.4654, 1.5199, 1.5989, 1.5632], device='cuda:3'), covar=tensor([1.1144, 1.7291, 1.3491, 1.6475, 1.7268, 1.1928, 2.0702, 1.2604], device='cuda:3'), in_proj_covar=tensor([0.0228, 0.0256, 0.0252, 0.0268, 0.0244, 0.0218, 0.0279, 0.0219], 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') 2023-03-26 01:26:27,570 INFO [finetune.py:976] (3/7) Epoch 3, batch 2200, loss[loss=0.2186, simple_loss=0.2777, pruned_loss=0.07978, over 4761.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3033, pruned_loss=0.1005, over 955960.93 frames. ], batch size: 26, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:26:27,712 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.3741, 1.5470, 1.6858, 1.0072, 1.5839, 1.8070, 1.8585, 1.5510], device='cuda:3'), covar=tensor([0.0833, 0.0668, 0.0478, 0.0530, 0.0380, 0.0431, 0.0302, 0.0523], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0158, 0.0118, 0.0136, 0.0133, 0.0120, 0.0147, 0.0144], device='cuda:3'), out_proj_covar=tensor([9.8272e-05, 1.1752e-04, 8.6482e-05, 9.9919e-05, 9.6361e-05, 8.8466e-05, 1.0966e-04, 1.0687e-04], device='cuda:3') 2023-03-26 01:26:43,338 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6097, 1.4552, 1.5845, 1.8924, 1.5675, 3.1989, 1.1875, 1.5355], device='cuda:3'), covar=tensor([0.0921, 0.1777, 0.1309, 0.0936, 0.1692, 0.0262, 0.1566, 0.1878], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0082, 0.0079, 0.0080, 0.0094, 0.0084, 0.0086, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 01:27:00,298 INFO [finetune.py:976] (3/7) Epoch 3, batch 2250, loss[loss=0.2162, simple_loss=0.2847, pruned_loss=0.07388, over 4856.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.304, pruned_loss=0.1006, over 956135.43 frames. ], batch size: 31, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:27:08,389 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.215e+02 1.927e+02 2.166e+02 2.564e+02 5.587e+02, threshold=4.333e+02, percent-clipped=2.0 2023-03-26 01:27:24,306 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=13737.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:27:42,150 INFO [finetune.py:976] (3/7) Epoch 3, batch 2300, loss[loss=0.2104, simple_loss=0.2715, pruned_loss=0.0747, over 4919.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3042, pruned_loss=0.09976, over 957028.14 frames. ], batch size: 38, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:28:25,835 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=13798.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 01:28:36,682 INFO [finetune.py:976] (3/7) Epoch 3, batch 2350, loss[loss=0.276, simple_loss=0.3131, pruned_loss=0.1195, over 4799.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3005, pruned_loss=0.09846, over 955056.67 frames. ], batch size: 45, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:28:36,794 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3582, 1.4240, 1.3669, 0.7759, 1.6202, 1.4907, 1.4294, 1.3656], device='cuda:3'), covar=tensor([0.0667, 0.0704, 0.0738, 0.1026, 0.0645, 0.0701, 0.0699, 0.1102], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0134, 0.0146, 0.0131, 0.0111, 0.0145, 0.0149, 0.0165], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 01:28:50,257 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.589e+01 1.791e+02 2.182e+02 2.579e+02 6.380e+02, threshold=4.365e+02, percent-clipped=2.0 2023-03-26 01:29:27,911 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9530, 1.7899, 1.5288, 1.9299, 2.0626, 1.6427, 2.3808, 1.9265], device='cuda:3'), covar=tensor([0.2050, 0.4201, 0.4571, 0.4134, 0.3130, 0.2229, 0.3508, 0.2808], device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0195, 0.0238, 0.0253, 0.0218, 0.0183, 0.0207, 0.0187], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 01:29:31,312 INFO [finetune.py:976] (3/7) Epoch 3, batch 2400, loss[loss=0.2254, simple_loss=0.2766, pruned_loss=0.08705, over 4877.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.298, pruned_loss=0.09777, over 953389.27 frames. ], batch size: 31, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:30:15,689 INFO [finetune.py:976] (3/7) Epoch 3, batch 2450, loss[loss=0.2285, simple_loss=0.285, pruned_loss=0.08595, over 4819.00 frames. ], tot_loss[loss=0.244, simple_loss=0.2953, pruned_loss=0.09638, over 956228.76 frames. ], batch size: 39, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:30:26,396 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.03 vs. limit=5.0 2023-03-26 01:30:29,106 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.953e+01 1.906e+02 2.150e+02 2.578e+02 4.181e+02, threshold=4.299e+02, percent-clipped=0.0 2023-03-26 01:30:59,052 INFO [finetune.py:976] (3/7) Epoch 3, batch 2500, loss[loss=0.2198, simple_loss=0.2899, pruned_loss=0.07483, over 4895.00 frames. ], tot_loss[loss=0.247, simple_loss=0.2979, pruned_loss=0.09806, over 955084.13 frames. ], batch size: 32, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:31:21,483 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.7319, 1.6247, 1.6356, 1.1802, 1.7927, 1.8626, 1.8490, 1.5635], device='cuda:3'), covar=tensor([0.0927, 0.0625, 0.0523, 0.0572, 0.0381, 0.0505, 0.0355, 0.0632], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0157, 0.0118, 0.0136, 0.0132, 0.0119, 0.0146, 0.0143], device='cuda:3'), out_proj_covar=tensor([9.8036e-05, 1.1708e-04, 8.5745e-05, 9.9785e-05, 9.5436e-05, 8.8216e-05, 1.0903e-04, 1.0623e-04], device='cuda:3') 2023-03-26 01:31:40,714 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5765, 1.4108, 1.0863, 0.3180, 1.2348, 1.4125, 1.2179, 1.4125], device='cuda:3'), covar=tensor([0.0689, 0.0640, 0.1070, 0.1714, 0.0959, 0.1779, 0.1723, 0.0667], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0197, 0.0204, 0.0189, 0.0215, 0.0209, 0.0216, 0.0200], 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') 2023-03-26 01:31:53,428 INFO [finetune.py:976] (3/7) Epoch 3, batch 2550, loss[loss=0.2729, simple_loss=0.3153, pruned_loss=0.1152, over 4897.00 frames. ], tot_loss[loss=0.252, simple_loss=0.303, pruned_loss=0.1005, over 954483.34 frames. ], batch size: 32, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:32:01,952 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.3264, 2.9224, 3.0514, 3.2457, 3.0951, 2.9036, 3.3965, 1.1132], device='cuda:3'), covar=tensor([0.1064, 0.0879, 0.1054, 0.1145, 0.1561, 0.1540, 0.0953, 0.4901], device='cuda:3'), in_proj_covar=tensor([0.0365, 0.0242, 0.0272, 0.0292, 0.0337, 0.0283, 0.0308, 0.0299], 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') 2023-03-26 01:32:02,448 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.924e+01 1.822e+02 2.075e+02 2.520e+02 4.375e+02, threshold=4.150e+02, percent-clipped=1.0 2023-03-26 01:32:19,254 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7048, 1.1855, 0.8547, 1.6286, 1.9615, 1.1195, 1.3697, 1.5778], device='cuda:3'), covar=tensor([0.1686, 0.2292, 0.2179, 0.1267, 0.2137, 0.2195, 0.1531, 0.2126], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0099, 0.0117, 0.0094, 0.0124, 0.0098, 0.0100, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 01:32:30,058 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9258, 1.6209, 2.5478, 3.8265, 2.6970, 2.6425, 0.9256, 2.9871], device='cuda:3'), covar=tensor([0.1914, 0.1668, 0.1300, 0.0463, 0.0836, 0.1661, 0.2099, 0.0611], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0120, 0.0139, 0.0165, 0.0105, 0.0146, 0.0131, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 01:32:30,595 INFO [finetune.py:976] (3/7) Epoch 3, batch 2600, loss[loss=0.2291, simple_loss=0.2891, pruned_loss=0.08456, over 4798.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3041, pruned_loss=0.1009, over 954590.24 frames. ], batch size: 41, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:32:31,436 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.28 vs. limit=5.0 2023-03-26 01:32:38,338 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-26 01:32:41,347 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14072.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:33:04,942 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14093.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 01:33:16,421 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14103.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:33:18,155 INFO [finetune.py:976] (3/7) Epoch 3, batch 2650, loss[loss=0.2455, simple_loss=0.3041, pruned_loss=0.09344, over 4819.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3041, pruned_loss=0.1004, over 955268.99 frames. ], batch size: 33, lr: 3.98e-03, grad_scale: 16.0 2023-03-26 01:33:28,080 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8902, 1.6559, 1.3680, 1.8005, 1.9155, 1.5879, 2.1463, 1.8075], device='cuda:3'), covar=tensor([0.2242, 0.4537, 0.5333, 0.4719, 0.3525, 0.2430, 0.4444, 0.3296], device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0196, 0.0239, 0.0254, 0.0220, 0.0184, 0.0208, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 01:33:34,829 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.159e+02 1.796e+02 2.195e+02 2.771e+02 4.502e+02, threshold=4.390e+02, percent-clipped=2.0 2023-03-26 01:33:37,259 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14120.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 01:33:55,980 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.72 vs. limit=5.0 2023-03-26 01:33:56,208 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14133.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:34:18,023 INFO [finetune.py:976] (3/7) Epoch 3, batch 2700, loss[loss=0.2093, simple_loss=0.2579, pruned_loss=0.08039, over 4795.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3019, pruned_loss=0.09894, over 955888.31 frames. ], batch size: 25, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:34:28,736 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14164.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:34:50,949 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14181.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 01:34:57,081 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2924, 2.0808, 2.2039, 1.3045, 2.3514, 2.5858, 2.1395, 2.1607], device='cuda:3'), covar=tensor([0.1172, 0.0817, 0.0475, 0.0795, 0.0543, 0.0722, 0.0559, 0.0756], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0157, 0.0118, 0.0135, 0.0132, 0.0119, 0.0146, 0.0143], device='cuda:3'), out_proj_covar=tensor([9.7996e-05, 1.1688e-04, 8.5758e-05, 9.9368e-05, 9.5397e-05, 8.7804e-05, 1.0930e-04, 1.0578e-04], device='cuda:3') 2023-03-26 01:35:02,808 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-03-26 01:35:13,611 INFO [finetune.py:976] (3/7) Epoch 3, batch 2750, loss[loss=0.2606, simple_loss=0.3048, pruned_loss=0.1081, over 4874.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.2985, pruned_loss=0.09781, over 955617.91 frames. ], batch size: 31, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:35:20,820 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.010e+02 1.613e+02 1.949e+02 2.418e+02 3.837e+02, threshold=3.898e+02, percent-clipped=0.0 2023-03-26 01:35:22,042 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.0602, 0.9789, 1.0356, 0.3499, 0.7232, 1.1570, 1.2005, 1.0845], device='cuda:3'), covar=tensor([0.1011, 0.0590, 0.0482, 0.0696, 0.0548, 0.0513, 0.0354, 0.0529], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0157, 0.0118, 0.0136, 0.0132, 0.0119, 0.0147, 0.0143], device='cuda:3'), out_proj_covar=tensor([9.8238e-05, 1.1715e-04, 8.5829e-05, 9.9739e-05, 9.5576e-05, 8.7894e-05, 1.0947e-04, 1.0589e-04], device='cuda:3') 2023-03-26 01:35:44,210 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5028, 1.3920, 2.0139, 2.8541, 1.9507, 2.2269, 1.0494, 2.3306], device='cuda:3'), covar=tensor([0.1857, 0.1663, 0.1128, 0.0679, 0.0885, 0.1219, 0.1769, 0.0680], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0119, 0.0138, 0.0165, 0.0104, 0.0145, 0.0130, 0.0107], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 01:35:50,233 INFO [finetune.py:976] (3/7) Epoch 3, batch 2800, loss[loss=0.1971, simple_loss=0.256, pruned_loss=0.0691, over 4828.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.2942, pruned_loss=0.09614, over 955921.64 frames. ], batch size: 39, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:36:23,753 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5305, 1.4569, 2.0777, 3.2339, 2.2276, 2.3182, 0.9251, 2.5134], device='cuda:3'), covar=tensor([0.1984, 0.1682, 0.1412, 0.0552, 0.0884, 0.1493, 0.2092, 0.0715], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0120, 0.0138, 0.0165, 0.0105, 0.0145, 0.0130, 0.0107], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 01:36:35,463 INFO [finetune.py:976] (3/7) Epoch 3, batch 2850, loss[loss=0.1812, simple_loss=0.235, pruned_loss=0.06368, over 4798.00 frames. ], tot_loss[loss=0.2428, simple_loss=0.2935, pruned_loss=0.09604, over 955607.33 frames. ], batch size: 29, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:36:47,133 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-03-26 01:36:48,617 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.210e+02 1.714e+02 2.069e+02 2.397e+02 3.427e+02, threshold=4.138e+02, percent-clipped=0.0 2023-03-26 01:37:20,820 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0015, 1.9711, 1.9432, 1.4332, 2.2447, 2.2444, 2.1016, 1.6155], device='cuda:3'), covar=tensor([0.0743, 0.0679, 0.0877, 0.1010, 0.0453, 0.0691, 0.0728, 0.1272], device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0133, 0.0146, 0.0130, 0.0110, 0.0143, 0.0149, 0.0164], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 01:37:26,620 INFO [finetune.py:976] (3/7) Epoch 3, batch 2900, loss[loss=0.2308, simple_loss=0.2914, pruned_loss=0.0851, over 4930.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.2968, pruned_loss=0.09804, over 955746.67 frames. ], batch size: 33, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:38:06,341 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14393.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:38:25,744 INFO [finetune.py:976] (3/7) Epoch 3, batch 2950, loss[loss=0.2608, simple_loss=0.3107, pruned_loss=0.1055, over 4784.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.2998, pruned_loss=0.09876, over 955155.76 frames. ], batch size: 51, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:38:35,818 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14413.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:38:37,077 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8211, 1.7208, 1.6766, 1.8301, 1.4954, 4.3875, 1.7924, 2.3717], device='cuda:3'), covar=tensor([0.3478, 0.2398, 0.2025, 0.2255, 0.1718, 0.0101, 0.2568, 0.1409], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0111, 0.0116, 0.0119, 0.0116, 0.0097, 0.0101, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 01:38:38,198 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.306e+02 1.859e+02 2.169e+02 2.721e+02 5.785e+02, threshold=4.339e+02, percent-clipped=3.0 2023-03-26 01:38:50,262 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14428.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:39:02,086 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=14441.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:39:20,294 INFO [finetune.py:976] (3/7) Epoch 3, batch 3000, loss[loss=0.299, simple_loss=0.3337, pruned_loss=0.1322, over 4214.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3014, pruned_loss=0.09932, over 953558.40 frames. ], batch size: 66, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:39:20,294 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 01:39:31,920 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7080, 1.4962, 1.5227, 1.7490, 2.0779, 1.7165, 1.1097, 1.4779], device='cuda:3'), covar=tensor([0.2520, 0.2687, 0.2266, 0.1970, 0.1903, 0.1369, 0.3222, 0.2021], device='cuda:3'), in_proj_covar=tensor([0.0227, 0.0206, 0.0193, 0.0180, 0.0229, 0.0170, 0.0210, 0.0184], 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') 2023-03-26 01:39:32,311 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.8214, 3.3829, 3.4884, 3.6872, 3.5479, 3.4125, 3.9007, 1.3646], device='cuda:3'), covar=tensor([0.0903, 0.0825, 0.0857, 0.1008, 0.1449, 0.1427, 0.0693, 0.4863], device='cuda:3'), in_proj_covar=tensor([0.0368, 0.0244, 0.0274, 0.0295, 0.0340, 0.0285, 0.0311, 0.0301], 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') 2023-03-26 01:39:34,191 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8457, 1.0077, 1.7572, 1.6223, 1.5875, 1.5107, 1.3937, 1.6161], device='cuda:3'), covar=tensor([0.8115, 1.4099, 1.1218, 1.0869, 1.3270, 0.8891, 1.5040, 1.0112], device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0254, 0.0252, 0.0266, 0.0243, 0.0217, 0.0277, 0.0218], 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') 2023-03-26 01:39:37,117 INFO [finetune.py:1010] (3/7) Epoch 3, validation: loss=0.1777, simple_loss=0.2485, pruned_loss=0.05342, over 2265189.00 frames. 2023-03-26 01:39:37,117 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 01:39:42,138 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14459.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:39:56,812 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14474.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:40:03,361 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14476.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 01:40:13,816 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2023-03-26 01:40:15,396 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14487.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:40:17,836 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.38 vs. limit=5.0 2023-03-26 01:40:36,152 INFO [finetune.py:976] (3/7) Epoch 3, batch 3050, loss[loss=0.2575, simple_loss=0.3197, pruned_loss=0.09767, over 4809.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3016, pruned_loss=0.09853, over 954533.18 frames. ], batch size: 45, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:40:44,357 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7449, 1.1827, 1.6101, 1.5044, 1.4229, 1.4276, 1.4262, 1.4738], device='cuda:3'), covar=tensor([0.9649, 1.6389, 1.2328, 1.4128, 1.5330, 1.0684, 1.7991, 1.1637], device='cuda:3'), in_proj_covar=tensor([0.0227, 0.0255, 0.0253, 0.0267, 0.0244, 0.0218, 0.0279, 0.0219], 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') 2023-03-26 01:40:52,886 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.290e+02 1.934e+02 2.277e+02 2.724e+02 4.940e+02, threshold=4.554e+02, percent-clipped=2.0 2023-03-26 01:40:57,870 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0211, 1.8622, 1.4619, 1.9632, 2.0801, 1.7210, 2.4294, 1.9843], device='cuda:3'), covar=tensor([0.2216, 0.4271, 0.5282, 0.4385, 0.3572, 0.2426, 0.4225, 0.2851], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0196, 0.0239, 0.0254, 0.0221, 0.0184, 0.0209, 0.0189], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 01:41:17,994 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14548.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:41:22,792 INFO [finetune.py:976] (3/7) Epoch 3, batch 3100, loss[loss=0.2302, simple_loss=0.2612, pruned_loss=0.09959, over 4116.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.2996, pruned_loss=0.09751, over 954305.41 frames. ], batch size: 17, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:41:51,358 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9168, 1.6391, 1.3962, 1.6131, 1.6100, 1.6203, 1.5592, 2.4989], device='cuda:3'), covar=tensor([1.0950, 1.2239, 0.9117, 1.2242, 1.0169, 0.6276, 1.2117, 0.3330], device='cuda:3'), in_proj_covar=tensor([0.0270, 0.0246, 0.0218, 0.0282, 0.0234, 0.0195, 0.0237, 0.0183], 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') 2023-03-26 01:42:10,663 INFO [finetune.py:976] (3/7) Epoch 3, batch 3150, loss[loss=0.2385, simple_loss=0.2621, pruned_loss=0.1074, over 4454.00 frames. ], tot_loss[loss=0.245, simple_loss=0.2968, pruned_loss=0.09663, over 955303.62 frames. ], batch size: 19, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:42:18,344 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.181e+02 1.775e+02 2.189e+02 2.683e+02 4.981e+02, threshold=4.378e+02, percent-clipped=2.0 2023-03-26 01:42:28,641 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7501, 1.5195, 1.5066, 1.3641, 1.8288, 1.5376, 1.8005, 1.7051], device='cuda:3'), covar=tensor([0.2112, 0.4091, 0.4715, 0.3799, 0.3366, 0.2207, 0.3476, 0.2759], device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0195, 0.0238, 0.0254, 0.0220, 0.0184, 0.0208, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 01:43:00,563 INFO [finetune.py:976] (3/7) Epoch 3, batch 3200, loss[loss=0.2119, simple_loss=0.2725, pruned_loss=0.07564, over 4846.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.292, pruned_loss=0.09471, over 955616.10 frames. ], batch size: 47, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:43:14,921 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-03-26 01:43:27,380 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7482, 1.7464, 1.6400, 1.7705, 1.1061, 3.9540, 1.6709, 2.2094], device='cuda:3'), covar=tensor([0.3319, 0.2221, 0.1941, 0.2197, 0.1958, 0.0126, 0.2625, 0.1423], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0111, 0.0116, 0.0119, 0.0116, 0.0097, 0.0101, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 01:43:41,545 INFO [finetune.py:976] (3/7) Epoch 3, batch 3250, loss[loss=0.2372, simple_loss=0.2963, pruned_loss=0.08902, over 4917.00 frames. ], tot_loss[loss=0.2412, simple_loss=0.2922, pruned_loss=0.09507, over 954005.61 frames. ], batch size: 36, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:43:54,505 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.171e+02 1.718e+02 2.102e+02 2.544e+02 5.358e+02, threshold=4.204e+02, percent-clipped=1.0 2023-03-26 01:44:08,060 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14728.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:44:15,464 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14740.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:44:21,482 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-26 01:44:22,306 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.81 vs. limit=5.0 2023-03-26 01:44:31,815 INFO [finetune.py:976] (3/7) Epoch 3, batch 3300, loss[loss=0.2036, simple_loss=0.2441, pruned_loss=0.08159, over 4079.00 frames. ], tot_loss[loss=0.246, simple_loss=0.2969, pruned_loss=0.09753, over 950888.55 frames. ], batch size: 17, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:44:35,949 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14759.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:44:42,747 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14769.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:44:48,716 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=14776.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:44:48,745 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14776.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 01:44:54,873 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1694, 0.7645, 0.9256, 0.9668, 1.2644, 1.2834, 1.1013, 1.0019], device='cuda:3'), covar=tensor([0.0235, 0.0428, 0.0652, 0.0383, 0.0257, 0.0329, 0.0279, 0.0396], device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0113, 0.0135, 0.0116, 0.0104, 0.0098, 0.0088, 0.0108], device='cuda:3'), out_proj_covar=tensor([6.4675e-05, 8.9395e-05, 1.0845e-04, 9.1885e-05, 8.2369e-05, 7.2874e-05, 6.8237e-05, 8.4578e-05], device='cuda:3') 2023-03-26 01:45:09,028 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14801.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:45:16,878 INFO [finetune.py:976] (3/7) Epoch 3, batch 3350, loss[loss=0.2812, simple_loss=0.3181, pruned_loss=0.1222, over 4219.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.2994, pruned_loss=0.09875, over 947746.36 frames. ], batch size: 65, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:45:17,532 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=14807.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:45:29,677 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.053e+02 1.682e+02 2.084e+02 2.593e+02 4.183e+02, threshold=4.169e+02, percent-clipped=0.0 2023-03-26 01:45:39,453 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=14824.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 01:45:52,839 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14843.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:46:01,259 INFO [finetune.py:976] (3/7) Epoch 3, batch 3400, loss[loss=0.3149, simple_loss=0.3532, pruned_loss=0.1383, over 4921.00 frames. ], tot_loss[loss=0.251, simple_loss=0.302, pruned_loss=0.1, over 948400.66 frames. ], batch size: 42, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:46:41,112 INFO [finetune.py:976] (3/7) Epoch 3, batch 3450, loss[loss=0.2122, simple_loss=0.2599, pruned_loss=0.08231, over 4794.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.2996, pruned_loss=0.09861, over 949475.80 frames. ], batch size: 29, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:46:53,149 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.352e+02 1.935e+02 2.237e+02 2.692e+02 3.962e+02, threshold=4.475e+02, percent-clipped=0.0 2023-03-26 01:47:08,445 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.2492, 3.6750, 3.8805, 4.0543, 3.9839, 3.7538, 4.3292, 1.4687], device='cuda:3'), covar=tensor([0.0706, 0.0726, 0.0713, 0.0914, 0.1180, 0.1322, 0.0601, 0.4919], device='cuda:3'), in_proj_covar=tensor([0.0367, 0.0244, 0.0275, 0.0294, 0.0339, 0.0286, 0.0310, 0.0301], 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') 2023-03-26 01:47:27,613 INFO [finetune.py:976] (3/7) Epoch 3, batch 3500, loss[loss=0.2331, simple_loss=0.2862, pruned_loss=0.09001, over 4874.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.2973, pruned_loss=0.09713, over 952280.36 frames. ], batch size: 35, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:48:09,622 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6923, 1.5071, 1.4437, 1.7820, 2.0913, 1.7187, 1.1283, 1.4056], device='cuda:3'), covar=tensor([0.2512, 0.2593, 0.2168, 0.1903, 0.2233, 0.1349, 0.3312, 0.1972], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0208, 0.0195, 0.0182, 0.0233, 0.0172, 0.0212, 0.0186], 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') 2023-03-26 01:48:19,805 INFO [finetune.py:976] (3/7) Epoch 3, batch 3550, loss[loss=0.2276, simple_loss=0.2574, pruned_loss=0.09892, over 4068.00 frames. ], tot_loss[loss=0.2412, simple_loss=0.2929, pruned_loss=0.09478, over 953492.39 frames. ], batch size: 17, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:48:26,975 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.116e+02 1.780e+02 2.199e+02 2.800e+02 5.904e+02, threshold=4.398e+02, percent-clipped=2.0 2023-03-26 01:48:38,303 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6425, 1.5901, 1.2709, 1.4009, 1.8310, 1.8501, 1.6059, 1.3478], device='cuda:3'), covar=tensor([0.0294, 0.0380, 0.0554, 0.0377, 0.0226, 0.0402, 0.0338, 0.0478], device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0113, 0.0135, 0.0116, 0.0104, 0.0098, 0.0089, 0.0108], device='cuda:3'), out_proj_covar=tensor([6.4691e-05, 8.9530e-05, 1.0872e-04, 9.1907e-05, 8.2628e-05, 7.2893e-05, 6.8370e-05, 8.4395e-05], device='cuda:3') 2023-03-26 01:49:01,215 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9973, 1.8054, 1.4907, 1.8784, 1.9356, 1.5826, 2.2477, 1.8834], device='cuda:3'), covar=tensor([0.2004, 0.4254, 0.4782, 0.4306, 0.3399, 0.2436, 0.4400, 0.2857], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0198, 0.0241, 0.0256, 0.0223, 0.0187, 0.0211, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 01:49:03,541 INFO [finetune.py:976] (3/7) Epoch 3, batch 3600, loss[loss=0.2047, simple_loss=0.2766, pruned_loss=0.06645, over 4899.00 frames. ], tot_loss[loss=0.238, simple_loss=0.2901, pruned_loss=0.093, over 955495.77 frames. ], batch size: 36, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:49:12,120 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15069.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:49:23,341 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.1058, 4.4141, 4.6450, 4.8607, 4.7818, 4.4975, 5.1824, 1.6095], device='cuda:3'), covar=tensor([0.0683, 0.0763, 0.0663, 0.0970, 0.1176, 0.1477, 0.0500, 0.5783], device='cuda:3'), in_proj_covar=tensor([0.0366, 0.0243, 0.0275, 0.0292, 0.0339, 0.0286, 0.0309, 0.0301], 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') 2023-03-26 01:49:42,970 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15096.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:49:49,067 INFO [finetune.py:976] (3/7) Epoch 3, batch 3650, loss[loss=0.2815, simple_loss=0.3345, pruned_loss=0.1142, over 4907.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.2924, pruned_loss=0.0941, over 954298.52 frames. ], batch size: 43, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:49:56,316 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.204e+02 1.927e+02 2.238e+02 2.686e+02 4.916e+02, threshold=4.476e+02, percent-clipped=1.0 2023-03-26 01:49:56,389 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=15117.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:50:10,634 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0042, 1.7686, 1.5096, 1.7703, 1.7121, 1.7039, 1.5857, 2.5562], device='cuda:3'), covar=tensor([1.1908, 1.1247, 0.8978, 1.1720, 0.9290, 0.6320, 1.1454, 0.3578], device='cuda:3'), in_proj_covar=tensor([0.0271, 0.0249, 0.0218, 0.0283, 0.0234, 0.0196, 0.0238, 0.0185], 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') 2023-03-26 01:50:28,377 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15143.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:50:41,688 INFO [finetune.py:976] (3/7) Epoch 3, batch 3700, loss[loss=0.2565, simple_loss=0.3086, pruned_loss=0.1022, over 4925.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.2963, pruned_loss=0.09597, over 954861.18 frames. ], batch size: 42, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:51:00,719 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.63 vs. limit=5.0 2023-03-26 01:51:16,178 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=15191.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:51:19,599 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15195.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:51:29,141 INFO [finetune.py:976] (3/7) Epoch 3, batch 3750, loss[loss=0.2511, simple_loss=0.3031, pruned_loss=0.09952, over 4876.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.2985, pruned_loss=0.09737, over 954648.06 frames. ], batch size: 34, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:51:40,515 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.151e+02 1.801e+02 2.153e+02 2.622e+02 6.720e+02, threshold=4.305e+02, percent-clipped=1.0 2023-03-26 01:52:33,793 INFO [finetune.py:976] (3/7) Epoch 3, batch 3800, loss[loss=0.2605, simple_loss=0.3085, pruned_loss=0.1063, over 4821.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3005, pruned_loss=0.09787, over 955161.86 frames. ], batch size: 33, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:52:33,912 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15256.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 01:53:22,428 INFO [finetune.py:976] (3/7) Epoch 3, batch 3850, loss[loss=0.2426, simple_loss=0.2883, pruned_loss=0.09849, over 4901.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.2987, pruned_loss=0.09669, over 955409.69 frames. ], batch size: 35, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:53:39,318 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.276e+02 1.876e+02 2.253e+02 2.579e+02 5.032e+02, threshold=4.505e+02, percent-clipped=1.0 2023-03-26 01:53:49,546 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6667, 1.1218, 1.5089, 1.4583, 1.3880, 1.3640, 1.4007, 1.4116], device='cuda:3'), covar=tensor([0.8835, 1.4188, 1.1181, 1.2814, 1.3720, 0.9918, 1.6094, 1.0232], device='cuda:3'), in_proj_covar=tensor([0.0228, 0.0255, 0.0253, 0.0267, 0.0243, 0.0218, 0.0279, 0.0220], 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') 2023-03-26 01:54:26,485 INFO [finetune.py:976] (3/7) Epoch 3, batch 3900, loss[loss=0.2375, simple_loss=0.2872, pruned_loss=0.09388, over 4819.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.2955, pruned_loss=0.09545, over 955604.40 frames. ], batch size: 41, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:54:39,893 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7452, 1.5116, 1.4891, 1.5428, 1.8393, 1.4866, 2.0712, 1.6826], device='cuda:3'), covar=tensor([0.2172, 0.4190, 0.4570, 0.3908, 0.3240, 0.2246, 0.4154, 0.2836], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0196, 0.0239, 0.0255, 0.0221, 0.0185, 0.0209, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 01:55:10,137 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15396.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:55:16,819 INFO [finetune.py:976] (3/7) Epoch 3, batch 3950, loss[loss=0.2202, simple_loss=0.2713, pruned_loss=0.08449, over 4901.00 frames. ], tot_loss[loss=0.2374, simple_loss=0.2899, pruned_loss=0.09242, over 956812.03 frames. ], batch size: 43, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:55:25,261 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.128e+02 1.678e+02 2.164e+02 2.472e+02 4.231e+02, threshold=4.328e+02, percent-clipped=0.0 2023-03-26 01:55:25,414 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7593, 1.6727, 1.6168, 1.9796, 2.4251, 1.8951, 1.3172, 1.5171], device='cuda:3'), covar=tensor([0.2508, 0.2488, 0.2060, 0.1882, 0.1873, 0.1315, 0.3118, 0.2060], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0207, 0.0195, 0.0181, 0.0231, 0.0171, 0.0212, 0.0185], 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') 2023-03-26 01:55:28,933 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7969, 3.9188, 3.7258, 1.9242, 4.0738, 3.0951, 0.9876, 2.9093], device='cuda:3'), covar=tensor([0.2237, 0.2169, 0.1518, 0.3436, 0.0893, 0.0908, 0.4396, 0.1437], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0168, 0.0165, 0.0129, 0.0155, 0.0121, 0.0146, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 01:55:52,198 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=15444.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 01:55:53,354 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-26 01:55:59,831 INFO [finetune.py:976] (3/7) Epoch 3, batch 4000, loss[loss=0.2174, simple_loss=0.2723, pruned_loss=0.08123, over 4774.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.2878, pruned_loss=0.0917, over 956482.29 frames. ], batch size: 26, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:56:24,655 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-26 01:57:05,461 INFO [finetune.py:976] (3/7) Epoch 3, batch 4050, loss[loss=0.2131, simple_loss=0.2776, pruned_loss=0.07427, over 4925.00 frames. ], tot_loss[loss=0.2394, simple_loss=0.2921, pruned_loss=0.09339, over 956670.98 frames. ], batch size: 33, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:57:20,271 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.142e+02 1.798e+02 2.110e+02 2.647e+02 5.396e+02, threshold=4.219e+02, percent-clipped=2.0 2023-03-26 01:57:37,066 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1427, 1.8648, 2.6187, 1.5912, 2.2959, 2.4199, 1.8337, 2.6045], device='cuda:3'), covar=tensor([0.1739, 0.2332, 0.1736, 0.2477, 0.1045, 0.1784, 0.2725, 0.1066], device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0205, 0.0204, 0.0196, 0.0181, 0.0225, 0.0215, 0.0203], 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') 2023-03-26 01:57:38,928 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1728, 1.9660, 1.5957, 2.0977, 1.9083, 1.7864, 1.7942, 3.0488], device='cuda:3'), covar=tensor([1.1163, 1.1685, 0.9060, 1.2000, 0.9801, 0.6262, 1.1618, 0.3145], device='cuda:3'), in_proj_covar=tensor([0.0273, 0.0249, 0.0218, 0.0284, 0.0235, 0.0196, 0.0239, 0.0185], 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') 2023-03-26 01:57:40,732 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.1354, 2.6560, 2.3539, 1.3903, 2.6336, 2.1914, 1.9945, 2.3531], device='cuda:3'), covar=tensor([0.1029, 0.1037, 0.2025, 0.2707, 0.2104, 0.2684, 0.2499, 0.1467], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0199, 0.0205, 0.0190, 0.0218, 0.0212, 0.0218, 0.0202], 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') 2023-03-26 01:57:47,154 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7059, 1.6162, 1.5070, 1.7525, 1.3657, 3.6463, 1.4385, 2.1690], device='cuda:3'), covar=tensor([0.3571, 0.2406, 0.2037, 0.2140, 0.1750, 0.0188, 0.2579, 0.1270], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0111, 0.0116, 0.0120, 0.0116, 0.0097, 0.0102, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 01:57:58,328 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15551.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 01:58:06,352 INFO [finetune.py:976] (3/7) Epoch 3, batch 4100, loss[loss=0.2388, simple_loss=0.2977, pruned_loss=0.08998, over 4828.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.2948, pruned_loss=0.09425, over 955288.75 frames. ], batch size: 39, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:58:48,424 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7734, 1.6434, 1.3389, 1.4086, 1.8716, 1.9565, 1.6319, 1.4687], device='cuda:3'), covar=tensor([0.0249, 0.0346, 0.0522, 0.0373, 0.0232, 0.0311, 0.0392, 0.0397], device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0113, 0.0136, 0.0116, 0.0104, 0.0098, 0.0089, 0.0108], device='cuda:3'), out_proj_covar=tensor([6.5023e-05, 8.9353e-05, 1.0920e-04, 9.1637e-05, 8.2690e-05, 7.2781e-05, 6.8859e-05, 8.4468e-05], device='cuda:3') 2023-03-26 01:59:02,792 INFO [finetune.py:976] (3/7) Epoch 3, batch 4150, loss[loss=0.2934, simple_loss=0.3283, pruned_loss=0.1293, over 4822.00 frames. ], tot_loss[loss=0.2412, simple_loss=0.2946, pruned_loss=0.09386, over 954991.87 frames. ], batch size: 33, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 01:59:10,586 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.238e+02 1.791e+02 2.157e+02 2.467e+02 4.537e+02, threshold=4.313e+02, percent-clipped=1.0 2023-03-26 01:59:50,065 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-03-26 01:59:51,465 INFO [finetune.py:976] (3/7) Epoch 3, batch 4200, loss[loss=0.2369, simple_loss=0.2988, pruned_loss=0.08751, over 4789.00 frames. ], tot_loss[loss=0.2411, simple_loss=0.2951, pruned_loss=0.09348, over 954773.72 frames. ], batch size: 51, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 02:00:21,934 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7880, 1.9227, 1.8748, 1.3304, 2.1024, 2.0728, 1.9558, 1.6629], device='cuda:3'), covar=tensor([0.0745, 0.0616, 0.0763, 0.1012, 0.0495, 0.0697, 0.0717, 0.1124], device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0134, 0.0146, 0.0132, 0.0112, 0.0145, 0.0149, 0.0165], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 02:00:32,636 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 2023-03-26 02:00:45,023 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5125, 1.4251, 2.1156, 3.1959, 2.1744, 2.3202, 0.8171, 2.3514], device='cuda:3'), covar=tensor([0.1825, 0.1579, 0.1305, 0.0522, 0.0859, 0.1643, 0.1982, 0.0707], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0118, 0.0137, 0.0164, 0.0103, 0.0143, 0.0128, 0.0106], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 02:00:53,255 INFO [finetune.py:976] (3/7) Epoch 3, batch 4250, loss[loss=0.2562, simple_loss=0.3112, pruned_loss=0.1006, over 4775.00 frames. ], tot_loss[loss=0.2398, simple_loss=0.2933, pruned_loss=0.0932, over 954698.03 frames. ], batch size: 29, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 02:01:00,003 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.228e+02 1.731e+02 2.073e+02 2.469e+02 5.386e+02, threshold=4.147e+02, percent-clipped=2.0 2023-03-26 02:01:24,588 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-26 02:01:43,529 INFO [finetune.py:976] (3/7) Epoch 3, batch 4300, loss[loss=0.2617, simple_loss=0.2941, pruned_loss=0.1147, over 4284.00 frames. ], tot_loss[loss=0.2394, simple_loss=0.2918, pruned_loss=0.09346, over 954092.75 frames. ], batch size: 65, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 02:02:02,652 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-26 02:02:20,287 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9788, 1.8587, 1.5493, 2.0273, 2.0305, 1.6964, 2.4023, 1.9704], device='cuda:3'), covar=tensor([0.2001, 0.3968, 0.4170, 0.3764, 0.2758, 0.2076, 0.3740, 0.2600], device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0195, 0.0238, 0.0254, 0.0221, 0.0185, 0.0209, 0.0189], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 02:02:43,903 INFO [finetune.py:976] (3/7) Epoch 3, batch 4350, loss[loss=0.3235, simple_loss=0.3519, pruned_loss=0.1475, over 4834.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.288, pruned_loss=0.09208, over 953761.13 frames. ], batch size: 33, lr: 3.98e-03, grad_scale: 32.0 2023-03-26 02:03:01,713 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.218e+02 1.712e+02 2.007e+02 2.460e+02 4.679e+02, threshold=4.015e+02, percent-clipped=1.0 2023-03-26 02:03:24,164 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-26 02:03:33,439 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15851.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:03:36,425 INFO [finetune.py:976] (3/7) Epoch 3, batch 4400, loss[loss=0.301, simple_loss=0.3458, pruned_loss=0.1281, over 4873.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.2902, pruned_loss=0.09348, over 954735.42 frames. ], batch size: 34, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:04:05,493 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15883.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:04:17,888 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5324, 2.3357, 1.9621, 1.0709, 2.1212, 1.9566, 1.6592, 2.1992], device='cuda:3'), covar=tensor([0.0854, 0.0932, 0.1662, 0.2523, 0.1709, 0.2392, 0.2559, 0.1096], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0198, 0.0204, 0.0190, 0.0218, 0.0210, 0.0217, 0.0201], 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') 2023-03-26 02:04:20,878 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=15899.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:04:30,835 INFO [finetune.py:976] (3/7) Epoch 3, batch 4450, loss[loss=0.2717, simple_loss=0.3221, pruned_loss=0.1107, over 4707.00 frames. ], tot_loss[loss=0.242, simple_loss=0.2942, pruned_loss=0.09493, over 952085.14 frames. ], batch size: 59, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:04:48,084 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.107e+02 1.872e+02 2.224e+02 2.526e+02 5.583e+02, threshold=4.448e+02, percent-clipped=1.0 2023-03-26 02:05:16,005 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15944.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:05:23,247 INFO [finetune.py:976] (3/7) Epoch 3, batch 4500, loss[loss=0.2418, simple_loss=0.3072, pruned_loss=0.08819, over 4791.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.2971, pruned_loss=0.09592, over 954794.11 frames. ], batch size: 51, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:06:17,316 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16001.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:06:24,310 INFO [finetune.py:976] (3/7) Epoch 3, batch 4550, loss[loss=0.2726, simple_loss=0.3242, pruned_loss=0.1105, over 4806.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.2992, pruned_loss=0.09676, over 954321.28 frames. ], batch size: 41, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:06:36,870 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.237e+02 1.760e+02 2.085e+02 2.487e+02 3.865e+02, threshold=4.170e+02, percent-clipped=0.0 2023-03-26 02:07:05,903 INFO [finetune.py:976] (3/7) Epoch 3, batch 4600, loss[loss=0.194, simple_loss=0.259, pruned_loss=0.06456, over 4864.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.2971, pruned_loss=0.09517, over 954918.21 frames. ], batch size: 31, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:07:07,949 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-26 02:07:11,905 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16062.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:07:35,229 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16085.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:07:59,849 INFO [finetune.py:976] (3/7) Epoch 3, batch 4650, loss[loss=0.202, simple_loss=0.2735, pruned_loss=0.06527, over 4909.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.2943, pruned_loss=0.09425, over 955048.09 frames. ], batch size: 43, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:08:03,850 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8940, 1.6132, 1.3554, 1.5668, 1.5906, 1.5131, 1.4971, 2.4451], device='cuda:3'), covar=tensor([1.1186, 1.1454, 0.8740, 1.0738, 0.8986, 0.6013, 1.0529, 0.3256], device='cuda:3'), in_proj_covar=tensor([0.0272, 0.0248, 0.0217, 0.0282, 0.0234, 0.0195, 0.0237, 0.0185], 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') 2023-03-26 02:08:07,261 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.007e+02 1.817e+02 2.201e+02 2.598e+02 3.850e+02, threshold=4.403e+02, percent-clipped=0.0 2023-03-26 02:08:30,572 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1795, 2.2640, 2.2068, 1.4841, 2.4455, 2.4397, 2.4204, 1.9630], device='cuda:3'), covar=tensor([0.0717, 0.0624, 0.0774, 0.1096, 0.0508, 0.0707, 0.0643, 0.1121], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0134, 0.0146, 0.0131, 0.0111, 0.0144, 0.0148, 0.0164], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 02:08:39,877 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16146.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:08:42,101 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16149.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:08:52,384 INFO [finetune.py:976] (3/7) Epoch 3, batch 4700, loss[loss=0.2109, simple_loss=0.2565, pruned_loss=0.08259, over 4819.00 frames. ], tot_loss[loss=0.2387, simple_loss=0.2912, pruned_loss=0.09316, over 956364.34 frames. ], batch size: 51, lr: 3.97e-03, grad_scale: 64.0 2023-03-26 02:09:26,186 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.39 vs. limit=5.0 2023-03-26 02:09:30,569 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.66 vs. limit=5.0 2023-03-26 02:09:41,914 INFO [finetune.py:976] (3/7) Epoch 3, batch 4750, loss[loss=0.3401, simple_loss=0.3528, pruned_loss=0.1637, over 3938.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.2882, pruned_loss=0.0918, over 956541.40 frames. ], batch size: 65, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:09:44,976 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16210.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:09:49,753 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.299e+02 1.759e+02 2.065e+02 2.416e+02 4.123e+02, threshold=4.129e+02, percent-clipped=0.0 2023-03-26 02:10:03,619 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16239.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:10:05,419 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3782, 1.1848, 1.4807, 2.4048, 1.6639, 2.0998, 0.8814, 1.9556], device='cuda:3'), covar=tensor([0.2048, 0.1813, 0.1435, 0.0868, 0.1094, 0.1262, 0.1833, 0.0922], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0120, 0.0139, 0.0166, 0.0104, 0.0144, 0.0130, 0.0108], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 02:10:16,138 INFO [finetune.py:976] (3/7) Epoch 3, batch 4800, loss[loss=0.2759, simple_loss=0.3404, pruned_loss=0.1056, over 4821.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.2919, pruned_loss=0.09324, over 956971.15 frames. ], batch size: 51, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:11:10,320 INFO [finetune.py:976] (3/7) Epoch 3, batch 4850, loss[loss=0.2855, simple_loss=0.3215, pruned_loss=0.1248, over 4826.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.2953, pruned_loss=0.0945, over 954730.43 frames. ], batch size: 33, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:11:18,710 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.278e+02 1.868e+02 2.289e+02 2.628e+02 4.977e+02, threshold=4.577e+02, percent-clipped=4.0 2023-03-26 02:11:53,000 INFO [finetune.py:976] (3/7) Epoch 3, batch 4900, loss[loss=0.3472, simple_loss=0.375, pruned_loss=0.1597, over 4147.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.2986, pruned_loss=0.09606, over 956008.82 frames. ], batch size: 65, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:11:53,737 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16357.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:11:57,248 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.24 vs. limit=5.0 2023-03-26 02:12:02,745 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-26 02:12:04,212 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16371.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:12:35,109 INFO [finetune.py:976] (3/7) Epoch 3, batch 4950, loss[loss=0.2426, simple_loss=0.296, pruned_loss=0.09458, over 4892.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3002, pruned_loss=0.09659, over 957504.27 frames. ], batch size: 43, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:12:43,740 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.165e+02 1.793e+02 2.170e+02 2.564e+02 4.726e+02, threshold=4.340e+02, percent-clipped=1.0 2023-03-26 02:12:53,404 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16432.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:12:59,265 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16441.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:13:11,535 INFO [finetune.py:976] (3/7) Epoch 3, batch 5000, loss[loss=0.2438, simple_loss=0.2893, pruned_loss=0.09918, over 4771.00 frames. ], tot_loss[loss=0.244, simple_loss=0.2972, pruned_loss=0.09544, over 956506.14 frames. ], batch size: 28, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:13:26,564 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0946, 2.1037, 2.1588, 1.4806, 2.3741, 2.4757, 2.2286, 1.9694], device='cuda:3'), covar=tensor([0.0768, 0.0721, 0.0778, 0.1114, 0.0485, 0.0685, 0.0832, 0.0994], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0133, 0.0144, 0.0129, 0.0110, 0.0142, 0.0148, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 02:13:45,830 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.98 vs. limit=5.0 2023-03-26 02:14:08,361 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16505.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:14:08,895 INFO [finetune.py:976] (3/7) Epoch 3, batch 5050, loss[loss=0.2468, simple_loss=0.3212, pruned_loss=0.08617, over 4824.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.2959, pruned_loss=0.09575, over 954816.76 frames. ], batch size: 41, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:14:27,727 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.054e+02 1.680e+02 2.024e+02 2.446e+02 4.498e+02, threshold=4.048e+02, percent-clipped=1.0 2023-03-26 02:14:49,228 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16539.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:14:55,111 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6712, 1.5242, 1.5776, 1.6021, 0.8832, 3.2471, 1.2075, 1.6606], device='cuda:3'), covar=tensor([0.3377, 0.2442, 0.2070, 0.2274, 0.2237, 0.0194, 0.2918, 0.1495], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0112, 0.0116, 0.0119, 0.0116, 0.0097, 0.0101, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 02:15:09,826 INFO [finetune.py:976] (3/7) Epoch 3, batch 5100, loss[loss=0.2507, simple_loss=0.2884, pruned_loss=0.1065, over 4869.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.2921, pruned_loss=0.09415, over 956099.22 frames. ], batch size: 34, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:15:36,571 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=16587.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:15:44,378 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0670, 1.4313, 0.7769, 1.9368, 2.3649, 1.4913, 1.6559, 1.8465], device='cuda:3'), covar=tensor([0.1443, 0.2105, 0.2534, 0.1150, 0.1835, 0.2013, 0.1435, 0.2028], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0098, 0.0117, 0.0093, 0.0124, 0.0097, 0.0100, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 02:15:52,949 INFO [finetune.py:976] (3/7) Epoch 3, batch 5150, loss[loss=0.2907, simple_loss=0.344, pruned_loss=0.1187, over 4822.00 frames. ], tot_loss[loss=0.239, simple_loss=0.2911, pruned_loss=0.09348, over 952593.64 frames. ], batch size: 39, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:16:05,234 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1102, 1.6112, 2.2685, 4.0044, 2.7646, 2.7657, 0.7824, 3.1593], device='cuda:3'), covar=tensor([0.1839, 0.1782, 0.1690, 0.0613, 0.0878, 0.1563, 0.2372, 0.0632], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0120, 0.0139, 0.0166, 0.0104, 0.0144, 0.0130, 0.0107], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 02:16:12,054 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.182e+02 1.762e+02 2.113e+02 2.590e+02 4.768e+02, threshold=4.226e+02, percent-clipped=2.0 2023-03-26 02:16:47,967 INFO [finetune.py:976] (3/7) Epoch 3, batch 5200, loss[loss=0.268, simple_loss=0.3262, pruned_loss=0.1049, over 4805.00 frames. ], tot_loss[loss=0.2406, simple_loss=0.2935, pruned_loss=0.09389, over 953590.07 frames. ], batch size: 41, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:16:48,668 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16657.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:16:59,969 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6099, 1.4490, 1.2921, 1.2045, 1.7570, 1.7878, 1.5698, 1.2783], device='cuda:3'), covar=tensor([0.0290, 0.0300, 0.0634, 0.0383, 0.0217, 0.0404, 0.0277, 0.0408], device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0114, 0.0137, 0.0118, 0.0105, 0.0100, 0.0091, 0.0109], device='cuda:3'), out_proj_covar=tensor([6.5845e-05, 9.0485e-05, 1.1060e-04, 9.3333e-05, 8.3089e-05, 7.4385e-05, 6.9957e-05, 8.5062e-05], device='cuda:3') 2023-03-26 02:17:19,011 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-26 02:17:40,755 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=16705.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:17:41,307 INFO [finetune.py:976] (3/7) Epoch 3, batch 5250, loss[loss=0.2757, simple_loss=0.3167, pruned_loss=0.1174, over 4759.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.2957, pruned_loss=0.09406, over 953091.99 frames. ], batch size: 28, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:17:41,997 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7722, 1.2099, 0.7989, 1.6571, 2.1155, 1.3801, 1.5150, 1.7069], device='cuda:3'), covar=tensor([0.1554, 0.2210, 0.2365, 0.1317, 0.1938, 0.2131, 0.1444, 0.2059], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0099, 0.0118, 0.0094, 0.0125, 0.0098, 0.0101, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 02:17:53,304 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.211e+02 1.833e+02 2.103e+02 2.647e+02 4.683e+02, threshold=4.205e+02, percent-clipped=1.0 2023-03-26 02:17:54,294 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.93 vs. limit=5.0 2023-03-26 02:18:00,562 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16727.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:18:12,249 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16741.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:18:16,031 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 2023-03-26 02:18:20,130 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.8636, 4.2229, 4.4265, 4.6493, 4.5631, 4.2616, 4.9539, 1.5514], device='cuda:3'), covar=tensor([0.0660, 0.0702, 0.0654, 0.0773, 0.1181, 0.1306, 0.0506, 0.5336], device='cuda:3'), in_proj_covar=tensor([0.0366, 0.0246, 0.0279, 0.0296, 0.0344, 0.0290, 0.0312, 0.0303], 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') 2023-03-26 02:18:21,890 INFO [finetune.py:976] (3/7) Epoch 3, batch 5300, loss[loss=0.2129, simple_loss=0.2714, pruned_loss=0.07718, over 4754.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.2972, pruned_loss=0.09446, over 953805.70 frames. ], batch size: 26, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:18:49,199 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=16789.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:19:07,696 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16805.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:19:08,225 INFO [finetune.py:976] (3/7) Epoch 3, batch 5350, loss[loss=0.1886, simple_loss=0.253, pruned_loss=0.06212, over 4789.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.2967, pruned_loss=0.0942, over 952559.44 frames. ], batch size: 29, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:19:21,977 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.184e+02 1.880e+02 2.227e+02 2.511e+02 3.677e+02, threshold=4.454e+02, percent-clipped=0.0 2023-03-26 02:19:40,172 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2023-03-26 02:19:46,294 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=16853.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:19:48,087 INFO [finetune.py:976] (3/7) Epoch 3, batch 5400, loss[loss=0.2966, simple_loss=0.3234, pruned_loss=0.1349, over 4166.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.2945, pruned_loss=0.0934, over 952772.21 frames. ], batch size: 65, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:19:51,859 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5313, 1.4138, 2.0344, 3.0801, 2.2336, 2.1846, 1.1532, 2.4388], device='cuda:3'), covar=tensor([0.1847, 0.1661, 0.1311, 0.0704, 0.0810, 0.1529, 0.1868, 0.0679], device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0119, 0.0138, 0.0166, 0.0104, 0.0144, 0.0130, 0.0106], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 02:19:55,560 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16868.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 02:20:04,496 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7302, 1.3597, 1.0534, 0.2491, 1.2858, 1.4904, 1.3697, 1.4669], device='cuda:3'), covar=tensor([0.0908, 0.0817, 0.1237, 0.1983, 0.1287, 0.2251, 0.2137, 0.0816], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0200, 0.0205, 0.0192, 0.0217, 0.0212, 0.0219, 0.0202], 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') 2023-03-26 02:20:31,377 INFO [finetune.py:976] (3/7) Epoch 3, batch 5450, loss[loss=0.2333, simple_loss=0.2846, pruned_loss=0.09101, over 4847.00 frames. ], tot_loss[loss=0.2381, simple_loss=0.2916, pruned_loss=0.09236, over 954455.67 frames. ], batch size: 44, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:20:31,476 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16906.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:20:38,654 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.130e+02 1.665e+02 2.000e+02 2.450e+02 4.433e+02, threshold=4.000e+02, percent-clipped=0.0 2023-03-26 02:20:46,023 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16929.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 02:21:14,237 INFO [finetune.py:976] (3/7) Epoch 3, batch 5500, loss[loss=0.21, simple_loss=0.2649, pruned_loss=0.0776, over 4798.00 frames. ], tot_loss[loss=0.2337, simple_loss=0.2871, pruned_loss=0.09013, over 956951.13 frames. ], batch size: 25, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:21:26,193 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16967.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 02:21:38,392 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.22 vs. limit=5.0 2023-03-26 02:22:07,777 INFO [finetune.py:976] (3/7) Epoch 3, batch 5550, loss[loss=0.3085, simple_loss=0.3513, pruned_loss=0.1329, over 4907.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.2908, pruned_loss=0.09229, over 958484.98 frames. ], batch size: 43, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:22:15,686 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.144e+02 1.712e+02 2.015e+02 2.380e+02 4.122e+02, threshold=4.030e+02, percent-clipped=1.0 2023-03-26 02:22:21,769 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17027.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:22:24,240 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9940, 1.3218, 0.8685, 1.8020, 2.2414, 1.3476, 1.6597, 1.7998], device='cuda:3'), covar=tensor([0.1562, 0.2162, 0.2512, 0.1246, 0.1907, 0.2128, 0.1482, 0.2044], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0099, 0.0118, 0.0094, 0.0124, 0.0098, 0.0101, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 02:22:53,362 INFO [finetune.py:976] (3/7) Epoch 3, batch 5600, loss[loss=0.2403, simple_loss=0.3011, pruned_loss=0.08973, over 4769.00 frames. ], tot_loss[loss=0.2406, simple_loss=0.2947, pruned_loss=0.09327, over 954507.98 frames. ], batch size: 54, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:23:10,682 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=17075.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:23:38,817 INFO [finetune.py:976] (3/7) Epoch 3, batch 5650, loss[loss=0.2508, simple_loss=0.318, pruned_loss=0.09185, over 4823.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.2964, pruned_loss=0.0932, over 954746.59 frames. ], batch size: 40, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:23:45,814 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.238e+02 1.755e+02 2.152e+02 2.681e+02 4.789e+02, threshold=4.305e+02, percent-clipped=1.0 2023-03-26 02:23:58,126 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8735, 3.7134, 3.6167, 1.8602, 3.9287, 2.9691, 1.2262, 2.6546], device='cuda:3'), covar=tensor([0.2008, 0.1946, 0.1600, 0.3232, 0.0889, 0.0946, 0.4159, 0.1457], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0168, 0.0162, 0.0128, 0.0153, 0.0120, 0.0145, 0.0120], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 02:24:08,962 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7702, 2.4077, 2.8923, 1.6332, 2.9825, 2.9576, 2.5101, 2.7006], device='cuda:3'), covar=tensor([0.1047, 0.0768, 0.0433, 0.0687, 0.0402, 0.0578, 0.0416, 0.0447], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0157, 0.0117, 0.0136, 0.0131, 0.0120, 0.0145, 0.0144], device='cuda:3'), out_proj_covar=tensor([9.7460e-05, 1.1656e-04, 8.5047e-05, 9.9616e-05, 9.4778e-05, 8.9260e-05, 1.0805e-04, 1.0663e-04], device='cuda:3') 2023-03-26 02:24:15,377 INFO [finetune.py:976] (3/7) Epoch 3, batch 5700, loss[loss=0.1847, simple_loss=0.2301, pruned_loss=0.06966, over 4276.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.2916, pruned_loss=0.09285, over 933932.62 frames. ], batch size: 18, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:24:56,869 INFO [finetune.py:976] (3/7) Epoch 4, batch 0, loss[loss=0.3116, simple_loss=0.3484, pruned_loss=0.1374, over 4828.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3484, pruned_loss=0.1374, over 4828.00 frames. ], batch size: 49, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:24:56,869 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 02:25:05,085 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.2260, 1.4845, 1.5498, 0.7973, 1.3887, 1.6770, 1.7159, 1.4871], device='cuda:3'), covar=tensor([0.1011, 0.0604, 0.0490, 0.0484, 0.0507, 0.0454, 0.0367, 0.0694], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0157, 0.0117, 0.0136, 0.0132, 0.0121, 0.0146, 0.0145], device='cuda:3'), out_proj_covar=tensor([9.8181e-05, 1.1719e-04, 8.5409e-05, 1.0017e-04, 9.5355e-05, 9.0035e-05, 1.0877e-04, 1.0736e-04], device='cuda:3') 2023-03-26 02:25:18,215 INFO [finetune.py:1010] (3/7) Epoch 4, validation: loss=0.1768, simple_loss=0.2473, pruned_loss=0.0532, over 2265189.00 frames. 2023-03-26 02:25:18,216 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 02:25:22,712 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17189.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:25:30,440 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-03-26 02:25:55,799 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.188e+02 1.713e+02 2.128e+02 2.708e+02 4.853e+02, threshold=4.257e+02, percent-clipped=3.0 2023-03-26 02:26:00,041 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17224.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 02:26:03,133 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1922, 2.6306, 1.9264, 1.6608, 2.6317, 2.6235, 2.5869, 2.2173], device='cuda:3'), covar=tensor([0.0754, 0.0535, 0.1036, 0.1151, 0.0789, 0.0789, 0.0769, 0.1028], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0132, 0.0143, 0.0128, 0.0110, 0.0141, 0.0146, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 02:26:05,894 INFO [finetune.py:976] (3/7) Epoch 4, batch 50, loss[loss=0.2567, simple_loss=0.3106, pruned_loss=0.1015, over 4835.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.2994, pruned_loss=0.09887, over 215314.64 frames. ], batch size: 44, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:26:29,254 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17250.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:26:32,877 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9228, 1.8010, 1.6811, 1.9754, 2.5547, 2.1188, 1.4659, 1.5504], device='cuda:3'), covar=tensor([0.2717, 0.2597, 0.2296, 0.2167, 0.2037, 0.1246, 0.3125, 0.2226], device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0209, 0.0198, 0.0183, 0.0233, 0.0173, 0.0214, 0.0186], 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') 2023-03-26 02:26:36,459 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17262.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 02:26:55,371 INFO [finetune.py:976] (3/7) Epoch 4, batch 100, loss[loss=0.2062, simple_loss=0.2568, pruned_loss=0.07779, over 4765.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.2878, pruned_loss=0.09223, over 378656.71 frames. ], batch size: 26, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:27:20,977 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6742, 1.7962, 2.0060, 1.8925, 1.9810, 4.4029, 1.6186, 2.0775], device='cuda:3'), covar=tensor([0.0958, 0.1658, 0.1154, 0.1076, 0.1503, 0.0183, 0.1398, 0.1579], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0081, 0.0077, 0.0079, 0.0092, 0.0083, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 02:27:26,862 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.228e+02 1.697e+02 1.982e+02 2.273e+02 3.827e+02, threshold=3.964e+02, percent-clipped=0.0 2023-03-26 02:27:34,328 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1188, 1.2293, 0.8885, 2.0427, 2.4287, 1.7507, 1.5192, 1.9610], device='cuda:3'), covar=tensor([0.1586, 0.2237, 0.2504, 0.1220, 0.1998, 0.2026, 0.1576, 0.2067], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0099, 0.0118, 0.0095, 0.0125, 0.0098, 0.0101, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 02:27:35,552 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6549, 2.3405, 1.9654, 1.0579, 2.1957, 2.0283, 1.7847, 2.0613], device='cuda:3'), covar=tensor([0.0892, 0.0978, 0.1751, 0.2505, 0.1653, 0.2873, 0.2400, 0.1165], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0200, 0.0205, 0.0191, 0.0217, 0.0210, 0.0219, 0.0202], 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') 2023-03-26 02:27:36,644 INFO [finetune.py:976] (3/7) Epoch 4, batch 150, loss[loss=0.191, simple_loss=0.2551, pruned_loss=0.06343, over 4812.00 frames. ], tot_loss[loss=0.232, simple_loss=0.2836, pruned_loss=0.09024, over 508135.64 frames. ], batch size: 25, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:28:01,910 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17362.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:28:25,904 INFO [finetune.py:976] (3/7) Epoch 4, batch 200, loss[loss=0.2385, simple_loss=0.2933, pruned_loss=0.09181, over 4840.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.2809, pruned_loss=0.0886, over 607979.66 frames. ], batch size: 49, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:28:26,029 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17383.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 02:28:55,774 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.314e+02 1.771e+02 2.098e+02 2.514e+02 4.657e+02, threshold=4.195e+02, percent-clipped=1.0 2023-03-26 02:29:01,229 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17423.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:29:09,066 INFO [finetune.py:976] (3/7) Epoch 4, batch 250, loss[loss=0.1817, simple_loss=0.2373, pruned_loss=0.06305, over 4800.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.2868, pruned_loss=0.09047, over 686461.82 frames. ], batch size: 25, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:29:17,846 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17444.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 02:29:26,232 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6966, 1.4328, 2.1726, 3.3355, 2.3750, 2.2976, 0.8970, 2.5665], device='cuda:3'), covar=tensor([0.1812, 0.1618, 0.1269, 0.0571, 0.0788, 0.1339, 0.2030, 0.0620], device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0118, 0.0137, 0.0165, 0.0103, 0.0142, 0.0129, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 02:29:32,353 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8850, 2.0142, 1.8425, 1.2804, 2.1584, 2.1216, 2.1016, 1.7309], device='cuda:3'), covar=tensor([0.0787, 0.0703, 0.0913, 0.1143, 0.0522, 0.0782, 0.0823, 0.1184], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0133, 0.0144, 0.0129, 0.0111, 0.0142, 0.0147, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 02:29:49,002 INFO [finetune.py:976] (3/7) Epoch 4, batch 300, loss[loss=0.2363, simple_loss=0.2858, pruned_loss=0.09345, over 4866.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.2942, pruned_loss=0.09314, over 749250.54 frames. ], batch size: 31, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:30:35,038 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.302e+02 1.964e+02 2.271e+02 2.699e+02 6.272e+02, threshold=4.542e+02, percent-clipped=2.0 2023-03-26 02:30:39,309 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17524.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 02:30:44,607 INFO [finetune.py:976] (3/7) Epoch 4, batch 350, loss[loss=0.2674, simple_loss=0.3055, pruned_loss=0.1147, over 4716.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.2974, pruned_loss=0.09459, over 795734.29 frames. ], batch size: 54, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:30:53,138 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17545.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:31:10,224 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17562.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 02:31:16,312 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=17572.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 02:31:23,351 INFO [finetune.py:976] (3/7) Epoch 4, batch 400, loss[loss=0.2005, simple_loss=0.2559, pruned_loss=0.07254, over 4818.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.2968, pruned_loss=0.09427, over 830949.61 frames. ], batch size: 25, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:31:46,314 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=17610.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:31:51,079 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.188e+02 1.790e+02 1.987e+02 2.567e+02 5.687e+02, threshold=3.975e+02, percent-clipped=1.0 2023-03-26 02:32:10,368 INFO [finetune.py:976] (3/7) Epoch 4, batch 450, loss[loss=0.2255, simple_loss=0.2778, pruned_loss=0.08662, over 4885.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.2944, pruned_loss=0.09268, over 859525.12 frames. ], batch size: 35, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:32:29,120 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-26 02:33:00,230 INFO [finetune.py:976] (3/7) Epoch 4, batch 500, loss[loss=0.245, simple_loss=0.2964, pruned_loss=0.09678, over 4895.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.2911, pruned_loss=0.09214, over 879697.55 frames. ], batch size: 35, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:33:11,519 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17700.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:33:24,343 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.264e+02 1.808e+02 2.091e+02 2.485e+02 4.480e+02, threshold=4.181e+02, percent-clipped=1.0 2023-03-26 02:33:24,433 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17718.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:33:31,861 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17730.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:33:34,059 INFO [finetune.py:976] (3/7) Epoch 4, batch 550, loss[loss=0.2367, simple_loss=0.2863, pruned_loss=0.0935, over 4827.00 frames. ], tot_loss[loss=0.2345, simple_loss=0.2876, pruned_loss=0.09071, over 895963.29 frames. ], batch size: 33, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:33:37,769 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17739.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 02:33:44,319 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9508, 1.7874, 2.3620, 1.5137, 1.9907, 2.2307, 1.7341, 2.5276], device='cuda:3'), covar=tensor([0.1712, 0.2151, 0.1495, 0.2376, 0.1208, 0.1701, 0.2596, 0.0982], device='cuda:3'), in_proj_covar=tensor([0.0207, 0.0208, 0.0205, 0.0198, 0.0184, 0.0227, 0.0217, 0.0205], 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') 2023-03-26 02:34:03,939 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17761.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:34:17,629 INFO [finetune.py:976] (3/7) Epoch 4, batch 600, loss[loss=0.2574, simple_loss=0.3229, pruned_loss=0.09588, over 4811.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.2887, pruned_loss=0.09058, over 908185.29 frames. ], batch size: 38, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:34:22,577 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17791.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:34:39,732 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0465, 1.3344, 0.9173, 2.0416, 2.3464, 1.5562, 1.7935, 1.9739], device='cuda:3'), covar=tensor([0.1484, 0.2072, 0.2175, 0.1063, 0.1851, 0.1956, 0.1263, 0.1847], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0099, 0.0117, 0.0094, 0.0125, 0.0098, 0.0101, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 02:34:41,400 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.036e+02 1.729e+02 2.101e+02 2.480e+02 7.519e+02, threshold=4.202e+02, percent-clipped=3.0 2023-03-26 02:34:46,960 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5129, 1.3735, 1.4495, 1.3692, 0.8652, 2.1865, 0.7876, 1.2928], device='cuda:3'), covar=tensor([0.3003, 0.2154, 0.1787, 0.2164, 0.1905, 0.0344, 0.2409, 0.1174], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0112, 0.0116, 0.0120, 0.0116, 0.0097, 0.0101, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 02:34:50,414 INFO [finetune.py:976] (3/7) Epoch 4, batch 650, loss[loss=0.2348, simple_loss=0.3019, pruned_loss=0.08383, over 4829.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.2922, pruned_loss=0.09244, over 917578.13 frames. ], batch size: 47, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:34:59,958 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17845.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:35:18,617 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-26 02:35:31,914 INFO [finetune.py:976] (3/7) Epoch 4, batch 700, loss[loss=0.1896, simple_loss=0.2493, pruned_loss=0.06491, over 3963.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.2948, pruned_loss=0.09344, over 923899.51 frames. ], batch size: 17, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:35:46,369 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=17893.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:36:18,031 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.240e+02 1.804e+02 2.027e+02 2.461e+02 4.855e+02, threshold=4.055e+02, percent-clipped=2.0 2023-03-26 02:36:34,755 INFO [finetune.py:976] (3/7) Epoch 4, batch 750, loss[loss=0.3108, simple_loss=0.3336, pruned_loss=0.144, over 4823.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.2953, pruned_loss=0.09325, over 931197.47 frames. ], batch size: 39, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:37:08,559 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17960.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:37:30,390 INFO [finetune.py:976] (3/7) Epoch 4, batch 800, loss[loss=0.2441, simple_loss=0.2826, pruned_loss=0.1028, over 4803.00 frames. ], tot_loss[loss=0.2397, simple_loss=0.2946, pruned_loss=0.09245, over 936571.12 frames. ], batch size: 25, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:38:05,562 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.155e+02 1.784e+02 2.191e+02 2.808e+02 5.190e+02, threshold=4.382e+02, percent-clipped=3.0 2023-03-26 02:38:05,684 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18018.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:38:08,550 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18021.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:38:09,863 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.96 vs. limit=5.0 2023-03-26 02:38:20,834 INFO [finetune.py:976] (3/7) Epoch 4, batch 850, loss[loss=0.2059, simple_loss=0.261, pruned_loss=0.07542, over 4828.00 frames. ], tot_loss[loss=0.2372, simple_loss=0.2912, pruned_loss=0.09157, over 940818.33 frames. ], batch size: 33, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:38:26,595 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18039.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 02:38:26,810 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-26 02:38:39,176 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18056.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:38:43,928 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.3497, 1.4927, 1.5267, 0.8889, 1.4137, 1.7334, 1.7224, 1.3865], device='cuda:3'), covar=tensor([0.0999, 0.0522, 0.0591, 0.0560, 0.0606, 0.0507, 0.0396, 0.0589], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0159, 0.0117, 0.0138, 0.0134, 0.0123, 0.0147, 0.0145], device='cuda:3'), out_proj_covar=tensor([9.8537e-05, 1.1809e-04, 8.5377e-05, 1.0129e-04, 9.7212e-05, 9.1131e-05, 1.0991e-04, 1.0767e-04], device='cuda:3') 2023-03-26 02:38:45,706 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=18066.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:38:57,811 INFO [finetune.py:976] (3/7) Epoch 4, batch 900, loss[loss=0.1949, simple_loss=0.246, pruned_loss=0.07186, over 4808.00 frames. ], tot_loss[loss=0.2336, simple_loss=0.2878, pruned_loss=0.08974, over 944522.31 frames. ], batch size: 25, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:38:59,693 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18086.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:39:00,270 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=18087.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 02:39:19,671 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.218e+02 1.720e+02 1.921e+02 2.312e+02 4.297e+02, threshold=3.842e+02, percent-clipped=0.0 2023-03-26 02:39:28,070 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-03-26 02:39:35,996 INFO [finetune.py:976] (3/7) Epoch 4, batch 950, loss[loss=0.1854, simple_loss=0.2449, pruned_loss=0.06297, over 4786.00 frames. ], tot_loss[loss=0.2305, simple_loss=0.285, pruned_loss=0.08801, over 947366.38 frames. ], batch size: 26, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:40:07,923 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.6868, 1.8114, 2.1034, 1.3482, 1.8587, 2.1541, 2.0616, 1.7730], device='cuda:3'), covar=tensor([0.0924, 0.0600, 0.0324, 0.0495, 0.0382, 0.0567, 0.0329, 0.0505], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0158, 0.0117, 0.0138, 0.0133, 0.0122, 0.0147, 0.0145], device='cuda:3'), out_proj_covar=tensor([9.8408e-05, 1.1778e-04, 8.5075e-05, 1.0144e-04, 9.6654e-05, 9.0824e-05, 1.0967e-04, 1.0721e-04], device='cuda:3') 2023-03-26 02:40:22,942 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18174.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:40:24,036 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18175.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:40:29,235 INFO [finetune.py:976] (3/7) Epoch 4, batch 1000, loss[loss=0.2866, simple_loss=0.3449, pruned_loss=0.1142, over 4752.00 frames. ], tot_loss[loss=0.2346, simple_loss=0.2891, pruned_loss=0.09009, over 948676.77 frames. ], batch size: 54, lr: 3.97e-03, grad_scale: 64.0 2023-03-26 02:40:44,149 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7645, 1.6002, 1.3321, 1.4196, 1.5763, 1.5035, 1.5379, 2.3649], device='cuda:3'), covar=tensor([0.9883, 0.9844, 0.7390, 1.0365, 0.8332, 0.5046, 0.9011, 0.3054], device='cuda:3'), in_proj_covar=tensor([0.0275, 0.0250, 0.0218, 0.0283, 0.0235, 0.0196, 0.0239, 0.0188], 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') 2023-03-26 02:41:07,310 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.077e+02 1.690e+02 2.099e+02 2.471e+02 3.966e+02, threshold=4.198e+02, percent-clipped=1.0 2023-03-26 02:41:28,636 INFO [finetune.py:976] (3/7) Epoch 4, batch 1050, loss[loss=0.2789, simple_loss=0.3331, pruned_loss=0.1123, over 4207.00 frames. ], tot_loss[loss=0.2368, simple_loss=0.2918, pruned_loss=0.09093, over 948001.59 frames. ], batch size: 65, lr: 3.97e-03, grad_scale: 64.0 2023-03-26 02:41:29,950 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18235.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:41:30,592 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18236.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:41:33,075 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9549, 1.8082, 1.4467, 1.8823, 1.7739, 1.7359, 1.6907, 2.7107], device='cuda:3'), covar=tensor([1.0117, 1.2000, 0.8176, 1.1782, 0.9424, 0.5724, 1.1321, 0.3174], device='cuda:3'), in_proj_covar=tensor([0.0274, 0.0250, 0.0218, 0.0282, 0.0234, 0.0196, 0.0238, 0.0187], 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') 2023-03-26 02:42:18,619 INFO [finetune.py:976] (3/7) Epoch 4, batch 1100, loss[loss=0.2651, simple_loss=0.3155, pruned_loss=0.1073, over 4913.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.2933, pruned_loss=0.09197, over 946173.99 frames. ], batch size: 36, lr: 3.97e-03, grad_scale: 64.0 2023-03-26 02:42:39,981 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6675, 1.5150, 1.5161, 1.5591, 1.0874, 3.5087, 1.2886, 1.7348], device='cuda:3'), covar=tensor([0.3414, 0.2376, 0.2082, 0.2300, 0.1995, 0.0171, 0.2628, 0.1361], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0111, 0.0116, 0.0119, 0.0115, 0.0096, 0.0100, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 02:42:49,579 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18316.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:42:50,749 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.170e+02 1.851e+02 2.220e+02 2.754e+02 4.687e+02, threshold=4.440e+02, percent-clipped=1.0 2023-03-26 02:43:08,652 INFO [finetune.py:976] (3/7) Epoch 4, batch 1150, loss[loss=0.2504, simple_loss=0.3043, pruned_loss=0.09823, over 4922.00 frames. ], tot_loss[loss=0.238, simple_loss=0.2931, pruned_loss=0.09149, over 950009.89 frames. ], batch size: 38, lr: 3.97e-03, grad_scale: 64.0 2023-03-26 02:43:14,698 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-03-26 02:43:19,809 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1229, 3.6004, 3.7860, 3.9997, 3.8263, 3.6476, 4.1783, 1.3102], device='cuda:3'), covar=tensor([0.0682, 0.0753, 0.0792, 0.0878, 0.1138, 0.1349, 0.0704, 0.4777], device='cuda:3'), in_proj_covar=tensor([0.0362, 0.0245, 0.0279, 0.0295, 0.0342, 0.0286, 0.0310, 0.0302], 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') 2023-03-26 02:43:21,094 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5894, 2.1727, 1.8987, 1.1420, 2.1843, 1.9566, 1.5472, 1.9726], device='cuda:3'), covar=tensor([0.0793, 0.1080, 0.1924, 0.2139, 0.1615, 0.2232, 0.2209, 0.1204], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0199, 0.0203, 0.0190, 0.0217, 0.0210, 0.0218, 0.0200], 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') 2023-03-26 02:43:25,368 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18356.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:44:03,368 INFO [finetune.py:976] (3/7) Epoch 4, batch 1200, loss[loss=0.256, simple_loss=0.3016, pruned_loss=0.1052, over 4908.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.2915, pruned_loss=0.09139, over 951193.35 frames. ], batch size: 37, lr: 3.97e-03, grad_scale: 64.0 2023-03-26 02:44:05,826 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18386.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:44:27,174 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4587, 2.2631, 2.1361, 2.4481, 3.1289, 2.4244, 2.2240, 2.0089], device='cuda:3'), covar=tensor([0.2468, 0.2266, 0.2043, 0.1949, 0.1840, 0.1158, 0.2752, 0.1994], device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0208, 0.0197, 0.0183, 0.0232, 0.0173, 0.0213, 0.0186], 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') 2023-03-26 02:44:28,300 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=18404.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:44:28,782 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-03-26 02:44:47,472 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.035e+02 1.754e+02 2.082e+02 2.492e+02 3.668e+02, threshold=4.164e+02, percent-clipped=0.0 2023-03-26 02:45:07,910 INFO [finetune.py:976] (3/7) Epoch 4, batch 1250, loss[loss=0.1814, simple_loss=0.2542, pruned_loss=0.05436, over 4934.00 frames. ], tot_loss[loss=0.2347, simple_loss=0.2889, pruned_loss=0.09019, over 949760.88 frames. ], batch size: 33, lr: 3.97e-03, grad_scale: 64.0 2023-03-26 02:45:09,086 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=18434.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:45:38,725 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6360, 1.4496, 1.3525, 1.6717, 1.9239, 1.6619, 1.0991, 1.2904], device='cuda:3'), covar=tensor([0.2751, 0.2651, 0.2457, 0.2120, 0.2042, 0.1371, 0.3407, 0.2328], device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0209, 0.0197, 0.0184, 0.0233, 0.0174, 0.0214, 0.0187], 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') 2023-03-26 02:45:59,532 INFO [finetune.py:976] (3/7) Epoch 4, batch 1300, loss[loss=0.1621, simple_loss=0.2248, pruned_loss=0.04971, over 4704.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.2844, pruned_loss=0.08806, over 948386.19 frames. ], batch size: 23, lr: 3.97e-03, grad_scale: 64.0 2023-03-26 02:46:00,762 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9371, 1.3144, 1.0562, 1.7384, 2.2580, 1.5176, 1.6038, 1.9050], device='cuda:3'), covar=tensor([0.1463, 0.2086, 0.2101, 0.1206, 0.1909, 0.1879, 0.1405, 0.1909], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0100, 0.0117, 0.0095, 0.0126, 0.0098, 0.0102, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 02:46:10,683 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18490.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:46:22,660 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18508.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 02:46:29,709 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.069e+02 1.658e+02 2.008e+02 2.632e+02 4.281e+02, threshold=4.017e+02, percent-clipped=1.0 2023-03-26 02:46:36,934 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18530.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:46:37,495 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18531.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:46:41,946 INFO [finetune.py:976] (3/7) Epoch 4, batch 1350, loss[loss=0.2318, simple_loss=0.3012, pruned_loss=0.08122, over 4907.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.2848, pruned_loss=0.0888, over 948872.35 frames. ], batch size: 43, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:46:55,203 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18551.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:47:11,772 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18569.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 02:47:25,815 INFO [finetune.py:976] (3/7) Epoch 4, batch 1400, loss[loss=0.2274, simple_loss=0.2758, pruned_loss=0.08945, over 4768.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.29, pruned_loss=0.09103, over 951133.21 frames. ], batch size: 26, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:47:26,504 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9101, 1.9995, 1.8986, 1.3235, 2.1155, 2.2380, 2.0435, 1.7330], device='cuda:3'), covar=tensor([0.0677, 0.0577, 0.0830, 0.1102, 0.0469, 0.0581, 0.0677, 0.0987], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0133, 0.0144, 0.0129, 0.0110, 0.0142, 0.0147, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 02:47:31,095 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-26 02:47:53,782 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18616.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:47:55,468 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.253e+02 1.863e+02 2.207e+02 2.712e+02 5.337e+02, threshold=4.415e+02, percent-clipped=2.0 2023-03-26 02:48:10,032 INFO [finetune.py:976] (3/7) Epoch 4, batch 1450, loss[loss=0.3135, simple_loss=0.3567, pruned_loss=0.1352, over 4892.00 frames. ], tot_loss[loss=0.2369, simple_loss=0.2916, pruned_loss=0.09108, over 952833.47 frames. ], batch size: 37, lr: 3.97e-03, grad_scale: 32.0 2023-03-26 02:48:38,583 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8405, 2.4759, 2.3263, 1.4505, 2.4511, 2.0226, 1.6982, 2.2067], device='cuda:3'), covar=tensor([0.1078, 0.0962, 0.1715, 0.2346, 0.2077, 0.2462, 0.2278, 0.1288], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0200, 0.0203, 0.0191, 0.0218, 0.0211, 0.0220, 0.0202], 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') 2023-03-26 02:48:43,321 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=18664.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:48:47,595 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0998, 1.7920, 2.4138, 1.5749, 2.2618, 2.1701, 1.8044, 2.5593], device='cuda:3'), covar=tensor([0.1438, 0.1985, 0.1337, 0.1938, 0.0935, 0.1653, 0.2127, 0.0834], device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0206, 0.0203, 0.0196, 0.0182, 0.0226, 0.0215, 0.0204], 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') 2023-03-26 02:48:48,461 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-26 02:48:49,426 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18674.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:48:56,989 INFO [finetune.py:976] (3/7) Epoch 4, batch 1500, loss[loss=0.2751, simple_loss=0.3241, pruned_loss=0.113, over 4882.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.292, pruned_loss=0.09076, over 954226.12 frames. ], batch size: 43, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 02:49:36,307 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.377e+02 1.810e+02 2.122e+02 2.509e+02 6.153e+02, threshold=4.245e+02, percent-clipped=1.0 2023-03-26 02:49:36,437 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2504, 1.3223, 1.6007, 1.1354, 1.2830, 1.3981, 1.3306, 1.6031], device='cuda:3'), covar=tensor([0.1475, 0.2291, 0.1387, 0.1526, 0.1113, 0.1433, 0.2739, 0.1017], device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0207, 0.0203, 0.0197, 0.0182, 0.0227, 0.0215, 0.0204], 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') 2023-03-26 02:49:54,650 INFO [finetune.py:976] (3/7) Epoch 4, batch 1550, loss[loss=0.2504, simple_loss=0.2972, pruned_loss=0.1018, over 4801.00 frames. ], tot_loss[loss=0.2354, simple_loss=0.2907, pruned_loss=0.08999, over 950678.36 frames. ], batch size: 39, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 02:49:55,975 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18735.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:50:24,705 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.3321, 3.7908, 3.9020, 4.2055, 4.0466, 3.8151, 4.4028, 1.4433], device='cuda:3'), covar=tensor([0.0645, 0.0681, 0.0799, 0.0721, 0.1099, 0.1409, 0.0633, 0.4861], device='cuda:3'), in_proj_covar=tensor([0.0357, 0.0242, 0.0275, 0.0290, 0.0335, 0.0283, 0.0306, 0.0297], 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') 2023-03-26 02:50:39,060 INFO [finetune.py:976] (3/7) Epoch 4, batch 1600, loss[loss=0.2507, simple_loss=0.3047, pruned_loss=0.09837, over 4920.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.2879, pruned_loss=0.08868, over 950788.45 frames. ], batch size: 43, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 02:50:42,306 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6237, 1.5302, 1.5278, 1.6103, 1.0498, 3.1457, 1.2255, 1.7968], device='cuda:3'), covar=tensor([0.3294, 0.2312, 0.1946, 0.2215, 0.1991, 0.0207, 0.2774, 0.1299], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0112, 0.0116, 0.0120, 0.0116, 0.0097, 0.0101, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 02:51:19,524 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.176e+02 1.645e+02 2.011e+02 2.399e+02 5.772e+02, threshold=4.021e+02, percent-clipped=1.0 2023-03-26 02:51:30,366 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18830.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:51:30,962 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18831.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:51:32,085 INFO [finetune.py:976] (3/7) Epoch 4, batch 1650, loss[loss=0.1943, simple_loss=0.2518, pruned_loss=0.06839, over 4857.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.285, pruned_loss=0.08713, over 953792.84 frames. ], batch size: 47, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 02:51:48,114 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18846.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:52:05,189 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18864.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 02:52:13,739 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9568, 1.8212, 1.7761, 2.0761, 1.5188, 4.5092, 1.7775, 2.4303], device='cuda:3'), covar=tensor([0.3641, 0.2426, 0.2064, 0.2219, 0.1817, 0.0095, 0.2506, 0.1289], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0112, 0.0117, 0.0120, 0.0117, 0.0097, 0.0101, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 02:52:23,039 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=18878.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:52:23,636 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=18879.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:52:25,972 INFO [finetune.py:976] (3/7) Epoch 4, batch 1700, loss[loss=0.2286, simple_loss=0.2892, pruned_loss=0.08404, over 4826.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.2838, pruned_loss=0.08723, over 954317.81 frames. ], batch size: 30, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 02:52:31,212 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-26 02:53:00,726 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.073e+02 1.767e+02 2.149e+02 2.599e+02 5.673e+02, threshold=4.299e+02, percent-clipped=2.0 2023-03-26 02:53:07,537 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18930.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:53:09,235 INFO [finetune.py:976] (3/7) Epoch 4, batch 1750, loss[loss=0.2359, simple_loss=0.2935, pruned_loss=0.08915, over 4897.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.2866, pruned_loss=0.08876, over 951216.37 frames. ], batch size: 35, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 02:53:11,221 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9130, 1.7837, 1.4355, 1.9141, 1.8774, 1.5731, 2.3208, 1.8443], device='cuda:3'), covar=tensor([0.2036, 0.3661, 0.4527, 0.3681, 0.3384, 0.2336, 0.3828, 0.2788], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0194, 0.0238, 0.0255, 0.0223, 0.0185, 0.0210, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 02:53:33,396 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8047, 3.9634, 3.7812, 1.9467, 4.0677, 2.8886, 1.1204, 2.7626], device='cuda:3'), covar=tensor([0.2305, 0.2186, 0.1735, 0.3276, 0.0860, 0.0988, 0.4447, 0.1768], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0170, 0.0164, 0.0129, 0.0156, 0.0123, 0.0146, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 02:53:50,573 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6535, 3.6943, 3.5941, 1.8922, 3.8055, 2.7864, 0.8190, 2.6309], device='cuda:3'), covar=tensor([0.2627, 0.1700, 0.1604, 0.2992, 0.1030, 0.1020, 0.4443, 0.1395], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0170, 0.0164, 0.0129, 0.0156, 0.0123, 0.0146, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 02:53:52,364 INFO [finetune.py:976] (3/7) Epoch 4, batch 1800, loss[loss=0.253, simple_loss=0.3105, pruned_loss=0.09771, over 4908.00 frames. ], tot_loss[loss=0.2348, simple_loss=0.2901, pruned_loss=0.08975, over 952758.46 frames. ], batch size: 36, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 02:53:57,497 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18991.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:54:06,965 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4066, 2.1566, 1.8326, 1.0711, 2.1204, 1.9435, 1.5481, 1.9780], device='cuda:3'), covar=tensor([0.0918, 0.0795, 0.1550, 0.1934, 0.1235, 0.2010, 0.2108, 0.1045], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0199, 0.0202, 0.0189, 0.0216, 0.0208, 0.0218, 0.0200], 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') 2023-03-26 02:54:32,249 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.317e+02 1.869e+02 2.115e+02 2.590e+02 5.981e+02, threshold=4.230e+02, percent-clipped=1.0 2023-03-26 02:54:50,224 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19030.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:54:52,013 INFO [finetune.py:976] (3/7) Epoch 4, batch 1850, loss[loss=0.2051, simple_loss=0.269, pruned_loss=0.07059, over 4883.00 frames. ], tot_loss[loss=0.237, simple_loss=0.2919, pruned_loss=0.09104, over 951806.21 frames. ], batch size: 32, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 02:55:00,202 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.79 vs. limit=5.0 2023-03-26 02:55:39,895 INFO [finetune.py:976] (3/7) Epoch 4, batch 1900, loss[loss=0.2606, simple_loss=0.318, pruned_loss=0.1016, over 4735.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.2935, pruned_loss=0.09095, over 951706.57 frames. ], batch size: 59, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 02:55:50,714 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7344, 1.5374, 2.1063, 3.3624, 2.4193, 2.4026, 0.9892, 2.6586], device='cuda:3'), covar=tensor([0.1815, 0.1675, 0.1335, 0.0633, 0.0781, 0.1362, 0.2099, 0.0647], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0119, 0.0137, 0.0166, 0.0104, 0.0143, 0.0129, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 02:55:56,280 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 2023-03-26 02:56:12,754 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.301e+02 1.752e+02 2.082e+02 2.658e+02 3.786e+02, threshold=4.164e+02, percent-clipped=0.0 2023-03-26 02:56:12,959 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-03-26 02:56:25,543 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19129.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 02:56:33,396 INFO [finetune.py:976] (3/7) Epoch 4, batch 1950, loss[loss=0.2265, simple_loss=0.2652, pruned_loss=0.09392, over 4491.00 frames. ], tot_loss[loss=0.2345, simple_loss=0.2903, pruned_loss=0.08935, over 952470.20 frames. ], batch size: 19, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 02:56:35,875 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4971, 1.5352, 1.8075, 1.7809, 1.6750, 3.6308, 1.3912, 1.6072], device='cuda:3'), covar=tensor([0.1108, 0.1734, 0.1184, 0.1108, 0.1680, 0.0233, 0.1526, 0.1782], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0082, 0.0078, 0.0080, 0.0094, 0.0083, 0.0086, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 02:56:41,417 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19146.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:56:53,408 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19164.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 02:57:09,558 INFO [finetune.py:976] (3/7) Epoch 4, batch 2000, loss[loss=0.2669, simple_loss=0.309, pruned_loss=0.1124, over 4812.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.2878, pruned_loss=0.08862, over 953515.88 frames. ], batch size: 39, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 02:57:14,702 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19190.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 02:57:17,078 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=19194.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:57:23,264 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8116, 1.9336, 1.7232, 1.1373, 2.0647, 2.0197, 1.8854, 1.6896], device='cuda:3'), covar=tensor([0.0618, 0.0573, 0.0792, 0.1047, 0.0561, 0.0656, 0.0666, 0.0994], device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0134, 0.0146, 0.0130, 0.0112, 0.0144, 0.0148, 0.0164], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 02:57:35,168 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=19212.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 02:57:39,348 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.251e+02 1.681e+02 2.009e+02 2.396e+02 5.395e+02, threshold=4.017e+02, percent-clipped=3.0 2023-03-26 02:57:49,438 INFO [finetune.py:976] (3/7) Epoch 4, batch 2050, loss[loss=0.207, simple_loss=0.2676, pruned_loss=0.07319, over 4806.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.2836, pruned_loss=0.08702, over 951694.61 frames. ], batch size: 29, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 02:58:31,807 INFO [finetune.py:976] (3/7) Epoch 4, batch 2100, loss[loss=0.2013, simple_loss=0.2617, pruned_loss=0.07043, over 4837.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.2842, pruned_loss=0.0875, over 952520.82 frames. ], batch size: 33, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 02:58:34,245 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19286.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:59:05,562 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-26 02:59:09,863 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.136e+02 1.737e+02 2.057e+02 2.371e+02 3.601e+02, threshold=4.115e+02, percent-clipped=0.0 2023-03-26 02:59:17,249 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.79 vs. limit=5.0 2023-03-26 02:59:25,781 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19330.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 02:59:27,977 INFO [finetune.py:976] (3/7) Epoch 4, batch 2150, loss[loss=0.2312, simple_loss=0.2916, pruned_loss=0.08544, over 4773.00 frames. ], tot_loss[loss=0.2328, simple_loss=0.2879, pruned_loss=0.08884, over 955086.67 frames. ], batch size: 54, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:00:10,912 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=19378.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:00:13,945 INFO [finetune.py:976] (3/7) Epoch 4, batch 2200, loss[loss=0.2424, simple_loss=0.2997, pruned_loss=0.09259, over 4928.00 frames. ], tot_loss[loss=0.2335, simple_loss=0.2895, pruned_loss=0.08871, over 956393.25 frames. ], batch size: 38, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:00:33,376 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19394.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:01:05,355 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.065e+02 1.822e+02 2.092e+02 2.549e+02 4.918e+02, threshold=4.184e+02, percent-clipped=2.0 2023-03-26 03:01:19,366 INFO [finetune.py:976] (3/7) Epoch 4, batch 2250, loss[loss=0.2345, simple_loss=0.2745, pruned_loss=0.0972, over 4025.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.2917, pruned_loss=0.08976, over 956807.52 frames. ], batch size: 17, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:01:42,174 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2975, 1.2331, 1.5457, 1.0707, 1.2115, 1.3326, 1.2597, 1.5507], device='cuda:3'), covar=tensor([0.1214, 0.2010, 0.1206, 0.1471, 0.0940, 0.1362, 0.2512, 0.0851], device='cuda:3'), in_proj_covar=tensor([0.0208, 0.0208, 0.0204, 0.0198, 0.0183, 0.0227, 0.0216, 0.0206], 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') 2023-03-26 03:01:46,302 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19455.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:02:09,425 INFO [finetune.py:976] (3/7) Epoch 4, batch 2300, loss[loss=0.2147, simple_loss=0.2802, pruned_loss=0.07465, over 4763.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.2926, pruned_loss=0.08959, over 956448.90 frames. ], batch size: 27, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:02:12,902 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19485.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 03:02:37,770 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.125e+02 1.717e+02 2.025e+02 2.639e+02 4.089e+02, threshold=4.050e+02, percent-clipped=0.0 2023-03-26 03:02:45,332 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6889, 1.5458, 1.5784, 1.6429, 1.1557, 3.6203, 1.3783, 1.9965], device='cuda:3'), covar=tensor([0.3363, 0.2428, 0.2106, 0.2344, 0.1986, 0.0146, 0.2649, 0.1345], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0112, 0.0116, 0.0120, 0.0116, 0.0097, 0.0101, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 03:02:53,439 INFO [finetune.py:976] (3/7) Epoch 4, batch 2350, loss[loss=0.2218, simple_loss=0.2646, pruned_loss=0.08949, over 4828.00 frames. ], tot_loss[loss=0.2329, simple_loss=0.2889, pruned_loss=0.08847, over 955923.29 frames. ], batch size: 33, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:03:37,780 INFO [finetune.py:976] (3/7) Epoch 4, batch 2400, loss[loss=0.2435, simple_loss=0.3099, pruned_loss=0.08855, over 4824.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.2851, pruned_loss=0.08702, over 954969.12 frames. ], batch size: 39, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:03:40,263 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19586.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:03:40,881 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8845, 1.6928, 1.6147, 1.8157, 1.1823, 4.3397, 1.5096, 2.1409], device='cuda:3'), covar=tensor([0.3268, 0.2354, 0.2110, 0.2222, 0.1869, 0.0091, 0.2659, 0.1384], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0112, 0.0117, 0.0120, 0.0117, 0.0097, 0.0101, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 03:03:49,285 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6951, 1.4555, 1.9667, 1.9085, 1.7501, 4.1455, 1.4024, 1.7366], device='cuda:3'), covar=tensor([0.1071, 0.1971, 0.1426, 0.1202, 0.1737, 0.0204, 0.1783, 0.1925], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0082, 0.0079, 0.0080, 0.0093, 0.0084, 0.0086, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 03:04:10,018 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19618.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:04:10,537 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.076e+02 1.651e+02 1.936e+02 2.390e+02 3.810e+02, threshold=3.872e+02, percent-clipped=0.0 2023-03-26 03:04:19,620 INFO [finetune.py:976] (3/7) Epoch 4, batch 2450, loss[loss=0.1986, simple_loss=0.2589, pruned_loss=0.06912, over 4797.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.2812, pruned_loss=0.08547, over 954377.82 frames. ], batch size: 25, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:04:20,303 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=19634.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:04:33,697 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6009, 1.2898, 1.7066, 1.8643, 1.4880, 3.2732, 1.1961, 1.3941], device='cuda:3'), covar=tensor([0.1228, 0.2326, 0.1518, 0.1184, 0.1912, 0.0318, 0.2087, 0.2377], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0082, 0.0078, 0.0080, 0.0093, 0.0084, 0.0086, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 03:04:59,501 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.26 vs. limit=5.0 2023-03-26 03:04:59,939 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19679.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:05:02,276 INFO [finetune.py:976] (3/7) Epoch 4, batch 2500, loss[loss=0.2366, simple_loss=0.3064, pruned_loss=0.08338, over 4742.00 frames. ], tot_loss[loss=0.23, simple_loss=0.2845, pruned_loss=0.0877, over 954664.01 frames. ], batch size: 54, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:05:23,526 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.44 vs. limit=5.0 2023-03-26 03:05:30,656 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.296e+02 1.721e+02 2.075e+02 2.470e+02 4.533e+02, threshold=4.150e+02, percent-clipped=4.0 2023-03-26 03:05:38,217 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-26 03:05:45,256 INFO [finetune.py:976] (3/7) Epoch 4, batch 2550, loss[loss=0.2113, simple_loss=0.2861, pruned_loss=0.06822, over 4896.00 frames. ], tot_loss[loss=0.2337, simple_loss=0.2893, pruned_loss=0.08907, over 956007.66 frames. ], batch size: 37, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:05:46,082 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-03-26 03:05:58,475 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19750.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:05:58,765 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-03-26 03:06:31,488 INFO [finetune.py:976] (3/7) Epoch 4, batch 2600, loss[loss=0.2176, simple_loss=0.2668, pruned_loss=0.0842, over 4860.00 frames. ], tot_loss[loss=0.2346, simple_loss=0.2901, pruned_loss=0.08958, over 956211.25 frames. ], batch size: 31, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:06:33,307 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19785.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 03:07:15,920 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.335e+02 1.804e+02 2.222e+02 2.871e+02 4.406e+02, threshold=4.445e+02, percent-clipped=2.0 2023-03-26 03:07:33,381 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0120, 1.8635, 1.4977, 1.8500, 2.0292, 1.6180, 2.3320, 1.9681], device='cuda:3'), covar=tensor([0.1860, 0.3760, 0.4382, 0.4217, 0.3047, 0.2226, 0.4142, 0.2649], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0194, 0.0238, 0.0256, 0.0224, 0.0187, 0.0211, 0.0189], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 03:07:34,471 INFO [finetune.py:976] (3/7) Epoch 4, batch 2650, loss[loss=0.2355, simple_loss=0.3085, pruned_loss=0.08126, over 4854.00 frames. ], tot_loss[loss=0.2368, simple_loss=0.2926, pruned_loss=0.09049, over 956598.86 frames. ], batch size: 44, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:07:34,541 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=19833.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 03:07:36,423 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.83 vs. limit=5.0 2023-03-26 03:08:05,184 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1291, 1.8173, 1.5023, 1.9524, 1.8261, 1.7803, 1.7627, 2.7979], device='cuda:3'), covar=tensor([0.8904, 1.1272, 0.7505, 1.0565, 0.9035, 0.5006, 1.0999, 0.3009], device='cuda:3'), in_proj_covar=tensor([0.0276, 0.0251, 0.0219, 0.0283, 0.0235, 0.0196, 0.0239, 0.0189], 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') 2023-03-26 03:08:18,116 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7325, 1.7891, 1.5894, 1.8999, 2.3439, 1.8178, 1.5423, 1.4356], device='cuda:3'), covar=tensor([0.2458, 0.2270, 0.1999, 0.1747, 0.1908, 0.1344, 0.2894, 0.1993], device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0211, 0.0199, 0.0185, 0.0234, 0.0175, 0.0216, 0.0188], 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') 2023-03-26 03:08:34,283 INFO [finetune.py:976] (3/7) Epoch 4, batch 2700, loss[loss=0.2187, simple_loss=0.2573, pruned_loss=0.09004, over 4225.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.2912, pruned_loss=0.08997, over 955964.30 frames. ], batch size: 18, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:08:58,100 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.62 vs. limit=5.0 2023-03-26 03:09:16,013 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.181e+02 1.716e+02 2.005e+02 2.489e+02 3.950e+02, threshold=4.009e+02, percent-clipped=0.0 2023-03-26 03:09:26,957 INFO [finetune.py:976] (3/7) Epoch 4, batch 2750, loss[loss=0.2221, simple_loss=0.2721, pruned_loss=0.08612, over 4824.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.2872, pruned_loss=0.08821, over 956176.73 frames. ], batch size: 49, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:09:40,804 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5771, 1.3788, 1.3323, 1.2419, 1.7322, 1.7221, 1.6078, 1.3145], device='cuda:3'), covar=tensor([0.0237, 0.0349, 0.0553, 0.0347, 0.0209, 0.0342, 0.0313, 0.0409], device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0114, 0.0138, 0.0118, 0.0104, 0.0099, 0.0092, 0.0109], device='cuda:3'), out_proj_covar=tensor([6.6579e-05, 9.0110e-05, 1.1128e-04, 9.3341e-05, 8.2547e-05, 7.3838e-05, 7.0618e-05, 8.5410e-05], device='cuda:3') 2023-03-26 03:09:54,795 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19974.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:10:00,205 INFO [finetune.py:976] (3/7) Epoch 4, batch 2800, loss[loss=0.2045, simple_loss=0.2554, pruned_loss=0.0768, over 4895.00 frames. ], tot_loss[loss=0.228, simple_loss=0.2831, pruned_loss=0.08647, over 955999.32 frames. ], batch size: 32, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:10:24,997 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.030e+02 1.781e+02 2.091e+02 2.518e+02 3.954e+02, threshold=4.183e+02, percent-clipped=0.0 2023-03-26 03:10:26,235 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20020.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:10:34,543 INFO [finetune.py:976] (3/7) Epoch 4, batch 2850, loss[loss=0.247, simple_loss=0.3056, pruned_loss=0.09416, over 4813.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.282, pruned_loss=0.08648, over 955060.69 frames. ], batch size: 41, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:10:45,851 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20050.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:11:21,582 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20081.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:11:22,669 INFO [finetune.py:976] (3/7) Epoch 4, batch 2900, loss[loss=0.2438, simple_loss=0.3026, pruned_loss=0.09255, over 4819.00 frames. ], tot_loss[loss=0.2299, simple_loss=0.2848, pruned_loss=0.08751, over 952647.17 frames. ], batch size: 40, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:11:31,832 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20089.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:11:42,687 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=20098.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:11:56,189 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20112.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:12:05,858 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.342e+02 1.862e+02 2.201e+02 2.748e+02 4.534e+02, threshold=4.402e+02, percent-clipped=1.0 2023-03-26 03:12:27,413 INFO [finetune.py:976] (3/7) Epoch 4, batch 2950, loss[loss=0.2799, simple_loss=0.3325, pruned_loss=0.1136, over 4794.00 frames. ], tot_loss[loss=0.2326, simple_loss=0.288, pruned_loss=0.08867, over 953145.58 frames. ], batch size: 51, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:12:48,259 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20150.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:13:10,786 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20173.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:13:18,822 INFO [finetune.py:976] (3/7) Epoch 4, batch 3000, loss[loss=0.2687, simple_loss=0.3142, pruned_loss=0.1116, over 4894.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.2898, pruned_loss=0.08938, over 953883.14 frames. ], batch size: 35, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:13:18,822 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 03:13:27,571 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5988, 1.4918, 1.5378, 1.5985, 1.0418, 2.9927, 1.1816, 1.7887], device='cuda:3'), covar=tensor([0.3385, 0.2356, 0.2090, 0.2337, 0.2096, 0.0239, 0.2802, 0.1404], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0112, 0.0116, 0.0120, 0.0117, 0.0097, 0.0101, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 03:13:34,007 INFO [finetune.py:1010] (3/7) Epoch 4, validation: loss=0.169, simple_loss=0.2409, pruned_loss=0.04857, over 2265189.00 frames. 2023-03-26 03:13:34,008 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 03:13:58,695 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20204.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:14:18,154 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.180e+02 1.917e+02 2.134e+02 2.804e+02 4.274e+02, threshold=4.268e+02, percent-clipped=0.0 2023-03-26 03:14:27,243 INFO [finetune.py:976] (3/7) Epoch 4, batch 3050, loss[loss=0.2534, simple_loss=0.3068, pruned_loss=0.09996, over 4818.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.2916, pruned_loss=0.09038, over 954245.39 frames. ], batch size: 30, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:14:49,595 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9691, 1.8294, 1.5202, 2.0766, 1.9867, 1.6715, 2.3839, 1.9940], device='cuda:3'), covar=tensor([0.1922, 0.3952, 0.4428, 0.3745, 0.3159, 0.2144, 0.4145, 0.2523], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0195, 0.0238, 0.0256, 0.0225, 0.0187, 0.0211, 0.0189], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 03:14:59,755 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0210, 1.9781, 1.4461, 2.1823, 2.0537, 1.6602, 2.9708, 1.9941], device='cuda:3'), covar=tensor([0.1834, 0.3437, 0.4312, 0.3924, 0.3184, 0.2012, 0.2859, 0.2565], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0195, 0.0239, 0.0257, 0.0225, 0.0188, 0.0212, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 03:15:06,971 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20265.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:15:16,616 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20274.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:15:22,248 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.77 vs. limit=5.0 2023-03-26 03:15:24,345 INFO [finetune.py:976] (3/7) Epoch 4, batch 3100, loss[loss=0.2013, simple_loss=0.2543, pruned_loss=0.07418, over 4811.00 frames. ], tot_loss[loss=0.233, simple_loss=0.2886, pruned_loss=0.08868, over 954742.96 frames. ], batch size: 41, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:16:01,239 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.111e+02 1.549e+02 1.969e+02 2.570e+02 5.632e+02, threshold=3.937e+02, percent-clipped=1.0 2023-03-26 03:16:03,106 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=20322.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:16:14,161 INFO [finetune.py:976] (3/7) Epoch 4, batch 3150, loss[loss=0.2271, simple_loss=0.2693, pruned_loss=0.09247, over 4809.00 frames. ], tot_loss[loss=0.2298, simple_loss=0.2849, pruned_loss=0.08733, over 956618.40 frames. ], batch size: 39, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:16:33,356 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9079, 1.6764, 1.5395, 1.6928, 1.9718, 1.6210, 2.1004, 1.8466], device='cuda:3'), covar=tensor([0.1980, 0.3668, 0.4510, 0.3602, 0.3239, 0.2325, 0.3531, 0.2851], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0194, 0.0237, 0.0255, 0.0224, 0.0187, 0.0210, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 03:16:56,930 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20376.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:17:01,162 INFO [finetune.py:976] (3/7) Epoch 4, batch 3200, loss[loss=0.1833, simple_loss=0.2439, pruned_loss=0.06136, over 4107.00 frames. ], tot_loss[loss=0.2273, simple_loss=0.2822, pruned_loss=0.08618, over 956763.33 frames. ], batch size: 17, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:17:04,823 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20388.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:17:07,274 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.12 vs. limit=5.0 2023-03-26 03:17:28,180 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2516, 2.0007, 1.8576, 2.0361, 2.2491, 1.9451, 2.4537, 2.1516], device='cuda:3'), covar=tensor([0.1785, 0.3667, 0.4186, 0.3534, 0.2799, 0.1988, 0.3803, 0.2445], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0195, 0.0239, 0.0256, 0.0226, 0.0188, 0.0211, 0.0189], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 03:17:40,213 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20418.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:17:40,663 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.143e+02 1.625e+02 1.959e+02 2.342e+02 5.079e+02, threshold=3.919e+02, percent-clipped=3.0 2023-03-26 03:17:57,126 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-26 03:17:58,486 INFO [finetune.py:976] (3/7) Epoch 4, batch 3250, loss[loss=0.232, simple_loss=0.3015, pruned_loss=0.08126, over 4826.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.2827, pruned_loss=0.08626, over 955647.22 frames. ], batch size: 33, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:18:00,248 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-03-26 03:18:01,659 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8582, 1.6705, 1.3315, 1.5593, 1.6048, 1.5127, 1.5103, 2.3852], device='cuda:3'), covar=tensor([0.9118, 0.9732, 0.7347, 0.9908, 0.7558, 0.5161, 0.9090, 0.3283], device='cuda:3'), in_proj_covar=tensor([0.0278, 0.0253, 0.0220, 0.0285, 0.0237, 0.0197, 0.0242, 0.0191], 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') 2023-03-26 03:18:06,351 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20445.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:18:08,882 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20449.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:18:21,895 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20468.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:18:29,628 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20479.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:18:31,929 INFO [finetune.py:976] (3/7) Epoch 4, batch 3300, loss[loss=0.2588, simple_loss=0.3182, pruned_loss=0.0997, over 4945.00 frames. ], tot_loss[loss=0.2319, simple_loss=0.2875, pruned_loss=0.08815, over 956713.20 frames. ], batch size: 38, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:19:05,035 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-03-26 03:19:09,459 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.164e+02 1.750e+02 2.039e+02 2.534e+02 4.074e+02, threshold=4.078e+02, percent-clipped=2.0 2023-03-26 03:19:15,668 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7921, 1.5834, 1.2991, 1.5602, 1.5744, 1.4848, 1.5111, 2.2998], device='cuda:3'), covar=tensor([0.8060, 0.8728, 0.6580, 0.8534, 0.7231, 0.4642, 0.7997, 0.2924], device='cuda:3'), in_proj_covar=tensor([0.0277, 0.0253, 0.0220, 0.0284, 0.0237, 0.0198, 0.0241, 0.0191], 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') 2023-03-26 03:19:29,492 INFO [finetune.py:976] (3/7) Epoch 4, batch 3350, loss[loss=0.2903, simple_loss=0.3228, pruned_loss=0.1289, over 4143.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.2903, pruned_loss=0.08893, over 955633.26 frames. ], batch size: 65, lr: 3.96e-03, grad_scale: 64.0 2023-03-26 03:19:40,753 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5322, 1.2742, 1.6159, 1.7043, 1.5144, 3.2443, 1.1136, 1.5207], device='cuda:3'), covar=tensor([0.1015, 0.1987, 0.1607, 0.1163, 0.1782, 0.0309, 0.1800, 0.1973], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0082, 0.0078, 0.0080, 0.0093, 0.0084, 0.0086, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 03:19:49,080 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2979, 1.8556, 2.6712, 1.8674, 2.3366, 2.4003, 1.8925, 2.5197], device='cuda:3'), covar=tensor([0.1044, 0.1678, 0.1227, 0.1708, 0.0656, 0.1208, 0.1906, 0.0707], device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0207, 0.0203, 0.0197, 0.0181, 0.0225, 0.0215, 0.0203], 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') 2023-03-26 03:19:59,031 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20560.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:20:17,266 INFO [finetune.py:976] (3/7) Epoch 4, batch 3400, loss[loss=0.2032, simple_loss=0.2696, pruned_loss=0.06843, over 4775.00 frames. ], tot_loss[loss=0.2344, simple_loss=0.2909, pruned_loss=0.08893, over 955791.40 frames. ], batch size: 51, lr: 3.96e-03, grad_scale: 64.0 2023-03-26 03:20:20,428 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20588.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:20:26,684 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-03-26 03:20:46,762 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.150e+02 1.735e+02 2.048e+02 2.538e+02 3.974e+02, threshold=4.096e+02, percent-clipped=0.0 2023-03-26 03:21:05,098 INFO [finetune.py:976] (3/7) Epoch 4, batch 3450, loss[loss=0.2338, simple_loss=0.2876, pruned_loss=0.09, over 4933.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.2917, pruned_loss=0.08904, over 957385.87 frames. ], batch size: 33, lr: 3.96e-03, grad_scale: 64.0 2023-03-26 03:21:22,517 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20649.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:21:36,679 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2052, 1.7341, 2.4447, 3.8033, 2.8424, 2.6246, 0.6359, 3.0943], device='cuda:3'), covar=tensor([0.1574, 0.1621, 0.1428, 0.0536, 0.0700, 0.1591, 0.2301, 0.0479], device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0120, 0.0138, 0.0168, 0.0105, 0.0144, 0.0131, 0.0106], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 03:21:45,699 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20676.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:21:51,862 INFO [finetune.py:976] (3/7) Epoch 4, batch 3500, loss[loss=0.2527, simple_loss=0.3083, pruned_loss=0.09849, over 4828.00 frames. ], tot_loss[loss=0.2322, simple_loss=0.2881, pruned_loss=0.08815, over 954527.35 frames. ], batch size: 39, lr: 3.96e-03, grad_scale: 64.0 2023-03-26 03:22:12,391 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6816, 1.4081, 1.9367, 3.1381, 2.1345, 2.2701, 0.9899, 2.5055], device='cuda:3'), covar=tensor([0.1927, 0.1919, 0.1641, 0.0712, 0.0978, 0.1787, 0.2244, 0.0753], device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0120, 0.0137, 0.0168, 0.0105, 0.0144, 0.0130, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 03:22:31,512 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.236e+02 1.688e+02 2.022e+02 2.523e+02 5.341e+02, threshold=4.043e+02, percent-clipped=2.0 2023-03-26 03:22:35,129 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=20724.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:22:42,637 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2023-03-26 03:22:43,663 INFO [finetune.py:976] (3/7) Epoch 4, batch 3550, loss[loss=0.2347, simple_loss=0.2801, pruned_loss=0.09468, over 4863.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.2856, pruned_loss=0.08788, over 956088.61 frames. ], batch size: 34, lr: 3.96e-03, grad_scale: 64.0 2023-03-26 03:22:56,118 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20744.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:22:56,774 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20745.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:23:24,107 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20768.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:23:33,949 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20774.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:23:41,428 INFO [finetune.py:976] (3/7) Epoch 4, batch 3600, loss[loss=0.2761, simple_loss=0.3238, pruned_loss=0.1142, over 4886.00 frames. ], tot_loss[loss=0.2294, simple_loss=0.2834, pruned_loss=0.08774, over 953198.00 frames. ], batch size: 34, lr: 3.96e-03, grad_scale: 64.0 2023-03-26 03:23:52,915 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=20793.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:24:03,900 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.05 vs. limit=5.0 2023-03-26 03:24:24,358 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=20816.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:24:31,498 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.828e+01 1.831e+02 2.152e+02 2.506e+02 5.159e+02, threshold=4.304e+02, percent-clipped=1.0 2023-03-26 03:24:50,040 INFO [finetune.py:976] (3/7) Epoch 4, batch 3650, loss[loss=0.2986, simple_loss=0.3391, pruned_loss=0.129, over 4906.00 frames. ], tot_loss[loss=0.2309, simple_loss=0.2858, pruned_loss=0.08802, over 953969.47 frames. ], batch size: 43, lr: 3.96e-03, grad_scale: 64.0 2023-03-26 03:25:24,801 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20860.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:25:52,754 INFO [finetune.py:976] (3/7) Epoch 4, batch 3700, loss[loss=0.2205, simple_loss=0.2812, pruned_loss=0.07994, over 4924.00 frames. ], tot_loss[loss=0.235, simple_loss=0.2899, pruned_loss=0.09004, over 952777.90 frames. ], batch size: 33, lr: 3.96e-03, grad_scale: 64.0 2023-03-26 03:26:17,161 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=20908.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:26:19,512 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0582, 1.6882, 1.7543, 1.8396, 1.6229, 1.7084, 1.8184, 1.7016], device='cuda:3'), covar=tensor([0.7906, 1.0971, 0.8789, 1.0067, 1.1142, 0.7862, 1.3725, 0.8067], device='cuda:3'), in_proj_covar=tensor([0.0228, 0.0251, 0.0256, 0.0261, 0.0241, 0.0217, 0.0276, 0.0221], 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') 2023-03-26 03:26:24,268 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.096e+02 1.710e+02 2.022e+02 2.626e+02 5.956e+02, threshold=4.044e+02, percent-clipped=2.0 2023-03-26 03:26:34,612 INFO [finetune.py:976] (3/7) Epoch 4, batch 3750, loss[loss=0.2721, simple_loss=0.3229, pruned_loss=0.1107, over 4814.00 frames. ], tot_loss[loss=0.2376, simple_loss=0.293, pruned_loss=0.09113, over 953353.22 frames. ], batch size: 39, lr: 3.96e-03, grad_scale: 64.0 2023-03-26 03:26:46,553 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20944.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:27:21,961 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9625, 1.8141, 1.4985, 1.9357, 2.0348, 1.6445, 2.3310, 1.9466], device='cuda:3'), covar=tensor([0.2023, 0.4039, 0.4482, 0.3804, 0.3129, 0.2199, 0.4401, 0.2513], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0196, 0.0238, 0.0257, 0.0227, 0.0188, 0.0212, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 03:27:33,087 INFO [finetune.py:976] (3/7) Epoch 4, batch 3800, loss[loss=0.1996, simple_loss=0.2619, pruned_loss=0.06859, over 4762.00 frames. ], tot_loss[loss=0.236, simple_loss=0.2919, pruned_loss=0.09007, over 953396.19 frames. ], batch size: 28, lr: 3.96e-03, grad_scale: 64.0 2023-03-26 03:28:01,108 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-26 03:28:14,228 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.257e+02 1.718e+02 1.983e+02 2.372e+02 3.493e+02, threshold=3.966e+02, percent-clipped=0.0 2023-03-26 03:28:29,875 INFO [finetune.py:976] (3/7) Epoch 4, batch 3850, loss[loss=0.2195, simple_loss=0.2748, pruned_loss=0.08207, over 4676.00 frames. ], tot_loss[loss=0.2335, simple_loss=0.29, pruned_loss=0.08844, over 954755.01 frames. ], batch size: 23, lr: 3.96e-03, grad_scale: 64.0 2023-03-26 03:28:37,705 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21044.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:28:45,302 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21052.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:29:05,451 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-03-26 03:29:08,438 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21074.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:29:19,704 INFO [finetune.py:976] (3/7) Epoch 4, batch 3900, loss[loss=0.2043, simple_loss=0.261, pruned_loss=0.0738, over 4770.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.2876, pruned_loss=0.08795, over 955507.07 frames. ], batch size: 28, lr: 3.96e-03, grad_scale: 64.0 2023-03-26 03:29:26,806 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=21092.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:29:43,718 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21113.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:29:46,793 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21118.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:29:47,848 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.038e+02 1.709e+02 1.984e+02 2.370e+02 6.134e+02, threshold=3.968e+02, percent-clipped=2.0 2023-03-26 03:29:49,148 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=21122.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:29:59,343 INFO [finetune.py:976] (3/7) Epoch 4, batch 3950, loss[loss=0.2011, simple_loss=0.2642, pruned_loss=0.06906, over 4832.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.2832, pruned_loss=0.08587, over 954647.27 frames. ], batch size: 33, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:30:47,945 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21179.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:30:50,289 INFO [finetune.py:976] (3/7) Epoch 4, batch 4000, loss[loss=0.222, simple_loss=0.2827, pruned_loss=0.08068, over 4916.00 frames. ], tot_loss[loss=0.2271, simple_loss=0.2823, pruned_loss=0.08595, over 955879.40 frames. ], batch size: 37, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:31:19,059 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.151e+02 1.740e+02 2.148e+02 2.427e+02 4.357e+02, threshold=4.296e+02, percent-clipped=1.0 2023-03-26 03:31:27,483 INFO [finetune.py:976] (3/7) Epoch 4, batch 4050, loss[loss=0.2398, simple_loss=0.3096, pruned_loss=0.08494, over 4813.00 frames. ], tot_loss[loss=0.2305, simple_loss=0.2862, pruned_loss=0.08739, over 956478.33 frames. ], batch size: 38, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:31:35,255 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21244.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:32:09,113 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-26 03:32:12,118 INFO [finetune.py:976] (3/7) Epoch 4, batch 4100, loss[loss=0.2334, simple_loss=0.2789, pruned_loss=0.09397, over 4058.00 frames. ], tot_loss[loss=0.2336, simple_loss=0.2897, pruned_loss=0.08876, over 955388.92 frames. ], batch size: 65, lr: 3.96e-03, grad_scale: 32.0 2023-03-26 03:32:23,704 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=21292.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:32:44,648 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21312.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:32:53,584 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.029e+02 1.825e+02 2.074e+02 2.571e+02 5.101e+02, threshold=4.147e+02, percent-clipped=2.0 2023-03-26 03:33:04,207 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21328.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:33:05,424 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 2023-03-26 03:33:07,023 INFO [finetune.py:976] (3/7) Epoch 4, batch 4150, loss[loss=0.228, simple_loss=0.2978, pruned_loss=0.07905, over 4789.00 frames. ], tot_loss[loss=0.2345, simple_loss=0.2908, pruned_loss=0.08908, over 955314.97 frames. ], batch size: 51, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:33:45,530 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21373.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:33:52,599 INFO [finetune.py:976] (3/7) Epoch 4, batch 4200, loss[loss=0.2149, simple_loss=0.269, pruned_loss=0.08035, over 4782.00 frames. ], tot_loss[loss=0.2354, simple_loss=0.2919, pruned_loss=0.08945, over 956151.87 frames. ], batch size: 26, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:34:02,010 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21389.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:34:21,944 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21408.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:34:32,322 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2654, 1.9875, 2.7317, 1.7278, 2.4120, 2.5195, 1.9631, 2.6724], device='cuda:3'), covar=tensor([0.1885, 0.2337, 0.1911, 0.2688, 0.0969, 0.1688, 0.2661, 0.1032], device='cuda:3'), in_proj_covar=tensor([0.0207, 0.0208, 0.0204, 0.0197, 0.0184, 0.0226, 0.0215, 0.0205], 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') 2023-03-26 03:34:39,316 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.013e+02 1.725e+02 2.024e+02 2.533e+02 3.913e+02, threshold=4.049e+02, percent-clipped=0.0 2023-03-26 03:34:50,756 INFO [finetune.py:976] (3/7) Epoch 4, batch 4250, loss[loss=0.2506, simple_loss=0.3018, pruned_loss=0.09977, over 4861.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.2885, pruned_loss=0.08782, over 954753.63 frames. ], batch size: 34, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:34:54,020 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1489, 1.9736, 1.6331, 2.0708, 1.9385, 1.8276, 1.8153, 2.8008], device='cuda:3'), covar=tensor([0.7536, 0.9804, 0.6509, 0.8702, 0.7990, 0.4522, 0.9390, 0.2701], device='cuda:3'), in_proj_covar=tensor([0.0279, 0.0254, 0.0220, 0.0283, 0.0237, 0.0198, 0.0241, 0.0192], 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') 2023-03-26 03:35:18,394 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21474.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:35:24,284 INFO [finetune.py:976] (3/7) Epoch 4, batch 4300, loss[loss=0.2024, simple_loss=0.2669, pruned_loss=0.06896, over 4804.00 frames. ], tot_loss[loss=0.2292, simple_loss=0.2849, pruned_loss=0.08673, over 955146.15 frames. ], batch size: 51, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:35:39,844 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.2031, 3.6656, 3.8614, 4.0534, 3.9706, 3.7318, 4.2787, 1.3808], device='cuda:3'), covar=tensor([0.0816, 0.0764, 0.0814, 0.0821, 0.1217, 0.1339, 0.0729, 0.5078], device='cuda:3'), in_proj_covar=tensor([0.0356, 0.0242, 0.0276, 0.0293, 0.0337, 0.0284, 0.0306, 0.0298], 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') 2023-03-26 03:35:53,225 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.244e+02 1.671e+02 2.065e+02 2.560e+02 4.445e+02, threshold=4.130e+02, percent-clipped=1.0 2023-03-26 03:36:01,103 INFO [finetune.py:976] (3/7) Epoch 4, batch 4350, loss[loss=0.2103, simple_loss=0.2625, pruned_loss=0.07899, over 4829.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.2818, pruned_loss=0.08548, over 955559.39 frames. ], batch size: 30, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:36:01,805 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3132, 1.3569, 1.3434, 1.4571, 1.5570, 2.9756, 1.3252, 1.5477], device='cuda:3'), covar=tensor([0.0983, 0.1703, 0.1173, 0.1052, 0.1518, 0.0271, 0.1429, 0.1550], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0082, 0.0078, 0.0080, 0.0093, 0.0083, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 03:36:04,123 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2671, 2.3410, 2.3052, 1.6262, 2.5942, 2.6170, 2.4538, 2.0133], device='cuda:3'), covar=tensor([0.0683, 0.0540, 0.0681, 0.0932, 0.0392, 0.0635, 0.0604, 0.0929], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0133, 0.0146, 0.0129, 0.0112, 0.0145, 0.0148, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 03:36:40,601 INFO [finetune.py:976] (3/7) Epoch 4, batch 4400, loss[loss=0.2979, simple_loss=0.3516, pruned_loss=0.1221, over 4805.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.2819, pruned_loss=0.08558, over 955124.15 frames. ], batch size: 41, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:36:48,988 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21595.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:37:05,948 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21613.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:37:15,011 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.167e+02 1.721e+02 2.036e+02 2.627e+02 4.967e+02, threshold=4.072e+02, percent-clipped=2.0 2023-03-26 03:37:22,898 INFO [finetune.py:976] (3/7) Epoch 4, batch 4450, loss[loss=0.2514, simple_loss=0.3126, pruned_loss=0.09513, over 4814.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.2858, pruned_loss=0.08715, over 955040.34 frames. ], batch size: 40, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:37:36,701 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2812, 2.0545, 2.1954, 0.9197, 2.3610, 2.5858, 2.1787, 2.0213], device='cuda:3'), covar=tensor([0.0926, 0.0711, 0.0423, 0.0748, 0.0334, 0.0446, 0.0373, 0.0520], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0156, 0.0117, 0.0135, 0.0131, 0.0120, 0.0145, 0.0144], device='cuda:3'), out_proj_covar=tensor([9.6595e-05, 1.1571e-04, 8.5092e-05, 9.9217e-05, 9.4447e-05, 8.8998e-05, 1.0829e-04, 1.0640e-04], device='cuda:3') 2023-03-26 03:37:38,565 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21656.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:37:51,471 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21668.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:38:01,190 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21674.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:38:09,621 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.5051, 3.9036, 4.0767, 4.3144, 4.2102, 3.9448, 4.6110, 1.4736], device='cuda:3'), covar=tensor([0.0835, 0.0832, 0.0869, 0.1062, 0.1274, 0.1501, 0.0632, 0.5430], device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0243, 0.0277, 0.0293, 0.0337, 0.0285, 0.0307, 0.0300], 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') 2023-03-26 03:38:12,008 INFO [finetune.py:976] (3/7) Epoch 4, batch 4500, loss[loss=0.1881, simple_loss=0.237, pruned_loss=0.06963, over 4231.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.2871, pruned_loss=0.08674, over 954399.66 frames. ], batch size: 18, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:38:12,675 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21684.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:38:31,684 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.6017, 2.5833, 2.5418, 1.3567, 2.7595, 2.4219, 1.9694, 2.2454], device='cuda:3'), covar=tensor([0.1117, 0.1494, 0.2236, 0.2797, 0.2032, 0.2213, 0.2811, 0.1707], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0201, 0.0203, 0.0191, 0.0219, 0.0210, 0.0220, 0.0201], 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') 2023-03-26 03:38:41,406 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21708.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:38:42,059 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7191, 0.9560, 1.5948, 1.4379, 1.3693, 1.3518, 1.3003, 1.4779], device='cuda:3'), covar=tensor([0.6128, 0.8695, 0.7025, 0.8128, 0.8756, 0.6234, 0.9978, 0.6228], device='cuda:3'), in_proj_covar=tensor([0.0227, 0.0251, 0.0255, 0.0261, 0.0240, 0.0217, 0.0276, 0.0221], 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') 2023-03-26 03:38:49,076 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.156e+02 1.721e+02 2.032e+02 2.543e+02 5.339e+02, threshold=4.063e+02, percent-clipped=3.0 2023-03-26 03:38:58,598 INFO [finetune.py:976] (3/7) Epoch 4, batch 4550, loss[loss=0.2003, simple_loss=0.267, pruned_loss=0.06679, over 4901.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.2879, pruned_loss=0.08718, over 955814.46 frames. ], batch size: 37, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:39:11,355 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7998, 1.6487, 1.4836, 1.5992, 1.8906, 1.5425, 2.0805, 1.7644], device='cuda:3'), covar=tensor([0.1735, 0.3151, 0.3868, 0.3292, 0.2925, 0.2057, 0.3285, 0.2336], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0193, 0.0235, 0.0253, 0.0224, 0.0186, 0.0209, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 03:39:14,117 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=21756.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:39:18,451 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21763.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:39:30,501 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21774.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:39:41,971 INFO [finetune.py:976] (3/7) Epoch 4, batch 4600, loss[loss=0.1905, simple_loss=0.2522, pruned_loss=0.0644, over 4829.00 frames. ], tot_loss[loss=0.2305, simple_loss=0.287, pruned_loss=0.08701, over 954875.92 frames. ], batch size: 33, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:40:25,680 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.219e+02 1.789e+02 2.167e+02 2.687e+02 4.147e+02, threshold=4.334e+02, percent-clipped=1.0 2023-03-26 03:40:32,331 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=21822.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:40:33,597 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21824.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:40:44,392 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8389, 1.6913, 2.1211, 1.4771, 1.8864, 2.0894, 1.7632, 2.2557], device='cuda:3'), covar=tensor([0.1331, 0.2037, 0.1152, 0.1683, 0.0861, 0.1309, 0.2158, 0.0796], device='cuda:3'), in_proj_covar=tensor([0.0204, 0.0204, 0.0201, 0.0194, 0.0182, 0.0223, 0.0213, 0.0202], 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') 2023-03-26 03:40:44,878 INFO [finetune.py:976] (3/7) Epoch 4, batch 4650, loss[loss=0.2118, simple_loss=0.2494, pruned_loss=0.08708, over 4064.00 frames. ], tot_loss[loss=0.2292, simple_loss=0.285, pruned_loss=0.08673, over 954473.51 frames. ], batch size: 17, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:41:06,181 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21850.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:41:09,255 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8195, 1.7801, 1.7662, 1.7747, 1.3767, 3.7531, 1.5195, 2.1469], device='cuda:3'), covar=tensor([0.3300, 0.2187, 0.1850, 0.2190, 0.1789, 0.0181, 0.2323, 0.1227], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0113, 0.0116, 0.0120, 0.0116, 0.0097, 0.0101, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 03:41:19,493 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8486, 1.6936, 1.4240, 1.5769, 1.5845, 1.5629, 1.5505, 2.3455], device='cuda:3'), covar=tensor([0.7708, 0.8049, 0.6117, 0.7686, 0.6829, 0.4642, 0.7782, 0.2723], device='cuda:3'), in_proj_covar=tensor([0.0280, 0.0255, 0.0220, 0.0285, 0.0238, 0.0199, 0.0242, 0.0193], 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') 2023-03-26 03:41:28,342 INFO [finetune.py:976] (3/7) Epoch 4, batch 4700, loss[loss=0.2232, simple_loss=0.2647, pruned_loss=0.09088, over 4713.00 frames. ], tot_loss[loss=0.226, simple_loss=0.2817, pruned_loss=0.08512, over 955072.63 frames. ], batch size: 59, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:41:33,177 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-26 03:41:43,347 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8202, 0.8839, 1.6502, 1.5410, 1.4296, 1.3735, 1.3540, 1.5260], device='cuda:3'), covar=tensor([0.5603, 0.8548, 0.7346, 0.7462, 0.8502, 0.6644, 0.9943, 0.6335], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0252, 0.0256, 0.0261, 0.0242, 0.0218, 0.0278, 0.0222], 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') 2023-03-26 03:42:02,069 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21911.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:42:08,439 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.035e+02 1.749e+02 2.081e+02 2.546e+02 7.973e+02, threshold=4.162e+02, percent-clipped=1.0 2023-03-26 03:42:16,629 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-03-26 03:42:16,912 INFO [finetune.py:976] (3/7) Epoch 4, batch 4750, loss[loss=0.1919, simple_loss=0.238, pruned_loss=0.07296, over 4376.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.2797, pruned_loss=0.0847, over 951622.81 frames. ], batch size: 19, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:42:28,926 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4338, 2.4906, 2.3372, 1.8215, 2.6598, 2.6093, 2.5438, 2.2077], device='cuda:3'), covar=tensor([0.0649, 0.0508, 0.0650, 0.0825, 0.0504, 0.0611, 0.0545, 0.0898], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0132, 0.0144, 0.0128, 0.0110, 0.0143, 0.0146, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 03:42:29,498 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21951.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:42:46,660 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21968.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:42:53,027 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21969.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:42:53,715 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21970.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:43:02,051 INFO [finetune.py:976] (3/7) Epoch 4, batch 4800, loss[loss=0.2489, simple_loss=0.291, pruned_loss=0.1034, over 4812.00 frames. ], tot_loss[loss=0.2271, simple_loss=0.2825, pruned_loss=0.08591, over 950251.66 frames. ], batch size: 25, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:43:02,774 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21984.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:43:15,955 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.0914, 2.5469, 2.4703, 1.2715, 2.5884, 2.1634, 1.8409, 2.3680], device='cuda:3'), covar=tensor([0.0890, 0.1117, 0.1724, 0.2625, 0.2023, 0.2559, 0.2676, 0.1336], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0204, 0.0206, 0.0194, 0.0222, 0.0212, 0.0224, 0.0203], 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') 2023-03-26 03:43:34,457 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=22016.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:43:37,302 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.143e+02 1.769e+02 1.987e+02 2.631e+02 5.032e+02, threshold=3.974e+02, percent-clipped=2.0 2023-03-26 03:43:42,978 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-26 03:43:44,131 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22031.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:43:44,661 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=22032.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:43:45,184 INFO [finetune.py:976] (3/7) Epoch 4, batch 4850, loss[loss=0.1892, simple_loss=0.2547, pruned_loss=0.06189, over 4926.00 frames. ], tot_loss[loss=0.2292, simple_loss=0.2849, pruned_loss=0.0868, over 948757.94 frames. ], batch size: 33, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:43:55,788 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.74 vs. limit=5.0 2023-03-26 03:44:19,578 INFO [finetune.py:976] (3/7) Epoch 4, batch 4900, loss[loss=0.2181, simple_loss=0.291, pruned_loss=0.07262, over 4888.00 frames. ], tot_loss[loss=0.2306, simple_loss=0.2866, pruned_loss=0.08723, over 950351.39 frames. ], batch size: 35, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:44:39,949 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6585, 1.4905, 1.5233, 1.5296, 0.9664, 3.6013, 1.3783, 1.9712], device='cuda:3'), covar=tensor([0.3346, 0.2427, 0.2026, 0.2383, 0.1979, 0.0174, 0.2671, 0.1329], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0112, 0.0116, 0.0120, 0.0116, 0.0097, 0.0101, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 03:44:41,736 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7873, 1.2448, 1.5942, 1.5948, 1.3583, 1.3689, 1.5620, 1.4117], device='cuda:3'), covar=tensor([0.6267, 0.9594, 0.7809, 0.8494, 0.9675, 0.7475, 1.0684, 0.7679], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0252, 0.0257, 0.0262, 0.0242, 0.0218, 0.0278, 0.0223], 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') 2023-03-26 03:44:49,810 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7967, 1.6075, 2.2968, 1.4675, 2.0087, 2.1506, 1.6662, 2.2271], device='cuda:3'), covar=tensor([0.1646, 0.2145, 0.1429, 0.2303, 0.0975, 0.1659, 0.2583, 0.0974], device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0205, 0.0203, 0.0196, 0.0183, 0.0224, 0.0215, 0.0204], 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') 2023-03-26 03:45:00,548 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22119.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:45:01,045 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.093e+02 1.764e+02 2.232e+02 2.515e+02 4.523e+02, threshold=4.464e+02, percent-clipped=3.0 2023-03-26 03:45:18,661 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22130.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:45:20,457 INFO [finetune.py:976] (3/7) Epoch 4, batch 4950, loss[loss=0.1851, simple_loss=0.2614, pruned_loss=0.05435, over 4785.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.2865, pruned_loss=0.08639, over 951326.50 frames. ], batch size: 29, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:45:53,668 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-26 03:46:02,235 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7329, 1.6797, 1.5517, 1.8220, 2.3108, 1.7752, 1.6675, 1.3600], device='cuda:3'), covar=tensor([0.2353, 0.2267, 0.1929, 0.1827, 0.1891, 0.1227, 0.2577, 0.1922], device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0209, 0.0198, 0.0185, 0.0236, 0.0175, 0.0215, 0.0188], 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') 2023-03-26 03:46:12,752 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22176.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:46:19,504 INFO [finetune.py:976] (3/7) Epoch 4, batch 5000, loss[loss=0.2256, simple_loss=0.2806, pruned_loss=0.08531, over 4909.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.285, pruned_loss=0.08593, over 953534.82 frames. ], batch size: 36, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:46:20,337 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.43 vs. limit=5.0 2023-03-26 03:46:22,147 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8561, 1.6817, 1.3718, 1.5912, 1.5613, 1.5292, 1.4890, 2.3532], device='cuda:3'), covar=tensor([0.7558, 0.8075, 0.6261, 0.7395, 0.6769, 0.4220, 0.7753, 0.2750], device='cuda:3'), in_proj_covar=tensor([0.0279, 0.0254, 0.0219, 0.0284, 0.0237, 0.0198, 0.0242, 0.0193], 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') 2023-03-26 03:46:24,513 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22191.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:46:35,050 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22206.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:46:37,157 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.58 vs. limit=5.0 2023-03-26 03:46:40,525 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22215.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:46:43,480 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.178e+02 1.598e+02 2.028e+02 2.482e+02 4.524e+02, threshold=4.056e+02, percent-clipped=1.0 2023-03-26 03:46:59,736 INFO [finetune.py:976] (3/7) Epoch 4, batch 5050, loss[loss=0.2289, simple_loss=0.285, pruned_loss=0.08644, over 4812.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.2834, pruned_loss=0.08581, over 954921.45 frames. ], batch size: 40, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:47:02,289 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22237.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:47:02,299 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22237.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:47:16,635 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22251.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:47:22,084 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7037, 1.4745, 1.3103, 1.0895, 1.4675, 1.4642, 1.4059, 2.0429], device='cuda:3'), covar=tensor([0.7673, 0.7646, 0.5954, 0.7158, 0.5966, 0.3991, 0.7033, 0.2835], device='cuda:3'), in_proj_covar=tensor([0.0279, 0.0254, 0.0219, 0.0283, 0.0236, 0.0198, 0.0241, 0.0193], 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') 2023-03-26 03:47:28,160 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22269.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:47:32,942 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22276.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:47:40,348 INFO [finetune.py:976] (3/7) Epoch 4, batch 5100, loss[loss=0.1861, simple_loss=0.2457, pruned_loss=0.06326, over 4811.00 frames. ], tot_loss[loss=0.2233, simple_loss=0.279, pruned_loss=0.08381, over 952336.45 frames. ], batch size: 25, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:47:44,819 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.68 vs. limit=5.0 2023-03-26 03:47:46,894 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3533, 1.4945, 1.4311, 1.5186, 1.5791, 3.0279, 1.4676, 1.6492], device='cuda:3'), covar=tensor([0.1027, 0.1731, 0.1115, 0.1083, 0.1615, 0.0287, 0.1406, 0.1632], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0082, 0.0078, 0.0081, 0.0093, 0.0084, 0.0086, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 03:47:49,897 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22298.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:47:50,454 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=22299.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:47:53,454 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7262, 1.5519, 2.1276, 1.3978, 1.8965, 2.0566, 1.5502, 2.0930], device='cuda:3'), covar=tensor([0.1397, 0.2167, 0.1285, 0.1949, 0.0953, 0.1513, 0.2643, 0.1002], device='cuda:3'), in_proj_covar=tensor([0.0205, 0.0205, 0.0202, 0.0196, 0.0183, 0.0224, 0.0215, 0.0204], 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') 2023-03-26 03:48:03,281 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=22317.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:48:05,003 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.872e+01 1.620e+02 1.853e+02 2.165e+02 3.345e+02, threshold=3.706e+02, percent-clipped=0.0 2023-03-26 03:48:08,736 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22326.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:48:13,376 INFO [finetune.py:976] (3/7) Epoch 4, batch 5150, loss[loss=0.2941, simple_loss=0.3377, pruned_loss=0.1252, over 4902.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.2805, pruned_loss=0.08556, over 951709.63 frames. ], batch size: 35, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:48:14,906 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-26 03:48:51,908 INFO [finetune.py:976] (3/7) Epoch 4, batch 5200, loss[loss=0.2643, simple_loss=0.3193, pruned_loss=0.1046, over 4850.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.2841, pruned_loss=0.08753, over 949836.52 frames. ], batch size: 44, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:48:53,218 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22385.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:49:26,405 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22419.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:49:26,883 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.279e+02 1.811e+02 2.155e+02 2.737e+02 4.498e+02, threshold=4.310e+02, percent-clipped=4.0 2023-03-26 03:49:40,159 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0474, 1.6468, 1.8891, 1.8726, 1.6520, 1.7035, 1.8270, 1.7566], device='cuda:3'), covar=tensor([0.7595, 1.0004, 0.7240, 0.9050, 0.9562, 0.7598, 1.1794, 0.7147], device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0252, 0.0257, 0.0261, 0.0242, 0.0219, 0.0278, 0.0223], 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') 2023-03-26 03:49:40,614 INFO [finetune.py:976] (3/7) Epoch 4, batch 5250, loss[loss=0.1674, simple_loss=0.2295, pruned_loss=0.05262, over 4889.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.2853, pruned_loss=0.08744, over 948558.29 frames. ], batch size: 32, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:49:54,903 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22446.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:50:18,143 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=22467.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:50:27,826 INFO [finetune.py:976] (3/7) Epoch 4, batch 5300, loss[loss=0.2158, simple_loss=0.2822, pruned_loss=0.07473, over 4834.00 frames. ], tot_loss[loss=0.2304, simple_loss=0.2863, pruned_loss=0.08723, over 950316.51 frames. ], batch size: 47, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:50:30,467 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22486.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:50:49,039 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22506.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:51:10,354 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.196e+02 1.644e+02 2.088e+02 2.525e+02 4.526e+02, threshold=4.176e+02, percent-clipped=1.0 2023-03-26 03:51:18,953 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-03-26 03:51:29,036 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22532.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:51:29,576 INFO [finetune.py:976] (3/7) Epoch 4, batch 5350, loss[loss=0.2372, simple_loss=0.2983, pruned_loss=0.08805, over 4891.00 frames. ], tot_loss[loss=0.2293, simple_loss=0.2858, pruned_loss=0.08641, over 950225.29 frames. ], batch size: 43, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:51:43,431 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=22554.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:52:04,997 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22571.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:52:20,638 INFO [finetune.py:976] (3/7) Epoch 4, batch 5400, loss[loss=0.2248, simple_loss=0.2831, pruned_loss=0.08327, over 4892.00 frames. ], tot_loss[loss=0.226, simple_loss=0.2825, pruned_loss=0.08474, over 951556.15 frames. ], batch size: 43, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:52:24,439 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-03-26 03:52:27,242 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22593.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:52:45,275 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.025e+02 1.598e+02 1.961e+02 2.261e+02 4.832e+02, threshold=3.922e+02, percent-clipped=2.0 2023-03-26 03:52:50,432 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22626.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:52:54,550 INFO [finetune.py:976] (3/7) Epoch 4, batch 5450, loss[loss=0.1941, simple_loss=0.2541, pruned_loss=0.06704, over 4898.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.28, pruned_loss=0.08411, over 952678.05 frames. ], batch size: 32, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:52:55,900 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5871, 2.2213, 1.9760, 1.0049, 2.2583, 2.0263, 1.7716, 2.1049], device='cuda:3'), covar=tensor([0.0804, 0.0979, 0.1750, 0.2342, 0.1639, 0.1853, 0.1964, 0.1180], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0202, 0.0205, 0.0192, 0.0220, 0.0211, 0.0221, 0.0201], 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') 2023-03-26 03:52:58,315 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22639.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:53:18,745 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.2126, 4.5523, 4.7131, 5.0559, 4.9116, 4.6931, 5.3571, 1.7858], device='cuda:3'), covar=tensor([0.0700, 0.0798, 0.0744, 0.0747, 0.1128, 0.1240, 0.0511, 0.4907], device='cuda:3'), in_proj_covar=tensor([0.0359, 0.0244, 0.0276, 0.0293, 0.0339, 0.0285, 0.0307, 0.0299], 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') 2023-03-26 03:53:19,963 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.8408, 3.3913, 3.5369, 3.7352, 3.6017, 3.4527, 3.9404, 1.2946], device='cuda:3'), covar=tensor([0.0917, 0.0883, 0.0878, 0.1056, 0.1314, 0.1400, 0.0833, 0.4982], device='cuda:3'), in_proj_covar=tensor([0.0359, 0.0244, 0.0276, 0.0293, 0.0339, 0.0285, 0.0307, 0.0300], 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') 2023-03-26 03:53:30,710 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=22674.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:53:37,508 INFO [finetune.py:976] (3/7) Epoch 4, batch 5500, loss[loss=0.2472, simple_loss=0.2925, pruned_loss=0.101, over 4741.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.2775, pruned_loss=0.08288, over 955473.11 frames. ], batch size: 59, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:53:48,402 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22700.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 03:53:49,656 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8091, 1.0241, 1.5647, 1.5768, 1.4045, 1.3860, 1.4703, 1.4565], device='cuda:3'), covar=tensor([0.5620, 0.8714, 0.7233, 0.7843, 0.8829, 0.6774, 0.9801, 0.6543], device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0252, 0.0257, 0.0261, 0.0241, 0.0219, 0.0277, 0.0223], 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') 2023-03-26 03:54:00,990 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.350e+02 1.787e+02 2.088e+02 2.580e+02 6.017e+02, threshold=4.176e+02, percent-clipped=5.0 2023-03-26 03:54:01,871 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3635, 1.2634, 1.2750, 1.2768, 0.7705, 2.1437, 0.6974, 1.2687], device='cuda:3'), covar=tensor([0.3357, 0.2402, 0.2113, 0.2406, 0.2128, 0.0396, 0.2969, 0.1439], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0112, 0.0117, 0.0120, 0.0116, 0.0097, 0.0101, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 03:54:16,194 INFO [finetune.py:976] (3/7) Epoch 4, batch 5550, loss[loss=0.2375, simple_loss=0.2834, pruned_loss=0.09574, over 4723.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.2799, pruned_loss=0.0839, over 956915.66 frames. ], batch size: 59, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:54:23,405 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22741.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:54:55,507 INFO [finetune.py:976] (3/7) Epoch 4, batch 5600, loss[loss=0.2504, simple_loss=0.3191, pruned_loss=0.09078, over 4831.00 frames. ], tot_loss[loss=0.2292, simple_loss=0.2859, pruned_loss=0.08627, over 956025.16 frames. ], batch size: 47, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:54:57,333 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22786.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:55:01,933 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22794.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:55:28,583 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.180e+02 1.750e+02 2.147e+02 2.459e+02 4.993e+02, threshold=4.295e+02, percent-clipped=1.0 2023-03-26 03:55:31,576 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22825.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:55:40,677 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22832.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:55:41,208 INFO [finetune.py:976] (3/7) Epoch 4, batch 5650, loss[loss=0.1979, simple_loss=0.2636, pruned_loss=0.06616, over 4712.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.2879, pruned_loss=0.08622, over 954539.48 frames. ], batch size: 23, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:55:41,971 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=22834.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:55:52,346 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0250, 1.8160, 1.4528, 2.0184, 2.0360, 1.7046, 2.3332, 1.9455], device='cuda:3'), covar=tensor([0.2083, 0.3729, 0.4813, 0.4005, 0.3398, 0.2277, 0.3813, 0.2755], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0193, 0.0236, 0.0253, 0.0225, 0.0187, 0.0209, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 03:56:06,188 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22855.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:56:21,394 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22871.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:56:31,135 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=22880.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:56:33,111 INFO [finetune.py:976] (3/7) Epoch 4, batch 5700, loss[loss=0.2401, simple_loss=0.2739, pruned_loss=0.1031, over 4166.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.2829, pruned_loss=0.08495, over 936409.04 frames. ], batch size: 18, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:56:34,949 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22886.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:56:41,775 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22893.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:57:20,678 INFO [finetune.py:976] (3/7) Epoch 5, batch 0, loss[loss=0.1896, simple_loss=0.2711, pruned_loss=0.0541, over 4793.00 frames. ], tot_loss[loss=0.1896, simple_loss=0.2711, pruned_loss=0.0541, over 4793.00 frames. ], batch size: 29, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:57:20,678 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 03:57:30,666 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5493, 1.2162, 1.4179, 1.2904, 1.7069, 1.6328, 1.5362, 1.3658], device='cuda:3'), covar=tensor([0.0312, 0.0362, 0.0508, 0.0336, 0.0284, 0.0376, 0.0279, 0.0412], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0115, 0.0138, 0.0120, 0.0105, 0.0101, 0.0092, 0.0110], device='cuda:3'), out_proj_covar=tensor([6.7570e-05, 9.0440e-05, 1.1130e-04, 9.4555e-05, 8.3660e-05, 7.5062e-05, 7.0309e-05, 8.6112e-05], device='cuda:3') 2023-03-26 03:57:37,527 INFO [finetune.py:1010] (3/7) Epoch 5, validation: loss=0.1701, simple_loss=0.2413, pruned_loss=0.0494, over 2265189.00 frames. 2023-03-26 03:57:37,528 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 03:57:48,821 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=22919.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:57:48,964 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-26 03:57:49,349 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.844e+01 1.616e+02 1.840e+02 2.309e+02 3.969e+02, threshold=3.680e+02, percent-clipped=0.0 2023-03-26 03:57:51,315 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9611, 1.7924, 1.4237, 1.7245, 1.7138, 1.6710, 1.6834, 2.4085], device='cuda:3'), covar=tensor([0.7001, 0.7700, 0.6050, 0.7876, 0.6587, 0.4150, 0.7391, 0.2649], device='cuda:3'), in_proj_covar=tensor([0.0278, 0.0254, 0.0220, 0.0284, 0.0236, 0.0198, 0.0241, 0.0194], 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') 2023-03-26 03:58:02,583 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=22941.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:58:29,018 INFO [finetune.py:976] (3/7) Epoch 5, batch 50, loss[loss=0.224, simple_loss=0.289, pruned_loss=0.0795, over 4819.00 frames. ], tot_loss[loss=0.2292, simple_loss=0.2871, pruned_loss=0.08562, over 216266.24 frames. ], batch size: 38, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:59:05,722 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22995.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 03:59:17,857 INFO [finetune.py:976] (3/7) Epoch 5, batch 100, loss[loss=0.1969, simple_loss=0.2517, pruned_loss=0.07108, over 4798.00 frames. ], tot_loss[loss=0.2236, simple_loss=0.2792, pruned_loss=0.08404, over 379844.78 frames. ], batch size: 45, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:59:23,735 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.085e+02 1.764e+02 2.029e+02 2.456e+02 6.922e+02, threshold=4.057e+02, percent-clipped=5.0 2023-03-26 03:59:36,982 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23041.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 03:59:45,167 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7660, 1.4283, 0.9323, 1.7320, 2.2320, 1.4243, 1.7214, 1.8051], device='cuda:3'), covar=tensor([0.1501, 0.1985, 0.2190, 0.1169, 0.1891, 0.1958, 0.1329, 0.1840], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0099, 0.0117, 0.0093, 0.0124, 0.0097, 0.0101, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 03:59:51,323 INFO [finetune.py:976] (3/7) Epoch 5, batch 150, loss[loss=0.2362, simple_loss=0.2777, pruned_loss=0.09733, over 4751.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2748, pruned_loss=0.08288, over 506859.74 frames. ], batch size: 23, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 03:59:57,982 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0653, 1.2677, 1.0350, 1.9791, 2.4811, 1.8318, 1.7049, 2.0511], device='cuda:3'), covar=tensor([0.1473, 0.2011, 0.2165, 0.1155, 0.1813, 0.2011, 0.1375, 0.1780], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0099, 0.0117, 0.0093, 0.0124, 0.0097, 0.0101, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 04:00:05,809 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7097, 1.8228, 1.7183, 1.0418, 2.0712, 1.8388, 1.7577, 1.6006], device='cuda:3'), covar=tensor([0.0677, 0.0657, 0.0761, 0.0995, 0.0508, 0.0784, 0.0712, 0.1073], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0133, 0.0144, 0.0129, 0.0111, 0.0144, 0.0148, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 04:00:08,778 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=23089.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:00:16,070 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3201, 2.0713, 1.6867, 0.7857, 1.8693, 1.8956, 1.6193, 1.9498], device='cuda:3'), covar=tensor([0.0786, 0.0695, 0.1216, 0.1792, 0.1088, 0.1957, 0.1995, 0.0781], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0201, 0.0203, 0.0190, 0.0217, 0.0209, 0.0221, 0.0200], 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') 2023-03-26 04:00:16,820 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9993, 1.7449, 2.5057, 1.6001, 2.1857, 2.4111, 1.7732, 2.5069], device='cuda:3'), covar=tensor([0.1640, 0.2282, 0.1263, 0.2201, 0.1030, 0.1768, 0.2667, 0.0914], device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0206, 0.0203, 0.0198, 0.0184, 0.0225, 0.0216, 0.0205], 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') 2023-03-26 04:00:23,412 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23108.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:00:30,573 INFO [finetune.py:976] (3/7) Epoch 5, batch 200, loss[loss=0.1644, simple_loss=0.225, pruned_loss=0.05195, over 4776.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2744, pruned_loss=0.08215, over 606686.45 frames. ], batch size: 28, lr: 3.95e-03, grad_scale: 64.0 2023-03-26 04:00:42,592 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.047e+02 1.662e+02 1.994e+02 2.595e+02 4.858e+02, threshold=3.989e+02, percent-clipped=3.0 2023-03-26 04:01:01,381 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23150.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:01:09,515 INFO [finetune.py:976] (3/7) Epoch 5, batch 250, loss[loss=0.1899, simple_loss=0.2445, pruned_loss=0.06762, over 4752.00 frames. ], tot_loss[loss=0.2244, simple_loss=0.2796, pruned_loss=0.08461, over 682377.27 frames. ], batch size: 23, lr: 3.95e-03, grad_scale: 64.0 2023-03-26 04:01:17,604 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.53 vs. limit=5.0 2023-03-26 04:01:18,066 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23169.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:01:31,046 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23181.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:01:37,462 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-26 04:02:00,760 INFO [finetune.py:976] (3/7) Epoch 5, batch 300, loss[loss=0.2195, simple_loss=0.2821, pruned_loss=0.07848, over 4872.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.2841, pruned_loss=0.08633, over 743439.14 frames. ], batch size: 34, lr: 3.95e-03, grad_scale: 64.0 2023-03-26 04:02:11,926 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23217.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:02:20,057 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.235e+02 1.736e+02 2.142e+02 2.597e+02 5.294e+02, threshold=4.284e+02, percent-clipped=3.0 2023-03-26 04:02:31,208 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23229.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:03:01,013 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-26 04:03:03,007 INFO [finetune.py:976] (3/7) Epoch 5, batch 350, loss[loss=0.2023, simple_loss=0.2639, pruned_loss=0.07035, over 4838.00 frames. ], tot_loss[loss=0.2304, simple_loss=0.2867, pruned_loss=0.08709, over 792257.16 frames. ], batch size: 47, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 04:03:20,896 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23278.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:03:34,250 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23290.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:03:41,694 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23295.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:03:51,625 INFO [finetune.py:976] (3/7) Epoch 5, batch 400, loss[loss=0.2409, simple_loss=0.3013, pruned_loss=0.09025, over 4914.00 frames. ], tot_loss[loss=0.231, simple_loss=0.2876, pruned_loss=0.08719, over 828431.22 frames. ], batch size: 33, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 04:03:58,184 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.903e+01 1.731e+02 2.116e+02 2.565e+02 5.981e+02, threshold=4.232e+02, percent-clipped=1.0 2023-03-26 04:04:13,595 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=23343.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:04:24,904 INFO [finetune.py:976] (3/7) Epoch 5, batch 450, loss[loss=0.2142, simple_loss=0.2732, pruned_loss=0.07763, over 4756.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.2856, pruned_loss=0.08631, over 857497.11 frames. ], batch size: 26, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 04:04:26,810 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9835, 1.8261, 1.5137, 1.8957, 2.0067, 1.6659, 2.3296, 1.9549], device='cuda:3'), covar=tensor([0.1629, 0.3356, 0.3823, 0.3463, 0.2822, 0.1834, 0.3770, 0.2229], device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0192, 0.0235, 0.0252, 0.0224, 0.0186, 0.0208, 0.0187], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 04:05:10,554 INFO [finetune.py:976] (3/7) Epoch 5, batch 500, loss[loss=0.2481, simple_loss=0.2869, pruned_loss=0.1047, over 4830.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.2823, pruned_loss=0.085, over 878665.60 frames. ], batch size: 41, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 04:05:11,116 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.81 vs. limit=5.0 2023-03-26 04:05:16,627 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.006e+02 1.775e+02 2.029e+02 2.615e+02 5.539e+02, threshold=4.057e+02, percent-clipped=1.0 2023-03-26 04:05:42,938 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23450.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:05:52,277 INFO [finetune.py:976] (3/7) Epoch 5, batch 550, loss[loss=0.2168, simple_loss=0.2708, pruned_loss=0.08135, over 4892.00 frames. ], tot_loss[loss=0.2214, simple_loss=0.2773, pruned_loss=0.08275, over 895405.23 frames. ], batch size: 32, lr: 3.95e-03, grad_scale: 32.0 2023-03-26 04:05:52,993 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5281, 1.4556, 1.7979, 1.9364, 1.5852, 3.2021, 1.2042, 1.5166], device='cuda:3'), covar=tensor([0.0929, 0.1679, 0.1344, 0.0946, 0.1583, 0.0250, 0.1500, 0.1677], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0082, 0.0078, 0.0080, 0.0093, 0.0084, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 04:05:54,204 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23464.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:06:15,691 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23481.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:06:23,748 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23488.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:06:30,291 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=23498.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:06:31,649 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-26 04:06:38,164 INFO [finetune.py:976] (3/7) Epoch 5, batch 600, loss[loss=0.2003, simple_loss=0.2611, pruned_loss=0.06978, over 4911.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.2769, pruned_loss=0.08201, over 909516.50 frames. ], batch size: 36, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:06:44,765 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.135e+02 1.759e+02 2.043e+02 2.434e+02 4.744e+02, threshold=4.086e+02, percent-clipped=2.0 2023-03-26 04:06:51,177 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=23529.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:07:18,737 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23549.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 04:07:25,882 INFO [finetune.py:976] (3/7) Epoch 5, batch 650, loss[loss=0.2019, simple_loss=0.2584, pruned_loss=0.07272, over 4771.00 frames. ], tot_loss[loss=0.2236, simple_loss=0.2802, pruned_loss=0.08352, over 916038.64 frames. ], batch size: 27, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:07:33,759 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23573.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:07:43,056 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23585.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:08:14,419 INFO [finetune.py:976] (3/7) Epoch 5, batch 700, loss[loss=0.2417, simple_loss=0.293, pruned_loss=0.09519, over 4870.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.2833, pruned_loss=0.08492, over 924608.16 frames. ], batch size: 31, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:08:30,908 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.053e+02 1.701e+02 2.127e+02 2.576e+02 5.648e+02, threshold=4.253e+02, percent-clipped=2.0 2023-03-26 04:09:25,209 INFO [finetune.py:976] (3/7) Epoch 5, batch 750, loss[loss=0.2416, simple_loss=0.3033, pruned_loss=0.08992, over 4844.00 frames. ], tot_loss[loss=0.2271, simple_loss=0.2841, pruned_loss=0.08511, over 930986.76 frames. ], batch size: 44, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:09:29,990 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2023-03-26 04:09:36,346 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.04 vs. limit=2.0 2023-03-26 04:10:02,096 INFO [finetune.py:976] (3/7) Epoch 5, batch 800, loss[loss=0.2218, simple_loss=0.285, pruned_loss=0.07933, over 4896.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.285, pruned_loss=0.08527, over 938052.14 frames. ], batch size: 37, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:10:08,708 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.149e+02 1.635e+02 1.954e+02 2.429e+02 4.773e+02, threshold=3.908e+02, percent-clipped=1.0 2023-03-26 04:10:20,872 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-03-26 04:10:56,824 INFO [finetune.py:976] (3/7) Epoch 5, batch 850, loss[loss=0.1959, simple_loss=0.2672, pruned_loss=0.06226, over 4773.00 frames. ], tot_loss[loss=0.2254, simple_loss=0.2828, pruned_loss=0.08398, over 943161.53 frames. ], batch size: 26, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:11:04,280 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23764.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:11:54,658 INFO [finetune.py:976] (3/7) Epoch 5, batch 900, loss[loss=0.2418, simple_loss=0.2994, pruned_loss=0.09215, over 4727.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.2811, pruned_loss=0.08376, over 947050.33 frames. ], batch size: 23, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:11:55,340 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=23812.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:11:59,621 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4423, 2.9408, 2.8120, 1.2898, 3.0133, 2.1441, 0.6776, 1.9059], device='cuda:3'), covar=tensor([0.2169, 0.1911, 0.1832, 0.3343, 0.1282, 0.1192, 0.3894, 0.1566], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0170, 0.0163, 0.0128, 0.0155, 0.0122, 0.0145, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 04:12:00,783 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.157e+02 1.642e+02 1.957e+02 2.389e+02 4.840e+02, threshold=3.913e+02, percent-clipped=2.0 2023-03-26 04:12:21,815 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23844.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 04:12:37,545 INFO [finetune.py:976] (3/7) Epoch 5, batch 950, loss[loss=0.2146, simple_loss=0.2596, pruned_loss=0.0848, over 4038.00 frames. ], tot_loss[loss=0.2221, simple_loss=0.279, pruned_loss=0.08263, over 950529.29 frames. ], batch size: 65, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:12:44,903 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23873.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:12:52,631 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23885.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:13:28,831 INFO [finetune.py:976] (3/7) Epoch 5, batch 1000, loss[loss=0.1966, simple_loss=0.2635, pruned_loss=0.06483, over 4821.00 frames. ], tot_loss[loss=0.2235, simple_loss=0.2809, pruned_loss=0.08307, over 952443.32 frames. ], batch size: 25, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:13:38,585 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.160e+02 1.702e+02 2.066e+02 2.385e+02 5.722e+02, threshold=4.131e+02, percent-clipped=3.0 2023-03-26 04:13:38,656 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=23921.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:13:47,709 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=23933.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:14:14,857 INFO [finetune.py:976] (3/7) Epoch 5, batch 1050, loss[loss=0.2122, simple_loss=0.2735, pruned_loss=0.07545, over 4757.00 frames. ], tot_loss[loss=0.2231, simple_loss=0.2814, pruned_loss=0.08235, over 951675.82 frames. ], batch size: 26, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:14:17,902 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0879, 1.4795, 0.9748, 1.9688, 2.3302, 1.5776, 1.9980, 1.8467], device='cuda:3'), covar=tensor([0.1591, 0.2199, 0.2513, 0.1286, 0.2050, 0.2214, 0.1394, 0.2251], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0100, 0.0118, 0.0094, 0.0126, 0.0099, 0.0102, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-26 04:14:58,018 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6681, 1.5327, 1.3743, 1.6417, 2.0126, 1.6409, 1.2027, 1.3709], device='cuda:3'), covar=tensor([0.2498, 0.2385, 0.2136, 0.2045, 0.1940, 0.1339, 0.3100, 0.2025], device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0210, 0.0199, 0.0185, 0.0236, 0.0175, 0.0216, 0.0188], 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') 2023-03-26 04:15:19,195 INFO [finetune.py:976] (3/7) Epoch 5, batch 1100, loss[loss=0.2409, simple_loss=0.2988, pruned_loss=0.09149, over 4913.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.2824, pruned_loss=0.08267, over 950426.88 frames. ], batch size: 36, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:15:28,406 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.244e+02 1.825e+02 2.109e+02 2.589e+02 5.024e+02, threshold=4.219e+02, percent-clipped=4.0 2023-03-26 04:15:38,233 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24037.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:15:42,280 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.2983, 4.5188, 4.7733, 5.1717, 4.9191, 4.6841, 5.4069, 1.6418], device='cuda:3'), covar=tensor([0.0750, 0.0889, 0.0626, 0.0743, 0.1448, 0.1566, 0.0464, 0.5392], device='cuda:3'), in_proj_covar=tensor([0.0357, 0.0243, 0.0274, 0.0290, 0.0337, 0.0284, 0.0305, 0.0298], 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') 2023-03-26 04:15:54,485 INFO [finetune.py:976] (3/7) Epoch 5, batch 1150, loss[loss=0.207, simple_loss=0.2661, pruned_loss=0.07396, over 4807.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.2843, pruned_loss=0.08395, over 951759.72 frames. ], batch size: 33, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:16:04,916 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.20 vs. limit=5.0 2023-03-26 04:16:18,919 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24098.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:16:28,019 INFO [finetune.py:976] (3/7) Epoch 5, batch 1200, loss[loss=0.2574, simple_loss=0.3092, pruned_loss=0.1029, over 4873.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.2832, pruned_loss=0.08411, over 951195.00 frames. ], batch size: 34, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:16:37,230 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.193e+02 1.721e+02 2.129e+02 2.606e+02 7.150e+02, threshold=4.257e+02, percent-clipped=3.0 2023-03-26 04:16:51,890 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=24144.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 04:17:03,454 INFO [finetune.py:976] (3/7) Epoch 5, batch 1250, loss[loss=0.206, simple_loss=0.2679, pruned_loss=0.07209, over 4780.00 frames. ], tot_loss[loss=0.2231, simple_loss=0.2804, pruned_loss=0.08289, over 952326.42 frames. ], batch size: 28, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:17:17,594 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-03-26 04:17:28,511 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=24192.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:17:31,185 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.26 vs. limit=5.0 2023-03-26 04:17:42,423 INFO [finetune.py:976] (3/7) Epoch 5, batch 1300, loss[loss=0.19, simple_loss=0.2504, pruned_loss=0.06476, over 4792.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2767, pruned_loss=0.08082, over 953020.27 frames. ], batch size: 51, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:17:56,617 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.098e+02 1.607e+02 1.851e+02 2.364e+02 3.844e+02, threshold=3.702e+02, percent-clipped=0.0 2023-03-26 04:18:34,752 INFO [finetune.py:976] (3/7) Epoch 5, batch 1350, loss[loss=0.2211, simple_loss=0.2736, pruned_loss=0.08435, over 4720.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2761, pruned_loss=0.08064, over 952773.02 frames. ], batch size: 23, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:19:02,006 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1196, 1.9235, 1.5199, 2.1077, 1.8682, 1.7561, 1.7078, 2.8975], device='cuda:3'), covar=tensor([0.7629, 0.8901, 0.6535, 0.8651, 0.7351, 0.4504, 0.8628, 0.2683], device='cuda:3'), in_proj_covar=tensor([0.0280, 0.0255, 0.0220, 0.0283, 0.0236, 0.0199, 0.0242, 0.0195], 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') 2023-03-26 04:19:12,805 INFO [finetune.py:976] (3/7) Epoch 5, batch 1400, loss[loss=0.2514, simple_loss=0.3126, pruned_loss=0.09511, over 4874.00 frames. ], tot_loss[loss=0.2217, simple_loss=0.2792, pruned_loss=0.0821, over 951251.04 frames. ], batch size: 34, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:19:21,620 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.173e+02 1.714e+02 2.138e+02 2.571e+02 4.877e+02, threshold=4.276e+02, percent-clipped=6.0 2023-03-26 04:19:46,852 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-26 04:19:56,385 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0731, 2.2995, 1.9880, 1.4037, 2.2816, 2.2577, 2.1866, 1.8699], device='cuda:3'), covar=tensor([0.0712, 0.0554, 0.0817, 0.1002, 0.0506, 0.0724, 0.0702, 0.1013], device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0134, 0.0146, 0.0129, 0.0112, 0.0145, 0.0148, 0.0164], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 04:19:56,901 INFO [finetune.py:976] (3/7) Epoch 5, batch 1450, loss[loss=0.2316, simple_loss=0.2909, pruned_loss=0.0862, over 4828.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.2819, pruned_loss=0.08335, over 952171.91 frames. ], batch size: 30, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:20:20,193 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24393.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:20:31,289 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.3101, 1.4798, 1.6813, 0.8749, 1.4541, 1.8412, 1.7692, 1.4649], device='cuda:3'), covar=tensor([0.0893, 0.0571, 0.0447, 0.0535, 0.0399, 0.0523, 0.0327, 0.0631], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0159, 0.0119, 0.0136, 0.0133, 0.0123, 0.0148, 0.0145], device='cuda:3'), out_proj_covar=tensor([9.8781e-05, 1.1823e-04, 8.6730e-05, 9.9604e-05, 9.6326e-05, 9.1130e-05, 1.0988e-04, 1.0759e-04], device='cuda:3') 2023-03-26 04:20:31,769 INFO [finetune.py:976] (3/7) Epoch 5, batch 1500, loss[loss=0.235, simple_loss=0.289, pruned_loss=0.09044, over 4771.00 frames. ], tot_loss[loss=0.2271, simple_loss=0.2847, pruned_loss=0.08478, over 953809.31 frames. ], batch size: 26, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:20:38,327 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.023e+02 1.776e+02 2.138e+02 2.564e+02 4.291e+02, threshold=4.276e+02, percent-clipped=1.0 2023-03-26 04:20:59,126 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.4436, 4.0273, 4.1559, 4.1376, 3.9692, 3.7936, 4.5643, 1.5420], device='cuda:3'), covar=tensor([0.1201, 0.1465, 0.1476, 0.1812, 0.1969, 0.2110, 0.0965, 0.7104], device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0244, 0.0273, 0.0291, 0.0337, 0.0284, 0.0305, 0.0298], 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') 2023-03-26 04:21:13,454 INFO [finetune.py:976] (3/7) Epoch 5, batch 1550, loss[loss=0.2132, simple_loss=0.2828, pruned_loss=0.07178, over 4769.00 frames. ], tot_loss[loss=0.2273, simple_loss=0.2852, pruned_loss=0.08469, over 954325.34 frames. ], batch size: 26, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:21:29,422 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-03-26 04:21:47,121 INFO [finetune.py:976] (3/7) Epoch 5, batch 1600, loss[loss=0.2064, simple_loss=0.2702, pruned_loss=0.07127, over 4819.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.2824, pruned_loss=0.08312, over 954023.05 frames. ], batch size: 41, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:21:58,798 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.205e+02 1.770e+02 2.018e+02 2.552e+02 5.194e+02, threshold=4.037e+02, percent-clipped=4.0 2023-03-26 04:22:13,488 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-03-26 04:22:21,116 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24547.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:22:21,990 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.36 vs. limit=5.0 2023-03-26 04:22:33,860 INFO [finetune.py:976] (3/7) Epoch 5, batch 1650, loss[loss=0.2168, simple_loss=0.2804, pruned_loss=0.07659, over 4701.00 frames. ], tot_loss[loss=0.2225, simple_loss=0.2798, pruned_loss=0.08259, over 955913.94 frames. ], batch size: 23, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:22:40,462 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-26 04:22:43,191 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24569.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:22:57,353 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.01 vs. limit=5.0 2023-03-26 04:23:17,410 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24608.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:23:22,440 INFO [finetune.py:976] (3/7) Epoch 5, batch 1700, loss[loss=0.2649, simple_loss=0.32, pruned_loss=0.105, over 4831.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.2773, pruned_loss=0.08144, over 955200.83 frames. ], batch size: 39, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:23:31,257 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.092e+02 1.673e+02 1.915e+02 2.251e+02 4.027e+02, threshold=3.830e+02, percent-clipped=0.0 2023-03-26 04:23:42,118 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24630.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:24:06,425 INFO [finetune.py:976] (3/7) Epoch 5, batch 1750, loss[loss=0.3291, simple_loss=0.3603, pruned_loss=0.149, over 4768.00 frames. ], tot_loss[loss=0.2236, simple_loss=0.2804, pruned_loss=0.08342, over 956111.04 frames. ], batch size: 59, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:24:22,662 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6805, 1.4628, 1.4999, 1.5840, 0.8975, 3.3269, 1.2255, 1.7970], device='cuda:3'), covar=tensor([0.3371, 0.2489, 0.2181, 0.2428, 0.2094, 0.0171, 0.2728, 0.1343], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0113, 0.0116, 0.0121, 0.0116, 0.0097, 0.0100, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 04:24:28,022 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=24693.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:24:34,569 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.8524, 3.3111, 3.5068, 3.7253, 3.5750, 3.3924, 3.9234, 1.2959], device='cuda:3'), covar=tensor([0.0954, 0.1031, 0.1080, 0.1155, 0.1514, 0.1679, 0.0912, 0.5101], device='cuda:3'), in_proj_covar=tensor([0.0360, 0.0245, 0.0276, 0.0294, 0.0340, 0.0286, 0.0306, 0.0300], 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') 2023-03-26 04:24:39,307 INFO [finetune.py:976] (3/7) Epoch 5, batch 1800, loss[loss=0.2603, simple_loss=0.3201, pruned_loss=0.1002, over 4862.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.2833, pruned_loss=0.084, over 957626.12 frames. ], batch size: 34, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:24:45,828 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.243e+02 1.800e+02 2.166e+02 2.491e+02 4.201e+02, threshold=4.331e+02, percent-clipped=2.0 2023-03-26 04:24:59,915 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=24741.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:25:06,987 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8050, 1.2307, 0.9220, 1.5774, 1.9991, 1.2826, 1.5699, 1.7303], device='cuda:3'), covar=tensor([0.1488, 0.2031, 0.2076, 0.1181, 0.2080, 0.2070, 0.1348, 0.2019], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0099, 0.0117, 0.0093, 0.0124, 0.0097, 0.0101, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 04:25:12,943 INFO [finetune.py:976] (3/7) Epoch 5, batch 1850, loss[loss=0.2231, simple_loss=0.2904, pruned_loss=0.07785, over 4818.00 frames. ], tot_loss[loss=0.228, simple_loss=0.2857, pruned_loss=0.08515, over 955922.30 frames. ], batch size: 39, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:25:15,498 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24765.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:25:46,396 INFO [finetune.py:976] (3/7) Epoch 5, batch 1900, loss[loss=0.2276, simple_loss=0.2861, pruned_loss=0.08452, over 4860.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.2865, pruned_loss=0.08512, over 955280.12 frames. ], batch size: 31, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:25:52,457 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.150e+02 1.802e+02 2.061e+02 2.489e+02 6.200e+02, threshold=4.122e+02, percent-clipped=1.0 2023-03-26 04:25:57,986 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24826.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:26:11,674 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4435, 1.4943, 1.4346, 0.7517, 1.6894, 1.4854, 1.4822, 1.3906], device='cuda:3'), covar=tensor([0.0696, 0.0724, 0.0748, 0.1005, 0.0679, 0.0821, 0.0674, 0.1279], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0133, 0.0145, 0.0128, 0.0112, 0.0144, 0.0147, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 04:26:20,828 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24848.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:26:29,634 INFO [finetune.py:976] (3/7) Epoch 5, batch 1950, loss[loss=0.2286, simple_loss=0.2881, pruned_loss=0.08456, over 4925.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.2842, pruned_loss=0.08379, over 956089.89 frames. ], batch size: 38, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:26:31,558 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3597, 2.1672, 1.6205, 0.7403, 1.8350, 1.8682, 1.6574, 1.9078], device='cuda:3'), covar=tensor([0.0852, 0.0853, 0.1648, 0.2306, 0.1618, 0.2597, 0.2298, 0.0973], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0201, 0.0204, 0.0191, 0.0217, 0.0209, 0.0221, 0.0200], 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') 2023-03-26 04:26:50,652 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9765, 1.8048, 2.3783, 1.4535, 2.1857, 2.2220, 1.7615, 2.4489], device='cuda:3'), covar=tensor([0.1621, 0.2170, 0.1685, 0.2405, 0.1022, 0.1825, 0.2768, 0.0970], device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0206, 0.0202, 0.0196, 0.0183, 0.0223, 0.0216, 0.0203], 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') 2023-03-26 04:26:57,652 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24903.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:27:01,828 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24909.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:27:02,942 INFO [finetune.py:976] (3/7) Epoch 5, batch 2000, loss[loss=0.1893, simple_loss=0.2584, pruned_loss=0.06012, over 4792.00 frames. ], tot_loss[loss=0.2229, simple_loss=0.2811, pruned_loss=0.08236, over 955829.79 frames. ], batch size: 29, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:27:13,004 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.094e+02 1.611e+02 2.012e+02 2.424e+02 3.709e+02, threshold=4.024e+02, percent-clipped=0.0 2023-03-26 04:27:15,550 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24925.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:27:50,325 INFO [finetune.py:976] (3/7) Epoch 5, batch 2050, loss[loss=0.231, simple_loss=0.277, pruned_loss=0.09254, over 4757.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2778, pruned_loss=0.08106, over 955774.36 frames. ], batch size: 26, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:28:12,190 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24995.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:28:23,337 INFO [finetune.py:976] (3/7) Epoch 5, batch 2100, loss[loss=0.1906, simple_loss=0.2532, pruned_loss=0.06406, over 4926.00 frames. ], tot_loss[loss=0.219, simple_loss=0.2765, pruned_loss=0.08068, over 955364.23 frames. ], batch size: 33, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:28:39,037 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.247e+01 1.663e+02 1.989e+02 2.457e+02 4.446e+02, threshold=3.978e+02, percent-clipped=1.0 2023-03-26 04:29:08,279 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25056.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 04:29:11,159 INFO [finetune.py:976] (3/7) Epoch 5, batch 2150, loss[loss=0.286, simple_loss=0.3294, pruned_loss=0.1213, over 4825.00 frames. ], tot_loss[loss=0.2217, simple_loss=0.2793, pruned_loss=0.08208, over 954760.98 frames. ], batch size: 40, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:29:45,150 INFO [finetune.py:976] (3/7) Epoch 5, batch 2200, loss[loss=0.2816, simple_loss=0.3306, pruned_loss=0.1164, over 4858.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.2823, pruned_loss=0.08278, over 955354.40 frames. ], batch size: 31, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:29:52,263 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 1.691e+02 1.983e+02 2.301e+02 4.176e+02, threshold=3.967e+02, percent-clipped=1.0 2023-03-26 04:29:52,343 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25121.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:30:18,608 INFO [finetune.py:976] (3/7) Epoch 5, batch 2250, loss[loss=0.2087, simple_loss=0.2879, pruned_loss=0.06473, over 4848.00 frames. ], tot_loss[loss=0.2251, simple_loss=0.2838, pruned_loss=0.08322, over 956379.96 frames. ], batch size: 49, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:30:46,355 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25203.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:30:47,787 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25204.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:30:53,042 INFO [finetune.py:976] (3/7) Epoch 5, batch 2300, loss[loss=0.1989, simple_loss=0.2551, pruned_loss=0.07131, over 4787.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.2848, pruned_loss=0.08354, over 954297.36 frames. ], batch size: 25, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:31:05,177 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.234e+02 1.840e+02 2.117e+02 2.638e+02 5.911e+02, threshold=4.234e+02, percent-clipped=5.0 2023-03-26 04:31:14,021 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25225.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:31:35,994 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=25251.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:31:42,890 INFO [finetune.py:976] (3/7) Epoch 5, batch 2350, loss[loss=0.1951, simple_loss=0.2525, pruned_loss=0.0688, over 4796.00 frames. ], tot_loss[loss=0.2226, simple_loss=0.2814, pruned_loss=0.08188, over 952813.67 frames. ], batch size: 29, lr: 3.94e-03, grad_scale: 64.0 2023-03-26 04:31:51,255 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=25273.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:32:00,142 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6291, 1.4544, 1.4875, 1.6213, 0.8817, 3.2514, 1.2460, 1.7035], device='cuda:3'), covar=tensor([0.3407, 0.2375, 0.2137, 0.2276, 0.2147, 0.0206, 0.2829, 0.1401], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0114, 0.0118, 0.0122, 0.0118, 0.0098, 0.0101, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 04:32:16,768 INFO [finetune.py:976] (3/7) Epoch 5, batch 2400, loss[loss=0.2143, simple_loss=0.2746, pruned_loss=0.07699, over 4895.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.2781, pruned_loss=0.08108, over 954456.62 frames. ], batch size: 32, lr: 3.94e-03, grad_scale: 64.0 2023-03-26 04:32:23,858 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.140e+02 1.631e+02 1.900e+02 2.318e+02 5.058e+02, threshold=3.799e+02, percent-clipped=1.0 2023-03-26 04:32:28,455 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4303, 1.9434, 1.6210, 0.6411, 1.8513, 1.9421, 1.4662, 1.8265], device='cuda:3'), covar=tensor([0.0644, 0.1186, 0.1564, 0.2355, 0.1335, 0.1868, 0.2354, 0.1069], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0200, 0.0203, 0.0189, 0.0216, 0.0209, 0.0220, 0.0199], 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') 2023-03-26 04:32:58,160 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25351.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 04:33:04,187 INFO [finetune.py:976] (3/7) Epoch 5, batch 2450, loss[loss=0.2669, simple_loss=0.3074, pruned_loss=0.1131, over 4791.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2754, pruned_loss=0.08078, over 953354.38 frames. ], batch size: 29, lr: 3.94e-03, grad_scale: 64.0 2023-03-26 04:33:06,050 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8269, 1.6345, 2.1758, 3.3908, 2.4457, 2.3888, 0.9022, 2.7558], device='cuda:3'), covar=tensor([0.1690, 0.1431, 0.1303, 0.0512, 0.0720, 0.1470, 0.2042, 0.0608], device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0119, 0.0136, 0.0166, 0.0102, 0.0143, 0.0128, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 04:33:14,793 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0450, 1.9815, 1.7132, 2.1178, 1.9637, 1.8447, 1.7981, 2.7499], device='cuda:3'), covar=tensor([0.6464, 0.8412, 0.5275, 0.7631, 0.7081, 0.3787, 0.8245, 0.2315], device='cuda:3'), in_proj_covar=tensor([0.0280, 0.0254, 0.0221, 0.0283, 0.0237, 0.0198, 0.0242, 0.0196], 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') 2023-03-26 04:33:35,618 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5119, 0.9461, 0.8059, 1.3948, 2.0117, 0.7132, 1.2549, 1.4583], device='cuda:3'), covar=tensor([0.1383, 0.2009, 0.1663, 0.1048, 0.1601, 0.1760, 0.1363, 0.1720], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0098, 0.0116, 0.0093, 0.0123, 0.0096, 0.0101, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 04:34:02,098 INFO [finetune.py:976] (3/7) Epoch 5, batch 2500, loss[loss=0.1936, simple_loss=0.2642, pruned_loss=0.06147, over 4906.00 frames. ], tot_loss[loss=0.2174, simple_loss=0.2748, pruned_loss=0.07999, over 954003.26 frames. ], batch size: 37, lr: 3.94e-03, grad_scale: 64.0 2023-03-26 04:34:18,834 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.080e+02 1.793e+02 2.115e+02 2.620e+02 5.379e+02, threshold=4.229e+02, percent-clipped=6.0 2023-03-26 04:34:18,926 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25421.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:34:26,166 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25428.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 04:34:27,811 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-03-26 04:34:47,670 INFO [finetune.py:976] (3/7) Epoch 5, batch 2550, loss[loss=0.19, simple_loss=0.2494, pruned_loss=0.06531, over 4723.00 frames. ], tot_loss[loss=0.2212, simple_loss=0.2789, pruned_loss=0.0817, over 953836.02 frames. ], batch size: 23, lr: 3.94e-03, grad_scale: 64.0 2023-03-26 04:34:53,567 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=25469.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:34:54,260 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4962, 1.2095, 1.2788, 1.2591, 1.6541, 1.6270, 1.4598, 1.2874], device='cuda:3'), covar=tensor([0.0289, 0.0330, 0.0556, 0.0317, 0.0218, 0.0359, 0.0268, 0.0360], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0113, 0.0139, 0.0119, 0.0106, 0.0102, 0.0092, 0.0110], device='cuda:3'), out_proj_covar=tensor([6.8483e-05, 8.9237e-05, 1.1197e-04, 9.3955e-05, 8.3889e-05, 7.5654e-05, 7.0233e-05, 8.5914e-05], device='cuda:3') 2023-03-26 04:35:07,249 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25489.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 04:35:16,166 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25504.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:35:20,665 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.71 vs. limit=5.0 2023-03-26 04:35:20,834 INFO [finetune.py:976] (3/7) Epoch 5, batch 2600, loss[loss=0.2539, simple_loss=0.3147, pruned_loss=0.0966, over 4824.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.2816, pruned_loss=0.08288, over 953668.50 frames. ], batch size: 38, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:35:28,040 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.057e+02 1.693e+02 2.088e+02 2.425e+02 4.415e+02, threshold=4.177e+02, percent-clipped=1.0 2023-03-26 04:35:48,649 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=25552.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:35:54,565 INFO [finetune.py:976] (3/7) Epoch 5, batch 2650, loss[loss=0.2338, simple_loss=0.2939, pruned_loss=0.08689, over 4867.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.2833, pruned_loss=0.08326, over 954516.25 frames. ], batch size: 34, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:36:33,947 INFO [finetune.py:976] (3/7) Epoch 5, batch 2700, loss[loss=0.2365, simple_loss=0.3038, pruned_loss=0.08459, over 4813.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.2835, pruned_loss=0.08315, over 954835.44 frames. ], batch size: 41, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:36:50,928 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.205e+02 1.711e+02 2.002e+02 2.331e+02 3.948e+02, threshold=4.004e+02, percent-clipped=0.0 2023-03-26 04:37:22,110 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0531, 1.7933, 1.5580, 1.8786, 1.7907, 1.7169, 1.6932, 2.5681], device='cuda:3'), covar=tensor([0.6841, 0.8322, 0.5716, 0.7422, 0.6850, 0.4117, 0.7614, 0.2497], device='cuda:3'), in_proj_covar=tensor([0.0281, 0.0255, 0.0220, 0.0284, 0.0237, 0.0200, 0.0243, 0.0197], 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') 2023-03-26 04:37:26,219 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25651.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 04:37:32,346 INFO [finetune.py:976] (3/7) Epoch 5, batch 2750, loss[loss=0.2775, simple_loss=0.3227, pruned_loss=0.1161, over 4906.00 frames. ], tot_loss[loss=0.2228, simple_loss=0.2807, pruned_loss=0.08248, over 955081.68 frames. ], batch size: 36, lr: 3.94e-03, grad_scale: 32.0 2023-03-26 04:37:37,778 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4088, 1.3495, 1.5306, 2.4352, 1.6749, 2.2146, 0.9746, 2.0113], device='cuda:3'), covar=tensor([0.1897, 0.1608, 0.1290, 0.0726, 0.0966, 0.1113, 0.1677, 0.0793], device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0119, 0.0137, 0.0167, 0.0103, 0.0143, 0.0129, 0.0105], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 04:37:41,811 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-03-26 04:37:58,588 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=25699.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:38:07,649 INFO [finetune.py:976] (3/7) Epoch 5, batch 2800, loss[loss=0.2085, simple_loss=0.2692, pruned_loss=0.07385, over 4820.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2771, pruned_loss=0.08067, over 957876.75 frames. ], batch size: 38, lr: 3.93e-03, grad_scale: 32.0 2023-03-26 04:38:17,653 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4988, 1.3728, 1.4097, 1.3925, 0.7962, 2.2807, 0.7497, 1.2450], device='cuda:3'), covar=tensor([0.3478, 0.2514, 0.2060, 0.2532, 0.2173, 0.0383, 0.2775, 0.1445], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0114, 0.0117, 0.0122, 0.0118, 0.0098, 0.0101, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 04:38:23,890 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.133e+02 1.611e+02 1.888e+02 2.301e+02 3.388e+02, threshold=3.776e+02, percent-clipped=0.0 2023-03-26 04:38:42,770 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-26 04:39:03,402 INFO [finetune.py:976] (3/7) Epoch 5, batch 2850, loss[loss=0.1727, simple_loss=0.2357, pruned_loss=0.0548, over 4719.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2767, pruned_loss=0.08088, over 958482.80 frames. ], batch size: 23, lr: 3.93e-03, grad_scale: 32.0 2023-03-26 04:39:18,522 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25784.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 04:39:36,439 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0757, 1.9752, 1.5630, 2.0017, 2.0194, 1.7032, 2.3507, 2.1111], device='cuda:3'), covar=tensor([0.1660, 0.3157, 0.3977, 0.3528, 0.2944, 0.1927, 0.3792, 0.2196], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0191, 0.0233, 0.0252, 0.0227, 0.0186, 0.0208, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 04:39:37,534 INFO [finetune.py:976] (3/7) Epoch 5, batch 2900, loss[loss=0.2479, simple_loss=0.3097, pruned_loss=0.09305, over 4915.00 frames. ], tot_loss[loss=0.221, simple_loss=0.2787, pruned_loss=0.08167, over 958714.11 frames. ], batch size: 43, lr: 3.93e-03, grad_scale: 32.0 2023-03-26 04:39:44,762 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.053e+02 1.812e+02 2.065e+02 2.463e+02 5.082e+02, threshold=4.130e+02, percent-clipped=4.0 2023-03-26 04:39:59,081 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-26 04:40:09,042 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25858.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 04:40:10,731 INFO [finetune.py:976] (3/7) Epoch 5, batch 2950, loss[loss=0.2513, simple_loss=0.3119, pruned_loss=0.09537, over 4802.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.2811, pruned_loss=0.08186, over 957503.00 frames. ], batch size: 51, lr: 3.93e-03, grad_scale: 32.0 2023-03-26 04:40:35,711 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.74 vs. limit=5.0 2023-03-26 04:40:43,944 INFO [finetune.py:976] (3/7) Epoch 5, batch 3000, loss[loss=0.2097, simple_loss=0.2552, pruned_loss=0.08214, over 4731.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.2832, pruned_loss=0.08298, over 957207.15 frames. ], batch size: 23, lr: 3.93e-03, grad_scale: 32.0 2023-03-26 04:40:43,944 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 04:40:47,166 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8061, 1.7304, 2.1269, 1.3700, 1.7728, 2.0230, 1.7439, 2.1749], device='cuda:3'), covar=tensor([0.1360, 0.2081, 0.1209, 0.1764, 0.0918, 0.1266, 0.2500, 0.0805], device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0206, 0.0202, 0.0197, 0.0185, 0.0224, 0.0217, 0.0203], 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') 2023-03-26 04:40:54,558 INFO [finetune.py:1010] (3/7) Epoch 5, validation: loss=0.1652, simple_loss=0.2371, pruned_loss=0.04667, over 2265189.00 frames. 2023-03-26 04:40:54,558 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 04:41:00,100 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25919.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 04:41:01,808 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.188e+02 1.739e+02 2.096e+02 2.435e+02 4.160e+02, threshold=4.193e+02, percent-clipped=2.0 2023-03-26 04:41:03,182 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.0664, 0.9249, 0.9310, 1.0937, 1.1744, 1.1353, 1.0013, 0.9907], device='cuda:3'), covar=tensor([0.0311, 0.0321, 0.0557, 0.0276, 0.0266, 0.0436, 0.0343, 0.0411], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0114, 0.0139, 0.0119, 0.0106, 0.0102, 0.0092, 0.0110], device='cuda:3'), out_proj_covar=tensor([6.8425e-05, 8.9502e-05, 1.1199e-04, 9.4067e-05, 8.3789e-05, 7.6075e-05, 7.0147e-05, 8.5914e-05], device='cuda:3') 2023-03-26 04:41:27,994 INFO [finetune.py:976] (3/7) Epoch 5, batch 3050, loss[loss=0.2377, simple_loss=0.293, pruned_loss=0.09119, over 4859.00 frames. ], tot_loss[loss=0.225, simple_loss=0.2841, pruned_loss=0.08292, over 956289.60 frames. ], batch size: 31, lr: 3.93e-03, grad_scale: 32.0 2023-03-26 04:41:32,505 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.26 vs. limit=5.0 2023-03-26 04:41:38,244 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4045, 2.1721, 1.7365, 2.3012, 2.3152, 1.9264, 2.7972, 2.3012], device='cuda:3'), covar=tensor([0.1737, 0.3984, 0.4536, 0.4350, 0.3560, 0.2160, 0.3999, 0.2647], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0192, 0.0235, 0.0254, 0.0228, 0.0187, 0.0209, 0.0189], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 04:42:08,228 INFO [finetune.py:976] (3/7) Epoch 5, batch 3100, loss[loss=0.2074, simple_loss=0.2696, pruned_loss=0.0726, over 4828.00 frames. ], tot_loss[loss=0.2218, simple_loss=0.2808, pruned_loss=0.08136, over 958327.73 frames. ], batch size: 33, lr: 3.93e-03, grad_scale: 32.0 2023-03-26 04:42:25,415 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.176e+02 1.603e+02 1.879e+02 2.413e+02 4.411e+02, threshold=3.758e+02, percent-clipped=2.0 2023-03-26 04:42:35,916 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.78 vs. limit=2.0 2023-03-26 04:42:43,948 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8907, 1.8212, 1.6571, 1.7134, 1.3239, 3.9399, 1.7763, 2.1924], device='cuda:3'), covar=tensor([0.3049, 0.2062, 0.1872, 0.2096, 0.1518, 0.0152, 0.2320, 0.1188], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0114, 0.0117, 0.0122, 0.0117, 0.0098, 0.0101, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 04:43:10,313 INFO [finetune.py:976] (3/7) Epoch 5, batch 3150, loss[loss=0.1856, simple_loss=0.2357, pruned_loss=0.06773, over 4721.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2769, pruned_loss=0.08003, over 957540.85 frames. ], batch size: 23, lr: 3.93e-03, grad_scale: 32.0 2023-03-26 04:43:30,919 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26084.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 04:43:42,329 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6374, 1.4138, 0.9732, 0.2491, 1.2170, 1.4943, 1.3477, 1.3854], device='cuda:3'), covar=tensor([0.0851, 0.0922, 0.1481, 0.2080, 0.1500, 0.2504, 0.2490, 0.0935], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0201, 0.0204, 0.0190, 0.0218, 0.0209, 0.0222, 0.0201], 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') 2023-03-26 04:43:42,332 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26102.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:43:49,558 INFO [finetune.py:976] (3/7) Epoch 5, batch 3200, loss[loss=0.2333, simple_loss=0.2956, pruned_loss=0.08548, over 4903.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2738, pruned_loss=0.07876, over 957516.92 frames. ], batch size: 43, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 04:43:58,345 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.040e+02 1.599e+02 1.999e+02 2.453e+02 4.323e+02, threshold=3.997e+02, percent-clipped=1.0 2023-03-26 04:43:58,473 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7243, 2.2428, 2.2179, 1.3478, 2.4204, 2.0850, 1.6061, 1.9290], device='cuda:3'), covar=tensor([0.1086, 0.1073, 0.1844, 0.2211, 0.1765, 0.2031, 0.2286, 0.1525], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0202, 0.0205, 0.0191, 0.0219, 0.0211, 0.0223, 0.0202], 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') 2023-03-26 04:44:04,858 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=26132.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 04:44:37,844 INFO [finetune.py:976] (3/7) Epoch 5, batch 3250, loss[loss=0.1839, simple_loss=0.257, pruned_loss=0.05538, over 4897.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2738, pruned_loss=0.07879, over 957485.66 frames. ], batch size: 35, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 04:44:43,907 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26163.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:44:47,963 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26168.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:45:40,773 INFO [finetune.py:976] (3/7) Epoch 5, batch 3300, loss[loss=0.2324, simple_loss=0.2888, pruned_loss=0.08803, over 4182.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.2773, pruned_loss=0.08071, over 953689.20 frames. ], batch size: 65, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 04:45:47,648 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26214.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 04:45:56,626 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9284, 2.1244, 1.9857, 1.3261, 2.2994, 2.1910, 2.0731, 1.7673], device='cuda:3'), covar=tensor([0.0722, 0.0635, 0.0783, 0.0989, 0.0447, 0.0773, 0.0708, 0.1097], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0134, 0.0145, 0.0128, 0.0111, 0.0144, 0.0146, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 04:46:00,065 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.165e+02 1.721e+02 2.149e+02 2.490e+02 4.939e+02, threshold=4.298e+02, percent-clipped=1.0 2023-03-26 04:46:06,003 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26229.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:46:19,857 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26250.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:46:29,545 INFO [finetune.py:976] (3/7) Epoch 5, batch 3350, loss[loss=0.2118, simple_loss=0.2764, pruned_loss=0.07358, over 4775.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2785, pruned_loss=0.08099, over 951985.60 frames. ], batch size: 54, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 04:47:03,390 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 2023-03-26 04:47:06,386 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-03-26 04:47:12,173 INFO [finetune.py:976] (3/7) Epoch 5, batch 3400, loss[loss=0.2272, simple_loss=0.2799, pruned_loss=0.08724, over 4882.00 frames. ], tot_loss[loss=0.222, simple_loss=0.2806, pruned_loss=0.08173, over 951276.54 frames. ], batch size: 32, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 04:47:12,308 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26311.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:47:19,911 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.097e+02 1.624e+02 1.904e+02 2.361e+02 4.543e+02, threshold=3.807e+02, percent-clipped=1.0 2023-03-26 04:47:53,149 INFO [finetune.py:976] (3/7) Epoch 5, batch 3450, loss[loss=0.2377, simple_loss=0.301, pruned_loss=0.08725, over 4815.00 frames. ], tot_loss[loss=0.2222, simple_loss=0.2812, pruned_loss=0.08156, over 950892.84 frames. ], batch size: 38, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 04:48:37,958 INFO [finetune.py:976] (3/7) Epoch 5, batch 3500, loss[loss=0.2043, simple_loss=0.2674, pruned_loss=0.07058, over 4824.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.279, pruned_loss=0.081, over 952061.37 frames. ], batch size: 39, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 04:48:43,890 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26414.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:48:47,881 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-03-26 04:48:51,312 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.234e+02 1.586e+02 1.962e+02 2.229e+02 4.326e+02, threshold=3.925e+02, percent-clipped=1.0 2023-03-26 04:49:22,642 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.06 vs. limit=5.0 2023-03-26 04:49:23,746 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26458.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:49:25,518 INFO [finetune.py:976] (3/7) Epoch 5, batch 3550, loss[loss=0.1935, simple_loss=0.2493, pruned_loss=0.06886, over 4765.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2755, pruned_loss=0.07997, over 953786.47 frames. ], batch size: 28, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 04:49:34,681 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26475.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:50:11,928 INFO [finetune.py:976] (3/7) Epoch 5, batch 3600, loss[loss=0.2111, simple_loss=0.2633, pruned_loss=0.07949, over 4766.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2726, pruned_loss=0.07861, over 955679.72 frames. ], batch size: 27, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 04:50:13,890 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26514.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 04:50:19,781 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.118e+02 1.588e+02 1.920e+02 2.290e+02 3.397e+02, threshold=3.841e+02, percent-clipped=0.0 2023-03-26 04:50:20,508 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26524.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:50:25,169 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-26 04:50:42,181 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4726, 3.3721, 3.2504, 1.4352, 3.5256, 2.5666, 0.8434, 2.2818], device='cuda:3'), covar=tensor([0.2517, 0.1937, 0.1566, 0.3506, 0.1125, 0.1015, 0.4357, 0.1535], device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0173, 0.0165, 0.0129, 0.0156, 0.0123, 0.0147, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 04:50:55,027 INFO [finetune.py:976] (3/7) Epoch 5, batch 3650, loss[loss=0.2491, simple_loss=0.3201, pruned_loss=0.08907, over 4754.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2751, pruned_loss=0.07992, over 953919.18 frames. ], batch size: 54, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 04:50:55,489 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-03-26 04:50:55,700 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=26562.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 04:50:56,949 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6337, 1.5248, 1.4638, 1.6165, 1.3229, 3.7649, 1.3554, 2.0188], device='cuda:3'), covar=tensor([0.3380, 0.2457, 0.2191, 0.2337, 0.1824, 0.0155, 0.2800, 0.1374], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0113, 0.0116, 0.0121, 0.0117, 0.0097, 0.0100, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 04:51:26,462 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8256, 3.9830, 3.8059, 1.8888, 4.0620, 3.0175, 0.6430, 2.7661], device='cuda:3'), covar=tensor([0.2112, 0.1851, 0.1217, 0.3121, 0.1054, 0.0955, 0.4748, 0.1432], device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0173, 0.0165, 0.0129, 0.0156, 0.0123, 0.0147, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 04:51:31,368 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26606.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:51:34,854 INFO [finetune.py:976] (3/7) Epoch 5, batch 3700, loss[loss=0.2088, simple_loss=0.2752, pruned_loss=0.07126, over 4860.00 frames. ], tot_loss[loss=0.222, simple_loss=0.2802, pruned_loss=0.0819, over 954867.39 frames. ], batch size: 31, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 04:51:42,594 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.101e+02 1.866e+02 2.222e+02 2.784e+02 4.852e+02, threshold=4.444e+02, percent-clipped=6.0 2023-03-26 04:51:55,906 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5935, 0.5897, 1.4905, 1.3613, 1.3024, 1.2660, 1.2036, 1.3605], device='cuda:3'), covar=tensor([0.5643, 0.7965, 0.6553, 0.7035, 0.7714, 0.5707, 0.8709, 0.6160], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0247, 0.0254, 0.0258, 0.0241, 0.0218, 0.0274, 0.0223], 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') 2023-03-26 04:52:03,059 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8002, 1.7054, 1.4967, 1.8666, 2.3553, 1.8656, 1.5299, 1.4001], device='cuda:3'), covar=tensor([0.2425, 0.2399, 0.2155, 0.1932, 0.1930, 0.1270, 0.2762, 0.2138], device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0208, 0.0199, 0.0183, 0.0235, 0.0174, 0.0214, 0.0188], 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') 2023-03-26 04:52:07,673 INFO [finetune.py:976] (3/7) Epoch 5, batch 3750, loss[loss=0.1983, simple_loss=0.2537, pruned_loss=0.07139, over 4707.00 frames. ], tot_loss[loss=0.2232, simple_loss=0.2818, pruned_loss=0.08236, over 955331.31 frames. ], batch size: 23, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 04:52:17,093 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-26 04:52:46,387 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9978, 1.8816, 1.5320, 1.7784, 2.0491, 1.6782, 2.2873, 1.9030], device='cuda:3'), covar=tensor([0.1906, 0.3186, 0.4016, 0.3632, 0.3088, 0.2105, 0.4349, 0.2458], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0195, 0.0237, 0.0256, 0.0232, 0.0190, 0.0212, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 04:52:49,274 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5492, 2.3894, 1.9813, 1.0218, 2.1512, 1.9152, 1.7767, 2.0851], device='cuda:3'), covar=tensor([0.0793, 0.0822, 0.1653, 0.2258, 0.1644, 0.2401, 0.2179, 0.1145], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0203, 0.0205, 0.0192, 0.0220, 0.0212, 0.0224, 0.0202], 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') 2023-03-26 04:53:00,961 INFO [finetune.py:976] (3/7) Epoch 5, batch 3800, loss[loss=0.2306, simple_loss=0.2944, pruned_loss=0.08343, over 4799.00 frames. ], tot_loss[loss=0.2254, simple_loss=0.2838, pruned_loss=0.0835, over 954128.78 frames. ], batch size: 40, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 04:53:08,702 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.110e+02 1.710e+02 2.076e+02 2.649e+02 5.488e+02, threshold=4.152e+02, percent-clipped=2.0 2023-03-26 04:53:37,502 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2069, 1.7120, 2.0744, 2.0019, 1.7540, 1.8019, 1.8756, 1.8884], device='cuda:3'), covar=tensor([0.6773, 0.8618, 0.6460, 0.8437, 0.9285, 0.7174, 1.0692, 0.6621], device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0248, 0.0255, 0.0259, 0.0242, 0.0219, 0.0275, 0.0224], 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') 2023-03-26 04:53:39,793 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5802, 1.4825, 1.4595, 1.5857, 1.1133, 3.1053, 1.3299, 1.7317], device='cuda:3'), covar=tensor([0.3177, 0.2305, 0.1929, 0.2115, 0.1758, 0.0201, 0.2714, 0.1277], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0114, 0.0117, 0.0121, 0.0117, 0.0097, 0.0101, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 04:53:57,420 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26758.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:53:59,695 INFO [finetune.py:976] (3/7) Epoch 5, batch 3850, loss[loss=0.2179, simple_loss=0.281, pruned_loss=0.07737, over 4812.00 frames. ], tot_loss[loss=0.2226, simple_loss=0.2814, pruned_loss=0.08195, over 955426.82 frames. ], batch size: 40, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 04:54:11,292 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26770.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:55:00,184 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=26806.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:55:03,122 INFO [finetune.py:976] (3/7) Epoch 5, batch 3900, loss[loss=0.2029, simple_loss=0.256, pruned_loss=0.07493, over 4800.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2771, pruned_loss=0.0803, over 955144.47 frames. ], batch size: 45, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 04:55:15,091 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1450, 1.3673, 1.1495, 1.3115, 1.5190, 2.4938, 1.2746, 1.5246], device='cuda:3'), covar=tensor([0.1126, 0.1868, 0.1226, 0.1065, 0.1698, 0.0358, 0.1564, 0.1695], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0083, 0.0078, 0.0081, 0.0093, 0.0084, 0.0086, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 04:55:21,752 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.223e+02 1.797e+02 2.146e+02 2.516e+02 4.177e+02, threshold=4.292e+02, percent-clipped=1.0 2023-03-26 04:55:22,491 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26824.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:55:57,246 INFO [finetune.py:976] (3/7) Epoch 5, batch 3950, loss[loss=0.2405, simple_loss=0.304, pruned_loss=0.08847, over 4929.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2737, pruned_loss=0.07857, over 954662.54 frames. ], batch size: 38, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 04:56:06,713 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=26872.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:56:14,707 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7101, 1.8903, 1.4514, 1.4857, 2.0677, 2.1452, 1.9034, 1.8399], device='cuda:3'), covar=tensor([0.0376, 0.0381, 0.0566, 0.0357, 0.0280, 0.0555, 0.0398, 0.0319], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0112, 0.0137, 0.0117, 0.0104, 0.0100, 0.0091, 0.0108], device='cuda:3'), out_proj_covar=tensor([6.7233e-05, 8.7888e-05, 1.0995e-04, 9.2520e-05, 8.2591e-05, 7.4707e-05, 6.9348e-05, 8.4675e-05], device='cuda:3') 2023-03-26 04:56:29,761 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26904.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 04:56:30,927 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26906.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:56:34,935 INFO [finetune.py:976] (3/7) Epoch 5, batch 4000, loss[loss=0.2496, simple_loss=0.2977, pruned_loss=0.1008, over 4265.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2732, pruned_loss=0.079, over 952522.43 frames. ], batch size: 65, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 04:56:39,826 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-03-26 04:56:43,169 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.227e+02 1.662e+02 2.009e+02 2.453e+02 3.802e+02, threshold=4.018e+02, percent-clipped=0.0 2023-03-26 04:57:05,867 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26950.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:57:08,737 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=26954.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:57:18,758 INFO [finetune.py:976] (3/7) Epoch 5, batch 4050, loss[loss=0.1982, simple_loss=0.2407, pruned_loss=0.07782, over 4087.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2768, pruned_loss=0.08035, over 953260.92 frames. ], batch size: 17, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 04:57:26,846 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26965.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 04:57:59,970 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.3928, 1.4968, 1.5487, 0.7216, 1.4092, 1.7140, 1.7709, 1.3977], device='cuda:3'), covar=tensor([0.0917, 0.0501, 0.0401, 0.0612, 0.0417, 0.0452, 0.0273, 0.0630], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0158, 0.0120, 0.0136, 0.0133, 0.0124, 0.0148, 0.0146], device='cuda:3'), out_proj_covar=tensor([9.8043e-05, 1.1740e-04, 8.6992e-05, 9.9544e-05, 9.6206e-05, 9.1548e-05, 1.0999e-04, 1.0783e-04], device='cuda:3') 2023-03-26 04:58:10,822 INFO [finetune.py:976] (3/7) Epoch 5, batch 4100, loss[loss=0.2623, simple_loss=0.3213, pruned_loss=0.1017, over 4812.00 frames. ], tot_loss[loss=0.2229, simple_loss=0.2811, pruned_loss=0.08229, over 952092.91 frames. ], batch size: 38, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 04:58:11,440 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27011.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:58:19,550 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.223e+02 1.740e+02 2.111e+02 2.577e+02 5.326e+02, threshold=4.223e+02, percent-clipped=3.0 2023-03-26 04:58:28,472 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7718, 2.4576, 2.1846, 2.0361, 2.7391, 2.9959, 2.7032, 2.4242], device='cuda:3'), covar=tensor([0.0166, 0.0318, 0.0387, 0.0315, 0.0235, 0.0321, 0.0223, 0.0318], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0112, 0.0137, 0.0117, 0.0104, 0.0101, 0.0091, 0.0109], device='cuda:3'), out_proj_covar=tensor([6.7278e-05, 8.7850e-05, 1.1052e-04, 9.2634e-05, 8.2579e-05, 7.4935e-05, 6.9428e-05, 8.4940e-05], device='cuda:3') 2023-03-26 04:58:58,625 INFO [finetune.py:976] (3/7) Epoch 5, batch 4150, loss[loss=0.2803, simple_loss=0.3419, pruned_loss=0.1093, over 4817.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.2821, pruned_loss=0.08269, over 952968.66 frames. ], batch size: 40, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 04:59:04,721 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-26 04:59:05,640 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27070.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:59:32,488 INFO [finetune.py:976] (3/7) Epoch 5, batch 4200, loss[loss=0.2338, simple_loss=0.2866, pruned_loss=0.09053, over 4789.00 frames. ], tot_loss[loss=0.2225, simple_loss=0.2817, pruned_loss=0.08167, over 953234.18 frames. ], batch size: 25, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 04:59:37,722 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=27118.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 04:59:41,614 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.001e+02 1.682e+02 1.955e+02 2.295e+02 5.538e+02, threshold=3.911e+02, percent-clipped=3.0 2023-03-26 04:59:57,832 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27148.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:00:10,426 INFO [finetune.py:976] (3/7) Epoch 5, batch 4250, loss[loss=0.2027, simple_loss=0.2612, pruned_loss=0.07207, over 4906.00 frames. ], tot_loss[loss=0.221, simple_loss=0.2795, pruned_loss=0.08124, over 951381.56 frames. ], batch size: 36, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 05:01:08,485 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27209.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:01:09,570 INFO [finetune.py:976] (3/7) Epoch 5, batch 4300, loss[loss=0.187, simple_loss=0.249, pruned_loss=0.06254, over 4910.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2759, pruned_loss=0.07997, over 951343.50 frames. ], batch size: 46, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 05:01:26,945 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.146e+02 1.762e+02 2.023e+02 2.453e+02 1.035e+03, threshold=4.046e+02, percent-clipped=2.0 2023-03-26 05:01:33,586 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6406, 1.6635, 2.1335, 2.0593, 1.7862, 4.0950, 1.3669, 1.8869], device='cuda:3'), covar=tensor([0.1303, 0.2290, 0.1385, 0.1303, 0.1914, 0.0269, 0.2062, 0.2132], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0082, 0.0077, 0.0080, 0.0093, 0.0083, 0.0086, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 05:01:59,786 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27260.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 05:02:00,300 INFO [finetune.py:976] (3/7) Epoch 5, batch 4350, loss[loss=0.2234, simple_loss=0.2714, pruned_loss=0.08772, over 4821.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2722, pruned_loss=0.07834, over 952191.21 frames. ], batch size: 40, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 05:02:30,556 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27306.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:02:33,997 INFO [finetune.py:976] (3/7) Epoch 5, batch 4400, loss[loss=0.2177, simple_loss=0.2801, pruned_loss=0.07761, over 4803.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2743, pruned_loss=0.07995, over 953562.06 frames. ], batch size: 25, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 05:02:41,212 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.108e+02 1.601e+02 1.888e+02 2.389e+02 3.644e+02, threshold=3.775e+02, percent-clipped=0.0 2023-03-26 05:03:07,515 INFO [finetune.py:976] (3/7) Epoch 5, batch 4450, loss[loss=0.267, simple_loss=0.3284, pruned_loss=0.1028, over 4856.00 frames. ], tot_loss[loss=0.2218, simple_loss=0.2794, pruned_loss=0.08205, over 953565.90 frames. ], batch size: 44, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 05:03:11,616 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-03-26 05:03:11,986 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.58 vs. limit=5.0 2023-03-26 05:03:12,743 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-03-26 05:03:34,980 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6328, 0.6232, 1.5220, 1.3809, 1.3523, 1.3025, 1.2632, 1.3939], device='cuda:3'), covar=tensor([0.5373, 0.7515, 0.6171, 0.6906, 0.7136, 0.5657, 0.8244, 0.5763], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0248, 0.0255, 0.0258, 0.0242, 0.0218, 0.0275, 0.0223], 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') 2023-03-26 05:03:36,668 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-03-26 05:03:37,230 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9839, 1.8565, 1.5474, 1.8118, 2.0189, 1.6286, 2.2761, 2.0015], device='cuda:3'), covar=tensor([0.1933, 0.3359, 0.4306, 0.3530, 0.3069, 0.2135, 0.3845, 0.2425], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0193, 0.0236, 0.0254, 0.0230, 0.0189, 0.0211, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 05:03:40,731 INFO [finetune.py:976] (3/7) Epoch 5, batch 4500, loss[loss=0.2616, simple_loss=0.3104, pruned_loss=0.1064, over 4230.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.2816, pruned_loss=0.08312, over 952737.47 frames. ], batch size: 65, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 05:03:48,442 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.291e+02 1.723e+02 2.077e+02 2.543e+02 6.449e+02, threshold=4.154e+02, percent-clipped=4.0 2023-03-26 05:04:09,020 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7134, 1.3674, 0.9428, 1.5890, 1.9748, 1.1361, 1.5649, 1.7061], device='cuda:3'), covar=tensor([0.1519, 0.2101, 0.2138, 0.1244, 0.2109, 0.2259, 0.1442, 0.1975], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0098, 0.0116, 0.0094, 0.0124, 0.0097, 0.0101, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 05:04:14,225 INFO [finetune.py:976] (3/7) Epoch 5, batch 4550, loss[loss=0.2488, simple_loss=0.3011, pruned_loss=0.09827, over 4907.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.283, pruned_loss=0.08301, over 955440.92 frames. ], batch size: 36, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 05:04:42,681 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27504.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:04:47,378 INFO [finetune.py:976] (3/7) Epoch 5, batch 4600, loss[loss=0.2008, simple_loss=0.2655, pruned_loss=0.06802, over 4788.00 frames. ], tot_loss[loss=0.2229, simple_loss=0.2813, pruned_loss=0.08223, over 952935.57 frames. ], batch size: 29, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 05:04:55,105 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.119e+02 1.596e+02 2.009e+02 2.524e+02 8.514e+02, threshold=4.018e+02, percent-clipped=5.0 2023-03-26 05:05:12,318 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6144, 1.4656, 1.3144, 1.3934, 1.8219, 1.7381, 1.5790, 1.3157], device='cuda:3'), covar=tensor([0.0276, 0.0276, 0.0581, 0.0323, 0.0207, 0.0394, 0.0296, 0.0376], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0111, 0.0137, 0.0116, 0.0104, 0.0100, 0.0090, 0.0108], device='cuda:3'), out_proj_covar=tensor([6.6974e-05, 8.7265e-05, 1.0986e-04, 9.2138e-05, 8.1869e-05, 7.4309e-05, 6.8925e-05, 8.4501e-05], device='cuda:3') 2023-03-26 05:05:20,043 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27560.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 05:05:20,538 INFO [finetune.py:976] (3/7) Epoch 5, batch 4650, loss[loss=0.1973, simple_loss=0.2547, pruned_loss=0.06992, over 4773.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.2776, pruned_loss=0.0803, over 953735.23 frames. ], batch size: 26, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 05:05:23,551 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0874, 1.9690, 1.6174, 1.9301, 2.1389, 1.7800, 2.4143, 2.1035], device='cuda:3'), covar=tensor([0.1679, 0.3091, 0.4063, 0.3482, 0.2907, 0.1951, 0.3919, 0.2150], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0192, 0.0234, 0.0252, 0.0228, 0.0188, 0.0209, 0.0188], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 05:06:07,094 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27606.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:06:13,639 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=27608.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 05:06:15,420 INFO [finetune.py:976] (3/7) Epoch 5, batch 4700, loss[loss=0.1662, simple_loss=0.2178, pruned_loss=0.05728, over 4028.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2746, pruned_loss=0.07919, over 953675.32 frames. ], batch size: 17, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 05:06:16,772 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9351, 1.2894, 1.6503, 1.7224, 1.5225, 1.5490, 1.5665, 1.6713], device='cuda:3'), covar=tensor([0.6772, 0.8909, 0.6831, 0.7945, 0.8704, 0.6634, 0.9709, 0.6370], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0247, 0.0255, 0.0257, 0.0241, 0.0218, 0.0275, 0.0223], 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') 2023-03-26 05:06:27,906 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.051e+02 1.549e+02 1.903e+02 2.293e+02 3.137e+02, threshold=3.806e+02, percent-clipped=0.0 2023-03-26 05:07:04,518 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5876, 1.4279, 1.4638, 1.5382, 0.9581, 3.0485, 1.0935, 1.5126], device='cuda:3'), covar=tensor([0.3425, 0.2543, 0.2166, 0.2357, 0.2093, 0.0228, 0.2891, 0.1466], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0114, 0.0117, 0.0122, 0.0117, 0.0098, 0.0101, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 05:07:05,704 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=27654.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:07:14,261 INFO [finetune.py:976] (3/7) Epoch 5, batch 4750, loss[loss=0.1545, simple_loss=0.2165, pruned_loss=0.04622, over 4279.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.2717, pruned_loss=0.07784, over 955240.77 frames. ], batch size: 18, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 05:07:56,929 INFO [finetune.py:976] (3/7) Epoch 5, batch 4800, loss[loss=0.2258, simple_loss=0.3047, pruned_loss=0.07341, over 4826.00 frames. ], tot_loss[loss=0.2174, simple_loss=0.2756, pruned_loss=0.07958, over 955029.13 frames. ], batch size: 40, lr: 3.93e-03, grad_scale: 16.0 2023-03-26 05:07:59,316 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27714.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 05:08:04,702 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.165e+02 1.708e+02 2.066e+02 2.363e+02 4.852e+02, threshold=4.133e+02, percent-clipped=3.0 2023-03-26 05:08:22,118 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.2869, 3.6365, 3.8177, 4.0934, 3.9819, 3.7056, 4.3645, 1.3239], device='cuda:3'), covar=tensor([0.0737, 0.0891, 0.0835, 0.0947, 0.1268, 0.1617, 0.0657, 0.5502], device='cuda:3'), in_proj_covar=tensor([0.0357, 0.0245, 0.0278, 0.0294, 0.0338, 0.0286, 0.0306, 0.0300], 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') 2023-03-26 05:08:30,288 INFO [finetune.py:976] (3/7) Epoch 5, batch 4850, loss[loss=0.2738, simple_loss=0.3294, pruned_loss=0.1091, over 4750.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.28, pruned_loss=0.08093, over 955388.04 frames. ], batch size: 59, lr: 3.92e-03, grad_scale: 16.0 2023-03-26 05:08:38,294 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2899, 1.4244, 1.5079, 1.6736, 1.4429, 3.0991, 1.2963, 1.4598], device='cuda:3'), covar=tensor([0.0999, 0.1643, 0.1278, 0.0970, 0.1597, 0.0282, 0.1428, 0.1626], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0082, 0.0077, 0.0079, 0.0092, 0.0083, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 05:08:39,489 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27775.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 05:08:59,121 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27804.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:09:04,313 INFO [finetune.py:976] (3/7) Epoch 5, batch 4900, loss[loss=0.2774, simple_loss=0.3302, pruned_loss=0.1122, over 4879.00 frames. ], tot_loss[loss=0.2227, simple_loss=0.282, pruned_loss=0.08174, over 953790.80 frames. ], batch size: 35, lr: 3.92e-03, grad_scale: 16.0 2023-03-26 05:09:12,047 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.126e+02 1.628e+02 1.864e+02 2.335e+02 3.818e+02, threshold=3.728e+02, percent-clipped=0.0 2023-03-26 05:09:12,210 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8209, 1.6962, 1.4768, 1.6219, 1.8972, 1.5241, 2.0370, 1.8361], device='cuda:3'), covar=tensor([0.1839, 0.3126, 0.4141, 0.3621, 0.3175, 0.2132, 0.3530, 0.2454], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0193, 0.0236, 0.0254, 0.0230, 0.0189, 0.0211, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 05:09:15,244 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 2023-03-26 05:09:30,673 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=27852.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:09:37,159 INFO [finetune.py:976] (3/7) Epoch 5, batch 4950, loss[loss=0.237, simple_loss=0.2983, pruned_loss=0.08781, over 4777.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.2828, pruned_loss=0.08194, over 953565.46 frames. ], batch size: 28, lr: 3.92e-03, grad_scale: 16.0 2023-03-26 05:09:49,900 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7235, 1.6136, 1.4896, 1.7945, 1.9921, 1.8140, 1.1308, 1.4871], device='cuda:3'), covar=tensor([0.2329, 0.2303, 0.2012, 0.1857, 0.1735, 0.1202, 0.2926, 0.1985], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0212, 0.0203, 0.0187, 0.0239, 0.0178, 0.0217, 0.0192], 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') 2023-03-26 05:10:09,333 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7378, 1.5022, 1.5538, 1.6300, 1.2356, 3.4744, 1.4099, 1.7732], device='cuda:3'), covar=tensor([0.3373, 0.2480, 0.2121, 0.2439, 0.1816, 0.0171, 0.2779, 0.1396], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0114, 0.0118, 0.0122, 0.0117, 0.0098, 0.0101, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 05:10:10,462 INFO [finetune.py:976] (3/7) Epoch 5, batch 5000, loss[loss=0.2643, simple_loss=0.3322, pruned_loss=0.0982, over 4810.00 frames. ], tot_loss[loss=0.221, simple_loss=0.2802, pruned_loss=0.08091, over 953971.98 frames. ], batch size: 38, lr: 3.92e-03, grad_scale: 16.0 2023-03-26 05:10:19,084 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.230e+02 1.618e+02 1.837e+02 2.301e+02 4.829e+02, threshold=3.674e+02, percent-clipped=1.0 2023-03-26 05:10:31,200 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0967, 2.0498, 1.6298, 2.1191, 2.1289, 1.7380, 2.6212, 2.0987], device='cuda:3'), covar=tensor([0.1733, 0.3108, 0.3936, 0.3454, 0.3010, 0.1966, 0.3659, 0.2339], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0193, 0.0236, 0.0255, 0.0231, 0.0189, 0.0211, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 05:10:40,144 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7654, 1.5606, 1.3743, 1.3053, 1.9208, 1.9496, 1.7649, 1.4351], device='cuda:3'), covar=tensor([0.0300, 0.0386, 0.0556, 0.0426, 0.0212, 0.0425, 0.0313, 0.0444], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0112, 0.0137, 0.0117, 0.0104, 0.0100, 0.0091, 0.0109], device='cuda:3'), out_proj_covar=tensor([6.7717e-05, 8.7935e-05, 1.1037e-04, 9.2828e-05, 8.2463e-05, 7.4558e-05, 6.9514e-05, 8.5061e-05], device='cuda:3') 2023-03-26 05:10:43,560 INFO [finetune.py:976] (3/7) Epoch 5, batch 5050, loss[loss=0.2245, simple_loss=0.279, pruned_loss=0.08506, over 4892.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2776, pruned_loss=0.08069, over 955297.03 frames. ], batch size: 35, lr: 3.92e-03, grad_scale: 16.0 2023-03-26 05:11:48,643 INFO [finetune.py:976] (3/7) Epoch 5, batch 5100, loss[loss=0.1819, simple_loss=0.2449, pruned_loss=0.05946, over 4797.00 frames. ], tot_loss[loss=0.216, simple_loss=0.2742, pruned_loss=0.07888, over 956954.14 frames. ], batch size: 45, lr: 3.92e-03, grad_scale: 16.0 2023-03-26 05:12:02,951 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.177e+02 1.639e+02 1.875e+02 2.408e+02 3.954e+02, threshold=3.749e+02, percent-clipped=2.0 2023-03-26 05:12:32,814 INFO [finetune.py:976] (3/7) Epoch 5, batch 5150, loss[loss=0.1942, simple_loss=0.2597, pruned_loss=0.06435, over 4749.00 frames. ], tot_loss[loss=0.217, simple_loss=0.2747, pruned_loss=0.07971, over 956412.45 frames. ], batch size: 27, lr: 3.92e-03, grad_scale: 16.0 2023-03-26 05:12:38,922 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28070.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 05:13:06,308 INFO [finetune.py:976] (3/7) Epoch 5, batch 5200, loss[loss=0.2397, simple_loss=0.2951, pruned_loss=0.0921, over 4916.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2771, pruned_loss=0.08019, over 956898.73 frames. ], batch size: 36, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:13:14,538 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.308e+02 1.748e+02 1.996e+02 2.342e+02 5.311e+02, threshold=3.992e+02, percent-clipped=1.0 2023-03-26 05:13:17,607 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0681, 1.5691, 2.3862, 1.6467, 2.0904, 2.2441, 1.6103, 2.4238], device='cuda:3'), covar=tensor([0.1262, 0.2262, 0.1405, 0.2043, 0.0909, 0.1451, 0.2890, 0.0926], device='cuda:3'), in_proj_covar=tensor([0.0205, 0.0204, 0.0200, 0.0195, 0.0186, 0.0222, 0.0216, 0.0204], 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') 2023-03-26 05:13:35,301 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4927, 2.1211, 1.6546, 0.7356, 1.7931, 1.9706, 1.8077, 1.9686], device='cuda:3'), covar=tensor([0.0760, 0.0943, 0.1605, 0.2429, 0.1717, 0.2372, 0.2127, 0.0964], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0201, 0.0204, 0.0191, 0.0219, 0.0209, 0.0222, 0.0201], 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') 2023-03-26 05:13:39,348 INFO [finetune.py:976] (3/7) Epoch 5, batch 5250, loss[loss=0.211, simple_loss=0.2676, pruned_loss=0.0772, over 4862.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.2793, pruned_loss=0.08172, over 954912.98 frames. ], batch size: 31, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:14:12,315 INFO [finetune.py:976] (3/7) Epoch 5, batch 5300, loss[loss=0.2497, simple_loss=0.3085, pruned_loss=0.09542, over 4920.00 frames. ], tot_loss[loss=0.2228, simple_loss=0.2812, pruned_loss=0.0822, over 955591.87 frames. ], batch size: 38, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:14:19,559 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.181e+02 1.725e+02 1.957e+02 2.435e+02 6.444e+02, threshold=3.915e+02, percent-clipped=2.0 2023-03-26 05:14:24,305 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28229.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:14:51,796 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28255.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 05:14:56,275 INFO [finetune.py:976] (3/7) Epoch 5, batch 5350, loss[loss=0.2995, simple_loss=0.3212, pruned_loss=0.1389, over 4061.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.2807, pruned_loss=0.08122, over 951136.57 frames. ], batch size: 65, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:15:13,525 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28287.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:15:15,776 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28290.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:15:28,418 INFO [finetune.py:976] (3/7) Epoch 5, batch 5400, loss[loss=0.1599, simple_loss=0.2285, pruned_loss=0.04566, over 4888.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2776, pruned_loss=0.0797, over 951699.87 frames. ], batch size: 32, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:15:31,992 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28316.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 05:15:33,813 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3412, 2.1984, 1.7828, 2.3914, 2.2746, 1.9418, 2.8807, 2.2631], device='cuda:3'), covar=tensor([0.1754, 0.3568, 0.4091, 0.3905, 0.3131, 0.2085, 0.4045, 0.2516], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0193, 0.0236, 0.0255, 0.0231, 0.0189, 0.0211, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 05:15:36,016 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.900e+01 1.568e+02 1.878e+02 2.260e+02 3.573e+02, threshold=3.756e+02, percent-clipped=0.0 2023-03-26 05:15:53,765 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28348.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:16:00,330 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5396, 1.2989, 2.0445, 3.1911, 2.1133, 2.4434, 0.8404, 2.6306], device='cuda:3'), covar=tensor([0.2214, 0.2253, 0.1649, 0.0978, 0.1179, 0.1717, 0.2500, 0.0795], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0120, 0.0137, 0.0167, 0.0104, 0.0143, 0.0129, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 05:16:01,446 INFO [finetune.py:976] (3/7) Epoch 5, batch 5450, loss[loss=0.1856, simple_loss=0.2451, pruned_loss=0.06308, over 4826.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2749, pruned_loss=0.07894, over 952364.26 frames. ], batch size: 39, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:16:18,312 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28370.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 05:16:59,842 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7671, 1.2434, 0.7491, 1.6155, 2.1677, 1.4542, 1.5084, 1.8675], device='cuda:3'), covar=tensor([0.1400, 0.1988, 0.2237, 0.1171, 0.1877, 0.2114, 0.1361, 0.1822], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0098, 0.0116, 0.0094, 0.0124, 0.0097, 0.0100, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 05:17:02,776 INFO [finetune.py:976] (3/7) Epoch 5, batch 5500, loss[loss=0.2147, simple_loss=0.2677, pruned_loss=0.08086, over 4905.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2722, pruned_loss=0.07792, over 953095.25 frames. ], batch size: 35, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:17:12,777 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=28418.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 05:17:21,152 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.059e+02 1.651e+02 2.036e+02 2.478e+02 5.642e+02, threshold=4.072e+02, percent-clipped=3.0 2023-03-26 05:17:32,689 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-26 05:18:07,648 INFO [finetune.py:976] (3/7) Epoch 5, batch 5550, loss[loss=0.2578, simple_loss=0.3089, pruned_loss=0.1033, over 4819.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.2736, pruned_loss=0.07853, over 953817.07 frames. ], batch size: 40, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:18:36,696 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28486.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:18:45,467 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28492.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:19:02,057 INFO [finetune.py:976] (3/7) Epoch 5, batch 5600, loss[loss=0.2608, simple_loss=0.3227, pruned_loss=0.09943, over 4919.00 frames. ], tot_loss[loss=0.2202, simple_loss=0.2793, pruned_loss=0.08057, over 954538.57 frames. ], batch size: 37, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:19:13,255 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-26 05:19:14,568 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.120e+02 1.753e+02 2.098e+02 2.591e+02 4.684e+02, threshold=4.196e+02, percent-clipped=2.0 2023-03-26 05:19:19,276 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7698, 1.6209, 1.6872, 1.6802, 1.2236, 3.8446, 1.4948, 2.0230], device='cuda:3'), covar=tensor([0.3324, 0.2451, 0.1956, 0.2297, 0.1919, 0.0149, 0.2505, 0.1302], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0114, 0.0118, 0.0122, 0.0117, 0.0098, 0.0101, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 05:19:40,175 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28547.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:19:46,447 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1467, 1.7094, 1.8164, 1.9190, 1.6115, 1.6681, 1.8874, 1.8019], device='cuda:3'), covar=tensor([0.6142, 0.8557, 0.7114, 0.8206, 0.9797, 0.7120, 1.0713, 0.6627], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0248, 0.0255, 0.0257, 0.0241, 0.0219, 0.0274, 0.0224], 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') 2023-03-26 05:19:47,576 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28553.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:19:52,221 INFO [finetune.py:976] (3/7) Epoch 5, batch 5650, loss[loss=0.2548, simple_loss=0.3041, pruned_loss=0.1027, over 4799.00 frames. ], tot_loss[loss=0.2221, simple_loss=0.2817, pruned_loss=0.08123, over 954764.52 frames. ], batch size: 25, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:20:02,302 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9376, 2.4408, 2.1482, 1.0996, 2.2431, 2.2294, 1.8224, 2.0565], device='cuda:3'), covar=tensor([0.0703, 0.0984, 0.1502, 0.2276, 0.1572, 0.2168, 0.2295, 0.1121], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0199, 0.0201, 0.0189, 0.0217, 0.0208, 0.0220, 0.0200], 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') 2023-03-26 05:20:06,327 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28585.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:20:10,331 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-03-26 05:20:18,808 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.68 vs. limit=5.0 2023-03-26 05:20:21,893 INFO [finetune.py:976] (3/7) Epoch 5, batch 5700, loss[loss=0.1704, simple_loss=0.2231, pruned_loss=0.05884, over 3574.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2773, pruned_loss=0.07989, over 941652.74 frames. ], batch size: 15, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:20:21,937 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28611.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 05:20:29,013 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.136e+02 1.580e+02 1.894e+02 2.210e+02 5.665e+02, threshold=3.789e+02, percent-clipped=1.0 2023-03-26 05:20:53,188 INFO [finetune.py:976] (3/7) Epoch 6, batch 0, loss[loss=0.2031, simple_loss=0.2783, pruned_loss=0.06393, over 4819.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2783, pruned_loss=0.06393, over 4819.00 frames. ], batch size: 39, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:20:53,188 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 05:20:57,116 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4175, 1.5634, 1.4871, 1.6067, 1.6937, 2.9679, 1.4274, 1.6812], device='cuda:3'), covar=tensor([0.1005, 0.1670, 0.1066, 0.1023, 0.1490, 0.0330, 0.1382, 0.1594], device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0083, 0.0078, 0.0081, 0.0094, 0.0084, 0.0087, 0.0081], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 05:21:08,971 INFO [finetune.py:1010] (3/7) Epoch 6, validation: loss=0.1659, simple_loss=0.2379, pruned_loss=0.04693, over 2265189.00 frames. 2023-03-26 05:21:08,972 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 05:21:15,235 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28643.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:21:15,283 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1345, 2.2135, 2.1312, 1.4685, 2.3466, 2.3244, 2.2790, 1.8064], device='cuda:3'), covar=tensor([0.0736, 0.0657, 0.0842, 0.1022, 0.0462, 0.0801, 0.0701, 0.1179], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0136, 0.0146, 0.0131, 0.0114, 0.0145, 0.0149, 0.0166], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 05:21:18,007 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0550, 2.0946, 2.0692, 1.4259, 2.2350, 2.2266, 2.1298, 1.7134], device='cuda:3'), covar=tensor([0.0742, 0.0639, 0.0777, 0.1110, 0.0531, 0.0723, 0.0659, 0.1184], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0136, 0.0146, 0.0131, 0.0114, 0.0145, 0.0148, 0.0166], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 05:21:29,755 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5003, 1.3951, 1.3594, 1.4602, 0.9706, 2.8157, 1.1066, 1.6336], device='cuda:3'), covar=tensor([0.3430, 0.2344, 0.2084, 0.2322, 0.1993, 0.0264, 0.2845, 0.1351], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0114, 0.0118, 0.0121, 0.0117, 0.0097, 0.0101, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 05:21:59,715 INFO [finetune.py:976] (3/7) Epoch 6, batch 50, loss[loss=0.2104, simple_loss=0.2649, pruned_loss=0.07792, over 4064.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2796, pruned_loss=0.07911, over 215623.29 frames. ], batch size: 66, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:22:19,626 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3452, 1.4372, 1.4266, 1.4373, 1.5491, 2.9592, 1.3304, 1.5283], device='cuda:3'), covar=tensor([0.1044, 0.1771, 0.1148, 0.1098, 0.1624, 0.0280, 0.1473, 0.1739], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0082, 0.0078, 0.0080, 0.0093, 0.0084, 0.0087, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 05:22:30,502 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.152e+02 1.667e+02 1.978e+02 2.446e+02 6.098e+02, threshold=3.955e+02, percent-clipped=3.0 2023-03-26 05:22:41,773 INFO [finetune.py:976] (3/7) Epoch 6, batch 100, loss[loss=0.2193, simple_loss=0.27, pruned_loss=0.0843, over 4901.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.2754, pruned_loss=0.07755, over 381163.90 frames. ], batch size: 36, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:22:52,795 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9336, 2.1400, 2.4359, 2.2436, 2.3560, 4.7102, 1.8232, 2.3965], device='cuda:3'), covar=tensor([0.0982, 0.1583, 0.0987, 0.0979, 0.1337, 0.0179, 0.1337, 0.1564], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0082, 0.0078, 0.0080, 0.0093, 0.0084, 0.0087, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 05:23:07,236 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-03-26 05:23:15,381 INFO [finetune.py:976] (3/7) Epoch 6, batch 150, loss[loss=0.2032, simple_loss=0.2606, pruned_loss=0.07291, over 4820.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.2734, pruned_loss=0.07866, over 506252.93 frames. ], batch size: 41, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:23:15,555 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.69 vs. limit=5.0 2023-03-26 05:23:37,612 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.084e+02 1.593e+02 1.877e+02 2.260e+02 4.734e+02, threshold=3.755e+02, percent-clipped=1.0 2023-03-26 05:23:39,492 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5252, 1.4637, 1.9244, 1.2155, 1.7069, 1.6811, 1.4226, 1.8700], device='cuda:3'), covar=tensor([0.1395, 0.2118, 0.1282, 0.1965, 0.0881, 0.1626, 0.2753, 0.1009], device='cuda:3'), in_proj_covar=tensor([0.0208, 0.0207, 0.0203, 0.0199, 0.0188, 0.0224, 0.0218, 0.0205], 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') 2023-03-26 05:23:48,134 INFO [finetune.py:976] (3/7) Epoch 6, batch 200, loss[loss=0.2041, simple_loss=0.2763, pruned_loss=0.06594, over 4822.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.2727, pruned_loss=0.07891, over 606600.15 frames. ], batch size: 39, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:23:50,544 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28842.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:23:55,122 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28848.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:24:03,981 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28861.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:24:07,041 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2248, 1.8434, 2.6716, 4.0195, 2.9086, 2.6896, 0.8711, 3.2006], device='cuda:3'), covar=tensor([0.1833, 0.1608, 0.1360, 0.0621, 0.0837, 0.1602, 0.2291, 0.0558], device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0119, 0.0136, 0.0166, 0.0103, 0.0143, 0.0128, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 05:24:19,119 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28885.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:24:26,390 INFO [finetune.py:976] (3/7) Epoch 6, batch 250, loss[loss=0.2348, simple_loss=0.3013, pruned_loss=0.08414, over 4901.00 frames. ], tot_loss[loss=0.217, simple_loss=0.2755, pruned_loss=0.07924, over 683473.39 frames. ], batch size: 35, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:24:47,333 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28911.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 05:25:02,683 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28922.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:25:03,159 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.192e+02 1.635e+02 1.950e+02 2.422e+02 4.878e+02, threshold=3.900e+02, percent-clipped=5.0 2023-03-26 05:25:03,889 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0517, 1.7770, 2.5581, 4.0833, 2.8885, 2.6658, 0.9173, 3.0920], device='cuda:3'), covar=tensor([0.1903, 0.1627, 0.1433, 0.0487, 0.0827, 0.1709, 0.2153, 0.0650], device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0119, 0.0136, 0.0165, 0.0103, 0.0142, 0.0128, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 05:25:09,280 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=28933.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:25:13,897 INFO [finetune.py:976] (3/7) Epoch 6, batch 300, loss[loss=0.2108, simple_loss=0.2801, pruned_loss=0.07069, over 4889.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2767, pruned_loss=0.07938, over 744887.84 frames. ], batch size: 35, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:25:16,444 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28943.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:25:28,661 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=28959.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 05:25:47,451 INFO [finetune.py:976] (3/7) Epoch 6, batch 350, loss[loss=0.2454, simple_loss=0.3086, pruned_loss=0.09109, over 4814.00 frames. ], tot_loss[loss=0.2212, simple_loss=0.2799, pruned_loss=0.08126, over 790777.98 frames. ], batch size: 39, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:25:49,412 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=28991.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:26:03,147 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29002.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:26:28,284 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.301e+02 1.836e+02 2.201e+02 2.620e+02 4.241e+02, threshold=4.402e+02, percent-clipped=2.0 2023-03-26 05:26:38,005 INFO [finetune.py:976] (3/7) Epoch 6, batch 400, loss[loss=0.2267, simple_loss=0.3013, pruned_loss=0.07603, over 4813.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.2801, pruned_loss=0.08061, over 826320.32 frames. ], batch size: 38, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:26:56,426 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8720, 1.7382, 1.4617, 1.6435, 1.6011, 1.5807, 1.6285, 2.3638], device='cuda:3'), covar=tensor([0.6420, 0.7045, 0.5272, 0.7492, 0.6283, 0.3860, 0.6592, 0.2436], device='cuda:3'), in_proj_covar=tensor([0.0281, 0.0257, 0.0219, 0.0283, 0.0238, 0.0201, 0.0244, 0.0199], 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') 2023-03-26 05:27:01,616 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29063.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:27:17,310 INFO [finetune.py:976] (3/7) Epoch 6, batch 450, loss[loss=0.1996, simple_loss=0.2619, pruned_loss=0.06861, over 4820.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.2804, pruned_loss=0.08088, over 856468.64 frames. ], batch size: 38, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:27:29,266 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.0541, 1.3019, 1.1052, 1.2663, 1.4123, 2.4711, 1.1783, 1.3991], device='cuda:3'), covar=tensor([0.1087, 0.1813, 0.1304, 0.1060, 0.1604, 0.0423, 0.1539, 0.1736], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0082, 0.0077, 0.0080, 0.0092, 0.0083, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 05:27:29,280 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4558, 1.4093, 1.4871, 0.7797, 1.5777, 1.5068, 1.5450, 1.3534], device='cuda:3'), covar=tensor([0.0600, 0.0771, 0.0651, 0.1028, 0.0773, 0.0703, 0.0582, 0.1140], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0135, 0.0144, 0.0128, 0.0113, 0.0143, 0.0146, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 05:27:41,547 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-26 05:27:45,346 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.158e+02 1.708e+02 1.996e+02 2.284e+02 5.200e+02, threshold=3.993e+02, percent-clipped=1.0 2023-03-26 05:27:55,106 INFO [finetune.py:976] (3/7) Epoch 6, batch 500, loss[loss=0.2138, simple_loss=0.2689, pruned_loss=0.0793, over 4705.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.277, pruned_loss=0.07917, over 877746.64 frames. ], batch size: 23, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:27:57,522 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29142.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:28:01,655 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29148.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:28:21,957 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.05 vs. limit=5.0 2023-03-26 05:28:28,352 INFO [finetune.py:976] (3/7) Epoch 6, batch 550, loss[loss=0.2626, simple_loss=0.2989, pruned_loss=0.1132, over 4690.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2734, pruned_loss=0.07754, over 896361.14 frames. ], batch size: 23, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:28:28,996 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=29190.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:28:34,887 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=29196.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:28:54,944 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1564, 2.0391, 2.1268, 0.7968, 2.3351, 2.5612, 2.1514, 2.0406], device='cuda:3'), covar=tensor([0.1037, 0.0820, 0.0577, 0.0871, 0.0525, 0.0652, 0.0545, 0.0813], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0160, 0.0122, 0.0138, 0.0134, 0.0125, 0.0148, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.8922e-05, 1.1838e-04, 8.8211e-05, 1.0093e-04, 9.6449e-05, 9.2262e-05, 1.0959e-04, 1.0870e-04], device='cuda:3') 2023-03-26 05:29:00,096 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29217.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:29:04,678 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.109e+02 1.638e+02 2.076e+02 2.525e+02 4.090e+02, threshold=4.153e+02, percent-clipped=1.0 2023-03-26 05:29:18,500 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29236.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:29:20,238 INFO [finetune.py:976] (3/7) Epoch 6, batch 600, loss[loss=0.1593, simple_loss=0.2251, pruned_loss=0.04672, over 4743.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2736, pruned_loss=0.07801, over 909889.24 frames. ], batch size: 27, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:29:20,983 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.5316, 1.6263, 1.7225, 0.8548, 1.6373, 1.8426, 1.9150, 1.5271], device='cuda:3'), covar=tensor([0.0826, 0.0607, 0.0457, 0.0603, 0.0449, 0.0508, 0.0304, 0.0638], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0161, 0.0122, 0.0139, 0.0135, 0.0125, 0.0149, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.9401e-05, 1.1888e-04, 8.8761e-05, 1.0126e-04, 9.7002e-05, 9.2751e-05, 1.1013e-04, 1.0910e-04], device='cuda:3') 2023-03-26 05:29:51,244 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.2008, 3.5852, 3.7711, 4.0383, 3.9380, 3.7346, 4.2484, 1.3263], device='cuda:3'), covar=tensor([0.0738, 0.0754, 0.0785, 0.0843, 0.1174, 0.1407, 0.0709, 0.5261], device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0241, 0.0273, 0.0291, 0.0334, 0.0282, 0.0302, 0.0297], 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') 2023-03-26 05:30:08,967 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1143, 1.2742, 0.8642, 2.0562, 2.4322, 1.8331, 1.7248, 2.0437], device='cuda:3'), covar=tensor([0.1427, 0.2168, 0.2161, 0.1145, 0.1899, 0.1981, 0.1399, 0.1925], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0099, 0.0116, 0.0093, 0.0124, 0.0097, 0.0101, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 05:30:22,478 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-26 05:30:24,096 INFO [finetune.py:976] (3/7) Epoch 6, batch 650, loss[loss=0.2516, simple_loss=0.3154, pruned_loss=0.09387, over 4816.00 frames. ], tot_loss[loss=0.219, simple_loss=0.2781, pruned_loss=0.07995, over 922452.94 frames. ], batch size: 39, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:30:34,116 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29297.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:30:43,492 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1478, 1.6357, 1.9864, 1.8932, 1.7007, 1.6973, 1.8480, 1.7691], device='cuda:3'), covar=tensor([0.6655, 0.8124, 0.5999, 0.7891, 0.8699, 0.6747, 0.9746, 0.6228], device='cuda:3'), in_proj_covar=tensor([0.0228, 0.0246, 0.0253, 0.0255, 0.0240, 0.0217, 0.0272, 0.0222], 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') 2023-03-26 05:30:54,627 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5279, 1.3981, 1.8173, 1.8361, 1.6109, 3.5499, 1.2322, 1.5929], device='cuda:3'), covar=tensor([0.0989, 0.1798, 0.1185, 0.1031, 0.1597, 0.0238, 0.1592, 0.1785], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0082, 0.0078, 0.0080, 0.0093, 0.0084, 0.0086, 0.0081], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 05:31:13,222 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.225e+02 1.735e+02 1.982e+02 2.438e+02 4.583e+02, threshold=3.965e+02, percent-clipped=2.0 2023-03-26 05:31:33,680 INFO [finetune.py:976] (3/7) Epoch 6, batch 700, loss[loss=0.249, simple_loss=0.3097, pruned_loss=0.09412, over 4069.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.2784, pruned_loss=0.07991, over 929263.79 frames. ], batch size: 65, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:31:46,288 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29358.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:32:17,029 INFO [finetune.py:976] (3/7) Epoch 6, batch 750, loss[loss=0.1921, simple_loss=0.2515, pruned_loss=0.06634, over 4740.00 frames. ], tot_loss[loss=0.2206, simple_loss=0.2802, pruned_loss=0.08044, over 937112.59 frames. ], batch size: 23, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:32:25,452 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6314, 1.5281, 1.4838, 1.6482, 1.0384, 3.6226, 1.4434, 2.0360], device='cuda:3'), covar=tensor([0.3349, 0.2469, 0.2094, 0.2332, 0.2043, 0.0152, 0.2638, 0.1336], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0114, 0.0118, 0.0122, 0.0117, 0.0099, 0.0101, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 05:32:26,068 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.1596, 1.3928, 1.3956, 0.6414, 1.1495, 1.5545, 1.5567, 1.3578], device='cuda:3'), covar=tensor([0.0955, 0.0524, 0.0463, 0.0595, 0.0467, 0.0517, 0.0358, 0.0652], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0160, 0.0122, 0.0138, 0.0134, 0.0125, 0.0149, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.9148e-05, 1.1841e-04, 8.8548e-05, 1.0082e-04, 9.6548e-05, 9.2117e-05, 1.1013e-04, 1.0878e-04], device='cuda:3') 2023-03-26 05:32:48,605 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2142, 2.0947, 1.8778, 1.0289, 1.9521, 1.7772, 1.6199, 1.9332], device='cuda:3'), covar=tensor([0.0794, 0.0531, 0.1081, 0.1733, 0.1036, 0.1616, 0.1666, 0.0752], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0199, 0.0200, 0.0189, 0.0216, 0.0208, 0.0219, 0.0199], 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') 2023-03-26 05:32:59,102 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.656e+01 1.814e+02 2.109e+02 2.501e+02 5.044e+02, threshold=4.217e+02, percent-clipped=3.0 2023-03-26 05:33:26,483 INFO [finetune.py:976] (3/7) Epoch 6, batch 800, loss[loss=0.18, simple_loss=0.2458, pruned_loss=0.05705, over 4808.00 frames. ], tot_loss[loss=0.2191, simple_loss=0.2794, pruned_loss=0.0794, over 941993.92 frames. ], batch size: 41, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:34:10,554 INFO [finetune.py:976] (3/7) Epoch 6, batch 850, loss[loss=0.2329, simple_loss=0.2826, pruned_loss=0.09167, over 4745.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2765, pruned_loss=0.07856, over 945661.81 frames. ], batch size: 27, lr: 3.92e-03, grad_scale: 32.0 2023-03-26 05:34:43,559 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29517.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:34:52,274 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.241e+02 1.650e+02 2.021e+02 2.394e+02 5.702e+02, threshold=4.042e+02, percent-clipped=1.0 2023-03-26 05:35:13,147 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4999, 1.3708, 1.3854, 1.4389, 1.0666, 2.9565, 1.1332, 1.5968], device='cuda:3'), covar=tensor([0.3514, 0.2564, 0.2168, 0.2427, 0.1917, 0.0229, 0.2723, 0.1357], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0115, 0.0118, 0.0122, 0.0118, 0.0099, 0.0102, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 05:35:14,735 INFO [finetune.py:976] (3/7) Epoch 6, batch 900, loss[loss=0.2453, simple_loss=0.3039, pruned_loss=0.09332, over 4828.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2745, pruned_loss=0.07846, over 947766.34 frames. ], batch size: 39, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:35:32,165 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29551.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:35:46,205 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=29565.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:35:46,290 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9577, 1.8423, 1.5560, 1.7475, 1.7268, 1.7175, 1.7138, 2.5081], device='cuda:3'), covar=tensor([0.6065, 0.6691, 0.4812, 0.6648, 0.6103, 0.3549, 0.6312, 0.2276], device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0258, 0.0219, 0.0284, 0.0240, 0.0202, 0.0245, 0.0200], 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') 2023-03-26 05:36:10,949 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3862, 2.8428, 2.2644, 1.7853, 2.9895, 3.0772, 2.7158, 2.4928], device='cuda:3'), covar=tensor([0.0768, 0.0584, 0.0929, 0.1013, 0.0413, 0.0697, 0.0733, 0.0945], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0136, 0.0145, 0.0128, 0.0114, 0.0144, 0.0147, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 05:36:13,633 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-26 05:36:14,403 INFO [finetune.py:976] (3/7) Epoch 6, batch 950, loss[loss=0.21, simple_loss=0.2589, pruned_loss=0.08051, over 4870.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2718, pruned_loss=0.07735, over 949336.90 frames. ], batch size: 31, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:36:21,375 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29592.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:36:43,998 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29612.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:36:56,223 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.131e+02 1.661e+02 1.977e+02 2.359e+02 4.237e+02, threshold=3.954e+02, percent-clipped=2.0 2023-03-26 05:37:17,826 INFO [finetune.py:976] (3/7) Epoch 6, batch 1000, loss[loss=0.2643, simple_loss=0.3095, pruned_loss=0.1095, over 4930.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2752, pruned_loss=0.07867, over 950312.45 frames. ], batch size: 33, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:37:35,983 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9728, 1.8302, 1.5412, 1.7883, 1.7559, 1.7095, 1.7756, 2.5052], device='cuda:3'), covar=tensor([0.6468, 0.7478, 0.5214, 0.6527, 0.5877, 0.3852, 0.6499, 0.2442], device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0259, 0.0220, 0.0285, 0.0240, 0.0203, 0.0246, 0.0200], 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') 2023-03-26 05:37:40,148 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29658.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:37:47,499 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29662.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:37:55,360 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.3561, 4.6527, 4.8033, 5.1369, 4.9915, 4.7687, 5.4450, 1.6517], device='cuda:3'), covar=tensor([0.0710, 0.0766, 0.0845, 0.0835, 0.1263, 0.1409, 0.0542, 0.5263], device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0243, 0.0276, 0.0293, 0.0336, 0.0284, 0.0304, 0.0297], 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') 2023-03-26 05:38:20,623 INFO [finetune.py:976] (3/7) Epoch 6, batch 1050, loss[loss=0.2061, simple_loss=0.2763, pruned_loss=0.06797, over 4863.00 frames. ], tot_loss[loss=0.2169, simple_loss=0.2768, pruned_loss=0.07848, over 950641.74 frames. ], batch size: 34, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:38:40,533 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-26 05:38:41,606 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=29706.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:39:02,909 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.157e+02 1.795e+02 2.095e+02 2.533e+02 7.754e+02, threshold=4.191e+02, percent-clipped=4.0 2023-03-26 05:39:03,058 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29723.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 05:39:20,381 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-26 05:39:23,374 INFO [finetune.py:976] (3/7) Epoch 6, batch 1100, loss[loss=0.2201, simple_loss=0.2781, pruned_loss=0.08111, over 4902.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2804, pruned_loss=0.08059, over 952209.28 frames. ], batch size: 37, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:39:28,624 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6235, 1.5186, 1.3290, 1.1319, 1.6803, 1.4031, 1.9924, 1.6385], device='cuda:3'), covar=tensor([0.1567, 0.2484, 0.3677, 0.3090, 0.2722, 0.1826, 0.2361, 0.2213], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0191, 0.0235, 0.0252, 0.0229, 0.0188, 0.0210, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 05:40:24,913 INFO [finetune.py:976] (3/7) Epoch 6, batch 1150, loss[loss=0.1831, simple_loss=0.2641, pruned_loss=0.05106, over 4948.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2804, pruned_loss=0.08016, over 952997.59 frames. ], batch size: 29, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:40:37,971 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29802.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:41:01,070 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.060e+02 1.742e+02 2.057e+02 2.377e+02 6.600e+02, threshold=4.115e+02, percent-clipped=1.0 2023-03-26 05:41:11,692 INFO [finetune.py:976] (3/7) Epoch 6, batch 1200, loss[loss=0.2604, simple_loss=0.2867, pruned_loss=0.117, over 4189.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2781, pruned_loss=0.07948, over 950579.80 frames. ], batch size: 65, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:41:28,761 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29863.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:41:45,292 INFO [finetune.py:976] (3/7) Epoch 6, batch 1250, loss[loss=0.2764, simple_loss=0.3073, pruned_loss=0.1227, over 4198.00 frames. ], tot_loss[loss=0.2159, simple_loss=0.2754, pruned_loss=0.07821, over 950805.32 frames. ], batch size: 65, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:41:47,164 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29892.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:41:57,712 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29907.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:42:00,185 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-26 05:42:04,919 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-03-26 05:42:07,774 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.170e+02 1.638e+02 1.953e+02 2.201e+02 4.150e+02, threshold=3.906e+02, percent-clipped=1.0 2023-03-26 05:42:18,520 INFO [finetune.py:976] (3/7) Epoch 6, batch 1300, loss[loss=0.1843, simple_loss=0.2365, pruned_loss=0.06609, over 4837.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2719, pruned_loss=0.07662, over 953582.48 frames. ], batch size: 25, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:42:19,141 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=29940.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:42:38,247 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8212, 1.5454, 2.0689, 1.4781, 2.0410, 2.1023, 1.5593, 2.2184], device='cuda:3'), covar=tensor([0.1200, 0.1961, 0.1287, 0.1773, 0.0712, 0.1225, 0.2555, 0.0692], device='cuda:3'), in_proj_covar=tensor([0.0205, 0.0204, 0.0198, 0.0195, 0.0184, 0.0220, 0.0215, 0.0201], 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') 2023-03-26 05:42:38,270 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9178, 1.7526, 1.4835, 1.9207, 1.9550, 1.6111, 2.2407, 1.8957], device='cuda:3'), covar=tensor([0.1758, 0.3110, 0.3764, 0.2927, 0.2796, 0.1914, 0.3858, 0.2166], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0192, 0.0236, 0.0253, 0.0230, 0.0190, 0.0211, 0.0191], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 05:42:53,738 INFO [finetune.py:976] (3/7) Epoch 6, batch 1350, loss[loss=0.2586, simple_loss=0.316, pruned_loss=0.1006, over 4800.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2733, pruned_loss=0.07813, over 953712.88 frames. ], batch size: 45, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:43:22,436 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30018.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 05:43:25,362 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.107e+02 1.660e+02 2.040e+02 2.486e+02 4.804e+02, threshold=4.081e+02, percent-clipped=2.0 2023-03-26 05:43:35,491 INFO [finetune.py:976] (3/7) Epoch 6, batch 1400, loss[loss=0.2511, simple_loss=0.297, pruned_loss=0.1027, over 4774.00 frames. ], tot_loss[loss=0.217, simple_loss=0.2766, pruned_loss=0.07872, over 953965.45 frames. ], batch size: 28, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:43:54,648 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-03-26 05:44:14,310 INFO [finetune.py:976] (3/7) Epoch 6, batch 1450, loss[loss=0.2996, simple_loss=0.3421, pruned_loss=0.1285, over 4250.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2788, pruned_loss=0.07901, over 954626.83 frames. ], batch size: 65, lr: 3.91e-03, grad_scale: 64.0 2023-03-26 05:44:58,650 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.184e+02 1.825e+02 2.216e+02 2.642e+02 7.386e+02, threshold=4.431e+02, percent-clipped=2.0 2023-03-26 05:44:59,432 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30125.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:45:04,849 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.79 vs. limit=5.0 2023-03-26 05:45:16,732 INFO [finetune.py:976] (3/7) Epoch 6, batch 1500, loss[loss=0.2148, simple_loss=0.2871, pruned_loss=0.07126, over 4822.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.2816, pruned_loss=0.08066, over 954804.80 frames. ], batch size: 39, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:45:38,716 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30158.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:45:59,674 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30186.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:46:01,366 INFO [finetune.py:976] (3/7) Epoch 6, batch 1550, loss[loss=0.184, simple_loss=0.2525, pruned_loss=0.05771, over 4862.00 frames. ], tot_loss[loss=0.2197, simple_loss=0.2802, pruned_loss=0.07964, over 954880.91 frames. ], batch size: 34, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:46:12,977 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30207.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:46:14,206 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30209.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:46:25,110 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.142e+02 1.698e+02 2.038e+02 2.458e+02 3.858e+02, threshold=4.076e+02, percent-clipped=0.0 2023-03-26 05:46:34,721 INFO [finetune.py:976] (3/7) Epoch 6, batch 1600, loss[loss=0.2243, simple_loss=0.2852, pruned_loss=0.08164, over 4755.00 frames. ], tot_loss[loss=0.2167, simple_loss=0.2764, pruned_loss=0.07845, over 953999.79 frames. ], batch size: 59, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:46:34,836 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7802, 1.7923, 1.7226, 1.0948, 2.0043, 1.9586, 1.8237, 1.6746], device='cuda:3'), covar=tensor([0.0665, 0.0694, 0.0782, 0.1018, 0.0588, 0.0704, 0.0716, 0.1114], device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0137, 0.0148, 0.0130, 0.0115, 0.0146, 0.0149, 0.0164], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 05:46:39,561 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-03-26 05:46:42,507 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6531, 1.7491, 1.9867, 1.8735, 2.0044, 4.2726, 1.6485, 2.0843], device='cuda:3'), covar=tensor([0.0919, 0.1711, 0.1147, 0.1039, 0.1381, 0.0162, 0.1324, 0.1529], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0082, 0.0077, 0.0080, 0.0092, 0.0083, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 05:46:44,947 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=30255.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:46:56,016 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30270.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:47:08,007 INFO [finetune.py:976] (3/7) Epoch 6, batch 1650, loss[loss=0.1534, simple_loss=0.223, pruned_loss=0.04193, over 4765.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2736, pruned_loss=0.07741, over 954332.04 frames. ], batch size: 28, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:47:10,513 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0403, 2.2016, 1.7750, 1.3788, 2.3699, 2.3222, 2.0637, 1.8806], device='cuda:3'), covar=tensor([0.0789, 0.0618, 0.1039, 0.1126, 0.0447, 0.0719, 0.0775, 0.1097], device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0138, 0.0149, 0.0132, 0.0116, 0.0147, 0.0150, 0.0166], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 05:47:21,459 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-03-26 05:47:24,413 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4326, 1.4075, 1.2562, 1.2883, 1.7180, 1.5868, 1.4796, 1.2298], device='cuda:3'), covar=tensor([0.0304, 0.0286, 0.0531, 0.0302, 0.0196, 0.0427, 0.0257, 0.0401], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0111, 0.0138, 0.0118, 0.0104, 0.0100, 0.0091, 0.0109], device='cuda:3'), out_proj_covar=tensor([6.8523e-05, 8.7688e-05, 1.1069e-04, 9.2930e-05, 8.1705e-05, 7.4646e-05, 6.9435e-05, 8.5209e-05], device='cuda:3') 2023-03-26 05:47:27,133 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30318.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 05:47:31,616 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.138e+02 1.710e+02 1.999e+02 2.408e+02 3.997e+02, threshold=3.998e+02, percent-clipped=0.0 2023-03-26 05:47:41,358 INFO [finetune.py:976] (3/7) Epoch 6, batch 1700, loss[loss=0.2163, simple_loss=0.2789, pruned_loss=0.07683, over 4936.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2717, pruned_loss=0.07688, over 956891.28 frames. ], batch size: 33, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:47:55,166 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30360.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:47:57,065 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9662, 1.2428, 1.7512, 1.7496, 1.5550, 1.5663, 1.5919, 1.5991], device='cuda:3'), covar=tensor([0.4877, 0.6985, 0.5601, 0.5947, 0.7486, 0.5445, 0.8023, 0.5646], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0246, 0.0254, 0.0256, 0.0241, 0.0219, 0.0273, 0.0223], 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') 2023-03-26 05:47:58,774 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=30366.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:48:27,034 INFO [finetune.py:976] (3/7) Epoch 6, batch 1750, loss[loss=0.2489, simple_loss=0.3278, pruned_loss=0.08499, over 4865.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2727, pruned_loss=0.07682, over 955526.52 frames. ], batch size: 44, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:48:48,759 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30421.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:48:50,965 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.220e+01 1.748e+02 2.265e+02 2.778e+02 4.150e+02, threshold=4.530e+02, percent-clipped=1.0 2023-03-26 05:48:51,671 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0576, 1.4730, 1.1240, 2.0457, 2.2824, 1.9304, 1.7288, 1.9465], device='cuda:3'), covar=tensor([0.1170, 0.1814, 0.1985, 0.0955, 0.1765, 0.2188, 0.1209, 0.1583], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0099, 0.0116, 0.0094, 0.0124, 0.0097, 0.0101, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 05:48:52,762 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.07 vs. limit=2.0 2023-03-26 05:48:56,638 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7228, 1.6988, 1.4026, 1.4932, 1.9730, 2.0580, 1.8093, 1.4220], device='cuda:3'), covar=tensor([0.0299, 0.0328, 0.0532, 0.0361, 0.0211, 0.0353, 0.0251, 0.0475], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0110, 0.0136, 0.0117, 0.0103, 0.0099, 0.0090, 0.0108], device='cuda:3'), out_proj_covar=tensor([6.7867e-05, 8.6979e-05, 1.0934e-04, 9.2223e-05, 8.1112e-05, 7.3650e-05, 6.8908e-05, 8.4440e-05], device='cuda:3') 2023-03-26 05:49:00,662 INFO [finetune.py:976] (3/7) Epoch 6, batch 1800, loss[loss=0.2203, simple_loss=0.291, pruned_loss=0.0748, over 4755.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2754, pruned_loss=0.07728, over 955233.24 frames. ], batch size: 54, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:49:13,233 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30458.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:49:15,706 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4477, 0.9683, 0.8370, 1.3336, 1.8843, 0.6969, 1.1714, 1.3711], device='cuda:3'), covar=tensor([0.1693, 0.2388, 0.1785, 0.1273, 0.2119, 0.2153, 0.1659, 0.2166], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0099, 0.0116, 0.0094, 0.0125, 0.0097, 0.0101, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 05:49:28,998 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30481.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:49:33,833 INFO [finetune.py:976] (3/7) Epoch 6, batch 1850, loss[loss=0.2179, simple_loss=0.2828, pruned_loss=0.07651, over 4732.00 frames. ], tot_loss[loss=0.217, simple_loss=0.2774, pruned_loss=0.07834, over 955576.07 frames. ], batch size: 59, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:49:44,725 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=30506.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:50:03,764 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.078e+02 1.671e+02 2.010e+02 2.577e+02 4.739e+02, threshold=4.020e+02, percent-clipped=1.0 2023-03-26 05:50:22,792 INFO [finetune.py:976] (3/7) Epoch 6, batch 1900, loss[loss=0.1993, simple_loss=0.2705, pruned_loss=0.06403, over 4826.00 frames. ], tot_loss[loss=0.218, simple_loss=0.2787, pruned_loss=0.07861, over 956275.85 frames. ], batch size: 30, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:50:33,196 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.50 vs. limit=5.0 2023-03-26 05:50:39,630 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2630, 2.2182, 1.9289, 1.6890, 2.5430, 2.6884, 2.3195, 2.0848], device='cuda:3'), covar=tensor([0.0339, 0.0342, 0.0527, 0.0395, 0.0290, 0.0424, 0.0386, 0.0412], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0111, 0.0136, 0.0116, 0.0103, 0.0099, 0.0090, 0.0108], device='cuda:3'), out_proj_covar=tensor([6.7438e-05, 8.7026e-05, 1.0915e-04, 9.1792e-05, 8.1145e-05, 7.3471e-05, 6.8737e-05, 8.4281e-05], device='cuda:3') 2023-03-26 05:50:52,224 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30565.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:51:02,772 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1064, 1.8836, 1.5813, 1.8779, 2.0842, 1.7290, 2.3988, 2.0219], device='cuda:3'), covar=tensor([0.1634, 0.3028, 0.3839, 0.3112, 0.2796, 0.2004, 0.3391, 0.2534], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0192, 0.0236, 0.0253, 0.0231, 0.0190, 0.0211, 0.0191], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 05:51:20,333 INFO [finetune.py:976] (3/7) Epoch 6, batch 1950, loss[loss=0.2362, simple_loss=0.2901, pruned_loss=0.09114, over 4285.00 frames. ], tot_loss[loss=0.2159, simple_loss=0.2759, pruned_loss=0.07794, over 954940.65 frames. ], batch size: 65, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:51:40,754 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-03-26 05:51:54,977 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-26 05:51:58,194 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.083e+02 1.556e+02 1.883e+02 2.215e+02 4.222e+02, threshold=3.767e+02, percent-clipped=2.0 2023-03-26 05:51:58,927 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8629, 1.6345, 1.5968, 1.7870, 1.3586, 3.7340, 1.4561, 2.1743], device='cuda:3'), covar=tensor([0.3046, 0.2411, 0.2065, 0.2178, 0.1747, 0.0165, 0.2493, 0.1247], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0114, 0.0118, 0.0121, 0.0117, 0.0099, 0.0101, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 05:52:04,796 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.3499, 1.4927, 1.5487, 0.7789, 1.4208, 1.7107, 1.7907, 1.4182], device='cuda:3'), covar=tensor([0.0872, 0.0477, 0.0510, 0.0550, 0.0438, 0.0431, 0.0254, 0.0524], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0159, 0.0122, 0.0138, 0.0133, 0.0124, 0.0147, 0.0146], device='cuda:3'), out_proj_covar=tensor([9.8645e-05, 1.1726e-04, 8.8389e-05, 1.0094e-04, 9.5997e-05, 9.1888e-05, 1.0925e-04, 1.0804e-04], device='cuda:3') 2023-03-26 05:52:09,311 INFO [finetune.py:976] (3/7) Epoch 6, batch 2000, loss[loss=0.2137, simple_loss=0.2658, pruned_loss=0.08075, over 4819.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.274, pruned_loss=0.07793, over 954007.77 frames. ], batch size: 30, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:52:19,609 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30655.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:52:56,172 INFO [finetune.py:976] (3/7) Epoch 6, batch 2050, loss[loss=0.1616, simple_loss=0.2291, pruned_loss=0.04705, over 4755.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2707, pruned_loss=0.07657, over 955281.02 frames. ], batch size: 27, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:53:14,339 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30716.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:53:14,407 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30716.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 05:53:20,056 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.195e+02 1.687e+02 1.889e+02 2.318e+02 4.112e+02, threshold=3.779e+02, percent-clipped=3.0 2023-03-26 05:53:40,713 INFO [finetune.py:976] (3/7) Epoch 6, batch 2100, loss[loss=0.2841, simple_loss=0.3305, pruned_loss=0.1188, over 4056.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2708, pruned_loss=0.07632, over 955185.86 frames. ], batch size: 65, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:53:52,629 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-03-26 05:54:13,623 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30781.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:54:18,836 INFO [finetune.py:976] (3/7) Epoch 6, batch 2150, loss[loss=0.2401, simple_loss=0.3019, pruned_loss=0.08916, over 4830.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2765, pruned_loss=0.07882, over 956083.56 frames. ], batch size: 30, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:54:42,039 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.093e+02 1.708e+02 2.011e+02 2.625e+02 4.679e+02, threshold=4.022e+02, percent-clipped=7.0 2023-03-26 05:54:46,133 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=30829.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:54:52,229 INFO [finetune.py:976] (3/7) Epoch 6, batch 2200, loss[loss=0.2072, simple_loss=0.2691, pruned_loss=0.07261, over 4858.00 frames. ], tot_loss[loss=0.2188, simple_loss=0.2784, pruned_loss=0.0796, over 956617.61 frames. ], batch size: 31, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:55:16,387 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30865.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:55:31,243 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.8210, 3.3724, 3.4864, 3.7065, 3.5776, 3.3355, 3.9099, 1.2431], device='cuda:3'), covar=tensor([0.0793, 0.0801, 0.0805, 0.0941, 0.1186, 0.1496, 0.0725, 0.4779], device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0243, 0.0276, 0.0292, 0.0331, 0.0282, 0.0304, 0.0297], 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') 2023-03-26 05:55:37,465 INFO [finetune.py:976] (3/7) Epoch 6, batch 2250, loss[loss=0.2336, simple_loss=0.2977, pruned_loss=0.08479, over 4861.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.2804, pruned_loss=0.08026, over 954799.36 frames. ], batch size: 34, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:55:49,185 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6178, 0.6074, 1.5293, 1.3484, 1.2908, 1.2742, 1.2646, 1.3875], device='cuda:3'), covar=tensor([0.4653, 0.6247, 0.5159, 0.5563, 0.6012, 0.4790, 0.6612, 0.4728], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0246, 0.0254, 0.0256, 0.0242, 0.0219, 0.0273, 0.0224], 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') 2023-03-26 05:55:53,170 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30903.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:56:10,383 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=30913.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:56:14,010 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6610, 4.2234, 3.9792, 2.3208, 4.3008, 3.3497, 0.9278, 2.9224], device='cuda:3'), covar=tensor([0.3271, 0.1525, 0.1613, 0.2844, 0.0836, 0.0856, 0.4360, 0.1348], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0170, 0.0162, 0.0127, 0.0155, 0.0122, 0.0144, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 05:56:22,248 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.065e+02 1.611e+02 1.958e+02 2.302e+02 5.232e+02, threshold=3.915e+02, percent-clipped=2.0 2023-03-26 05:56:22,995 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30925.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:56:34,305 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30934.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:56:42,904 INFO [finetune.py:976] (3/7) Epoch 6, batch 2300, loss[loss=0.1768, simple_loss=0.2528, pruned_loss=0.05042, over 4907.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.281, pruned_loss=0.0802, over 955998.01 frames. ], batch size: 37, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:56:42,981 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4788, 4.0312, 3.7384, 1.9670, 4.0879, 3.0453, 0.8363, 2.8154], device='cuda:3'), covar=tensor([0.2675, 0.1782, 0.1623, 0.3111, 0.0921, 0.0926, 0.4390, 0.1341], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0170, 0.0163, 0.0128, 0.0156, 0.0123, 0.0145, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 05:57:16,227 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30964.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:57:46,774 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30986.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:57:49,007 INFO [finetune.py:976] (3/7) Epoch 6, batch 2350, loss[loss=0.2617, simple_loss=0.3105, pruned_loss=0.1065, over 4912.00 frames. ], tot_loss[loss=0.2184, simple_loss=0.2782, pruned_loss=0.07931, over 953613.69 frames. ], batch size: 46, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:57:57,498 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30995.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:58:08,628 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-26 05:58:19,348 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31011.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 05:58:19,373 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31011.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 05:58:28,250 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31016.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:58:33,048 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.055e+02 1.621e+02 1.904e+02 2.250e+02 3.149e+02, threshold=3.808e+02, percent-clipped=0.0 2023-03-26 05:58:53,285 INFO [finetune.py:976] (3/7) Epoch 6, batch 2400, loss[loss=0.1882, simple_loss=0.2635, pruned_loss=0.05646, over 4822.00 frames. ], tot_loss[loss=0.2166, simple_loss=0.276, pruned_loss=0.07857, over 954181.80 frames. ], batch size: 25, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 05:59:16,908 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=31064.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 05:59:24,050 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31072.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 05:59:44,204 INFO [finetune.py:976] (3/7) Epoch 6, batch 2450, loss[loss=0.2107, simple_loss=0.2657, pruned_loss=0.07779, over 4827.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2724, pruned_loss=0.07706, over 954116.01 frames. ], batch size: 39, lr: 3.91e-03, grad_scale: 32.0 2023-03-26 06:00:21,949 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.316e+02 1.762e+02 2.221e+02 2.552e+02 5.044e+02, threshold=4.442e+02, percent-clipped=4.0 2023-03-26 06:00:24,492 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3034, 2.0481, 2.7800, 1.7749, 2.5713, 2.5585, 1.9258, 2.6718], device='cuda:3'), covar=tensor([0.1868, 0.2518, 0.1764, 0.2583, 0.1032, 0.1793, 0.2799, 0.1033], device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0206, 0.0199, 0.0195, 0.0183, 0.0221, 0.0217, 0.0202], 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') 2023-03-26 06:00:30,924 INFO [finetune.py:976] (3/7) Epoch 6, batch 2500, loss[loss=0.1864, simple_loss=0.2428, pruned_loss=0.06503, over 4770.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2733, pruned_loss=0.07787, over 953284.78 frames. ], batch size: 28, lr: 3.91e-03, grad_scale: 16.0 2023-03-26 06:01:00,111 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.7514, 3.4071, 3.5397, 3.5083, 3.3241, 3.2274, 3.8706, 1.2454], device='cuda:3'), covar=tensor([0.1321, 0.1460, 0.1429, 0.1854, 0.2165, 0.2227, 0.1096, 0.6934], device='cuda:3'), in_proj_covar=tensor([0.0359, 0.0245, 0.0278, 0.0294, 0.0335, 0.0285, 0.0304, 0.0299], 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') 2023-03-26 06:01:06,652 INFO [finetune.py:976] (3/7) Epoch 6, batch 2550, loss[loss=0.2015, simple_loss=0.2696, pruned_loss=0.06671, over 4890.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2766, pruned_loss=0.07929, over 953072.20 frames. ], batch size: 32, lr: 3.91e-03, grad_scale: 16.0 2023-03-26 06:01:26,847 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.57 vs. limit=5.0 2023-03-26 06:01:50,621 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.038e+02 1.680e+02 2.034e+02 2.315e+02 3.655e+02, threshold=4.067e+02, percent-clipped=0.0 2023-03-26 06:02:04,702 INFO [finetune.py:976] (3/7) Epoch 6, batch 2600, loss[loss=0.2149, simple_loss=0.2821, pruned_loss=0.07385, over 4919.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2774, pruned_loss=0.07889, over 951783.64 frames. ], batch size: 38, lr: 3.91e-03, grad_scale: 16.0 2023-03-26 06:02:33,424 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31259.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:03:04,531 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31281.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:03:09,263 INFO [finetune.py:976] (3/7) Epoch 6, batch 2650, loss[loss=0.219, simple_loss=0.2841, pruned_loss=0.07695, over 4890.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.2802, pruned_loss=0.07999, over 952187.72 frames. ], batch size: 43, lr: 3.91e-03, grad_scale: 16.0 2023-03-26 06:03:09,906 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31290.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:03:16,462 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1268, 2.1721, 2.0096, 1.4751, 2.2804, 2.2612, 2.1385, 1.8727], device='cuda:3'), covar=tensor([0.0564, 0.0546, 0.0749, 0.0957, 0.0474, 0.0742, 0.0709, 0.0957], device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0135, 0.0145, 0.0129, 0.0114, 0.0145, 0.0146, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 06:03:38,138 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31311.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:03:47,924 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.100e+02 1.761e+02 2.106e+02 2.462e+02 3.966e+02, threshold=4.213e+02, percent-clipped=0.0 2023-03-26 06:03:52,730 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9110, 1.9070, 1.6116, 1.4940, 2.1700, 2.2870, 2.0339, 1.8297], device='cuda:3'), covar=tensor([0.0317, 0.0390, 0.0570, 0.0380, 0.0274, 0.0506, 0.0267, 0.0403], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0111, 0.0136, 0.0116, 0.0103, 0.0099, 0.0090, 0.0108], device='cuda:3'), out_proj_covar=tensor([6.7124e-05, 8.7120e-05, 1.0905e-04, 9.1607e-05, 8.0807e-05, 7.3825e-05, 6.8309e-05, 8.4114e-05], device='cuda:3') 2023-03-26 06:03:59,089 INFO [finetune.py:976] (3/7) Epoch 6, batch 2700, loss[loss=0.2213, simple_loss=0.2772, pruned_loss=0.08271, over 4819.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2785, pruned_loss=0.079, over 952672.34 frames. ], batch size: 33, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:04:09,176 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.5199, 1.4506, 1.5878, 0.9057, 1.6320, 1.8421, 1.6944, 1.4862], device='cuda:3'), covar=tensor([0.1044, 0.0812, 0.0537, 0.0619, 0.0492, 0.0435, 0.0413, 0.0617], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0158, 0.0119, 0.0137, 0.0132, 0.0124, 0.0146, 0.0144], device='cuda:3'), out_proj_covar=tensor([9.7905e-05, 1.1648e-04, 8.6544e-05, 1.0002e-04, 9.4728e-05, 9.1591e-05, 1.0817e-04, 1.0658e-04], device='cuda:3') 2023-03-26 06:04:23,339 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=31359.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:04:33,876 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31367.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 06:04:55,838 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3781, 1.4784, 1.4445, 1.6224, 1.4857, 2.8551, 1.3243, 1.6261], device='cuda:3'), covar=tensor([0.0913, 0.1570, 0.1169, 0.0917, 0.1387, 0.0303, 0.1314, 0.1458], device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0082, 0.0078, 0.0080, 0.0093, 0.0083, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 06:05:04,058 INFO [finetune.py:976] (3/7) Epoch 6, batch 2750, loss[loss=0.2513, simple_loss=0.2817, pruned_loss=0.1105, over 4239.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2761, pruned_loss=0.07839, over 955101.77 frames. ], batch size: 65, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:05:28,054 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31409.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:05:46,888 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.163e+02 1.636e+02 1.989e+02 2.265e+02 3.886e+02, threshold=3.977e+02, percent-clipped=0.0 2023-03-26 06:05:56,330 INFO [finetune.py:976] (3/7) Epoch 6, batch 2800, loss[loss=0.231, simple_loss=0.2911, pruned_loss=0.08546, over 4851.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2714, pruned_loss=0.07648, over 957139.31 frames. ], batch size: 44, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:05:59,571 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.40 vs. limit=5.0 2023-03-26 06:06:07,760 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8173, 1.5817, 1.4662, 1.1850, 1.5832, 1.5514, 1.5565, 2.1541], device='cuda:3'), covar=tensor([0.5419, 0.5627, 0.4101, 0.5063, 0.4708, 0.2888, 0.4895, 0.2119], device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0257, 0.0219, 0.0283, 0.0239, 0.0202, 0.0245, 0.0200], 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') 2023-03-26 06:06:10,272 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-26 06:06:12,350 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.2687, 1.4762, 1.5644, 0.8840, 1.4555, 1.7053, 1.7241, 1.4227], device='cuda:3'), covar=tensor([0.1016, 0.0567, 0.0474, 0.0572, 0.0461, 0.0539, 0.0293, 0.0588], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0159, 0.0121, 0.0138, 0.0133, 0.0125, 0.0148, 0.0145], device='cuda:3'), out_proj_covar=tensor([9.8718e-05, 1.1766e-04, 8.7427e-05, 1.0089e-04, 9.5565e-05, 9.2616e-05, 1.0939e-04, 1.0735e-04], device='cuda:3') 2023-03-26 06:06:16,532 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31470.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:06:42,531 INFO [finetune.py:976] (3/7) Epoch 6, batch 2850, loss[loss=0.2538, simple_loss=0.3018, pruned_loss=0.1029, over 4014.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2693, pruned_loss=0.07535, over 956058.74 frames. ], batch size: 65, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:06:43,873 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31491.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 06:07:26,133 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.153e+02 1.576e+02 1.974e+02 2.520e+02 5.037e+02, threshold=3.948e+02, percent-clipped=2.0 2023-03-26 06:07:41,444 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9013, 1.8019, 1.9096, 1.2022, 2.0144, 1.9391, 1.8972, 1.5394], device='cuda:3'), covar=tensor([0.0616, 0.0686, 0.0672, 0.0960, 0.0545, 0.0748, 0.0591, 0.1154], device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0135, 0.0146, 0.0129, 0.0115, 0.0145, 0.0147, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 06:07:42,540 INFO [finetune.py:976] (3/7) Epoch 6, batch 2900, loss[loss=0.2068, simple_loss=0.2762, pruned_loss=0.06867, over 4827.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2725, pruned_loss=0.07636, over 956116.63 frames. ], batch size: 40, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:07:43,336 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.82 vs. limit=5.0 2023-03-26 06:07:59,556 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31552.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 06:08:04,326 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31559.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:08:15,062 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6295, 1.4922, 1.4809, 1.5618, 1.1979, 3.5515, 1.5305, 2.0625], device='cuda:3'), covar=tensor([0.3439, 0.2501, 0.2135, 0.2305, 0.1802, 0.0176, 0.2586, 0.1264], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0115, 0.0119, 0.0123, 0.0118, 0.0099, 0.0102, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 06:08:19,206 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31581.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:08:27,360 INFO [finetune.py:976] (3/7) Epoch 6, batch 2950, loss[loss=0.2098, simple_loss=0.2614, pruned_loss=0.07912, over 4865.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2745, pruned_loss=0.07683, over 955059.46 frames. ], batch size: 31, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:08:33,428 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31590.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:08:47,781 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=31607.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:08:55,268 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9243, 1.7875, 1.6014, 1.7966, 1.8719, 1.6023, 2.1599, 1.8735], device='cuda:3'), covar=tensor([0.1322, 0.2446, 0.2808, 0.2504, 0.2242, 0.1550, 0.2855, 0.1837], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0193, 0.0237, 0.0254, 0.0232, 0.0192, 0.0212, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 06:08:59,952 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.136e+02 1.780e+02 2.105e+02 2.565e+02 4.082e+02, threshold=4.210e+02, percent-clipped=1.0 2023-03-26 06:09:02,950 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=31629.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:09:09,343 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=31638.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:09:09,893 INFO [finetune.py:976] (3/7) Epoch 6, batch 3000, loss[loss=0.2014, simple_loss=0.2789, pruned_loss=0.06196, over 4738.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2773, pruned_loss=0.07858, over 954412.22 frames. ], batch size: 54, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:09:09,893 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 06:09:15,444 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5114, 1.2741, 1.2511, 1.3301, 1.6409, 1.5799, 1.3956, 1.2251], device='cuda:3'), covar=tensor([0.0269, 0.0335, 0.0618, 0.0316, 0.0242, 0.0388, 0.0316, 0.0450], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0111, 0.0137, 0.0116, 0.0103, 0.0100, 0.0090, 0.0108], device='cuda:3'), out_proj_covar=tensor([6.7290e-05, 8.7190e-05, 1.0945e-04, 9.1587e-05, 8.1257e-05, 7.4104e-05, 6.8327e-05, 8.4034e-05], device='cuda:3') 2023-03-26 06:09:16,759 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6284, 1.6155, 2.0106, 1.3007, 1.6700, 1.8230, 1.5508, 2.0665], device='cuda:3'), covar=tensor([0.1413, 0.2278, 0.1364, 0.2004, 0.1006, 0.1439, 0.3183, 0.0898], device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0206, 0.0199, 0.0196, 0.0184, 0.0221, 0.0218, 0.0203], 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') 2023-03-26 06:09:23,490 INFO [finetune.py:1010] (3/7) Epoch 6, validation: loss=0.1625, simple_loss=0.2344, pruned_loss=0.04534, over 2265189.00 frames. 2023-03-26 06:09:23,490 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 06:09:33,013 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.72 vs. limit=5.0 2023-03-26 06:09:55,818 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31667.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 06:10:13,517 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0287, 1.4799, 0.8313, 1.8512, 2.1052, 1.8157, 1.6269, 1.8058], device='cuda:3'), covar=tensor([0.1970, 0.2719, 0.2841, 0.1572, 0.2584, 0.2541, 0.2001, 0.2868], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0099, 0.0116, 0.0093, 0.0125, 0.0097, 0.0101, 0.0093], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 06:10:19,291 INFO [finetune.py:976] (3/7) Epoch 6, batch 3050, loss[loss=0.1923, simple_loss=0.2618, pruned_loss=0.06146, over 4893.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.2794, pruned_loss=0.07966, over 954748.80 frames. ], batch size: 32, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:10:20,014 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4248, 1.3903, 1.3665, 1.6047, 1.5670, 3.0446, 1.3861, 1.6658], device='cuda:3'), covar=tensor([0.0975, 0.1794, 0.1114, 0.0975, 0.1623, 0.0302, 0.1447, 0.1630], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0082, 0.0077, 0.0079, 0.0092, 0.0084, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 06:10:27,504 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1897, 2.0752, 1.6802, 0.7683, 1.8945, 1.8080, 1.6394, 1.8654], device='cuda:3'), covar=tensor([0.0853, 0.0715, 0.1421, 0.2155, 0.1254, 0.1947, 0.2015, 0.0899], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0202, 0.0202, 0.0190, 0.0217, 0.0209, 0.0221, 0.0200], 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') 2023-03-26 06:10:29,959 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31704.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:10:34,228 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9486, 2.0595, 2.0530, 1.2780, 2.2728, 2.1956, 2.0028, 1.7962], device='cuda:3'), covar=tensor([0.0661, 0.0604, 0.0666, 0.0973, 0.0511, 0.0635, 0.0638, 0.1017], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0135, 0.0144, 0.0129, 0.0113, 0.0145, 0.0146, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 06:10:36,649 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=31715.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 06:10:43,059 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.029e+02 1.708e+02 2.072e+02 2.420e+02 4.871e+02, threshold=4.144e+02, percent-clipped=1.0 2023-03-26 06:10:44,408 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2337, 1.9454, 1.4576, 0.6715, 1.6609, 1.9368, 1.7981, 1.7542], device='cuda:3'), covar=tensor([0.0694, 0.0736, 0.1119, 0.1793, 0.1231, 0.1597, 0.1630, 0.0769], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0202, 0.0201, 0.0190, 0.0217, 0.0209, 0.0221, 0.0200], 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') 2023-03-26 06:10:59,686 INFO [finetune.py:976] (3/7) Epoch 6, batch 3100, loss[loss=0.2182, simple_loss=0.2862, pruned_loss=0.07515, over 4909.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2774, pruned_loss=0.07888, over 955165.61 frames. ], batch size: 37, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:11:17,252 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5768, 3.4065, 3.2027, 1.4111, 3.4772, 2.5453, 0.8066, 2.3579], device='cuda:3'), covar=tensor([0.2303, 0.1846, 0.1711, 0.3581, 0.1255, 0.1069, 0.4416, 0.1505], device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0172, 0.0163, 0.0128, 0.0156, 0.0123, 0.0145, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 06:11:20,233 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31765.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:11:20,279 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31765.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:11:35,800 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31788.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:11:36,281 INFO [finetune.py:976] (3/7) Epoch 6, batch 3150, loss[loss=0.2199, simple_loss=0.2703, pruned_loss=0.08474, over 4861.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2743, pruned_loss=0.07771, over 956453.33 frames. ], batch size: 49, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:11:48,782 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-26 06:12:00,893 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31823.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:12:01,992 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.748e+01 1.682e+02 2.020e+02 2.535e+02 4.605e+02, threshold=4.040e+02, percent-clipped=1.0 2023-03-26 06:12:05,661 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1075, 1.9644, 2.1799, 1.1543, 2.3008, 2.5561, 2.1850, 1.9891], device='cuda:3'), covar=tensor([0.0962, 0.0705, 0.0473, 0.0684, 0.0533, 0.0553, 0.0454, 0.0645], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0158, 0.0121, 0.0137, 0.0133, 0.0124, 0.0147, 0.0145], device='cuda:3'), out_proj_covar=tensor([9.8205e-05, 1.1675e-04, 8.7371e-05, 1.0032e-04, 9.5853e-05, 9.1954e-05, 1.0890e-04, 1.0681e-04], device='cuda:3') 2023-03-26 06:12:15,951 INFO [finetune.py:976] (3/7) Epoch 6, batch 3200, loss[loss=0.1994, simple_loss=0.2642, pruned_loss=0.06729, over 4848.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2705, pruned_loss=0.07572, over 956307.62 frames. ], batch size: 44, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:12:25,648 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31847.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 06:12:32,373 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31849.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:13:06,130 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31878.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:13:11,207 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31884.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:13:14,085 INFO [finetune.py:976] (3/7) Epoch 6, batch 3250, loss[loss=0.2944, simple_loss=0.3387, pruned_loss=0.1251, over 4742.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2726, pruned_loss=0.0772, over 952939.35 frames. ], batch size: 54, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:13:40,884 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.56 vs. limit=5.0 2023-03-26 06:13:50,740 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.8717, 1.5005, 1.4506, 0.9291, 1.5883, 1.6778, 1.6901, 1.4149], device='cuda:3'), covar=tensor([0.0672, 0.0552, 0.0423, 0.0511, 0.0447, 0.0437, 0.0308, 0.0521], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0157, 0.0120, 0.0137, 0.0133, 0.0124, 0.0147, 0.0144], device='cuda:3'), out_proj_covar=tensor([9.7938e-05, 1.1618e-04, 8.6918e-05, 9.9924e-05, 9.5248e-05, 9.1514e-05, 1.0860e-04, 1.0669e-04], device='cuda:3') 2023-03-26 06:13:51,942 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31915.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:14:01,794 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.169e+02 1.695e+02 2.048e+02 2.376e+02 4.231e+02, threshold=4.096e+02, percent-clipped=1.0 2023-03-26 06:14:17,633 INFO [finetune.py:976] (3/7) Epoch 6, batch 3300, loss[loss=0.2492, simple_loss=0.3127, pruned_loss=0.09282, over 4789.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2764, pruned_loss=0.07833, over 951781.60 frames. ], batch size: 51, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:14:17,761 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31939.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:14:29,143 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31950.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:14:35,335 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7184, 1.2365, 0.8035, 1.5547, 2.0678, 1.3314, 1.5178, 1.6175], device='cuda:3'), covar=tensor([0.1619, 0.2230, 0.2326, 0.1325, 0.2094, 0.2272, 0.1443, 0.2150], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0098, 0.0115, 0.0093, 0.0124, 0.0096, 0.0100, 0.0093], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 06:14:46,579 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31976.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:14:54,839 INFO [finetune.py:976] (3/7) Epoch 6, batch 3350, loss[loss=0.2402, simple_loss=0.3039, pruned_loss=0.08824, over 4727.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2779, pruned_loss=0.0787, over 953567.97 frames. ], batch size: 59, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:15:00,970 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4538, 2.2540, 1.8309, 2.5081, 2.4588, 1.9779, 2.9861, 2.4371], device='cuda:3'), covar=tensor([0.1492, 0.3303, 0.3863, 0.3439, 0.2934, 0.1872, 0.3401, 0.2195], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0191, 0.0235, 0.0253, 0.0231, 0.0190, 0.0211, 0.0190], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 06:15:11,486 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32011.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:15:13,907 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8463, 1.8292, 1.6395, 2.0263, 2.4963, 1.9842, 1.4805, 1.4797], device='cuda:3'), covar=tensor([0.2369, 0.2121, 0.2087, 0.1808, 0.1868, 0.1175, 0.2728, 0.2115], device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0209, 0.0203, 0.0185, 0.0237, 0.0175, 0.0213, 0.0190], 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') 2023-03-26 06:15:22,687 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.186e+02 1.743e+02 2.097e+02 2.540e+02 4.089e+02, threshold=4.194e+02, percent-clipped=0.0 2023-03-26 06:15:41,803 INFO [finetune.py:976] (3/7) Epoch 6, batch 3400, loss[loss=0.1734, simple_loss=0.2463, pruned_loss=0.0503, over 4750.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2771, pruned_loss=0.07792, over 951356.58 frames. ], batch size: 27, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:15:51,304 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6860, 1.2150, 0.9340, 1.5503, 2.0558, 1.2576, 1.4725, 1.6165], device='cuda:3'), covar=tensor([0.1496, 0.2093, 0.2121, 0.1223, 0.1903, 0.2035, 0.1419, 0.1915], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0098, 0.0115, 0.0093, 0.0124, 0.0096, 0.0100, 0.0093], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 06:16:04,449 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32060.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:16:13,849 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32065.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:16:41,401 INFO [finetune.py:976] (3/7) Epoch 6, batch 3450, loss[loss=0.2323, simple_loss=0.2868, pruned_loss=0.08896, over 4859.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2763, pruned_loss=0.07736, over 951809.49 frames. ], batch size: 31, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:16:42,608 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.3108, 3.7866, 3.9399, 4.1961, 4.0872, 3.8573, 4.4068, 1.3759], device='cuda:3'), covar=tensor([0.0751, 0.0741, 0.0815, 0.0950, 0.1101, 0.1292, 0.0592, 0.5042], device='cuda:3'), in_proj_covar=tensor([0.0356, 0.0244, 0.0276, 0.0294, 0.0334, 0.0285, 0.0303, 0.0299], 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') 2023-03-26 06:17:07,732 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=32113.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:17:16,937 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.122e+02 1.573e+02 1.926e+02 2.516e+02 4.351e+02, threshold=3.853e+02, percent-clipped=2.0 2023-03-26 06:17:20,133 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-03-26 06:17:25,471 INFO [finetune.py:976] (3/7) Epoch 6, batch 3500, loss[loss=0.2217, simple_loss=0.2723, pruned_loss=0.08557, over 4892.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2753, pruned_loss=0.07746, over 955137.71 frames. ], batch size: 32, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:17:29,069 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32144.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:17:30,901 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32147.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 06:18:09,982 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32179.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:18:15,990 INFO [finetune.py:976] (3/7) Epoch 6, batch 3550, loss[loss=0.1995, simple_loss=0.2598, pruned_loss=0.06964, over 4901.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.272, pruned_loss=0.07622, over 954762.98 frames. ], batch size: 36, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:18:19,693 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=32195.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 06:18:40,466 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.015e+02 1.617e+02 1.889e+02 2.318e+02 4.823e+02, threshold=3.777e+02, percent-clipped=2.0 2023-03-26 06:18:52,029 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32234.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:19:00,028 INFO [finetune.py:976] (3/7) Epoch 6, batch 3600, loss[loss=0.2549, simple_loss=0.2971, pruned_loss=0.1063, over 4896.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2685, pruned_loss=0.07474, over 955287.46 frames. ], batch size: 32, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:19:36,022 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32271.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:19:56,037 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7683, 1.2162, 0.8346, 1.5256, 2.0625, 1.3808, 1.3270, 1.5979], device='cuda:3'), covar=tensor([0.1517, 0.2194, 0.2303, 0.1293, 0.2037, 0.2037, 0.1559, 0.2020], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0097, 0.0114, 0.0092, 0.0123, 0.0095, 0.0100, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 06:19:57,774 INFO [finetune.py:976] (3/7) Epoch 6, batch 3650, loss[loss=0.2353, simple_loss=0.2994, pruned_loss=0.0856, over 4856.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.273, pruned_loss=0.07721, over 954982.23 frames. ], batch size: 44, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:20:14,265 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32306.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:20:26,743 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.020e+02 1.776e+02 2.129e+02 2.587e+02 5.126e+02, threshold=4.259e+02, percent-clipped=5.0 2023-03-26 06:20:43,053 INFO [finetune.py:976] (3/7) Epoch 6, batch 3700, loss[loss=0.1725, simple_loss=0.244, pruned_loss=0.05052, over 4799.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2754, pruned_loss=0.07743, over 955726.71 frames. ], batch size: 29, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:20:56,522 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32360.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:21:16,590 INFO [finetune.py:976] (3/7) Epoch 6, batch 3750, loss[loss=0.1896, simple_loss=0.2347, pruned_loss=0.07221, over 4085.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2762, pruned_loss=0.07737, over 954541.08 frames. ], batch size: 17, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:21:37,052 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=32408.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:21:48,323 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.045e+02 1.626e+02 1.929e+02 2.294e+02 3.909e+02, threshold=3.857e+02, percent-clipped=0.0 2023-03-26 06:21:58,659 INFO [finetune.py:976] (3/7) Epoch 6, batch 3800, loss[loss=0.2477, simple_loss=0.2928, pruned_loss=0.1013, over 4913.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2789, pruned_loss=0.07904, over 955929.89 frames. ], batch size: 37, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:22:01,790 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32444.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:22:24,390 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32479.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:22:31,249 INFO [finetune.py:976] (3/7) Epoch 6, batch 3850, loss[loss=0.1871, simple_loss=0.2508, pruned_loss=0.06175, over 4891.00 frames. ], tot_loss[loss=0.2169, simple_loss=0.2774, pruned_loss=0.07817, over 956315.15 frames. ], batch size: 35, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:22:33,639 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=32492.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:22:53,999 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.617e+01 1.620e+02 2.056e+02 2.694e+02 5.558e+02, threshold=4.113e+02, percent-clipped=4.0 2023-03-26 06:22:55,288 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=32527.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:22:57,745 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-26 06:23:00,111 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32534.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:23:04,473 INFO [finetune.py:976] (3/7) Epoch 6, batch 3900, loss[loss=0.2167, simple_loss=0.2769, pruned_loss=0.07824, over 4940.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2738, pruned_loss=0.07701, over 956723.55 frames. ], batch size: 38, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:23:19,379 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1401, 3.5946, 3.7650, 3.9942, 3.9043, 3.6959, 4.2242, 1.4862], device='cuda:3'), covar=tensor([0.0701, 0.0729, 0.0748, 0.0757, 0.1020, 0.1277, 0.0546, 0.4688], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0239, 0.0272, 0.0291, 0.0330, 0.0281, 0.0299, 0.0295], 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') 2023-03-26 06:23:19,391 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5887, 1.4496, 2.0901, 3.1669, 2.1417, 2.2880, 0.9948, 2.4669], device='cuda:3'), covar=tensor([0.1638, 0.1478, 0.1240, 0.0536, 0.0775, 0.1493, 0.1823, 0.0668], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0119, 0.0136, 0.0167, 0.0103, 0.0142, 0.0129, 0.0104], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 06:23:24,752 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32571.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:23:31,335 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=32582.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:23:36,052 INFO [finetune.py:976] (3/7) Epoch 6, batch 3950, loss[loss=0.2381, simple_loss=0.2781, pruned_loss=0.09909, over 4339.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.27, pruned_loss=0.07543, over 956772.07 frames. ], batch size: 19, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:23:53,914 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32606.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:24:07,661 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=32619.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:24:11,253 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.084e+02 1.672e+02 2.141e+02 2.509e+02 4.478e+02, threshold=4.281e+02, percent-clipped=2.0 2023-03-26 06:24:20,844 INFO [finetune.py:976] (3/7) Epoch 6, batch 4000, loss[loss=0.1846, simple_loss=0.2504, pruned_loss=0.05935, over 4903.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.269, pruned_loss=0.07516, over 956701.79 frames. ], batch size: 43, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:24:20,982 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8018, 1.7756, 1.6795, 2.0237, 2.1022, 2.0145, 1.3870, 1.4625], device='cuda:3'), covar=tensor([0.2156, 0.1931, 0.1717, 0.1544, 0.1836, 0.1059, 0.2568, 0.1805], device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0209, 0.0202, 0.0184, 0.0237, 0.0174, 0.0212, 0.0190], 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') 2023-03-26 06:24:30,955 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=32654.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:24:31,628 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5830, 1.4710, 1.4655, 1.5339, 1.0435, 3.2609, 1.3765, 1.8097], device='cuda:3'), covar=tensor([0.3178, 0.2347, 0.1992, 0.2208, 0.1965, 0.0225, 0.2744, 0.1305], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0115, 0.0118, 0.0122, 0.0118, 0.0099, 0.0102, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 06:25:04,487 INFO [finetune.py:976] (3/7) Epoch 6, batch 4050, loss[loss=0.2119, simple_loss=0.2928, pruned_loss=0.06547, over 4916.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2741, pruned_loss=0.07786, over 954380.80 frames. ], batch size: 42, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:25:12,868 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.4598, 2.9400, 2.6671, 1.4925, 2.8457, 2.5608, 2.3615, 2.4726], device='cuda:3'), covar=tensor([0.0800, 0.0961, 0.1714, 0.2264, 0.1693, 0.1681, 0.1814, 0.1154], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0201, 0.0200, 0.0187, 0.0214, 0.0207, 0.0219, 0.0198], 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') 2023-03-26 06:25:28,440 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.148e+02 1.831e+02 2.133e+02 2.637e+02 5.226e+02, threshold=4.267e+02, percent-clipped=1.0 2023-03-26 06:25:42,720 INFO [finetune.py:976] (3/7) Epoch 6, batch 4100, loss[loss=0.2517, simple_loss=0.3152, pruned_loss=0.09407, over 4837.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.276, pruned_loss=0.07841, over 952581.61 frames. ], batch size: 47, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:26:34,201 INFO [finetune.py:976] (3/7) Epoch 6, batch 4150, loss[loss=0.1482, simple_loss=0.2159, pruned_loss=0.04031, over 4728.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2786, pruned_loss=0.07985, over 951657.62 frames. ], batch size: 23, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:27:18,097 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.203e+02 1.784e+02 2.149e+02 2.523e+02 4.029e+02, threshold=4.299e+02, percent-clipped=0.0 2023-03-26 06:27:21,156 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5854, 1.4245, 1.4167, 1.4753, 1.3059, 3.4756, 1.6317, 2.0920], device='cuda:3'), covar=tensor([0.4098, 0.3058, 0.2497, 0.2847, 0.1832, 0.0244, 0.2514, 0.1255], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0115, 0.0118, 0.0123, 0.0119, 0.0099, 0.0102, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 06:27:37,296 INFO [finetune.py:976] (3/7) Epoch 6, batch 4200, loss[loss=0.1917, simple_loss=0.2566, pruned_loss=0.0634, over 4894.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2786, pruned_loss=0.07939, over 951386.54 frames. ], batch size: 36, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:28:34,720 INFO [finetune.py:976] (3/7) Epoch 6, batch 4250, loss[loss=0.2476, simple_loss=0.2939, pruned_loss=0.1006, over 4823.00 frames. ], tot_loss[loss=0.216, simple_loss=0.2758, pruned_loss=0.07811, over 949219.30 frames. ], batch size: 40, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:28:40,563 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-26 06:29:25,479 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.113e+02 1.595e+02 1.920e+02 2.303e+02 3.727e+02, threshold=3.841e+02, percent-clipped=0.0 2023-03-26 06:29:44,594 INFO [finetune.py:976] (3/7) Epoch 6, batch 4300, loss[loss=0.2011, simple_loss=0.2592, pruned_loss=0.0715, over 4904.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2722, pruned_loss=0.07629, over 952382.68 frames. ], batch size: 46, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:30:17,422 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32966.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:30:47,045 INFO [finetune.py:976] (3/7) Epoch 6, batch 4350, loss[loss=0.1362, simple_loss=0.2046, pruned_loss=0.0339, over 4834.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2678, pruned_loss=0.07437, over 954820.83 frames. ], batch size: 25, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:30:49,811 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.29 vs. limit=5.0 2023-03-26 06:31:32,014 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.139e+02 1.730e+02 2.041e+02 2.589e+02 3.941e+02, threshold=4.082e+02, percent-clipped=1.0 2023-03-26 06:31:33,345 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33027.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:31:50,660 INFO [finetune.py:976] (3/7) Epoch 6, batch 4400, loss[loss=0.2637, simple_loss=0.3185, pruned_loss=0.1045, over 4179.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2686, pruned_loss=0.07471, over 954600.89 frames. ], batch size: 65, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:31:52,689 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-03-26 06:32:00,125 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5480, 1.5683, 1.3001, 1.3371, 1.7334, 1.6820, 1.5395, 1.2997], device='cuda:3'), covar=tensor([0.0259, 0.0280, 0.0499, 0.0315, 0.0186, 0.0520, 0.0320, 0.0397], device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0110, 0.0135, 0.0114, 0.0102, 0.0098, 0.0089, 0.0107], device='cuda:3'), out_proj_covar=tensor([6.6613e-05, 8.6402e-05, 1.0812e-04, 9.0167e-05, 8.0504e-05, 7.2943e-05, 6.7925e-05, 8.3406e-05], device='cuda:3') 2023-03-26 06:32:31,731 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33069.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:32:36,541 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9152, 1.2604, 1.8002, 1.7503, 1.6230, 1.5916, 1.6433, 1.5968], device='cuda:3'), covar=tensor([0.4916, 0.6250, 0.5516, 0.5605, 0.6915, 0.4903, 0.6763, 0.4891], device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0245, 0.0254, 0.0256, 0.0242, 0.0219, 0.0273, 0.0224], 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') 2023-03-26 06:32:54,583 INFO [finetune.py:976] (3/7) Epoch 6, batch 4450, loss[loss=0.2152, simple_loss=0.2812, pruned_loss=0.07458, over 4802.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2727, pruned_loss=0.07635, over 954549.77 frames. ], batch size: 45, lr: 3.90e-03, grad_scale: 16.0 2023-03-26 06:33:39,235 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.174e+02 1.714e+02 2.125e+02 2.690e+02 4.211e+02, threshold=4.250e+02, percent-clipped=1.0 2023-03-26 06:33:47,518 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33130.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:33:58,504 INFO [finetune.py:976] (3/7) Epoch 6, batch 4500, loss[loss=0.1771, simple_loss=0.2328, pruned_loss=0.06067, over 4694.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.275, pruned_loss=0.07729, over 957056.98 frames. ], batch size: 23, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:34:19,043 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33155.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:34:19,053 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.2049, 1.3138, 1.3712, 0.6142, 1.1390, 1.5507, 1.6189, 1.2407], device='cuda:3'), covar=tensor([0.0884, 0.0573, 0.0476, 0.0503, 0.0495, 0.0471, 0.0300, 0.0587], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0159, 0.0122, 0.0139, 0.0134, 0.0125, 0.0148, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.7805e-05, 1.1734e-04, 8.8274e-05, 1.0125e-04, 9.6000e-05, 9.2417e-05, 1.0997e-04, 1.0834e-04], device='cuda:3') 2023-03-26 06:35:01,062 INFO [finetune.py:976] (3/7) Epoch 6, batch 4550, loss[loss=0.2779, simple_loss=0.317, pruned_loss=0.1194, over 4146.00 frames. ], tot_loss[loss=0.2166, simple_loss=0.2769, pruned_loss=0.07814, over 954997.14 frames. ], batch size: 66, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:35:33,150 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33216.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:35:44,645 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 1.755e+02 2.084e+02 2.335e+02 4.621e+02, threshold=4.168e+02, percent-clipped=2.0 2023-03-26 06:36:05,046 INFO [finetune.py:976] (3/7) Epoch 6, batch 4600, loss[loss=0.1731, simple_loss=0.2362, pruned_loss=0.05502, over 4780.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2761, pruned_loss=0.07732, over 955861.95 frames. ], batch size: 29, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:37:07,545 INFO [finetune.py:976] (3/7) Epoch 6, batch 4650, loss[loss=0.2297, simple_loss=0.2909, pruned_loss=0.08422, over 4891.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2732, pruned_loss=0.07633, over 956562.44 frames. ], batch size: 35, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:37:48,442 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33322.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:37:50,592 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.141e+02 1.572e+02 1.961e+02 2.434e+02 3.752e+02, threshold=3.921e+02, percent-clipped=0.0 2023-03-26 06:38:10,951 INFO [finetune.py:976] (3/7) Epoch 6, batch 4700, loss[loss=0.1547, simple_loss=0.2307, pruned_loss=0.0393, over 4753.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2697, pruned_loss=0.0746, over 956205.11 frames. ], batch size: 27, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:38:29,941 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1279, 1.9722, 1.3833, 0.5701, 1.6867, 1.7400, 1.5285, 1.7808], device='cuda:3'), covar=tensor([0.0907, 0.0759, 0.1429, 0.1968, 0.1295, 0.2292, 0.2372, 0.0825], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0201, 0.0199, 0.0188, 0.0215, 0.0207, 0.0219, 0.0198], 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') 2023-03-26 06:39:20,542 INFO [finetune.py:976] (3/7) Epoch 6, batch 4750, loss[loss=0.1793, simple_loss=0.2456, pruned_loss=0.0565, over 4894.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2664, pruned_loss=0.07334, over 956833.71 frames. ], batch size: 32, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:40:04,157 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.123e+02 1.633e+02 1.917e+02 2.350e+02 3.562e+02, threshold=3.835e+02, percent-clipped=0.0 2023-03-26 06:40:04,244 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33425.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:40:24,234 INFO [finetune.py:976] (3/7) Epoch 6, batch 4800, loss[loss=0.1801, simple_loss=0.2582, pruned_loss=0.05104, over 4776.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2701, pruned_loss=0.07513, over 952942.95 frames. ], batch size: 28, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:40:35,531 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.4051, 4.5667, 4.8815, 5.2559, 5.1035, 4.8055, 5.5035, 1.7861], device='cuda:3'), covar=tensor([0.0681, 0.0893, 0.0614, 0.0931, 0.1111, 0.1449, 0.0493, 0.5161], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0238, 0.0272, 0.0291, 0.0332, 0.0281, 0.0300, 0.0296], 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') 2023-03-26 06:41:07,492 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33472.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:41:28,615 INFO [finetune.py:976] (3/7) Epoch 6, batch 4850, loss[loss=0.2003, simple_loss=0.2407, pruned_loss=0.08001, over 3967.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2741, pruned_loss=0.07638, over 952130.36 frames. ], batch size: 17, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:41:59,904 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33511.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:42:14,019 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.198e+02 1.841e+02 2.130e+02 2.477e+02 4.983e+02, threshold=4.260e+02, percent-clipped=3.0 2023-03-26 06:42:24,098 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33533.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:42:32,810 INFO [finetune.py:976] (3/7) Epoch 6, batch 4900, loss[loss=0.2504, simple_loss=0.3188, pruned_loss=0.091, over 4805.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2757, pruned_loss=0.07755, over 951078.10 frames. ], batch size: 38, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:43:36,446 INFO [finetune.py:976] (3/7) Epoch 6, batch 4950, loss[loss=0.1939, simple_loss=0.2662, pruned_loss=0.0608, over 4892.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2766, pruned_loss=0.07732, over 954144.48 frames. ], batch size: 43, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:44:20,460 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33622.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:44:22,686 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.562e+01 1.670e+02 2.087e+02 2.382e+02 5.310e+02, threshold=4.173e+02, percent-clipped=3.0 2023-03-26 06:44:41,928 INFO [finetune.py:976] (3/7) Epoch 6, batch 5000, loss[loss=0.165, simple_loss=0.234, pruned_loss=0.04807, over 4747.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.2745, pruned_loss=0.0764, over 955499.13 frames. ], batch size: 28, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:45:14,312 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=33670.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:45:19,059 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33677.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:45:26,105 INFO [finetune.py:976] (3/7) Epoch 6, batch 5050, loss[loss=0.2168, simple_loss=0.2763, pruned_loss=0.0787, over 4867.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2733, pruned_loss=0.07654, over 955044.37 frames. ], batch size: 31, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:45:26,840 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8239, 1.7205, 1.6265, 1.7882, 1.2972, 4.0649, 1.8141, 2.3076], device='cuda:3'), covar=tensor([0.3116, 0.2324, 0.1920, 0.2069, 0.1662, 0.0120, 0.2298, 0.1195], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0115, 0.0118, 0.0122, 0.0118, 0.0098, 0.0102, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 06:45:50,298 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.151e+02 1.621e+02 1.877e+02 2.380e+02 3.773e+02, threshold=3.754e+02, percent-clipped=0.0 2023-03-26 06:45:50,424 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33725.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:45:56,358 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5371, 1.6575, 1.3099, 1.4782, 1.7935, 1.7780, 1.5864, 1.3579], device='cuda:3'), covar=tensor([0.0320, 0.0231, 0.0548, 0.0280, 0.0200, 0.0406, 0.0338, 0.0380], device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0109, 0.0135, 0.0114, 0.0102, 0.0098, 0.0089, 0.0106], device='cuda:3'), out_proj_covar=tensor([6.6705e-05, 8.5406e-05, 1.0811e-04, 9.0028e-05, 8.0454e-05, 7.2603e-05, 6.7422e-05, 8.3005e-05], device='cuda:3') 2023-03-26 06:45:58,846 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33738.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 06:45:59,300 INFO [finetune.py:976] (3/7) Epoch 6, batch 5100, loss[loss=0.2627, simple_loss=0.3137, pruned_loss=0.1059, over 4837.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2696, pruned_loss=0.07495, over 956752.28 frames. ], batch size: 47, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:46:22,563 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=33773.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:46:24,783 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-03-26 06:46:29,768 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.7785, 3.3479, 3.4720, 3.6551, 3.5648, 3.3730, 3.8941, 1.2266], device='cuda:3'), covar=tensor([0.1049, 0.0824, 0.0833, 0.1178, 0.1553, 0.1603, 0.0849, 0.5171], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0241, 0.0273, 0.0293, 0.0333, 0.0283, 0.0301, 0.0296], 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') 2023-03-26 06:46:32,749 INFO [finetune.py:976] (3/7) Epoch 6, batch 5150, loss[loss=0.2158, simple_loss=0.2742, pruned_loss=0.07876, over 4836.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2695, pruned_loss=0.07487, over 955652.04 frames. ], batch size: 33, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:46:40,225 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5213, 1.6452, 1.2776, 1.3934, 1.8467, 1.8176, 1.6580, 1.3771], device='cuda:3'), covar=tensor([0.0365, 0.0353, 0.0575, 0.0395, 0.0221, 0.0518, 0.0298, 0.0463], device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0109, 0.0136, 0.0115, 0.0103, 0.0098, 0.0089, 0.0107], device='cuda:3'), out_proj_covar=tensor([6.7211e-05, 8.5952e-05, 1.0875e-04, 9.0449e-05, 8.0889e-05, 7.3118e-05, 6.7743e-05, 8.3526e-05], device='cuda:3') 2023-03-26 06:46:48,246 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33811.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:46:54,033 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1580, 1.8713, 2.0719, 0.8835, 2.3513, 2.6545, 2.0684, 1.9065], device='cuda:3'), covar=tensor([0.1098, 0.0976, 0.0554, 0.0916, 0.0529, 0.0505, 0.0569, 0.0801], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0158, 0.0121, 0.0139, 0.0133, 0.0125, 0.0148, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.7451e-05, 1.1717e-04, 8.8073e-05, 1.0163e-04, 9.5286e-05, 9.2083e-05, 1.0967e-04, 1.0825e-04], device='cuda:3') 2023-03-26 06:46:57,549 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.284e+02 1.768e+02 2.107e+02 2.614e+02 3.782e+02, threshold=4.214e+02, percent-clipped=1.0 2023-03-26 06:46:59,476 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33828.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:47:06,626 INFO [finetune.py:976] (3/7) Epoch 6, batch 5200, loss[loss=0.2335, simple_loss=0.2839, pruned_loss=0.09159, over 4896.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.273, pruned_loss=0.07628, over 954160.15 frames. ], batch size: 35, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:47:19,509 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=33859.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:47:35,983 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.61 vs. limit=5.0 2023-03-26 06:47:44,970 INFO [finetune.py:976] (3/7) Epoch 6, batch 5250, loss[loss=0.233, simple_loss=0.2867, pruned_loss=0.08965, over 4100.00 frames. ], tot_loss[loss=0.2147, simple_loss=0.2752, pruned_loss=0.07705, over 952485.85 frames. ], batch size: 65, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:47:55,309 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-26 06:47:59,527 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 2023-03-26 06:47:59,830 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.9981, 1.3355, 1.0954, 1.2964, 1.4750, 2.4025, 1.2022, 1.4945], device='cuda:3'), covar=tensor([0.0885, 0.1435, 0.0971, 0.0838, 0.1372, 0.0376, 0.1245, 0.1378], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0081, 0.0077, 0.0079, 0.0092, 0.0084, 0.0084, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 06:48:10,024 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.099e+02 1.728e+02 2.112e+02 2.559e+02 4.196e+02, threshold=4.224e+02, percent-clipped=0.0 2023-03-26 06:48:17,930 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.69 vs. limit=5.0 2023-03-26 06:48:18,714 INFO [finetune.py:976] (3/7) Epoch 6, batch 5300, loss[loss=0.2244, simple_loss=0.2865, pruned_loss=0.08118, over 4823.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2772, pruned_loss=0.07784, over 953621.83 frames. ], batch size: 39, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:48:19,399 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33939.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:48:53,985 INFO [finetune.py:976] (3/7) Epoch 6, batch 5350, loss[loss=0.1895, simple_loss=0.2523, pruned_loss=0.06333, over 4918.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2765, pruned_loss=0.07715, over 954409.92 frames. ], batch size: 38, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:49:07,550 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34000.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:49:40,308 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.035e+02 1.730e+02 2.013e+02 2.437e+02 5.230e+02, threshold=4.026e+02, percent-clipped=3.0 2023-03-26 06:49:50,527 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34033.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 06:49:59,344 INFO [finetune.py:976] (3/7) Epoch 6, batch 5400, loss[loss=0.1722, simple_loss=0.2324, pruned_loss=0.05601, over 4455.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2746, pruned_loss=0.07695, over 956552.37 frames. ], batch size: 19, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:50:02,485 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34044.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:50:51,575 INFO [finetune.py:976] (3/7) Epoch 6, batch 5450, loss[loss=0.1489, simple_loss=0.2228, pruned_loss=0.03747, over 4770.00 frames. ], tot_loss[loss=0.2111, simple_loss=0.2717, pruned_loss=0.07528, over 957718.08 frames. ], batch size: 28, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:51:04,654 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34105.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:51:17,056 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.064e+02 1.588e+02 1.777e+02 2.163e+02 4.002e+02, threshold=3.553e+02, percent-clipped=0.0 2023-03-26 06:51:19,937 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34128.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:51:27,293 INFO [finetune.py:976] (3/7) Epoch 6, batch 5500, loss[loss=0.2246, simple_loss=0.2732, pruned_loss=0.088, over 4768.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2681, pruned_loss=0.07419, over 957162.39 frames. ], batch size: 23, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:51:31,017 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34145.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:51:36,921 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3640, 1.2057, 1.1771, 1.1775, 1.5291, 1.4489, 1.3403, 1.1679], device='cuda:3'), covar=tensor([0.0290, 0.0309, 0.0616, 0.0303, 0.0237, 0.0513, 0.0261, 0.0417], device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0111, 0.0138, 0.0116, 0.0104, 0.0099, 0.0090, 0.0108], device='cuda:3'), out_proj_covar=tensor([6.7933e-05, 8.6961e-05, 1.1027e-04, 9.1623e-05, 8.1992e-05, 7.3770e-05, 6.8746e-05, 8.4399e-05], device='cuda:3') 2023-03-26 06:51:50,704 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=34176.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:51:59,801 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.3022, 3.7376, 3.8718, 4.1495, 4.0793, 3.8433, 4.4338, 1.3647], device='cuda:3'), covar=tensor([0.0810, 0.0892, 0.0930, 0.1088, 0.1178, 0.1452, 0.0658, 0.5398], device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0241, 0.0273, 0.0293, 0.0333, 0.0284, 0.0302, 0.0295], 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') 2023-03-26 06:52:00,938 INFO [finetune.py:976] (3/7) Epoch 6, batch 5550, loss[loss=0.2784, simple_loss=0.3266, pruned_loss=0.1151, over 4812.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2693, pruned_loss=0.07444, over 955181.18 frames. ], batch size: 45, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:52:15,690 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34206.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:52:37,391 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.199e+02 1.703e+02 1.979e+02 2.376e+02 4.570e+02, threshold=3.959e+02, percent-clipped=2.0 2023-03-26 06:52:55,406 INFO [finetune.py:976] (3/7) Epoch 6, batch 5600, loss[loss=0.2173, simple_loss=0.2767, pruned_loss=0.07896, over 4844.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2741, pruned_loss=0.07611, over 953135.85 frames. ], batch size: 47, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:53:24,706 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7525, 1.3425, 1.0557, 1.7256, 2.0593, 1.4257, 1.5741, 1.7924], device='cuda:3'), covar=tensor([0.1616, 0.2176, 0.2046, 0.1226, 0.2148, 0.2025, 0.1498, 0.1916], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0098, 0.0115, 0.0092, 0.0124, 0.0095, 0.0100, 0.0093], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 06:53:54,499 INFO [finetune.py:976] (3/7) Epoch 6, batch 5650, loss[loss=0.2118, simple_loss=0.2781, pruned_loss=0.07277, over 4898.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.2787, pruned_loss=0.07746, over 952169.08 frames. ], batch size: 35, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:53:58,431 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34295.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:54:35,325 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.805e+01 1.675e+02 2.029e+02 2.481e+02 4.265e+02, threshold=4.057e+02, percent-clipped=3.0 2023-03-26 06:54:40,163 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34333.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:54:48,433 INFO [finetune.py:976] (3/7) Epoch 6, batch 5700, loss[loss=0.1632, simple_loss=0.222, pruned_loss=0.05219, over 4258.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2737, pruned_loss=0.07676, over 932751.36 frames. ], batch size: 18, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:55:16,275 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8799, 1.7726, 1.7282, 1.8669, 1.4426, 3.3580, 1.6737, 2.1505], device='cuda:3'), covar=tensor([0.3168, 0.2435, 0.1950, 0.2244, 0.1849, 0.0256, 0.2315, 0.1149], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0114, 0.0118, 0.0122, 0.0117, 0.0098, 0.0101, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 06:55:38,389 INFO [finetune.py:976] (3/7) Epoch 7, batch 0, loss[loss=0.2676, simple_loss=0.3192, pruned_loss=0.108, over 4813.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3192, pruned_loss=0.108, over 4813.00 frames. ], batch size: 55, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:55:38,389 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 06:55:55,928 INFO [finetune.py:1010] (3/7) Epoch 7, validation: loss=0.165, simple_loss=0.2365, pruned_loss=0.04677, over 2265189.00 frames. 2023-03-26 06:55:55,928 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 06:56:14,683 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=34381.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:56:36,914 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34400.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:56:59,209 INFO [finetune.py:976] (3/7) Epoch 7, batch 50, loss[loss=0.2344, simple_loss=0.2952, pruned_loss=0.08677, over 4812.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2763, pruned_loss=0.0775, over 216838.17 frames. ], batch size: 33, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:56:59,316 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1060, 1.3359, 1.2061, 1.3652, 1.4211, 2.4023, 1.2275, 1.4772], device='cuda:3'), covar=tensor([0.0973, 0.1618, 0.1107, 0.0904, 0.1506, 0.0368, 0.1371, 0.1498], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0081, 0.0076, 0.0079, 0.0092, 0.0083, 0.0084, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 06:57:09,544 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.105e+02 1.600e+02 2.022e+02 2.565e+02 5.766e+02, threshold=4.045e+02, percent-clipped=4.0 2023-03-26 06:57:19,387 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-03-26 06:57:30,164 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-03-26 06:58:05,579 INFO [finetune.py:976] (3/7) Epoch 7, batch 100, loss[loss=0.1834, simple_loss=0.2522, pruned_loss=0.05735, over 4759.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2689, pruned_loss=0.07382, over 380804.42 frames. ], batch size: 28, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:58:45,252 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34501.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 06:59:06,470 INFO [finetune.py:976] (3/7) Epoch 7, batch 150, loss[loss=0.2137, simple_loss=0.2613, pruned_loss=0.08304, over 4693.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.266, pruned_loss=0.07431, over 507690.66 frames. ], batch size: 23, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 06:59:17,659 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.007e+02 1.586e+02 1.902e+02 2.328e+02 6.438e+02, threshold=3.804e+02, percent-clipped=3.0 2023-03-26 06:59:21,505 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2068, 2.0458, 1.7317, 2.1065, 2.1864, 1.8869, 2.5530, 2.1510], device='cuda:3'), covar=tensor([0.1589, 0.3104, 0.3792, 0.3507, 0.2859, 0.1986, 0.4153, 0.2169], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0192, 0.0237, 0.0257, 0.0234, 0.0193, 0.0214, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 07:00:03,513 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9185, 1.3551, 1.7536, 1.7587, 1.5485, 1.5684, 1.6511, 1.6293], device='cuda:3'), covar=tensor([0.4781, 0.5971, 0.5011, 0.5539, 0.6782, 0.5177, 0.7241, 0.4851], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0243, 0.0252, 0.0254, 0.0241, 0.0218, 0.0271, 0.0224], 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') 2023-03-26 07:00:10,782 INFO [finetune.py:976] (3/7) Epoch 7, batch 200, loss[loss=0.197, simple_loss=0.2496, pruned_loss=0.07217, over 4761.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2679, pruned_loss=0.0765, over 608242.10 frames. ], batch size: 27, lr: 3.89e-03, grad_scale: 32.0 2023-03-26 07:00:42,233 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34593.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:00:43,444 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34595.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:01:00,775 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5272, 1.3964, 1.2850, 1.5436, 1.8208, 1.5364, 1.0960, 1.2818], device='cuda:3'), covar=tensor([0.2370, 0.2343, 0.2123, 0.1855, 0.2078, 0.1270, 0.3001, 0.2008], device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0209, 0.0203, 0.0187, 0.0238, 0.0176, 0.0213, 0.0191], 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') 2023-03-26 07:01:13,423 INFO [finetune.py:976] (3/7) Epoch 7, batch 250, loss[loss=0.1859, simple_loss=0.2468, pruned_loss=0.06252, over 4773.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2718, pruned_loss=0.07804, over 683505.01 frames. ], batch size: 28, lr: 3.88e-03, grad_scale: 32.0 2023-03-26 07:01:22,726 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34622.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:01:24,451 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.284e+02 1.737e+02 2.023e+02 2.550e+02 3.958e+02, threshold=4.047e+02, percent-clipped=1.0 2023-03-26 07:01:25,172 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34626.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:01:45,026 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=34643.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:01:51,974 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1267, 2.1838, 2.0725, 1.5174, 2.3501, 2.2875, 2.2838, 1.9162], device='cuda:3'), covar=tensor([0.0664, 0.0572, 0.0770, 0.0956, 0.0430, 0.0761, 0.0643, 0.1014], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0135, 0.0145, 0.0129, 0.0114, 0.0146, 0.0147, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 07:01:57,404 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34654.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:02:14,157 INFO [finetune.py:976] (3/7) Epoch 7, batch 300, loss[loss=0.2773, simple_loss=0.332, pruned_loss=0.1113, over 4745.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2758, pruned_loss=0.07839, over 742865.39 frames. ], batch size: 59, lr: 3.88e-03, grad_scale: 32.0 2023-03-26 07:02:21,562 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5175, 1.3801, 1.2153, 1.2925, 1.6401, 1.5847, 1.4093, 1.1886], device='cuda:3'), covar=tensor([0.0256, 0.0273, 0.0595, 0.0311, 0.0208, 0.0404, 0.0320, 0.0445], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0112, 0.0140, 0.0117, 0.0105, 0.0101, 0.0092, 0.0110], device='cuda:3'), out_proj_covar=tensor([6.9120e-05, 8.8286e-05, 1.1197e-04, 9.2425e-05, 8.2981e-05, 7.4870e-05, 6.9678e-05, 8.5705e-05], device='cuda:3') 2023-03-26 07:02:34,479 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34683.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 07:02:36,955 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34687.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:02:46,205 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8675, 4.5128, 4.2908, 2.5600, 4.7308, 3.5209, 0.9454, 3.1932], device='cuda:3'), covar=tensor([0.2324, 0.1621, 0.1379, 0.2820, 0.0745, 0.0805, 0.4403, 0.1282], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0170, 0.0161, 0.0127, 0.0154, 0.0121, 0.0144, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 07:02:55,167 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34700.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:03:06,657 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34710.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 07:03:15,679 INFO [finetune.py:976] (3/7) Epoch 7, batch 350, loss[loss=0.2301, simple_loss=0.295, pruned_loss=0.08263, over 4813.00 frames. ], tot_loss[loss=0.2172, simple_loss=0.2771, pruned_loss=0.07866, over 790103.53 frames. ], batch size: 40, lr: 3.88e-03, grad_scale: 32.0 2023-03-26 07:03:27,162 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.146e+02 1.664e+02 2.043e+02 2.496e+02 5.690e+02, threshold=4.087e+02, percent-clipped=3.0 2023-03-26 07:03:54,745 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=34748.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:04:16,439 INFO [finetune.py:976] (3/7) Epoch 7, batch 400, loss[loss=0.2135, simple_loss=0.2783, pruned_loss=0.07434, over 4909.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2781, pruned_loss=0.07848, over 828258.49 frames. ], batch size: 38, lr: 3.88e-03, grad_scale: 32.0 2023-03-26 07:04:24,036 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34771.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 07:04:24,604 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1021, 3.5666, 3.7602, 3.9121, 3.8359, 3.6549, 4.2121, 1.4227], device='cuda:3'), covar=tensor([0.0797, 0.0884, 0.0788, 0.1000, 0.1168, 0.1351, 0.0693, 0.5229], device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0243, 0.0275, 0.0295, 0.0335, 0.0284, 0.0303, 0.0298], 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') 2023-03-26 07:04:55,056 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34801.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:05:05,996 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2808, 1.7826, 2.0470, 1.9971, 1.7919, 1.8839, 1.9531, 1.9049], device='cuda:3'), covar=tensor([0.5562, 0.6768, 0.5619, 0.6786, 0.7559, 0.5646, 0.8847, 0.5397], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0244, 0.0252, 0.0254, 0.0242, 0.0218, 0.0271, 0.0225], 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') 2023-03-26 07:05:14,340 INFO [finetune.py:976] (3/7) Epoch 7, batch 450, loss[loss=0.1989, simple_loss=0.265, pruned_loss=0.06639, over 4908.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.275, pruned_loss=0.07631, over 856738.00 frames. ], batch size: 36, lr: 3.88e-03, grad_scale: 32.0 2023-03-26 07:05:15,663 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7373, 1.1927, 0.8714, 1.6543, 2.1451, 1.3845, 1.5390, 1.6584], device='cuda:3'), covar=tensor([0.1590, 0.2223, 0.2294, 0.1311, 0.1948, 0.2019, 0.1541, 0.2028], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0098, 0.0115, 0.0093, 0.0124, 0.0096, 0.0100, 0.0093], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 07:05:25,982 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.133e+01 1.634e+02 1.827e+02 2.211e+02 3.915e+02, threshold=3.654e+02, percent-clipped=0.0 2023-03-26 07:05:40,040 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8134, 1.6700, 1.5148, 1.8167, 2.3077, 1.8625, 1.2739, 1.4773], device='cuda:3'), covar=tensor([0.2222, 0.2172, 0.2054, 0.1791, 0.1632, 0.1153, 0.2635, 0.1969], device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0207, 0.0202, 0.0185, 0.0236, 0.0175, 0.0210, 0.0189], 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') 2023-03-26 07:05:50,916 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=34849.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:06:11,591 INFO [finetune.py:976] (3/7) Epoch 7, batch 500, loss[loss=0.1919, simple_loss=0.2449, pruned_loss=0.06939, over 4800.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2717, pruned_loss=0.07541, over 880773.63 frames. ], batch size: 25, lr: 3.88e-03, grad_scale: 32.0 2023-03-26 07:07:15,207 INFO [finetune.py:976] (3/7) Epoch 7, batch 550, loss[loss=0.2016, simple_loss=0.2601, pruned_loss=0.07156, over 4925.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2687, pruned_loss=0.07416, over 896324.13 frames. ], batch size: 36, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:07:26,484 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.074e+02 1.634e+02 2.013e+02 2.383e+02 4.182e+02, threshold=4.026e+02, percent-clipped=3.0 2023-03-26 07:07:26,614 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5086, 1.3875, 1.3756, 1.4766, 1.0256, 2.8632, 1.1351, 1.6873], device='cuda:3'), covar=tensor([0.3210, 0.2245, 0.2000, 0.2235, 0.1913, 0.0259, 0.2963, 0.1259], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0114, 0.0117, 0.0122, 0.0117, 0.0098, 0.0101, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 07:07:53,608 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34949.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:08:14,352 INFO [finetune.py:976] (3/7) Epoch 7, batch 600, loss[loss=0.1785, simple_loss=0.2462, pruned_loss=0.05539, over 4758.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2714, pruned_loss=0.07556, over 909352.18 frames. ], batch size: 27, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:08:31,412 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34978.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 07:08:34,795 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34982.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:08:52,142 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-26 07:09:16,190 INFO [finetune.py:976] (3/7) Epoch 7, batch 650, loss[loss=0.1698, simple_loss=0.2261, pruned_loss=0.05674, over 4061.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2727, pruned_loss=0.0758, over 917502.16 frames. ], batch size: 17, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:09:27,411 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.303e+02 1.723e+02 2.031e+02 2.472e+02 3.902e+02, threshold=4.061e+02, percent-clipped=0.0 2023-03-26 07:10:17,351 INFO [finetune.py:976] (3/7) Epoch 7, batch 700, loss[loss=0.2241, simple_loss=0.2931, pruned_loss=0.07752, over 4841.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.274, pruned_loss=0.07538, over 925568.73 frames. ], batch size: 47, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:10:17,422 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35066.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 07:10:46,479 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35092.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:10:55,249 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.74 vs. limit=5.0 2023-03-26 07:11:11,734 INFO [finetune.py:976] (3/7) Epoch 7, batch 750, loss[loss=0.1603, simple_loss=0.2238, pruned_loss=0.04835, over 4753.00 frames. ], tot_loss[loss=0.212, simple_loss=0.2742, pruned_loss=0.07489, over 931842.27 frames. ], batch size: 27, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:11:23,577 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.067e+02 1.641e+02 2.027e+02 2.429e+02 3.682e+02, threshold=4.054e+02, percent-clipped=0.0 2023-03-26 07:11:55,024 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6631, 1.4450, 1.4879, 1.5933, 1.1309, 3.1873, 1.2250, 1.7770], device='cuda:3'), covar=tensor([0.3133, 0.2358, 0.2030, 0.2199, 0.1707, 0.0213, 0.2612, 0.1261], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0114, 0.0117, 0.0122, 0.0117, 0.0098, 0.0101, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 07:12:01,751 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35153.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:12:14,611 INFO [finetune.py:976] (3/7) Epoch 7, batch 800, loss[loss=0.1989, simple_loss=0.2743, pruned_loss=0.06177, over 4898.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2735, pruned_loss=0.07505, over 937640.23 frames. ], batch size: 35, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:12:16,452 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2374, 2.1375, 1.7477, 2.2234, 2.0693, 1.9202, 1.9662, 2.9992], device='cuda:3'), covar=tensor([0.6311, 0.7752, 0.5196, 0.7088, 0.6424, 0.3905, 0.7390, 0.2462], device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0257, 0.0220, 0.0281, 0.0239, 0.0204, 0.0244, 0.0203], 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') 2023-03-26 07:13:17,536 INFO [finetune.py:976] (3/7) Epoch 7, batch 850, loss[loss=0.2005, simple_loss=0.2608, pruned_loss=0.07013, over 4757.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2711, pruned_loss=0.07416, over 943792.61 frames. ], batch size: 26, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:13:17,708 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.87 vs. limit=5.0 2023-03-26 07:13:27,728 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.064e+02 1.645e+02 1.948e+02 2.227e+02 3.525e+02, threshold=3.897e+02, percent-clipped=0.0 2023-03-26 07:13:55,203 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35249.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:14:06,050 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-03-26 07:14:14,656 INFO [finetune.py:976] (3/7) Epoch 7, batch 900, loss[loss=0.1979, simple_loss=0.2574, pruned_loss=0.06915, over 4869.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2675, pruned_loss=0.07277, over 948023.84 frames. ], batch size: 31, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:14:23,621 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8302, 1.6013, 1.6637, 1.7385, 1.3029, 4.1620, 1.5902, 2.2448], device='cuda:3'), covar=tensor([0.3181, 0.2324, 0.1942, 0.2098, 0.1697, 0.0110, 0.2462, 0.1257], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0114, 0.0118, 0.0122, 0.0117, 0.0098, 0.0101, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 07:14:32,803 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35278.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 07:14:35,288 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35282.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:14:55,292 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=35297.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:15:16,598 INFO [finetune.py:976] (3/7) Epoch 7, batch 950, loss[loss=0.1904, simple_loss=0.2644, pruned_loss=0.0582, over 4758.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2654, pruned_loss=0.07209, over 947478.27 frames. ], batch size: 26, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:15:22,345 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.8647, 3.3640, 3.5267, 3.7029, 3.6179, 3.4102, 3.9192, 1.1802], device='cuda:3'), covar=tensor([0.0876, 0.0861, 0.0932, 0.1092, 0.1348, 0.1619, 0.0821, 0.5382], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0240, 0.0272, 0.0291, 0.0331, 0.0282, 0.0302, 0.0295], 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') 2023-03-26 07:15:31,367 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.193e+02 1.515e+02 1.811e+02 2.306e+02 3.628e+02, threshold=3.621e+02, percent-clipped=0.0 2023-03-26 07:15:31,440 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=35326.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:15:33,895 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=35330.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:15:38,913 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2424, 1.8136, 2.0532, 2.0098, 1.7818, 1.8115, 1.9098, 1.8667], device='cuda:3'), covar=tensor([0.5214, 0.7069, 0.5959, 0.6808, 0.7932, 0.5636, 0.8319, 0.5430], device='cuda:3'), in_proj_covar=tensor([0.0228, 0.0242, 0.0254, 0.0254, 0.0242, 0.0218, 0.0271, 0.0224], 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') 2023-03-26 07:15:50,926 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35345.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 07:16:18,537 INFO [finetune.py:976] (3/7) Epoch 7, batch 1000, loss[loss=0.2187, simple_loss=0.2691, pruned_loss=0.08412, over 4912.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2677, pruned_loss=0.07321, over 949690.64 frames. ], batch size: 32, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:16:18,647 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35366.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 07:16:21,142 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6244, 1.5879, 2.0608, 1.9443, 1.8204, 3.5074, 1.4850, 1.7269], device='cuda:3'), covar=tensor([0.0993, 0.1779, 0.1572, 0.0987, 0.1465, 0.0302, 0.1450, 0.1604], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0081, 0.0077, 0.0079, 0.0092, 0.0083, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 07:16:27,448 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35374.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:17:08,113 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35406.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 07:17:17,933 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=35414.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 07:17:19,085 INFO [finetune.py:976] (3/7) Epoch 7, batch 1050, loss[loss=0.2068, simple_loss=0.2807, pruned_loss=0.06644, over 4811.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2729, pruned_loss=0.07509, over 948556.59 frames. ], batch size: 39, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:17:30,142 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.274e+02 1.696e+02 1.924e+02 2.371e+02 5.787e+02, threshold=3.848e+02, percent-clipped=4.0 2023-03-26 07:17:41,402 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35435.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:17:43,769 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.63 vs. limit=5.0 2023-03-26 07:17:59,814 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35448.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:18:20,379 INFO [finetune.py:976] (3/7) Epoch 7, batch 1100, loss[loss=0.2008, simple_loss=0.2601, pruned_loss=0.07074, over 4756.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2759, pruned_loss=0.07659, over 949389.36 frames. ], batch size: 28, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:19:22,492 INFO [finetune.py:976] (3/7) Epoch 7, batch 1150, loss[loss=0.1698, simple_loss=0.233, pruned_loss=0.05333, over 4779.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2766, pruned_loss=0.07675, over 950794.60 frames. ], batch size: 26, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:19:33,762 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.207e+02 1.764e+02 2.068e+02 2.432e+02 4.937e+02, threshold=4.137e+02, percent-clipped=2.0 2023-03-26 07:19:41,264 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5324, 1.4135, 1.9112, 2.9617, 1.9809, 2.1960, 0.9693, 2.4384], device='cuda:3'), covar=tensor([0.1900, 0.1585, 0.1297, 0.0656, 0.0886, 0.1308, 0.1931, 0.0627], device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0118, 0.0136, 0.0165, 0.0102, 0.0141, 0.0129, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 07:19:41,299 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35530.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:19:42,914 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-26 07:20:24,974 INFO [finetune.py:976] (3/7) Epoch 7, batch 1200, loss[loss=0.2013, simple_loss=0.2716, pruned_loss=0.06553, over 4805.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2747, pruned_loss=0.07586, over 950593.20 frames. ], batch size: 41, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:20:38,795 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7828, 1.4572, 2.3428, 3.4549, 2.3521, 2.5235, 0.8877, 2.6762], device='cuda:3'), covar=tensor([0.1742, 0.1534, 0.1327, 0.0581, 0.0792, 0.1379, 0.2058, 0.0600], device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0117, 0.0135, 0.0165, 0.0101, 0.0140, 0.0128, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 07:20:43,340 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-26 07:20:51,619 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35591.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:20:55,483 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9306, 4.3478, 4.1974, 2.1897, 4.4585, 3.4072, 0.7563, 3.1860], device='cuda:3'), covar=tensor([0.2633, 0.1625, 0.1431, 0.3185, 0.0839, 0.0873, 0.4761, 0.1388], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0173, 0.0162, 0.0128, 0.0156, 0.0122, 0.0146, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 07:21:18,408 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2502, 2.2232, 1.8470, 1.6180, 2.4182, 2.6208, 2.2364, 1.9980], device='cuda:3'), covar=tensor([0.0298, 0.0360, 0.0477, 0.0376, 0.0260, 0.0444, 0.0313, 0.0402], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0111, 0.0138, 0.0116, 0.0104, 0.0100, 0.0091, 0.0109], device='cuda:3'), out_proj_covar=tensor([6.9121e-05, 8.7388e-05, 1.1045e-04, 9.1722e-05, 8.2185e-05, 7.3987e-05, 6.9139e-05, 8.5076e-05], device='cuda:3') 2023-03-26 07:21:20,131 INFO [finetune.py:976] (3/7) Epoch 7, batch 1250, loss[loss=0.2609, simple_loss=0.3022, pruned_loss=0.1098, over 4273.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2711, pruned_loss=0.07485, over 950418.84 frames. ], batch size: 65, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:21:31,425 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.265e+02 1.688e+02 2.018e+02 2.672e+02 1.298e+03, threshold=4.035e+02, percent-clipped=4.0 2023-03-26 07:22:18,864 INFO [finetune.py:976] (3/7) Epoch 7, batch 1300, loss[loss=0.2634, simple_loss=0.3109, pruned_loss=0.1079, over 4211.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2678, pruned_loss=0.07331, over 951897.99 frames. ], batch size: 65, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:23:06,117 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35701.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 07:23:25,711 INFO [finetune.py:976] (3/7) Epoch 7, batch 1350, loss[loss=0.1688, simple_loss=0.2341, pruned_loss=0.05182, over 4771.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2683, pruned_loss=0.07381, over 954212.99 frames. ], batch size: 26, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:23:29,239 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5156, 1.5380, 1.8213, 1.9981, 1.5811, 3.4292, 1.3587, 1.5513], device='cuda:3'), covar=tensor([0.1039, 0.1744, 0.1283, 0.0991, 0.1601, 0.0235, 0.1468, 0.1731], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0081, 0.0077, 0.0079, 0.0092, 0.0083, 0.0084, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 07:23:38,295 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.109e+01 1.659e+02 1.871e+02 2.249e+02 4.421e+02, threshold=3.743e+02, percent-clipped=1.0 2023-03-26 07:23:40,812 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35730.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:24:06,321 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35748.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:24:25,448 INFO [finetune.py:976] (3/7) Epoch 7, batch 1400, loss[loss=0.2194, simple_loss=0.2792, pruned_loss=0.07979, over 4746.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2739, pruned_loss=0.07593, over 955233.66 frames. ], batch size: 27, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:24:33,472 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35771.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:25:01,855 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=35796.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:25:04,165 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.0401, 4.3610, 4.5741, 4.8494, 4.7075, 4.4261, 5.0982, 1.6100], device='cuda:3'), covar=tensor([0.0605, 0.0779, 0.0719, 0.0798, 0.1087, 0.1459, 0.0588, 0.5440], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0241, 0.0272, 0.0292, 0.0331, 0.0283, 0.0303, 0.0295], 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') 2023-03-26 07:25:20,922 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35811.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:25:23,857 INFO [finetune.py:976] (3/7) Epoch 7, batch 1450, loss[loss=0.1999, simple_loss=0.2686, pruned_loss=0.06566, over 4744.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2754, pruned_loss=0.07555, over 956641.52 frames. ], batch size: 54, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:25:33,652 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.204e+02 1.737e+02 2.011e+02 2.560e+02 4.083e+02, threshold=4.021e+02, percent-clipped=3.0 2023-03-26 07:25:43,689 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35832.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:25:51,856 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5593, 1.6770, 1.2342, 1.4311, 1.7876, 1.7114, 1.4852, 1.3396], device='cuda:3'), covar=tensor([0.0291, 0.0264, 0.0584, 0.0332, 0.0208, 0.0400, 0.0307, 0.0386], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0111, 0.0138, 0.0116, 0.0104, 0.0100, 0.0091, 0.0109], device='cuda:3'), out_proj_covar=tensor([6.8873e-05, 8.7373e-05, 1.1047e-04, 9.1584e-05, 8.2238e-05, 7.4230e-05, 6.9186e-05, 8.5319e-05], device='cuda:3') 2023-03-26 07:26:24,785 INFO [finetune.py:976] (3/7) Epoch 7, batch 1500, loss[loss=0.2033, simple_loss=0.2735, pruned_loss=0.06658, over 4725.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2773, pruned_loss=0.07674, over 956436.48 frames. ], batch size: 54, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:26:32,648 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35872.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:26:47,876 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35886.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:27:11,196 INFO [finetune.py:976] (3/7) Epoch 7, batch 1550, loss[loss=0.2095, simple_loss=0.2695, pruned_loss=0.0748, over 4857.00 frames. ], tot_loss[loss=0.2145, simple_loss=0.2763, pruned_loss=0.07635, over 955848.29 frames. ], batch size: 31, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:27:17,687 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.940e+01 1.555e+02 1.902e+02 2.352e+02 4.828e+02, threshold=3.804e+02, percent-clipped=1.0 2023-03-26 07:27:45,065 INFO [finetune.py:976] (3/7) Epoch 7, batch 1600, loss[loss=0.1988, simple_loss=0.2498, pruned_loss=0.0739, over 4708.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.274, pruned_loss=0.0752, over 957812.53 frames. ], batch size: 23, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:28:25,879 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36001.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 07:28:35,457 INFO [finetune.py:976] (3/7) Epoch 7, batch 1650, loss[loss=0.2271, simple_loss=0.2884, pruned_loss=0.08288, over 4764.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2707, pruned_loss=0.07391, over 958059.01 frames. ], batch size: 26, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:28:41,918 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.242e+02 1.569e+02 1.867e+02 2.342e+02 3.778e+02, threshold=3.734e+02, percent-clipped=0.0 2023-03-26 07:28:50,686 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36030.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:29:09,237 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=36049.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 07:29:25,208 INFO [finetune.py:976] (3/7) Epoch 7, batch 1700, loss[loss=0.179, simple_loss=0.25, pruned_loss=0.05399, over 4902.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2692, pruned_loss=0.07417, over 956186.59 frames. ], batch size: 37, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:29:39,788 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=36078.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:29:58,262 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7972, 2.0222, 1.5679, 1.4409, 2.1652, 2.1113, 1.9447, 1.8293], device='cuda:3'), covar=tensor([0.0368, 0.0365, 0.0548, 0.0407, 0.0280, 0.0527, 0.0319, 0.0416], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0110, 0.0138, 0.0116, 0.0104, 0.0099, 0.0091, 0.0109], device='cuda:3'), out_proj_covar=tensor([6.8814e-05, 8.6870e-05, 1.1004e-04, 9.1143e-05, 8.1695e-05, 7.3955e-05, 6.8855e-05, 8.4868e-05], device='cuda:3') 2023-03-26 07:30:15,631 INFO [finetune.py:976] (3/7) Epoch 7, batch 1750, loss[loss=0.2358, simple_loss=0.3091, pruned_loss=0.08126, over 4815.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2709, pruned_loss=0.07484, over 955102.94 frames. ], batch size: 45, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:30:27,792 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.035e+02 1.647e+02 1.960e+02 2.452e+02 4.962e+02, threshold=3.920e+02, percent-clipped=3.0 2023-03-26 07:30:28,485 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36127.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:30:55,300 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7881, 4.0141, 3.8139, 1.7086, 4.0997, 3.0429, 0.8190, 2.7779], device='cuda:3'), covar=tensor([0.2418, 0.2078, 0.1479, 0.3561, 0.0951, 0.0986, 0.4785, 0.1452], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0172, 0.0162, 0.0129, 0.0155, 0.0122, 0.0146, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 07:31:18,545 INFO [finetune.py:976] (3/7) Epoch 7, batch 1800, loss[loss=0.2094, simple_loss=0.271, pruned_loss=0.07385, over 4871.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2743, pruned_loss=0.0758, over 956020.11 frames. ], batch size: 34, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:31:19,217 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36167.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:31:38,136 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36181.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:31:47,170 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36186.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:32:09,278 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6786, 2.3038, 2.0035, 0.9441, 2.1808, 1.9560, 1.7390, 2.1792], device='cuda:3'), covar=tensor([0.0803, 0.1033, 0.1557, 0.2405, 0.1589, 0.2598, 0.2431, 0.1015], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0200, 0.0197, 0.0187, 0.0214, 0.0204, 0.0218, 0.0196], 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') 2023-03-26 07:32:21,148 INFO [finetune.py:976] (3/7) Epoch 7, batch 1850, loss[loss=0.2216, simple_loss=0.2849, pruned_loss=0.07915, over 4909.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.276, pruned_loss=0.07632, over 957178.98 frames. ], batch size: 36, lr: 3.88e-03, grad_scale: 16.0 2023-03-26 07:32:21,293 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7302, 2.5539, 2.5638, 2.9608, 3.3532, 2.7417, 2.6069, 2.2850], device='cuda:3'), covar=tensor([0.1962, 0.1832, 0.1509, 0.1353, 0.1453, 0.0922, 0.1981, 0.1726], device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0209, 0.0204, 0.0186, 0.0239, 0.0176, 0.0213, 0.0192], 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') 2023-03-26 07:32:33,456 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.247e+02 1.736e+02 2.131e+02 2.651e+02 6.216e+02, threshold=4.263e+02, percent-clipped=3.0 2023-03-26 07:32:40,444 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=36234.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:32:50,327 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36242.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:33:20,732 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-26 07:33:21,420 INFO [finetune.py:976] (3/7) Epoch 7, batch 1900, loss[loss=0.2023, simple_loss=0.2569, pruned_loss=0.07389, over 4891.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2772, pruned_loss=0.07654, over 955508.92 frames. ], batch size: 32, lr: 3.87e-03, grad_scale: 16.0 2023-03-26 07:33:41,232 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-26 07:34:25,473 INFO [finetune.py:976] (3/7) Epoch 7, batch 1950, loss[loss=0.2712, simple_loss=0.3109, pruned_loss=0.1158, over 4753.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2755, pruned_loss=0.07589, over 954845.25 frames. ], batch size: 59, lr: 3.87e-03, grad_scale: 16.0 2023-03-26 07:34:36,926 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.969e+01 1.685e+02 2.051e+02 2.475e+02 4.640e+02, threshold=4.103e+02, percent-clipped=3.0 2023-03-26 07:35:26,475 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-03-26 07:35:28,761 INFO [finetune.py:976] (3/7) Epoch 7, batch 2000, loss[loss=0.1874, simple_loss=0.2548, pruned_loss=0.06, over 4784.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2713, pruned_loss=0.07411, over 954924.73 frames. ], batch size: 29, lr: 3.87e-03, grad_scale: 16.0 2023-03-26 07:36:26,428 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.08 vs. limit=5.0 2023-03-26 07:36:30,438 INFO [finetune.py:976] (3/7) Epoch 7, batch 2050, loss[loss=0.2359, simple_loss=0.2804, pruned_loss=0.09568, over 4850.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2678, pruned_loss=0.07277, over 954272.89 frames. ], batch size: 47, lr: 3.87e-03, grad_scale: 16.0 2023-03-26 07:36:43,670 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 7.910e+01 1.532e+02 1.893e+02 2.218e+02 7.941e+02, threshold=3.786e+02, percent-clipped=2.0 2023-03-26 07:36:43,819 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7556, 1.5474, 1.4518, 1.7818, 1.9427, 1.8469, 1.1232, 1.4223], device='cuda:3'), covar=tensor([0.2168, 0.2218, 0.2012, 0.1658, 0.1736, 0.1125, 0.2813, 0.2004], device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0209, 0.0204, 0.0186, 0.0238, 0.0176, 0.0212, 0.0191], 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') 2023-03-26 07:36:44,381 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36427.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:37:34,745 INFO [finetune.py:976] (3/7) Epoch 7, batch 2100, loss[loss=0.2224, simple_loss=0.2971, pruned_loss=0.07381, over 4818.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2683, pruned_loss=0.07317, over 954405.72 frames. ], batch size: 39, lr: 3.87e-03, grad_scale: 16.0 2023-03-26 07:37:35,450 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36467.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:37:45,794 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=36475.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:37:47,710 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.5403, 3.0201, 2.6931, 1.5122, 2.8818, 2.4673, 2.2539, 2.4504], device='cuda:3'), covar=tensor([0.0737, 0.0933, 0.1581, 0.2241, 0.1729, 0.1809, 0.1947, 0.1199], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0202, 0.0199, 0.0188, 0.0216, 0.0206, 0.0220, 0.0197], 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') 2023-03-26 07:38:15,570 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5880, 1.4702, 1.9456, 2.7876, 1.8888, 2.1805, 1.1584, 2.3305], device='cuda:3'), covar=tensor([0.1547, 0.1234, 0.1066, 0.0640, 0.0813, 0.1235, 0.1562, 0.0581], device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0118, 0.0135, 0.0166, 0.0102, 0.0142, 0.0129, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 07:38:36,450 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=36515.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:38:36,994 INFO [finetune.py:976] (3/7) Epoch 7, batch 2150, loss[loss=0.2737, simple_loss=0.3297, pruned_loss=0.1088, over 4912.00 frames. ], tot_loss[loss=0.2115, simple_loss=0.2727, pruned_loss=0.07514, over 956714.20 frames. ], batch size: 36, lr: 3.87e-03, grad_scale: 16.0 2023-03-26 07:38:48,005 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.129e+02 1.787e+02 2.211e+02 2.590e+02 5.595e+02, threshold=4.423e+02, percent-clipped=4.0 2023-03-26 07:39:00,124 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36537.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:39:15,521 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5456, 1.3536, 1.4328, 1.4787, 1.0451, 3.2576, 1.1738, 1.7615], device='cuda:3'), covar=tensor([0.3339, 0.2505, 0.2233, 0.2510, 0.1964, 0.0201, 0.2779, 0.1402], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0115, 0.0118, 0.0122, 0.0116, 0.0099, 0.0101, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 07:39:23,264 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7728, 1.5334, 2.0351, 1.4023, 1.7894, 2.0011, 1.4626, 2.0150], device='cuda:3'), covar=tensor([0.1513, 0.2256, 0.1436, 0.2122, 0.0980, 0.1570, 0.3026, 0.0976], device='cuda:3'), in_proj_covar=tensor([0.0200, 0.0204, 0.0196, 0.0195, 0.0180, 0.0219, 0.0216, 0.0200], 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') 2023-03-26 07:39:35,927 INFO [finetune.py:976] (3/7) Epoch 7, batch 2200, loss[loss=0.2047, simple_loss=0.2656, pruned_loss=0.07192, over 4893.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2751, pruned_loss=0.07583, over 955933.87 frames. ], batch size: 35, lr: 3.87e-03, grad_scale: 16.0 2023-03-26 07:40:25,402 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4614, 1.5734, 1.8254, 1.7950, 1.6105, 3.4844, 1.3993, 1.6149], device='cuda:3'), covar=tensor([0.0975, 0.1697, 0.1082, 0.1080, 0.1579, 0.0265, 0.1446, 0.1708], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0081, 0.0076, 0.0079, 0.0092, 0.0083, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 07:40:25,420 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36605.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:40:27,769 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36608.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:40:35,342 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36612.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:40:38,177 INFO [finetune.py:976] (3/7) Epoch 7, batch 2250, loss[loss=0.2341, simple_loss=0.2956, pruned_loss=0.08633, over 4911.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2745, pruned_loss=0.0754, over 954968.17 frames. ], batch size: 36, lr: 3.87e-03, grad_scale: 16.0 2023-03-26 07:40:49,911 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.198e+02 1.735e+02 1.950e+02 2.446e+02 5.153e+02, threshold=3.899e+02, percent-clipped=1.0 2023-03-26 07:41:32,990 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.69 vs. limit=5.0 2023-03-26 07:41:41,215 INFO [finetune.py:976] (3/7) Epoch 7, batch 2300, loss[loss=0.2058, simple_loss=0.2764, pruned_loss=0.06761, over 4810.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2747, pruned_loss=0.07539, over 954143.23 frames. ], batch size: 39, lr: 3.87e-03, grad_scale: 16.0 2023-03-26 07:41:41,323 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36666.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:41:43,624 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36669.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:41:51,491 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36673.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:42:03,075 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4582, 2.2151, 1.7190, 0.7800, 1.9030, 1.9387, 1.6926, 2.0448], device='cuda:3'), covar=tensor([0.0715, 0.0752, 0.1426, 0.2226, 0.1528, 0.2423, 0.2172, 0.0868], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0202, 0.0200, 0.0188, 0.0217, 0.0207, 0.0221, 0.0198], 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') 2023-03-26 07:42:18,124 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36700.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 07:42:18,703 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.9100, 3.3396, 3.5872, 3.7371, 3.6765, 3.4238, 3.9661, 1.6748], device='cuda:3'), covar=tensor([0.0732, 0.0799, 0.0800, 0.0889, 0.1088, 0.1277, 0.0685, 0.4360], device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0241, 0.0271, 0.0290, 0.0328, 0.0281, 0.0301, 0.0293], 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') 2023-03-26 07:42:36,646 INFO [finetune.py:976] (3/7) Epoch 7, batch 2350, loss[loss=0.1955, simple_loss=0.2564, pruned_loss=0.06728, over 4808.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2718, pruned_loss=0.07435, over 954697.38 frames. ], batch size: 25, lr: 3.87e-03, grad_scale: 16.0 2023-03-26 07:42:47,524 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7152, 1.5034, 1.3805, 1.1244, 1.5065, 1.5129, 1.4802, 2.0844], device='cuda:3'), covar=tensor([0.5540, 0.5669, 0.4475, 0.5653, 0.5058, 0.3036, 0.5132, 0.2428], device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0259, 0.0220, 0.0282, 0.0240, 0.0205, 0.0244, 0.0204], 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') 2023-03-26 07:42:49,194 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.103e+02 1.509e+02 1.866e+02 2.321e+02 4.735e+02, threshold=3.732e+02, percent-clipped=2.0 2023-03-26 07:43:25,429 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-03-26 07:43:29,351 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36761.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 07:43:37,712 INFO [finetune.py:976] (3/7) Epoch 7, batch 2400, loss[loss=0.1899, simple_loss=0.257, pruned_loss=0.06144, over 4766.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2692, pruned_loss=0.07359, over 955496.63 frames. ], batch size: 28, lr: 3.87e-03, grad_scale: 16.0 2023-03-26 07:44:20,075 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6391, 1.4796, 1.4734, 1.5956, 1.1326, 3.3319, 1.2622, 1.8255], device='cuda:3'), covar=tensor([0.3311, 0.2415, 0.2142, 0.2354, 0.1913, 0.0198, 0.2756, 0.1359], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0115, 0.0118, 0.0122, 0.0116, 0.0098, 0.0101, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 07:44:28,538 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36805.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 07:44:40,195 INFO [finetune.py:976] (3/7) Epoch 7, batch 2450, loss[loss=0.1565, simple_loss=0.22, pruned_loss=0.04649, over 4781.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2665, pruned_loss=0.07273, over 952346.77 frames. ], batch size: 27, lr: 3.87e-03, grad_scale: 16.0 2023-03-26 07:44:51,698 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 1.798e+02 2.140e+02 2.594e+02 4.660e+02, threshold=4.281e+02, percent-clipped=3.0 2023-03-26 07:45:10,238 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36837.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:45:49,129 INFO [finetune.py:976] (3/7) Epoch 7, batch 2500, loss[loss=0.2125, simple_loss=0.2752, pruned_loss=0.07485, over 4813.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2677, pruned_loss=0.07311, over 954750.00 frames. ], batch size: 38, lr: 3.87e-03, grad_scale: 16.0 2023-03-26 07:45:49,258 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36866.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 07:46:12,541 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=36885.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:46:44,080 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36911.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:46:52,309 INFO [finetune.py:976] (3/7) Epoch 7, batch 2550, loss[loss=0.2071, simple_loss=0.271, pruned_loss=0.07162, over 4893.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2715, pruned_loss=0.07459, over 951685.44 frames. ], batch size: 32, lr: 3.87e-03, grad_scale: 32.0 2023-03-26 07:47:03,265 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.261e+02 1.620e+02 1.912e+02 2.307e+02 6.491e+02, threshold=3.825e+02, percent-clipped=1.0 2023-03-26 07:47:45,946 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36958.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:47:47,752 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36961.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:47:55,199 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36964.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:47:56,365 INFO [finetune.py:976] (3/7) Epoch 7, batch 2600, loss[loss=0.1628, simple_loss=0.2243, pruned_loss=0.05065, over 3946.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2738, pruned_loss=0.07524, over 949896.72 frames. ], batch size: 17, lr: 3.87e-03, grad_scale: 32.0 2023-03-26 07:47:57,653 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36968.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:47:58,952 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8958, 1.9002, 1.3450, 1.9912, 1.9201, 1.6328, 2.6841, 1.9430], device='cuda:3'), covar=tensor([0.1687, 0.2612, 0.3841, 0.3232, 0.3091, 0.1874, 0.2331, 0.2208], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0191, 0.0236, 0.0255, 0.0235, 0.0193, 0.0212, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 07:48:05,153 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36972.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:48:41,490 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37004.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:48:53,702 INFO [finetune.py:976] (3/7) Epoch 7, batch 2650, loss[loss=0.2684, simple_loss=0.3191, pruned_loss=0.1088, over 4808.00 frames. ], tot_loss[loss=0.2125, simple_loss=0.2747, pruned_loss=0.07515, over 949292.32 frames. ], batch size: 40, lr: 3.87e-03, grad_scale: 32.0 2023-03-26 07:49:01,197 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37019.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:49:05,453 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.083e+02 1.627e+02 1.954e+02 2.393e+02 3.704e+02, threshold=3.907e+02, percent-clipped=0.0 2023-03-26 07:49:12,077 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9164, 1.3961, 1.8804, 1.7746, 1.6006, 1.6106, 1.7001, 1.6586], device='cuda:3'), covar=tensor([0.4864, 0.5986, 0.4641, 0.5707, 0.6651, 0.5228, 0.6737, 0.4594], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0242, 0.0255, 0.0254, 0.0243, 0.0220, 0.0272, 0.0225], 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') 2023-03-26 07:49:34,589 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.0529, 0.9311, 0.9412, 1.1366, 1.1871, 1.1383, 1.0066, 1.0026], device='cuda:3'), covar=tensor([0.0316, 0.0313, 0.0582, 0.0278, 0.0271, 0.0463, 0.0324, 0.0449], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0111, 0.0140, 0.0117, 0.0105, 0.0100, 0.0091, 0.0110], device='cuda:3'), out_proj_covar=tensor([6.9515e-05, 8.7504e-05, 1.1161e-04, 9.1844e-05, 8.2551e-05, 7.4687e-05, 6.9132e-05, 8.5521e-05], device='cuda:3') 2023-03-26 07:49:41,702 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37056.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 07:49:47,115 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2051, 1.9102, 2.7385, 1.7379, 2.4908, 2.4446, 1.9331, 2.6792], device='cuda:3'), covar=tensor([0.1446, 0.2161, 0.1631, 0.2203, 0.0825, 0.1821, 0.2635, 0.0854], device='cuda:3'), in_proj_covar=tensor([0.0202, 0.0203, 0.0197, 0.0195, 0.0181, 0.0220, 0.0216, 0.0200], 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') 2023-03-26 07:49:47,736 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37065.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:49:48,223 INFO [finetune.py:976] (3/7) Epoch 7, batch 2700, loss[loss=0.2249, simple_loss=0.2753, pruned_loss=0.08723, over 4785.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2741, pruned_loss=0.07519, over 949981.05 frames. ], batch size: 29, lr: 3.87e-03, grad_scale: 32.0 2023-03-26 07:50:22,104 INFO [finetune.py:976] (3/7) Epoch 7, batch 2750, loss[loss=0.2133, simple_loss=0.2646, pruned_loss=0.08098, over 4868.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2719, pruned_loss=0.07439, over 952574.39 frames. ], batch size: 34, lr: 3.87e-03, grad_scale: 32.0 2023-03-26 07:50:28,698 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.127e+02 1.629e+02 1.991e+02 2.307e+02 4.303e+02, threshold=3.983e+02, percent-clipped=1.0 2023-03-26 07:50:41,154 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.92 vs. limit=5.0 2023-03-26 07:50:58,190 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37161.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 07:51:03,501 INFO [finetune.py:976] (3/7) Epoch 7, batch 2800, loss[loss=0.1798, simple_loss=0.2449, pruned_loss=0.05733, over 4910.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2679, pruned_loss=0.07301, over 954250.27 frames. ], batch size: 37, lr: 3.87e-03, grad_scale: 32.0 2023-03-26 07:52:00,007 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6264, 1.4582, 1.3684, 1.6463, 1.7137, 1.6378, 0.8860, 1.3843], device='cuda:3'), covar=tensor([0.2154, 0.2153, 0.1886, 0.1655, 0.1621, 0.1193, 0.2839, 0.1837], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0209, 0.0204, 0.0187, 0.0240, 0.0178, 0.0214, 0.0192], 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') 2023-03-26 07:52:09,302 INFO [finetune.py:976] (3/7) Epoch 7, batch 2850, loss[loss=0.1964, simple_loss=0.2523, pruned_loss=0.07026, over 4806.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2652, pruned_loss=0.07202, over 952433.00 frames. ], batch size: 25, lr: 3.87e-03, grad_scale: 32.0 2023-03-26 07:52:09,886 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7184, 3.6856, 3.4590, 1.8054, 3.7873, 2.9035, 0.8661, 2.5836], device='cuda:3'), covar=tensor([0.2454, 0.2575, 0.1959, 0.3669, 0.1246, 0.1067, 0.4868, 0.1902], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0173, 0.0162, 0.0129, 0.0153, 0.0123, 0.0146, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 07:52:20,862 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.100e+02 1.582e+02 1.929e+02 2.327e+02 4.539e+02, threshold=3.857e+02, percent-clipped=3.0 2023-03-26 07:53:00,976 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37261.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:53:02,784 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37264.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:53:03,916 INFO [finetune.py:976] (3/7) Epoch 7, batch 2900, loss[loss=0.3041, simple_loss=0.3483, pruned_loss=0.13, over 4804.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2706, pruned_loss=0.07501, over 953392.16 frames. ], batch size: 51, lr: 3.87e-03, grad_scale: 32.0 2023-03-26 07:53:09,378 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37267.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:53:09,982 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37268.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:53:09,998 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37268.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:54:04,930 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=37309.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:54:06,759 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=37312.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:54:07,994 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37314.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:54:13,844 INFO [finetune.py:976] (3/7) Epoch 7, batch 2950, loss[loss=0.2081, simple_loss=0.2846, pruned_loss=0.06575, over 4863.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2735, pruned_loss=0.07595, over 953702.79 frames. ], batch size: 44, lr: 3.87e-03, grad_scale: 32.0 2023-03-26 07:54:13,910 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=37316.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:54:17,027 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8131, 1.6918, 1.5752, 1.9291, 2.4254, 1.8851, 1.6162, 1.4909], device='cuda:3'), covar=tensor([0.2487, 0.2392, 0.2257, 0.1887, 0.1801, 0.1352, 0.2692, 0.2244], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0210, 0.0205, 0.0187, 0.0240, 0.0178, 0.0214, 0.0193], 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') 2023-03-26 07:54:25,443 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.193e+02 1.702e+02 2.045e+02 2.514e+02 5.908e+02, threshold=4.090e+02, percent-clipped=3.0 2023-03-26 07:54:27,402 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37329.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:54:58,643 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1941, 2.2551, 2.2842, 1.5277, 2.4271, 2.3625, 2.3025, 1.9113], device='cuda:3'), covar=tensor([0.0707, 0.0632, 0.0815, 0.0956, 0.0470, 0.0751, 0.0656, 0.1059], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0135, 0.0146, 0.0128, 0.0114, 0.0146, 0.0148, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 07:55:00,453 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37356.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 07:55:05,687 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37360.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:55:08,128 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37364.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:55:09,239 INFO [finetune.py:976] (3/7) Epoch 7, batch 3000, loss[loss=0.2142, simple_loss=0.2953, pruned_loss=0.06658, over 4816.00 frames. ], tot_loss[loss=0.213, simple_loss=0.2742, pruned_loss=0.07588, over 952683.86 frames. ], batch size: 39, lr: 3.87e-03, grad_scale: 32.0 2023-03-26 07:55:09,239 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 07:55:17,551 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8722, 1.7748, 1.9653, 1.1880, 1.9744, 2.0166, 1.9350, 1.6164], device='cuda:3'), covar=tensor([0.0616, 0.0664, 0.0694, 0.0999, 0.0658, 0.0710, 0.0673, 0.1066], device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0135, 0.0146, 0.0128, 0.0114, 0.0146, 0.0148, 0.0164], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 07:55:18,345 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.1565, 1.4382, 1.4999, 0.6153, 1.3386, 1.6130, 1.6851, 1.3527], device='cuda:3'), covar=tensor([0.1127, 0.0668, 0.0542, 0.0739, 0.0533, 0.0795, 0.0401, 0.0924], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0156, 0.0120, 0.0137, 0.0131, 0.0124, 0.0145, 0.0145], device='cuda:3'), out_proj_covar=tensor([9.6111e-05, 1.1481e-04, 8.6984e-05, 9.9453e-05, 9.3983e-05, 9.1066e-05, 1.0714e-04, 1.0661e-04], device='cuda:3') 2023-03-26 07:55:25,742 INFO [finetune.py:1010] (3/7) Epoch 7, validation: loss=0.161, simple_loss=0.2327, pruned_loss=0.04464, over 2265189.00 frames. 2023-03-26 07:55:25,743 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 07:55:38,025 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0250, 1.8365, 1.6477, 1.7694, 2.1312, 1.7767, 2.2457, 2.0266], device='cuda:3'), covar=tensor([0.1707, 0.2807, 0.3905, 0.3199, 0.2921, 0.1960, 0.3467, 0.2242], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0191, 0.0235, 0.0256, 0.0234, 0.0192, 0.0212, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 07:55:54,503 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=37404.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 07:56:02,292 INFO [finetune.py:976] (3/7) Epoch 7, batch 3050, loss[loss=0.2328, simple_loss=0.2897, pruned_loss=0.088, over 4885.00 frames. ], tot_loss[loss=0.2126, simple_loss=0.2749, pruned_loss=0.07519, over 953569.80 frames. ], batch size: 35, lr: 3.87e-03, grad_scale: 32.0 2023-03-26 07:56:04,300 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 2023-03-26 07:56:11,536 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37425.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:56:12,025 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.173e+02 1.581e+02 1.871e+02 2.387e+02 4.591e+02, threshold=3.742e+02, percent-clipped=1.0 2023-03-26 07:56:48,981 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37461.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 07:56:57,054 INFO [finetune.py:976] (3/7) Epoch 7, batch 3100, loss[loss=0.1752, simple_loss=0.2381, pruned_loss=0.05616, over 4767.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2713, pruned_loss=0.07345, over 953002.08 frames. ], batch size: 26, lr: 3.87e-03, grad_scale: 32.0 2023-03-26 07:57:42,546 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1224, 2.0495, 1.5992, 2.1386, 2.1232, 1.7846, 2.5459, 2.0926], device='cuda:3'), covar=tensor([0.1645, 0.3096, 0.3884, 0.3302, 0.2966, 0.1986, 0.3638, 0.2270], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0191, 0.0235, 0.0256, 0.0235, 0.0192, 0.0212, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 07:57:52,635 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=37509.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 07:58:01,946 INFO [finetune.py:976] (3/7) Epoch 7, batch 3150, loss[loss=0.2119, simple_loss=0.2728, pruned_loss=0.07552, over 4874.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2693, pruned_loss=0.07346, over 950935.03 frames. ], batch size: 31, lr: 3.87e-03, grad_scale: 32.0 2023-03-26 07:58:13,088 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.225e+02 1.704e+02 2.041e+02 2.515e+02 5.799e+02, threshold=4.081e+02, percent-clipped=3.0 2023-03-26 07:58:34,810 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-26 07:59:05,775 INFO [finetune.py:976] (3/7) Epoch 7, batch 3200, loss[loss=0.2051, simple_loss=0.2671, pruned_loss=0.07152, over 4716.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.266, pruned_loss=0.07215, over 951808.36 frames. ], batch size: 59, lr: 3.87e-03, grad_scale: 32.0 2023-03-26 07:59:06,471 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37567.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 07:59:38,043 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8024, 3.6679, 3.5519, 1.9165, 3.8049, 2.9183, 0.7768, 2.6094], device='cuda:3'), covar=tensor([0.2546, 0.2053, 0.1500, 0.3109, 0.0935, 0.0952, 0.4566, 0.1531], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0171, 0.0160, 0.0127, 0.0152, 0.0121, 0.0145, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 08:00:08,911 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37614.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:00:14,411 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=37615.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:00:14,954 INFO [finetune.py:976] (3/7) Epoch 7, batch 3250, loss[loss=0.2309, simple_loss=0.2903, pruned_loss=0.08579, over 4905.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2669, pruned_loss=0.07275, over 950972.33 frames. ], batch size: 36, lr: 3.87e-03, grad_scale: 32.0 2023-03-26 08:00:19,923 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37624.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:00:26,119 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.216e+02 1.664e+02 1.918e+02 2.274e+02 4.430e+02, threshold=3.836e+02, percent-clipped=1.0 2023-03-26 08:00:26,875 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4626, 2.7848, 2.4994, 1.7949, 2.9238, 2.8057, 2.6867, 2.4007], device='cuda:3'), covar=tensor([0.0682, 0.0567, 0.0789, 0.0989, 0.0458, 0.0724, 0.0681, 0.0927], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0132, 0.0142, 0.0126, 0.0112, 0.0143, 0.0145, 0.0160], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 08:00:36,204 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4543, 2.1827, 1.8163, 0.8636, 2.0962, 1.8849, 1.6519, 1.9638], device='cuda:3'), covar=tensor([0.0983, 0.0871, 0.1573, 0.2306, 0.1453, 0.2361, 0.2373, 0.1117], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0200, 0.0198, 0.0186, 0.0216, 0.0204, 0.0219, 0.0197], 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') 2023-03-26 08:01:09,578 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37660.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:01:10,704 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=37662.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:01:18,268 INFO [finetune.py:976] (3/7) Epoch 7, batch 3300, loss[loss=0.2258, simple_loss=0.2995, pruned_loss=0.07604, over 4804.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2706, pruned_loss=0.07411, over 953016.79 frames. ], batch size: 41, lr: 3.87e-03, grad_scale: 32.0 2023-03-26 08:01:20,322 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.31 vs. limit=5.0 2023-03-26 08:01:39,547 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7542, 1.6161, 1.6307, 1.6855, 1.2273, 4.1025, 1.5685, 2.1993], device='cuda:3'), covar=tensor([0.3156, 0.2295, 0.2030, 0.2174, 0.1692, 0.0127, 0.2493, 0.1272], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0114, 0.0118, 0.0122, 0.0116, 0.0098, 0.0101, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0003, 0.0005, 0.0004], device='cuda:3') 2023-03-26 08:02:12,454 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=37708.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:02:22,477 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2714, 1.2710, 1.5914, 1.1715, 1.3994, 1.4164, 1.2456, 1.5361], device='cuda:3'), covar=tensor([0.0899, 0.1850, 0.1050, 0.1244, 0.0742, 0.1020, 0.2665, 0.0701], device='cuda:3'), in_proj_covar=tensor([0.0201, 0.0203, 0.0197, 0.0195, 0.0180, 0.0219, 0.0217, 0.0201], 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') 2023-03-26 08:02:22,974 INFO [finetune.py:976] (3/7) Epoch 7, batch 3350, loss[loss=0.2808, simple_loss=0.3342, pruned_loss=0.1137, over 4753.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2719, pruned_loss=0.07404, over 954908.70 frames. ], batch size: 54, lr: 3.87e-03, grad_scale: 32.0 2023-03-26 08:02:25,470 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37720.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:02:34,111 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.166e+02 1.767e+02 2.019e+02 2.457e+02 5.992e+02, threshold=4.038e+02, percent-clipped=4.0 2023-03-26 08:02:45,330 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3477, 1.4434, 1.4392, 1.5690, 1.5123, 2.8649, 1.2920, 1.5112], device='cuda:3'), covar=tensor([0.0983, 0.1636, 0.1113, 0.0998, 0.1495, 0.0341, 0.1387, 0.1630], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0082, 0.0076, 0.0079, 0.0092, 0.0083, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 08:03:16,436 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9842, 2.1424, 2.1132, 1.4453, 2.2896, 2.1826, 2.0574, 1.7780], device='cuda:3'), covar=tensor([0.0787, 0.0640, 0.0857, 0.1037, 0.0512, 0.0861, 0.0806, 0.1099], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0133, 0.0143, 0.0126, 0.0112, 0.0144, 0.0146, 0.0160], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 08:03:28,096 INFO [finetune.py:976] (3/7) Epoch 7, batch 3400, loss[loss=0.2046, simple_loss=0.2733, pruned_loss=0.06794, over 4865.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.273, pruned_loss=0.07441, over 956056.73 frames. ], batch size: 31, lr: 3.87e-03, grad_scale: 32.0 2023-03-26 08:04:32,120 INFO [finetune.py:976] (3/7) Epoch 7, batch 3450, loss[loss=0.1831, simple_loss=0.253, pruned_loss=0.05656, over 4813.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2719, pruned_loss=0.07326, over 956279.82 frames. ], batch size: 38, lr: 3.87e-03, grad_scale: 32.0 2023-03-26 08:04:43,341 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.130e+02 1.715e+02 1.991e+02 2.496e+02 6.747e+02, threshold=3.982e+02, percent-clipped=3.0 2023-03-26 08:05:00,479 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8010, 3.6891, 3.4438, 1.6167, 3.6777, 2.7859, 0.6594, 2.3998], device='cuda:3'), covar=tensor([0.2392, 0.1943, 0.1663, 0.3681, 0.1160, 0.1041, 0.4802, 0.1673], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0172, 0.0162, 0.0128, 0.0153, 0.0122, 0.0145, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 08:05:36,356 INFO [finetune.py:976] (3/7) Epoch 7, batch 3500, loss[loss=0.1675, simple_loss=0.2258, pruned_loss=0.0546, over 4765.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2694, pruned_loss=0.07298, over 954435.39 frames. ], batch size: 27, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:05:45,673 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-03-26 08:05:54,129 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4654, 2.3086, 2.8316, 1.8057, 2.7221, 2.8081, 2.1922, 2.8501], device='cuda:3'), covar=tensor([0.1593, 0.1936, 0.1625, 0.2299, 0.1014, 0.1731, 0.2494, 0.0934], device='cuda:3'), in_proj_covar=tensor([0.0203, 0.0204, 0.0198, 0.0195, 0.0181, 0.0221, 0.0218, 0.0202], 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') 2023-03-26 08:06:28,676 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9408, 1.8184, 1.4994, 1.7272, 1.9053, 1.5848, 2.1377, 1.8747], device='cuda:3'), covar=tensor([0.1566, 0.2404, 0.3706, 0.2971, 0.3071, 0.1900, 0.3829, 0.2129], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0190, 0.0235, 0.0255, 0.0234, 0.0193, 0.0212, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 08:06:41,048 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-26 08:06:41,166 INFO [finetune.py:976] (3/7) Epoch 7, batch 3550, loss[loss=0.1886, simple_loss=0.2463, pruned_loss=0.06548, over 4812.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2665, pruned_loss=0.07203, over 954843.45 frames. ], batch size: 41, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:06:51,853 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37924.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:06:58,534 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.177e+02 1.559e+02 1.846e+02 2.185e+02 4.242e+02, threshold=3.693e+02, percent-clipped=1.0 2023-03-26 08:07:01,531 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-26 08:07:52,118 INFO [finetune.py:976] (3/7) Epoch 7, batch 3600, loss[loss=0.1814, simple_loss=0.2448, pruned_loss=0.059, over 4828.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2646, pruned_loss=0.07122, over 954156.33 frames. ], batch size: 30, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:07:56,329 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=37972.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:08:06,515 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37979.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:08:59,065 INFO [finetune.py:976] (3/7) Epoch 7, batch 3650, loss[loss=0.2897, simple_loss=0.3377, pruned_loss=0.1208, over 4824.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2663, pruned_loss=0.07221, over 951936.99 frames. ], batch size: 51, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:09:07,230 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38020.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:09:10,799 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.158e+02 1.699e+02 2.068e+02 2.418e+02 4.148e+02, threshold=4.136e+02, percent-clipped=4.0 2023-03-26 08:09:20,546 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2274, 1.8448, 2.6067, 1.6175, 2.4710, 2.4952, 1.8708, 2.5777], device='cuda:3'), covar=tensor([0.1417, 0.2256, 0.1948, 0.2578, 0.0875, 0.1697, 0.2701, 0.0911], device='cuda:3'), in_proj_covar=tensor([0.0202, 0.0203, 0.0197, 0.0195, 0.0181, 0.0220, 0.0216, 0.0201], 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') 2023-03-26 08:09:27,925 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38040.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:09:46,796 INFO [finetune.py:976] (3/7) Epoch 7, batch 3700, loss[loss=0.2033, simple_loss=0.267, pruned_loss=0.06976, over 4928.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2696, pruned_loss=0.07324, over 952047.82 frames. ], batch size: 38, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:09:48,561 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=38068.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:10:19,916 INFO [finetune.py:976] (3/7) Epoch 7, batch 3750, loss[loss=0.17, simple_loss=0.2443, pruned_loss=0.04787, over 4809.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2718, pruned_loss=0.07411, over 952245.79 frames. ], batch size: 45, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:10:24,834 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.21 vs. limit=5.0 2023-03-26 08:10:26,924 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.737e+01 1.627e+02 1.982e+02 2.503e+02 4.763e+02, threshold=3.965e+02, percent-clipped=1.0 2023-03-26 08:10:57,174 INFO [finetune.py:976] (3/7) Epoch 7, batch 3800, loss[loss=0.1208, simple_loss=0.1921, pruned_loss=0.02471, over 4764.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2722, pruned_loss=0.07446, over 950535.97 frames. ], batch size: 23, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:11:16,640 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5533, 1.3906, 1.8777, 3.2932, 2.2069, 2.2298, 0.9649, 2.5519], device='cuda:3'), covar=tensor([0.1910, 0.1566, 0.1441, 0.0558, 0.0844, 0.1531, 0.1853, 0.0624], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0118, 0.0135, 0.0166, 0.0102, 0.0140, 0.0128, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 08:11:30,369 INFO [finetune.py:976] (3/7) Epoch 7, batch 3850, loss[loss=0.2358, simple_loss=0.2888, pruned_loss=0.09146, over 4870.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2703, pruned_loss=0.07298, over 951751.93 frames. ], batch size: 34, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:11:43,043 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.105e+02 1.610e+02 2.090e+02 2.406e+02 4.877e+02, threshold=4.181e+02, percent-clipped=2.0 2023-03-26 08:11:45,786 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-26 08:12:25,294 INFO [finetune.py:976] (3/7) Epoch 7, batch 3900, loss[loss=0.2074, simple_loss=0.2759, pruned_loss=0.06944, over 4910.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2673, pruned_loss=0.07184, over 951526.87 frames. ], batch size: 35, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:13:16,529 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.87 vs. limit=5.0 2023-03-26 08:13:28,056 INFO [finetune.py:976] (3/7) Epoch 7, batch 3950, loss[loss=0.1751, simple_loss=0.2486, pruned_loss=0.05077, over 4766.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2637, pruned_loss=0.07012, over 951654.91 frames. ], batch size: 26, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:13:45,513 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.011e+02 1.693e+02 1.988e+02 2.374e+02 4.679e+02, threshold=3.976e+02, percent-clipped=1.0 2023-03-26 08:13:56,366 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38335.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:14:21,707 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38360.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:14:25,157 INFO [finetune.py:976] (3/7) Epoch 7, batch 4000, loss[loss=0.2011, simple_loss=0.2682, pruned_loss=0.06706, over 4809.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2641, pruned_loss=0.07086, over 950555.12 frames. ], batch size: 40, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:15:29,791 INFO [finetune.py:976] (3/7) Epoch 7, batch 4050, loss[loss=0.208, simple_loss=0.2807, pruned_loss=0.06769, over 4763.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2687, pruned_loss=0.07343, over 948827.71 frames. ], batch size: 28, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:15:33,993 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38421.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:15:42,069 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.248e+02 1.778e+02 2.129e+02 2.625e+02 5.238e+02, threshold=4.258e+02, percent-clipped=5.0 2023-03-26 08:16:24,540 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-03-26 08:16:32,436 INFO [finetune.py:976] (3/7) Epoch 7, batch 4100, loss[loss=0.2159, simple_loss=0.2787, pruned_loss=0.07655, over 4860.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.272, pruned_loss=0.07429, over 950191.67 frames. ], batch size: 31, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:17:02,888 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-26 08:17:31,591 INFO [finetune.py:976] (3/7) Epoch 7, batch 4150, loss[loss=0.268, simple_loss=0.3259, pruned_loss=0.1051, over 4126.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2739, pruned_loss=0.07495, over 953255.03 frames. ], batch size: 66, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:17:43,680 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.188e+02 1.723e+02 2.145e+02 2.598e+02 6.605e+02, threshold=4.291e+02, percent-clipped=2.0 2023-03-26 08:18:32,089 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-03-26 08:18:34,212 INFO [finetune.py:976] (3/7) Epoch 7, batch 4200, loss[loss=0.2702, simple_loss=0.313, pruned_loss=0.1137, over 4149.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2747, pruned_loss=0.07543, over 951651.38 frames. ], batch size: 65, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:19:34,111 INFO [finetune.py:976] (3/7) Epoch 7, batch 4250, loss[loss=0.2129, simple_loss=0.2782, pruned_loss=0.07386, over 4830.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2716, pruned_loss=0.07385, over 952720.83 frames. ], batch size: 39, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:19:44,841 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.083e+02 1.606e+02 1.980e+02 2.259e+02 5.740e+02, threshold=3.960e+02, percent-clipped=2.0 2023-03-26 08:19:55,315 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4470, 1.5047, 1.5581, 1.7275, 1.6724, 3.2083, 1.4441, 1.6074], device='cuda:3'), covar=tensor([0.1054, 0.1735, 0.1102, 0.0985, 0.1567, 0.0280, 0.1393, 0.1672], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0081, 0.0076, 0.0078, 0.0092, 0.0083, 0.0084, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 08:20:02,301 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38635.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:20:38,696 INFO [finetune.py:976] (3/7) Epoch 7, batch 4300, loss[loss=0.1981, simple_loss=0.2608, pruned_loss=0.06768, over 4791.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2691, pruned_loss=0.07369, over 954291.89 frames. ], batch size: 26, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:20:59,384 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=38683.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:21:11,482 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-03-26 08:21:41,238 INFO [finetune.py:976] (3/7) Epoch 7, batch 4350, loss[loss=0.1887, simple_loss=0.2473, pruned_loss=0.06506, over 4829.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2654, pruned_loss=0.07225, over 954479.12 frames. ], batch size: 25, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:21:41,327 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38716.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:21:52,343 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.223e+02 1.679e+02 1.871e+02 2.197e+02 5.866e+02, threshold=3.741e+02, percent-clipped=4.0 2023-03-26 08:21:53,096 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2232, 2.0271, 1.6215, 2.1480, 2.1326, 1.7711, 2.4074, 2.1414], device='cuda:3'), covar=tensor([0.1553, 0.2789, 0.3960, 0.3093, 0.2947, 0.2064, 0.3392, 0.2118], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0190, 0.0234, 0.0253, 0.0233, 0.0192, 0.0212, 0.0192], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 08:22:43,656 INFO [finetune.py:976] (3/7) Epoch 7, batch 4400, loss[loss=0.2632, simple_loss=0.3157, pruned_loss=0.1054, over 4120.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2666, pruned_loss=0.07284, over 951669.59 frames. ], batch size: 65, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:23:41,098 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6659, 2.9037, 2.3574, 1.9014, 2.8208, 2.8075, 2.7480, 2.3368], device='cuda:3'), covar=tensor([0.0649, 0.0555, 0.0873, 0.1012, 0.0437, 0.0796, 0.0723, 0.1056], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0134, 0.0144, 0.0126, 0.0114, 0.0146, 0.0146, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 08:23:42,200 INFO [finetune.py:976] (3/7) Epoch 7, batch 4450, loss[loss=0.2326, simple_loss=0.3065, pruned_loss=0.07938, over 4743.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2711, pruned_loss=0.07402, over 950872.64 frames. ], batch size: 54, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:23:51,889 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.5313, 3.9045, 4.1538, 4.3768, 4.2775, 3.9922, 4.5753, 1.4055], device='cuda:3'), covar=tensor([0.0682, 0.0799, 0.0789, 0.0853, 0.1069, 0.1430, 0.0636, 0.5156], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0243, 0.0275, 0.0294, 0.0332, 0.0282, 0.0304, 0.0295], 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') 2023-03-26 08:23:51,918 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38823.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:23:53,601 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.165e+02 1.712e+02 1.965e+02 2.330e+02 4.727e+02, threshold=3.929e+02, percent-clipped=4.0 2023-03-26 08:24:21,831 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8208, 1.7142, 1.5487, 1.9629, 2.5003, 1.9611, 1.6971, 1.4606], device='cuda:3'), covar=tensor([0.2226, 0.2012, 0.1906, 0.1583, 0.1651, 0.1128, 0.2338, 0.1879], device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0208, 0.0204, 0.0186, 0.0238, 0.0177, 0.0213, 0.0191], 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') 2023-03-26 08:24:44,780 INFO [finetune.py:976] (3/7) Epoch 7, batch 4500, loss[loss=0.243, simple_loss=0.2964, pruned_loss=0.09479, over 4767.00 frames. ], tot_loss[loss=0.2107, simple_loss=0.2727, pruned_loss=0.07441, over 951759.22 frames. ], batch size: 28, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:25:05,705 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.0625, 4.3688, 4.5764, 4.8556, 4.7349, 4.5377, 5.1636, 1.3788], device='cuda:3'), covar=tensor([0.0759, 0.0906, 0.0772, 0.0904, 0.1272, 0.1386, 0.0564, 0.5599], device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0242, 0.0273, 0.0293, 0.0331, 0.0281, 0.0303, 0.0294], 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') 2023-03-26 08:25:06,351 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38884.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:25:49,098 INFO [finetune.py:976] (3/7) Epoch 7, batch 4550, loss[loss=0.2328, simple_loss=0.2916, pruned_loss=0.08699, over 4847.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2734, pruned_loss=0.07446, over 952869.51 frames. ], batch size: 49, lr: 3.86e-03, grad_scale: 64.0 2023-03-26 08:25:59,510 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.041e+02 1.671e+02 2.000e+02 2.524e+02 3.434e+02, threshold=4.000e+02, percent-clipped=0.0 2023-03-26 08:26:47,278 INFO [finetune.py:976] (3/7) Epoch 7, batch 4600, loss[loss=0.1645, simple_loss=0.2343, pruned_loss=0.04736, over 4728.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.272, pruned_loss=0.07317, over 954156.66 frames. ], batch size: 54, lr: 3.86e-03, grad_scale: 64.0 2023-03-26 08:27:38,937 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-03-26 08:27:54,986 INFO [finetune.py:976] (3/7) Epoch 7, batch 4650, loss[loss=0.1736, simple_loss=0.246, pruned_loss=0.05057, over 4898.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2688, pruned_loss=0.07212, over 954318.90 frames. ], batch size: 32, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:27:55,105 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39016.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:28:06,170 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.761e+01 1.505e+02 1.924e+02 2.345e+02 4.238e+02, threshold=3.847e+02, percent-clipped=2.0 2023-03-26 08:28:28,401 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-03-26 08:28:50,939 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=39064.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:28:57,994 INFO [finetune.py:976] (3/7) Epoch 7, batch 4700, loss[loss=0.1721, simple_loss=0.2356, pruned_loss=0.05432, over 4765.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2666, pruned_loss=0.07137, over 955227.81 frames. ], batch size: 27, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:29:41,871 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.91 vs. limit=5.0 2023-03-26 08:29:56,018 INFO [finetune.py:976] (3/7) Epoch 7, batch 4750, loss[loss=0.2415, simple_loss=0.2951, pruned_loss=0.09393, over 4906.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2624, pruned_loss=0.06959, over 954361.69 frames. ], batch size: 32, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:29:58,985 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8321, 1.1013, 1.7210, 1.6189, 1.5243, 1.5110, 1.5304, 1.6114], device='cuda:3'), covar=tensor([0.4304, 0.5775, 0.4763, 0.5399, 0.6488, 0.4854, 0.6446, 0.4582], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0241, 0.0253, 0.0253, 0.0243, 0.0219, 0.0271, 0.0225], 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') 2023-03-26 08:30:05,797 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5091, 3.4020, 3.2242, 1.5837, 3.4890, 2.5684, 0.6977, 2.3181], device='cuda:3'), covar=tensor([0.2540, 0.2054, 0.1648, 0.3533, 0.1184, 0.1115, 0.4644, 0.1598], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0173, 0.0162, 0.0129, 0.0154, 0.0122, 0.0147, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 08:30:08,812 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.221e+02 1.609e+02 1.819e+02 2.206e+02 4.512e+02, threshold=3.638e+02, percent-clipped=2.0 2023-03-26 08:30:58,730 INFO [finetune.py:976] (3/7) Epoch 7, batch 4800, loss[loss=0.214, simple_loss=0.2892, pruned_loss=0.06939, over 4895.00 frames. ], tot_loss[loss=0.204, simple_loss=0.266, pruned_loss=0.07098, over 954912.72 frames. ], batch size: 35, lr: 3.86e-03, grad_scale: 32.0 2023-03-26 08:31:12,897 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39179.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:31:22,013 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39185.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:31:56,919 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-03-26 08:31:57,086 INFO [finetune.py:976] (3/7) Epoch 7, batch 4850, loss[loss=0.2436, simple_loss=0.3008, pruned_loss=0.0932, over 4889.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2691, pruned_loss=0.07268, over 952801.39 frames. ], batch size: 35, lr: 3.86e-03, grad_scale: 16.0 2023-03-26 08:32:06,049 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.092e+02 1.691e+02 2.004e+02 2.499e+02 4.240e+02, threshold=4.008e+02, percent-clipped=2.0 2023-03-26 08:32:12,758 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4934, 1.5379, 1.3185, 1.5100, 1.7340, 1.6275, 1.5353, 1.2947], device='cuda:3'), covar=tensor([0.0294, 0.0257, 0.0499, 0.0227, 0.0205, 0.0437, 0.0254, 0.0375], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0111, 0.0140, 0.0116, 0.0104, 0.0101, 0.0091, 0.0110], device='cuda:3'), out_proj_covar=tensor([6.9467e-05, 8.7534e-05, 1.1212e-04, 9.0990e-05, 8.2127e-05, 7.5158e-05, 6.9184e-05, 8.5534e-05], device='cuda:3') 2023-03-26 08:32:15,700 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5319, 1.6790, 0.8017, 2.4918, 2.5961, 1.8886, 2.0624, 2.2521], device='cuda:3'), covar=tensor([0.1270, 0.1915, 0.2229, 0.0981, 0.1775, 0.2055, 0.1370, 0.1829], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0098, 0.0114, 0.0093, 0.0124, 0.0096, 0.0100, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 08:32:18,155 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39246.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:32:21,158 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9656, 1.2268, 1.6547, 1.7374, 1.5667, 1.5659, 1.6000, 1.6143], device='cuda:3'), covar=tensor([0.5073, 0.6348, 0.5360, 0.5329, 0.6549, 0.5109, 0.7193, 0.4988], device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0242, 0.0254, 0.0253, 0.0243, 0.0220, 0.0272, 0.0225], 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') 2023-03-26 08:32:30,570 INFO [finetune.py:976] (3/7) Epoch 7, batch 4900, loss[loss=0.2066, simple_loss=0.289, pruned_loss=0.06207, over 4758.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2704, pruned_loss=0.07298, over 952708.74 frames. ], batch size: 26, lr: 3.86e-03, grad_scale: 16.0 2023-03-26 08:32:32,914 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.2016, 3.6572, 3.8645, 4.0172, 3.9570, 3.7943, 4.2899, 1.3767], device='cuda:3'), covar=tensor([0.0691, 0.0794, 0.0723, 0.0804, 0.1142, 0.1331, 0.0637, 0.4935], device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0242, 0.0273, 0.0291, 0.0329, 0.0280, 0.0302, 0.0294], 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') 2023-03-26 08:32:43,213 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39284.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:32:52,740 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-03-26 08:33:03,517 INFO [finetune.py:976] (3/7) Epoch 7, batch 4950, loss[loss=0.1818, simple_loss=0.2452, pruned_loss=0.05923, over 4895.00 frames. ], tot_loss[loss=0.211, simple_loss=0.2727, pruned_loss=0.07462, over 950669.62 frames. ], batch size: 35, lr: 3.86e-03, grad_scale: 16.0 2023-03-26 08:33:12,240 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4739, 1.4692, 1.6409, 1.8021, 1.5637, 3.2947, 1.3519, 1.6927], device='cuda:3'), covar=tensor([0.0969, 0.1782, 0.1147, 0.0955, 0.1719, 0.0287, 0.1543, 0.1740], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0082, 0.0076, 0.0078, 0.0092, 0.0084, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 08:33:12,728 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.085e+02 1.619e+02 1.980e+02 2.423e+02 3.796e+02, threshold=3.961e+02, percent-clipped=0.0 2023-03-26 08:33:19,014 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.04 vs. limit=2.0 2023-03-26 08:33:24,238 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39345.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:33:24,588 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-03-26 08:33:37,226 INFO [finetune.py:976] (3/7) Epoch 7, batch 5000, loss[loss=0.145, simple_loss=0.2078, pruned_loss=0.04112, over 4779.00 frames. ], tot_loss[loss=0.2085, simple_loss=0.2704, pruned_loss=0.07329, over 953601.56 frames. ], batch size: 26, lr: 3.86e-03, grad_scale: 16.0 2023-03-26 08:33:37,947 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.5901, 1.6712, 1.6261, 0.8960, 1.7386, 1.9697, 1.9193, 1.4870], device='cuda:3'), covar=tensor([0.0831, 0.0564, 0.0413, 0.0550, 0.0379, 0.0436, 0.0276, 0.0655], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0155, 0.0120, 0.0136, 0.0130, 0.0123, 0.0143, 0.0144], device='cuda:3'), out_proj_covar=tensor([9.5564e-05, 1.1404e-04, 8.6548e-05, 9.8784e-05, 9.3529e-05, 9.0811e-05, 1.0523e-04, 1.0629e-04], device='cuda:3') 2023-03-26 08:33:41,490 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0784, 1.9359, 1.5798, 1.9534, 2.1369, 1.7537, 2.3922, 2.1650], device='cuda:3'), covar=tensor([0.1526, 0.3043, 0.3568, 0.3247, 0.2759, 0.1854, 0.4023, 0.2055], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0191, 0.0235, 0.0255, 0.0235, 0.0193, 0.0212, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 08:34:10,433 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4564, 2.9848, 2.4561, 1.8562, 2.8400, 2.7892, 2.6538, 2.4505], device='cuda:3'), covar=tensor([0.0739, 0.0473, 0.0807, 0.0983, 0.0415, 0.0849, 0.0700, 0.0860], device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0135, 0.0145, 0.0127, 0.0114, 0.0146, 0.0147, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 08:34:10,935 INFO [finetune.py:976] (3/7) Epoch 7, batch 5050, loss[loss=0.1977, simple_loss=0.2551, pruned_loss=0.07014, over 4833.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.267, pruned_loss=0.07226, over 953199.92 frames. ], batch size: 30, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:34:19,594 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.096e+02 1.623e+02 1.955e+02 2.404e+02 3.498e+02, threshold=3.910e+02, percent-clipped=0.0 2023-03-26 08:34:52,727 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39463.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:34:54,408 INFO [finetune.py:976] (3/7) Epoch 7, batch 5100, loss[loss=0.2104, simple_loss=0.266, pruned_loss=0.07739, over 4918.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2649, pruned_loss=0.07156, over 953020.08 frames. ], batch size: 36, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:35:05,350 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39479.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:35:39,560 INFO [finetune.py:976] (3/7) Epoch 7, batch 5150, loss[loss=0.2457, simple_loss=0.3123, pruned_loss=0.08955, over 4808.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2641, pruned_loss=0.07125, over 953477.16 frames. ], batch size: 41, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:35:47,021 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39524.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:35:48,748 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=39527.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:35:49,815 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.405e+01 1.710e+02 2.016e+02 2.412e+02 5.054e+02, threshold=4.032e+02, percent-clipped=2.0 2023-03-26 08:36:08,539 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39541.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:36:24,413 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 2023-03-26 08:36:24,640 INFO [finetune.py:976] (3/7) Epoch 7, batch 5200, loss[loss=0.2932, simple_loss=0.3403, pruned_loss=0.123, over 4190.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2659, pruned_loss=0.07113, over 950684.13 frames. ], batch size: 65, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:37:07,820 INFO [finetune.py:976] (3/7) Epoch 7, batch 5250, loss[loss=0.1985, simple_loss=0.2621, pruned_loss=0.06749, over 4814.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2677, pruned_loss=0.07129, over 949085.31 frames. ], batch size: 51, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:37:15,022 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.077e+02 1.709e+02 2.070e+02 2.577e+02 5.953e+02, threshold=4.140e+02, percent-clipped=1.0 2023-03-26 08:37:25,993 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39639.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 08:37:26,592 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39640.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:37:43,681 INFO [finetune.py:976] (3/7) Epoch 7, batch 5300, loss[loss=0.2096, simple_loss=0.2663, pruned_loss=0.07642, over 4762.00 frames. ], tot_loss[loss=0.2084, simple_loss=0.271, pruned_loss=0.07293, over 950442.15 frames. ], batch size: 27, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:38:17,434 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39700.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 08:38:32,614 INFO [finetune.py:976] (3/7) Epoch 7, batch 5350, loss[loss=0.2083, simple_loss=0.2721, pruned_loss=0.07222, over 4925.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.271, pruned_loss=0.07221, over 952177.75 frames. ], batch size: 33, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:38:40,830 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.014e+02 1.565e+02 1.855e+02 2.323e+02 5.491e+02, threshold=3.710e+02, percent-clipped=1.0 2023-03-26 08:39:15,939 INFO [finetune.py:976] (3/7) Epoch 7, batch 5400, loss[loss=0.2072, simple_loss=0.2669, pruned_loss=0.07378, over 4817.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2687, pruned_loss=0.07147, over 951526.72 frames. ], batch size: 41, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:39:16,662 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39767.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:39:23,332 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-03-26 08:39:26,677 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4475, 1.5456, 1.5840, 1.6661, 1.6829, 2.8204, 1.3633, 1.6093], device='cuda:3'), covar=tensor([0.0892, 0.1464, 0.1171, 0.0876, 0.1258, 0.0328, 0.1242, 0.1404], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0081, 0.0075, 0.0078, 0.0092, 0.0083, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 08:39:50,823 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39815.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:39:51,314 INFO [finetune.py:976] (3/7) Epoch 7, batch 5450, loss[loss=0.187, simple_loss=0.25, pruned_loss=0.06199, over 4815.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2659, pruned_loss=0.07057, over 951742.11 frames. ], batch size: 51, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:39:53,200 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39819.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:40:03,208 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7548, 1.5723, 1.4167, 1.2070, 1.5810, 1.6013, 1.5513, 2.0809], device='cuda:3'), covar=tensor([0.4975, 0.4997, 0.3987, 0.4578, 0.4335, 0.2816, 0.4304, 0.1990], device='cuda:3'), in_proj_covar=tensor([0.0281, 0.0256, 0.0219, 0.0278, 0.0239, 0.0204, 0.0243, 0.0203], 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') 2023-03-26 08:40:03,635 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.339e+01 1.599e+02 1.928e+02 2.299e+02 3.698e+02, threshold=3.856e+02, percent-clipped=0.0 2023-03-26 08:40:03,741 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39828.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:40:12,028 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-26 08:40:15,682 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1559, 1.7610, 2.0485, 2.0201, 1.7292, 1.7976, 1.9302, 1.8833], device='cuda:3'), covar=tensor([0.5164, 0.6607, 0.5166, 0.6198, 0.7455, 0.5469, 0.7587, 0.4957], device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0243, 0.0256, 0.0254, 0.0245, 0.0222, 0.0274, 0.0227], 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') 2023-03-26 08:40:16,807 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39841.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:40:54,100 INFO [finetune.py:976] (3/7) Epoch 7, batch 5500, loss[loss=0.1925, simple_loss=0.2449, pruned_loss=0.07005, over 4822.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2628, pruned_loss=0.06961, over 953443.74 frames. ], batch size: 30, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:41:05,338 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39876.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:41:18,430 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=39889.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:41:57,092 INFO [finetune.py:976] (3/7) Epoch 7, batch 5550, loss[loss=0.2023, simple_loss=0.2718, pruned_loss=0.06634, over 4913.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2675, pruned_loss=0.07254, over 954490.27 frames. ], batch size: 43, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:42:09,838 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.056e+02 1.647e+02 1.995e+02 2.278e+02 3.177e+02, threshold=3.991e+02, percent-clipped=0.0 2023-03-26 08:42:28,009 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39940.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:42:58,490 INFO [finetune.py:976] (3/7) Epoch 7, batch 5600, loss[loss=0.2032, simple_loss=0.2568, pruned_loss=0.07483, over 4739.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2711, pruned_loss=0.0735, over 956001.39 frames. ], batch size: 23, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:43:20,787 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=39988.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:43:30,226 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39995.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 08:43:52,202 INFO [finetune.py:976] (3/7) Epoch 7, batch 5650, loss[loss=0.2196, simple_loss=0.2939, pruned_loss=0.07266, over 4802.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2743, pruned_loss=0.07468, over 954673.66 frames. ], batch size: 45, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:44:08,617 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8414, 1.1870, 1.6304, 1.6991, 1.4655, 1.5171, 1.5543, 1.5420], device='cuda:3'), covar=tensor([0.5237, 0.6281, 0.5557, 0.5462, 0.6969, 0.5577, 0.6755, 0.5127], device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0242, 0.0256, 0.0254, 0.0245, 0.0221, 0.0273, 0.0227], 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') 2023-03-26 08:44:09,054 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.182e+02 1.647e+02 1.995e+02 2.469e+02 4.643e+02, threshold=3.989e+02, percent-clipped=3.0 2023-03-26 08:44:29,840 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.8587, 1.8632, 2.1305, 1.3627, 2.0189, 2.1735, 2.1875, 1.8093], device='cuda:3'), covar=tensor([0.0922, 0.0642, 0.0436, 0.0667, 0.0444, 0.0673, 0.0318, 0.0546], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0158, 0.0121, 0.0138, 0.0133, 0.0126, 0.0146, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.7082e-05, 1.1638e-04, 8.7430e-05, 1.0026e-04, 9.5574e-05, 9.2435e-05, 1.0731e-04, 1.0825e-04], device='cuda:3') 2023-03-26 08:44:33,123 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1679, 1.9489, 1.7676, 1.9363, 1.8563, 1.9091, 1.8794, 2.6603], device='cuda:3'), covar=tensor([0.5350, 0.5677, 0.4201, 0.4968, 0.5209, 0.3140, 0.5521, 0.2006], device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0259, 0.0221, 0.0281, 0.0242, 0.0206, 0.0246, 0.0206], 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') 2023-03-26 08:44:50,691 INFO [finetune.py:976] (3/7) Epoch 7, batch 5700, loss[loss=0.1837, simple_loss=0.2385, pruned_loss=0.06441, over 4051.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2703, pruned_loss=0.07449, over 936129.61 frames. ], batch size: 17, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:45:04,311 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.7180, 1.8468, 1.9883, 1.1902, 1.9223, 2.1136, 2.1799, 1.7577], device='cuda:3'), covar=tensor([0.0969, 0.0596, 0.0386, 0.0625, 0.0409, 0.0654, 0.0285, 0.0568], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0157, 0.0121, 0.0137, 0.0132, 0.0125, 0.0145, 0.0146], device='cuda:3'), out_proj_covar=tensor([9.6737e-05, 1.1601e-04, 8.7068e-05, 9.9917e-05, 9.5063e-05, 9.2174e-05, 1.0693e-04, 1.0780e-04], device='cuda:3') 2023-03-26 08:45:42,067 INFO [finetune.py:976] (3/7) Epoch 8, batch 0, loss[loss=0.2457, simple_loss=0.2919, pruned_loss=0.09976, over 4870.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.2919, pruned_loss=0.09976, over 4870.00 frames. ], batch size: 32, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:45:42,068 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 08:45:49,245 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4101, 1.2277, 1.2362, 1.3598, 1.6221, 1.4895, 1.3775, 1.2122], device='cuda:3'), covar=tensor([0.0389, 0.0325, 0.0632, 0.0261, 0.0255, 0.0450, 0.0363, 0.0470], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0113, 0.0140, 0.0116, 0.0105, 0.0101, 0.0092, 0.0110], device='cuda:3'), out_proj_covar=tensor([7.0022e-05, 8.8500e-05, 1.1228e-04, 9.1282e-05, 8.2386e-05, 7.5241e-05, 6.9340e-05, 8.5588e-05], device='cuda:3') 2023-03-26 08:45:57,865 INFO [finetune.py:1010] (3/7) Epoch 8, validation: loss=0.1624, simple_loss=0.234, pruned_loss=0.04544, over 2265189.00 frames. 2023-03-26 08:45:57,866 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 08:46:20,314 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40119.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:46:20,722 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-26 08:46:26,526 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40123.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:46:29,511 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.031e+02 1.580e+02 2.018e+02 2.508e+02 5.130e+02, threshold=4.036e+02, percent-clipped=1.0 2023-03-26 08:46:41,216 INFO [finetune.py:976] (3/7) Epoch 8, batch 50, loss[loss=0.2117, simple_loss=0.2693, pruned_loss=0.0771, over 4843.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2773, pruned_loss=0.07453, over 217849.78 frames. ], batch size: 44, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:46:44,273 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9123, 1.6692, 1.7404, 1.8612, 1.3363, 4.6232, 1.6795, 2.4817], device='cuda:3'), covar=tensor([0.3359, 0.2648, 0.2139, 0.2329, 0.1881, 0.0120, 0.2379, 0.1287], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0116, 0.0120, 0.0123, 0.0117, 0.0099, 0.0101, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 08:47:08,172 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=40167.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:47:10,677 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40171.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:47:13,610 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-26 08:47:14,328 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1999, 2.6631, 2.5946, 1.2716, 2.8299, 2.0672, 0.9286, 1.8502], device='cuda:3'), covar=tensor([0.3073, 0.2034, 0.1801, 0.2970, 0.1352, 0.1127, 0.3419, 0.1502], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0172, 0.0161, 0.0130, 0.0155, 0.0122, 0.0145, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 08:47:26,488 INFO [finetune.py:976] (3/7) Epoch 8, batch 100, loss[loss=0.1795, simple_loss=0.2502, pruned_loss=0.05436, over 4898.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2699, pruned_loss=0.07306, over 382309.00 frames. ], batch size: 35, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:47:27,193 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40195.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:47:32,198 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.6839, 1.5306, 1.4445, 0.8803, 1.5011, 1.7224, 1.7292, 1.4577], device='cuda:3'), covar=tensor([0.0845, 0.0515, 0.0523, 0.0594, 0.0412, 0.0464, 0.0265, 0.0587], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0157, 0.0121, 0.0137, 0.0133, 0.0126, 0.0146, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.6892e-05, 1.1612e-04, 8.7611e-05, 1.0006e-04, 9.5540e-05, 9.2499e-05, 1.0733e-04, 1.0818e-04], device='cuda:3') 2023-03-26 08:47:42,316 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40218.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:47:48,307 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.035e+02 1.570e+02 1.832e+02 2.394e+02 3.868e+02, threshold=3.663e+02, percent-clipped=0.0 2023-03-26 08:47:59,303 INFO [finetune.py:976] (3/7) Epoch 8, batch 150, loss[loss=0.1977, simple_loss=0.2434, pruned_loss=0.07606, over 4825.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2641, pruned_loss=0.07075, over 508439.55 frames. ], batch size: 25, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:48:07,721 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40256.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:48:18,417 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.1947, 2.6878, 2.5791, 1.3209, 2.7326, 2.2830, 2.1704, 2.4012], device='cuda:3'), covar=tensor([0.1200, 0.1074, 0.1792, 0.2464, 0.1778, 0.2532, 0.2075, 0.1312], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0201, 0.0203, 0.0188, 0.0219, 0.0207, 0.0224, 0.0198], 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') 2023-03-26 08:48:18,419 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40272.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:48:22,640 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40279.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:48:33,049 INFO [finetune.py:976] (3/7) Epoch 8, batch 200, loss[loss=0.2493, simple_loss=0.3061, pruned_loss=0.09624, over 4800.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2632, pruned_loss=0.07102, over 607924.56 frames. ], batch size: 45, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:48:33,761 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40295.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 08:48:55,738 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.120e+02 1.652e+02 1.957e+02 2.371e+02 3.958e+02, threshold=3.914e+02, percent-clipped=3.0 2023-03-26 08:48:57,114 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40330.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:48:58,977 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40333.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:49:05,954 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=40343.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 08:49:06,467 INFO [finetune.py:976] (3/7) Epoch 8, batch 250, loss[loss=0.2714, simple_loss=0.3284, pruned_loss=0.1071, over 4825.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.2671, pruned_loss=0.07203, over 685778.90 frames. ], batch size: 38, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:49:37,962 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40391.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:49:40,040 INFO [finetune.py:976] (3/7) Epoch 8, batch 300, loss[loss=0.2448, simple_loss=0.2871, pruned_loss=0.1012, over 4879.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.271, pruned_loss=0.07409, over 745485.86 frames. ], batch size: 31, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:49:59,864 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40423.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:50:07,992 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.190e+02 1.680e+02 2.022e+02 2.440e+02 4.521e+02, threshold=4.043e+02, percent-clipped=1.0 2023-03-26 08:50:27,980 INFO [finetune.py:976] (3/7) Epoch 8, batch 350, loss[loss=0.1596, simple_loss=0.2195, pruned_loss=0.04985, over 4709.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2715, pruned_loss=0.07414, over 792805.32 frames. ], batch size: 23, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:51:00,930 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.95 vs. limit=5.0 2023-03-26 08:51:01,400 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=40471.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:51:01,448 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40471.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:51:26,213 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40492.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:51:27,355 INFO [finetune.py:976] (3/7) Epoch 8, batch 400, loss[loss=0.2681, simple_loss=0.3221, pruned_loss=0.107, over 4765.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2721, pruned_loss=0.0739, over 827279.99 frames. ], batch size: 28, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:51:52,934 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=40519.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:51:58,844 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.015e+02 1.662e+02 2.008e+02 2.590e+02 4.107e+02, threshold=4.016e+02, percent-clipped=2.0 2023-03-26 08:52:11,112 INFO [finetune.py:976] (3/7) Epoch 8, batch 450, loss[loss=0.1719, simple_loss=0.2391, pruned_loss=0.05234, over 4816.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.271, pruned_loss=0.07337, over 856362.67 frames. ], batch size: 39, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:52:21,157 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40551.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:52:22,008 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-26 08:52:26,842 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40553.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:52:38,701 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4414, 2.3997, 1.9580, 2.5652, 2.4116, 2.1271, 2.9686, 2.5328], device='cuda:3'), covar=tensor([0.1579, 0.2947, 0.3670, 0.3149, 0.2882, 0.1791, 0.3932, 0.2227], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0189, 0.0235, 0.0255, 0.0235, 0.0193, 0.0211, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 08:52:41,172 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40574.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:52:54,277 INFO [finetune.py:976] (3/7) Epoch 8, batch 500, loss[loss=0.1877, simple_loss=0.2459, pruned_loss=0.06475, over 4834.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2686, pruned_loss=0.07231, over 880676.47 frames. ], batch size: 30, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:53:17,850 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.064e+02 1.648e+02 1.946e+02 2.379e+02 4.476e+02, threshold=3.892e+02, percent-clipped=1.0 2023-03-26 08:53:17,929 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40628.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:53:28,115 INFO [finetune.py:976] (3/7) Epoch 8, batch 550, loss[loss=0.1938, simple_loss=0.2555, pruned_loss=0.06608, over 4819.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2663, pruned_loss=0.07175, over 896799.16 frames. ], batch size: 33, lr: 3.85e-03, grad_scale: 16.0 2023-03-26 08:53:32,518 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40651.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:53:56,757 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40686.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:54:01,540 INFO [finetune.py:976] (3/7) Epoch 8, batch 600, loss[loss=0.2356, simple_loss=0.2995, pruned_loss=0.08585, over 4922.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2674, pruned_loss=0.07161, over 911407.53 frames. ], batch size: 38, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 08:54:14,594 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40712.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:54:15,386 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-26 08:54:24,593 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.141e+02 1.757e+02 2.080e+02 2.524e+02 4.426e+02, threshold=4.160e+02, percent-clipped=1.0 2023-03-26 08:54:34,714 INFO [finetune.py:976] (3/7) Epoch 8, batch 650, loss[loss=0.1706, simple_loss=0.2346, pruned_loss=0.05323, over 4827.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2704, pruned_loss=0.07271, over 921709.72 frames. ], batch size: 25, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 08:55:08,423 INFO [finetune.py:976] (3/7) Epoch 8, batch 700, loss[loss=0.2394, simple_loss=0.2885, pruned_loss=0.09516, over 4798.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2706, pruned_loss=0.07242, over 928828.39 frames. ], batch size: 25, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 08:55:10,749 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-03-26 08:55:16,443 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1289, 1.9576, 1.6536, 1.8963, 2.0891, 1.8360, 2.3190, 2.0910], device='cuda:3'), covar=tensor([0.1489, 0.2645, 0.3640, 0.3034, 0.2808, 0.1817, 0.3542, 0.2115], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0190, 0.0235, 0.0255, 0.0235, 0.0193, 0.0211, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 08:55:19,315 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1835, 2.0384, 1.7153, 2.2311, 2.2670, 1.9508, 2.4887, 2.1896], device='cuda:3'), covar=tensor([0.1482, 0.2900, 0.3808, 0.2930, 0.2747, 0.1726, 0.3692, 0.2138], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0190, 0.0235, 0.0255, 0.0235, 0.0193, 0.0211, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 08:55:20,552 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9863, 1.6934, 2.3825, 1.5117, 2.1653, 2.0770, 1.6921, 2.3545], device='cuda:3'), covar=tensor([0.1517, 0.2113, 0.1558, 0.2336, 0.0959, 0.1649, 0.2670, 0.0903], device='cuda:3'), in_proj_covar=tensor([0.0205, 0.0207, 0.0200, 0.0197, 0.0183, 0.0222, 0.0222, 0.0203], 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') 2023-03-26 08:55:25,055 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1615, 3.6100, 3.7664, 3.9966, 3.8915, 3.6823, 4.2787, 1.3057], device='cuda:3'), covar=tensor([0.0783, 0.0789, 0.0795, 0.0911, 0.1317, 0.1490, 0.0689, 0.5285], device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0240, 0.0272, 0.0289, 0.0330, 0.0278, 0.0300, 0.0293], 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') 2023-03-26 08:55:31,876 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.288e+02 1.702e+02 1.948e+02 2.422e+02 4.930e+02, threshold=3.896e+02, percent-clipped=3.0 2023-03-26 08:55:40,851 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9393, 1.8822, 1.6885, 1.9752, 2.5052, 1.9778, 1.8477, 1.4261], device='cuda:3'), covar=tensor([0.2308, 0.2140, 0.1871, 0.1686, 0.2087, 0.1228, 0.2420, 0.1968], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0209, 0.0205, 0.0188, 0.0240, 0.0178, 0.0215, 0.0193], 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') 2023-03-26 08:55:51,723 INFO [finetune.py:976] (3/7) Epoch 8, batch 750, loss[loss=0.2339, simple_loss=0.303, pruned_loss=0.08238, over 4894.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.271, pruned_loss=0.07235, over 933938.95 frames. ], batch size: 36, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 08:55:54,219 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40848.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:56:01,149 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40851.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:56:20,353 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40865.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:56:30,279 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0863, 1.9601, 1.5543, 0.7011, 1.7165, 1.7278, 1.4986, 1.8160], device='cuda:3'), covar=tensor([0.0864, 0.0651, 0.1341, 0.1824, 0.1243, 0.2162, 0.2279, 0.0832], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0201, 0.0202, 0.0189, 0.0219, 0.0207, 0.0224, 0.0199], 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') 2023-03-26 08:56:32,130 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40874.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:56:55,830 INFO [finetune.py:976] (3/7) Epoch 8, batch 800, loss[loss=0.2101, simple_loss=0.2659, pruned_loss=0.07718, over 4820.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2692, pruned_loss=0.07082, over 938442.12 frames. ], batch size: 39, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 08:57:03,992 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=40899.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:57:04,650 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40900.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:57:23,554 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=40922.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:57:26,543 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40926.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:57:27,573 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.008e+02 1.606e+02 1.985e+02 2.397e+02 9.945e+02, threshold=3.971e+02, percent-clipped=3.0 2023-03-26 08:57:27,708 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40928.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:57:45,602 INFO [finetune.py:976] (3/7) Epoch 8, batch 850, loss[loss=0.1485, simple_loss=0.2151, pruned_loss=0.04099, over 4908.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2659, pruned_loss=0.06985, over 942688.27 frames. ], batch size: 43, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 08:58:00,373 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40961.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 08:58:05,826 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8945, 1.7510, 1.6720, 2.0189, 2.4888, 2.0435, 1.4086, 1.5827], device='cuda:3'), covar=tensor([0.2372, 0.2273, 0.2006, 0.1830, 0.1925, 0.1156, 0.2898, 0.2046], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0210, 0.0205, 0.0189, 0.0241, 0.0179, 0.0216, 0.0194], 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') 2023-03-26 08:58:10,971 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=40976.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:58:18,120 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40986.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:58:22,822 INFO [finetune.py:976] (3/7) Epoch 8, batch 900, loss[loss=0.2147, simple_loss=0.2589, pruned_loss=0.08527, over 4811.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2637, pruned_loss=0.06925, over 947672.99 frames. ], batch size: 25, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 08:58:25,182 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40997.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:58:31,500 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41007.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:58:46,165 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.012e+02 1.525e+02 1.868e+02 2.283e+02 3.598e+02, threshold=3.736e+02, percent-clipped=0.0 2023-03-26 08:58:50,349 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=41034.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:58:56,787 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-03-26 08:58:56,855 INFO [finetune.py:976] (3/7) Epoch 8, batch 950, loss[loss=0.2061, simple_loss=0.2778, pruned_loss=0.06724, over 4797.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2637, pruned_loss=0.06965, over 951164.85 frames. ], batch size: 29, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 08:58:57,583 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41045.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:59:06,016 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41058.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:59:21,894 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 2023-03-26 08:59:30,583 INFO [finetune.py:976] (3/7) Epoch 8, batch 1000, loss[loss=0.3214, simple_loss=0.3499, pruned_loss=0.1464, over 4109.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2672, pruned_loss=0.07154, over 952961.45 frames. ], batch size: 65, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 08:59:36,020 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41102.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:59:38,365 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41106.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 08:59:39,584 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4362, 1.2468, 1.1742, 1.4438, 1.5292, 1.4339, 0.7904, 1.1850], device='cuda:3'), covar=tensor([0.2351, 0.2462, 0.2151, 0.1770, 0.1887, 0.1339, 0.3029, 0.1985], device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0211, 0.0207, 0.0189, 0.0242, 0.0180, 0.0217, 0.0195], 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') 2023-03-26 08:59:52,966 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.084e+02 1.652e+02 2.000e+02 2.359e+02 4.809e+02, threshold=4.000e+02, percent-clipped=2.0 2023-03-26 09:00:04,080 INFO [finetune.py:976] (3/7) Epoch 8, batch 1050, loss[loss=0.1663, simple_loss=0.2371, pruned_loss=0.04777, over 4770.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2699, pruned_loss=0.07218, over 953728.28 frames. ], batch size: 28, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 09:00:06,628 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41148.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:00:16,282 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41163.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:00:18,703 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2838, 1.6396, 0.8902, 2.2325, 2.6303, 1.9062, 2.1830, 2.1623], device='cuda:3'), covar=tensor([0.1508, 0.2126, 0.2239, 0.1159, 0.1973, 0.1982, 0.1445, 0.2072], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0097, 0.0114, 0.0092, 0.0123, 0.0096, 0.0101, 0.0093], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 09:00:26,985 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7927, 1.0556, 1.7219, 1.5614, 1.4222, 1.4159, 1.4600, 1.6119], device='cuda:3'), covar=tensor([0.4067, 0.5568, 0.4371, 0.5172, 0.6053, 0.4361, 0.6111, 0.4017], device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0241, 0.0254, 0.0254, 0.0244, 0.0221, 0.0272, 0.0227], 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') 2023-03-26 09:00:37,444 INFO [finetune.py:976] (3/7) Epoch 8, batch 1100, loss[loss=0.2315, simple_loss=0.2796, pruned_loss=0.09174, over 4767.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.271, pruned_loss=0.07267, over 951848.09 frames. ], batch size: 28, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 09:00:38,741 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=41196.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:00:54,969 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41221.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:00:59,686 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.124e+02 1.750e+02 2.155e+02 2.664e+02 4.791e+02, threshold=4.309e+02, percent-clipped=2.0 2023-03-26 09:01:06,296 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5063, 2.4009, 2.0112, 0.9384, 2.1945, 1.8746, 1.8141, 2.1319], device='cuda:3'), covar=tensor([0.0883, 0.0770, 0.1486, 0.2284, 0.1493, 0.2377, 0.2076, 0.1008], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0201, 0.0203, 0.0190, 0.0219, 0.0208, 0.0224, 0.0200], 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') 2023-03-26 09:01:17,460 INFO [finetune.py:976] (3/7) Epoch 8, batch 1150, loss[loss=0.2454, simple_loss=0.2977, pruned_loss=0.09653, over 4814.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2729, pruned_loss=0.07333, over 954324.17 frames. ], batch size: 33, lr: 3.84e-03, grad_scale: 32.0 2023-03-26 09:01:32,473 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41256.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 09:01:45,012 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9794, 1.9019, 1.6648, 1.9642, 1.8589, 1.8101, 1.8325, 2.6478], device='cuda:3'), covar=tensor([0.5826, 0.7019, 0.4596, 0.6349, 0.6160, 0.3378, 0.6520, 0.2226], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0257, 0.0220, 0.0279, 0.0241, 0.0205, 0.0243, 0.0205], 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') 2023-03-26 09:02:15,112 INFO [finetune.py:976] (3/7) Epoch 8, batch 1200, loss[loss=0.1705, simple_loss=0.2373, pruned_loss=0.0519, over 4766.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2714, pruned_loss=0.07262, over 954856.50 frames. ], batch size: 26, lr: 3.84e-03, grad_scale: 32.0 2023-03-26 09:02:24,085 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41307.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:02:37,270 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.297e+01 1.635e+02 1.914e+02 2.289e+02 4.123e+02, threshold=3.829e+02, percent-clipped=0.0 2023-03-26 09:02:51,371 INFO [finetune.py:976] (3/7) Epoch 8, batch 1250, loss[loss=0.1948, simple_loss=0.2572, pruned_loss=0.06624, over 4942.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2688, pruned_loss=0.07172, over 955741.68 frames. ], batch size: 38, lr: 3.84e-03, grad_scale: 32.0 2023-03-26 09:03:02,677 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41353.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:03:04,420 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=41355.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:03:32,985 INFO [finetune.py:976] (3/7) Epoch 8, batch 1300, loss[loss=0.2148, simple_loss=0.2792, pruned_loss=0.07519, over 4894.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2657, pruned_loss=0.07026, over 955713.57 frames. ], batch size: 35, lr: 3.84e-03, grad_scale: 32.0 2023-03-26 09:03:37,902 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41401.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:03:45,552 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7442, 1.1721, 1.0634, 1.6277, 2.0374, 1.3536, 1.5348, 1.6597], device='cuda:3'), covar=tensor([0.1295, 0.2001, 0.1799, 0.1078, 0.1899, 0.2067, 0.1357, 0.1640], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0097, 0.0114, 0.0092, 0.0124, 0.0096, 0.0101, 0.0093], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-26 09:03:56,247 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.077e+02 1.674e+02 1.900e+02 2.309e+02 4.379e+02, threshold=3.799e+02, percent-clipped=1.0 2023-03-26 09:04:06,255 INFO [finetune.py:976] (3/7) Epoch 8, batch 1350, loss[loss=0.1615, simple_loss=0.2272, pruned_loss=0.04794, over 4814.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2656, pruned_loss=0.07029, over 955548.59 frames. ], batch size: 25, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 09:04:16,298 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41458.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:04:39,875 INFO [finetune.py:976] (3/7) Epoch 8, batch 1400, loss[loss=0.2279, simple_loss=0.2939, pruned_loss=0.08093, over 4751.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2691, pruned_loss=0.07129, over 957526.44 frames. ], batch size: 27, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 09:04:56,047 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.74 vs. limit=5.0 2023-03-26 09:04:58,598 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41521.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:05:03,328 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.146e+02 1.704e+02 2.004e+02 2.444e+02 3.700e+02, threshold=4.008e+02, percent-clipped=0.0 2023-03-26 09:05:03,883 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-03-26 09:05:12,585 INFO [finetune.py:976] (3/7) Epoch 8, batch 1450, loss[loss=0.2704, simple_loss=0.3166, pruned_loss=0.1121, over 4864.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2696, pruned_loss=0.07096, over 956455.49 frames. ], batch size: 49, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 09:05:21,944 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41556.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 09:05:30,811 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=41569.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:05:41,835 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-03-26 09:05:46,314 INFO [finetune.py:976] (3/7) Epoch 8, batch 1500, loss[loss=0.1782, simple_loss=0.254, pruned_loss=0.05123, over 4764.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2693, pruned_loss=0.0705, over 957455.88 frames. ], batch size: 28, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 09:05:53,543 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=41604.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:06:10,825 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.231e+02 1.635e+02 1.924e+02 2.365e+02 3.634e+02, threshold=3.848e+02, percent-clipped=0.0 2023-03-26 09:06:22,441 INFO [finetune.py:976] (3/7) Epoch 8, batch 1550, loss[loss=0.2185, simple_loss=0.2908, pruned_loss=0.07309, over 4750.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2709, pruned_loss=0.07105, over 957354.75 frames. ], batch size: 26, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 09:06:34,417 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41653.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:06:35,049 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41654.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:07:19,720 INFO [finetune.py:976] (3/7) Epoch 8, batch 1600, loss[loss=0.2099, simple_loss=0.2644, pruned_loss=0.07766, over 4913.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2687, pruned_loss=0.07037, over 955790.68 frames. ], batch size: 36, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 09:07:25,549 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=41701.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:07:25,576 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41701.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:07:37,223 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 2023-03-26 09:07:39,588 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41715.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 09:07:48,907 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.136e+02 1.581e+02 1.949e+02 2.490e+02 4.755e+02, threshold=3.899e+02, percent-clipped=2.0 2023-03-26 09:07:58,439 INFO [finetune.py:976] (3/7) Epoch 8, batch 1650, loss[loss=0.2249, simple_loss=0.2725, pruned_loss=0.08867, over 4905.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2652, pruned_loss=0.0692, over 956765.48 frames. ], batch size: 37, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 09:08:01,529 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=41749.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:08:09,827 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41758.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:08:42,692 INFO [finetune.py:976] (3/7) Epoch 8, batch 1700, loss[loss=0.1716, simple_loss=0.2341, pruned_loss=0.05452, over 4817.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2623, pruned_loss=0.06801, over 956186.88 frames. ], batch size: 25, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 09:08:48,205 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-03-26 09:08:50,309 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=41806.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:09:06,690 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.178e+02 1.763e+02 2.034e+02 2.335e+02 4.675e+02, threshold=4.069e+02, percent-clipped=2.0 2023-03-26 09:09:16,758 INFO [finetune.py:976] (3/7) Epoch 8, batch 1750, loss[loss=0.2165, simple_loss=0.2868, pruned_loss=0.07306, over 4901.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2657, pruned_loss=0.06951, over 957926.78 frames. ], batch size: 37, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 09:09:19,317 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8255, 3.9527, 3.8282, 1.9363, 4.0381, 2.9726, 0.6697, 2.7214], device='cuda:3'), covar=tensor([0.2347, 0.1939, 0.1354, 0.3402, 0.0870, 0.1038, 0.4949, 0.1565], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0173, 0.0161, 0.0130, 0.0156, 0.0123, 0.0146, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 09:09:36,492 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9710, 1.3897, 0.8693, 1.8515, 2.1840, 1.4291, 1.8619, 1.7029], device='cuda:3'), covar=tensor([0.1356, 0.2110, 0.2134, 0.1135, 0.1938, 0.2069, 0.1344, 0.1888], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0098, 0.0115, 0.0092, 0.0124, 0.0096, 0.0101, 0.0093], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-26 09:09:50,595 INFO [finetune.py:976] (3/7) Epoch 8, batch 1800, loss[loss=0.21, simple_loss=0.2717, pruned_loss=0.07416, over 4896.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2692, pruned_loss=0.07135, over 958853.53 frames. ], batch size: 35, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 09:09:58,008 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41906.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:10:13,609 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.102e+02 1.792e+02 2.103e+02 2.633e+02 4.479e+02, threshold=4.207e+02, percent-clipped=2.0 2023-03-26 09:10:23,655 INFO [finetune.py:976] (3/7) Epoch 8, batch 1850, loss[loss=0.2002, simple_loss=0.2711, pruned_loss=0.06464, over 4897.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2699, pruned_loss=0.07165, over 957804.28 frames. ], batch size: 35, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 09:10:23,785 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7081, 1.5430, 1.3985, 1.4257, 1.8720, 1.7616, 1.6101, 1.3387], device='cuda:3'), covar=tensor([0.0262, 0.0318, 0.0557, 0.0348, 0.0208, 0.0446, 0.0343, 0.0428], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0113, 0.0141, 0.0118, 0.0105, 0.0102, 0.0092, 0.0111], device='cuda:3'), out_proj_covar=tensor([7.1267e-05, 8.8714e-05, 1.1305e-04, 9.2749e-05, 8.2887e-05, 7.5610e-05, 6.9745e-05, 8.5895e-05], device='cuda:3') 2023-03-26 09:10:26,655 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41948.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 09:10:27,948 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4606, 1.3531, 1.2424, 1.5073, 1.5646, 1.5659, 0.9153, 1.2593], device='cuda:3'), covar=tensor([0.2538, 0.2360, 0.2131, 0.1746, 0.1813, 0.1279, 0.2876, 0.2076], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0210, 0.0205, 0.0189, 0.0240, 0.0178, 0.0214, 0.0193], 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') 2023-03-26 09:10:38,713 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41967.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:10:57,329 INFO [finetune.py:976] (3/7) Epoch 8, batch 1900, loss[loss=0.1637, simple_loss=0.2339, pruned_loss=0.04674, over 4751.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2695, pruned_loss=0.07122, over 954115.36 frames. ], batch size: 26, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 09:11:08,252 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42009.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 09:11:08,801 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42010.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 09:11:22,116 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.207e+02 1.556e+02 1.920e+02 2.218e+02 3.872e+02, threshold=3.841e+02, percent-clipped=0.0 2023-03-26 09:11:32,122 INFO [finetune.py:976] (3/7) Epoch 8, batch 1950, loss[loss=0.1994, simple_loss=0.2704, pruned_loss=0.06424, over 4760.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2687, pruned_loss=0.0705, over 952826.77 frames. ], batch size: 27, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 09:11:32,241 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3805, 1.3530, 1.3979, 0.6653, 1.5646, 1.4431, 1.3579, 1.2775], device='cuda:3'), covar=tensor([0.0638, 0.0794, 0.0741, 0.1057, 0.0676, 0.0778, 0.0749, 0.1208], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0131, 0.0141, 0.0124, 0.0112, 0.0141, 0.0143, 0.0159], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 09:12:00,742 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-03-26 09:12:30,959 INFO [finetune.py:976] (3/7) Epoch 8, batch 2000, loss[loss=0.179, simple_loss=0.2461, pruned_loss=0.05596, over 4724.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2664, pruned_loss=0.06994, over 953901.46 frames. ], batch size: 59, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 09:12:39,863 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7374, 1.6070, 1.6182, 1.5636, 1.1744, 3.4022, 1.4068, 1.9897], device='cuda:3'), covar=tensor([0.3428, 0.2682, 0.2126, 0.2487, 0.2052, 0.0233, 0.2711, 0.1295], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0115, 0.0119, 0.0123, 0.0117, 0.0099, 0.0101, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 09:12:56,440 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.048e+02 1.506e+02 1.840e+02 2.176e+02 3.856e+02, threshold=3.679e+02, percent-clipped=1.0 2023-03-26 09:13:06,570 INFO [finetune.py:976] (3/7) Epoch 8, batch 2050, loss[loss=0.1712, simple_loss=0.2369, pruned_loss=0.05271, over 4776.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2635, pruned_loss=0.06935, over 954884.70 frames. ], batch size: 26, lr: 3.84e-03, grad_scale: 16.0 2023-03-26 09:13:35,947 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-26 09:13:53,279 INFO [finetune.py:976] (3/7) Epoch 8, batch 2100, loss[loss=0.2713, simple_loss=0.3263, pruned_loss=0.1082, over 4825.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2636, pruned_loss=0.07013, over 954312.66 frames. ], batch size: 40, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:14:16,279 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.091e+02 1.710e+02 1.945e+02 2.376e+02 4.149e+02, threshold=3.889e+02, percent-clipped=2.0 2023-03-26 09:14:26,550 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9707, 1.2245, 1.8753, 1.7720, 1.6627, 1.5995, 1.6614, 1.7234], device='cuda:3'), covar=tensor([0.4925, 0.6223, 0.5374, 0.5596, 0.6869, 0.5518, 0.7111, 0.4869], device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0241, 0.0255, 0.0254, 0.0245, 0.0222, 0.0272, 0.0227], 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') 2023-03-26 09:14:26,998 INFO [finetune.py:976] (3/7) Epoch 8, batch 2150, loss[loss=0.1638, simple_loss=0.2324, pruned_loss=0.04762, over 4737.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.268, pruned_loss=0.0715, over 954293.09 frames. ], batch size: 23, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:14:31,754 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.5893, 1.4956, 1.5088, 0.8742, 1.5821, 1.7640, 1.7044, 1.3776], device='cuda:3'), covar=tensor([0.1086, 0.0679, 0.0490, 0.0637, 0.0470, 0.0618, 0.0359, 0.0750], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0155, 0.0120, 0.0135, 0.0131, 0.0125, 0.0144, 0.0146], device='cuda:3'), out_proj_covar=tensor([9.5550e-05, 1.1422e-04, 8.6716e-05, 9.8414e-05, 9.3851e-05, 9.1706e-05, 1.0627e-04, 1.0771e-04], device='cuda:3') 2023-03-26 09:14:38,882 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42262.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:15:18,053 INFO [finetune.py:976] (3/7) Epoch 8, batch 2200, loss[loss=0.2218, simple_loss=0.2885, pruned_loss=0.07758, over 4824.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2707, pruned_loss=0.07243, over 954197.02 frames. ], batch size: 33, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:15:25,342 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42304.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 09:15:29,023 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42310.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 09:15:34,473 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8541, 1.0703, 1.7206, 1.7298, 1.5223, 1.4936, 1.5632, 1.5564], device='cuda:3'), covar=tensor([0.4158, 0.5482, 0.4484, 0.4634, 0.5525, 0.4502, 0.6058, 0.4333], device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0241, 0.0255, 0.0255, 0.0245, 0.0222, 0.0273, 0.0227], 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') 2023-03-26 09:15:45,966 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.121e+02 1.543e+02 1.921e+02 2.479e+02 5.347e+02, threshold=3.843e+02, percent-clipped=1.0 2023-03-26 09:16:07,395 INFO [finetune.py:976] (3/7) Epoch 8, batch 2250, loss[loss=0.2321, simple_loss=0.2875, pruned_loss=0.08837, over 4722.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2722, pruned_loss=0.07335, over 952721.36 frames. ], batch size: 54, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:16:27,422 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=42358.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:16:47,904 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42379.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:17:09,028 INFO [finetune.py:976] (3/7) Epoch 8, batch 2300, loss[loss=0.1728, simple_loss=0.2516, pruned_loss=0.04697, over 4786.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2713, pruned_loss=0.07235, over 955099.89 frames. ], batch size: 51, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:17:57,231 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.935e+01 1.473e+02 1.816e+02 2.175e+02 3.275e+02, threshold=3.633e+02, percent-clipped=0.0 2023-03-26 09:18:06,508 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42440.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:18:09,263 INFO [finetune.py:976] (3/7) Epoch 8, batch 2350, loss[loss=0.1775, simple_loss=0.2387, pruned_loss=0.05811, over 4229.00 frames. ], tot_loss[loss=0.205, simple_loss=0.268, pruned_loss=0.071, over 955378.09 frames. ], batch size: 65, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:18:35,850 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-26 09:18:49,244 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-03-26 09:18:51,873 INFO [finetune.py:976] (3/7) Epoch 8, batch 2400, loss[loss=0.1828, simple_loss=0.2506, pruned_loss=0.05747, over 4809.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2645, pruned_loss=0.06992, over 956342.86 frames. ], batch size: 25, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:19:02,409 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42506.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:19:16,750 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-26 09:19:25,681 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.076e+02 1.517e+02 1.798e+02 2.223e+02 5.682e+02, threshold=3.597e+02, percent-clipped=2.0 2023-03-26 09:19:28,657 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-26 09:19:35,402 INFO [finetune.py:976] (3/7) Epoch 8, batch 2450, loss[loss=0.1778, simple_loss=0.2405, pruned_loss=0.05757, over 4762.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2623, pruned_loss=0.06954, over 956297.92 frames. ], batch size: 27, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:19:47,992 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42562.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:19:51,987 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42567.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:20:08,910 INFO [finetune.py:976] (3/7) Epoch 8, batch 2500, loss[loss=0.204, simple_loss=0.2719, pruned_loss=0.0681, over 4845.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2651, pruned_loss=0.07107, over 953898.60 frames. ], batch size: 49, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:20:16,170 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42604.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 09:20:20,731 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=42610.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:20:33,719 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.056e+02 1.709e+02 1.979e+02 2.316e+02 5.134e+02, threshold=3.959e+02, percent-clipped=4.0 2023-03-26 09:20:42,883 INFO [finetune.py:976] (3/7) Epoch 8, batch 2550, loss[loss=0.2002, simple_loss=0.2651, pruned_loss=0.06764, over 4924.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2714, pruned_loss=0.07392, over 955526.22 frames. ], batch size: 33, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:20:46,539 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42649.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:20:53,417 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=42652.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 09:21:16,913 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.0095, 0.9250, 0.9047, 1.1261, 1.2087, 1.1044, 0.9801, 0.9635], device='cuda:3'), covar=tensor([0.0299, 0.0267, 0.0573, 0.0239, 0.0238, 0.0370, 0.0285, 0.0357], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0112, 0.0141, 0.0117, 0.0105, 0.0102, 0.0092, 0.0110], device='cuda:3'), out_proj_covar=tensor([7.0999e-05, 8.8057e-05, 1.1287e-04, 9.2091e-05, 8.2767e-05, 7.5704e-05, 6.9407e-05, 8.5598e-05], device='cuda:3') 2023-03-26 09:21:25,357 INFO [finetune.py:976] (3/7) Epoch 8, batch 2600, loss[loss=0.2104, simple_loss=0.2751, pruned_loss=0.07287, over 4905.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.272, pruned_loss=0.07407, over 955622.92 frames. ], batch size: 36, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:21:36,661 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42710.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:21:49,602 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.153e+02 1.771e+02 2.168e+02 2.787e+02 4.495e+02, threshold=4.337e+02, percent-clipped=4.0 2023-03-26 09:21:53,823 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42735.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:21:54,482 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3695, 2.1877, 2.7137, 1.6268, 2.4842, 2.6633, 1.9953, 2.9039], device='cuda:3'), covar=tensor([0.1552, 0.1908, 0.1710, 0.2496, 0.1047, 0.1649, 0.2579, 0.0841], device='cuda:3'), in_proj_covar=tensor([0.0201, 0.0204, 0.0196, 0.0195, 0.0181, 0.0219, 0.0218, 0.0199], 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') 2023-03-26 09:21:59,147 INFO [finetune.py:976] (3/7) Epoch 8, batch 2650, loss[loss=0.1684, simple_loss=0.2418, pruned_loss=0.04748, over 4789.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.2731, pruned_loss=0.07373, over 956021.68 frames. ], batch size: 29, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:22:29,289 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-03-26 09:22:40,399 INFO [finetune.py:976] (3/7) Epoch 8, batch 2700, loss[loss=0.1951, simple_loss=0.2693, pruned_loss=0.0605, over 4824.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2718, pruned_loss=0.07237, over 956743.49 frames. ], batch size: 39, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:22:59,294 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8805, 1.2310, 1.8216, 1.6823, 1.5013, 1.5141, 1.6286, 1.6633], device='cuda:3'), covar=tensor([0.4020, 0.4822, 0.3936, 0.4259, 0.5365, 0.4104, 0.5258, 0.3704], device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0242, 0.0255, 0.0255, 0.0246, 0.0223, 0.0274, 0.0228], 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') 2023-03-26 09:23:27,242 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1258, 1.7692, 2.4258, 1.5423, 2.2961, 2.3606, 1.7397, 2.5512], device='cuda:3'), covar=tensor([0.1357, 0.2033, 0.1345, 0.2223, 0.0894, 0.1435, 0.2437, 0.0738], device='cuda:3'), in_proj_covar=tensor([0.0202, 0.0204, 0.0197, 0.0196, 0.0182, 0.0220, 0.0219, 0.0201], 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') 2023-03-26 09:23:27,703 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.159e+02 1.604e+02 1.897e+02 2.218e+02 3.599e+02, threshold=3.793e+02, percent-clipped=0.0 2023-03-26 09:23:46,890 INFO [finetune.py:976] (3/7) Epoch 8, batch 2750, loss[loss=0.2273, simple_loss=0.276, pruned_loss=0.08927, over 4853.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.269, pruned_loss=0.07183, over 957637.88 frames. ], batch size: 44, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:23:55,937 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42859.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:24:02,854 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42862.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:24:14,742 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42879.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:24:21,965 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.69 vs. limit=5.0 2023-03-26 09:24:31,693 INFO [finetune.py:976] (3/7) Epoch 8, batch 2800, loss[loss=0.1776, simple_loss=0.2414, pruned_loss=0.05688, over 4895.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2643, pruned_loss=0.06963, over 954928.09 frames. ], batch size: 32, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:24:38,695 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-26 09:24:41,481 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6367, 1.4653, 1.8756, 1.2546, 1.5528, 1.7040, 1.4663, 1.9794], device='cuda:3'), covar=tensor([0.1106, 0.1849, 0.1230, 0.1562, 0.0863, 0.1272, 0.2388, 0.0725], device='cuda:3'), in_proj_covar=tensor([0.0204, 0.0206, 0.0199, 0.0198, 0.0184, 0.0223, 0.0221, 0.0202], 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') 2023-03-26 09:24:48,556 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42920.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:24:54,868 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.137e+02 1.632e+02 1.941e+02 2.379e+02 3.960e+02, threshold=3.882e+02, percent-clipped=2.0 2023-03-26 09:25:02,624 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42940.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:25:04,920 INFO [finetune.py:976] (3/7) Epoch 8, batch 2850, loss[loss=0.1916, simple_loss=0.2596, pruned_loss=0.06182, over 4808.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2645, pruned_loss=0.07035, over 956266.86 frames. ], batch size: 45, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:25:38,296 INFO [finetune.py:976] (3/7) Epoch 8, batch 2900, loss[loss=0.2123, simple_loss=0.2785, pruned_loss=0.07302, over 4911.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2686, pruned_loss=0.07205, over 956544.19 frames. ], batch size: 37, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:25:45,628 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43005.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:26:03,402 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.926e+01 1.731e+02 1.970e+02 2.373e+02 5.777e+02, threshold=3.941e+02, percent-clipped=2.0 2023-03-26 09:26:13,023 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43035.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:26:18,880 INFO [finetune.py:976] (3/7) Epoch 8, batch 2950, loss[loss=0.2358, simple_loss=0.3053, pruned_loss=0.08314, over 4835.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2702, pruned_loss=0.07184, over 957139.54 frames. ], batch size: 47, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:26:44,526 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=43083.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:26:49,176 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3921, 1.3472, 1.7498, 2.8752, 1.9812, 2.1474, 0.9341, 2.2502], device='cuda:3'), covar=tensor([0.1710, 0.1452, 0.1259, 0.0574, 0.0846, 0.1361, 0.1743, 0.0658], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0117, 0.0134, 0.0166, 0.0102, 0.0139, 0.0127, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 09:26:52,571 INFO [finetune.py:976] (3/7) Epoch 8, batch 3000, loss[loss=0.2366, simple_loss=0.2966, pruned_loss=0.08836, over 4921.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2715, pruned_loss=0.07255, over 956198.50 frames. ], batch size: 33, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:26:52,571 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 09:26:54,456 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4602, 1.6699, 1.4381, 1.7027, 1.7472, 3.0462, 1.5059, 1.7465], device='cuda:3'), covar=tensor([0.0870, 0.1503, 0.1036, 0.0864, 0.1301, 0.0318, 0.1239, 0.1468], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0081, 0.0075, 0.0078, 0.0092, 0.0082, 0.0084, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 09:26:58,583 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8110, 1.7102, 1.5988, 1.8823, 2.2116, 1.7810, 1.4702, 1.6061], device='cuda:3'), covar=tensor([0.1849, 0.1951, 0.1725, 0.1564, 0.1469, 0.1122, 0.2410, 0.1618], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0210, 0.0206, 0.0188, 0.0241, 0.0180, 0.0215, 0.0194], 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') 2023-03-26 09:26:59,339 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9182, 1.0585, 1.9207, 1.7219, 1.6045, 1.5357, 1.5913, 1.6897], device='cuda:3'), covar=tensor([0.4341, 0.5390, 0.4349, 0.5020, 0.6008, 0.4529, 0.6359, 0.4156], device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0241, 0.0253, 0.0254, 0.0245, 0.0222, 0.0273, 0.0227], 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') 2023-03-26 09:27:10,878 INFO [finetune.py:1010] (3/7) Epoch 8, validation: loss=0.16, simple_loss=0.2311, pruned_loss=0.04446, over 2265189.00 frames. 2023-03-26 09:27:10,879 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 09:27:49,850 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.022e+02 1.686e+02 2.049e+02 2.426e+02 3.920e+02, threshold=4.099e+02, percent-clipped=0.0 2023-03-26 09:27:52,958 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1450, 2.0691, 1.6302, 2.0009, 1.9254, 1.8857, 1.9987, 2.7063], device='cuda:3'), covar=tensor([0.5239, 0.5749, 0.4581, 0.5466, 0.4687, 0.3295, 0.5438, 0.2036], device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0258, 0.0220, 0.0279, 0.0240, 0.0205, 0.0244, 0.0206], 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') 2023-03-26 09:27:54,165 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7166, 2.3028, 1.9992, 0.9192, 2.2089, 2.0592, 1.9380, 2.2366], device='cuda:3'), covar=tensor([0.0823, 0.1036, 0.1697, 0.2196, 0.1803, 0.2146, 0.1972, 0.0994], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0201, 0.0202, 0.0187, 0.0218, 0.0206, 0.0223, 0.0198], 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') 2023-03-26 09:27:57,683 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6641, 1.8160, 1.8957, 1.9512, 1.8144, 3.6555, 1.6161, 1.7817], device='cuda:3'), covar=tensor([0.0942, 0.1563, 0.0990, 0.0899, 0.1434, 0.0265, 0.1361, 0.1551], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0080, 0.0075, 0.0078, 0.0091, 0.0082, 0.0084, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 09:28:00,453 INFO [finetune.py:976] (3/7) Epoch 8, batch 3050, loss[loss=0.2124, simple_loss=0.2432, pruned_loss=0.09077, over 4017.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2731, pruned_loss=0.07308, over 956479.12 frames. ], batch size: 17, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:28:13,425 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43162.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:28:16,471 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.2705, 2.9249, 3.0219, 3.1967, 3.0232, 2.8864, 3.3285, 0.9997], device='cuda:3'), covar=tensor([0.1179, 0.1023, 0.1159, 0.1253, 0.1736, 0.1769, 0.1154, 0.5243], device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0241, 0.0275, 0.0293, 0.0329, 0.0282, 0.0300, 0.0292], 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') 2023-03-26 09:28:36,035 INFO [finetune.py:976] (3/7) Epoch 8, batch 3100, loss[loss=0.1621, simple_loss=0.2383, pruned_loss=0.04293, over 4897.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2698, pruned_loss=0.07218, over 957408.91 frames. ], batch size: 36, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:28:52,833 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=43210.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:29:01,134 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43215.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:29:14,666 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.165e+02 1.652e+02 1.930e+02 2.337e+02 4.149e+02, threshold=3.860e+02, percent-clipped=1.0 2023-03-26 09:29:23,817 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43235.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:29:34,859 INFO [finetune.py:976] (3/7) Epoch 8, batch 3150, loss[loss=0.2036, simple_loss=0.266, pruned_loss=0.07058, over 4823.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2656, pruned_loss=0.07052, over 954178.35 frames. ], batch size: 38, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:30:24,889 INFO [finetune.py:976] (3/7) Epoch 8, batch 3200, loss[loss=0.2181, simple_loss=0.2728, pruned_loss=0.0817, over 4739.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2623, pruned_loss=0.06899, over 956218.67 frames. ], batch size: 23, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:30:32,647 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43305.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:30:49,478 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.070e+02 1.668e+02 2.087e+02 2.541e+02 1.424e+03, threshold=4.174e+02, percent-clipped=3.0 2023-03-26 09:31:03,656 INFO [finetune.py:976] (3/7) Epoch 8, batch 3250, loss[loss=0.2564, simple_loss=0.3161, pruned_loss=0.09836, over 4743.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2632, pruned_loss=0.06959, over 956247.87 frames. ], batch size: 54, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:31:15,389 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=43353.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:31:56,290 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6918, 2.4733, 2.3447, 1.3601, 2.4252, 2.0058, 1.8910, 2.2805], device='cuda:3'), covar=tensor([0.1254, 0.0813, 0.1666, 0.2148, 0.1793, 0.2262, 0.2084, 0.1148], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0201, 0.0202, 0.0188, 0.0220, 0.0207, 0.0224, 0.0199], 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') 2023-03-26 09:31:59,778 INFO [finetune.py:976] (3/7) Epoch 8, batch 3300, loss[loss=0.1817, simple_loss=0.23, pruned_loss=0.06669, over 4421.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.267, pruned_loss=0.07093, over 953157.79 frames. ], batch size: 19, lr: 3.83e-03, grad_scale: 16.0 2023-03-26 09:32:13,202 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1106, 1.9414, 2.2312, 0.8315, 2.4096, 2.5500, 2.1123, 1.9270], device='cuda:3'), covar=tensor([0.1155, 0.0871, 0.0443, 0.0856, 0.0563, 0.0831, 0.0502, 0.0739], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0155, 0.0119, 0.0136, 0.0132, 0.0125, 0.0145, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.5587e-05, 1.1439e-04, 8.6067e-05, 9.8577e-05, 9.4593e-05, 9.1461e-05, 1.0622e-04, 1.0793e-04], device='cuda:3') 2023-03-26 09:32:22,299 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5021, 1.9134, 2.9795, 1.8303, 2.4754, 2.8193, 2.1329, 2.9506], device='cuda:3'), covar=tensor([0.1489, 0.2394, 0.1469, 0.2555, 0.1209, 0.1602, 0.2898, 0.0942], device='cuda:3'), in_proj_covar=tensor([0.0199, 0.0202, 0.0196, 0.0194, 0.0180, 0.0219, 0.0218, 0.0199], 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') 2023-03-26 09:32:45,622 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.223e+02 1.691e+02 2.029e+02 2.462e+02 4.055e+02, threshold=4.059e+02, percent-clipped=0.0 2023-03-26 09:33:04,669 INFO [finetune.py:976] (3/7) Epoch 8, batch 3350, loss[loss=0.1871, simple_loss=0.2621, pruned_loss=0.0561, over 4815.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2695, pruned_loss=0.07206, over 951278.91 frames. ], batch size: 38, lr: 3.83e-03, grad_scale: 32.0 2023-03-26 09:33:34,786 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.36 vs. limit=5.0 2023-03-26 09:33:45,174 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9649, 1.8251, 1.6417, 1.7898, 1.7212, 1.7291, 1.7835, 2.4038], device='cuda:3'), covar=tensor([0.4153, 0.4476, 0.3396, 0.4047, 0.4250, 0.2490, 0.4304, 0.1592], device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0257, 0.0219, 0.0278, 0.0239, 0.0204, 0.0243, 0.0205], 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') 2023-03-26 09:33:51,120 INFO [finetune.py:976] (3/7) Epoch 8, batch 3400, loss[loss=0.1916, simple_loss=0.2601, pruned_loss=0.06149, over 4703.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2716, pruned_loss=0.07277, over 952186.73 frames. ], batch size: 59, lr: 3.83e-03, grad_scale: 32.0 2023-03-26 09:34:00,408 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9369, 1.8211, 1.5109, 1.6315, 1.7208, 1.7275, 1.7811, 2.4239], device='cuda:3'), covar=tensor([0.4725, 0.4848, 0.4043, 0.4988, 0.4834, 0.2802, 0.4824, 0.1906], device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0257, 0.0220, 0.0278, 0.0239, 0.0205, 0.0243, 0.0205], 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') 2023-03-26 09:34:04,484 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43514.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:34:05,086 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43515.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:34:10,489 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.35 vs. limit=5.0 2023-03-26 09:34:15,393 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.089e+02 1.653e+02 1.866e+02 2.233e+02 4.638e+02, threshold=3.733e+02, percent-clipped=1.0 2023-03-26 09:34:18,551 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.66 vs. limit=2.0 2023-03-26 09:34:19,644 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43535.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:34:25,010 INFO [finetune.py:976] (3/7) Epoch 8, batch 3450, loss[loss=0.1905, simple_loss=0.2589, pruned_loss=0.06109, over 4828.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2721, pruned_loss=0.07309, over 953048.79 frames. ], batch size: 47, lr: 3.83e-03, grad_scale: 32.0 2023-03-26 09:34:43,308 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=43563.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:34:55,916 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43575.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:35:02,197 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=43583.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:35:08,906 INFO [finetune.py:976] (3/7) Epoch 8, batch 3500, loss[loss=0.2172, simple_loss=0.2587, pruned_loss=0.08781, over 4732.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2698, pruned_loss=0.0727, over 953287.86 frames. ], batch size: 23, lr: 3.83e-03, grad_scale: 32.0 2023-03-26 09:35:31,782 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6425, 1.4953, 1.5017, 1.6619, 1.3048, 3.5378, 1.3943, 2.0466], device='cuda:3'), covar=tensor([0.3281, 0.2446, 0.2112, 0.2193, 0.1808, 0.0177, 0.2708, 0.1205], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0118, 0.0122, 0.0116, 0.0097, 0.0100, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 09:35:33,682 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0339, 1.8585, 1.6125, 1.7176, 1.7062, 1.7000, 1.7795, 2.5226], device='cuda:3'), covar=tensor([0.4878, 0.5247, 0.3914, 0.5014, 0.5043, 0.2882, 0.4912, 0.2016], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0259, 0.0222, 0.0280, 0.0241, 0.0206, 0.0245, 0.0207], 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') 2023-03-26 09:35:34,548 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.199e+02 1.655e+02 2.028e+02 2.395e+02 4.370e+02, threshold=4.057e+02, percent-clipped=4.0 2023-03-26 09:35:44,696 INFO [finetune.py:976] (3/7) Epoch 8, batch 3550, loss[loss=0.1342, simple_loss=0.2066, pruned_loss=0.03086, over 4875.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2661, pruned_loss=0.07094, over 955637.21 frames. ], batch size: 31, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:36:18,005 INFO [finetune.py:976] (3/7) Epoch 8, batch 3600, loss[loss=0.1691, simple_loss=0.2399, pruned_loss=0.04916, over 4765.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2634, pruned_loss=0.06998, over 956197.67 frames. ], batch size: 28, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:36:40,280 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.156e+02 1.641e+02 1.836e+02 2.142e+02 3.900e+02, threshold=3.673e+02, percent-clipped=0.0 2023-03-26 09:36:44,128 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-03-26 09:37:03,252 INFO [finetune.py:976] (3/7) Epoch 8, batch 3650, loss[loss=0.2758, simple_loss=0.3368, pruned_loss=0.1074, over 4833.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2659, pruned_loss=0.07078, over 955828.41 frames. ], batch size: 47, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:37:14,435 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2539, 2.1777, 1.7536, 2.1935, 2.3247, 1.9280, 2.6170, 2.1783], device='cuda:3'), covar=tensor([0.1591, 0.2477, 0.3604, 0.2718, 0.2596, 0.1954, 0.2956, 0.2218], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0188, 0.0233, 0.0254, 0.0236, 0.0194, 0.0211, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 09:37:35,839 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9216, 1.6318, 2.2998, 1.5559, 2.1799, 2.1033, 1.6067, 2.3141], device='cuda:3'), covar=tensor([0.1337, 0.1982, 0.1526, 0.2065, 0.0849, 0.1592, 0.2811, 0.0858], device='cuda:3'), in_proj_covar=tensor([0.0201, 0.0204, 0.0197, 0.0195, 0.0181, 0.0220, 0.0219, 0.0200], 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') 2023-03-26 09:37:41,733 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43787.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:37:53,053 INFO [finetune.py:976] (3/7) Epoch 8, batch 3700, loss[loss=0.2288, simple_loss=0.2909, pruned_loss=0.08334, over 4932.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2692, pruned_loss=0.07196, over 955488.25 frames. ], batch size: 42, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:38:37,141 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.066e+01 1.637e+02 2.055e+02 2.501e+02 4.825e+02, threshold=4.110e+02, percent-clipped=4.0 2023-03-26 09:38:56,855 INFO [finetune.py:976] (3/7) Epoch 8, batch 3750, loss[loss=0.1904, simple_loss=0.2679, pruned_loss=0.0565, over 4835.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2699, pruned_loss=0.07179, over 953588.89 frames. ], batch size: 47, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:39:00,433 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43848.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:39:13,804 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43870.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:39:30,691 INFO [finetune.py:976] (3/7) Epoch 8, batch 3800, loss[loss=0.1739, simple_loss=0.2386, pruned_loss=0.05459, over 4801.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.27, pruned_loss=0.07155, over 953417.56 frames. ], batch size: 25, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:39:40,882 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43909.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:40:01,520 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.167e+02 1.597e+02 1.980e+02 2.453e+02 5.062e+02, threshold=3.959e+02, percent-clipped=3.0 2023-03-26 09:40:16,149 INFO [finetune.py:976] (3/7) Epoch 8, batch 3850, loss[loss=0.2449, simple_loss=0.2996, pruned_loss=0.09515, over 4822.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2684, pruned_loss=0.07094, over 953648.63 frames. ], batch size: 30, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:40:26,732 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43959.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:40:33,729 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43970.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:40:54,210 INFO [finetune.py:976] (3/7) Epoch 8, batch 3900, loss[loss=0.1993, simple_loss=0.2537, pruned_loss=0.07248, over 4906.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2647, pruned_loss=0.06921, over 955802.28 frames. ], batch size: 35, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:41:17,072 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44020.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:41:22,369 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.038e+02 1.530e+02 1.807e+02 2.225e+02 3.627e+02, threshold=3.614e+02, percent-clipped=0.0 2023-03-26 09:41:32,027 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9930, 2.0032, 2.1300, 1.2616, 2.1649, 2.0798, 1.9357, 1.7149], device='cuda:3'), covar=tensor([0.0544, 0.0633, 0.0613, 0.0915, 0.0541, 0.0770, 0.0655, 0.1146], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0132, 0.0144, 0.0125, 0.0115, 0.0143, 0.0144, 0.0160], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 09:41:32,530 INFO [finetune.py:976] (3/7) Epoch 8, batch 3950, loss[loss=0.197, simple_loss=0.2632, pruned_loss=0.06536, over 4865.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2617, pruned_loss=0.06839, over 955543.25 frames. ], batch size: 31, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:42:06,485 INFO [finetune.py:976] (3/7) Epoch 8, batch 4000, loss[loss=0.1894, simple_loss=0.253, pruned_loss=0.06293, over 4767.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2611, pruned_loss=0.06829, over 954992.44 frames. ], batch size: 27, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:42:10,722 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7697, 3.9507, 3.7736, 1.9126, 4.0528, 3.0008, 1.1569, 2.8411], device='cuda:3'), covar=tensor([0.2637, 0.1992, 0.1447, 0.3482, 0.1044, 0.1065, 0.4294, 0.1523], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0172, 0.0160, 0.0129, 0.0155, 0.0123, 0.0146, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 09:42:37,513 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.105e+02 1.793e+02 2.145e+02 2.588e+02 4.712e+02, threshold=4.291e+02, percent-clipped=10.0 2023-03-26 09:42:56,497 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44143.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:42:57,049 INFO [finetune.py:976] (3/7) Epoch 8, batch 4050, loss[loss=0.2022, simple_loss=0.2739, pruned_loss=0.06522, over 4718.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2653, pruned_loss=0.0702, over 953814.60 frames. ], batch size: 59, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:43:31,080 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44170.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:43:32,290 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2322, 1.2915, 1.2070, 1.3872, 1.3963, 2.3824, 1.1735, 1.4442], device='cuda:3'), covar=tensor([0.0966, 0.1847, 0.1083, 0.0910, 0.1735, 0.0423, 0.1617, 0.1766], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0082, 0.0076, 0.0079, 0.0093, 0.0083, 0.0086, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 09:43:39,849 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9398, 1.9603, 1.9829, 1.3363, 1.9083, 2.0186, 1.9506, 1.5837], device='cuda:3'), covar=tensor([0.0664, 0.0654, 0.0730, 0.1045, 0.0659, 0.0833, 0.0700, 0.1226], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0133, 0.0145, 0.0126, 0.0116, 0.0145, 0.0145, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 09:43:49,919 INFO [finetune.py:976] (3/7) Epoch 8, batch 4100, loss[loss=0.2638, simple_loss=0.3118, pruned_loss=0.1079, over 4903.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2688, pruned_loss=0.07178, over 952592.97 frames. ], batch size: 36, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:43:55,549 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0912, 2.1860, 2.1418, 1.4111, 2.2300, 2.1001, 2.0649, 1.8136], device='cuda:3'), covar=tensor([0.0667, 0.0643, 0.0783, 0.1062, 0.0514, 0.0859, 0.0821, 0.1168], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0133, 0.0145, 0.0126, 0.0116, 0.0145, 0.0146, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 09:44:15,790 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=44218.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:44:22,434 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.086e+02 1.696e+02 1.895e+02 2.360e+02 4.949e+02, threshold=3.791e+02, percent-clipped=1.0 2023-03-26 09:44:31,519 INFO [finetune.py:976] (3/7) Epoch 8, batch 4150, loss[loss=0.2368, simple_loss=0.2977, pruned_loss=0.08791, over 4875.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.27, pruned_loss=0.07181, over 953183.97 frames. ], batch size: 32, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:44:46,754 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44265.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:45:07,274 INFO [finetune.py:976] (3/7) Epoch 8, batch 4200, loss[loss=0.1989, simple_loss=0.2659, pruned_loss=0.06598, over 4916.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2698, pruned_loss=0.07151, over 951508.29 frames. ], batch size: 38, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:45:28,245 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7592, 1.7126, 1.5791, 2.0437, 2.4034, 1.9540, 1.5837, 1.3891], device='cuda:3'), covar=tensor([0.2284, 0.2204, 0.1934, 0.1561, 0.1803, 0.1129, 0.2471, 0.1979], device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0208, 0.0205, 0.0187, 0.0239, 0.0178, 0.0213, 0.0193], 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') 2023-03-26 09:45:30,607 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44315.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:45:40,277 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-26 09:45:40,453 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.242e+02 1.746e+02 1.975e+02 2.337e+02 5.106e+02, threshold=3.951e+02, percent-clipped=1.0 2023-03-26 09:45:54,906 INFO [finetune.py:976] (3/7) Epoch 8, batch 4250, loss[loss=0.2066, simple_loss=0.2577, pruned_loss=0.07778, over 4908.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2683, pruned_loss=0.07094, over 954262.28 frames. ], batch size: 46, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:46:32,566 INFO [finetune.py:976] (3/7) Epoch 8, batch 4300, loss[loss=0.1566, simple_loss=0.2266, pruned_loss=0.0433, over 4743.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2644, pruned_loss=0.06945, over 953796.12 frames. ], batch size: 27, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:46:56,824 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.088e+02 1.550e+02 1.851e+02 2.365e+02 4.860e+02, threshold=3.701e+02, percent-clipped=2.0 2023-03-26 09:47:05,831 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44443.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:47:06,351 INFO [finetune.py:976] (3/7) Epoch 8, batch 4350, loss[loss=0.2791, simple_loss=0.3143, pruned_loss=0.122, over 4841.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2621, pruned_loss=0.06901, over 950865.37 frames. ], batch size: 51, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:47:19,547 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2708, 2.3414, 2.1000, 2.5027, 2.9433, 2.3235, 2.0725, 1.8314], device='cuda:3'), covar=tensor([0.2273, 0.1958, 0.1877, 0.1635, 0.1802, 0.1126, 0.2379, 0.1951], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0210, 0.0207, 0.0188, 0.0241, 0.0180, 0.0215, 0.0195], 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') 2023-03-26 09:47:37,686 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=44491.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:47:39,433 INFO [finetune.py:976] (3/7) Epoch 8, batch 4400, loss[loss=0.2007, simple_loss=0.2704, pruned_loss=0.0655, over 4864.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2635, pruned_loss=0.06994, over 952983.38 frames. ], batch size: 34, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:47:57,235 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-03-26 09:48:17,053 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44527.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:48:18,612 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.261e+02 1.746e+02 1.995e+02 2.515e+02 6.158e+02, threshold=3.991e+02, percent-clipped=2.0 2023-03-26 09:48:28,720 INFO [finetune.py:976] (3/7) Epoch 8, batch 4450, loss[loss=0.248, simple_loss=0.3154, pruned_loss=0.09032, over 4809.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2684, pruned_loss=0.07171, over 953985.60 frames. ], batch size: 45, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:48:37,005 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44550.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:48:46,310 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1642, 2.1559, 2.1769, 0.9083, 2.3909, 2.5727, 2.2401, 1.9702], device='cuda:3'), covar=tensor([0.1065, 0.0672, 0.0438, 0.0794, 0.0455, 0.0699, 0.0380, 0.0716], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0155, 0.0120, 0.0136, 0.0132, 0.0125, 0.0146, 0.0148], device='cuda:3'), out_proj_covar=tensor([9.5902e-05, 1.1423e-04, 8.7031e-05, 9.8780e-05, 9.4813e-05, 9.2118e-05, 1.0704e-04, 1.0883e-04], device='cuda:3') 2023-03-26 09:48:51,543 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44565.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:49:19,355 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44588.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:49:28,022 INFO [finetune.py:976] (3/7) Epoch 8, batch 4500, loss[loss=0.2475, simple_loss=0.3071, pruned_loss=0.09401, over 4802.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2698, pruned_loss=0.07183, over 954492.85 frames. ], batch size: 41, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:49:47,691 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44611.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:49:48,883 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=44613.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:49:50,664 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44615.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:49:52,927 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7612, 1.6498, 1.9597, 2.0254, 1.8445, 3.7345, 1.4838, 1.8704], device='cuda:3'), covar=tensor([0.0881, 0.1714, 0.0987, 0.0942, 0.1455, 0.0257, 0.1481, 0.1636], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0081, 0.0075, 0.0079, 0.0092, 0.0083, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 09:50:00,002 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.305e+02 1.720e+02 2.026e+02 2.465e+02 5.780e+02, threshold=4.053e+02, percent-clipped=2.0 2023-03-26 09:50:03,070 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0167, 1.9001, 1.5084, 1.8284, 1.8676, 1.7639, 1.8597, 2.5307], device='cuda:3'), covar=tensor([0.5569, 0.5371, 0.4620, 0.5539, 0.5147, 0.3492, 0.5073, 0.2411], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0259, 0.0220, 0.0280, 0.0240, 0.0205, 0.0244, 0.0207], 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') 2023-03-26 09:50:10,549 INFO [finetune.py:976] (3/7) Epoch 8, batch 4550, loss[loss=0.2062, simple_loss=0.2703, pruned_loss=0.07103, over 4679.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2706, pruned_loss=0.07192, over 954405.05 frames. ], batch size: 23, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:50:27,740 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=44663.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:50:52,807 INFO [finetune.py:976] (3/7) Epoch 8, batch 4600, loss[loss=0.1849, simple_loss=0.2582, pruned_loss=0.0558, over 4908.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2684, pruned_loss=0.07045, over 952918.69 frames. ], batch size: 37, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:51:15,458 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.778e+01 1.566e+02 1.839e+02 2.114e+02 3.234e+02, threshold=3.678e+02, percent-clipped=0.0 2023-03-26 09:51:19,732 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8748, 1.8543, 1.7466, 2.1018, 2.3532, 2.0597, 1.6104, 1.6065], device='cuda:3'), covar=tensor([0.2171, 0.2041, 0.1801, 0.1565, 0.1808, 0.1103, 0.2610, 0.1919], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0210, 0.0206, 0.0188, 0.0241, 0.0179, 0.0215, 0.0195], 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') 2023-03-26 09:51:25,985 INFO [finetune.py:976] (3/7) Epoch 8, batch 4650, loss[loss=0.1923, simple_loss=0.2599, pruned_loss=0.06232, over 4817.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2656, pruned_loss=0.06966, over 954462.03 frames. ], batch size: 30, lr: 3.82e-03, grad_scale: 32.0 2023-03-26 09:51:59,448 INFO [finetune.py:976] (3/7) Epoch 8, batch 4700, loss[loss=0.1769, simple_loss=0.2373, pruned_loss=0.05824, over 4873.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2623, pruned_loss=0.06855, over 956512.15 frames. ], batch size: 34, lr: 3.82e-03, grad_scale: 16.0 2023-03-26 09:52:22,743 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 1.499e+02 1.877e+02 2.319e+02 4.193e+02, threshold=3.754e+02, percent-clipped=1.0 2023-03-26 09:52:32,230 INFO [finetune.py:976] (3/7) Epoch 8, batch 4750, loss[loss=0.1977, simple_loss=0.2642, pruned_loss=0.06557, over 4836.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2597, pruned_loss=0.06745, over 956722.83 frames. ], batch size: 33, lr: 3.82e-03, grad_scale: 16.0 2023-03-26 09:52:58,205 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44883.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:53:05,273 INFO [finetune.py:976] (3/7) Epoch 8, batch 4800, loss[loss=0.1949, simple_loss=0.2748, pruned_loss=0.05751, over 4822.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2646, pruned_loss=0.07018, over 956763.98 frames. ], batch size: 40, lr: 3.82e-03, grad_scale: 16.0 2023-03-26 09:53:15,870 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44906.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:53:15,956 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9949, 1.3184, 1.8357, 1.8523, 1.5977, 1.6050, 1.7239, 1.7384], device='cuda:3'), covar=tensor([0.4676, 0.5339, 0.4421, 0.4665, 0.6105, 0.4663, 0.6057, 0.4251], device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0242, 0.0254, 0.0256, 0.0248, 0.0224, 0.0275, 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') 2023-03-26 09:53:41,287 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.083e+02 1.589e+02 1.932e+02 2.383e+02 4.430e+02, threshold=3.864e+02, percent-clipped=2.0 2023-03-26 09:53:55,401 INFO [finetune.py:976] (3/7) Epoch 8, batch 4850, loss[loss=0.2163, simple_loss=0.2747, pruned_loss=0.07893, over 4816.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2682, pruned_loss=0.07082, over 957646.46 frames. ], batch size: 30, lr: 3.82e-03, grad_scale: 16.0 2023-03-26 09:54:24,946 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9658, 1.7914, 1.5311, 1.7332, 1.7174, 1.6551, 1.7515, 2.4159], device='cuda:3'), covar=tensor([0.4936, 0.5626, 0.4170, 0.5237, 0.4851, 0.2945, 0.5204, 0.2220], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0259, 0.0221, 0.0279, 0.0240, 0.0206, 0.0244, 0.0207], 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') 2023-03-26 09:54:57,803 INFO [finetune.py:976] (3/7) Epoch 8, batch 4900, loss[loss=0.2108, simple_loss=0.2872, pruned_loss=0.06722, over 4855.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2704, pruned_loss=0.07153, over 957155.01 frames. ], batch size: 44, lr: 3.82e-03, grad_scale: 16.0 2023-03-26 09:55:20,218 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.2671, 1.3030, 1.3309, 0.7241, 1.2124, 1.5180, 1.5641, 1.2036], device='cuda:3'), covar=tensor([0.0917, 0.0505, 0.0443, 0.0516, 0.0434, 0.0625, 0.0277, 0.0648], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0158, 0.0123, 0.0138, 0.0133, 0.0126, 0.0147, 0.0150], device='cuda:3'), out_proj_covar=tensor([9.7603e-05, 1.1640e-04, 8.8708e-05, 1.0027e-04, 9.5529e-05, 9.2611e-05, 1.0820e-04, 1.1028e-04], device='cuda:3') 2023-03-26 09:55:25,592 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.182e+02 1.694e+02 2.008e+02 2.325e+02 4.035e+02, threshold=4.016e+02, percent-clipped=1.0 2023-03-26 09:55:44,370 INFO [finetune.py:976] (3/7) Epoch 8, batch 4950, loss[loss=0.1888, simple_loss=0.2642, pruned_loss=0.0567, over 4851.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.271, pruned_loss=0.0714, over 956942.31 frames. ], batch size: 44, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 09:55:57,042 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45056.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:56:13,222 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45081.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:56:21,541 INFO [finetune.py:976] (3/7) Epoch 8, batch 5000, loss[loss=0.1948, simple_loss=0.2518, pruned_loss=0.0689, over 4804.00 frames. ], tot_loss[loss=0.205, simple_loss=0.2688, pruned_loss=0.07064, over 958505.14 frames. ], batch size: 40, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 09:56:37,061 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45117.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 09:56:45,213 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.085e+02 1.621e+02 1.919e+02 2.483e+02 3.797e+02, threshold=3.837e+02, percent-clipped=0.0 2023-03-26 09:56:52,661 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45142.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:56:53,773 INFO [finetune.py:976] (3/7) Epoch 8, batch 5050, loss[loss=0.1557, simple_loss=0.222, pruned_loss=0.04472, over 4905.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2658, pruned_loss=0.06958, over 957214.45 frames. ], batch size: 32, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 09:56:54,364 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7017, 2.4266, 2.4011, 1.5805, 2.5011, 2.0498, 1.8463, 2.2225], device='cuda:3'), covar=tensor([0.1005, 0.0693, 0.1450, 0.1781, 0.1703, 0.1771, 0.1870, 0.1047], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0200, 0.0200, 0.0187, 0.0215, 0.0204, 0.0220, 0.0195], 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') 2023-03-26 09:57:14,722 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2864, 2.1577, 1.7586, 2.3303, 2.2649, 1.8949, 2.6161, 2.2495], device='cuda:3'), covar=tensor([0.1445, 0.2511, 0.3423, 0.2787, 0.2582, 0.1788, 0.2812, 0.2114], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0189, 0.0233, 0.0253, 0.0235, 0.0193, 0.0211, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 09:57:19,974 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45183.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:57:26,577 INFO [finetune.py:976] (3/7) Epoch 8, batch 5100, loss[loss=0.1969, simple_loss=0.2551, pruned_loss=0.06939, over 4132.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2627, pruned_loss=0.06836, over 957514.86 frames. ], batch size: 18, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 09:57:35,008 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45206.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:57:55,114 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.141e+02 1.630e+02 1.903e+02 2.262e+02 3.588e+02, threshold=3.806e+02, percent-clipped=0.0 2023-03-26 09:57:55,792 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=45231.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:58:04,089 INFO [finetune.py:976] (3/7) Epoch 8, batch 5150, loss[loss=0.2238, simple_loss=0.285, pruned_loss=0.0813, over 4842.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2622, pruned_loss=0.0686, over 955706.97 frames. ], batch size: 49, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 09:58:06,017 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6984, 1.5022, 2.1214, 3.2108, 2.2253, 2.3805, 1.1004, 2.5034], device='cuda:3'), covar=tensor([0.1985, 0.1744, 0.1653, 0.0855, 0.0943, 0.1637, 0.1982, 0.0786], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0118, 0.0136, 0.0167, 0.0104, 0.0140, 0.0128, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:3') 2023-03-26 09:58:11,270 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=45254.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 09:58:40,359 INFO [finetune.py:976] (3/7) Epoch 8, batch 5200, loss[loss=0.1728, simple_loss=0.2472, pruned_loss=0.04924, over 4865.00 frames. ], tot_loss[loss=0.2023, simple_loss=0.2652, pruned_loss=0.06974, over 953218.91 frames. ], batch size: 31, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 09:59:04,299 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9302, 4.4254, 4.2352, 2.2818, 4.5060, 3.2716, 0.8624, 3.1303], device='cuda:3'), covar=tensor([0.2596, 0.1845, 0.1360, 0.3256, 0.0787, 0.0916, 0.4660, 0.1394], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0172, 0.0159, 0.0128, 0.0155, 0.0122, 0.0146, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 09:59:09,668 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.225e+02 1.659e+02 1.911e+02 2.326e+02 4.760e+02, threshold=3.822e+02, percent-clipped=1.0 2023-03-26 09:59:11,081 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3666, 2.2493, 1.7634, 2.4199, 2.3456, 1.9111, 2.8043, 2.3234], device='cuda:3'), covar=tensor([0.1424, 0.2778, 0.3526, 0.3061, 0.2819, 0.1913, 0.3697, 0.2094], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0190, 0.0235, 0.0255, 0.0237, 0.0194, 0.0212, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 09:59:18,770 INFO [finetune.py:976] (3/7) Epoch 8, batch 5250, loss[loss=0.2103, simple_loss=0.2739, pruned_loss=0.0733, over 4903.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.267, pruned_loss=0.06989, over 954131.06 frames. ], batch size: 37, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 09:59:27,675 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45348.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 09:59:29,880 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.68 vs. limit=5.0 2023-03-26 09:59:52,491 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8704, 3.8312, 3.7579, 2.0015, 3.9471, 2.8222, 0.7032, 2.7929], device='cuda:3'), covar=tensor([0.2335, 0.1610, 0.1346, 0.2882, 0.0884, 0.0999, 0.4387, 0.1255], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0172, 0.0160, 0.0128, 0.0156, 0.0122, 0.0146, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 09:59:56,799 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-03-26 10:00:03,245 INFO [finetune.py:976] (3/7) Epoch 8, batch 5300, loss[loss=0.2248, simple_loss=0.288, pruned_loss=0.08078, over 4742.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.268, pruned_loss=0.07024, over 953247.93 frames. ], batch size: 54, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 10:00:11,405 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6517, 0.6619, 1.4875, 1.3997, 1.3406, 1.2506, 1.2823, 1.4871], device='cuda:3'), covar=tensor([0.4787, 0.5591, 0.5280, 0.5046, 0.6314, 0.4923, 0.6453, 0.4870], device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0240, 0.0253, 0.0254, 0.0246, 0.0222, 0.0272, 0.0227], 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') 2023-03-26 10:00:16,218 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45409.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 10:00:16,342 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-26 10:00:18,468 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45412.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 10:00:30,690 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.820e+01 1.516e+02 1.858e+02 2.399e+02 4.469e+02, threshold=3.716e+02, percent-clipped=2.0 2023-03-26 10:00:39,966 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45437.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:00:49,408 INFO [finetune.py:976] (3/7) Epoch 8, batch 5350, loss[loss=0.1919, simple_loss=0.2588, pruned_loss=0.06252, over 4727.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.268, pruned_loss=0.06983, over 954123.72 frames. ], batch size: 54, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 10:00:53,044 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1499, 2.2109, 2.1397, 1.6191, 2.2155, 2.2451, 2.1899, 1.9235], device='cuda:3'), covar=tensor([0.0634, 0.0596, 0.0850, 0.0913, 0.0541, 0.0896, 0.0764, 0.1073], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0133, 0.0145, 0.0125, 0.0116, 0.0144, 0.0145, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 10:01:54,084 INFO [finetune.py:976] (3/7) Epoch 8, batch 5400, loss[loss=0.178, simple_loss=0.2414, pruned_loss=0.05726, over 4779.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2656, pruned_loss=0.06925, over 955491.64 frames. ], batch size: 26, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 10:02:34,314 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.106e+02 1.640e+02 2.019e+02 2.439e+02 4.086e+02, threshold=4.037e+02, percent-clipped=1.0 2023-03-26 10:02:54,208 INFO [finetune.py:976] (3/7) Epoch 8, batch 5450, loss[loss=0.1925, simple_loss=0.2589, pruned_loss=0.06304, over 4870.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2629, pruned_loss=0.06831, over 956619.72 frames. ], batch size: 31, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 10:02:56,734 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9170, 3.9863, 3.7620, 2.0411, 4.0927, 3.0232, 0.8402, 2.9459], device='cuda:3'), covar=tensor([0.2415, 0.1978, 0.1572, 0.3327, 0.0992, 0.1030, 0.4820, 0.1437], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0173, 0.0160, 0.0129, 0.0156, 0.0122, 0.0146, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 10:03:54,238 INFO [finetune.py:976] (3/7) Epoch 8, batch 5500, loss[loss=0.2036, simple_loss=0.2612, pruned_loss=0.07298, over 4831.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2615, pruned_loss=0.06847, over 957976.29 frames. ], batch size: 30, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 10:04:18,373 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.004e+02 1.550e+02 1.829e+02 2.210e+02 3.729e+02, threshold=3.658e+02, percent-clipped=0.0 2023-03-26 10:04:28,410 INFO [finetune.py:976] (3/7) Epoch 8, batch 5550, loss[loss=0.1851, simple_loss=0.2462, pruned_loss=0.06196, over 4887.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2641, pruned_loss=0.06968, over 956401.01 frames. ], batch size: 32, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 10:04:46,981 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2052, 1.9026, 1.9290, 0.8777, 2.1389, 2.4226, 1.9989, 1.8491], device='cuda:3'), covar=tensor([0.0871, 0.0709, 0.0542, 0.0771, 0.0437, 0.0578, 0.0517, 0.0748], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0158, 0.0122, 0.0137, 0.0134, 0.0126, 0.0148, 0.0149], device='cuda:3'), out_proj_covar=tensor([9.7086e-05, 1.1661e-04, 8.8059e-05, 9.9754e-05, 9.5651e-05, 9.2528e-05, 1.0867e-04, 1.0990e-04], device='cuda:3') 2023-03-26 10:05:19,427 INFO [finetune.py:976] (3/7) Epoch 8, batch 5600, loss[loss=0.2378, simple_loss=0.2975, pruned_loss=0.08904, over 4764.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2683, pruned_loss=0.07105, over 956007.54 frames. ], batch size: 54, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 10:05:24,101 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45702.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:05:25,271 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45704.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 10:05:29,919 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45712.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 10:05:36,358 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5691, 1.1126, 0.9329, 1.5910, 1.9721, 1.2304, 1.3868, 1.4688], device='cuda:3'), covar=tensor([0.1665, 0.2302, 0.2040, 0.1184, 0.2110, 0.2069, 0.1564, 0.2042], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0097, 0.0114, 0.0092, 0.0123, 0.0096, 0.0100, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 10:05:40,353 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.219e+02 1.739e+02 1.993e+02 2.505e+02 5.014e+02, threshold=3.987e+02, percent-clipped=3.0 2023-03-26 10:05:44,493 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45737.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:05:48,900 INFO [finetune.py:976] (3/7) Epoch 8, batch 5650, loss[loss=0.2278, simple_loss=0.2976, pruned_loss=0.07902, over 4816.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2703, pruned_loss=0.07122, over 955210.97 frames. ], batch size: 38, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 10:05:58,591 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=45760.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:06:00,363 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45763.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:06:06,754 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45774.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:06:13,161 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=45785.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:06:18,811 INFO [finetune.py:976] (3/7) Epoch 8, batch 5700, loss[loss=0.2276, simple_loss=0.2736, pruned_loss=0.09081, over 4418.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2673, pruned_loss=0.07154, over 939133.65 frames. ], batch size: 19, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 10:06:26,780 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-26 10:06:54,867 INFO [finetune.py:976] (3/7) Epoch 9, batch 0, loss[loss=0.2249, simple_loss=0.285, pruned_loss=0.08234, over 4841.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.285, pruned_loss=0.08234, over 4841.00 frames. ], batch size: 49, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 10:06:54,867 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 10:07:00,913 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6736, 1.4527, 2.0298, 2.8213, 1.9679, 2.2774, 0.9823, 2.2449], device='cuda:3'), covar=tensor([0.1772, 0.1572, 0.1184, 0.0674, 0.0934, 0.1197, 0.1815, 0.0679], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0118, 0.0135, 0.0166, 0.0103, 0.0140, 0.0127, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 10:07:12,131 INFO [finetune.py:1010] (3/7) Epoch 9, validation: loss=0.1616, simple_loss=0.233, pruned_loss=0.04515, over 2265189.00 frames. 2023-03-26 10:07:12,132 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 10:07:17,601 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.132e+01 1.600e+02 1.914e+02 2.307e+02 4.538e+02, threshold=3.829e+02, percent-clipped=2.0 2023-03-26 10:07:22,773 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45835.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:07:46,513 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45859.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:07:55,658 INFO [finetune.py:976] (3/7) Epoch 9, batch 50, loss[loss=0.1797, simple_loss=0.2567, pruned_loss=0.05134, over 4919.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2694, pruned_loss=0.07232, over 215138.65 frames. ], batch size: 46, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 10:07:58,008 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4903, 2.0917, 2.8649, 4.4258, 3.1168, 2.9211, 1.5162, 3.6034], device='cuda:3'), covar=tensor([0.1567, 0.1473, 0.1277, 0.0545, 0.0728, 0.1683, 0.1556, 0.0465], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0118, 0.0134, 0.0166, 0.0103, 0.0139, 0.0127, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 10:08:03,838 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-26 10:08:05,116 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 2023-03-26 10:08:23,329 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.85 vs. limit=5.0 2023-03-26 10:08:35,696 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8231, 1.3584, 0.8521, 1.6929, 2.1556, 1.3836, 1.5183, 1.6717], device='cuda:3'), covar=tensor([0.1506, 0.2034, 0.2114, 0.1154, 0.1967, 0.2012, 0.1506, 0.2013], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0097, 0.0114, 0.0092, 0.0123, 0.0096, 0.0100, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 10:08:35,727 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45920.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:08:36,814 INFO [finetune.py:976] (3/7) Epoch 9, batch 100, loss[loss=0.2097, simple_loss=0.2596, pruned_loss=0.07988, over 4692.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2612, pruned_loss=0.06845, over 379986.93 frames. ], batch size: 59, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 10:08:42,569 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.088e+02 1.760e+02 2.008e+02 2.423e+02 3.807e+02, threshold=4.016e+02, percent-clipped=0.0 2023-03-26 10:09:10,374 INFO [finetune.py:976] (3/7) Epoch 9, batch 150, loss[loss=0.1723, simple_loss=0.2412, pruned_loss=0.05164, over 4771.00 frames. ], tot_loss[loss=0.1953, simple_loss=0.2572, pruned_loss=0.06672, over 508648.57 frames. ], batch size: 28, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 10:09:32,062 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46004.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 10:09:49,162 INFO [finetune.py:976] (3/7) Epoch 9, batch 200, loss[loss=0.166, simple_loss=0.2389, pruned_loss=0.04658, over 4853.00 frames. ], tot_loss[loss=0.1947, simple_loss=0.2561, pruned_loss=0.0666, over 606448.31 frames. ], batch size: 49, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 10:09:58,660 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.086e+02 1.678e+02 2.056e+02 2.461e+02 4.455e+02, threshold=4.113e+02, percent-clipped=4.0 2023-03-26 10:10:22,925 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=46052.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 10:10:26,560 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46058.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:10:36,486 INFO [finetune.py:976] (3/7) Epoch 9, batch 250, loss[loss=0.2025, simple_loss=0.2746, pruned_loss=0.06523, over 4705.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2611, pruned_loss=0.06814, over 684725.97 frames. ], batch size: 59, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 10:10:52,271 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.97 vs. limit=5.0 2023-03-26 10:10:55,076 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-03-26 10:10:56,475 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-26 10:11:03,440 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-03-26 10:11:08,967 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-03-26 10:11:09,157 INFO [finetune.py:976] (3/7) Epoch 9, batch 300, loss[loss=0.2513, simple_loss=0.3016, pruned_loss=0.1005, over 4839.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.265, pruned_loss=0.06932, over 744306.20 frames. ], batch size: 44, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 10:11:14,967 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.304e+02 1.737e+02 2.021e+02 2.354e+02 3.684e+02, threshold=4.042e+02, percent-clipped=0.0 2023-03-26 10:11:15,051 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46130.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:11:30,617 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6142, 1.5039, 1.8885, 1.9254, 1.6931, 3.0906, 1.3563, 1.6377], device='cuda:3'), covar=tensor([0.1109, 0.2049, 0.1652, 0.1103, 0.1671, 0.0378, 0.1894, 0.2089], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0082, 0.0076, 0.0079, 0.0093, 0.0084, 0.0086, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 10:11:33,609 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6035, 1.6386, 1.4173, 1.8028, 1.9590, 1.7427, 1.2192, 1.3774], device='cuda:3'), covar=tensor([0.2192, 0.1990, 0.1877, 0.1556, 0.1758, 0.1173, 0.2526, 0.1965], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0207, 0.0204, 0.0186, 0.0239, 0.0177, 0.0212, 0.0194], 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') 2023-03-26 10:11:41,958 INFO [finetune.py:976] (3/7) Epoch 9, batch 350, loss[loss=0.2364, simple_loss=0.3007, pruned_loss=0.08605, over 4925.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2672, pruned_loss=0.07018, over 789636.30 frames. ], batch size: 33, lr: 3.81e-03, grad_scale: 16.0 2023-03-26 10:12:11,368 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46215.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:12:17,563 INFO [finetune.py:976] (3/7) Epoch 9, batch 400, loss[loss=0.2094, simple_loss=0.2841, pruned_loss=0.06738, over 4887.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2693, pruned_loss=0.07041, over 827886.48 frames. ], batch size: 43, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:12:23,439 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.951e+01 1.676e+02 2.053e+02 2.418e+02 4.627e+02, threshold=4.106e+02, percent-clipped=2.0 2023-03-26 10:13:00,669 INFO [finetune.py:976] (3/7) Epoch 9, batch 450, loss[loss=0.1791, simple_loss=0.2463, pruned_loss=0.05597, over 4831.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2675, pruned_loss=0.06965, over 855980.38 frames. ], batch size: 30, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:13:03,233 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46276.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:13:35,968 INFO [finetune.py:976] (3/7) Epoch 9, batch 500, loss[loss=0.2537, simple_loss=0.2982, pruned_loss=0.1046, over 4863.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2655, pruned_loss=0.06936, over 880562.19 frames. ], batch size: 44, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:13:45,335 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.146e+02 1.586e+02 1.900e+02 2.408e+02 3.619e+02, threshold=3.799e+02, percent-clipped=0.0 2023-03-26 10:13:54,717 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46337.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:13:59,927 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-03-26 10:14:08,382 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46358.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:14:09,075 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-03-26 10:14:15,594 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.7597, 4.0908, 4.3608, 4.5563, 4.4904, 4.2415, 4.8125, 1.6012], device='cuda:3'), covar=tensor([0.0642, 0.0832, 0.0658, 0.0782, 0.1058, 0.1328, 0.0565, 0.5051], device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0243, 0.0276, 0.0292, 0.0332, 0.0281, 0.0301, 0.0294], 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') 2023-03-26 10:14:17,261 INFO [finetune.py:976] (3/7) Epoch 9, batch 550, loss[loss=0.2751, simple_loss=0.3205, pruned_loss=0.1148, over 4856.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2623, pruned_loss=0.06852, over 897756.18 frames. ], batch size: 44, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:14:39,978 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=46406.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:14:43,742 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1183, 1.9859, 2.0074, 0.8929, 2.2509, 2.5666, 2.0918, 1.8857], device='cuda:3'), covar=tensor([0.0956, 0.0747, 0.0557, 0.0777, 0.0568, 0.0568, 0.0554, 0.0691], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0156, 0.0121, 0.0136, 0.0132, 0.0125, 0.0146, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.6270e-05, 1.1497e-04, 8.7723e-05, 9.8639e-05, 9.4303e-05, 9.1793e-05, 1.0692e-04, 1.0814e-04], device='cuda:3') 2023-03-26 10:14:50,106 INFO [finetune.py:976] (3/7) Epoch 9, batch 600, loss[loss=0.1913, simple_loss=0.2601, pruned_loss=0.06125, over 4817.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2621, pruned_loss=0.06865, over 909263.19 frames. ], batch size: 51, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:14:54,846 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.149e+02 1.702e+02 1.969e+02 2.390e+02 4.680e+02, threshold=3.938e+02, percent-clipped=3.0 2023-03-26 10:14:54,941 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46430.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:15:08,738 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4049, 2.1379, 1.8459, 2.4010, 2.1969, 2.0749, 2.0505, 3.1692], device='cuda:3'), covar=tensor([0.5069, 0.6875, 0.4535, 0.5782, 0.5204, 0.3222, 0.6203, 0.1915], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0261, 0.0222, 0.0280, 0.0243, 0.0208, 0.0245, 0.0209], 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') 2023-03-26 10:15:36,213 INFO [finetune.py:976] (3/7) Epoch 9, batch 650, loss[loss=0.19, simple_loss=0.2493, pruned_loss=0.06536, over 4770.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2637, pruned_loss=0.06887, over 918433.86 frames. ], batch size: 26, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:15:40,463 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=46478.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:16:05,380 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46515.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:16:09,544 INFO [finetune.py:976] (3/7) Epoch 9, batch 700, loss[loss=0.3099, simple_loss=0.3541, pruned_loss=0.1329, over 4130.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2672, pruned_loss=0.0706, over 926345.39 frames. ], batch size: 65, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:16:14,888 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.046e+02 1.666e+02 2.010e+02 2.529e+02 4.289e+02, threshold=4.019e+02, percent-clipped=2.0 2023-03-26 10:16:28,029 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.4935, 1.5953, 1.6432, 0.9520, 1.6965, 1.8402, 1.8526, 1.4020], device='cuda:3'), covar=tensor([0.0897, 0.0563, 0.0466, 0.0577, 0.0448, 0.0667, 0.0324, 0.0729], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0157, 0.0122, 0.0137, 0.0133, 0.0127, 0.0147, 0.0148], device='cuda:3'), out_proj_covar=tensor([9.7012e-05, 1.1590e-04, 8.8219e-05, 9.9374e-05, 9.5214e-05, 9.2810e-05, 1.0779e-04, 1.0924e-04], device='cuda:3') 2023-03-26 10:16:37,471 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=46563.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:16:42,850 INFO [finetune.py:976] (3/7) Epoch 9, batch 750, loss[loss=0.2067, simple_loss=0.2785, pruned_loss=0.06749, over 4875.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2679, pruned_loss=0.07012, over 933961.86 frames. ], batch size: 43, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:16:58,829 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46595.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:17:03,707 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46602.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:17:15,994 INFO [finetune.py:976] (3/7) Epoch 9, batch 800, loss[loss=0.1946, simple_loss=0.2429, pruned_loss=0.07312, over 4466.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2671, pruned_loss=0.06956, over 937495.05 frames. ], batch size: 19, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:17:20,824 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.216e+01 1.561e+02 1.863e+02 2.153e+02 3.377e+02, threshold=3.726e+02, percent-clipped=0.0 2023-03-26 10:17:22,504 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46632.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:17:24,436 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6615, 0.6674, 1.6199, 1.5293, 1.4598, 1.3878, 1.4608, 1.5062], device='cuda:3'), covar=tensor([0.4149, 0.5022, 0.4221, 0.4387, 0.5310, 0.4083, 0.5171, 0.4031], device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0242, 0.0255, 0.0256, 0.0248, 0.0224, 0.0274, 0.0230], 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') 2023-03-26 10:17:39,094 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46656.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 10:17:43,752 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46663.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:17:49,096 INFO [finetune.py:976] (3/7) Epoch 9, batch 850, loss[loss=0.1949, simple_loss=0.2581, pruned_loss=0.06584, over 4161.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2648, pruned_loss=0.0684, over 942958.18 frames. ], batch size: 65, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:17:57,646 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8984, 1.3600, 1.7098, 1.7177, 1.5819, 1.5475, 1.6903, 1.6361], device='cuda:3'), covar=tensor([0.5892, 0.6684, 0.6123, 0.6298, 0.7853, 0.5962, 0.7924, 0.5878], device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0242, 0.0255, 0.0256, 0.0248, 0.0224, 0.0274, 0.0230], 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') 2023-03-26 10:18:34,822 INFO [finetune.py:976] (3/7) Epoch 9, batch 900, loss[loss=0.229, simple_loss=0.2905, pruned_loss=0.08378, over 4839.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2624, pruned_loss=0.06753, over 946825.66 frames. ], batch size: 47, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:18:39,658 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.365e+01 1.501e+02 1.768e+02 2.130e+02 3.855e+02, threshold=3.537e+02, percent-clipped=1.0 2023-03-26 10:18:52,978 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-03-26 10:19:02,910 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5959, 1.2213, 1.1834, 1.1842, 1.6786, 1.7434, 1.6350, 1.2362], device='cuda:3'), covar=tensor([0.0277, 0.0354, 0.0724, 0.0415, 0.0224, 0.0335, 0.0250, 0.0431], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0109, 0.0138, 0.0114, 0.0101, 0.0100, 0.0090, 0.0107], device='cuda:3'), out_proj_covar=tensor([6.9232e-05, 8.5075e-05, 1.1063e-04, 8.9873e-05, 7.9694e-05, 7.3931e-05, 6.8230e-05, 8.3001e-05], device='cuda:3') 2023-03-26 10:19:10,157 INFO [finetune.py:976] (3/7) Epoch 9, batch 950, loss[loss=0.1938, simple_loss=0.2548, pruned_loss=0.06646, over 4933.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2608, pruned_loss=0.06715, over 949859.24 frames. ], batch size: 33, lr: 3.80e-03, grad_scale: 32.0 2023-03-26 10:19:15,160 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1542, 1.2599, 1.6097, 0.9557, 1.1651, 1.3907, 1.2561, 1.5860], device='cuda:3'), covar=tensor([0.1275, 0.2180, 0.1303, 0.1728, 0.1101, 0.1329, 0.2900, 0.0844], device='cuda:3'), in_proj_covar=tensor([0.0202, 0.0206, 0.0198, 0.0195, 0.0183, 0.0221, 0.0221, 0.0204], 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') 2023-03-26 10:19:18,789 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4388, 1.3590, 1.5415, 2.4630, 1.6980, 2.0987, 0.8948, 2.0584], device='cuda:3'), covar=tensor([0.1731, 0.1508, 0.1201, 0.0742, 0.0905, 0.1084, 0.1548, 0.0680], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0117, 0.0133, 0.0165, 0.0102, 0.0138, 0.0126, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 10:19:21,157 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6690, 3.6882, 3.5596, 1.6086, 3.7671, 2.8110, 0.5729, 2.5167], device='cuda:3'), covar=tensor([0.2323, 0.2000, 0.1533, 0.3696, 0.1065, 0.1019, 0.4949, 0.1529], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0172, 0.0159, 0.0128, 0.0156, 0.0121, 0.0145, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 10:19:36,038 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46810.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:19:36,727 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.85 vs. limit=5.0 2023-03-26 10:19:44,266 INFO [finetune.py:976] (3/7) Epoch 9, batch 1000, loss[loss=0.1684, simple_loss=0.2365, pruned_loss=0.05019, over 4882.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2641, pruned_loss=0.06846, over 951558.53 frames. ], batch size: 32, lr: 3.80e-03, grad_scale: 32.0 2023-03-26 10:19:49,060 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.128e+02 1.690e+02 1.992e+02 2.479e+02 5.334e+02, threshold=3.983e+02, percent-clipped=2.0 2023-03-26 10:20:18,983 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1573, 1.9750, 2.1426, 0.9871, 2.3222, 2.5974, 2.0822, 1.9082], device='cuda:3'), covar=tensor([0.1043, 0.0860, 0.0487, 0.0720, 0.0588, 0.0633, 0.0611, 0.0772], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0156, 0.0120, 0.0134, 0.0131, 0.0125, 0.0145, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.5822e-05, 1.1483e-04, 8.6664e-05, 9.7688e-05, 9.4010e-05, 9.1360e-05, 1.0672e-04, 1.0800e-04], device='cuda:3') 2023-03-26 10:20:22,584 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46871.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:20:23,062 INFO [finetune.py:976] (3/7) Epoch 9, batch 1050, loss[loss=0.1916, simple_loss=0.2549, pruned_loss=0.0642, over 4884.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.267, pruned_loss=0.06929, over 953339.68 frames. ], batch size: 32, lr: 3.80e-03, grad_scale: 32.0 2023-03-26 10:21:05,008 INFO [finetune.py:976] (3/7) Epoch 9, batch 1100, loss[loss=0.1883, simple_loss=0.2681, pruned_loss=0.05422, over 4924.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2681, pruned_loss=0.06957, over 953987.83 frames. ], batch size: 33, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:21:10,490 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.274e+02 1.654e+02 2.044e+02 2.534e+02 3.814e+02, threshold=4.089e+02, percent-clipped=0.0 2023-03-26 10:21:11,185 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46932.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:21:16,132 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-03-26 10:21:23,032 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46951.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 10:21:28,257 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46958.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:21:30,603 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1206, 3.5269, 3.7591, 3.9641, 3.8819, 3.6382, 4.1975, 1.2518], device='cuda:3'), covar=tensor([0.0757, 0.0798, 0.0825, 0.1072, 0.1173, 0.1509, 0.0710, 0.5412], device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0242, 0.0276, 0.0293, 0.0331, 0.0281, 0.0300, 0.0293], 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') 2023-03-26 10:21:37,702 INFO [finetune.py:976] (3/7) Epoch 9, batch 1150, loss[loss=0.1893, simple_loss=0.2631, pruned_loss=0.05771, over 4889.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2692, pruned_loss=0.07, over 955497.65 frames. ], batch size: 35, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:21:42,583 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=46980.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:22:02,620 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-26 10:22:08,474 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0771, 1.9710, 1.6138, 1.8923, 1.8339, 1.8002, 1.8353, 2.6198], device='cuda:3'), covar=tensor([0.5119, 0.5517, 0.4176, 0.5118, 0.4926, 0.2997, 0.4791, 0.2021], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0260, 0.0222, 0.0280, 0.0242, 0.0208, 0.0244, 0.0208], 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') 2023-03-26 10:22:10,618 INFO [finetune.py:976] (3/7) Epoch 9, batch 1200, loss[loss=0.2136, simple_loss=0.2724, pruned_loss=0.07745, over 4923.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2676, pruned_loss=0.06974, over 955828.87 frames. ], batch size: 33, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:22:16,140 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.814e+01 1.625e+02 2.018e+02 2.406e+02 3.989e+02, threshold=4.036e+02, percent-clipped=0.0 2023-03-26 10:22:43,588 INFO [finetune.py:976] (3/7) Epoch 9, batch 1250, loss[loss=0.1912, simple_loss=0.2612, pruned_loss=0.06056, over 4860.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2645, pruned_loss=0.06865, over 955457.65 frames. ], batch size: 34, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:22:56,205 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7845, 1.2542, 0.7723, 1.6911, 2.1060, 1.4112, 1.6473, 1.5683], device='cuda:3'), covar=tensor([0.1488, 0.2158, 0.2262, 0.1199, 0.2035, 0.2069, 0.1439, 0.2117], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0096, 0.0113, 0.0092, 0.0122, 0.0096, 0.0100, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 10:22:58,667 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1217, 1.8624, 1.5949, 1.8110, 1.8298, 1.7343, 1.7568, 2.5843], device='cuda:3'), covar=tensor([0.4466, 0.4826, 0.3903, 0.4577, 0.4542, 0.2686, 0.4707, 0.1957], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0260, 0.0223, 0.0280, 0.0243, 0.0209, 0.0245, 0.0208], 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') 2023-03-26 10:23:21,437 INFO [finetune.py:976] (3/7) Epoch 9, batch 1300, loss[loss=0.2077, simple_loss=0.2673, pruned_loss=0.074, over 4902.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.261, pruned_loss=0.06744, over 956587.17 frames. ], batch size: 37, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:23:31,783 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.169e+01 1.658e+02 1.956e+02 2.404e+02 5.414e+02, threshold=3.912e+02, percent-clipped=2.0 2023-03-26 10:23:57,908 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47166.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:24:03,031 INFO [finetune.py:976] (3/7) Epoch 9, batch 1350, loss[loss=0.2187, simple_loss=0.2984, pruned_loss=0.06946, over 4725.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2613, pruned_loss=0.06724, over 958225.48 frames. ], batch size: 59, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:24:36,401 INFO [finetune.py:976] (3/7) Epoch 9, batch 1400, loss[loss=0.2024, simple_loss=0.2644, pruned_loss=0.07024, over 4765.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2647, pruned_loss=0.06813, over 958303.53 frames. ], batch size: 26, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:24:42,788 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.151e+02 1.739e+02 2.002e+02 2.347e+02 4.228e+02, threshold=4.005e+02, percent-clipped=2.0 2023-03-26 10:24:43,507 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9397, 1.3595, 0.9957, 1.8670, 2.2236, 1.5759, 1.7653, 1.6029], device='cuda:3'), covar=tensor([0.1614, 0.2316, 0.2181, 0.1241, 0.2017, 0.2044, 0.1595, 0.2372], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0096, 0.0113, 0.0092, 0.0122, 0.0095, 0.0100, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 10:24:49,015 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47241.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:24:55,130 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47251.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:24:55,143 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7877, 1.4654, 2.2057, 1.4049, 1.9834, 2.0692, 1.5293, 2.2332], device='cuda:3'), covar=tensor([0.1233, 0.2362, 0.1388, 0.1936, 0.0911, 0.1389, 0.2827, 0.0796], device='cuda:3'), in_proj_covar=tensor([0.0198, 0.0204, 0.0196, 0.0193, 0.0180, 0.0217, 0.0217, 0.0202], 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') 2023-03-26 10:24:55,229 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-03-26 10:24:59,394 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47258.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:25:09,255 INFO [finetune.py:976] (3/7) Epoch 9, batch 1450, loss[loss=0.2197, simple_loss=0.2803, pruned_loss=0.07958, over 4897.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2669, pruned_loss=0.0691, over 957648.47 frames. ], batch size: 37, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:25:27,568 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=47299.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:25:29,415 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47302.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:25:36,207 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=47306.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:25:46,810 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-03-26 10:25:53,871 INFO [finetune.py:976] (3/7) Epoch 9, batch 1500, loss[loss=0.2226, simple_loss=0.2968, pruned_loss=0.07423, over 4882.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2695, pruned_loss=0.07054, over 956086.56 frames. ], batch size: 35, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:26:00,803 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.103e+02 1.648e+02 1.983e+02 2.305e+02 4.092e+02, threshold=3.967e+02, percent-clipped=1.0 2023-03-26 10:26:25,129 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4254, 1.2951, 1.3603, 1.3362, 0.8396, 2.1986, 0.7492, 1.2750], device='cuda:3'), covar=tensor([0.3328, 0.2536, 0.2090, 0.2406, 0.2001, 0.0339, 0.2889, 0.1402], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0115, 0.0119, 0.0123, 0.0116, 0.0098, 0.0100, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 10:26:26,878 INFO [finetune.py:976] (3/7) Epoch 9, batch 1550, loss[loss=0.2616, simple_loss=0.3116, pruned_loss=0.1058, over 4884.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2699, pruned_loss=0.07094, over 954534.90 frames. ], batch size: 35, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:26:33,204 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47379.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:26:50,621 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-26 10:26:51,165 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1134, 2.1683, 2.1403, 1.5394, 2.2096, 2.3166, 2.2709, 1.8838], device='cuda:3'), covar=tensor([0.0630, 0.0580, 0.0777, 0.0878, 0.0604, 0.0732, 0.0617, 0.0991], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0132, 0.0145, 0.0125, 0.0117, 0.0145, 0.0145, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 10:26:51,592 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2023-03-26 10:27:00,841 INFO [finetune.py:976] (3/7) Epoch 9, batch 1600, loss[loss=0.1921, simple_loss=0.2532, pruned_loss=0.06549, over 4853.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2676, pruned_loss=0.07013, over 955525.27 frames. ], batch size: 44, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:27:00,981 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2071, 1.9923, 1.6304, 2.2852, 2.0951, 1.8576, 2.5539, 2.1629], device='cuda:3'), covar=tensor([0.1363, 0.2587, 0.3414, 0.2654, 0.2631, 0.1709, 0.3356, 0.1888], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0188, 0.0232, 0.0253, 0.0236, 0.0193, 0.0211, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 10:27:03,007 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.25 vs. limit=5.0 2023-03-26 10:27:07,814 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.155e+02 1.632e+02 1.991e+02 2.321e+02 5.028e+02, threshold=3.982e+02, percent-clipped=1.0 2023-03-26 10:27:14,447 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47440.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:27:16,135 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4501, 1.4245, 1.3408, 1.4786, 1.7985, 1.6660, 1.4824, 1.2598], device='cuda:3'), covar=tensor([0.0312, 0.0246, 0.0522, 0.0243, 0.0174, 0.0379, 0.0292, 0.0330], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0109, 0.0140, 0.0115, 0.0102, 0.0101, 0.0091, 0.0108], device='cuda:3'), out_proj_covar=tensor([6.9975e-05, 8.5556e-05, 1.1147e-04, 9.0154e-05, 8.0428e-05, 7.4777e-05, 6.9245e-05, 8.3328e-05], device='cuda:3') 2023-03-26 10:27:30,111 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7226, 1.5379, 2.3242, 3.6117, 2.3498, 2.4130, 0.9642, 2.8396], device='cuda:3'), covar=tensor([0.1799, 0.1621, 0.1378, 0.0629, 0.0900, 0.1439, 0.2194, 0.0605], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0118, 0.0135, 0.0166, 0.0103, 0.0139, 0.0127, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 10:27:30,702 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47466.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:27:34,229 INFO [finetune.py:976] (3/7) Epoch 9, batch 1650, loss[loss=0.1389, simple_loss=0.2123, pruned_loss=0.03278, over 4822.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2634, pruned_loss=0.06827, over 953847.62 frames. ], batch size: 39, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:28:02,970 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=47514.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:28:07,771 INFO [finetune.py:976] (3/7) Epoch 9, batch 1700, loss[loss=0.1672, simple_loss=0.2324, pruned_loss=0.05104, over 4901.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2614, pruned_loss=0.06789, over 955801.98 frames. ], batch size: 35, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:28:13,223 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.017e+02 1.617e+02 1.864e+02 2.209e+02 5.015e+02, threshold=3.728e+02, percent-clipped=2.0 2023-03-26 10:28:36,331 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6882, 1.5060, 2.2343, 3.0738, 2.1229, 2.2486, 1.1847, 2.3608], device='cuda:3'), covar=tensor([0.1495, 0.1401, 0.1053, 0.0524, 0.0781, 0.2005, 0.1580, 0.0626], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0117, 0.0134, 0.0165, 0.0102, 0.0138, 0.0126, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 10:28:47,728 INFO [finetune.py:976] (3/7) Epoch 9, batch 1750, loss[loss=0.2251, simple_loss=0.303, pruned_loss=0.07359, over 4903.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2646, pruned_loss=0.06886, over 956293.06 frames. ], batch size: 43, lr: 3.80e-03, grad_scale: 16.0 2023-03-26 10:29:13,719 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47597.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:29:29,377 INFO [finetune.py:976] (3/7) Epoch 9, batch 1800, loss[loss=0.1726, simple_loss=0.254, pruned_loss=0.04563, over 4775.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2687, pruned_loss=0.07037, over 957096.60 frames. ], batch size: 29, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:29:30,780 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-03-26 10:29:34,865 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.172e+02 1.736e+02 1.994e+02 2.573e+02 6.193e+02, threshold=3.989e+02, percent-clipped=4.0 2023-03-26 10:30:02,651 INFO [finetune.py:976] (3/7) Epoch 9, batch 1850, loss[loss=0.2198, simple_loss=0.2846, pruned_loss=0.07748, over 4749.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2699, pruned_loss=0.07081, over 955182.83 frames. ], batch size: 27, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:30:02,784 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1217, 1.9468, 1.6275, 2.0265, 2.0698, 1.7473, 2.4048, 2.1679], device='cuda:3'), covar=tensor([0.1452, 0.2774, 0.3613, 0.3005, 0.2963, 0.1840, 0.3946, 0.1926], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0188, 0.0233, 0.0254, 0.0237, 0.0194, 0.0211, 0.0193], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 10:30:35,896 INFO [finetune.py:976] (3/7) Epoch 9, batch 1900, loss[loss=0.2378, simple_loss=0.2926, pruned_loss=0.09153, over 4881.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2721, pruned_loss=0.07157, over 956899.35 frames. ], batch size: 35, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:30:36,604 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0539, 1.5313, 2.3971, 3.9962, 2.8349, 2.6644, 0.7799, 3.2759], device='cuda:3'), covar=tensor([0.1699, 0.1775, 0.1416, 0.0459, 0.0758, 0.1522, 0.2122, 0.0423], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0118, 0.0134, 0.0166, 0.0102, 0.0139, 0.0127, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 10:30:46,261 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.162e+02 1.608e+02 1.835e+02 2.236e+02 3.803e+02, threshold=3.670e+02, percent-clipped=0.0 2023-03-26 10:30:53,012 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47735.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:30:54,873 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47738.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:31:03,041 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.0741, 0.8944, 0.8627, 1.1937, 1.2461, 1.1961, 1.0157, 0.9375], device='cuda:3'), covar=tensor([0.0342, 0.0306, 0.0634, 0.0277, 0.0292, 0.0424, 0.0329, 0.0414], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0110, 0.0140, 0.0115, 0.0103, 0.0101, 0.0092, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.0224e-05, 8.5901e-05, 1.1143e-04, 9.0181e-05, 8.0732e-05, 7.5021e-05, 6.9318e-05, 8.3232e-05], device='cuda:3') 2023-03-26 10:31:21,760 INFO [finetune.py:976] (3/7) Epoch 9, batch 1950, loss[loss=0.2138, simple_loss=0.2741, pruned_loss=0.07675, over 4913.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2692, pruned_loss=0.07026, over 955250.47 frames. ], batch size: 37, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:31:31,517 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47788.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:31:39,137 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47799.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:31:46,709 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47809.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 10:31:55,008 INFO [finetune.py:976] (3/7) Epoch 9, batch 2000, loss[loss=0.2184, simple_loss=0.277, pruned_loss=0.0799, over 4914.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2651, pruned_loss=0.06862, over 954095.47 frames. ], batch size: 36, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:32:00,455 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.198e+02 1.527e+02 1.824e+02 2.186e+02 3.277e+02, threshold=3.648e+02, percent-clipped=0.0 2023-03-26 10:32:11,888 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47849.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:32:35,764 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47870.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 10:32:36,814 INFO [finetune.py:976] (3/7) Epoch 9, batch 2050, loss[loss=0.1492, simple_loss=0.2163, pruned_loss=0.04104, over 4791.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2611, pruned_loss=0.06689, over 954740.58 frames. ], batch size: 29, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:32:57,835 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47897.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:33:02,084 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-03-26 10:33:15,823 INFO [finetune.py:976] (3/7) Epoch 9, batch 2100, loss[loss=0.1686, simple_loss=0.2326, pruned_loss=0.05229, over 4782.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2603, pruned_loss=0.06642, over 953309.99 frames. ], batch size: 26, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:33:21,286 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.115e+02 1.604e+02 1.917e+02 2.370e+02 5.169e+02, threshold=3.834e+02, percent-clipped=3.0 2023-03-26 10:33:29,786 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=47945.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:33:54,984 INFO [finetune.py:976] (3/7) Epoch 9, batch 2150, loss[loss=0.2244, simple_loss=0.3, pruned_loss=0.0744, over 4801.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2651, pruned_loss=0.06862, over 953817.07 frames. ], batch size: 51, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:34:18,610 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 2023-03-26 10:34:36,693 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-26 10:35:00,856 INFO [finetune.py:976] (3/7) Epoch 9, batch 2200, loss[loss=0.1782, simple_loss=0.2503, pruned_loss=0.05306, over 4820.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2673, pruned_loss=0.06995, over 953819.63 frames. ], batch size: 40, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:35:10,205 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2602, 2.0139, 1.8502, 1.9059, 2.2100, 1.9846, 2.3551, 2.2199], device='cuda:3'), covar=tensor([0.1249, 0.2291, 0.3069, 0.2449, 0.2555, 0.1573, 0.2626, 0.1831], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0189, 0.0233, 0.0254, 0.0237, 0.0194, 0.0212, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 10:35:11,848 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.338e+02 1.754e+02 2.052e+02 2.528e+02 4.321e+02, threshold=4.105e+02, percent-clipped=1.0 2023-03-26 10:35:18,729 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48035.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:35:27,964 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-26 10:36:02,992 INFO [finetune.py:976] (3/7) Epoch 9, batch 2250, loss[loss=0.2767, simple_loss=0.3288, pruned_loss=0.1123, over 4135.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2687, pruned_loss=0.07083, over 952368.84 frames. ], batch size: 65, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:36:03,123 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48072.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:36:20,660 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=48083.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:36:32,294 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48094.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:36:54,938 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8717, 1.2413, 1.8576, 1.7022, 1.5535, 1.5409, 1.6576, 1.6890], device='cuda:3'), covar=tensor([0.5022, 0.6087, 0.4752, 0.5585, 0.6374, 0.4921, 0.6743, 0.4557], device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0241, 0.0255, 0.0257, 0.0250, 0.0225, 0.0274, 0.0230], 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') 2023-03-26 10:37:02,671 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7554, 4.0970, 3.8459, 2.1624, 4.1748, 3.1354, 1.0109, 3.0168], device='cuda:3'), covar=tensor([0.2200, 0.1510, 0.1422, 0.2999, 0.0856, 0.0838, 0.4398, 0.1242], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0172, 0.0159, 0.0128, 0.0155, 0.0122, 0.0147, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 10:37:03,937 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1513, 1.8462, 1.6906, 1.7640, 1.8357, 1.8443, 1.8092, 2.6370], device='cuda:3'), covar=tensor([0.4836, 0.5704, 0.4210, 0.5169, 0.4842, 0.2900, 0.5181, 0.1924], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0259, 0.0221, 0.0279, 0.0242, 0.0208, 0.0244, 0.0209], 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') 2023-03-26 10:37:05,453 INFO [finetune.py:976] (3/7) Epoch 9, batch 2300, loss[loss=0.1691, simple_loss=0.2337, pruned_loss=0.05219, over 4783.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2687, pruned_loss=0.07007, over 954099.71 frames. ], batch size: 25, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:37:16,651 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.508e+01 1.652e+02 1.874e+02 2.360e+02 5.580e+02, threshold=3.748e+02, percent-clipped=1.0 2023-03-26 10:37:23,193 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48133.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:37:34,215 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48144.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:37:54,847 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48165.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 10:38:01,017 INFO [finetune.py:976] (3/7) Epoch 9, batch 2350, loss[loss=0.1576, simple_loss=0.239, pruned_loss=0.03811, over 4910.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2658, pruned_loss=0.06918, over 955134.94 frames. ], batch size: 36, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:38:24,259 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6287, 1.4114, 1.3358, 1.6301, 1.6305, 1.6706, 0.9114, 1.3850], device='cuda:3'), covar=tensor([0.2265, 0.2198, 0.1945, 0.1657, 0.1586, 0.1223, 0.2856, 0.1890], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0208, 0.0205, 0.0187, 0.0241, 0.0179, 0.0214, 0.0194], 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') 2023-03-26 10:38:35,577 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48212.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 10:38:43,033 INFO [finetune.py:976] (3/7) Epoch 9, batch 2400, loss[loss=0.1615, simple_loss=0.2357, pruned_loss=0.04371, over 4904.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.263, pruned_loss=0.06889, over 953365.09 frames. ], batch size: 35, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:38:44,535 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-26 10:38:49,468 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.159e+02 1.602e+02 2.015e+02 2.421e+02 3.465e+02, threshold=4.031e+02, percent-clipped=0.0 2023-03-26 10:39:18,904 INFO [finetune.py:976] (3/7) Epoch 9, batch 2450, loss[loss=0.1987, simple_loss=0.2686, pruned_loss=0.06438, over 4826.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2609, pruned_loss=0.0683, over 953210.85 frames. ], batch size: 33, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:39:19,630 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48273.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 10:39:31,825 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.5983, 1.7018, 1.6595, 1.0067, 1.7491, 1.9742, 1.9267, 1.4590], device='cuda:3'), covar=tensor([0.0954, 0.0565, 0.0503, 0.0534, 0.0407, 0.0528, 0.0345, 0.0685], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0156, 0.0121, 0.0135, 0.0132, 0.0126, 0.0146, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.6064e-05, 1.1486e-04, 8.7502e-05, 9.8045e-05, 9.4115e-05, 9.2386e-05, 1.0737e-04, 1.0829e-04], device='cuda:3') 2023-03-26 10:39:44,215 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-26 10:39:54,553 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9912, 1.9590, 1.6741, 2.0402, 1.9132, 1.8685, 1.8796, 2.6203], device='cuda:3'), covar=tensor([0.5092, 0.6759, 0.4394, 0.6206, 0.5919, 0.3028, 0.6022, 0.2220], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0260, 0.0222, 0.0280, 0.0243, 0.0209, 0.0245, 0.0210], 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') 2023-03-26 10:40:01,692 INFO [finetune.py:976] (3/7) Epoch 9, batch 2500, loss[loss=0.1978, simple_loss=0.2697, pruned_loss=0.06292, over 4859.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2631, pruned_loss=0.06938, over 954082.19 frames. ], batch size: 31, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:40:03,383 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48324.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:40:09,017 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.093e+02 1.681e+02 1.962e+02 2.420e+02 5.026e+02, threshold=3.923e+02, percent-clipped=2.0 2023-03-26 10:40:35,357 INFO [finetune.py:976] (3/7) Epoch 9, batch 2550, loss[loss=0.2043, simple_loss=0.2707, pruned_loss=0.06894, over 4925.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2665, pruned_loss=0.06956, over 954858.14 frames. ], batch size: 38, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:40:45,883 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48385.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:40:51,850 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48394.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:40:52,502 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6056, 1.7714, 1.5555, 1.9860, 2.1405, 1.8838, 1.5414, 1.3714], device='cuda:3'), covar=tensor([0.2597, 0.2092, 0.1955, 0.1721, 0.2104, 0.1341, 0.2780, 0.2124], device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0207, 0.0204, 0.0186, 0.0239, 0.0178, 0.0212, 0.0193], 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') 2023-03-26 10:40:53,651 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48397.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:41:08,895 INFO [finetune.py:976] (3/7) Epoch 9, batch 2600, loss[loss=0.2233, simple_loss=0.287, pruned_loss=0.07979, over 4725.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2675, pruned_loss=0.06974, over 953074.80 frames. ], batch size: 59, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:41:09,605 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48423.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:41:12,599 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48428.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:41:15,191 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.166e+02 1.754e+02 2.041e+02 2.553e+02 4.015e+02, threshold=4.083e+02, percent-clipped=1.0 2023-03-26 10:41:23,747 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=48442.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:41:25,437 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48444.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:41:31,561 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9384, 4.2000, 3.9160, 1.9280, 4.2747, 3.2093, 0.7216, 2.9103], device='cuda:3'), covar=tensor([0.2061, 0.1705, 0.1335, 0.3060, 0.0908, 0.0937, 0.4448, 0.1308], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0173, 0.0159, 0.0128, 0.0156, 0.0122, 0.0147, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 10:41:34,029 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48458.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:41:35,243 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9308, 1.7110, 1.5013, 1.5466, 1.6748, 1.6291, 1.6680, 2.3548], device='cuda:3'), covar=tensor([0.4658, 0.5451, 0.3913, 0.4749, 0.4882, 0.2929, 0.4733, 0.2012], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0259, 0.0221, 0.0279, 0.0242, 0.0208, 0.0245, 0.0209], 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') 2023-03-26 10:41:37,656 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3641, 1.3527, 1.4534, 0.7220, 1.5208, 1.4948, 1.4012, 1.2684], device='cuda:3'), covar=tensor([0.0610, 0.0715, 0.0692, 0.0966, 0.0722, 0.0744, 0.0643, 0.1101], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0132, 0.0144, 0.0124, 0.0117, 0.0144, 0.0144, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 10:41:38,227 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48465.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 10:41:42,381 INFO [finetune.py:976] (3/7) Epoch 9, batch 2650, loss[loss=0.1867, simple_loss=0.2518, pruned_loss=0.06081, over 4860.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2689, pruned_loss=0.07022, over 955899.75 frames. ], batch size: 34, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:41:56,005 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48484.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:42:06,173 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=48492.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:42:19,479 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=48513.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 10:42:22,610 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0378, 1.7812, 1.6486, 1.8357, 1.7600, 1.7681, 1.7690, 2.4700], device='cuda:3'), covar=tensor([0.4738, 0.6026, 0.3893, 0.4997, 0.5012, 0.2926, 0.5079, 0.1957], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0260, 0.0222, 0.0279, 0.0243, 0.0209, 0.0245, 0.0209], 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') 2023-03-26 10:42:24,896 INFO [finetune.py:976] (3/7) Epoch 9, batch 2700, loss[loss=0.1518, simple_loss=0.2222, pruned_loss=0.04071, over 4812.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2673, pruned_loss=0.06904, over 956031.56 frames. ], batch size: 40, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:42:30,333 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.052e+02 1.590e+02 1.873e+02 2.487e+02 4.580e+02, threshold=3.745e+02, percent-clipped=2.0 2023-03-26 10:43:07,969 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48568.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 10:43:10,318 INFO [finetune.py:976] (3/7) Epoch 9, batch 2750, loss[loss=0.1312, simple_loss=0.2073, pruned_loss=0.02749, over 4748.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2633, pruned_loss=0.0678, over 953052.92 frames. ], batch size: 26, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:43:10,475 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9615, 1.8040, 1.5148, 1.6611, 1.6845, 1.6551, 1.6627, 2.4126], device='cuda:3'), covar=tensor([0.4752, 0.5234, 0.4103, 0.5000, 0.4848, 0.2970, 0.4915, 0.2011], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0260, 0.0222, 0.0279, 0.0242, 0.0209, 0.0245, 0.0209], 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') 2023-03-26 10:43:24,006 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5898, 2.3313, 2.9066, 1.7432, 2.7069, 2.9199, 2.2666, 3.0475], device='cuda:3'), covar=tensor([0.1477, 0.2111, 0.1523, 0.2581, 0.0930, 0.1734, 0.2429, 0.0989], device='cuda:3'), in_proj_covar=tensor([0.0201, 0.0205, 0.0196, 0.0194, 0.0181, 0.0220, 0.0218, 0.0204], 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') 2023-03-26 10:43:29,165 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5278, 1.3723, 1.4059, 1.3713, 0.8247, 2.3268, 0.7948, 1.2893], device='cuda:3'), covar=tensor([0.3486, 0.2401, 0.2148, 0.2452, 0.2127, 0.0343, 0.2660, 0.1371], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0116, 0.0120, 0.0124, 0.0117, 0.0099, 0.0101, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 10:43:31,226 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.59 vs. limit=5.0 2023-03-26 10:43:45,668 INFO [finetune.py:976] (3/7) Epoch 9, batch 2800, loss[loss=0.158, simple_loss=0.2282, pruned_loss=0.04386, over 4831.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.26, pruned_loss=0.06656, over 954096.84 frames. ], batch size: 38, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:43:51,110 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.018e+02 1.496e+02 1.797e+02 2.211e+02 4.995e+02, threshold=3.593e+02, percent-clipped=1.0 2023-03-26 10:44:19,115 INFO [finetune.py:976] (3/7) Epoch 9, batch 2850, loss[loss=0.1887, simple_loss=0.2526, pruned_loss=0.06237, over 4891.00 frames. ], tot_loss[loss=0.1952, simple_loss=0.2584, pruned_loss=0.06601, over 954000.76 frames. ], batch size: 35, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:44:24,052 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48680.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:44:38,791 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.14 vs. limit=5.0 2023-03-26 10:45:04,573 INFO [finetune.py:976] (3/7) Epoch 9, batch 2900, loss[loss=0.1908, simple_loss=0.2765, pruned_loss=0.0526, over 4814.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2605, pruned_loss=0.0663, over 953695.89 frames. ], batch size: 40, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:45:08,310 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48728.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:45:10,029 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.274e+02 1.660e+02 1.926e+02 2.355e+02 4.281e+02, threshold=3.853e+02, percent-clipped=2.0 2023-03-26 10:45:16,931 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6425, 2.3318, 2.1261, 2.5704, 2.4644, 2.1978, 2.8813, 2.5275], device='cuda:3'), covar=tensor([0.1293, 0.2407, 0.3243, 0.2608, 0.2519, 0.1697, 0.2480, 0.1897], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0189, 0.0234, 0.0254, 0.0237, 0.0195, 0.0212, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 10:45:25,492 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48753.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:45:30,934 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-26 10:45:38,485 INFO [finetune.py:976] (3/7) Epoch 9, batch 2950, loss[loss=0.1744, simple_loss=0.2474, pruned_loss=0.05066, over 4896.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2643, pruned_loss=0.06738, over 953174.64 frames. ], batch size: 37, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:45:40,981 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=48776.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:45:42,827 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48779.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:45:46,138 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-26 10:46:11,327 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48821.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:46:11,795 INFO [finetune.py:976] (3/7) Epoch 9, batch 3000, loss[loss=0.2732, simple_loss=0.3157, pruned_loss=0.1154, over 4908.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2657, pruned_loss=0.06764, over 953842.64 frames. ], batch size: 36, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:46:11,795 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 10:46:22,398 INFO [finetune.py:1010] (3/7) Epoch 9, validation: loss=0.159, simple_loss=0.2302, pruned_loss=0.04393, over 2265189.00 frames. 2023-03-26 10:46:22,398 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 10:46:27,908 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.053e+02 1.631e+02 1.894e+02 2.277e+02 3.777e+02, threshold=3.789e+02, percent-clipped=0.0 2023-03-26 10:46:32,268 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48838.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:46:42,847 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.03 vs. limit=5.0 2023-03-26 10:46:52,190 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48868.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 10:46:54,487 INFO [finetune.py:976] (3/7) Epoch 9, batch 3050, loss[loss=0.2321, simple_loss=0.2838, pruned_loss=0.09017, over 4824.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2678, pruned_loss=0.06932, over 952003.79 frames. ], batch size: 47, lr: 3.79e-03, grad_scale: 16.0 2023-03-26 10:47:03,439 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48882.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:47:13,627 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48899.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:47:24,760 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=48916.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 10:47:29,279 INFO [finetune.py:976] (3/7) Epoch 9, batch 3100, loss[loss=0.1755, simple_loss=0.2277, pruned_loss=0.06162, over 4201.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2657, pruned_loss=0.06887, over 952765.54 frames. ], batch size: 18, lr: 3.79e-03, grad_scale: 32.0 2023-03-26 10:47:36,135 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.216e+02 1.624e+02 1.916e+02 2.206e+02 4.881e+02, threshold=3.833e+02, percent-clipped=2.0 2023-03-26 10:47:59,949 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-03-26 10:48:07,637 INFO [finetune.py:976] (3/7) Epoch 9, batch 3150, loss[loss=0.1824, simple_loss=0.2435, pruned_loss=0.0607, over 4793.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2632, pruned_loss=0.06791, over 952389.66 frames. ], batch size: 51, lr: 3.78e-03, grad_scale: 32.0 2023-03-26 10:48:16,690 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48980.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:48:33,970 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9779, 3.9519, 3.7344, 2.2863, 4.0017, 3.0820, 1.3897, 2.8695], device='cuda:3'), covar=tensor([0.2390, 0.1726, 0.1628, 0.3119, 0.1018, 0.1055, 0.4225, 0.1425], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0173, 0.0159, 0.0127, 0.0156, 0.0122, 0.0146, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 10:48:51,002 INFO [finetune.py:976] (3/7) Epoch 9, batch 3200, loss[loss=0.2096, simple_loss=0.2652, pruned_loss=0.07701, over 4718.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2603, pruned_loss=0.06667, over 952293.95 frames. ], batch size: 54, lr: 3.78e-03, grad_scale: 32.0 2023-03-26 10:48:55,660 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=49028.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:48:57,864 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.147e+02 1.665e+02 1.973e+02 2.326e+02 6.022e+02, threshold=3.945e+02, percent-clipped=4.0 2023-03-26 10:49:00,951 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1900, 2.0725, 1.7852, 2.1982, 1.9858, 2.0136, 1.9500, 2.7858], device='cuda:3'), covar=tensor([0.4809, 0.6104, 0.4117, 0.5570, 0.5431, 0.2953, 0.5942, 0.2051], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0259, 0.0221, 0.0280, 0.0242, 0.0209, 0.0246, 0.0210], 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') 2023-03-26 10:49:12,708 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49053.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:49:30,122 INFO [finetune.py:976] (3/7) Epoch 9, batch 3250, loss[loss=0.226, simple_loss=0.2895, pruned_loss=0.08123, over 4761.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2612, pruned_loss=0.06738, over 952790.51 frames. ], batch size: 59, lr: 3.78e-03, grad_scale: 32.0 2023-03-26 10:49:40,158 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49079.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:50:05,720 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=49101.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:50:21,989 INFO [finetune.py:976] (3/7) Epoch 9, batch 3300, loss[loss=0.1304, simple_loss=0.1949, pruned_loss=0.03297, over 4724.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.2652, pruned_loss=0.06907, over 953765.48 frames. ], batch size: 23, lr: 3.78e-03, grad_scale: 32.0 2023-03-26 10:50:26,596 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=49127.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:50:28,548 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.54 vs. limit=5.0 2023-03-26 10:50:28,939 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.155e+02 1.707e+02 1.914e+02 2.346e+02 3.542e+02, threshold=3.827e+02, percent-clipped=0.0 2023-03-26 10:50:56,005 INFO [finetune.py:976] (3/7) Epoch 9, batch 3350, loss[loss=0.159, simple_loss=0.2259, pruned_loss=0.04607, over 4773.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2681, pruned_loss=0.07023, over 954972.76 frames. ], batch size: 28, lr: 3.78e-03, grad_scale: 32.0 2023-03-26 10:50:58,575 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7079, 1.5093, 1.3974, 1.7920, 2.2898, 1.7686, 1.4349, 1.3701], device='cuda:3'), covar=tensor([0.2282, 0.2186, 0.2067, 0.1684, 0.1691, 0.1210, 0.2507, 0.2032], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0209, 0.0207, 0.0188, 0.0241, 0.0180, 0.0214, 0.0195], 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') 2023-03-26 10:50:59,112 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49177.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:51:17,573 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49194.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:51:41,873 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.43 vs. limit=5.0 2023-03-26 10:51:50,171 INFO [finetune.py:976] (3/7) Epoch 9, batch 3400, loss[loss=0.2024, simple_loss=0.2655, pruned_loss=0.06969, over 4898.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2695, pruned_loss=0.07069, over 954304.49 frames. ], batch size: 35, lr: 3.78e-03, grad_scale: 32.0 2023-03-26 10:51:59,365 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2437, 1.3691, 1.5828, 1.1316, 1.1644, 1.4800, 1.2860, 1.5920], device='cuda:3'), covar=tensor([0.1325, 0.2017, 0.1242, 0.1521, 0.1011, 0.1297, 0.2883, 0.0907], device='cuda:3'), in_proj_covar=tensor([0.0199, 0.0205, 0.0194, 0.0192, 0.0180, 0.0218, 0.0217, 0.0201], 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') 2023-03-26 10:51:59,827 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.353e+01 1.606e+02 1.878e+02 2.295e+02 4.525e+02, threshold=3.756e+02, percent-clipped=2.0 2023-03-26 10:52:10,023 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49236.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:52:40,930 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1775, 1.3491, 1.6072, 1.0086, 1.1591, 1.4278, 1.2985, 1.5608], device='cuda:3'), covar=tensor([0.1285, 0.2134, 0.1349, 0.1629, 0.0997, 0.1152, 0.2772, 0.0858], device='cuda:3'), in_proj_covar=tensor([0.0199, 0.0205, 0.0194, 0.0193, 0.0181, 0.0219, 0.0218, 0.0202], 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') 2023-03-26 10:52:55,272 INFO [finetune.py:976] (3/7) Epoch 9, batch 3450, loss[loss=0.1452, simple_loss=0.218, pruned_loss=0.0362, over 4750.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2692, pruned_loss=0.07016, over 954486.51 frames. ], batch size: 27, lr: 3.78e-03, grad_scale: 32.0 2023-03-26 10:53:32,517 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49297.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:53:32,844 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-03-26 10:53:53,354 INFO [finetune.py:976] (3/7) Epoch 9, batch 3500, loss[loss=0.2049, simple_loss=0.2666, pruned_loss=0.07158, over 4823.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2673, pruned_loss=0.06976, over 955791.24 frames. ], batch size: 38, lr: 3.78e-03, grad_scale: 32.0 2023-03-26 10:53:58,773 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.331e+01 1.641e+02 1.916e+02 2.289e+02 6.335e+02, threshold=3.833e+02, percent-clipped=2.0 2023-03-26 10:54:32,776 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-03-26 10:54:34,217 INFO [finetune.py:976] (3/7) Epoch 9, batch 3550, loss[loss=0.1515, simple_loss=0.2149, pruned_loss=0.04404, over 4827.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2636, pruned_loss=0.06832, over 955844.01 frames. ], batch size: 25, lr: 3.78e-03, grad_scale: 32.0 2023-03-26 10:55:09,817 INFO [finetune.py:976] (3/7) Epoch 9, batch 3600, loss[loss=0.1628, simple_loss=0.2425, pruned_loss=0.04159, over 4760.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2618, pruned_loss=0.06801, over 956756.53 frames. ], batch size: 27, lr: 3.78e-03, grad_scale: 32.0 2023-03-26 10:55:09,931 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49422.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:55:10,544 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.4895, 1.6060, 1.6771, 0.9122, 1.7726, 1.8770, 1.8988, 1.4538], device='cuda:3'), covar=tensor([0.0959, 0.0563, 0.0434, 0.0581, 0.0373, 0.0511, 0.0293, 0.0654], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0155, 0.0121, 0.0134, 0.0131, 0.0125, 0.0145, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.5212e-05, 1.1410e-04, 8.7053e-05, 9.7567e-05, 9.3595e-05, 9.1443e-05, 1.0637e-04, 1.0779e-04], device='cuda:3') 2023-03-26 10:55:15,221 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.012e+02 1.662e+02 2.002e+02 2.382e+02 4.044e+02, threshold=4.004e+02, percent-clipped=1.0 2023-03-26 10:55:15,969 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49432.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 10:55:27,915 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5580, 1.3881, 1.9890, 3.0325, 2.0330, 2.3070, 1.0179, 2.3832], device='cuda:3'), covar=tensor([0.1867, 0.1789, 0.1507, 0.0914, 0.0984, 0.1561, 0.2016, 0.0733], device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0118, 0.0135, 0.0167, 0.0103, 0.0140, 0.0127, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 10:55:38,989 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6042, 1.6040, 1.7270, 1.8581, 1.5775, 3.4395, 1.5095, 1.6786], device='cuda:3'), covar=tensor([0.0966, 0.1672, 0.1081, 0.0951, 0.1595, 0.0255, 0.1390, 0.1669], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0080, 0.0075, 0.0077, 0.0091, 0.0081, 0.0084, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 10:55:43,180 INFO [finetune.py:976] (3/7) Epoch 9, batch 3650, loss[loss=0.2307, simple_loss=0.2951, pruned_loss=0.08313, over 4837.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2654, pruned_loss=0.06946, over 956200.46 frames. ], batch size: 33, lr: 3.78e-03, grad_scale: 32.0 2023-03-26 10:55:46,381 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49477.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:55:50,096 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49483.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:55:56,197 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49493.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 10:55:56,753 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49494.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:55:58,147 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-26 10:56:17,055 INFO [finetune.py:976] (3/7) Epoch 9, batch 3700, loss[loss=0.1978, simple_loss=0.2678, pruned_loss=0.06391, over 4931.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2693, pruned_loss=0.07049, over 953164.44 frames. ], batch size: 38, lr: 3.78e-03, grad_scale: 32.0 2023-03-26 10:56:18,954 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=49525.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:56:22,513 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.088e+02 1.782e+02 2.076e+02 2.384e+02 4.659e+02, threshold=4.152e+02, percent-clipped=5.0 2023-03-26 10:56:28,071 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.4661, 1.6760, 1.7405, 1.0138, 1.8065, 2.0227, 1.9467, 1.5264], device='cuda:3'), covar=tensor([0.0914, 0.0581, 0.0385, 0.0565, 0.0375, 0.0482, 0.0316, 0.0595], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0155, 0.0121, 0.0135, 0.0131, 0.0125, 0.0146, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.5661e-05, 1.1441e-04, 8.7249e-05, 9.7722e-05, 9.4007e-05, 9.1824e-05, 1.0671e-04, 1.0803e-04], device='cuda:3') 2023-03-26 10:56:29,207 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=49542.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:56:29,883 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3769, 2.5703, 2.3652, 1.7122, 2.2983, 2.7238, 2.5643, 2.2188], device='cuda:3'), covar=tensor([0.0628, 0.0514, 0.0834, 0.1044, 0.1117, 0.0665, 0.0645, 0.0938], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0132, 0.0144, 0.0124, 0.0118, 0.0144, 0.0144, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 10:56:50,563 INFO [finetune.py:976] (3/7) Epoch 9, batch 3750, loss[loss=0.2258, simple_loss=0.2996, pruned_loss=0.07596, over 4852.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2712, pruned_loss=0.07124, over 954979.17 frames. ], batch size: 44, lr: 3.78e-03, grad_scale: 32.0 2023-03-26 10:57:02,678 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49592.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:57:28,146 INFO [finetune.py:976] (3/7) Epoch 9, batch 3800, loss[loss=0.237, simple_loss=0.3094, pruned_loss=0.08234, over 4814.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2717, pruned_loss=0.07145, over 955944.32 frames. ], batch size: 38, lr: 3.78e-03, grad_scale: 32.0 2023-03-26 10:57:39,042 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.087e+02 1.635e+02 1.863e+02 2.259e+02 4.048e+02, threshold=3.725e+02, percent-clipped=0.0 2023-03-26 10:58:12,406 INFO [finetune.py:976] (3/7) Epoch 9, batch 3850, loss[loss=0.1859, simple_loss=0.2432, pruned_loss=0.06429, over 4719.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2681, pruned_loss=0.06926, over 955405.91 frames. ], batch size: 59, lr: 3.78e-03, grad_scale: 32.0 2023-03-26 10:58:28,255 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7724, 1.5540, 1.4204, 1.1429, 1.5934, 1.5119, 1.5759, 2.1012], device='cuda:3'), covar=tensor([0.4850, 0.4821, 0.3633, 0.4561, 0.4070, 0.2720, 0.4088, 0.1963], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0261, 0.0222, 0.0282, 0.0243, 0.0209, 0.0246, 0.0211], 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') 2023-03-26 10:58:38,348 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3410, 1.2190, 1.6048, 2.4812, 1.6290, 2.2055, 0.8620, 2.0134], device='cuda:3'), covar=tensor([0.1817, 0.1585, 0.1164, 0.0665, 0.0933, 0.1152, 0.1575, 0.0661], device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0119, 0.0135, 0.0166, 0.0103, 0.0140, 0.0127, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 10:58:48,063 INFO [finetune.py:976] (3/7) Epoch 9, batch 3900, loss[loss=0.19, simple_loss=0.2462, pruned_loss=0.06688, over 4866.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2642, pruned_loss=0.06783, over 953576.87 frames. ], batch size: 31, lr: 3.78e-03, grad_scale: 32.0 2023-03-26 10:58:58,259 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.078e+02 1.638e+02 1.913e+02 2.415e+02 4.821e+02, threshold=3.825e+02, percent-clipped=2.0 2023-03-26 10:59:31,484 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49766.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:59:36,462 INFO [finetune.py:976] (3/7) Epoch 9, batch 3950, loss[loss=0.2237, simple_loss=0.276, pruned_loss=0.08565, over 4774.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2604, pruned_loss=0.06653, over 954374.55 frames. ], batch size: 54, lr: 3.78e-03, grad_scale: 32.0 2023-03-26 10:59:40,656 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49778.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 10:59:47,183 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49788.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 11:00:09,539 INFO [finetune.py:976] (3/7) Epoch 9, batch 4000, loss[loss=0.2009, simple_loss=0.2564, pruned_loss=0.07268, over 4798.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2612, pruned_loss=0.06712, over 956474.36 frames. ], batch size: 29, lr: 3.78e-03, grad_scale: 32.0 2023-03-26 11:00:13,726 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49827.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:00:16,515 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.178e+02 1.645e+02 2.017e+02 2.376e+02 6.319e+02, threshold=4.034e+02, percent-clipped=3.0 2023-03-26 11:00:42,827 INFO [finetune.py:976] (3/7) Epoch 9, batch 4050, loss[loss=0.215, simple_loss=0.283, pruned_loss=0.07348, over 4867.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.264, pruned_loss=0.06822, over 957840.47 frames. ], batch size: 31, lr: 3.78e-03, grad_scale: 32.0 2023-03-26 11:00:56,958 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49892.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:01:00,013 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7607, 4.8069, 4.4901, 2.8209, 4.8972, 3.6788, 0.9196, 3.3677], device='cuda:3'), covar=tensor([0.2616, 0.1896, 0.1434, 0.2726, 0.0740, 0.0881, 0.4755, 0.1236], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0176, 0.0162, 0.0130, 0.0158, 0.0124, 0.0148, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 11:01:15,998 INFO [finetune.py:976] (3/7) Epoch 9, batch 4100, loss[loss=0.2108, simple_loss=0.2701, pruned_loss=0.07576, over 4810.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2671, pruned_loss=0.06952, over 954697.44 frames. ], batch size: 25, lr: 3.78e-03, grad_scale: 32.0 2023-03-26 11:01:18,093 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-03-26 11:01:22,949 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.254e+02 1.718e+02 2.083e+02 2.512e+02 3.689e+02, threshold=4.166e+02, percent-clipped=0.0 2023-03-26 11:01:28,935 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=49940.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:01:48,770 INFO [finetune.py:976] (3/7) Epoch 9, batch 4150, loss[loss=0.2296, simple_loss=0.2864, pruned_loss=0.08639, over 4162.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2679, pruned_loss=0.06953, over 954881.79 frames. ], batch size: 65, lr: 3.78e-03, grad_scale: 16.0 2023-03-26 11:02:23,457 INFO [finetune.py:976] (3/7) Epoch 9, batch 4200, loss[loss=0.1718, simple_loss=0.2447, pruned_loss=0.04943, over 4787.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2676, pruned_loss=0.06877, over 953094.11 frames. ], batch size: 29, lr: 3.78e-03, grad_scale: 16.0 2023-03-26 11:02:30,129 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-26 11:02:31,378 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.147e+02 1.706e+02 2.002e+02 2.506e+02 6.230e+02, threshold=4.003e+02, percent-clipped=2.0 2023-03-26 11:03:15,804 INFO [finetune.py:976] (3/7) Epoch 9, batch 4250, loss[loss=0.1633, simple_loss=0.2228, pruned_loss=0.05192, over 4308.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2646, pruned_loss=0.06752, over 950093.50 frames. ], batch size: 18, lr: 3.78e-03, grad_scale: 16.0 2023-03-26 11:03:24,180 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50078.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:03:32,322 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50088.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 11:03:53,926 INFO [finetune.py:976] (3/7) Epoch 9, batch 4300, loss[loss=0.2185, simple_loss=0.2826, pruned_loss=0.07718, over 4890.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2626, pruned_loss=0.06715, over 951762.04 frames. ], batch size: 32, lr: 3.78e-03, grad_scale: 16.0 2023-03-26 11:03:53,996 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50122.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:03:56,366 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=50126.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:03:57,633 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50128.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:04:00,430 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.003e+02 1.608e+02 1.917e+02 2.228e+02 4.011e+02, threshold=3.835e+02, percent-clipped=1.0 2023-03-26 11:04:03,879 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=50136.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 11:04:49,371 INFO [finetune.py:976] (3/7) Epoch 9, batch 4350, loss[loss=0.2045, simple_loss=0.2697, pruned_loss=0.06965, over 4260.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.2594, pruned_loss=0.06599, over 952166.69 frames. ], batch size: 65, lr: 3.78e-03, grad_scale: 16.0 2023-03-26 11:05:17,865 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50189.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:05:30,424 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50197.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 11:06:02,455 INFO [finetune.py:976] (3/7) Epoch 9, batch 4400, loss[loss=0.2191, simple_loss=0.2835, pruned_loss=0.07735, over 4816.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2629, pruned_loss=0.0682, over 952546.53 frames. ], batch size: 40, lr: 3.78e-03, grad_scale: 16.0 2023-03-26 11:06:14,078 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.827e+01 1.714e+02 1.989e+02 2.480e+02 5.028e+02, threshold=3.977e+02, percent-clipped=2.0 2023-03-26 11:06:37,325 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4466, 1.4456, 1.8167, 1.7441, 1.5287, 3.2482, 1.3074, 1.5909], device='cuda:3'), covar=tensor([0.1031, 0.1748, 0.1252, 0.0992, 0.1591, 0.0240, 0.1442, 0.1636], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0080, 0.0075, 0.0077, 0.0090, 0.0081, 0.0084, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004], device='cuda:3') 2023-03-26 11:06:48,109 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50258.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 11:06:58,870 INFO [finetune.py:976] (3/7) Epoch 9, batch 4450, loss[loss=0.1925, simple_loss=0.262, pruned_loss=0.06149, over 4822.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2669, pruned_loss=0.06956, over 953116.45 frames. ], batch size: 30, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:07:17,763 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.3142, 1.4634, 1.4616, 0.7773, 1.4560, 1.7366, 1.7131, 1.2780], device='cuda:3'), covar=tensor([0.0831, 0.0458, 0.0429, 0.0508, 0.0426, 0.0499, 0.0236, 0.0626], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0156, 0.0122, 0.0135, 0.0133, 0.0127, 0.0146, 0.0148], device='cuda:3'), out_proj_covar=tensor([9.6635e-05, 1.1478e-04, 8.7582e-05, 9.8374e-05, 9.4889e-05, 9.2778e-05, 1.0733e-04, 1.0894e-04], device='cuda:3') 2023-03-26 11:07:32,436 INFO [finetune.py:976] (3/7) Epoch 9, batch 4500, loss[loss=0.2067, simple_loss=0.2571, pruned_loss=0.07819, over 4398.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2685, pruned_loss=0.07054, over 951491.95 frames. ], batch size: 19, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:07:38,448 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.049e+02 1.523e+02 1.896e+02 2.480e+02 4.445e+02, threshold=3.793e+02, percent-clipped=2.0 2023-03-26 11:08:05,990 INFO [finetune.py:976] (3/7) Epoch 9, batch 4550, loss[loss=0.195, simple_loss=0.2635, pruned_loss=0.06319, over 4887.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.27, pruned_loss=0.07079, over 953592.53 frames. ], batch size: 35, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:08:58,491 INFO [finetune.py:976] (3/7) Epoch 9, batch 4600, loss[loss=0.1823, simple_loss=0.2493, pruned_loss=0.05767, over 4796.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2689, pruned_loss=0.07018, over 953951.25 frames. ], batch size: 25, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:08:58,582 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50422.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:08:59,213 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6234, 1.5832, 1.5536, 1.5869, 1.2318, 2.7991, 1.2856, 1.8202], device='cuda:3'), covar=tensor([0.3390, 0.2475, 0.1967, 0.2406, 0.1591, 0.0376, 0.2627, 0.1009], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0119, 0.0123, 0.0116, 0.0098, 0.0099, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 11:09:06,261 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.176e+02 1.671e+02 1.983e+02 2.416e+02 3.848e+02, threshold=3.965e+02, percent-clipped=1.0 2023-03-26 11:09:34,657 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9644, 1.8170, 1.7040, 2.0257, 2.4124, 2.1253, 1.5126, 1.6207], device='cuda:3'), covar=tensor([0.2133, 0.1982, 0.1861, 0.1633, 0.1821, 0.1095, 0.2573, 0.1885], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0210, 0.0208, 0.0190, 0.0243, 0.0181, 0.0216, 0.0197], 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') 2023-03-26 11:09:39,359 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=50470.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:09:40,499 INFO [finetune.py:976] (3/7) Epoch 9, batch 4650, loss[loss=0.1638, simple_loss=0.2316, pruned_loss=0.04802, over 4820.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2653, pruned_loss=0.06857, over 954035.24 frames. ], batch size: 30, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:09:50,497 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50484.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:10:22,781 INFO [finetune.py:976] (3/7) Epoch 9, batch 4700, loss[loss=0.2081, simple_loss=0.2692, pruned_loss=0.07354, over 4905.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2621, pruned_loss=0.06741, over 955646.75 frames. ], batch size: 36, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:10:28,852 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2259, 1.7558, 2.9280, 4.2560, 2.9570, 2.8128, 1.0553, 3.3231], device='cuda:3'), covar=tensor([0.1638, 0.1566, 0.1121, 0.0437, 0.0699, 0.1296, 0.1909, 0.0481], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0117, 0.0133, 0.0164, 0.0102, 0.0139, 0.0126, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 11:10:29,350 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.001e+02 1.527e+02 1.850e+02 2.224e+02 3.838e+02, threshold=3.699e+02, percent-clipped=0.0 2023-03-26 11:10:38,044 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50546.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:10:42,733 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50553.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 11:10:48,634 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9576, 1.4180, 1.8928, 1.8375, 1.6788, 1.5898, 1.7929, 1.6836], device='cuda:3'), covar=tensor([0.4118, 0.4906, 0.4173, 0.4620, 0.5797, 0.4330, 0.5590, 0.4110], device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0241, 0.0254, 0.0257, 0.0251, 0.0226, 0.0274, 0.0231], 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') 2023-03-26 11:10:56,119 INFO [finetune.py:976] (3/7) Epoch 9, batch 4750, loss[loss=0.2264, simple_loss=0.2905, pruned_loss=0.08114, over 4869.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2606, pruned_loss=0.06686, over 956562.17 frames. ], batch size: 34, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:11:18,585 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50607.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:11:29,493 INFO [finetune.py:976] (3/7) Epoch 9, batch 4800, loss[loss=0.2177, simple_loss=0.2833, pruned_loss=0.0761, over 4912.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2626, pruned_loss=0.0681, over 955341.48 frames. ], batch size: 36, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:11:36,103 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.695e+01 1.587e+02 1.907e+02 2.189e+02 3.978e+02, threshold=3.813e+02, percent-clipped=2.0 2023-03-26 11:11:44,970 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 2023-03-26 11:12:03,076 INFO [finetune.py:976] (3/7) Epoch 9, batch 4850, loss[loss=0.1951, simple_loss=0.2642, pruned_loss=0.06298, over 4684.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2649, pruned_loss=0.06853, over 954208.89 frames. ], batch size: 23, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:12:34,619 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0756, 1.5989, 2.4361, 3.8276, 2.7139, 2.5817, 0.5615, 3.0240], device='cuda:3'), covar=tensor([0.1695, 0.1624, 0.1416, 0.0464, 0.0750, 0.1825, 0.2263, 0.0488], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0118, 0.0135, 0.0165, 0.0102, 0.0140, 0.0127, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 11:12:36,219 INFO [finetune.py:976] (3/7) Epoch 9, batch 4900, loss[loss=0.191, simple_loss=0.2609, pruned_loss=0.06057, over 4756.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2662, pruned_loss=0.06888, over 953962.96 frames. ], batch size: 26, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:12:42,294 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.028e+02 1.610e+02 1.915e+02 2.289e+02 4.400e+02, threshold=3.830e+02, percent-clipped=2.0 2023-03-26 11:13:08,690 INFO [finetune.py:976] (3/7) Epoch 9, batch 4950, loss[loss=0.1603, simple_loss=0.2288, pruned_loss=0.04592, over 4896.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2682, pruned_loss=0.06927, over 956349.22 frames. ], batch size: 32, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:13:15,355 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50782.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:13:16,538 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50784.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:13:33,487 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50808.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 11:13:34,070 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9563, 1.6302, 2.2561, 1.6015, 2.0525, 2.1952, 1.6604, 2.3365], device='cuda:3'), covar=tensor([0.1063, 0.1952, 0.1155, 0.1712, 0.0708, 0.1172, 0.2318, 0.0639], device='cuda:3'), in_proj_covar=tensor([0.0200, 0.0206, 0.0195, 0.0193, 0.0180, 0.0219, 0.0219, 0.0202], 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') 2023-03-26 11:13:51,976 INFO [finetune.py:976] (3/7) Epoch 9, batch 5000, loss[loss=0.1941, simple_loss=0.2448, pruned_loss=0.0717, over 4169.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.267, pruned_loss=0.06911, over 957358.14 frames. ], batch size: 18, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:14:03,063 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.212e+02 1.752e+02 2.044e+02 2.444e+02 6.074e+02, threshold=4.089e+02, percent-clipped=4.0 2023-03-26 11:14:03,135 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=50832.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:14:13,391 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50843.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:14:24,227 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50853.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 11:14:30,185 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6420, 1.0991, 0.9177, 1.6851, 2.0835, 1.5057, 1.4938, 1.5081], device='cuda:3'), covar=tensor([0.2159, 0.3051, 0.2659, 0.1623, 0.2252, 0.2703, 0.2045, 0.3099], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0096, 0.0114, 0.0093, 0.0121, 0.0096, 0.0100, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 11:14:32,067 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-26 11:14:37,403 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50869.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 11:14:40,058 INFO [finetune.py:976] (3/7) Epoch 9, batch 5050, loss[loss=0.1909, simple_loss=0.2399, pruned_loss=0.07094, over 4674.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2643, pruned_loss=0.06842, over 958039.03 frames. ], batch size: 23, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:15:00,247 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=50901.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 11:15:00,835 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50902.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:15:01,494 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6791, 1.5202, 1.5568, 1.6645, 1.1925, 3.3749, 1.3974, 1.8590], device='cuda:3'), covar=tensor([0.3342, 0.2344, 0.2070, 0.2253, 0.1847, 0.0190, 0.2510, 0.1258], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0114, 0.0119, 0.0122, 0.0115, 0.0098, 0.0099, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 11:15:21,256 INFO [finetune.py:976] (3/7) Epoch 9, batch 5100, loss[loss=0.1996, simple_loss=0.2649, pruned_loss=0.06714, over 4935.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2612, pruned_loss=0.06726, over 958350.77 frames. ], batch size: 38, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:15:29,784 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.163e+02 1.571e+02 1.989e+02 2.366e+02 5.072e+02, threshold=3.977e+02, percent-clipped=2.0 2023-03-26 11:15:38,834 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50946.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:15:55,122 INFO [finetune.py:976] (3/7) Epoch 9, batch 5150, loss[loss=0.2332, simple_loss=0.2967, pruned_loss=0.08483, over 4874.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2618, pruned_loss=0.0675, over 958343.28 frames. ], batch size: 34, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:15:55,846 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6680, 1.4347, 1.5941, 1.5601, 1.2463, 3.6339, 1.5227, 2.0870], device='cuda:3'), covar=tensor([0.3667, 0.2691, 0.2164, 0.2491, 0.1935, 0.0206, 0.2638, 0.1284], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0114, 0.0118, 0.0122, 0.0115, 0.0098, 0.0099, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 11:15:58,849 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.67 vs. limit=2.0 2023-03-26 11:15:59,407 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8107, 1.7355, 1.5110, 1.8769, 2.3476, 1.8725, 1.5695, 1.4187], device='cuda:3'), covar=tensor([0.2205, 0.2199, 0.2049, 0.1807, 0.1908, 0.1262, 0.2603, 0.2053], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0210, 0.0207, 0.0190, 0.0242, 0.0181, 0.0214, 0.0195], 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') 2023-03-26 11:16:04,443 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4132, 1.3103, 1.8064, 2.4355, 1.7130, 2.2201, 0.8440, 2.0010], device='cuda:3'), covar=tensor([0.1687, 0.1605, 0.1068, 0.0800, 0.0883, 0.1196, 0.1601, 0.0706], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0118, 0.0134, 0.0165, 0.0102, 0.0139, 0.0126, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 11:16:19,649 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51007.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:16:23,417 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7863, 1.6400, 1.4683, 1.4043, 1.7761, 1.5270, 1.8584, 1.7082], device='cuda:3'), covar=tensor([0.1474, 0.2213, 0.3304, 0.2706, 0.2932, 0.1870, 0.3172, 0.2079], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0188, 0.0234, 0.0255, 0.0239, 0.0196, 0.0212, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 11:16:29,134 INFO [finetune.py:976] (3/7) Epoch 9, batch 5200, loss[loss=0.1865, simple_loss=0.2557, pruned_loss=0.0586, over 4901.00 frames. ], tot_loss[loss=0.202, simple_loss=0.2662, pruned_loss=0.06888, over 958347.74 frames. ], batch size: 35, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:16:37,438 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.247e+02 1.691e+02 2.095e+02 2.506e+02 4.401e+02, threshold=4.191e+02, percent-clipped=2.0 2023-03-26 11:16:48,098 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7063, 1.1028, 0.8019, 1.6574, 2.0095, 1.3044, 1.4720, 1.5562], device='cuda:3'), covar=tensor([0.1458, 0.2118, 0.2128, 0.1131, 0.1957, 0.2217, 0.1419, 0.1924], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0097, 0.0114, 0.0093, 0.0121, 0.0096, 0.0100, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 11:17:12,593 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51062.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:17:18,565 INFO [finetune.py:976] (3/7) Epoch 9, batch 5250, loss[loss=0.1987, simple_loss=0.2709, pruned_loss=0.06327, over 4895.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2681, pruned_loss=0.06912, over 956633.93 frames. ], batch size: 43, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:17:41,124 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.2366, 3.6855, 3.8844, 4.0190, 3.9662, 3.7179, 4.3076, 1.2648], device='cuda:3'), covar=tensor([0.0803, 0.0831, 0.0800, 0.1072, 0.1251, 0.1498, 0.0767, 0.6140], device='cuda:3'), in_proj_covar=tensor([0.0357, 0.0247, 0.0281, 0.0297, 0.0336, 0.0286, 0.0307, 0.0302], 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') 2023-03-26 11:17:42,742 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-03-26 11:17:50,644 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.1310, 4.8367, 4.5108, 2.7038, 4.9299, 3.8104, 1.1151, 3.2936], device='cuda:3'), covar=tensor([0.2169, 0.1361, 0.1507, 0.2824, 0.0673, 0.0810, 0.4338, 0.1231], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0174, 0.0159, 0.0127, 0.0156, 0.0122, 0.0146, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 11:17:51,198 INFO [finetune.py:976] (3/7) Epoch 9, batch 5300, loss[loss=0.1725, simple_loss=0.2433, pruned_loss=0.05088, over 4747.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2691, pruned_loss=0.0693, over 956420.59 frames. ], batch size: 27, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:17:51,932 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51123.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:17:57,262 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.167e+02 1.689e+02 2.023e+02 2.414e+02 5.734e+02, threshold=4.045e+02, percent-clipped=1.0 2023-03-26 11:18:01,386 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51138.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:18:11,912 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8479, 1.6600, 1.5091, 1.8757, 2.4070, 1.8810, 1.6843, 1.4760], device='cuda:3'), covar=tensor([0.2115, 0.2091, 0.1866, 0.1718, 0.1747, 0.1180, 0.2429, 0.1829], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0209, 0.0207, 0.0189, 0.0242, 0.0181, 0.0214, 0.0195], 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') 2023-03-26 11:18:19,581 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51164.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 11:18:24,357 INFO [finetune.py:976] (3/7) Epoch 9, batch 5350, loss[loss=0.1726, simple_loss=0.2366, pruned_loss=0.05427, over 4921.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2686, pruned_loss=0.06884, over 954171.64 frames. ], batch size: 33, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:18:24,448 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51172.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 11:18:53,728 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9887, 1.4148, 0.8124, 1.9755, 2.2722, 1.5188, 1.8217, 1.8543], device='cuda:3'), covar=tensor([0.1430, 0.2088, 0.2191, 0.1122, 0.1786, 0.1799, 0.1374, 0.1895], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0096, 0.0114, 0.0092, 0.0121, 0.0096, 0.0100, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 11:18:56,212 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51202.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:19:18,523 INFO [finetune.py:976] (3/7) Epoch 9, batch 5400, loss[loss=0.213, simple_loss=0.2669, pruned_loss=0.07951, over 4816.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2645, pruned_loss=0.06777, over 953537.73 frames. ], batch size: 33, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:19:18,636 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4681, 1.3644, 1.5093, 0.9127, 1.5184, 1.4920, 1.4423, 1.1927], device='cuda:3'), covar=tensor([0.0762, 0.1034, 0.0832, 0.1018, 0.0862, 0.0837, 0.0807, 0.1864], device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0134, 0.0146, 0.0125, 0.0118, 0.0145, 0.0145, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 11:19:26,734 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.913e+01 1.608e+02 1.826e+02 2.251e+02 3.272e+02, threshold=3.651e+02, percent-clipped=0.0 2023-03-26 11:19:32,706 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51233.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 11:19:48,896 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=51250.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:20:03,229 INFO [finetune.py:976] (3/7) Epoch 9, batch 5450, loss[loss=0.1328, simple_loss=0.211, pruned_loss=0.02732, over 4753.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.261, pruned_loss=0.0663, over 955179.17 frames. ], batch size: 27, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:20:31,733 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51302.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:20:50,279 INFO [finetune.py:976] (3/7) Epoch 9, batch 5500, loss[loss=0.1548, simple_loss=0.2247, pruned_loss=0.04246, over 4795.00 frames. ], tot_loss[loss=0.1937, simple_loss=0.2576, pruned_loss=0.06491, over 956656.56 frames. ], batch size: 29, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:20:56,825 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.040e+02 1.581e+02 1.869e+02 2.249e+02 3.902e+02, threshold=3.738e+02, percent-clipped=2.0 2023-03-26 11:21:23,245 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5516, 1.5983, 1.3855, 1.4313, 1.9065, 1.7408, 1.5694, 1.4114], device='cuda:3'), covar=tensor([0.0321, 0.0323, 0.0544, 0.0344, 0.0199, 0.0550, 0.0316, 0.0417], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0110, 0.0141, 0.0117, 0.0103, 0.0103, 0.0092, 0.0109], device='cuda:3'), out_proj_covar=tensor([7.1193e-05, 8.6235e-05, 1.1206e-04, 9.1541e-05, 8.0991e-05, 7.6444e-05, 6.9872e-05, 8.4048e-05], device='cuda:3') 2023-03-26 11:21:48,954 INFO [finetune.py:976] (3/7) Epoch 9, batch 5550, loss[loss=0.1897, simple_loss=0.2716, pruned_loss=0.05384, over 4823.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2617, pruned_loss=0.06716, over 953689.37 frames. ], batch size: 40, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:21:59,530 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51382.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:22:07,961 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51392.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:22:24,403 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51418.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:22:27,577 INFO [finetune.py:976] (3/7) Epoch 9, batch 5600, loss[loss=0.2383, simple_loss=0.3041, pruned_loss=0.08625, over 4797.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2653, pruned_loss=0.06821, over 953710.84 frames. ], batch size: 51, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:22:33,287 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.035e+02 1.626e+02 2.005e+02 2.362e+02 4.096e+02, threshold=4.011e+02, percent-clipped=2.0 2023-03-26 11:22:36,869 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51438.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:22:39,812 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51443.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:22:46,037 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51453.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:22:52,434 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51464.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 11:22:57,071 INFO [finetune.py:976] (3/7) Epoch 9, batch 5650, loss[loss=0.203, simple_loss=0.2666, pruned_loss=0.06966, over 4869.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2674, pruned_loss=0.06847, over 952321.91 frames. ], batch size: 44, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:23:05,306 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=51486.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:23:10,714 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51495.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:23:20,776 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=51512.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 11:23:26,711 INFO [finetune.py:976] (3/7) Epoch 9, batch 5700, loss[loss=0.1514, simple_loss=0.224, pruned_loss=0.03934, over 4244.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2625, pruned_loss=0.068, over 932628.73 frames. ], batch size: 18, lr: 3.77e-03, grad_scale: 16.0 2023-03-26 11:23:30,371 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51528.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 11:23:32,893 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.987e+01 1.624e+02 1.963e+02 2.341e+02 6.572e+02, threshold=3.927e+02, percent-clipped=1.0 2023-03-26 11:23:33,078 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-03-26 11:23:40,187 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1132, 1.8883, 1.7122, 1.7280, 2.0506, 1.8320, 2.1573, 2.0592], device='cuda:3'), covar=tensor([0.1387, 0.2300, 0.3391, 0.2860, 0.2687, 0.1756, 0.3050, 0.1942], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0188, 0.0233, 0.0254, 0.0238, 0.0196, 0.0212, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 11:23:57,292 INFO [finetune.py:976] (3/7) Epoch 10, batch 0, loss[loss=0.2179, simple_loss=0.2868, pruned_loss=0.0745, over 4878.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2868, pruned_loss=0.0745, over 4878.00 frames. ], batch size: 43, lr: 3.76e-03, grad_scale: 16.0 2023-03-26 11:23:57,292 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 11:24:16,170 INFO [finetune.py:1010] (3/7) Epoch 10, validation: loss=0.1604, simple_loss=0.2317, pruned_loss=0.04451, over 2265189.00 frames. 2023-03-26 11:24:16,171 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 11:24:22,494 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51556.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:24:31,187 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-26 11:24:40,648 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-26 11:24:46,066 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0397, 1.8400, 1.4988, 1.7076, 1.7439, 1.7111, 1.7265, 2.4783], device='cuda:3'), covar=tensor([0.4482, 0.4989, 0.3697, 0.4613, 0.4706, 0.2752, 0.4785, 0.1885], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0259, 0.0222, 0.0280, 0.0243, 0.0209, 0.0245, 0.0213], 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') 2023-03-26 11:24:58,294 INFO [finetune.py:976] (3/7) Epoch 10, batch 50, loss[loss=0.1994, simple_loss=0.2613, pruned_loss=0.06874, over 4846.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2656, pruned_loss=0.06669, over 216091.00 frames. ], batch size: 44, lr: 3.76e-03, grad_scale: 16.0 2023-03-26 11:25:01,631 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51602.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:25:20,208 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.916e+01 1.735e+02 2.131e+02 2.642e+02 7.480e+02, threshold=4.262e+02, percent-clipped=4.0 2023-03-26 11:25:31,997 INFO [finetune.py:976] (3/7) Epoch 10, batch 100, loss[loss=0.1737, simple_loss=0.2421, pruned_loss=0.0527, over 4727.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2604, pruned_loss=0.06617, over 379427.96 frames. ], batch size: 23, lr: 3.76e-03, grad_scale: 16.0 2023-03-26 11:25:32,639 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=51650.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:25:44,185 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5919, 1.4541, 2.0611, 3.0897, 2.1063, 2.2639, 0.8409, 2.4432], device='cuda:3'), covar=tensor([0.1698, 0.1538, 0.1234, 0.0631, 0.0752, 0.1415, 0.1933, 0.0591], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0117, 0.0134, 0.0164, 0.0102, 0.0139, 0.0126, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 11:25:56,463 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6288, 1.6203, 2.2759, 3.5113, 2.3916, 2.5122, 1.2642, 2.7339], device='cuda:3'), covar=tensor([0.1752, 0.1480, 0.1290, 0.0562, 0.0776, 0.1352, 0.1651, 0.0558], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0117, 0.0134, 0.0164, 0.0102, 0.0139, 0.0126, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 11:26:04,770 INFO [finetune.py:976] (3/7) Epoch 10, batch 150, loss[loss=0.2206, simple_loss=0.2678, pruned_loss=0.08672, over 4802.00 frames. ], tot_loss[loss=0.1947, simple_loss=0.2576, pruned_loss=0.06586, over 509191.50 frames. ], batch size: 45, lr: 3.76e-03, grad_scale: 16.0 2023-03-26 11:26:16,285 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3530, 2.9118, 2.7981, 1.1838, 3.0525, 2.2716, 0.7516, 1.8688], device='cuda:3'), covar=tensor([0.2424, 0.2092, 0.1616, 0.3690, 0.1336, 0.1125, 0.4169, 0.1770], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0173, 0.0159, 0.0128, 0.0156, 0.0122, 0.0145, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 11:26:18,704 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51718.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:26:20,953 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.52 vs. limit=5.0 2023-03-26 11:26:26,616 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2020, 1.9746, 1.9516, 2.2640, 2.7424, 2.2003, 1.9972, 1.5968], device='cuda:3'), covar=tensor([0.2275, 0.2306, 0.1998, 0.1783, 0.1971, 0.1273, 0.2418, 0.2054], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0209, 0.0208, 0.0189, 0.0241, 0.0181, 0.0214, 0.0196], 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') 2023-03-26 11:26:33,392 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.046e+02 1.594e+02 1.858e+02 2.240e+02 3.308e+02, threshold=3.716e+02, percent-clipped=0.0 2023-03-26 11:26:37,113 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51738.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:26:47,597 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51747.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:26:48,167 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51748.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:26:52,005 INFO [finetune.py:976] (3/7) Epoch 10, batch 200, loss[loss=0.1864, simple_loss=0.2421, pruned_loss=0.06536, over 4693.00 frames. ], tot_loss[loss=0.1952, simple_loss=0.257, pruned_loss=0.06665, over 608863.72 frames. ], batch size: 23, lr: 3.76e-03, grad_scale: 16.0 2023-03-26 11:27:04,852 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=51766.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:27:25,424 INFO [finetune.py:976] (3/7) Epoch 10, batch 250, loss[loss=0.1961, simple_loss=0.2574, pruned_loss=0.06741, over 4890.00 frames. ], tot_loss[loss=0.1961, simple_loss=0.2589, pruned_loss=0.06667, over 686788.66 frames. ], batch size: 35, lr: 3.76e-03, grad_scale: 16.0 2023-03-26 11:27:32,991 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51808.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:27:45,687 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51828.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 11:27:48,002 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.029e+02 1.605e+02 1.930e+02 2.331e+02 5.576e+02, threshold=3.861e+02, percent-clipped=5.0 2023-03-26 11:27:58,881 INFO [finetune.py:976] (3/7) Epoch 10, batch 300, loss[loss=0.2371, simple_loss=0.2964, pruned_loss=0.08891, over 4859.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2634, pruned_loss=0.06815, over 746030.84 frames. ], batch size: 44, lr: 3.76e-03, grad_scale: 16.0 2023-03-26 11:28:00,157 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51851.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:28:04,969 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0302, 1.1694, 1.8981, 1.8489, 1.6592, 1.6066, 1.7503, 1.7269], device='cuda:3'), covar=tensor([0.3900, 0.5039, 0.4058, 0.4099, 0.5451, 0.3931, 0.5276, 0.3874], device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0241, 0.0255, 0.0257, 0.0252, 0.0227, 0.0276, 0.0230], 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') 2023-03-26 11:28:17,687 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=51876.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 11:28:31,956 INFO [finetune.py:976] (3/7) Epoch 10, batch 350, loss[loss=0.2426, simple_loss=0.2986, pruned_loss=0.09331, over 4790.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.266, pruned_loss=0.06865, over 794516.23 frames. ], batch size: 51, lr: 3.76e-03, grad_scale: 16.0 2023-03-26 11:28:46,750 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.45 vs. limit=5.0 2023-03-26 11:28:54,279 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.052e+02 1.678e+02 2.044e+02 2.443e+02 3.814e+02, threshold=4.089e+02, percent-clipped=0.0 2023-03-26 11:29:04,642 INFO [finetune.py:976] (3/7) Epoch 10, batch 400, loss[loss=0.1877, simple_loss=0.2522, pruned_loss=0.06156, over 4747.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2661, pruned_loss=0.068, over 830387.96 frames. ], batch size: 26, lr: 3.76e-03, grad_scale: 16.0 2023-03-26 11:29:12,697 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-03-26 11:29:17,476 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7391, 1.4615, 2.2664, 3.5703, 2.4637, 2.5834, 0.8730, 2.7269], device='cuda:3'), covar=tensor([0.1773, 0.1662, 0.1395, 0.0643, 0.0818, 0.1493, 0.2177, 0.0614], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0117, 0.0134, 0.0165, 0.0102, 0.0139, 0.0126, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 11:29:47,993 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.72 vs. limit=5.0 2023-03-26 11:29:56,982 INFO [finetune.py:976] (3/7) Epoch 10, batch 450, loss[loss=0.1862, simple_loss=0.2473, pruned_loss=0.0625, over 4929.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2651, pruned_loss=0.06793, over 859746.12 frames. ], batch size: 33, lr: 3.76e-03, grad_scale: 32.0 2023-03-26 11:30:13,861 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.59 vs. limit=5.0 2023-03-26 11:30:21,154 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 1.724e+02 2.025e+02 2.574e+02 4.346e+02, threshold=4.050e+02, percent-clipped=1.0 2023-03-26 11:30:24,956 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52038.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:30:26,198 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52040.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:30:30,967 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52048.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:30:31,463 INFO [finetune.py:976] (3/7) Epoch 10, batch 500, loss[loss=0.2003, simple_loss=0.2664, pruned_loss=0.0671, over 4827.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2619, pruned_loss=0.06662, over 878394.58 frames. ], batch size: 33, lr: 3.76e-03, grad_scale: 32.0 2023-03-26 11:30:56,806 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=52086.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:31:02,786 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=52096.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:31:04,604 INFO [finetune.py:976] (3/7) Epoch 10, batch 550, loss[loss=0.1738, simple_loss=0.2401, pruned_loss=0.05379, over 4909.00 frames. ], tot_loss[loss=0.195, simple_loss=0.2586, pruned_loss=0.0657, over 894528.84 frames. ], batch size: 36, lr: 3.76e-03, grad_scale: 32.0 2023-03-26 11:31:05,920 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52101.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:31:07,067 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52103.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:31:10,803 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1186, 2.1155, 1.7439, 2.2650, 2.0183, 1.9876, 1.9943, 2.8740], device='cuda:3'), covar=tensor([0.4608, 0.5349, 0.3767, 0.5425, 0.5099, 0.2738, 0.4921, 0.1787], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0259, 0.0222, 0.0279, 0.0243, 0.0209, 0.0244, 0.0212], 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') 2023-03-26 11:31:27,058 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.020e+02 1.591e+02 1.822e+02 2.163e+02 6.487e+02, threshold=3.643e+02, percent-clipped=1.0 2023-03-26 11:31:37,963 INFO [finetune.py:976] (3/7) Epoch 10, batch 600, loss[loss=0.185, simple_loss=0.2487, pruned_loss=0.06071, over 4906.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.2593, pruned_loss=0.06603, over 909068.03 frames. ], batch size: 36, lr: 3.76e-03, grad_scale: 32.0 2023-03-26 11:31:39,254 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52151.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:32:10,102 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1493, 2.0362, 1.6338, 2.0457, 1.9716, 1.9089, 1.9005, 2.8902], device='cuda:3'), covar=tensor([0.4425, 0.5676, 0.4050, 0.5363, 0.5270, 0.2739, 0.5242, 0.1779], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0259, 0.0222, 0.0280, 0.0244, 0.0209, 0.0245, 0.0213], 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') 2023-03-26 11:32:11,832 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5296, 1.4077, 1.4192, 1.5784, 0.9444, 3.2858, 1.2300, 1.7344], device='cuda:3'), covar=tensor([0.3362, 0.2410, 0.2110, 0.2198, 0.1912, 0.0182, 0.2698, 0.1241], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0116, 0.0120, 0.0123, 0.0116, 0.0099, 0.0099, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 11:32:15,957 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52193.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:32:19,561 INFO [finetune.py:976] (3/7) Epoch 10, batch 650, loss[loss=0.2144, simple_loss=0.2912, pruned_loss=0.06881, over 4900.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2633, pruned_loss=0.0673, over 917578.65 frames. ], batch size: 37, lr: 3.76e-03, grad_scale: 32.0 2023-03-26 11:32:19,620 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=52199.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:32:42,611 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.128e+02 1.681e+02 1.969e+02 2.336e+02 3.855e+02, threshold=3.938e+02, percent-clipped=2.0 2023-03-26 11:32:46,417 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8604, 1.6114, 2.1442, 1.9945, 1.9199, 4.4670, 1.4917, 1.9868], device='cuda:3'), covar=tensor([0.1000, 0.1830, 0.1257, 0.1072, 0.1654, 0.0217, 0.1609, 0.1786], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0082, 0.0076, 0.0078, 0.0091, 0.0082, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 11:32:53,495 INFO [finetune.py:976] (3/7) Epoch 10, batch 700, loss[loss=0.2145, simple_loss=0.2863, pruned_loss=0.07137, over 4834.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2644, pruned_loss=0.06647, over 927387.09 frames. ], batch size: 49, lr: 3.76e-03, grad_scale: 32.0 2023-03-26 11:32:54,960 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-03-26 11:32:56,642 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52254.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:33:26,706 INFO [finetune.py:976] (3/7) Epoch 10, batch 750, loss[loss=0.232, simple_loss=0.2887, pruned_loss=0.08761, over 4755.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2667, pruned_loss=0.0676, over 935806.05 frames. ], batch size: 27, lr: 3.76e-03, grad_scale: 32.0 2023-03-26 11:33:45,058 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52312.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:34:02,791 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.081e+02 1.612e+02 1.864e+02 2.364e+02 4.342e+02, threshold=3.728e+02, percent-clipped=1.0 2023-03-26 11:34:15,210 INFO [finetune.py:976] (3/7) Epoch 10, batch 800, loss[loss=0.1809, simple_loss=0.2524, pruned_loss=0.05476, over 4671.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2662, pruned_loss=0.06703, over 938561.10 frames. ], batch size: 23, lr: 3.76e-03, grad_scale: 32.0 2023-03-26 11:34:30,565 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52373.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:34:49,110 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52396.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:34:50,909 INFO [finetune.py:976] (3/7) Epoch 10, batch 850, loss[loss=0.1915, simple_loss=0.2574, pruned_loss=0.06282, over 4739.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2632, pruned_loss=0.06585, over 942976.21 frames. ], batch size: 54, lr: 3.76e-03, grad_scale: 32.0 2023-03-26 11:34:54,128 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52403.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:35:14,904 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.098e+02 1.556e+02 1.848e+02 2.239e+02 3.627e+02, threshold=3.695e+02, percent-clipped=0.0 2023-03-26 11:35:26,367 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-26 11:35:36,902 INFO [finetune.py:976] (3/7) Epoch 10, batch 900, loss[loss=0.1605, simple_loss=0.2327, pruned_loss=0.04419, over 4851.00 frames. ], tot_loss[loss=0.1949, simple_loss=0.2602, pruned_loss=0.06479, over 946650.19 frames. ], batch size: 44, lr: 3.76e-03, grad_scale: 32.0 2023-03-26 11:35:38,211 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=52451.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:35:59,106 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4901, 1.4700, 1.5542, 0.8892, 1.5528, 1.6200, 1.5241, 1.3896], device='cuda:3'), covar=tensor([0.0586, 0.0812, 0.0718, 0.0916, 0.0828, 0.0710, 0.0645, 0.1130], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0132, 0.0142, 0.0123, 0.0118, 0.0141, 0.0141, 0.0159], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 11:36:16,380 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-03-26 11:36:25,700 INFO [finetune.py:976] (3/7) Epoch 10, batch 950, loss[loss=0.1717, simple_loss=0.2354, pruned_loss=0.05402, over 4828.00 frames. ], tot_loss[loss=0.1946, simple_loss=0.2589, pruned_loss=0.06514, over 948220.12 frames. ], batch size: 30, lr: 3.76e-03, grad_scale: 32.0 2023-03-26 11:36:45,066 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1223, 1.8633, 1.7082, 1.9428, 1.8590, 1.8803, 1.8788, 2.6442], device='cuda:3'), covar=tensor([0.5170, 0.6458, 0.4342, 0.5367, 0.5252, 0.3043, 0.5274, 0.2010], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0259, 0.0222, 0.0279, 0.0243, 0.0208, 0.0245, 0.0212], 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') 2023-03-26 11:36:46,734 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.124e+02 1.551e+02 1.918e+02 2.238e+02 5.409e+02, threshold=3.837e+02, percent-clipped=4.0 2023-03-26 11:37:01,194 INFO [finetune.py:976] (3/7) Epoch 10, batch 1000, loss[loss=0.2375, simple_loss=0.298, pruned_loss=0.08854, over 4817.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2615, pruned_loss=0.06594, over 947445.13 frames. ], batch size: 47, lr: 3.76e-03, grad_scale: 32.0 2023-03-26 11:37:01,276 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52549.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:37:04,948 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52555.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 11:37:05,206 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 2023-03-26 11:37:07,367 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-26 11:38:00,467 INFO [finetune.py:976] (3/7) Epoch 10, batch 1050, loss[loss=0.2094, simple_loss=0.2778, pruned_loss=0.0705, over 4919.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2649, pruned_loss=0.067, over 949167.47 frames. ], batch size: 42, lr: 3.76e-03, grad_scale: 32.0 2023-03-26 11:38:09,725 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5134, 1.5693, 1.3800, 1.4699, 1.8377, 1.7684, 1.6103, 1.3632], device='cuda:3'), covar=tensor([0.0323, 0.0321, 0.0595, 0.0320, 0.0239, 0.0544, 0.0275, 0.0398], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0109, 0.0139, 0.0114, 0.0102, 0.0102, 0.0091, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.0390e-05, 8.4967e-05, 1.1073e-04, 8.9827e-05, 7.9787e-05, 7.5280e-05, 6.8853e-05, 8.3132e-05], device='cuda:3') 2023-03-26 11:38:19,738 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-26 11:38:21,492 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52616.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 11:38:31,486 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.240e+02 1.591e+02 1.928e+02 2.293e+02 3.930e+02, threshold=3.855e+02, percent-clipped=1.0 2023-03-26 11:38:38,204 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-03-26 11:38:44,943 INFO [finetune.py:976] (3/7) Epoch 10, batch 1100, loss[loss=0.2238, simple_loss=0.2887, pruned_loss=0.07948, over 4901.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2665, pruned_loss=0.06744, over 951892.10 frames. ], batch size: 37, lr: 3.76e-03, grad_scale: 32.0 2023-03-26 11:38:59,650 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52668.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:39:17,756 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52696.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:39:19,453 INFO [finetune.py:976] (3/7) Epoch 10, batch 1150, loss[loss=0.1805, simple_loss=0.2594, pruned_loss=0.05078, over 4905.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2656, pruned_loss=0.067, over 951992.12 frames. ], batch size: 37, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:39:40,840 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.071e+02 1.654e+02 1.930e+02 2.314e+02 4.484e+02, threshold=3.861e+02, percent-clipped=2.0 2023-03-26 11:39:48,713 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=52744.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:39:52,581 INFO [finetune.py:976] (3/7) Epoch 10, batch 1200, loss[loss=0.154, simple_loss=0.2272, pruned_loss=0.04037, over 4641.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2639, pruned_loss=0.06628, over 951617.93 frames. ], batch size: 20, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:39:56,827 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52752.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:40:35,761 INFO [finetune.py:976] (3/7) Epoch 10, batch 1250, loss[loss=0.2226, simple_loss=0.2689, pruned_loss=0.08813, over 4730.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2621, pruned_loss=0.06616, over 952440.21 frames. ], batch size: 59, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:40:47,086 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52813.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:41:05,425 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.045e+02 1.514e+02 1.794e+02 2.223e+02 4.744e+02, threshold=3.588e+02, percent-clipped=2.0 2023-03-26 11:41:19,447 INFO [finetune.py:976] (3/7) Epoch 10, batch 1300, loss[loss=0.1908, simple_loss=0.2569, pruned_loss=0.06237, over 4834.00 frames. ], tot_loss[loss=0.1941, simple_loss=0.2588, pruned_loss=0.06467, over 951034.34 frames. ], batch size: 33, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:41:19,553 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52849.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:41:50,815 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.5224, 1.5355, 1.5820, 0.9764, 1.6298, 1.8102, 1.8433, 1.3846], device='cuda:3'), covar=tensor([0.0855, 0.0519, 0.0438, 0.0503, 0.0371, 0.0585, 0.0280, 0.0636], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0154, 0.0121, 0.0134, 0.0131, 0.0125, 0.0145, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.5418e-05, 1.1352e-04, 8.7310e-05, 9.7169e-05, 9.3666e-05, 9.1461e-05, 1.0626e-04, 1.0772e-04], device='cuda:3') 2023-03-26 11:41:51,930 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=52897.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:41:53,103 INFO [finetune.py:976] (3/7) Epoch 10, batch 1350, loss[loss=0.2131, simple_loss=0.2778, pruned_loss=0.07419, over 4849.00 frames. ], tot_loss[loss=0.1958, simple_loss=0.2602, pruned_loss=0.0657, over 952954.07 frames. ], batch size: 44, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:42:02,437 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52911.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 11:42:10,717 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9857, 4.9033, 4.6357, 2.9393, 4.9672, 3.9567, 1.0376, 3.5375], device='cuda:3'), covar=tensor([0.2243, 0.2064, 0.1215, 0.2638, 0.0742, 0.0743, 0.4498, 0.1360], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0174, 0.0160, 0.0128, 0.0156, 0.0122, 0.0145, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 11:42:15,926 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.950e+01 1.660e+02 2.003e+02 2.564e+02 3.985e+02, threshold=4.006e+02, percent-clipped=2.0 2023-03-26 11:42:30,880 INFO [finetune.py:976] (3/7) Epoch 10, batch 1400, loss[loss=0.1586, simple_loss=0.2365, pruned_loss=0.04034, over 4850.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2637, pruned_loss=0.06664, over 951834.76 frames. ], batch size: 49, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:42:48,535 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52968.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:43:03,522 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9658, 1.7026, 2.2946, 1.5171, 2.1175, 2.2003, 1.7118, 2.3908], device='cuda:3'), covar=tensor([0.1330, 0.2028, 0.1433, 0.1854, 0.0942, 0.1426, 0.2652, 0.0950], device='cuda:3'), in_proj_covar=tensor([0.0198, 0.0206, 0.0195, 0.0193, 0.0179, 0.0217, 0.0218, 0.0203], 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') 2023-03-26 11:43:10,258 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6161, 1.5765, 1.4336, 1.7889, 2.2168, 1.7963, 1.4319, 1.3495], device='cuda:3'), covar=tensor([0.2603, 0.2412, 0.2364, 0.1852, 0.1963, 0.1411, 0.2822, 0.2297], device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0208, 0.0207, 0.0188, 0.0240, 0.0180, 0.0213, 0.0194], 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') 2023-03-26 11:43:14,419 INFO [finetune.py:976] (3/7) Epoch 10, batch 1450, loss[loss=0.1587, simple_loss=0.2342, pruned_loss=0.04158, over 4757.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2653, pruned_loss=0.06709, over 952272.79 frames. ], batch size: 26, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:43:35,022 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=53016.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:43:45,123 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.022e+02 1.605e+02 1.913e+02 2.318e+02 4.347e+02, threshold=3.826e+02, percent-clipped=3.0 2023-03-26 11:43:55,916 INFO [finetune.py:976] (3/7) Epoch 10, batch 1500, loss[loss=0.2386, simple_loss=0.3045, pruned_loss=0.08632, over 4874.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2659, pruned_loss=0.06755, over 953070.27 frames. ], batch size: 34, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:44:00,687 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53056.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:44:03,379 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 2023-03-26 11:44:12,232 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0959, 1.6431, 2.5609, 3.6563, 2.6209, 2.6612, 0.8980, 2.8107], device='cuda:3'), covar=tensor([0.1565, 0.1508, 0.1183, 0.0596, 0.0754, 0.1770, 0.2060, 0.0618], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0117, 0.0134, 0.0164, 0.0101, 0.0138, 0.0126, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 11:44:16,346 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4424, 1.5205, 1.2150, 1.4611, 1.8291, 1.6582, 1.4532, 1.2589], device='cuda:3'), covar=tensor([0.0317, 0.0286, 0.0587, 0.0273, 0.0203, 0.0468, 0.0308, 0.0411], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0108, 0.0137, 0.0114, 0.0100, 0.0101, 0.0090, 0.0107], device='cuda:3'), out_proj_covar=tensor([6.9635e-05, 8.4273e-05, 1.0917e-04, 8.9080e-05, 7.8654e-05, 7.4797e-05, 6.8455e-05, 8.2266e-05], device='cuda:3') 2023-03-26 11:44:20,648 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2354, 1.7774, 2.1699, 2.0592, 1.8232, 1.8277, 2.0054, 1.9717], device='cuda:3'), covar=tensor([0.4043, 0.4917, 0.3537, 0.4821, 0.5698, 0.4369, 0.5872, 0.3565], device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0239, 0.0252, 0.0255, 0.0250, 0.0226, 0.0272, 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') 2023-03-26 11:44:26,226 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-03-26 11:44:29,473 INFO [finetune.py:976] (3/7) Epoch 10, batch 1550, loss[loss=0.175, simple_loss=0.2479, pruned_loss=0.05108, over 4813.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2656, pruned_loss=0.06742, over 950456.45 frames. ], batch size: 40, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:44:35,500 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53108.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:44:41,525 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53117.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:44:52,482 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.115e+02 1.652e+02 2.005e+02 2.543e+02 4.651e+02, threshold=4.009e+02, percent-clipped=4.0 2023-03-26 11:45:03,047 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.92 vs. limit=5.0 2023-03-26 11:45:03,285 INFO [finetune.py:976] (3/7) Epoch 10, batch 1600, loss[loss=0.1589, simple_loss=0.2198, pruned_loss=0.04899, over 4905.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2622, pruned_loss=0.06635, over 950325.53 frames. ], batch size: 46, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:45:11,203 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6451, 3.4667, 3.2180, 1.4952, 3.4869, 2.5649, 0.7265, 2.3104], device='cuda:3'), covar=tensor([0.2418, 0.2083, 0.1756, 0.3474, 0.1248, 0.1153, 0.4602, 0.1711], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0173, 0.0160, 0.0128, 0.0156, 0.0123, 0.0145, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 11:45:48,101 INFO [finetune.py:976] (3/7) Epoch 10, batch 1650, loss[loss=0.1898, simple_loss=0.2467, pruned_loss=0.06644, over 4922.00 frames. ], tot_loss[loss=0.1952, simple_loss=0.2598, pruned_loss=0.06531, over 952574.83 frames. ], batch size: 37, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:45:56,035 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53211.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 11:46:10,716 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.364e+01 1.592e+02 1.774e+02 2.189e+02 3.836e+02, threshold=3.549e+02, percent-clipped=0.0 2023-03-26 11:46:16,365 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53241.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:46:23,558 INFO [finetune.py:976] (3/7) Epoch 10, batch 1700, loss[loss=0.2273, simple_loss=0.2849, pruned_loss=0.08484, over 4854.00 frames. ], tot_loss[loss=0.1945, simple_loss=0.2588, pruned_loss=0.06511, over 953679.63 frames. ], batch size: 49, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:46:29,715 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=53259.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 11:46:34,695 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=8.17 vs. limit=5.0 2023-03-26 11:46:54,746 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0360, 1.9895, 1.8316, 2.1295, 2.7066, 2.0761, 2.0479, 1.5767], device='cuda:3'), covar=tensor([0.2297, 0.2186, 0.1953, 0.1727, 0.1916, 0.1235, 0.2231, 0.2063], device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0205, 0.0204, 0.0185, 0.0237, 0.0177, 0.0210, 0.0193], 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') 2023-03-26 11:46:54,798 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-03-26 11:46:56,426 INFO [finetune.py:976] (3/7) Epoch 10, batch 1750, loss[loss=0.2364, simple_loss=0.2911, pruned_loss=0.09089, over 4862.00 frames. ], tot_loss[loss=0.1955, simple_loss=0.2596, pruned_loss=0.06576, over 953812.48 frames. ], batch size: 44, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:46:58,856 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53302.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:47:00,084 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53304.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:47:07,270 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53315.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:47:18,948 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.172e+02 1.605e+02 1.832e+02 2.176e+02 4.638e+02, threshold=3.664e+02, percent-clipped=2.0 2023-03-26 11:47:19,819 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-03-26 11:47:29,903 INFO [finetune.py:976] (3/7) Epoch 10, batch 1800, loss[loss=0.235, simple_loss=0.305, pruned_loss=0.08251, over 4764.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2635, pruned_loss=0.06636, over 953229.54 frames. ], batch size: 54, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:47:45,418 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53365.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:47:46,647 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53367.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:47:56,038 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53376.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:48:26,361 INFO [finetune.py:976] (3/7) Epoch 10, batch 1850, loss[loss=0.2215, simple_loss=0.289, pruned_loss=0.077, over 4811.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.265, pruned_loss=0.06724, over 953619.28 frames. ], batch size: 40, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:48:32,715 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53408.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:48:34,546 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4863, 1.6381, 1.2372, 1.5608, 1.9015, 1.7330, 1.4748, 1.3431], device='cuda:3'), covar=tensor([0.0342, 0.0272, 0.0573, 0.0251, 0.0197, 0.0452, 0.0338, 0.0399], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0108, 0.0138, 0.0114, 0.0101, 0.0101, 0.0091, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.0092e-05, 8.4746e-05, 1.1027e-04, 8.9491e-05, 7.9184e-05, 7.5116e-05, 6.9098e-05, 8.3071e-05], device='cuda:3') 2023-03-26 11:48:35,084 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53412.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:48:40,191 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-03-26 11:48:51,320 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53428.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 11:48:58,641 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.045e+02 1.752e+02 2.111e+02 2.637e+02 7.323e+02, threshold=4.222e+02, percent-clipped=6.0 2023-03-26 11:48:59,410 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8315, 1.7500, 1.5577, 1.6531, 2.0806, 2.0354, 1.6338, 1.5852], device='cuda:3'), covar=tensor([0.0327, 0.0273, 0.0518, 0.0284, 0.0196, 0.0399, 0.0377, 0.0423], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0108, 0.0138, 0.0114, 0.0101, 0.0101, 0.0091, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.0097e-05, 8.4558e-05, 1.1027e-04, 8.9358e-05, 7.9128e-05, 7.5003e-05, 6.9041e-05, 8.2979e-05], device='cuda:3') 2023-03-26 11:49:10,458 INFO [finetune.py:976] (3/7) Epoch 10, batch 1900, loss[loss=0.1888, simple_loss=0.2615, pruned_loss=0.05805, over 4755.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2651, pruned_loss=0.0666, over 954134.52 frames. ], batch size: 27, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:49:14,808 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=53456.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:49:16,573 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6521, 1.3114, 2.3003, 3.1312, 2.2127, 2.3859, 1.1811, 2.4978], device='cuda:3'), covar=tensor([0.1867, 0.1843, 0.1328, 0.0902, 0.0822, 0.1806, 0.1972, 0.0757], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0117, 0.0133, 0.0164, 0.0102, 0.0138, 0.0126, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 11:49:23,211 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.1471, 4.6273, 4.4633, 2.7793, 4.7554, 3.5858, 1.0419, 3.2410], device='cuda:3'), covar=tensor([0.2056, 0.1629, 0.1228, 0.2614, 0.0693, 0.0875, 0.4362, 0.1289], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0174, 0.0161, 0.0129, 0.0157, 0.0123, 0.0146, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 11:49:43,866 INFO [finetune.py:976] (3/7) Epoch 10, batch 1950, loss[loss=0.1625, simple_loss=0.224, pruned_loss=0.05048, over 4826.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2641, pruned_loss=0.06593, over 957073.08 frames. ], batch size: 38, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:49:51,113 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1106, 1.3781, 1.0967, 1.3841, 1.4551, 2.5069, 1.1590, 1.4317], device='cuda:3'), covar=tensor([0.1058, 0.1749, 0.1091, 0.0929, 0.1658, 0.0363, 0.1574, 0.1686], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0081, 0.0076, 0.0078, 0.0091, 0.0082, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 11:50:09,884 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.065e+02 1.535e+02 1.778e+02 2.101e+02 3.650e+02, threshold=3.555e+02, percent-clipped=0.0 2023-03-26 11:50:29,331 INFO [finetune.py:976] (3/7) Epoch 10, batch 2000, loss[loss=0.168, simple_loss=0.2259, pruned_loss=0.05503, over 4785.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.2609, pruned_loss=0.06547, over 957165.44 frames. ], batch size: 25, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:51:21,831 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53597.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:51:23,504 INFO [finetune.py:976] (3/7) Epoch 10, batch 2050, loss[loss=0.1987, simple_loss=0.25, pruned_loss=0.07375, over 4852.00 frames. ], tot_loss[loss=0.1928, simple_loss=0.2574, pruned_loss=0.06412, over 956465.62 frames. ], batch size: 44, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:51:35,978 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-03-26 11:51:44,832 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.079e+02 1.628e+02 1.908e+02 2.249e+02 5.707e+02, threshold=3.816e+02, percent-clipped=1.0 2023-03-26 11:51:56,176 INFO [finetune.py:976] (3/7) Epoch 10, batch 2100, loss[loss=0.1554, simple_loss=0.2214, pruned_loss=0.04473, over 4829.00 frames. ], tot_loss[loss=0.1934, simple_loss=0.2575, pruned_loss=0.06464, over 955676.84 frames. ], batch size: 25, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:52:03,969 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53660.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:52:13,508 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53671.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:52:37,544 INFO [finetune.py:976] (3/7) Epoch 10, batch 2150, loss[loss=0.2285, simple_loss=0.298, pruned_loss=0.07954, over 4799.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2622, pruned_loss=0.06728, over 952460.64 frames. ], batch size: 45, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:52:52,733 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53712.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:53:04,002 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53723.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 11:53:19,766 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.135e+02 1.764e+02 2.057e+02 2.459e+02 5.535e+02, threshold=4.114e+02, percent-clipped=2.0 2023-03-26 11:53:34,458 INFO [finetune.py:976] (3/7) Epoch 10, batch 2200, loss[loss=0.2512, simple_loss=0.2979, pruned_loss=0.1023, over 4795.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2642, pruned_loss=0.06812, over 951368.94 frames. ], batch size: 45, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:53:43,155 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=53760.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:54:07,987 INFO [finetune.py:976] (3/7) Epoch 10, batch 2250, loss[loss=0.2505, simple_loss=0.3107, pruned_loss=0.09511, over 4183.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2655, pruned_loss=0.06827, over 951643.16 frames. ], batch size: 65, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:54:30,209 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.107e+02 1.658e+02 1.958e+02 2.430e+02 3.560e+02, threshold=3.915e+02, percent-clipped=0.0 2023-03-26 11:54:36,613 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.61 vs. limit=5.0 2023-03-26 11:54:41,557 INFO [finetune.py:976] (3/7) Epoch 10, batch 2300, loss[loss=0.1893, simple_loss=0.2504, pruned_loss=0.06416, over 4744.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2656, pruned_loss=0.0675, over 949436.64 frames. ], batch size: 23, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:54:42,824 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6860, 1.3163, 1.0710, 1.7258, 2.0967, 1.4691, 1.5778, 1.6568], device='cuda:3'), covar=tensor([0.1402, 0.1857, 0.1884, 0.1072, 0.1874, 0.2117, 0.1313, 0.1810], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0094, 0.0111, 0.0091, 0.0120, 0.0094, 0.0098, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 11:54:47,642 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1321, 1.9902, 1.8355, 1.8313, 2.1486, 1.8325, 2.2842, 2.1396], device='cuda:3'), covar=tensor([0.1437, 0.2304, 0.3229, 0.2828, 0.2752, 0.1851, 0.2930, 0.1935], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0188, 0.0232, 0.0253, 0.0239, 0.0195, 0.0212, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 11:55:15,980 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53897.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:55:17,111 INFO [finetune.py:976] (3/7) Epoch 10, batch 2350, loss[loss=0.1707, simple_loss=0.2355, pruned_loss=0.0529, over 4822.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2631, pruned_loss=0.06641, over 951801.49 frames. ], batch size: 30, lr: 3.75e-03, grad_scale: 32.0 2023-03-26 11:55:37,889 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6459, 1.5380, 1.4880, 1.4502, 1.7409, 1.4576, 1.7707, 1.6316], device='cuda:3'), covar=tensor([0.1274, 0.1859, 0.2485, 0.2300, 0.2033, 0.1437, 0.2423, 0.1573], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0188, 0.0232, 0.0253, 0.0239, 0.0195, 0.0213, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 11:55:47,268 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.820e+01 1.627e+02 1.969e+02 2.442e+02 4.599e+02, threshold=3.938e+02, percent-clipped=2.0 2023-03-26 11:55:58,277 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=53945.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:56:05,532 INFO [finetune.py:976] (3/7) Epoch 10, batch 2400, loss[loss=0.224, simple_loss=0.2823, pruned_loss=0.08285, over 4942.00 frames. ], tot_loss[loss=0.196, simple_loss=0.2607, pruned_loss=0.06571, over 953808.12 frames. ], batch size: 33, lr: 3.74e-03, grad_scale: 32.0 2023-03-26 11:56:15,960 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53960.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:56:24,610 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53971.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:56:29,976 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7477, 1.6657, 2.1429, 2.0909, 1.8695, 4.5111, 1.4827, 1.8870], device='cuda:3'), covar=tensor([0.1051, 0.1849, 0.1141, 0.1010, 0.1560, 0.0221, 0.1648, 0.1762], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0081, 0.0076, 0.0078, 0.0091, 0.0083, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 11:56:41,931 INFO [finetune.py:976] (3/7) Epoch 10, batch 2450, loss[loss=0.1838, simple_loss=0.2385, pruned_loss=0.06461, over 4784.00 frames. ], tot_loss[loss=0.1939, simple_loss=0.2579, pruned_loss=0.06498, over 956056.05 frames. ], batch size: 29, lr: 3.74e-03, grad_scale: 64.0 2023-03-26 11:56:49,146 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=54008.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:56:56,921 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=54019.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:56:59,887 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=54023.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 11:57:05,184 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.081e+02 1.615e+02 1.942e+02 2.268e+02 4.833e+02, threshold=3.884e+02, percent-clipped=2.0 2023-03-26 11:57:16,019 INFO [finetune.py:976] (3/7) Epoch 10, batch 2500, loss[loss=0.1748, simple_loss=0.2461, pruned_loss=0.05176, over 4763.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2597, pruned_loss=0.06677, over 956638.87 frames. ], batch size: 28, lr: 3.74e-03, grad_scale: 64.0 2023-03-26 11:57:42,137 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=54071.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 11:57:42,188 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5256, 1.4050, 1.4778, 1.4521, 1.1370, 2.8772, 1.0967, 1.6222], device='cuda:3'), covar=tensor([0.3092, 0.2285, 0.1897, 0.2213, 0.1670, 0.0252, 0.2863, 0.1202], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0120, 0.0123, 0.0116, 0.0098, 0.0099, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 11:57:57,183 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4081, 1.4170, 1.2322, 1.3982, 1.6771, 1.5561, 1.3634, 1.1954], device='cuda:3'), covar=tensor([0.0334, 0.0264, 0.0523, 0.0293, 0.0206, 0.0404, 0.0338, 0.0362], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0108, 0.0138, 0.0114, 0.0101, 0.0102, 0.0091, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.0725e-05, 8.4521e-05, 1.1015e-04, 8.9557e-05, 7.9564e-05, 7.5989e-05, 6.9198e-05, 8.2826e-05], device='cuda:3') 2023-03-26 11:58:00,116 INFO [finetune.py:976] (3/7) Epoch 10, batch 2550, loss[loss=0.1735, simple_loss=0.2529, pruned_loss=0.04702, over 4898.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2635, pruned_loss=0.06777, over 955353.10 frames. ], batch size: 36, lr: 3.74e-03, grad_scale: 64.0 2023-03-26 11:58:00,251 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3952, 2.4472, 2.0979, 1.9249, 2.6928, 2.8650, 2.4031, 2.2700], device='cuda:3'), covar=tensor([0.0291, 0.0289, 0.0405, 0.0362, 0.0217, 0.0385, 0.0417, 0.0310], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0108, 0.0138, 0.0114, 0.0101, 0.0102, 0.0091, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.0735e-05, 8.4534e-05, 1.1011e-04, 8.9561e-05, 7.9563e-05, 7.6006e-05, 6.9174e-05, 8.2810e-05], device='cuda:3') 2023-03-26 11:58:22,512 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-03-26 11:58:23,927 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7202, 1.4982, 2.1062, 1.3870, 1.9275, 2.0811, 1.4475, 2.3235], device='cuda:3'), covar=tensor([0.1535, 0.2154, 0.1401, 0.2101, 0.0875, 0.1513, 0.2881, 0.0906], device='cuda:3'), in_proj_covar=tensor([0.0198, 0.0203, 0.0192, 0.0191, 0.0177, 0.0215, 0.0217, 0.0201], 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') 2023-03-26 11:58:35,815 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.936e+01 1.671e+02 2.051e+02 2.356e+02 3.900e+02, threshold=4.103e+02, percent-clipped=1.0 2023-03-26 11:58:46,743 INFO [finetune.py:976] (3/7) Epoch 10, batch 2600, loss[loss=0.2274, simple_loss=0.29, pruned_loss=0.08241, over 4843.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2647, pruned_loss=0.06751, over 955139.02 frames. ], batch size: 47, lr: 3.74e-03, grad_scale: 64.0 2023-03-26 11:59:06,377 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1451, 1.9723, 1.8168, 2.0882, 2.6365, 2.1093, 1.9023, 1.6964], device='cuda:3'), covar=tensor([0.2113, 0.2111, 0.1927, 0.1648, 0.1874, 0.1203, 0.2268, 0.1906], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0208, 0.0208, 0.0189, 0.0241, 0.0180, 0.0213, 0.0196], 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') 2023-03-26 11:59:08,317 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.96 vs. limit=5.0 2023-03-26 11:59:19,472 INFO [finetune.py:976] (3/7) Epoch 10, batch 2650, loss[loss=0.1962, simple_loss=0.2712, pruned_loss=0.06065, over 4687.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2666, pruned_loss=0.06779, over 957564.21 frames. ], batch size: 59, lr: 3.74e-03, grad_scale: 64.0 2023-03-26 11:59:43,742 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.585e+01 1.566e+02 1.779e+02 2.159e+02 3.883e+02, threshold=3.557e+02, percent-clipped=0.0 2023-03-26 11:59:45,087 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5973, 2.8841, 2.6417, 1.8057, 2.8073, 3.0169, 2.9662, 2.5652], device='cuda:3'), covar=tensor([0.0684, 0.0555, 0.0744, 0.1014, 0.0573, 0.0727, 0.0597, 0.0930], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0133, 0.0143, 0.0124, 0.0119, 0.0142, 0.0142, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 11:59:53,473 INFO [finetune.py:976] (3/7) Epoch 10, batch 2700, loss[loss=0.1951, simple_loss=0.2593, pruned_loss=0.06543, over 4936.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2656, pruned_loss=0.06697, over 956729.44 frames. ], batch size: 33, lr: 3.74e-03, grad_scale: 32.0 2023-03-26 12:00:20,006 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.2867, 4.5057, 4.8491, 5.1181, 4.9768, 4.7382, 5.3850, 1.6493], device='cuda:3'), covar=tensor([0.0727, 0.0866, 0.0811, 0.0877, 0.1346, 0.1577, 0.0635, 0.5729], device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0243, 0.0274, 0.0287, 0.0325, 0.0279, 0.0299, 0.0291], 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') 2023-03-26 12:00:26,576 INFO [finetune.py:976] (3/7) Epoch 10, batch 2750, loss[loss=0.1291, simple_loss=0.2033, pruned_loss=0.02744, over 4793.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2621, pruned_loss=0.06582, over 957301.51 frames. ], batch size: 29, lr: 3.74e-03, grad_scale: 32.0 2023-03-26 12:00:50,902 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.123e+02 1.545e+02 1.929e+02 2.415e+02 3.548e+02, threshold=3.859e+02, percent-clipped=0.0 2023-03-26 12:00:56,305 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0061, 1.9947, 2.2202, 1.2979, 2.1955, 2.2063, 2.1242, 1.7426], device='cuda:3'), covar=tensor([0.0632, 0.0659, 0.0607, 0.0918, 0.0533, 0.0659, 0.0564, 0.0982], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0132, 0.0142, 0.0123, 0.0118, 0.0141, 0.0141, 0.0160], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 12:00:56,332 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0303, 1.9839, 1.5252, 1.7858, 2.0744, 1.6883, 2.5864, 2.0164], device='cuda:3'), covar=tensor([0.1550, 0.2277, 0.3589, 0.3388, 0.2887, 0.1827, 0.2681, 0.2117], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0188, 0.0233, 0.0253, 0.0239, 0.0195, 0.0213, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 12:01:01,567 INFO [finetune.py:976] (3/7) Epoch 10, batch 2800, loss[loss=0.208, simple_loss=0.264, pruned_loss=0.07598, over 4905.00 frames. ], tot_loss[loss=0.1943, simple_loss=0.2589, pruned_loss=0.06489, over 959116.14 frames. ], batch size: 35, lr: 3.74e-03, grad_scale: 32.0 2023-03-26 12:01:48,148 INFO [finetune.py:976] (3/7) Epoch 10, batch 2850, loss[loss=0.2493, simple_loss=0.3086, pruned_loss=0.09499, over 4755.00 frames. ], tot_loss[loss=0.1925, simple_loss=0.2571, pruned_loss=0.06396, over 956431.74 frames. ], batch size: 54, lr: 3.74e-03, grad_scale: 32.0 2023-03-26 12:01:59,565 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-26 12:02:10,452 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.958e+01 1.628e+02 1.894e+02 2.190e+02 3.699e+02, threshold=3.787e+02, percent-clipped=0.0 2023-03-26 12:02:16,323 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7507, 1.0021, 1.7430, 1.6585, 1.5008, 1.4341, 1.5215, 1.6095], device='cuda:3'), covar=tensor([0.4417, 0.4669, 0.3881, 0.4118, 0.5387, 0.4098, 0.5155, 0.3777], device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0239, 0.0252, 0.0256, 0.0251, 0.0227, 0.0273, 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') 2023-03-26 12:02:22,196 INFO [finetune.py:976] (3/7) Epoch 10, batch 2900, loss[loss=0.2013, simple_loss=0.2748, pruned_loss=0.06395, over 4825.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2617, pruned_loss=0.06613, over 954695.71 frames. ], batch size: 40, lr: 3.74e-03, grad_scale: 32.0 2023-03-26 12:02:28,997 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5571, 1.3578, 1.9067, 3.0958, 2.1156, 2.1634, 0.7531, 2.3737], device='cuda:3'), covar=tensor([0.1923, 0.1802, 0.1622, 0.0870, 0.0885, 0.1632, 0.2288, 0.0747], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0117, 0.0134, 0.0164, 0.0101, 0.0138, 0.0126, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 12:02:42,528 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.63 vs. limit=5.0 2023-03-26 12:02:57,316 INFO [finetune.py:976] (3/7) Epoch 10, batch 2950, loss[loss=0.2253, simple_loss=0.2885, pruned_loss=0.08103, over 4755.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2646, pruned_loss=0.06657, over 953624.05 frames. ], batch size: 54, lr: 3.74e-03, grad_scale: 32.0 2023-03-26 12:03:18,744 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.146e+02 1.694e+02 2.010e+02 2.318e+02 4.609e+02, threshold=4.019e+02, percent-clipped=2.0 2023-03-26 12:03:27,584 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0855, 1.9094, 1.6635, 1.8914, 1.7955, 1.8042, 1.8251, 2.5794], device='cuda:3'), covar=tensor([0.4943, 0.5556, 0.4122, 0.5229, 0.5109, 0.2890, 0.4839, 0.2100], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0259, 0.0222, 0.0279, 0.0243, 0.0208, 0.0246, 0.0211], 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') 2023-03-26 12:03:39,999 INFO [finetune.py:976] (3/7) Epoch 10, batch 3000, loss[loss=0.1953, simple_loss=0.2765, pruned_loss=0.05706, over 4802.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2658, pruned_loss=0.06712, over 953916.64 frames. ], batch size: 51, lr: 3.74e-03, grad_scale: 32.0 2023-03-26 12:03:39,999 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 12:03:47,446 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.6715, 3.2208, 3.4235, 3.5478, 3.4183, 3.3013, 3.7122, 1.4892], device='cuda:3'), covar=tensor([0.0811, 0.0739, 0.0749, 0.0936, 0.1157, 0.1247, 0.0659, 0.4489], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0246, 0.0278, 0.0291, 0.0331, 0.0283, 0.0303, 0.0296], 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') 2023-03-26 12:03:56,628 INFO [finetune.py:1010] (3/7) Epoch 10, validation: loss=0.1584, simple_loss=0.2295, pruned_loss=0.04366, over 2265189.00 frames. 2023-03-26 12:03:56,629 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 12:03:59,553 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.08 vs. limit=5.0 2023-03-26 12:04:04,965 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9719, 1.6355, 2.4444, 3.8985, 2.6935, 2.6158, 0.6857, 3.0465], device='cuda:3'), covar=tensor([0.1813, 0.1710, 0.1445, 0.0567, 0.0796, 0.1521, 0.2231, 0.0533], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0118, 0.0135, 0.0166, 0.0102, 0.0139, 0.0127, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 12:04:29,066 INFO [finetune.py:976] (3/7) Epoch 10, batch 3050, loss[loss=0.1892, simple_loss=0.2658, pruned_loss=0.05625, over 4825.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.266, pruned_loss=0.0667, over 954157.44 frames. ], batch size: 39, lr: 3.74e-03, grad_scale: 32.0 2023-03-26 12:04:52,093 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.154e+02 1.606e+02 1.839e+02 2.259e+02 4.011e+02, threshold=3.679e+02, percent-clipped=0.0 2023-03-26 12:05:02,811 INFO [finetune.py:976] (3/7) Epoch 10, batch 3100, loss[loss=0.2542, simple_loss=0.3139, pruned_loss=0.09719, over 4861.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2645, pruned_loss=0.06666, over 953717.81 frames. ], batch size: 44, lr: 3.74e-03, grad_scale: 32.0 2023-03-26 12:05:13,587 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-26 12:05:14,426 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54664.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 12:05:36,461 INFO [finetune.py:976] (3/7) Epoch 10, batch 3150, loss[loss=0.1465, simple_loss=0.2175, pruned_loss=0.03771, over 4795.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2617, pruned_loss=0.06612, over 954008.20 frames. ], batch size: 25, lr: 3.74e-03, grad_scale: 32.0 2023-03-26 12:05:41,531 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-26 12:05:54,667 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54725.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 12:05:59,383 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.009e+02 1.653e+02 1.993e+02 2.298e+02 5.311e+02, threshold=3.986e+02, percent-clipped=2.0 2023-03-26 12:06:10,113 INFO [finetune.py:976] (3/7) Epoch 10, batch 3200, loss[loss=0.1882, simple_loss=0.2509, pruned_loss=0.0628, over 4912.00 frames. ], tot_loss[loss=0.1936, simple_loss=0.2577, pruned_loss=0.06479, over 954469.38 frames. ], batch size: 36, lr: 3.74e-03, grad_scale: 32.0 2023-03-26 12:06:24,261 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.36 vs. limit=5.0 2023-03-26 12:06:53,305 INFO [finetune.py:976] (3/7) Epoch 10, batch 3250, loss[loss=0.2113, simple_loss=0.2778, pruned_loss=0.07241, over 4855.00 frames. ], tot_loss[loss=0.1946, simple_loss=0.2583, pruned_loss=0.06547, over 956314.16 frames. ], batch size: 44, lr: 3.74e-03, grad_scale: 32.0 2023-03-26 12:07:26,174 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.100e+02 1.711e+02 2.094e+02 2.546e+02 5.601e+02, threshold=4.189e+02, percent-clipped=2.0 2023-03-26 12:07:34,852 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.63 vs. limit=5.0 2023-03-26 12:07:46,365 INFO [finetune.py:976] (3/7) Epoch 10, batch 3300, loss[loss=0.2209, simple_loss=0.2825, pruned_loss=0.07962, over 4809.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2624, pruned_loss=0.06708, over 956784.23 frames. ], batch size: 51, lr: 3.74e-03, grad_scale: 32.0 2023-03-26 12:07:55,839 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54862.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:08:18,162 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.74 vs. limit=5.0 2023-03-26 12:08:20,112 INFO [finetune.py:976] (3/7) Epoch 10, batch 3350, loss[loss=0.1955, simple_loss=0.2721, pruned_loss=0.05946, over 4824.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2649, pruned_loss=0.06797, over 955483.49 frames. ], batch size: 39, lr: 3.74e-03, grad_scale: 32.0 2023-03-26 12:08:48,180 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54923.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 12:08:57,773 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.164e+02 1.693e+02 1.965e+02 2.442e+02 4.084e+02, threshold=3.930e+02, percent-clipped=0.0 2023-03-26 12:08:59,164 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.56 vs. limit=5.0 2023-03-26 12:09:03,964 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8340, 1.3521, 0.9862, 1.7598, 2.1440, 1.4522, 1.6982, 1.6459], device='cuda:3'), covar=tensor([0.1438, 0.2087, 0.1952, 0.1146, 0.1909, 0.2015, 0.1401, 0.2072], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0096, 0.0113, 0.0092, 0.0122, 0.0095, 0.0099, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 12:09:07,544 INFO [finetune.py:976] (3/7) Epoch 10, batch 3400, loss[loss=0.2208, simple_loss=0.301, pruned_loss=0.07023, over 4891.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2665, pruned_loss=0.06806, over 954880.63 frames. ], batch size: 37, lr: 3.74e-03, grad_scale: 32.0 2023-03-26 12:09:43,489 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5346, 1.3912, 1.9031, 1.0895, 1.5448, 1.6808, 1.3245, 1.9818], device='cuda:3'), covar=tensor([0.1424, 0.2304, 0.1183, 0.1906, 0.1177, 0.1558, 0.2911, 0.1041], device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0203, 0.0192, 0.0191, 0.0177, 0.0213, 0.0216, 0.0199], 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') 2023-03-26 12:09:56,944 INFO [finetune.py:976] (3/7) Epoch 10, batch 3450, loss[loss=0.1803, simple_loss=0.2538, pruned_loss=0.05341, over 4764.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2656, pruned_loss=0.06714, over 953914.06 frames. ], batch size: 28, lr: 3.74e-03, grad_scale: 32.0 2023-03-26 12:09:57,141 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-26 12:10:16,049 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55020.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 12:10:16,677 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0954, 1.3537, 0.6374, 2.1016, 2.4363, 1.9074, 1.6680, 1.8833], device='cuda:3'), covar=tensor([0.1433, 0.2102, 0.2276, 0.1121, 0.1836, 0.1787, 0.1442, 0.1993], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0096, 0.0113, 0.0092, 0.0122, 0.0095, 0.0099, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 12:10:30,757 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.349e+01 1.534e+02 1.957e+02 2.350e+02 5.428e+02, threshold=3.914e+02, percent-clipped=3.0 2023-03-26 12:10:51,432 INFO [finetune.py:976] (3/7) Epoch 10, batch 3500, loss[loss=0.1893, simple_loss=0.2465, pruned_loss=0.06606, over 4817.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2632, pruned_loss=0.06645, over 954203.19 frames. ], batch size: 41, lr: 3.74e-03, grad_scale: 32.0 2023-03-26 12:11:21,087 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9771, 1.1959, 1.8951, 1.8117, 1.6581, 1.5699, 1.7765, 1.7615], device='cuda:3'), covar=tensor([0.3776, 0.4641, 0.4023, 0.4452, 0.5376, 0.4336, 0.5109, 0.3890], device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0238, 0.0251, 0.0256, 0.0250, 0.0226, 0.0273, 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') 2023-03-26 12:11:28,584 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0976, 1.9616, 1.6901, 2.0163, 2.1410, 1.7897, 2.3741, 2.1098], device='cuda:3'), covar=tensor([0.1330, 0.2273, 0.3085, 0.2781, 0.2329, 0.1616, 0.3208, 0.1864], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0188, 0.0234, 0.0255, 0.0240, 0.0197, 0.0213, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 12:11:36,219 INFO [finetune.py:976] (3/7) Epoch 10, batch 3550, loss[loss=0.1678, simple_loss=0.2493, pruned_loss=0.04319, over 4838.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.2607, pruned_loss=0.0655, over 955867.25 frames. ], batch size: 30, lr: 3.74e-03, grad_scale: 32.0 2023-03-26 12:11:40,542 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4736, 1.3325, 1.3893, 1.3417, 0.8101, 2.1935, 0.7552, 1.3166], device='cuda:3'), covar=tensor([0.3337, 0.2525, 0.2147, 0.2392, 0.2102, 0.0379, 0.2698, 0.1367], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0120, 0.0123, 0.0115, 0.0098, 0.0099, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 12:11:48,791 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.5826, 3.9636, 4.2409, 4.3804, 4.2671, 4.1540, 4.6619, 1.9251], device='cuda:3'), covar=tensor([0.0850, 0.0915, 0.0788, 0.1046, 0.1407, 0.1450, 0.0736, 0.5081], device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0244, 0.0275, 0.0290, 0.0329, 0.0283, 0.0301, 0.0296], 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') 2023-03-26 12:11:54,302 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-26 12:11:58,574 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.369e+01 1.530e+02 1.896e+02 2.280e+02 4.793e+02, threshold=3.791e+02, percent-clipped=5.0 2023-03-26 12:12:08,654 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0087, 1.8965, 1.5092, 0.7767, 1.7179, 1.7438, 1.4952, 1.8533], device='cuda:3'), covar=tensor([0.0671, 0.0514, 0.0834, 0.1215, 0.0868, 0.1389, 0.1432, 0.0551], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0200, 0.0200, 0.0186, 0.0215, 0.0206, 0.0223, 0.0196], 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') 2023-03-26 12:12:09,240 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55148.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:12:09,742 INFO [finetune.py:976] (3/7) Epoch 10, batch 3600, loss[loss=0.2082, simple_loss=0.2732, pruned_loss=0.07162, over 4860.00 frames. ], tot_loss[loss=0.194, simple_loss=0.2584, pruned_loss=0.06484, over 955489.27 frames. ], batch size: 34, lr: 3.74e-03, grad_scale: 32.0 2023-03-26 12:12:16,648 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.63 vs. limit=5.0 2023-03-26 12:12:26,013 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9340, 1.8552, 1.5533, 1.9496, 1.9865, 1.6379, 2.2960, 1.9495], device='cuda:3'), covar=tensor([0.1425, 0.2290, 0.3316, 0.2620, 0.2640, 0.1761, 0.3281, 0.1985], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0188, 0.0234, 0.0254, 0.0239, 0.0197, 0.0212, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 12:12:43,425 INFO [finetune.py:976] (3/7) Epoch 10, batch 3650, loss[loss=0.2727, simple_loss=0.3274, pruned_loss=0.109, over 4803.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2614, pruned_loss=0.06607, over 955831.82 frames. ], batch size: 40, lr: 3.74e-03, grad_scale: 32.0 2023-03-26 12:12:49,690 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55209.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:12:52,579 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55211.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:12:56,770 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55218.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 12:13:14,908 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.143e+02 1.616e+02 1.938e+02 2.270e+02 4.700e+02, threshold=3.875e+02, percent-clipped=1.0 2023-03-26 12:13:19,651 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.19 vs. limit=5.0 2023-03-26 12:13:26,510 INFO [finetune.py:976] (3/7) Epoch 10, batch 3700, loss[loss=0.2337, simple_loss=0.3103, pruned_loss=0.07857, over 4839.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2651, pruned_loss=0.06712, over 956157.36 frames. ], batch size: 49, lr: 3.73e-03, grad_scale: 32.0 2023-03-26 12:13:40,331 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55272.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:13:45,049 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5646, 1.4717, 1.7103, 1.7498, 1.5269, 3.3307, 1.3890, 1.5531], device='cuda:3'), covar=tensor([0.0948, 0.1811, 0.1117, 0.1002, 0.1612, 0.0278, 0.1522, 0.1720], device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0082, 0.0076, 0.0078, 0.0092, 0.0082, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 12:13:46,754 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3929, 1.2838, 1.2700, 1.3139, 0.9624, 2.3032, 0.8693, 1.3704], device='cuda:3'), covar=tensor([0.4051, 0.3045, 0.2422, 0.2956, 0.1913, 0.0479, 0.2636, 0.1256], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0115, 0.0120, 0.0123, 0.0115, 0.0098, 0.0099, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 12:14:00,007 INFO [finetune.py:976] (3/7) Epoch 10, batch 3750, loss[loss=0.2423, simple_loss=0.3031, pruned_loss=0.09069, over 4914.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2655, pruned_loss=0.06697, over 954667.72 frames. ], batch size: 37, lr: 3.73e-03, grad_scale: 32.0 2023-03-26 12:14:16,754 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55320.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 12:14:33,808 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.113e+02 1.583e+02 1.835e+02 2.150e+02 3.880e+02, threshold=3.669e+02, percent-clipped=1.0 2023-03-26 12:14:45,562 INFO [finetune.py:976] (3/7) Epoch 10, batch 3800, loss[loss=0.1978, simple_loss=0.2644, pruned_loss=0.06564, over 4767.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2672, pruned_loss=0.06771, over 954911.44 frames. ], batch size: 28, lr: 3.73e-03, grad_scale: 32.0 2023-03-26 12:14:57,815 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=55368.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 12:15:06,925 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.38 vs. limit=5.0 2023-03-26 12:15:27,044 INFO [finetune.py:976] (3/7) Epoch 10, batch 3850, loss[loss=0.2051, simple_loss=0.2724, pruned_loss=0.06891, over 4917.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2642, pruned_loss=0.06616, over 954770.12 frames. ], batch size: 37, lr: 3.73e-03, grad_scale: 32.0 2023-03-26 12:15:49,877 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.001e+02 1.578e+02 1.920e+02 2.344e+02 4.809e+02, threshold=3.839e+02, percent-clipped=3.0 2023-03-26 12:16:01,525 INFO [finetune.py:976] (3/7) Epoch 10, batch 3900, loss[loss=0.179, simple_loss=0.2345, pruned_loss=0.06179, over 4078.00 frames. ], tot_loss[loss=0.1946, simple_loss=0.2601, pruned_loss=0.06458, over 953846.10 frames. ], batch size: 17, lr: 3.73e-03, grad_scale: 32.0 2023-03-26 12:16:40,090 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4520, 2.3137, 1.8204, 2.4996, 2.3670, 2.0722, 2.9146, 2.4363], device='cuda:3'), covar=tensor([0.1363, 0.2699, 0.3297, 0.2851, 0.2723, 0.1678, 0.3221, 0.2131], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0187, 0.0232, 0.0252, 0.0239, 0.0195, 0.0211, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 12:16:44,730 INFO [finetune.py:976] (3/7) Epoch 10, batch 3950, loss[loss=0.1587, simple_loss=0.2375, pruned_loss=0.03998, over 4882.00 frames. ], tot_loss[loss=0.1928, simple_loss=0.2575, pruned_loss=0.064, over 954735.11 frames. ], batch size: 35, lr: 3.73e-03, grad_scale: 32.0 2023-03-26 12:16:48,769 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55504.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:16:58,347 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55518.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 12:17:13,740 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.106e+02 1.580e+02 1.857e+02 2.217e+02 3.906e+02, threshold=3.714e+02, percent-clipped=1.0 2023-03-26 12:17:35,772 INFO [finetune.py:976] (3/7) Epoch 10, batch 4000, loss[loss=0.1512, simple_loss=0.2092, pruned_loss=0.04662, over 4757.00 frames. ], tot_loss[loss=0.1938, simple_loss=0.2584, pruned_loss=0.06459, over 955404.80 frames. ], batch size: 26, lr: 3.73e-03, grad_scale: 32.0 2023-03-26 12:17:48,290 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=55566.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:17:48,909 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55567.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:18:05,791 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6788, 1.5634, 1.4666, 1.5514, 1.9383, 1.8776, 1.7735, 1.4546], device='cuda:3'), covar=tensor([0.0293, 0.0284, 0.0510, 0.0296, 0.0188, 0.0413, 0.0254, 0.0349], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0109, 0.0139, 0.0113, 0.0101, 0.0103, 0.0092, 0.0107], device='cuda:3'), out_proj_covar=tensor([7.0882e-05, 8.5344e-05, 1.1061e-04, 8.8845e-05, 7.9208e-05, 7.6170e-05, 6.9650e-05, 8.2129e-05], device='cuda:3') 2023-03-26 12:18:09,102 INFO [finetune.py:976] (3/7) Epoch 10, batch 4050, loss[loss=0.2417, simple_loss=0.3066, pruned_loss=0.08839, over 4902.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2627, pruned_loss=0.06657, over 955034.95 frames. ], batch size: 36, lr: 3.73e-03, grad_scale: 32.0 2023-03-26 12:18:34,632 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.120e+02 1.736e+02 2.138e+02 2.510e+02 4.140e+02, threshold=4.276e+02, percent-clipped=4.0 2023-03-26 12:18:44,820 INFO [finetune.py:976] (3/7) Epoch 10, batch 4100, loss[loss=0.1995, simple_loss=0.2734, pruned_loss=0.06282, over 4828.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2659, pruned_loss=0.06781, over 953548.48 frames. ], batch size: 47, lr: 3.73e-03, grad_scale: 32.0 2023-03-26 12:19:00,099 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9134, 1.8399, 1.5370, 1.7679, 1.9083, 1.6228, 2.1504, 1.8595], device='cuda:3'), covar=tensor([0.1455, 0.2378, 0.3236, 0.2692, 0.2843, 0.1805, 0.3657, 0.1964], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0188, 0.0233, 0.0254, 0.0240, 0.0196, 0.0213, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 12:19:17,499 INFO [finetune.py:976] (3/7) Epoch 10, batch 4150, loss[loss=0.1678, simple_loss=0.2395, pruned_loss=0.04801, over 4745.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2664, pruned_loss=0.06809, over 953536.08 frames. ], batch size: 27, lr: 3.73e-03, grad_scale: 32.0 2023-03-26 12:19:21,780 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8691, 1.6726, 1.4534, 1.3996, 1.6177, 1.6215, 1.6388, 2.2958], device='cuda:3'), covar=tensor([0.4501, 0.4700, 0.3606, 0.4673, 0.4344, 0.2628, 0.4223, 0.1919], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0259, 0.0223, 0.0279, 0.0242, 0.0209, 0.0245, 0.0212], 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') 2023-03-26 12:19:49,997 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.152e+02 1.688e+02 2.035e+02 2.462e+02 3.895e+02, threshold=4.069e+02, percent-clipped=0.0 2023-03-26 12:19:59,114 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55748.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:19:59,613 INFO [finetune.py:976] (3/7) Epoch 10, batch 4200, loss[loss=0.1703, simple_loss=0.2366, pruned_loss=0.05201, over 4744.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.266, pruned_loss=0.06755, over 952646.35 frames. ], batch size: 59, lr: 3.73e-03, grad_scale: 32.0 2023-03-26 12:20:53,373 INFO [finetune.py:976] (3/7) Epoch 10, batch 4250, loss[loss=0.2254, simple_loss=0.2823, pruned_loss=0.08427, over 4700.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2647, pruned_loss=0.06804, over 952000.26 frames. ], batch size: 54, lr: 3.73e-03, grad_scale: 32.0 2023-03-26 12:20:57,627 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55804.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:21:06,158 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55809.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:21:38,837 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.929e+01 1.631e+02 1.908e+02 2.201e+02 4.056e+02, threshold=3.816e+02, percent-clipped=0.0 2023-03-26 12:21:48,080 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55848.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:21:48,563 INFO [finetune.py:976] (3/7) Epoch 10, batch 4300, loss[loss=0.1572, simple_loss=0.2304, pruned_loss=0.04206, over 4772.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2608, pruned_loss=0.06613, over 953883.64 frames. ], batch size: 27, lr: 3.73e-03, grad_scale: 32.0 2023-03-26 12:21:50,382 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=55852.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:22:10,850 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55867.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:22:28,661 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.6050, 4.0062, 4.1994, 4.4326, 4.3604, 4.0418, 4.6463, 1.3642], device='cuda:3'), covar=tensor([0.0681, 0.0764, 0.0815, 0.0784, 0.1022, 0.1564, 0.0595, 0.5712], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0245, 0.0277, 0.0292, 0.0331, 0.0285, 0.0302, 0.0296], 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') 2023-03-26 12:22:36,977 INFO [finetune.py:976] (3/7) Epoch 10, batch 4350, loss[loss=0.1598, simple_loss=0.2363, pruned_loss=0.04163, over 4743.00 frames. ], tot_loss[loss=0.1925, simple_loss=0.2565, pruned_loss=0.06429, over 951474.20 frames. ], batch size: 27, lr: 3.73e-03, grad_scale: 32.0 2023-03-26 12:22:47,078 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 2023-03-26 12:22:48,867 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55909.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:22:58,948 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=55915.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:23:23,253 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.076e+02 1.622e+02 1.928e+02 2.410e+02 3.855e+02, threshold=3.856e+02, percent-clipped=1.0 2023-03-26 12:23:37,721 INFO [finetune.py:976] (3/7) Epoch 10, batch 4400, loss[loss=0.1361, simple_loss=0.1922, pruned_loss=0.03995, over 4390.00 frames. ], tot_loss[loss=0.1929, simple_loss=0.2569, pruned_loss=0.06438, over 949794.23 frames. ], batch size: 18, lr: 3.73e-03, grad_scale: 32.0 2023-03-26 12:24:11,803 INFO [finetune.py:976] (3/7) Epoch 10, batch 4450, loss[loss=0.1894, simple_loss=0.2657, pruned_loss=0.0565, over 4871.00 frames. ], tot_loss[loss=0.1958, simple_loss=0.2609, pruned_loss=0.06534, over 948458.65 frames. ], batch size: 34, lr: 3.73e-03, grad_scale: 32.0 2023-03-26 12:24:22,922 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1460, 2.6632, 2.5893, 1.2127, 2.8027, 2.0972, 0.6581, 1.8094], device='cuda:3'), covar=tensor([0.2297, 0.2254, 0.1843, 0.3486, 0.1262, 0.1142, 0.3977, 0.1580], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0173, 0.0158, 0.0127, 0.0154, 0.0121, 0.0144, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 12:24:25,771 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3259, 1.5796, 1.2450, 1.3943, 1.6838, 1.5864, 1.4276, 1.3580], device='cuda:3'), covar=tensor([0.0395, 0.0253, 0.0508, 0.0289, 0.0220, 0.0400, 0.0333, 0.0342], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0110, 0.0138, 0.0114, 0.0100, 0.0102, 0.0092, 0.0106], device='cuda:3'), out_proj_covar=tensor([7.0465e-05, 8.5514e-05, 1.1005e-04, 8.9036e-05, 7.8511e-05, 7.5756e-05, 6.9311e-05, 8.1889e-05], device='cuda:3') 2023-03-26 12:24:36,683 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.054e+02 1.562e+02 1.840e+02 2.258e+02 4.729e+02, threshold=3.681e+02, percent-clipped=2.0 2023-03-26 12:24:43,415 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3176, 2.1200, 1.6050, 2.0587, 2.1298, 1.8677, 2.4575, 2.2052], device='cuda:3'), covar=tensor([0.1414, 0.2344, 0.3786, 0.3460, 0.3209, 0.1856, 0.4447, 0.2048], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0187, 0.0232, 0.0253, 0.0238, 0.0195, 0.0211, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 12:24:46,911 INFO [finetune.py:976] (3/7) Epoch 10, batch 4500, loss[loss=0.1947, simple_loss=0.2539, pruned_loss=0.06777, over 4783.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2625, pruned_loss=0.06588, over 950333.46 frames. ], batch size: 29, lr: 3.73e-03, grad_scale: 32.0 2023-03-26 12:25:11,458 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.67 vs. limit=2.0 2023-03-26 12:25:31,194 INFO [finetune.py:976] (3/7) Epoch 10, batch 4550, loss[loss=0.2227, simple_loss=0.291, pruned_loss=0.07723, over 4889.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2645, pruned_loss=0.0669, over 952039.13 frames. ], batch size: 43, lr: 3.73e-03, grad_scale: 32.0 2023-03-26 12:25:34,311 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56104.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:25:34,988 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7277, 1.5980, 1.5180, 1.8081, 2.2987, 1.8847, 1.4945, 1.4214], device='cuda:3'), covar=tensor([0.2253, 0.2135, 0.1941, 0.1708, 0.1681, 0.1182, 0.2608, 0.1989], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0209, 0.0209, 0.0190, 0.0241, 0.0181, 0.0214, 0.0196], 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') 2023-03-26 12:25:53,266 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.089e+02 1.686e+02 1.941e+02 2.447e+02 3.858e+02, threshold=3.882e+02, percent-clipped=3.0 2023-03-26 12:26:02,522 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56145.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:26:04,873 INFO [finetune.py:976] (3/7) Epoch 10, batch 4600, loss[loss=0.1589, simple_loss=0.2346, pruned_loss=0.04162, over 4775.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2648, pruned_loss=0.06743, over 951107.41 frames. ], batch size: 28, lr: 3.73e-03, grad_scale: 32.0 2023-03-26 12:26:13,535 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3504, 2.0886, 2.3510, 1.0038, 2.5477, 2.7375, 2.2097, 2.1050], device='cuda:3'), covar=tensor([0.0955, 0.0753, 0.0471, 0.0835, 0.0636, 0.0565, 0.0500, 0.0668], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0155, 0.0122, 0.0134, 0.0132, 0.0125, 0.0146, 0.0148], device='cuda:3'), out_proj_covar=tensor([9.5793e-05, 1.1404e-04, 8.7971e-05, 9.6848e-05, 9.4450e-05, 9.1361e-05, 1.0663e-04, 1.0826e-04], device='cuda:3') 2023-03-26 12:26:40,467 INFO [finetune.py:976] (3/7) Epoch 10, batch 4650, loss[loss=0.2018, simple_loss=0.2673, pruned_loss=0.06816, over 4922.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2632, pruned_loss=0.06738, over 950951.21 frames. ], batch size: 36, lr: 3.73e-03, grad_scale: 32.0 2023-03-26 12:26:40,801 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-03-26 12:26:43,723 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56204.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:26:44,972 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56206.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:27:05,765 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-26 12:27:11,819 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.027e+02 1.561e+02 1.851e+02 2.355e+02 3.865e+02, threshold=3.702e+02, percent-clipped=0.0 2023-03-26 12:27:23,137 INFO [finetune.py:976] (3/7) Epoch 10, batch 4700, loss[loss=0.1808, simple_loss=0.2519, pruned_loss=0.05483, over 4942.00 frames. ], tot_loss[loss=0.1942, simple_loss=0.2586, pruned_loss=0.0649, over 954138.40 frames. ], batch size: 33, lr: 3.73e-03, grad_scale: 64.0 2023-03-26 12:27:42,959 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-03-26 12:28:09,015 INFO [finetune.py:976] (3/7) Epoch 10, batch 4750, loss[loss=0.1993, simple_loss=0.2651, pruned_loss=0.0668, over 4823.00 frames. ], tot_loss[loss=0.1926, simple_loss=0.2568, pruned_loss=0.06418, over 955540.93 frames. ], batch size: 39, lr: 3.73e-03, grad_scale: 64.0 2023-03-26 12:28:16,667 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.49 vs. limit=5.0 2023-03-26 12:28:30,220 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.051e+02 1.456e+02 1.800e+02 2.273e+02 6.888e+02, threshold=3.601e+02, percent-clipped=2.0 2023-03-26 12:28:42,334 INFO [finetune.py:976] (3/7) Epoch 10, batch 4800, loss[loss=0.186, simple_loss=0.234, pruned_loss=0.06898, over 2971.00 frames. ], tot_loss[loss=0.1944, simple_loss=0.2593, pruned_loss=0.0648, over 953017.09 frames. ], batch size: 12, lr: 3.73e-03, grad_scale: 64.0 2023-03-26 12:29:14,946 INFO [finetune.py:976] (3/7) Epoch 10, batch 4850, loss[loss=0.2004, simple_loss=0.2663, pruned_loss=0.0673, over 4855.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2638, pruned_loss=0.06658, over 952486.79 frames. ], batch size: 44, lr: 3.73e-03, grad_scale: 64.0 2023-03-26 12:29:19,699 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56404.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:29:26,349 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0634, 1.7455, 1.5081, 1.5269, 1.7097, 1.7668, 1.7265, 2.4717], device='cuda:3'), covar=tensor([0.4575, 0.4663, 0.3865, 0.4818, 0.4262, 0.2843, 0.4289, 0.1911], device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0257, 0.0221, 0.0277, 0.0241, 0.0208, 0.0245, 0.0212], 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') 2023-03-26 12:29:37,028 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.167e+02 1.738e+02 2.004e+02 2.451e+02 5.164e+02, threshold=4.009e+02, percent-clipped=2.0 2023-03-26 12:29:48,224 INFO [finetune.py:976] (3/7) Epoch 10, batch 4900, loss[loss=0.2143, simple_loss=0.277, pruned_loss=0.07582, over 4849.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2653, pruned_loss=0.06724, over 951114.78 frames. ], batch size: 49, lr: 3.73e-03, grad_scale: 64.0 2023-03-26 12:29:50,491 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=56452.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:30:07,142 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9161, 1.7648, 1.4831, 1.6422, 1.9422, 1.6131, 2.1130, 1.8566], device='cuda:3'), covar=tensor([0.1454, 0.2562, 0.3469, 0.2943, 0.2840, 0.1776, 0.3292, 0.2041], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0187, 0.0232, 0.0252, 0.0238, 0.0195, 0.0210, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 12:30:26,420 INFO [finetune.py:976] (3/7) Epoch 10, batch 4950, loss[loss=0.181, simple_loss=0.2486, pruned_loss=0.05673, over 4897.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2664, pruned_loss=0.06743, over 952344.87 frames. ], batch size: 36, lr: 3.72e-03, grad_scale: 64.0 2023-03-26 12:30:32,444 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56501.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:30:34,842 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56504.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:30:49,790 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56523.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 12:30:55,719 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.081e+02 1.612e+02 1.965e+02 2.275e+02 4.231e+02, threshold=3.931e+02, percent-clipped=1.0 2023-03-26 12:31:06,810 INFO [finetune.py:976] (3/7) Epoch 10, batch 5000, loss[loss=0.2082, simple_loss=0.2783, pruned_loss=0.06899, over 4900.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2647, pruned_loss=0.06663, over 954339.81 frames. ], batch size: 32, lr: 3.72e-03, grad_scale: 64.0 2023-03-26 12:31:08,677 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=56552.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:31:27,804 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7444, 1.5707, 1.9087, 2.9517, 2.0847, 2.3653, 0.9677, 2.4301], device='cuda:3'), covar=tensor([0.1699, 0.1398, 0.1341, 0.0705, 0.0843, 0.1235, 0.1852, 0.0611], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0116, 0.0134, 0.0164, 0.0100, 0.0137, 0.0125, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 12:31:29,657 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56584.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 12:31:39,222 INFO [finetune.py:976] (3/7) Epoch 10, batch 5050, loss[loss=0.202, simple_loss=0.2619, pruned_loss=0.07099, over 4875.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2621, pruned_loss=0.06593, over 955264.56 frames. ], batch size: 34, lr: 3.72e-03, grad_scale: 64.0 2023-03-26 12:31:59,152 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-26 12:32:04,809 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.243e+02 1.580e+02 1.790e+02 2.049e+02 5.062e+02, threshold=3.579e+02, percent-clipped=1.0 2023-03-26 12:32:14,687 INFO [finetune.py:976] (3/7) Epoch 10, batch 5100, loss[loss=0.1721, simple_loss=0.2426, pruned_loss=0.0508, over 4753.00 frames. ], tot_loss[loss=0.1937, simple_loss=0.2585, pruned_loss=0.06442, over 955877.91 frames. ], batch size: 27, lr: 3.72e-03, grad_scale: 64.0 2023-03-26 12:32:22,428 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9753, 1.3527, 2.0011, 1.8757, 1.6613, 1.6197, 1.7388, 1.8315], device='cuda:3'), covar=tensor([0.3058, 0.3475, 0.2808, 0.3152, 0.4090, 0.3401, 0.3729, 0.2711], device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0239, 0.0252, 0.0257, 0.0252, 0.0228, 0.0273, 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') 2023-03-26 12:32:55,109 INFO [finetune.py:976] (3/7) Epoch 10, batch 5150, loss[loss=0.2024, simple_loss=0.2692, pruned_loss=0.0678, over 4876.00 frames. ], tot_loss[loss=0.1942, simple_loss=0.2587, pruned_loss=0.06488, over 955678.91 frames. ], batch size: 32, lr: 3.72e-03, grad_scale: 64.0 2023-03-26 12:33:27,186 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.919e+01 1.632e+02 1.974e+02 2.331e+02 5.610e+02, threshold=3.948e+02, percent-clipped=3.0 2023-03-26 12:33:36,882 INFO [finetune.py:976] (3/7) Epoch 10, batch 5200, loss[loss=0.1921, simple_loss=0.2599, pruned_loss=0.06214, over 4904.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2626, pruned_loss=0.06648, over 955394.85 frames. ], batch size: 37, lr: 3.72e-03, grad_scale: 64.0 2023-03-26 12:33:52,252 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.0411, 4.6189, 4.3562, 2.6263, 4.7092, 3.5115, 0.6962, 3.2059], device='cuda:3'), covar=tensor([0.2204, 0.1611, 0.1274, 0.2782, 0.0875, 0.0857, 0.4820, 0.1354], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0175, 0.0160, 0.0130, 0.0157, 0.0122, 0.0146, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 12:34:09,698 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2137, 1.3280, 1.3548, 1.4893, 1.4437, 2.9863, 1.2647, 1.4247], device='cuda:3'), covar=tensor([0.1096, 0.1785, 0.1240, 0.0972, 0.1583, 0.0317, 0.1465, 0.1731], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0081, 0.0075, 0.0078, 0.0091, 0.0082, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 12:34:10,223 INFO [finetune.py:976] (3/7) Epoch 10, batch 5250, loss[loss=0.2461, simple_loss=0.308, pruned_loss=0.09213, over 4915.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2632, pruned_loss=0.06627, over 954510.84 frames. ], batch size: 42, lr: 3.72e-03, grad_scale: 64.0 2023-03-26 12:34:11,649 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56801.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:34:12,864 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56803.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:34:32,860 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-03-26 12:34:34,253 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.164e+02 1.706e+02 2.047e+02 2.503e+02 5.084e+02, threshold=4.093e+02, percent-clipped=2.0 2023-03-26 12:34:43,966 INFO [finetune.py:976] (3/7) Epoch 10, batch 5300, loss[loss=0.182, simple_loss=0.2369, pruned_loss=0.0636, over 4401.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2658, pruned_loss=0.06741, over 954105.64 frames. ], batch size: 19, lr: 3.72e-03, grad_scale: 64.0 2023-03-26 12:34:44,025 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=56849.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:34:50,139 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9033, 1.5504, 2.0770, 1.4789, 1.9051, 1.9576, 1.4759, 2.1951], device='cuda:3'), covar=tensor([0.1415, 0.2384, 0.1618, 0.1998, 0.1143, 0.1593, 0.3140, 0.0959], device='cuda:3'), in_proj_covar=tensor([0.0199, 0.0208, 0.0196, 0.0192, 0.0179, 0.0217, 0.0220, 0.0203], 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') 2023-03-26 12:34:53,654 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56864.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:35:04,627 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56879.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 12:35:17,628 INFO [finetune.py:976] (3/7) Epoch 10, batch 5350, loss[loss=0.1614, simple_loss=0.239, pruned_loss=0.0419, over 4745.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2659, pruned_loss=0.06671, over 954772.36 frames. ], batch size: 27, lr: 3.72e-03, grad_scale: 64.0 2023-03-26 12:35:23,287 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56908.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:35:25,159 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1874, 2.0712, 2.1847, 1.6277, 2.1485, 2.2863, 2.2837, 1.8071], device='cuda:3'), covar=tensor([0.0560, 0.0620, 0.0705, 0.0807, 0.0706, 0.0672, 0.0557, 0.1051], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0135, 0.0144, 0.0126, 0.0121, 0.0144, 0.0144, 0.0163], 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') 2023-03-26 12:35:49,116 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.924e+01 1.584e+02 1.842e+02 2.192e+02 3.665e+02, threshold=3.684e+02, percent-clipped=0.0 2023-03-26 12:36:02,276 INFO [finetune.py:976] (3/7) Epoch 10, batch 5400, loss[loss=0.1822, simple_loss=0.2432, pruned_loss=0.06062, over 4831.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2627, pruned_loss=0.06555, over 952235.25 frames. ], batch size: 30, lr: 3.72e-03, grad_scale: 64.0 2023-03-26 12:36:15,013 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56969.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:36:32,489 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56993.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:36:33,661 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56995.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:36:35,990 INFO [finetune.py:976] (3/7) Epoch 10, batch 5450, loss[loss=0.2098, simple_loss=0.2729, pruned_loss=0.07332, over 4870.00 frames. ], tot_loss[loss=0.1945, simple_loss=0.2595, pruned_loss=0.06478, over 950514.03 frames. ], batch size: 31, lr: 3.72e-03, grad_scale: 64.0 2023-03-26 12:36:57,713 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.053e+02 1.491e+02 1.807e+02 2.344e+02 4.842e+02, threshold=3.613e+02, percent-clipped=5.0 2023-03-26 12:37:09,488 INFO [finetune.py:976] (3/7) Epoch 10, batch 5500, loss[loss=0.1804, simple_loss=0.2495, pruned_loss=0.0557, over 4778.00 frames. ], tot_loss[loss=0.1921, simple_loss=0.2565, pruned_loss=0.0639, over 951999.55 frames. ], batch size: 29, lr: 3.72e-03, grad_scale: 64.0 2023-03-26 12:37:11,051 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-26 12:37:12,636 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57054.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:37:13,863 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57056.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:37:14,278 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-26 12:37:39,335 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-03-26 12:37:43,350 INFO [finetune.py:976] (3/7) Epoch 10, batch 5550, loss[loss=0.2342, simple_loss=0.2993, pruned_loss=0.08454, over 4843.00 frames. ], tot_loss[loss=0.1937, simple_loss=0.2586, pruned_loss=0.0644, over 952484.16 frames. ], batch size: 49, lr: 3.72e-03, grad_scale: 64.0 2023-03-26 12:38:06,740 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.869e+01 1.587e+02 1.788e+02 2.090e+02 3.209e+02, threshold=3.576e+02, percent-clipped=0.0 2023-03-26 12:38:07,999 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7125, 1.5423, 2.2032, 3.4877, 2.4094, 2.4121, 0.8688, 2.6727], device='cuda:3'), covar=tensor([0.1743, 0.1538, 0.1356, 0.0593, 0.0781, 0.1580, 0.2055, 0.0553], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0116, 0.0133, 0.0162, 0.0100, 0.0137, 0.0125, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 12:38:25,602 INFO [finetune.py:976] (3/7) Epoch 10, batch 5600, loss[loss=0.2598, simple_loss=0.3159, pruned_loss=0.1019, over 4746.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2632, pruned_loss=0.06571, over 952404.82 frames. ], batch size: 54, lr: 3.72e-03, grad_scale: 64.0 2023-03-26 12:38:35,044 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57159.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:38:36,216 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5654, 1.5203, 1.9776, 1.9064, 1.6963, 3.6826, 1.4227, 1.7692], device='cuda:3'), covar=tensor([0.1025, 0.1788, 0.1120, 0.1015, 0.1681, 0.0273, 0.1558, 0.1651], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0082, 0.0076, 0.0078, 0.0092, 0.0082, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 12:38:46,641 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57179.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 12:38:50,739 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57186.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:38:55,940 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-03-26 12:38:57,564 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57197.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:38:58,179 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8028, 1.7272, 1.5931, 1.8656, 2.2517, 1.9316, 1.5876, 1.5058], device='cuda:3'), covar=tensor([0.2127, 0.1915, 0.1782, 0.1586, 0.1786, 0.1084, 0.2329, 0.1792], device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0206, 0.0206, 0.0187, 0.0239, 0.0179, 0.0212, 0.0195], 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') 2023-03-26 12:38:58,646 INFO [finetune.py:976] (3/7) Epoch 10, batch 5650, loss[loss=0.215, simple_loss=0.2781, pruned_loss=0.076, over 4761.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2652, pruned_loss=0.06576, over 954379.09 frames. ], batch size: 59, lr: 3.72e-03, grad_scale: 64.0 2023-03-26 12:39:01,209 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4651, 1.2881, 1.3017, 1.4604, 1.6809, 1.5523, 1.3673, 1.2471], device='cuda:3'), covar=tensor([0.0276, 0.0300, 0.0511, 0.0273, 0.0219, 0.0404, 0.0313, 0.0359], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0109, 0.0138, 0.0113, 0.0100, 0.0102, 0.0092, 0.0107], device='cuda:3'), out_proj_covar=tensor([7.0604e-05, 8.4793e-05, 1.1002e-04, 8.8828e-05, 7.8249e-05, 7.5681e-05, 6.9362e-05, 8.2376e-05], device='cuda:3') 2023-03-26 12:39:15,286 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=57227.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 12:39:19,289 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.057e+02 1.551e+02 1.804e+02 2.162e+02 3.713e+02, threshold=3.608e+02, percent-clipped=1.0 2023-03-26 12:39:22,462 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.42 vs. limit=5.0 2023-03-26 12:39:24,865 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-03-26 12:39:25,340 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57244.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:39:27,106 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57247.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:39:28,225 INFO [finetune.py:976] (3/7) Epoch 10, batch 5700, loss[loss=0.1795, simple_loss=0.2517, pruned_loss=0.05359, over 4407.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.2606, pruned_loss=0.06543, over 933849.78 frames. ], batch size: 19, lr: 3.72e-03, grad_scale: 32.0 2023-03-26 12:39:28,321 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3475, 1.9524, 2.6284, 1.8655, 2.3904, 2.4990, 2.0042, 2.6935], device='cuda:3'), covar=tensor([0.1066, 0.1992, 0.1144, 0.1625, 0.0924, 0.1237, 0.2395, 0.0743], device='cuda:3'), in_proj_covar=tensor([0.0198, 0.0206, 0.0194, 0.0191, 0.0178, 0.0217, 0.0219, 0.0202], 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') 2023-03-26 12:39:34,007 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57258.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 12:39:37,764 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57264.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:40:00,696 INFO [finetune.py:976] (3/7) Epoch 11, batch 0, loss[loss=0.2274, simple_loss=0.2801, pruned_loss=0.0874, over 4906.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.2801, pruned_loss=0.0874, over 4906.00 frames. ], batch size: 36, lr: 3.72e-03, grad_scale: 16.0 2023-03-26 12:40:00,696 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 12:40:16,056 INFO [finetune.py:1010] (3/7) Epoch 11, validation: loss=0.1597, simple_loss=0.2306, pruned_loss=0.04438, over 2265189.00 frames. 2023-03-26 12:40:16,057 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 12:40:37,191 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57305.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:40:59,549 INFO [finetune.py:976] (3/7) Epoch 11, batch 50, loss[loss=0.1689, simple_loss=0.2399, pruned_loss=0.04895, over 4810.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2644, pruned_loss=0.06699, over 217523.55 frames. ], batch size: 39, lr: 3.72e-03, grad_scale: 16.0 2023-03-26 12:41:10,026 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.036e+02 1.580e+02 1.868e+02 2.535e+02 4.204e+02, threshold=3.735e+02, percent-clipped=3.0 2023-03-26 12:41:18,571 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57349.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:41:19,797 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57351.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:41:38,107 INFO [finetune.py:976] (3/7) Epoch 11, batch 100, loss[loss=0.1474, simple_loss=0.2194, pruned_loss=0.03768, over 4756.00 frames. ], tot_loss[loss=0.1912, simple_loss=0.2565, pruned_loss=0.06296, over 380830.30 frames. ], batch size: 26, lr: 3.72e-03, grad_scale: 16.0 2023-03-26 12:41:50,239 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5224, 2.8415, 2.5336, 1.8674, 2.6589, 2.9560, 2.7913, 2.3936], device='cuda:3'), covar=tensor([0.0581, 0.0553, 0.0721, 0.0865, 0.0640, 0.0577, 0.0590, 0.0930], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0133, 0.0144, 0.0125, 0.0120, 0.0144, 0.0145, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 12:41:51,422 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57398.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:42:11,568 INFO [finetune.py:976] (3/7) Epoch 11, batch 150, loss[loss=0.1798, simple_loss=0.2406, pruned_loss=0.0595, over 4873.00 frames. ], tot_loss[loss=0.1904, simple_loss=0.2541, pruned_loss=0.06328, over 510207.45 frames. ], batch size: 34, lr: 3.72e-03, grad_scale: 16.0 2023-03-26 12:42:16,966 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.085e+02 1.728e+02 2.070e+02 2.489e+02 4.280e+02, threshold=4.140e+02, percent-clipped=3.0 2023-03-26 12:42:31,653 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57459.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:42:31,679 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57459.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:42:33,519 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57462.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:42:44,008 INFO [finetune.py:976] (3/7) Epoch 11, batch 200, loss[loss=0.1633, simple_loss=0.2383, pruned_loss=0.04417, over 4766.00 frames. ], tot_loss[loss=0.1892, simple_loss=0.2527, pruned_loss=0.06289, over 610070.38 frames. ], batch size: 28, lr: 3.72e-03, grad_scale: 16.0 2023-03-26 12:42:56,411 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57495.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:43:03,708 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=57507.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:43:14,473 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57523.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:43:17,309 INFO [finetune.py:976] (3/7) Epoch 11, batch 250, loss[loss=0.1508, simple_loss=0.2203, pruned_loss=0.04061, over 4830.00 frames. ], tot_loss[loss=0.1906, simple_loss=0.2548, pruned_loss=0.06323, over 685986.52 frames. ], batch size: 25, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:43:19,751 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0676, 1.6896, 1.9938, 2.0009, 1.7375, 1.7452, 1.9170, 1.8431], device='cuda:3'), covar=tensor([0.4429, 0.4778, 0.3924, 0.4645, 0.5526, 0.4490, 0.6066, 0.3868], device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0238, 0.0253, 0.0256, 0.0252, 0.0229, 0.0273, 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') 2023-03-26 12:43:22,638 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.287e+02 1.584e+02 1.966e+02 2.356e+02 4.681e+02, threshold=3.932e+02, percent-clipped=1.0 2023-03-26 12:43:27,967 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57542.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:43:43,781 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57553.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 12:43:45,613 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57556.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:43:55,237 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57564.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:44:08,373 INFO [finetune.py:976] (3/7) Epoch 11, batch 300, loss[loss=0.1996, simple_loss=0.2535, pruned_loss=0.07287, over 4902.00 frames. ], tot_loss[loss=0.1943, simple_loss=0.2595, pruned_loss=0.06459, over 746733.62 frames. ], batch size: 32, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:44:24,433 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57600.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:44:31,736 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=57612.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:44:40,867 INFO [finetune.py:976] (3/7) Epoch 11, batch 350, loss[loss=0.2366, simple_loss=0.292, pruned_loss=0.09059, over 4741.00 frames. ], tot_loss[loss=0.1958, simple_loss=0.2611, pruned_loss=0.06524, over 792649.69 frames. ], batch size: 59, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:44:46,738 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.100e+02 1.573e+02 1.819e+02 2.403e+02 4.156e+02, threshold=3.639e+02, percent-clipped=1.0 2023-03-26 12:44:56,806 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57649.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:44:58,453 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57651.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:45:14,036 INFO [finetune.py:976] (3/7) Epoch 11, batch 400, loss[loss=0.1822, simple_loss=0.2554, pruned_loss=0.05454, over 4829.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2634, pruned_loss=0.06552, over 830170.64 frames. ], batch size: 47, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:45:30,905 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=57697.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:45:32,135 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=57699.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:45:35,182 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-26 12:45:49,608 INFO [finetune.py:976] (3/7) Epoch 11, batch 450, loss[loss=0.1621, simple_loss=0.2309, pruned_loss=0.04661, over 4787.00 frames. ], tot_loss[loss=0.197, simple_loss=0.263, pruned_loss=0.06552, over 858952.14 frames. ], batch size: 29, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:45:53,879 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57733.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:45:55,473 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.158e+02 1.602e+02 1.902e+02 2.220e+02 3.989e+02, threshold=3.804e+02, percent-clipped=2.0 2023-03-26 12:46:15,572 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57754.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:46:23,664 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-26 12:46:23,883 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9772, 1.7039, 2.2192, 1.5813, 2.0755, 2.0982, 1.5674, 2.3359], device='cuda:3'), covar=tensor([0.1407, 0.2058, 0.1449, 0.1944, 0.0946, 0.1645, 0.2959, 0.0891], device='cuda:3'), in_proj_covar=tensor([0.0200, 0.0206, 0.0195, 0.0192, 0.0179, 0.0217, 0.0219, 0.0203], 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') 2023-03-26 12:46:32,837 INFO [finetune.py:976] (3/7) Epoch 11, batch 500, loss[loss=0.1518, simple_loss=0.2227, pruned_loss=0.04047, over 4916.00 frames. ], tot_loss[loss=0.1941, simple_loss=0.2598, pruned_loss=0.06422, over 880399.37 frames. ], batch size: 37, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:46:45,702 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57794.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:47:01,346 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57818.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:47:06,698 INFO [finetune.py:976] (3/7) Epoch 11, batch 550, loss[loss=0.2052, simple_loss=0.2684, pruned_loss=0.07102, over 4838.00 frames. ], tot_loss[loss=0.1915, simple_loss=0.2566, pruned_loss=0.06318, over 895240.20 frames. ], batch size: 33, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:47:11,532 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.185e+02 1.635e+02 1.936e+02 2.160e+02 3.511e+02, threshold=3.871e+02, percent-clipped=0.0 2023-03-26 12:47:16,794 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57842.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:47:20,232 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.4914, 1.5753, 1.5904, 0.9457, 1.6189, 1.8376, 1.8037, 1.3858], device='cuda:3'), covar=tensor([0.0966, 0.0583, 0.0407, 0.0499, 0.0389, 0.0525, 0.0282, 0.0693], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0153, 0.0120, 0.0132, 0.0130, 0.0123, 0.0142, 0.0145], device='cuda:3'), out_proj_covar=tensor([9.4172e-05, 1.1240e-04, 8.6302e-05, 9.5122e-05, 9.3206e-05, 8.9835e-05, 1.0412e-04, 1.0622e-04], device='cuda:3') 2023-03-26 12:47:23,739 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57851.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:47:24,993 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57853.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 12:47:40,101 INFO [finetune.py:976] (3/7) Epoch 11, batch 600, loss[loss=0.2019, simple_loss=0.2725, pruned_loss=0.06559, over 4833.00 frames. ], tot_loss[loss=0.1909, simple_loss=0.2556, pruned_loss=0.06303, over 909286.56 frames. ], batch size: 33, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:47:40,294 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.16 vs. limit=5.0 2023-03-26 12:47:48,475 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=57890.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:47:56,179 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57900.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:47:56,761 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=57901.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:48:13,629 INFO [finetune.py:976] (3/7) Epoch 11, batch 650, loss[loss=0.2038, simple_loss=0.2736, pruned_loss=0.06703, over 4847.00 frames. ], tot_loss[loss=0.192, simple_loss=0.2577, pruned_loss=0.06318, over 919947.02 frames. ], batch size: 44, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:48:18,497 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.178e+02 1.568e+02 1.897e+02 2.360e+02 4.682e+02, threshold=3.793e+02, percent-clipped=3.0 2023-03-26 12:48:27,448 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=57948.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:48:33,954 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 2023-03-26 12:48:48,745 INFO [finetune.py:976] (3/7) Epoch 11, batch 700, loss[loss=0.2044, simple_loss=0.2668, pruned_loss=0.07099, over 4820.00 frames. ], tot_loss[loss=0.1949, simple_loss=0.2608, pruned_loss=0.0645, over 927551.81 frames. ], batch size: 39, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:49:44,561 INFO [finetune.py:976] (3/7) Epoch 11, batch 750, loss[loss=0.1542, simple_loss=0.2277, pruned_loss=0.04039, over 4768.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2634, pruned_loss=0.06632, over 932084.40 frames. ], batch size: 28, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:49:49,410 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.116e+01 1.579e+02 1.894e+02 2.321e+02 4.436e+02, threshold=3.789e+02, percent-clipped=3.0 2023-03-26 12:50:02,174 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58054.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:50:18,110 INFO [finetune.py:976] (3/7) Epoch 11, batch 800, loss[loss=0.1681, simple_loss=0.2406, pruned_loss=0.04777, over 4760.00 frames. ], tot_loss[loss=0.1952, simple_loss=0.2614, pruned_loss=0.06451, over 935971.95 frames. ], batch size: 27, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:50:25,465 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58089.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:50:33,870 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=58102.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:50:45,550 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58118.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:50:51,411 INFO [finetune.py:976] (3/7) Epoch 11, batch 850, loss[loss=0.2294, simple_loss=0.2809, pruned_loss=0.08897, over 4722.00 frames. ], tot_loss[loss=0.1943, simple_loss=0.2603, pruned_loss=0.06413, over 941246.28 frames. ], batch size: 23, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:50:56,226 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.394e+01 1.505e+02 1.749e+02 2.082e+02 4.545e+02, threshold=3.498e+02, percent-clipped=2.0 2023-03-26 12:50:56,922 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.2254, 2.6849, 2.6022, 1.3913, 2.6572, 2.3176, 2.3213, 2.4253], device='cuda:3'), covar=tensor([0.0905, 0.0992, 0.1670, 0.2263, 0.1872, 0.2013, 0.1935, 0.1241], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0201, 0.0202, 0.0188, 0.0216, 0.0209, 0.0224, 0.0197], 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') 2023-03-26 12:50:59,937 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58141.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:51:01,768 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58144.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:51:05,919 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58151.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:51:08,865 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1693, 1.8528, 1.4128, 0.5369, 1.6016, 1.8068, 1.7435, 1.6944], device='cuda:3'), covar=tensor([0.0810, 0.0861, 0.1452, 0.2019, 0.1297, 0.2177, 0.2050, 0.0878], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0201, 0.0202, 0.0188, 0.0216, 0.0208, 0.0223, 0.0197], 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') 2023-03-26 12:51:23,203 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=58166.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:51:33,582 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5953, 1.4132, 2.0245, 3.1815, 2.0773, 2.3193, 1.0155, 2.5544], device='cuda:3'), covar=tensor([0.1789, 0.1543, 0.1333, 0.0702, 0.0878, 0.1426, 0.1851, 0.0587], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0116, 0.0133, 0.0163, 0.0101, 0.0137, 0.0125, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 12:51:35,955 INFO [finetune.py:976] (3/7) Epoch 11, batch 900, loss[loss=0.1708, simple_loss=0.2334, pruned_loss=0.05404, over 4841.00 frames. ], tot_loss[loss=0.1931, simple_loss=0.2583, pruned_loss=0.06401, over 943869.54 frames. ], batch size: 44, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:51:57,410 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=58199.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:51:59,473 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58202.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:52:01,305 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58205.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:52:12,172 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7115, 1.6009, 1.5515, 1.6228, 1.1538, 3.4241, 1.4570, 1.9895], device='cuda:3'), covar=tensor([0.3417, 0.2387, 0.2093, 0.2362, 0.1794, 0.0188, 0.2786, 0.1235], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0116, 0.0121, 0.0123, 0.0115, 0.0098, 0.0099, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 12:52:17,045 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4947, 2.1863, 1.9673, 2.1014, 2.1391, 2.1331, 2.1374, 2.8384], device='cuda:3'), covar=tensor([0.3862, 0.4688, 0.3508, 0.4268, 0.4279, 0.2578, 0.4115, 0.1725], device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0259, 0.0222, 0.0277, 0.0242, 0.0209, 0.0246, 0.0214], 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') 2023-03-26 12:52:17,500 INFO [finetune.py:976] (3/7) Epoch 11, batch 950, loss[loss=0.1632, simple_loss=0.215, pruned_loss=0.05572, over 4156.00 frames. ], tot_loss[loss=0.1911, simple_loss=0.2562, pruned_loss=0.06306, over 946398.99 frames. ], batch size: 18, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:52:22,883 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.011e+02 1.516e+02 1.975e+02 2.310e+02 4.008e+02, threshold=3.950e+02, percent-clipped=1.0 2023-03-26 12:52:51,452 INFO [finetune.py:976] (3/7) Epoch 11, batch 1000, loss[loss=0.211, simple_loss=0.2795, pruned_loss=0.07127, over 4867.00 frames. ], tot_loss[loss=0.1948, simple_loss=0.2596, pruned_loss=0.06503, over 949062.31 frames. ], batch size: 34, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:53:18,563 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4555, 2.3089, 1.7953, 2.6528, 2.3947, 1.9578, 2.9087, 2.4145], device='cuda:3'), covar=tensor([0.1439, 0.2875, 0.3533, 0.2936, 0.2907, 0.1878, 0.3545, 0.2050], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0187, 0.0232, 0.0253, 0.0238, 0.0195, 0.0211, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 12:53:46,407 INFO [finetune.py:976] (3/7) Epoch 11, batch 1050, loss[loss=0.2081, simple_loss=0.2632, pruned_loss=0.0765, over 4808.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2623, pruned_loss=0.06614, over 947855.30 frames. ], batch size: 25, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:53:51,319 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.197e+02 1.617e+02 2.003e+02 2.375e+02 3.670e+02, threshold=4.006e+02, percent-clipped=0.0 2023-03-26 12:54:42,548 INFO [finetune.py:976] (3/7) Epoch 11, batch 1100, loss[loss=0.2229, simple_loss=0.2902, pruned_loss=0.07785, over 4842.00 frames. ], tot_loss[loss=0.1961, simple_loss=0.2618, pruned_loss=0.06515, over 948917.09 frames. ], batch size: 49, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:54:55,701 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58389.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:55:23,532 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7330, 1.6828, 1.4083, 1.9099, 2.1749, 1.9624, 1.3582, 1.4173], device='cuda:3'), covar=tensor([0.2345, 0.2103, 0.2034, 0.1746, 0.1865, 0.1133, 0.2619, 0.2068], device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0207, 0.0207, 0.0189, 0.0241, 0.0181, 0.0213, 0.0197], 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') 2023-03-26 12:55:35,371 INFO [finetune.py:976] (3/7) Epoch 11, batch 1150, loss[loss=0.1946, simple_loss=0.2599, pruned_loss=0.06462, over 4725.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2633, pruned_loss=0.06549, over 950206.66 frames. ], batch size: 59, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:55:40,650 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.092e+02 1.672e+02 1.870e+02 2.321e+02 4.403e+02, threshold=3.740e+02, percent-clipped=1.0 2023-03-26 12:55:41,941 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=58437.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:55:54,582 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6391, 1.5799, 1.5161, 1.5705, 1.0736, 3.4393, 1.4080, 1.8958], device='cuda:3'), covar=tensor([0.3950, 0.2892, 0.2266, 0.2766, 0.1838, 0.0251, 0.2453, 0.1209], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0119, 0.0122, 0.0115, 0.0097, 0.0098, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 12:56:08,438 INFO [finetune.py:976] (3/7) Epoch 11, batch 1200, loss[loss=0.1617, simple_loss=0.224, pruned_loss=0.04964, over 4809.00 frames. ], tot_loss[loss=0.1952, simple_loss=0.2615, pruned_loss=0.06446, over 950868.38 frames. ], batch size: 38, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:56:08,577 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7653, 1.6469, 1.4722, 1.8742, 2.0346, 1.8831, 1.2154, 1.5237], device='cuda:3'), covar=tensor([0.2186, 0.2132, 0.1976, 0.1579, 0.1717, 0.1120, 0.2650, 0.1986], device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0208, 0.0207, 0.0189, 0.0242, 0.0181, 0.0213, 0.0197], 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') 2023-03-26 12:56:15,631 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2533, 2.1414, 1.7212, 2.2873, 2.1960, 1.9032, 2.5813, 2.2325], device='cuda:3'), covar=tensor([0.1310, 0.2435, 0.3040, 0.2649, 0.2705, 0.1662, 0.3024, 0.1910], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0187, 0.0231, 0.0253, 0.0238, 0.0195, 0.0211, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 12:56:21,455 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58497.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:56:23,265 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58500.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:56:37,015 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.1772, 2.5332, 2.4319, 1.2872, 2.6641, 2.2434, 1.6854, 2.1214], device='cuda:3'), covar=tensor([0.0671, 0.0947, 0.1592, 0.1998, 0.1475, 0.1651, 0.2207, 0.1128], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0202, 0.0203, 0.0189, 0.0216, 0.0209, 0.0224, 0.0198], 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') 2023-03-26 12:56:40,496 INFO [finetune.py:976] (3/7) Epoch 11, batch 1250, loss[loss=0.1745, simple_loss=0.2423, pruned_loss=0.0534, over 4903.00 frames. ], tot_loss[loss=0.194, simple_loss=0.2594, pruned_loss=0.06431, over 951837.23 frames. ], batch size: 46, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:56:46,790 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.332e+01 1.581e+02 1.822e+02 2.261e+02 4.369e+02, threshold=3.644e+02, percent-clipped=3.0 2023-03-26 12:57:08,445 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9578, 1.4791, 1.9730, 1.8609, 1.7064, 1.6495, 1.8409, 1.8093], device='cuda:3'), covar=tensor([0.4180, 0.5012, 0.4221, 0.4551, 0.6188, 0.4444, 0.5748, 0.3955], device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0238, 0.0252, 0.0256, 0.0252, 0.0228, 0.0273, 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') 2023-03-26 12:57:15,451 INFO [finetune.py:976] (3/7) Epoch 11, batch 1300, loss[loss=0.1624, simple_loss=0.2252, pruned_loss=0.04976, over 4913.00 frames. ], tot_loss[loss=0.1925, simple_loss=0.2575, pruned_loss=0.06376, over 951086.73 frames. ], batch size: 43, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:57:30,719 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6318, 1.4859, 2.0188, 3.1575, 2.0566, 2.2752, 0.8701, 2.4407], device='cuda:3'), covar=tensor([0.1768, 0.1517, 0.1352, 0.0634, 0.0849, 0.1328, 0.2021, 0.0597], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0132, 0.0163, 0.0100, 0.0137, 0.0124, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 12:57:48,896 INFO [finetune.py:976] (3/7) Epoch 11, batch 1350, loss[loss=0.1942, simple_loss=0.246, pruned_loss=0.0712, over 4051.00 frames. ], tot_loss[loss=0.1951, simple_loss=0.2595, pruned_loss=0.06539, over 951075.62 frames. ], batch size: 17, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:57:54,733 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.484e+01 1.581e+02 1.914e+02 2.266e+02 4.857e+02, threshold=3.829e+02, percent-clipped=2.0 2023-03-26 12:58:20,754 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-26 12:58:23,950 INFO [finetune.py:976] (3/7) Epoch 11, batch 1400, loss[loss=0.2435, simple_loss=0.3049, pruned_loss=0.09101, over 4857.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2637, pruned_loss=0.06662, over 952201.15 frames. ], batch size: 44, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:58:26,989 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58681.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:58:56,047 INFO [finetune.py:976] (3/7) Epoch 11, batch 1450, loss[loss=0.1987, simple_loss=0.2739, pruned_loss=0.06177, over 4827.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2651, pruned_loss=0.06681, over 951602.42 frames. ], batch size: 47, lr: 3.71e-03, grad_scale: 16.0 2023-03-26 12:59:00,193 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5987, 1.4617, 1.4471, 1.5556, 1.2094, 3.3977, 1.4154, 1.8837], device='cuda:3'), covar=tensor([0.3310, 0.2431, 0.2178, 0.2399, 0.1806, 0.0184, 0.2927, 0.1345], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0116, 0.0120, 0.0123, 0.0115, 0.0098, 0.0099, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 12:59:01,960 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.107e+02 1.669e+02 2.009e+02 2.318e+02 4.324e+02, threshold=4.017e+02, percent-clipped=1.0 2023-03-26 12:59:07,261 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58742.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:59:35,096 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58775.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 12:59:36,228 INFO [finetune.py:976] (3/7) Epoch 11, batch 1500, loss[loss=0.2244, simple_loss=0.2928, pruned_loss=0.07794, over 4895.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2649, pruned_loss=0.06671, over 951450.11 frames. ], batch size: 43, lr: 3.70e-03, grad_scale: 16.0 2023-03-26 12:59:58,349 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58797.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:00:04,299 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58800.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:00:16,528 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7123, 1.7256, 1.5164, 1.8646, 2.3558, 1.9057, 1.6614, 1.4270], device='cuda:3'), covar=tensor([0.2105, 0.2018, 0.1828, 0.1572, 0.1641, 0.1182, 0.2339, 0.1827], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0208, 0.0208, 0.0190, 0.0242, 0.0182, 0.0213, 0.0198], 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') 2023-03-26 13:00:34,186 INFO [finetune.py:976] (3/7) Epoch 11, batch 1550, loss[loss=0.2134, simple_loss=0.2813, pruned_loss=0.07279, over 4851.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2659, pruned_loss=0.06732, over 951773.69 frames. ], batch size: 44, lr: 3.70e-03, grad_scale: 16.0 2023-03-26 13:00:39,949 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.185e+02 1.569e+02 1.959e+02 2.197e+02 4.059e+02, threshold=3.918e+02, percent-clipped=1.0 2023-03-26 13:00:41,215 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58836.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:00:45,972 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6300, 1.5317, 1.0740, 0.3408, 1.2800, 1.4648, 1.3778, 1.4153], device='cuda:3'), covar=tensor([0.0925, 0.0813, 0.1337, 0.1925, 0.1480, 0.2558, 0.2322, 0.0883], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0202, 0.0203, 0.0189, 0.0216, 0.0210, 0.0224, 0.0198], 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') 2023-03-26 13:00:47,625 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=58845.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:00:49,963 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=58848.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:01:07,921 INFO [finetune.py:976] (3/7) Epoch 11, batch 1600, loss[loss=0.2185, simple_loss=0.2819, pruned_loss=0.07757, over 4900.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2637, pruned_loss=0.0668, over 953467.60 frames. ], batch size: 35, lr: 3.70e-03, grad_scale: 16.0 2023-03-26 13:01:08,870 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.19 vs. limit=5.0 2023-03-26 13:01:20,301 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6144, 2.4760, 2.0345, 2.5722, 2.4479, 2.1630, 2.9914, 2.5377], device='cuda:3'), covar=tensor([0.1347, 0.2596, 0.3149, 0.2970, 0.2826, 0.1764, 0.3459, 0.1966], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0187, 0.0232, 0.0254, 0.0239, 0.0196, 0.0212, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 13:01:28,378 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58906.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:01:50,382 INFO [finetune.py:976] (3/7) Epoch 11, batch 1650, loss[loss=0.1777, simple_loss=0.2405, pruned_loss=0.05746, over 4813.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.2606, pruned_loss=0.06545, over 953443.90 frames. ], batch size: 51, lr: 3.70e-03, grad_scale: 16.0 2023-03-26 13:01:55,258 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.149e+02 1.664e+02 1.923e+02 2.390e+02 4.121e+02, threshold=3.846e+02, percent-clipped=1.0 2023-03-26 13:01:56,005 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2993, 2.1302, 1.7045, 2.2102, 2.2122, 1.9290, 2.4921, 2.2070], device='cuda:3'), covar=tensor([0.1470, 0.2566, 0.3594, 0.2938, 0.2850, 0.1908, 0.4048, 0.2181], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0187, 0.0232, 0.0254, 0.0239, 0.0196, 0.0211, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 13:02:18,252 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58967.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:02:23,506 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-26 13:02:24,167 INFO [finetune.py:976] (3/7) Epoch 11, batch 1700, loss[loss=0.193, simple_loss=0.2614, pruned_loss=0.06228, over 4817.00 frames. ], tot_loss[loss=0.1925, simple_loss=0.2571, pruned_loss=0.06395, over 954718.44 frames. ], batch size: 33, lr: 3.70e-03, grad_scale: 16.0 2023-03-26 13:02:37,399 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-03-26 13:02:57,898 INFO [finetune.py:976] (3/7) Epoch 11, batch 1750, loss[loss=0.1802, simple_loss=0.255, pruned_loss=0.05268, over 4766.00 frames. ], tot_loss[loss=0.1925, simple_loss=0.2574, pruned_loss=0.06377, over 955010.63 frames. ], batch size: 26, lr: 3.70e-03, grad_scale: 16.0 2023-03-26 13:03:01,710 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-03-26 13:03:02,754 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.140e+02 1.620e+02 1.895e+02 2.249e+02 5.052e+02, threshold=3.790e+02, percent-clipped=2.0 2023-03-26 13:03:04,077 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59037.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:03:17,618 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-26 13:03:19,385 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0010, 2.0315, 1.7855, 2.1374, 2.7327, 2.0595, 2.0189, 1.4774], device='cuda:3'), covar=tensor([0.2234, 0.2013, 0.1823, 0.1685, 0.1694, 0.1158, 0.2123, 0.1947], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0208, 0.0208, 0.0189, 0.0242, 0.0182, 0.0213, 0.0197], 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') 2023-03-26 13:03:33,672 INFO [finetune.py:976] (3/7) Epoch 11, batch 1800, loss[loss=0.2156, simple_loss=0.2751, pruned_loss=0.078, over 4772.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2627, pruned_loss=0.06561, over 956088.11 frames. ], batch size: 28, lr: 3.70e-03, grad_scale: 16.0 2023-03-26 13:03:52,181 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0294, 1.5622, 2.2506, 4.1400, 2.8291, 2.9330, 0.6461, 3.5740], device='cuda:3'), covar=tensor([0.2033, 0.2236, 0.1854, 0.0958, 0.0970, 0.1320, 0.2805, 0.0556], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0132, 0.0163, 0.0100, 0.0137, 0.0125, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 13:03:57,438 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-26 13:04:03,259 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-26 13:04:19,595 INFO [finetune.py:976] (3/7) Epoch 11, batch 1850, loss[loss=0.167, simple_loss=0.2513, pruned_loss=0.04137, over 4816.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.2624, pruned_loss=0.0651, over 956035.68 frames. ], batch size: 38, lr: 3.70e-03, grad_scale: 16.0 2023-03-26 13:04:22,111 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59131.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:04:24,431 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.037e+02 1.668e+02 2.065e+02 2.636e+02 4.490e+02, threshold=4.130e+02, percent-clipped=5.0 2023-03-26 13:04:29,731 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.40 vs. limit=5.0 2023-03-26 13:04:44,000 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59164.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:04:57,374 INFO [finetune.py:976] (3/7) Epoch 11, batch 1900, loss[loss=0.2697, simple_loss=0.3189, pruned_loss=0.1103, over 4814.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.264, pruned_loss=0.06557, over 957727.82 frames. ], batch size: 45, lr: 3.70e-03, grad_scale: 16.0 2023-03-26 13:04:58,186 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-26 13:05:46,838 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59225.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:05:47,935 INFO [finetune.py:976] (3/7) Epoch 11, batch 1950, loss[loss=0.2242, simple_loss=0.2737, pruned_loss=0.08737, over 4703.00 frames. ], tot_loss[loss=0.1949, simple_loss=0.2615, pruned_loss=0.0641, over 957159.82 frames. ], batch size: 59, lr: 3.70e-03, grad_scale: 16.0 2023-03-26 13:05:58,853 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7305, 1.6494, 1.5020, 1.3578, 1.8262, 1.5072, 1.8217, 1.6985], device='cuda:3'), covar=tensor([0.1325, 0.2034, 0.2803, 0.2452, 0.2489, 0.1607, 0.2949, 0.1805], device='cuda:3'), in_proj_covar=tensor([0.0174, 0.0186, 0.0231, 0.0253, 0.0238, 0.0195, 0.0210, 0.0194], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 13:05:59,306 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.090e+02 1.570e+02 1.817e+02 2.294e+02 4.310e+02, threshold=3.633e+02, percent-clipped=1.0 2023-03-26 13:06:25,605 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-26 13:06:29,849 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59262.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:06:51,936 INFO [finetune.py:976] (3/7) Epoch 11, batch 2000, loss[loss=0.1662, simple_loss=0.2321, pruned_loss=0.05013, over 4767.00 frames. ], tot_loss[loss=0.1923, simple_loss=0.2587, pruned_loss=0.06297, over 957642.20 frames. ], batch size: 26, lr: 3.70e-03, grad_scale: 32.0 2023-03-26 13:07:32,698 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2023-03-26 13:07:37,499 INFO [finetune.py:976] (3/7) Epoch 11, batch 2050, loss[loss=0.165, simple_loss=0.2372, pruned_loss=0.04644, over 4756.00 frames. ], tot_loss[loss=0.1903, simple_loss=0.2554, pruned_loss=0.06253, over 954877.11 frames. ], batch size: 27, lr: 3.70e-03, grad_scale: 32.0 2023-03-26 13:07:42,277 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.122e+01 1.513e+02 1.843e+02 2.174e+02 3.611e+02, threshold=3.686e+02, percent-clipped=0.0 2023-03-26 13:07:44,102 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59337.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:08:17,286 INFO [finetune.py:976] (3/7) Epoch 11, batch 2100, loss[loss=0.258, simple_loss=0.3109, pruned_loss=0.1026, over 4913.00 frames. ], tot_loss[loss=0.1907, simple_loss=0.2555, pruned_loss=0.06292, over 956099.86 frames. ], batch size: 36, lr: 3.70e-03, grad_scale: 32.0 2023-03-26 13:08:27,940 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=59385.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:08:32,798 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9511, 1.7822, 1.5650, 1.8372, 1.7314, 1.6711, 1.7268, 2.4754], device='cuda:3'), covar=tensor([0.4183, 0.4928, 0.3626, 0.4222, 0.4308, 0.2606, 0.4346, 0.1646], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0261, 0.0225, 0.0280, 0.0243, 0.0210, 0.0246, 0.0216], 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') 2023-03-26 13:08:45,148 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4742, 1.2373, 1.2968, 1.3711, 1.6820, 1.5680, 1.4274, 1.2377], device='cuda:3'), covar=tensor([0.0312, 0.0305, 0.0610, 0.0316, 0.0249, 0.0470, 0.0316, 0.0400], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0109, 0.0139, 0.0114, 0.0101, 0.0103, 0.0092, 0.0107], device='cuda:3'), out_proj_covar=tensor([7.0362e-05, 8.4643e-05, 1.1079e-04, 8.9077e-05, 7.9076e-05, 7.6633e-05, 7.0121e-05, 8.2697e-05], device='cuda:3') 2023-03-26 13:09:08,919 INFO [finetune.py:976] (3/7) Epoch 11, batch 2150, loss[loss=0.2026, simple_loss=0.2834, pruned_loss=0.06091, over 4899.00 frames. ], tot_loss[loss=0.1929, simple_loss=0.258, pruned_loss=0.06396, over 955045.02 frames. ], batch size: 43, lr: 3.70e-03, grad_scale: 32.0 2023-03-26 13:09:13,347 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59431.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:09:15,687 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.234e+02 1.596e+02 1.893e+02 2.254e+02 5.168e+02, threshold=3.786e+02, percent-clipped=3.0 2023-03-26 13:09:54,871 INFO [finetune.py:976] (3/7) Epoch 11, batch 2200, loss[loss=0.1624, simple_loss=0.2298, pruned_loss=0.0475, over 4802.00 frames. ], tot_loss[loss=0.1953, simple_loss=0.2612, pruned_loss=0.06468, over 955856.46 frames. ], batch size: 25, lr: 3.70e-03, grad_scale: 32.0 2023-03-26 13:09:56,664 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=59479.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:10:13,071 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59502.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:10:15,546 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6770, 1.6196, 1.5394, 1.7207, 1.1714, 3.6764, 1.3826, 2.1302], device='cuda:3'), covar=tensor([0.3243, 0.2391, 0.2138, 0.2308, 0.1789, 0.0158, 0.2478, 0.1134], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0119, 0.0122, 0.0114, 0.0098, 0.0098, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 13:10:25,070 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59520.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:10:30,584 INFO [finetune.py:976] (3/7) Epoch 11, batch 2250, loss[loss=0.2314, simple_loss=0.2892, pruned_loss=0.08681, over 4896.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2639, pruned_loss=0.06646, over 955274.91 frames. ], batch size: 32, lr: 3.70e-03, grad_scale: 32.0 2023-03-26 13:10:37,559 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.128e+02 1.729e+02 2.023e+02 2.518e+02 3.990e+02, threshold=4.047e+02, percent-clipped=2.0 2023-03-26 13:11:00,967 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.48 vs. limit=5.0 2023-03-26 13:11:02,693 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59562.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:11:03,337 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59563.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:11:13,470 INFO [finetune.py:976] (3/7) Epoch 11, batch 2300, loss[loss=0.1541, simple_loss=0.2247, pruned_loss=0.04172, over 4814.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2643, pruned_loss=0.06601, over 956069.49 frames. ], batch size: 39, lr: 3.70e-03, grad_scale: 32.0 2023-03-26 13:11:19,778 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-26 13:11:24,320 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0450, 1.6692, 2.4753, 1.5585, 2.1905, 2.3716, 1.6766, 2.4269], device='cuda:3'), covar=tensor([0.1363, 0.2164, 0.1110, 0.1992, 0.0836, 0.1263, 0.2818, 0.0828], device='cuda:3'), in_proj_covar=tensor([0.0198, 0.0206, 0.0194, 0.0190, 0.0178, 0.0215, 0.0217, 0.0202], 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') 2023-03-26 13:11:35,297 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=59610.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:11:47,097 INFO [finetune.py:976] (3/7) Epoch 11, batch 2350, loss[loss=0.2171, simple_loss=0.2875, pruned_loss=0.0733, over 4810.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2624, pruned_loss=0.06503, over 955785.15 frames. ], batch size: 40, lr: 3.70e-03, grad_scale: 32.0 2023-03-26 13:11:52,463 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.309e+01 1.451e+02 1.728e+02 2.097e+02 4.600e+02, threshold=3.455e+02, percent-clipped=1.0 2023-03-26 13:12:01,918 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-03-26 13:12:19,961 INFO [finetune.py:976] (3/7) Epoch 11, batch 2400, loss[loss=0.2001, simple_loss=0.2661, pruned_loss=0.06705, over 4723.00 frames. ], tot_loss[loss=0.1945, simple_loss=0.2599, pruned_loss=0.06452, over 957161.92 frames. ], batch size: 54, lr: 3.70e-03, grad_scale: 32.0 2023-03-26 13:12:53,270 INFO [finetune.py:976] (3/7) Epoch 11, batch 2450, loss[loss=0.1495, simple_loss=0.2145, pruned_loss=0.04227, over 4822.00 frames. ], tot_loss[loss=0.1913, simple_loss=0.2565, pruned_loss=0.06305, over 958294.84 frames. ], batch size: 33, lr: 3.70e-03, grad_scale: 32.0 2023-03-26 13:13:01,211 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.102e+02 1.641e+02 1.877e+02 2.149e+02 5.374e+02, threshold=3.753e+02, percent-clipped=2.0 2023-03-26 13:13:37,051 INFO [finetune.py:976] (3/7) Epoch 11, batch 2500, loss[loss=0.1456, simple_loss=0.2159, pruned_loss=0.03761, over 4823.00 frames. ], tot_loss[loss=0.1953, simple_loss=0.2598, pruned_loss=0.06537, over 960239.85 frames. ], batch size: 30, lr: 3.70e-03, grad_scale: 32.0 2023-03-26 13:14:29,698 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59820.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:14:33,891 INFO [finetune.py:976] (3/7) Epoch 11, batch 2550, loss[loss=0.2134, simple_loss=0.2855, pruned_loss=0.07063, over 4870.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2633, pruned_loss=0.06579, over 958942.50 frames. ], batch size: 44, lr: 3.70e-03, grad_scale: 32.0 2023-03-26 13:14:40,182 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.135e+02 1.636e+02 1.885e+02 2.323e+02 4.849e+02, threshold=3.771e+02, percent-clipped=2.0 2023-03-26 13:14:46,996 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-03-26 13:14:49,222 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59848.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:14:57,187 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59858.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:14:57,958 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.80 vs. limit=2.0 2023-03-26 13:15:03,807 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=59868.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:15:09,218 INFO [finetune.py:976] (3/7) Epoch 11, batch 2600, loss[loss=0.1923, simple_loss=0.252, pruned_loss=0.06634, over 4815.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.265, pruned_loss=0.06636, over 958703.11 frames. ], batch size: 33, lr: 3.70e-03, grad_scale: 32.0 2023-03-26 13:15:18,130 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59889.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:15:31,228 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59909.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:15:42,442 INFO [finetune.py:976] (3/7) Epoch 11, batch 2650, loss[loss=0.2103, simple_loss=0.2784, pruned_loss=0.0711, over 4810.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2661, pruned_loss=0.06685, over 956013.90 frames. ], batch size: 41, lr: 3.70e-03, grad_scale: 32.0 2023-03-26 13:15:47,331 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.166e+02 1.549e+02 1.976e+02 2.444e+02 3.877e+02, threshold=3.952e+02, percent-clipped=1.0 2023-03-26 13:16:03,037 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59950.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:16:29,561 INFO [finetune.py:976] (3/7) Epoch 11, batch 2700, loss[loss=0.2035, simple_loss=0.2713, pruned_loss=0.06784, over 4859.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2636, pruned_loss=0.06595, over 952984.72 frames. ], batch size: 34, lr: 3.70e-03, grad_scale: 32.0 2023-03-26 13:16:38,534 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2450, 1.3196, 1.5647, 1.0399, 1.2336, 1.4154, 1.3205, 1.5712], device='cuda:3'), covar=tensor([0.1218, 0.2291, 0.1283, 0.1590, 0.0931, 0.1260, 0.2948, 0.0873], device='cuda:3'), in_proj_covar=tensor([0.0198, 0.0206, 0.0194, 0.0190, 0.0178, 0.0215, 0.0218, 0.0201], 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') 2023-03-26 13:16:42,159 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-03-26 13:17:04,322 INFO [finetune.py:976] (3/7) Epoch 11, batch 2750, loss[loss=0.1696, simple_loss=0.2276, pruned_loss=0.05576, over 4816.00 frames. ], tot_loss[loss=0.1945, simple_loss=0.2597, pruned_loss=0.06461, over 953338.38 frames. ], batch size: 38, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:17:09,207 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.070e+02 1.603e+02 1.823e+02 2.284e+02 4.397e+02, threshold=3.646e+02, percent-clipped=1.0 2023-03-26 13:17:12,380 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60040.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:17:21,588 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60052.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:17:30,381 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60065.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:17:37,371 INFO [finetune.py:976] (3/7) Epoch 11, batch 2800, loss[loss=0.1418, simple_loss=0.2071, pruned_loss=0.03824, over 4777.00 frames. ], tot_loss[loss=0.1925, simple_loss=0.2572, pruned_loss=0.06388, over 954257.29 frames. ], batch size: 26, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:17:38,096 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60078.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:17:54,532 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60101.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 13:18:02,768 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60113.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 13:18:10,606 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60126.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:18:11,110 INFO [finetune.py:976] (3/7) Epoch 11, batch 2850, loss[loss=0.1899, simple_loss=0.2564, pruned_loss=0.06165, over 4796.00 frames. ], tot_loss[loss=0.1915, simple_loss=0.2558, pruned_loss=0.06361, over 954367.80 frames. ], batch size: 29, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:18:17,974 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.190e+02 1.579e+02 1.818e+02 2.348e+02 4.165e+02, threshold=3.636e+02, percent-clipped=3.0 2023-03-26 13:18:20,419 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8023, 1.7659, 2.0577, 1.9750, 1.9110, 3.5909, 1.7466, 1.8960], device='cuda:3'), covar=tensor([0.0887, 0.1625, 0.0995, 0.0936, 0.1470, 0.0279, 0.1298, 0.1502], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0081, 0.0075, 0.0078, 0.0092, 0.0081, 0.0085, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 13:18:21,042 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60139.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:18:33,366 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3895, 1.5615, 1.3029, 1.5238, 1.8452, 1.6260, 1.5229, 1.2639], device='cuda:3'), covar=tensor([0.0316, 0.0249, 0.0557, 0.0267, 0.0205, 0.0435, 0.0291, 0.0413], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0109, 0.0140, 0.0114, 0.0102, 0.0104, 0.0093, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.1103e-05, 8.4938e-05, 1.1142e-04, 8.9295e-05, 7.9718e-05, 7.6972e-05, 7.0418e-05, 8.3216e-05], device='cuda:3') 2023-03-26 13:18:39,234 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4596, 1.1038, 0.7357, 1.2999, 1.9040, 0.6987, 1.3071, 1.3889], device='cuda:3'), covar=tensor([0.1635, 0.2290, 0.1933, 0.1326, 0.2069, 0.2005, 0.1560, 0.2005], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0098, 0.0115, 0.0094, 0.0122, 0.0096, 0.0101, 0.0092], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-26 13:18:39,245 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60158.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:18:51,668 INFO [finetune.py:976] (3/7) Epoch 11, batch 2900, loss[loss=0.2478, simple_loss=0.3097, pruned_loss=0.09293, over 4807.00 frames. ], tot_loss[loss=0.1942, simple_loss=0.2592, pruned_loss=0.06462, over 955059.09 frames. ], batch size: 45, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:19:12,023 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60198.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:19:21,312 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60204.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:19:27,580 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=60206.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:19:50,829 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9275, 1.7726, 1.6735, 1.7671, 2.1298, 1.9674, 1.8073, 1.5520], device='cuda:3'), covar=tensor([0.0268, 0.0294, 0.0460, 0.0268, 0.0225, 0.0469, 0.0243, 0.0402], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0109, 0.0139, 0.0113, 0.0101, 0.0103, 0.0093, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.0583e-05, 8.4536e-05, 1.1092e-04, 8.8772e-05, 7.9144e-05, 7.6542e-05, 7.0327e-05, 8.2739e-05], device='cuda:3') 2023-03-26 13:19:51,301 INFO [finetune.py:976] (3/7) Epoch 11, batch 2950, loss[loss=0.1679, simple_loss=0.2286, pruned_loss=0.05361, over 4232.00 frames. ], tot_loss[loss=0.1956, simple_loss=0.2617, pruned_loss=0.06471, over 954489.48 frames. ], batch size: 18, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:20:00,140 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.062e+02 1.723e+02 2.035e+02 2.444e+02 4.360e+02, threshold=4.070e+02, percent-clipped=6.0 2023-03-26 13:20:05,702 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-03-26 13:20:06,686 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60245.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:20:16,643 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60259.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:20:21,377 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9000, 2.5046, 2.3619, 1.2304, 2.4912, 2.0845, 1.9351, 2.2454], device='cuda:3'), covar=tensor([0.1029, 0.0982, 0.1782, 0.2300, 0.1809, 0.2315, 0.2303, 0.1336], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0201, 0.0202, 0.0187, 0.0216, 0.0209, 0.0224, 0.0197], 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') 2023-03-26 13:20:28,425 INFO [finetune.py:976] (3/7) Epoch 11, batch 3000, loss[loss=0.1936, simple_loss=0.2656, pruned_loss=0.06085, over 4917.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2638, pruned_loss=0.06557, over 956634.18 frames. ], batch size: 42, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:20:28,425 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 13:20:37,536 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0959, 1.7947, 1.7420, 1.7727, 1.7970, 1.8289, 1.8156, 2.4971], device='cuda:3'), covar=tensor([0.4571, 0.5449, 0.3806, 0.4494, 0.4743, 0.2951, 0.4781, 0.2061], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0260, 0.0223, 0.0279, 0.0243, 0.0210, 0.0247, 0.0216], 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') 2023-03-26 13:20:38,899 INFO [finetune.py:1010] (3/7) Epoch 11, validation: loss=0.1572, simple_loss=0.2284, pruned_loss=0.04301, over 2265189.00 frames. 2023-03-26 13:20:38,899 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 13:21:13,696 INFO [finetune.py:976] (3/7) Epoch 11, batch 3050, loss[loss=0.202, simple_loss=0.2695, pruned_loss=0.06725, over 4855.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2658, pruned_loss=0.06616, over 958302.00 frames. ], batch size: 31, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:21:19,479 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.076e+02 1.587e+02 1.939e+02 2.482e+02 4.597e+02, threshold=3.877e+02, percent-clipped=2.0 2023-03-26 13:21:56,081 INFO [finetune.py:976] (3/7) Epoch 11, batch 3100, loss[loss=0.1913, simple_loss=0.2648, pruned_loss=0.05892, over 4769.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2629, pruned_loss=0.06499, over 957670.95 frames. ], batch size: 26, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:22:03,113 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-03-26 13:22:08,703 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60396.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 13:22:16,695 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60408.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 13:22:25,066 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60421.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:22:29,560 INFO [finetune.py:976] (3/7) Epoch 11, batch 3150, loss[loss=0.2201, simple_loss=0.2705, pruned_loss=0.08492, over 4805.00 frames. ], tot_loss[loss=0.1948, simple_loss=0.26, pruned_loss=0.06481, over 957146.27 frames. ], batch size: 25, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:22:34,360 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60434.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:22:34,872 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.045e+02 1.624e+02 1.838e+02 2.200e+02 4.980e+02, threshold=3.676e+02, percent-clipped=1.0 2023-03-26 13:22:52,332 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8445, 1.6467, 1.4615, 1.2533, 1.6386, 1.6449, 1.5981, 2.1536], device='cuda:3'), covar=tensor([0.4429, 0.4493, 0.3564, 0.4281, 0.4225, 0.2557, 0.4054, 0.2089], device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0258, 0.0222, 0.0277, 0.0242, 0.0208, 0.0245, 0.0214], 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') 2023-03-26 13:22:52,952 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2114, 1.7757, 2.2148, 2.1114, 1.8450, 1.8387, 1.9971, 2.0082], device='cuda:3'), covar=tensor([0.4393, 0.4970, 0.3935, 0.4910, 0.6112, 0.4646, 0.5999, 0.3980], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0240, 0.0255, 0.0260, 0.0256, 0.0231, 0.0275, 0.0233], 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') 2023-03-26 13:23:01,695 INFO [finetune.py:976] (3/7) Epoch 11, batch 3200, loss[loss=0.1874, simple_loss=0.2495, pruned_loss=0.0626, over 4900.00 frames. ], tot_loss[loss=0.1913, simple_loss=0.2562, pruned_loss=0.06319, over 958261.17 frames. ], batch size: 32, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:23:20,583 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60504.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:23:24,813 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.0520, 4.7013, 4.3877, 2.8964, 4.8144, 3.5289, 1.0376, 3.2766], device='cuda:3'), covar=tensor([0.2241, 0.1818, 0.1451, 0.2790, 0.0698, 0.0986, 0.4490, 0.1353], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0174, 0.0160, 0.0128, 0.0156, 0.0122, 0.0146, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 13:23:37,313 INFO [finetune.py:976] (3/7) Epoch 11, batch 3250, loss[loss=0.248, simple_loss=0.3223, pruned_loss=0.08684, over 4844.00 frames. ], tot_loss[loss=0.1931, simple_loss=0.2577, pruned_loss=0.06422, over 954054.17 frames. ], batch size: 47, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:23:48,950 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.174e+02 1.626e+02 1.982e+02 2.397e+02 3.737e+02, threshold=3.964e+02, percent-clipped=1.0 2023-03-26 13:23:59,842 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60545.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:24:00,470 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6673, 2.4511, 2.3842, 1.6042, 2.4728, 2.0324, 1.9096, 2.2914], device='cuda:3'), covar=tensor([0.1223, 0.0656, 0.1462, 0.1604, 0.1471, 0.1644, 0.1731, 0.0953], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0200, 0.0202, 0.0187, 0.0216, 0.0210, 0.0224, 0.0198], 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') 2023-03-26 13:24:04,034 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=60552.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:24:05,271 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60554.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:24:09,020 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 2023-03-26 13:24:27,367 INFO [finetune.py:976] (3/7) Epoch 11, batch 3300, loss[loss=0.1977, simple_loss=0.2718, pruned_loss=0.06173, over 4179.00 frames. ], tot_loss[loss=0.196, simple_loss=0.261, pruned_loss=0.06552, over 951135.04 frames. ], batch size: 65, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:24:45,616 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=60593.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:24:45,672 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9000, 1.7596, 1.8803, 1.0817, 1.8404, 1.9345, 1.7926, 1.4979], device='cuda:3'), covar=tensor([0.0468, 0.0674, 0.0618, 0.0931, 0.0709, 0.0635, 0.0626, 0.1165], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0134, 0.0142, 0.0125, 0.0121, 0.0144, 0.0145, 0.0164], 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') 2023-03-26 13:25:28,938 INFO [finetune.py:976] (3/7) Epoch 11, batch 3350, loss[loss=0.1514, simple_loss=0.2179, pruned_loss=0.04251, over 4720.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2621, pruned_loss=0.0656, over 953192.04 frames. ], batch size: 23, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:25:34,910 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.073e+02 1.701e+02 2.036e+02 2.469e+02 3.577e+02, threshold=4.071e+02, percent-clipped=0.0 2023-03-26 13:25:52,226 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0622, 1.9217, 1.6125, 1.9180, 1.9806, 1.6866, 2.2907, 2.0294], device='cuda:3'), covar=tensor([0.1339, 0.2378, 0.3054, 0.2759, 0.2578, 0.1673, 0.3019, 0.1839], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0186, 0.0231, 0.0253, 0.0238, 0.0196, 0.0210, 0.0196], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 13:26:02,928 INFO [finetune.py:976] (3/7) Epoch 11, batch 3400, loss[loss=0.2048, simple_loss=0.2814, pruned_loss=0.0641, over 4913.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.264, pruned_loss=0.06621, over 953252.51 frames. ], batch size: 37, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:26:16,518 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60696.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:26:24,736 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60708.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:26:30,775 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60718.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:26:32,538 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60721.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:26:36,599 INFO [finetune.py:976] (3/7) Epoch 11, batch 3450, loss[loss=0.1909, simple_loss=0.2522, pruned_loss=0.06483, over 4759.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2642, pruned_loss=0.06576, over 953851.96 frames. ], batch size: 26, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:26:41,004 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60734.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:26:41,141 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-26 13:26:41,510 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.902e+01 1.594e+02 1.892e+02 2.253e+02 3.493e+02, threshold=3.783e+02, percent-clipped=0.0 2023-03-26 13:26:52,730 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=60744.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:27:06,423 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=60756.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:27:25,457 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=60769.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:27:36,413 INFO [finetune.py:976] (3/7) Epoch 11, batch 3500, loss[loss=0.1788, simple_loss=0.243, pruned_loss=0.05733, over 4938.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.262, pruned_loss=0.06523, over 955400.49 frames. ], batch size: 38, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:27:37,748 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60779.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:27:39,439 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=60782.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:28:12,294 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.2365, 1.2507, 1.3373, 0.5606, 1.2206, 1.5147, 1.4991, 1.2521], device='cuda:3'), covar=tensor([0.1039, 0.0779, 0.0503, 0.0606, 0.0523, 0.0526, 0.0431, 0.0731], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0152, 0.0121, 0.0132, 0.0129, 0.0124, 0.0142, 0.0146], device='cuda:3'), out_proj_covar=tensor([9.3620e-05, 1.1167e-04, 8.7093e-05, 9.5406e-05, 9.2305e-05, 9.0382e-05, 1.0395e-04, 1.0654e-04], device='cuda:3') 2023-03-26 13:28:15,227 INFO [finetune.py:976] (3/7) Epoch 11, batch 3550, loss[loss=0.1475, simple_loss=0.2191, pruned_loss=0.0379, over 4826.00 frames. ], tot_loss[loss=0.194, simple_loss=0.259, pruned_loss=0.06448, over 955626.52 frames. ], batch size: 41, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:28:20,663 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.181e+02 1.566e+02 1.863e+02 2.348e+02 4.575e+02, threshold=3.726e+02, percent-clipped=4.0 2023-03-26 13:28:34,185 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60854.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:28:34,214 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1028, 1.9586, 1.9866, 1.2367, 2.0180, 2.0644, 1.9870, 1.6918], device='cuda:3'), covar=tensor([0.0532, 0.0633, 0.0708, 0.0930, 0.0642, 0.0717, 0.0608, 0.1117], device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0134, 0.0142, 0.0125, 0.0121, 0.0144, 0.0145, 0.0163], 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') 2023-03-26 13:28:49,073 INFO [finetune.py:976] (3/7) Epoch 11, batch 3600, loss[loss=0.2571, simple_loss=0.2966, pruned_loss=0.1088, over 4193.00 frames. ], tot_loss[loss=0.1927, simple_loss=0.257, pruned_loss=0.06424, over 953200.58 frames. ], batch size: 65, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:29:17,738 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=60902.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:29:39,498 INFO [finetune.py:976] (3/7) Epoch 11, batch 3650, loss[loss=0.2192, simple_loss=0.2943, pruned_loss=0.07204, over 4924.00 frames. ], tot_loss[loss=0.1944, simple_loss=0.259, pruned_loss=0.06495, over 952966.98 frames. ], batch size: 42, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:29:44,367 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.012e+02 1.638e+02 1.962e+02 2.312e+02 3.604e+02, threshold=3.924e+02, percent-clipped=0.0 2023-03-26 13:30:33,789 INFO [finetune.py:976] (3/7) Epoch 11, batch 3700, loss[loss=0.2161, simple_loss=0.2864, pruned_loss=0.07289, over 4140.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.262, pruned_loss=0.06567, over 952084.20 frames. ], batch size: 65, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:31:06,873 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5487, 2.3151, 1.7135, 0.7993, 1.9621, 2.1230, 1.8861, 2.0255], device='cuda:3'), covar=tensor([0.0828, 0.0762, 0.1582, 0.2000, 0.1242, 0.1918, 0.2063, 0.0869], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0200, 0.0202, 0.0186, 0.0215, 0.0209, 0.0223, 0.0197], 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') 2023-03-26 13:31:15,833 INFO [finetune.py:976] (3/7) Epoch 11, batch 3750, loss[loss=0.2448, simple_loss=0.3064, pruned_loss=0.09165, over 4820.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2625, pruned_loss=0.06528, over 952392.42 frames. ], batch size: 39, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:31:20,653 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.055e+02 1.587e+02 1.819e+02 2.276e+02 4.586e+02, threshold=3.638e+02, percent-clipped=1.0 2023-03-26 13:31:38,971 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61061.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:31:47,661 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61074.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:31:49,400 INFO [finetune.py:976] (3/7) Epoch 11, batch 3800, loss[loss=0.2013, simple_loss=0.2572, pruned_loss=0.07268, over 4862.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2634, pruned_loss=0.06551, over 951473.92 frames. ], batch size: 34, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:32:29,720 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61122.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:32:32,632 INFO [finetune.py:976] (3/7) Epoch 11, batch 3850, loss[loss=0.1676, simple_loss=0.2401, pruned_loss=0.04756, over 4848.00 frames. ], tot_loss[loss=0.195, simple_loss=0.2613, pruned_loss=0.0643, over 952701.17 frames. ], batch size: 44, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:32:37,920 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.075e+02 1.518e+02 1.864e+02 2.279e+02 4.215e+02, threshold=3.727e+02, percent-clipped=1.0 2023-03-26 13:33:02,470 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6971, 1.4241, 2.1789, 1.3643, 1.7593, 1.9394, 1.3773, 2.0792], device='cuda:3'), covar=tensor([0.1367, 0.1892, 0.0974, 0.1651, 0.0887, 0.1382, 0.2539, 0.0952], device='cuda:3'), in_proj_covar=tensor([0.0199, 0.0205, 0.0194, 0.0192, 0.0179, 0.0216, 0.0218, 0.0202], 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') 2023-03-26 13:33:04,306 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.3596, 1.5238, 1.6282, 0.8118, 1.5535, 1.7835, 1.8515, 1.4349], device='cuda:3'), covar=tensor([0.0871, 0.0630, 0.0506, 0.0559, 0.0440, 0.0620, 0.0336, 0.0639], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0154, 0.0122, 0.0132, 0.0130, 0.0125, 0.0144, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.4418e-05, 1.1274e-04, 8.7576e-05, 9.5773e-05, 9.3040e-05, 9.1142e-05, 1.0518e-04, 1.0719e-04], device='cuda:3') 2023-03-26 13:33:05,948 INFO [finetune.py:976] (3/7) Epoch 11, batch 3900, loss[loss=0.1715, simple_loss=0.2371, pruned_loss=0.05288, over 4779.00 frames. ], tot_loss[loss=0.1939, simple_loss=0.2594, pruned_loss=0.06422, over 953788.36 frames. ], batch size: 29, lr: 3.69e-03, grad_scale: 32.0 2023-03-26 13:33:39,748 INFO [finetune.py:976] (3/7) Epoch 11, batch 3950, loss[loss=0.1882, simple_loss=0.2527, pruned_loss=0.06181, over 4711.00 frames. ], tot_loss[loss=0.1907, simple_loss=0.2557, pruned_loss=0.06285, over 954515.99 frames. ], batch size: 59, lr: 3.68e-03, grad_scale: 32.0 2023-03-26 13:33:45,060 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.129e+02 1.570e+02 1.907e+02 2.309e+02 4.377e+02, threshold=3.813e+02, percent-clipped=3.0 2023-03-26 13:33:49,051 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-03-26 13:33:54,389 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.73 vs. limit=5.0 2023-03-26 13:34:03,118 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7170, 1.5896, 1.5032, 1.8211, 2.0599, 1.9135, 1.3931, 1.4634], device='cuda:3'), covar=tensor([0.2340, 0.2173, 0.2103, 0.1687, 0.1800, 0.1241, 0.2602, 0.1931], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0206, 0.0208, 0.0188, 0.0240, 0.0181, 0.0211, 0.0195], 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') 2023-03-26 13:34:12,191 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-26 13:34:12,380 INFO [finetune.py:976] (3/7) Epoch 11, batch 4000, loss[loss=0.1534, simple_loss=0.2075, pruned_loss=0.04967, over 4030.00 frames. ], tot_loss[loss=0.1915, simple_loss=0.2556, pruned_loss=0.06369, over 949541.84 frames. ], batch size: 17, lr: 3.68e-03, grad_scale: 64.0 2023-03-26 13:34:55,752 INFO [finetune.py:976] (3/7) Epoch 11, batch 4050, loss[loss=0.1883, simple_loss=0.265, pruned_loss=0.05574, over 4779.00 frames. ], tot_loss[loss=0.1937, simple_loss=0.2587, pruned_loss=0.06433, over 949322.30 frames. ], batch size: 29, lr: 3.68e-03, grad_scale: 64.0 2023-03-26 13:35:04,888 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.154e+02 1.652e+02 2.086e+02 2.571e+02 4.987e+02, threshold=4.171e+02, percent-clipped=6.0 2023-03-26 13:35:10,642 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-26 13:35:23,169 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6325, 3.3574, 3.1243, 1.3253, 3.4228, 2.4911, 0.8582, 2.1303], device='cuda:3'), covar=tensor([0.2343, 0.2222, 0.1703, 0.3555, 0.1253, 0.1189, 0.4183, 0.1745], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0175, 0.0160, 0.0129, 0.0156, 0.0122, 0.0147, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 13:35:33,871 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4649, 1.4814, 1.8872, 1.8481, 1.5652, 3.4212, 1.3221, 1.5759], device='cuda:3'), covar=tensor([0.1000, 0.1919, 0.1049, 0.0975, 0.1701, 0.0239, 0.1568, 0.1764], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0074, 0.0077, 0.0091, 0.0081, 0.0085, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 13:35:36,712 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5752, 1.6289, 2.2351, 2.0278, 1.8675, 3.5648, 1.4783, 1.8618], device='cuda:3'), covar=tensor([0.0903, 0.1591, 0.1333, 0.0832, 0.1344, 0.0274, 0.1356, 0.1474], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0074, 0.0077, 0.0091, 0.0081, 0.0085, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 13:35:36,767 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1362, 2.0416, 1.8399, 2.3260, 2.0141, 2.0097, 2.0128, 2.6846], device='cuda:3'), covar=tensor([0.4092, 0.5558, 0.3535, 0.4420, 0.4911, 0.2437, 0.4882, 0.1672], device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0260, 0.0223, 0.0278, 0.0243, 0.0210, 0.0245, 0.0215], 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') 2023-03-26 13:35:41,813 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61374.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:35:43,534 INFO [finetune.py:976] (3/7) Epoch 11, batch 4100, loss[loss=0.1631, simple_loss=0.2409, pruned_loss=0.04267, over 4807.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.2611, pruned_loss=0.06511, over 947623.12 frames. ], batch size: 45, lr: 3.68e-03, grad_scale: 64.0 2023-03-26 13:35:47,820 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6966, 0.7441, 1.6154, 1.4899, 1.3994, 1.3629, 1.4090, 1.5312], device='cuda:3'), covar=tensor([0.2917, 0.3360, 0.2836, 0.3083, 0.3813, 0.3073, 0.3802, 0.2762], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0239, 0.0254, 0.0259, 0.0256, 0.0231, 0.0275, 0.0232], 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') 2023-03-26 13:36:11,797 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61417.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:36:20,057 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=61422.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:36:26,637 INFO [finetune.py:976] (3/7) Epoch 11, batch 4150, loss[loss=0.1907, simple_loss=0.2634, pruned_loss=0.05899, over 4833.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2622, pruned_loss=0.06535, over 948414.50 frames. ], batch size: 49, lr: 3.68e-03, grad_scale: 32.0 2023-03-26 13:36:32,502 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.161e+02 1.629e+02 1.982e+02 2.519e+02 5.426e+02, threshold=3.964e+02, percent-clipped=4.0 2023-03-26 13:36:36,839 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6494, 1.4701, 1.0815, 0.2742, 1.2396, 1.4606, 1.3622, 1.4064], device='cuda:3'), covar=tensor([0.0901, 0.0826, 0.1357, 0.1973, 0.1419, 0.2170, 0.2279, 0.0815], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0197, 0.0199, 0.0184, 0.0214, 0.0205, 0.0220, 0.0195], 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') 2023-03-26 13:36:59,827 INFO [finetune.py:976] (3/7) Epoch 11, batch 4200, loss[loss=0.1776, simple_loss=0.2527, pruned_loss=0.05125, over 4804.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2635, pruned_loss=0.06527, over 951511.06 frames. ], batch size: 40, lr: 3.68e-03, grad_scale: 32.0 2023-03-26 13:37:35,243 INFO [finetune.py:976] (3/7) Epoch 11, batch 4250, loss[loss=0.1542, simple_loss=0.2214, pruned_loss=0.04354, over 4693.00 frames. ], tot_loss[loss=0.1941, simple_loss=0.2602, pruned_loss=0.06404, over 952590.99 frames. ], batch size: 59, lr: 3.68e-03, grad_scale: 32.0 2023-03-26 13:37:35,984 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8985, 1.8118, 1.6792, 2.0566, 2.3146, 2.1295, 1.5155, 1.5686], device='cuda:3'), covar=tensor([0.2389, 0.2162, 0.1972, 0.1743, 0.1793, 0.1115, 0.2630, 0.2095], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0207, 0.0209, 0.0189, 0.0241, 0.0182, 0.0212, 0.0197], 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') 2023-03-26 13:37:45,939 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.771e+01 1.547e+02 1.858e+02 2.245e+02 5.805e+02, threshold=3.715e+02, percent-clipped=2.0 2023-03-26 13:37:53,200 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-26 13:38:15,490 INFO [finetune.py:976] (3/7) Epoch 11, batch 4300, loss[loss=0.2217, simple_loss=0.2858, pruned_loss=0.07876, over 4824.00 frames. ], tot_loss[loss=0.1911, simple_loss=0.2565, pruned_loss=0.06286, over 951745.89 frames. ], batch size: 39, lr: 3.68e-03, grad_scale: 32.0 2023-03-26 13:38:45,455 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7229, 1.5446, 2.0001, 2.7958, 1.9739, 2.1170, 1.2751, 2.2637], device='cuda:3'), covar=tensor([0.1327, 0.1278, 0.1005, 0.0597, 0.0731, 0.1887, 0.1299, 0.0583], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0132, 0.0163, 0.0100, 0.0136, 0.0124, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 13:38:48,429 INFO [finetune.py:976] (3/7) Epoch 11, batch 4350, loss[loss=0.1744, simple_loss=0.252, pruned_loss=0.04838, over 4914.00 frames. ], tot_loss[loss=0.1887, simple_loss=0.2541, pruned_loss=0.0617, over 950673.47 frames. ], batch size: 36, lr: 3.68e-03, grad_scale: 32.0 2023-03-26 13:38:54,826 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.000e+02 1.580e+02 1.801e+02 2.212e+02 3.446e+02, threshold=3.603e+02, percent-clipped=0.0 2023-03-26 13:39:21,855 INFO [finetune.py:976] (3/7) Epoch 11, batch 4400, loss[loss=0.2259, simple_loss=0.265, pruned_loss=0.09336, over 4422.00 frames. ], tot_loss[loss=0.1907, simple_loss=0.2556, pruned_loss=0.06292, over 951379.57 frames. ], batch size: 19, lr: 3.68e-03, grad_scale: 32.0 2023-03-26 13:39:26,768 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9378, 3.9224, 3.6891, 1.8021, 3.9900, 3.0251, 0.7950, 2.5706], device='cuda:3'), covar=tensor([0.2250, 0.2024, 0.1471, 0.3658, 0.0948, 0.0995, 0.4688, 0.1631], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0173, 0.0158, 0.0128, 0.0154, 0.0121, 0.0145, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 13:39:33,572 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3721, 0.8883, 0.8324, 1.2984, 1.8025, 0.7121, 1.1628, 1.2784], device='cuda:3'), covar=tensor([0.1619, 0.2462, 0.1866, 0.1380, 0.2145, 0.2023, 0.1604, 0.2042], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0096, 0.0112, 0.0093, 0.0119, 0.0094, 0.0099, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 13:39:53,801 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61717.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:39:56,383 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-26 13:40:04,334 INFO [finetune.py:976] (3/7) Epoch 11, batch 4450, loss[loss=0.1963, simple_loss=0.2674, pruned_loss=0.0626, over 4857.00 frames. ], tot_loss[loss=0.1943, simple_loss=0.2599, pruned_loss=0.06431, over 952447.80 frames. ], batch size: 49, lr: 3.68e-03, grad_scale: 32.0 2023-03-26 13:40:07,501 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61732.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:40:14,299 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.223e+02 1.628e+02 1.977e+02 2.534e+02 3.640e+02, threshold=3.954e+02, percent-clipped=2.0 2023-03-26 13:40:23,520 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61743.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:40:30,719 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.44 vs. limit=5.0 2023-03-26 13:40:44,697 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9086, 1.8354, 1.7553, 1.8488, 1.5967, 3.7711, 1.7557, 2.2979], device='cuda:3'), covar=tensor([0.3822, 0.2641, 0.2134, 0.2722, 0.1652, 0.0278, 0.2277, 0.1104], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0119, 0.0123, 0.0114, 0.0098, 0.0098, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 13:40:49,786 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=61765.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:40:57,017 INFO [finetune.py:976] (3/7) Epoch 11, batch 4500, loss[loss=0.2038, simple_loss=0.2715, pruned_loss=0.06804, over 4878.00 frames. ], tot_loss[loss=0.1954, simple_loss=0.2613, pruned_loss=0.06477, over 951158.87 frames. ], batch size: 34, lr: 3.68e-03, grad_scale: 32.0 2023-03-26 13:41:07,206 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7390, 3.9639, 3.6386, 1.7315, 4.0291, 3.0075, 0.8949, 2.5213], device='cuda:3'), covar=tensor([0.2130, 0.1723, 0.1655, 0.3454, 0.0868, 0.0968, 0.4213, 0.1538], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0174, 0.0159, 0.0128, 0.0155, 0.0122, 0.0145, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 13:41:07,849 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61793.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:41:16,081 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61804.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:41:33,079 INFO [finetune.py:976] (3/7) Epoch 11, batch 4550, loss[loss=0.1573, simple_loss=0.2245, pruned_loss=0.04502, over 4738.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2621, pruned_loss=0.06531, over 951425.28 frames. ], batch size: 23, lr: 3.68e-03, grad_scale: 32.0 2023-03-26 13:41:43,495 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.535e+01 1.607e+02 1.951e+02 2.245e+02 3.846e+02, threshold=3.902e+02, percent-clipped=0.0 2023-03-26 13:41:58,789 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7096, 2.5702, 2.2446, 2.6129, 2.6290, 2.5018, 2.9232, 2.7098], device='cuda:3'), covar=tensor([0.1171, 0.1992, 0.2918, 0.2345, 0.2301, 0.1478, 0.2676, 0.1703], device='cuda:3'), in_proj_covar=tensor([0.0175, 0.0186, 0.0233, 0.0253, 0.0239, 0.0197, 0.0212, 0.0195], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 13:42:05,352 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-03-26 13:42:15,222 INFO [finetune.py:976] (3/7) Epoch 11, batch 4600, loss[loss=0.1713, simple_loss=0.2475, pruned_loss=0.04751, over 4900.00 frames. ], tot_loss[loss=0.1954, simple_loss=0.2615, pruned_loss=0.06469, over 951351.02 frames. ], batch size: 37, lr: 3.68e-03, grad_scale: 32.0 2023-03-26 13:42:38,000 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4192, 1.4729, 1.8986, 1.8134, 1.6445, 3.3333, 1.3923, 1.6709], device='cuda:3'), covar=tensor([0.0943, 0.1656, 0.1182, 0.0903, 0.1445, 0.0267, 0.1331, 0.1566], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0074, 0.0077, 0.0091, 0.0080, 0.0084, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 13:42:40,464 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61914.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:42:45,378 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-03-26 13:42:48,594 INFO [finetune.py:976] (3/7) Epoch 11, batch 4650, loss[loss=0.2308, simple_loss=0.2852, pruned_loss=0.08824, over 4848.00 frames. ], tot_loss[loss=0.1938, simple_loss=0.2594, pruned_loss=0.0641, over 953367.41 frames. ], batch size: 44, lr: 3.68e-03, grad_scale: 32.0 2023-03-26 13:42:56,047 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.057e+02 1.606e+02 1.934e+02 2.317e+02 5.626e+02, threshold=3.867e+02, percent-clipped=3.0 2023-03-26 13:43:31,739 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61975.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:43:32,825 INFO [finetune.py:976] (3/7) Epoch 11, batch 4700, loss[loss=0.2041, simple_loss=0.2571, pruned_loss=0.07551, over 4817.00 frames. ], tot_loss[loss=0.1911, simple_loss=0.2564, pruned_loss=0.06283, over 954561.23 frames. ], batch size: 40, lr: 3.68e-03, grad_scale: 32.0 2023-03-26 13:43:44,846 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7467, 1.7025, 1.5014, 1.9020, 2.1014, 1.9569, 1.5365, 1.4338], device='cuda:3'), covar=tensor([0.2069, 0.1905, 0.1815, 0.1499, 0.1813, 0.1106, 0.2381, 0.1817], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0207, 0.0208, 0.0189, 0.0242, 0.0181, 0.0212, 0.0196], 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') 2023-03-26 13:44:19,772 INFO [finetune.py:976] (3/7) Epoch 11, batch 4750, loss[loss=0.2041, simple_loss=0.27, pruned_loss=0.06905, over 4809.00 frames. ], tot_loss[loss=0.1895, simple_loss=0.2544, pruned_loss=0.06233, over 956173.29 frames. ], batch size: 41, lr: 3.68e-03, grad_scale: 32.0 2023-03-26 13:44:25,603 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.066e+02 1.474e+02 1.769e+02 2.148e+02 4.944e+02, threshold=3.539e+02, percent-clipped=1.0 2023-03-26 13:44:53,406 INFO [finetune.py:976] (3/7) Epoch 11, batch 4800, loss[loss=0.1873, simple_loss=0.2447, pruned_loss=0.06499, over 4735.00 frames. ], tot_loss[loss=0.1915, simple_loss=0.2566, pruned_loss=0.06318, over 957362.10 frames. ], batch size: 23, lr: 3.68e-03, grad_scale: 32.0 2023-03-26 13:45:06,490 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62088.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:45:13,195 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62099.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:45:40,255 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-26 13:45:41,876 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62122.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:45:50,591 INFO [finetune.py:976] (3/7) Epoch 11, batch 4850, loss[loss=0.1933, simple_loss=0.273, pruned_loss=0.05674, over 4827.00 frames. ], tot_loss[loss=0.1935, simple_loss=0.2589, pruned_loss=0.06402, over 954074.59 frames. ], batch size: 39, lr: 3.68e-03, grad_scale: 32.0 2023-03-26 13:46:01,539 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.225e+02 1.730e+02 2.037e+02 2.587e+02 8.043e+02, threshold=4.075e+02, percent-clipped=4.0 2023-03-26 13:46:34,572 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-26 13:46:45,228 INFO [finetune.py:976] (3/7) Epoch 11, batch 4900, loss[loss=0.1799, simple_loss=0.2666, pruned_loss=0.04659, over 4833.00 frames. ], tot_loss[loss=0.1955, simple_loss=0.2609, pruned_loss=0.06503, over 954139.31 frames. ], batch size: 47, lr: 3.68e-03, grad_scale: 32.0 2023-03-26 13:46:46,630 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-26 13:46:48,909 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62183.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:47:49,179 INFO [finetune.py:976] (3/7) Epoch 11, batch 4950, loss[loss=0.1999, simple_loss=0.249, pruned_loss=0.07537, over 4705.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.263, pruned_loss=0.06555, over 956356.73 frames. ], batch size: 23, lr: 3.68e-03, grad_scale: 32.0 2023-03-26 13:47:56,654 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.284e+02 1.728e+02 2.029e+02 2.471e+02 5.736e+02, threshold=4.057e+02, percent-clipped=2.0 2023-03-26 13:48:18,909 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62270.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:48:21,274 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62273.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:48:24,031 INFO [finetune.py:976] (3/7) Epoch 11, batch 5000, loss[loss=0.1825, simple_loss=0.2369, pruned_loss=0.06404, over 4781.00 frames. ], tot_loss[loss=0.1956, simple_loss=0.2612, pruned_loss=0.06495, over 957304.47 frames. ], batch size: 26, lr: 3.68e-03, grad_scale: 32.0 2023-03-26 13:48:54,765 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-26 13:48:55,448 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2661, 2.1100, 1.5021, 0.7416, 1.6870, 1.9208, 1.7922, 1.9359], device='cuda:3'), covar=tensor([0.1018, 0.0811, 0.1414, 0.1854, 0.1341, 0.2237, 0.2225, 0.0791], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0198, 0.0201, 0.0185, 0.0215, 0.0207, 0.0222, 0.0196], 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') 2023-03-26 13:48:57,126 INFO [finetune.py:976] (3/7) Epoch 11, batch 5050, loss[loss=0.2111, simple_loss=0.2719, pruned_loss=0.07519, over 4780.00 frames. ], tot_loss[loss=0.1921, simple_loss=0.2576, pruned_loss=0.06333, over 956744.12 frames. ], batch size: 51, lr: 3.68e-03, grad_scale: 16.0 2023-03-26 13:49:02,467 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62334.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:49:04,172 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.067e+02 1.504e+02 1.759e+02 2.068e+02 4.473e+02, threshold=3.518e+02, percent-clipped=1.0 2023-03-26 13:49:32,190 INFO [finetune.py:976] (3/7) Epoch 11, batch 5100, loss[loss=0.1728, simple_loss=0.2367, pruned_loss=0.05444, over 3948.00 frames. ], tot_loss[loss=0.1901, simple_loss=0.2551, pruned_loss=0.06261, over 954991.62 frames. ], batch size: 17, lr: 3.68e-03, grad_scale: 16.0 2023-03-26 13:49:37,102 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6007, 1.4808, 2.1814, 3.0968, 2.1940, 2.2376, 1.3651, 2.4054], device='cuda:3'), covar=tensor([0.1570, 0.1362, 0.1114, 0.0535, 0.0723, 0.1793, 0.1458, 0.0533], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0133, 0.0162, 0.0100, 0.0136, 0.0124, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 13:49:40,052 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62388.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:49:42,302 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62391.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:49:44,394 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.25 vs. limit=5.0 2023-03-26 13:49:47,626 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62399.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:49:54,268 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-03-26 13:50:05,685 INFO [finetune.py:976] (3/7) Epoch 11, batch 5150, loss[loss=0.1788, simple_loss=0.2663, pruned_loss=0.0456, over 4926.00 frames. ], tot_loss[loss=0.1901, simple_loss=0.2552, pruned_loss=0.06253, over 951898.87 frames. ], batch size: 38, lr: 3.67e-03, grad_scale: 16.0 2023-03-26 13:50:12,136 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=62436.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:50:12,674 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.110e+02 1.578e+02 2.001e+02 2.432e+02 3.455e+02, threshold=4.003e+02, percent-clipped=0.0 2023-03-26 13:50:26,775 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=62447.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:50:29,153 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3370, 2.9263, 2.8371, 1.3196, 3.0670, 2.2216, 0.8526, 1.8884], device='cuda:3'), covar=tensor([0.2212, 0.2470, 0.1944, 0.3551, 0.1391, 0.1166, 0.4127, 0.1709], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0173, 0.0158, 0.0128, 0.0155, 0.0121, 0.0145, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 13:50:30,447 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62452.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:50:55,269 INFO [finetune.py:976] (3/7) Epoch 11, batch 5200, loss[loss=0.2962, simple_loss=0.3456, pruned_loss=0.1234, over 4271.00 frames. ], tot_loss[loss=0.1951, simple_loss=0.2609, pruned_loss=0.0647, over 953158.32 frames. ], batch size: 65, lr: 3.67e-03, grad_scale: 16.0 2023-03-26 13:50:56,958 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62478.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:51:08,481 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.73 vs. limit=5.0 2023-03-26 13:51:36,855 INFO [finetune.py:976] (3/7) Epoch 11, batch 5250, loss[loss=0.1638, simple_loss=0.2409, pruned_loss=0.04341, over 4931.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2631, pruned_loss=0.0657, over 952757.08 frames. ], batch size: 38, lr: 3.67e-03, grad_scale: 16.0 2023-03-26 13:51:42,821 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8712, 1.8316, 1.6005, 1.9882, 2.4571, 2.0620, 1.7752, 1.5284], device='cuda:3'), covar=tensor([0.2326, 0.2062, 0.1976, 0.1613, 0.1675, 0.1105, 0.2244, 0.2022], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0208, 0.0209, 0.0188, 0.0242, 0.0182, 0.0212, 0.0197], 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') 2023-03-26 13:51:54,382 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.247e+02 1.618e+02 1.949e+02 2.406e+02 7.235e+02, threshold=3.897e+02, percent-clipped=3.0 2023-03-26 13:52:03,531 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62545.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:52:19,511 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62570.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:52:23,690 INFO [finetune.py:976] (3/7) Epoch 11, batch 5300, loss[loss=0.2245, simple_loss=0.2917, pruned_loss=0.0787, over 4741.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2644, pruned_loss=0.06548, over 954982.69 frames. ], batch size: 54, lr: 3.67e-03, grad_scale: 16.0 2023-03-26 13:52:29,617 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62585.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:52:44,285 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62606.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:52:52,191 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=62618.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:52:57,603 INFO [finetune.py:976] (3/7) Epoch 11, batch 5350, loss[loss=0.2166, simple_loss=0.2835, pruned_loss=0.07488, over 4910.00 frames. ], tot_loss[loss=0.1961, simple_loss=0.2627, pruned_loss=0.06472, over 953842.63 frames. ], batch size: 43, lr: 3.67e-03, grad_scale: 16.0 2023-03-26 13:52:58,901 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62629.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:53:04,200 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.067e+02 1.504e+02 1.845e+02 2.238e+02 3.589e+02, threshold=3.690e+02, percent-clipped=0.0 2023-03-26 13:53:10,770 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62646.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:53:24,292 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62666.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:53:30,762 INFO [finetune.py:976] (3/7) Epoch 11, batch 5400, loss[loss=0.1766, simple_loss=0.2418, pruned_loss=0.05572, over 4898.00 frames. ], tot_loss[loss=0.1948, simple_loss=0.261, pruned_loss=0.06435, over 954799.20 frames. ], batch size: 35, lr: 3.67e-03, grad_scale: 16.0 2023-03-26 13:53:53,373 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-03-26 13:54:04,661 INFO [finetune.py:976] (3/7) Epoch 11, batch 5450, loss[loss=0.1332, simple_loss=0.2059, pruned_loss=0.0302, over 4827.00 frames. ], tot_loss[loss=0.1929, simple_loss=0.2586, pruned_loss=0.06359, over 953211.34 frames. ], batch size: 51, lr: 3.67e-03, grad_scale: 16.0 2023-03-26 13:54:04,773 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62727.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:54:10,765 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.513e+01 1.463e+02 1.876e+02 2.335e+02 4.427e+02, threshold=3.751e+02, percent-clipped=2.0 2023-03-26 13:54:14,970 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3684, 1.3939, 1.5519, 1.6301, 1.6329, 3.0347, 1.4015, 1.5981], device='cuda:3'), covar=tensor([0.0969, 0.1802, 0.0986, 0.0950, 0.1515, 0.0282, 0.1427, 0.1693], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0075, 0.0078, 0.0092, 0.0081, 0.0085, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 13:54:17,807 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62747.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:54:28,426 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8718, 1.6391, 1.5050, 1.8038, 1.9578, 1.9355, 1.2374, 1.5522], device='cuda:3'), covar=tensor([0.2241, 0.2143, 0.2173, 0.1771, 0.1793, 0.1134, 0.2700, 0.1924], device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0208, 0.0209, 0.0189, 0.0243, 0.0182, 0.0213, 0.0197], 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') 2023-03-26 13:54:36,209 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62773.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:54:38,553 INFO [finetune.py:976] (3/7) Epoch 11, batch 5500, loss[loss=0.2024, simple_loss=0.2621, pruned_loss=0.07133, over 4829.00 frames. ], tot_loss[loss=0.191, simple_loss=0.2562, pruned_loss=0.06288, over 955395.78 frames. ], batch size: 40, lr: 3.67e-03, grad_scale: 16.0 2023-03-26 13:54:39,230 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62778.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:54:42,267 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6036, 1.4650, 1.8732, 1.9518, 1.5838, 3.5153, 1.4212, 1.6365], device='cuda:3'), covar=tensor([0.0964, 0.1848, 0.1012, 0.0913, 0.1680, 0.0246, 0.1509, 0.1708], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0074, 0.0078, 0.0092, 0.0081, 0.0085, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 13:55:11,797 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8756, 1.3080, 1.9084, 1.8672, 1.6065, 1.6001, 1.8041, 1.7365], device='cuda:3'), covar=tensor([0.4291, 0.4886, 0.4095, 0.4128, 0.5563, 0.4107, 0.5437, 0.3933], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0238, 0.0253, 0.0259, 0.0255, 0.0230, 0.0274, 0.0232], 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') 2023-03-26 13:55:12,350 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=62826.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:55:12,912 INFO [finetune.py:976] (3/7) Epoch 11, batch 5550, loss[loss=0.2167, simple_loss=0.2938, pruned_loss=0.06979, over 4854.00 frames. ], tot_loss[loss=0.1923, simple_loss=0.2578, pruned_loss=0.06337, over 956515.95 frames. ], batch size: 49, lr: 3.67e-03, grad_scale: 16.0 2023-03-26 13:55:17,991 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62834.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:55:19,877 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.162e+02 1.580e+02 1.841e+02 2.336e+02 5.980e+02, threshold=3.683e+02, percent-clipped=6.0 2023-03-26 13:55:30,346 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.1187, 4.6113, 4.5088, 2.4404, 4.7348, 3.5911, 0.7760, 3.2271], device='cuda:3'), covar=tensor([0.2106, 0.1979, 0.1246, 0.2975, 0.0879, 0.0819, 0.4494, 0.1432], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0173, 0.0158, 0.0128, 0.0155, 0.0121, 0.0145, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 13:56:07,695 INFO [finetune.py:976] (3/7) Epoch 11, batch 5600, loss[loss=0.1942, simple_loss=0.2637, pruned_loss=0.06239, over 4860.00 frames. ], tot_loss[loss=0.1949, simple_loss=0.2611, pruned_loss=0.06432, over 955500.73 frames. ], batch size: 31, lr: 3.67e-03, grad_scale: 16.0 2023-03-26 13:56:13,064 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8608, 1.6107, 1.4825, 1.1993, 1.6292, 1.6675, 1.5897, 2.1746], device='cuda:3'), covar=tensor([0.5042, 0.5236, 0.3856, 0.4804, 0.4679, 0.2791, 0.4527, 0.2253], device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0260, 0.0222, 0.0275, 0.0242, 0.0209, 0.0244, 0.0215], 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') 2023-03-26 13:56:22,194 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62901.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:56:37,254 INFO [finetune.py:976] (3/7) Epoch 11, batch 5650, loss[loss=0.213, simple_loss=0.2899, pruned_loss=0.06804, over 4907.00 frames. ], tot_loss[loss=0.1948, simple_loss=0.2617, pruned_loss=0.06394, over 954953.24 frames. ], batch size: 36, lr: 3.67e-03, grad_scale: 16.0 2023-03-26 13:56:38,488 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62929.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:56:48,736 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.326e+01 1.606e+02 1.910e+02 2.279e+02 4.497e+02, threshold=3.820e+02, percent-clipped=2.0 2023-03-26 13:56:51,137 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62941.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:57:18,748 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1704, 1.1556, 1.4501, 1.0381, 1.0543, 1.3661, 1.1738, 1.4157], device='cuda:3'), covar=tensor([0.1071, 0.1811, 0.0950, 0.1115, 0.0925, 0.0984, 0.2409, 0.0732], device='cuda:3'), in_proj_covar=tensor([0.0199, 0.0208, 0.0195, 0.0193, 0.0181, 0.0217, 0.0219, 0.0202], 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') 2023-03-26 13:57:23,398 INFO [finetune.py:976] (3/7) Epoch 11, batch 5700, loss[loss=0.2082, simple_loss=0.2579, pruned_loss=0.07921, over 4086.00 frames. ], tot_loss[loss=0.1938, simple_loss=0.259, pruned_loss=0.0643, over 935202.89 frames. ], batch size: 17, lr: 3.67e-03, grad_scale: 16.0 2023-03-26 13:57:23,432 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=62977.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:57:54,977 INFO [finetune.py:976] (3/7) Epoch 12, batch 0, loss[loss=0.1755, simple_loss=0.2386, pruned_loss=0.05617, over 4803.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2386, pruned_loss=0.05617, over 4803.00 frames. ], batch size: 25, lr: 3.67e-03, grad_scale: 16.0 2023-03-26 13:57:54,977 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 13:58:10,480 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0498, 1.7542, 1.6655, 1.6881, 1.8010, 1.7273, 1.7407, 2.4448], device='cuda:3'), covar=tensor([0.4383, 0.5572, 0.4016, 0.4565, 0.4667, 0.2790, 0.4592, 0.2000], device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0260, 0.0222, 0.0275, 0.0242, 0.0209, 0.0245, 0.0216], 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') 2023-03-26 13:58:11,587 INFO [finetune.py:1010] (3/7) Epoch 12, validation: loss=0.16, simple_loss=0.2305, pruned_loss=0.04472, over 2265189.00 frames. 2023-03-26 13:58:11,587 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 13:58:22,060 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63022.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:58:37,035 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.014e+02 1.590e+02 1.966e+02 2.351e+02 4.424e+02, threshold=3.931e+02, percent-clipped=2.0 2023-03-26 13:58:49,943 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63047.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:58:56,926 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-03-26 13:59:00,855 INFO [finetune.py:976] (3/7) Epoch 12, batch 50, loss[loss=0.2319, simple_loss=0.2905, pruned_loss=0.08665, over 4739.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2662, pruned_loss=0.0667, over 215952.43 frames. ], batch size: 26, lr: 3.67e-03, grad_scale: 16.0 2023-03-26 13:59:42,643 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=63095.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 13:59:54,682 INFO [finetune.py:976] (3/7) Epoch 12, batch 100, loss[loss=0.2029, simple_loss=0.2641, pruned_loss=0.07085, over 4894.00 frames. ], tot_loss[loss=0.1936, simple_loss=0.2582, pruned_loss=0.06449, over 377599.13 frames. ], batch size: 35, lr: 3.67e-03, grad_scale: 16.0 2023-03-26 14:00:15,389 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63129.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:00:20,608 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.4102, 3.7601, 3.9506, 4.2360, 4.1318, 3.8865, 4.5390, 1.4508], device='cuda:3'), covar=tensor([0.0766, 0.0877, 0.0809, 0.0996, 0.1201, 0.1509, 0.0613, 0.5397], device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0242, 0.0274, 0.0288, 0.0328, 0.0281, 0.0300, 0.0293], 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') 2023-03-26 14:00:21,134 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.120e+02 1.723e+02 1.978e+02 2.544e+02 5.107e+02, threshold=3.957e+02, percent-clipped=1.0 2023-03-26 14:00:43,436 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-26 14:00:50,144 INFO [finetune.py:976] (3/7) Epoch 12, batch 150, loss[loss=0.1563, simple_loss=0.2287, pruned_loss=0.04193, over 4780.00 frames. ], tot_loss[loss=0.1891, simple_loss=0.2531, pruned_loss=0.06249, over 507445.26 frames. ], batch size: 26, lr: 3.67e-03, grad_scale: 16.0 2023-03-26 14:01:47,576 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63201.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:01:56,082 INFO [finetune.py:976] (3/7) Epoch 12, batch 200, loss[loss=0.2291, simple_loss=0.2989, pruned_loss=0.0796, over 4752.00 frames. ], tot_loss[loss=0.1885, simple_loss=0.2527, pruned_loss=0.06218, over 608787.58 frames. ], batch size: 59, lr: 3.67e-03, grad_scale: 16.0 2023-03-26 14:02:17,463 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63221.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:02:32,829 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.005e+02 1.551e+02 1.870e+02 2.223e+02 3.918e+02, threshold=3.740e+02, percent-clipped=0.0 2023-03-26 14:02:41,478 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63241.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:02:46,789 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=63249.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:02:51,341 INFO [finetune.py:976] (3/7) Epoch 12, batch 250, loss[loss=0.2312, simple_loss=0.3153, pruned_loss=0.07357, over 4821.00 frames. ], tot_loss[loss=0.1934, simple_loss=0.2582, pruned_loss=0.06434, over 687346.00 frames. ], batch size: 40, lr: 3.67e-03, grad_scale: 16.0 2023-03-26 14:03:08,657 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63282.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:03:13,375 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=63289.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:03:14,401 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-26 14:03:17,593 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7182, 1.5220, 1.8548, 1.9543, 1.6738, 3.5495, 1.3865, 1.6382], device='cuda:3'), covar=tensor([0.0929, 0.1854, 0.1111, 0.0962, 0.1698, 0.0243, 0.1586, 0.1778], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0082, 0.0075, 0.0078, 0.0092, 0.0082, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 14:03:23,523 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6389, 0.6706, 1.6704, 1.5176, 1.4087, 1.3310, 1.4582, 1.5526], device='cuda:3'), covar=tensor([0.3388, 0.3803, 0.3068, 0.3087, 0.4109, 0.3247, 0.4048, 0.2952], device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0239, 0.0256, 0.0261, 0.0258, 0.0232, 0.0276, 0.0234], 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') 2023-03-26 14:03:23,972 INFO [finetune.py:976] (3/7) Epoch 12, batch 300, loss[loss=0.2455, simple_loss=0.2997, pruned_loss=0.09568, over 4834.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2618, pruned_loss=0.06531, over 746800.33 frames. ], batch size: 49, lr: 3.67e-03, grad_scale: 16.0 2023-03-26 14:03:40,225 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63322.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:03:49,456 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3905, 1.4089, 1.4754, 1.6115, 1.5292, 2.9622, 1.2976, 1.5316], device='cuda:3'), covar=tensor([0.0945, 0.1781, 0.1086, 0.0886, 0.1466, 0.0285, 0.1456, 0.1699], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0082, 0.0075, 0.0078, 0.0092, 0.0082, 0.0085, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 14:03:51,185 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.087e+02 1.663e+02 2.076e+02 2.406e+02 5.777e+02, threshold=4.151e+02, percent-clipped=4.0 2023-03-26 14:04:08,575 INFO [finetune.py:976] (3/7) Epoch 12, batch 350, loss[loss=0.2083, simple_loss=0.2757, pruned_loss=0.07045, over 4831.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2629, pruned_loss=0.06545, over 794023.94 frames. ], batch size: 47, lr: 3.67e-03, grad_scale: 16.0 2023-03-26 14:04:27,599 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=63370.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:04:57,885 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6105, 1.4807, 1.4103, 1.5334, 1.8830, 1.7986, 1.6044, 1.3666], device='cuda:3'), covar=tensor([0.0293, 0.0310, 0.0510, 0.0311, 0.0222, 0.0393, 0.0313, 0.0361], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0108, 0.0140, 0.0114, 0.0102, 0.0104, 0.0093, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.1874e-05, 8.4196e-05, 1.1106e-04, 8.8737e-05, 7.9320e-05, 7.6902e-05, 7.0533e-05, 8.3432e-05], device='cuda:3') 2023-03-26 14:04:59,619 INFO [finetune.py:976] (3/7) Epoch 12, batch 400, loss[loss=0.1606, simple_loss=0.2357, pruned_loss=0.04279, over 4923.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2638, pruned_loss=0.06517, over 830070.94 frames. ], batch size: 33, lr: 3.66e-03, grad_scale: 16.0 2023-03-26 14:05:10,721 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63420.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:05:11,934 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63422.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:05:16,647 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63429.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:05:21,322 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.052e+02 1.591e+02 1.854e+02 2.332e+02 4.296e+02, threshold=3.709e+02, percent-clipped=1.0 2023-03-26 14:05:38,141 INFO [finetune.py:976] (3/7) Epoch 12, batch 450, loss[loss=0.1839, simple_loss=0.2414, pruned_loss=0.0632, over 4193.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2631, pruned_loss=0.0655, over 856547.47 frames. ], batch size: 18, lr: 3.66e-03, grad_scale: 16.0 2023-03-26 14:05:56,658 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0625, 1.5548, 0.6023, 1.9655, 2.4929, 1.7369, 1.6482, 2.0670], device='cuda:3'), covar=tensor([0.1513, 0.2047, 0.2485, 0.1272, 0.1756, 0.2003, 0.1487, 0.1883], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0096, 0.0114, 0.0093, 0.0121, 0.0096, 0.0100, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 14:05:57,265 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=63477.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:05:59,757 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63481.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:06:00,977 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63483.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:06:10,051 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-26 14:06:11,748 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2573, 1.7331, 1.2601, 2.0194, 2.5748, 1.8512, 1.9204, 2.1436], device='cuda:3'), covar=tensor([0.1238, 0.1835, 0.1750, 0.1134, 0.1532, 0.1667, 0.1252, 0.1669], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0096, 0.0114, 0.0093, 0.0120, 0.0095, 0.0100, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 14:06:15,170 INFO [finetune.py:976] (3/7) Epoch 12, batch 500, loss[loss=0.1218, simple_loss=0.1924, pruned_loss=0.02558, over 4763.00 frames. ], tot_loss[loss=0.194, simple_loss=0.2599, pruned_loss=0.06406, over 877875.10 frames. ], batch size: 27, lr: 3.66e-03, grad_scale: 16.0 2023-03-26 14:06:25,457 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4891, 1.3582, 1.3947, 1.3293, 0.6916, 2.2302, 0.7496, 1.1369], device='cuda:3'), covar=tensor([0.3458, 0.2558, 0.2245, 0.2513, 0.2261, 0.0382, 0.2781, 0.1480], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0115, 0.0119, 0.0123, 0.0115, 0.0098, 0.0098, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 14:06:27,331 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-26 14:06:37,049 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.336e+01 1.553e+02 1.855e+02 2.331e+02 4.193e+02, threshold=3.711e+02, percent-clipped=1.0 2023-03-26 14:06:37,196 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6757, 1.6051, 1.4200, 1.8040, 2.1205, 1.8526, 1.4266, 1.3579], device='cuda:3'), covar=tensor([0.2311, 0.2112, 0.2076, 0.1620, 0.1829, 0.1215, 0.2637, 0.2029], device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0207, 0.0209, 0.0189, 0.0242, 0.0182, 0.0212, 0.0197], 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') 2023-03-26 14:06:47,233 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1043, 3.5520, 3.7344, 3.8759, 3.8500, 3.6463, 4.1748, 1.3641], device='cuda:3'), covar=tensor([0.0806, 0.0962, 0.0808, 0.1073, 0.1272, 0.1401, 0.0745, 0.5498], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0245, 0.0277, 0.0291, 0.0331, 0.0284, 0.0302, 0.0295], 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') 2023-03-26 14:06:48,874 INFO [finetune.py:976] (3/7) Epoch 12, batch 550, loss[loss=0.1654, simple_loss=0.2319, pruned_loss=0.04944, over 4894.00 frames. ], tot_loss[loss=0.1928, simple_loss=0.2583, pruned_loss=0.06361, over 898036.77 frames. ], batch size: 32, lr: 3.66e-03, grad_scale: 16.0 2023-03-26 14:06:58,440 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63569.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:07:03,814 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63577.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:07:10,291 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63586.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:07:20,610 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4282, 1.3513, 1.4324, 0.8085, 1.5053, 1.4347, 1.4159, 1.2528], device='cuda:3'), covar=tensor([0.0605, 0.0782, 0.0711, 0.0989, 0.0819, 0.0755, 0.0651, 0.1229], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0133, 0.0140, 0.0124, 0.0120, 0.0142, 0.0142, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 14:07:22,328 INFO [finetune.py:976] (3/7) Epoch 12, batch 600, loss[loss=0.1529, simple_loss=0.2204, pruned_loss=0.04266, over 4706.00 frames. ], tot_loss[loss=0.1935, simple_loss=0.2591, pruned_loss=0.06399, over 910063.53 frames. ], batch size: 23, lr: 3.66e-03, grad_scale: 16.0 2023-03-26 14:07:40,190 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63630.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:07:44,849 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.064e+02 1.685e+02 2.017e+02 2.531e+02 3.696e+02, threshold=4.034e+02, percent-clipped=0.0 2023-03-26 14:07:51,108 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63647.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:07:56,392 INFO [finetune.py:976] (3/7) Epoch 12, batch 650, loss[loss=0.1721, simple_loss=0.2405, pruned_loss=0.05184, over 4796.00 frames. ], tot_loss[loss=0.1953, simple_loss=0.2612, pruned_loss=0.06472, over 917446.20 frames. ], batch size: 25, lr: 3.66e-03, grad_scale: 16.0 2023-03-26 14:08:29,419 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8139, 1.7237, 1.5972, 1.9491, 2.3189, 1.9712, 1.5201, 1.5374], device='cuda:3'), covar=tensor([0.2317, 0.2085, 0.1948, 0.1732, 0.1708, 0.1204, 0.2439, 0.1934], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0209, 0.0211, 0.0191, 0.0244, 0.0183, 0.0214, 0.0198], 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') 2023-03-26 14:08:29,866 INFO [finetune.py:976] (3/7) Epoch 12, batch 700, loss[loss=0.1774, simple_loss=0.2456, pruned_loss=0.05458, over 4751.00 frames. ], tot_loss[loss=0.1955, simple_loss=0.2622, pruned_loss=0.06445, over 926156.32 frames. ], batch size: 28, lr: 3.66e-03, grad_scale: 16.0 2023-03-26 14:08:38,503 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-26 14:08:57,062 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8944, 4.3328, 4.1656, 2.3372, 4.5138, 3.3927, 0.6292, 3.0048], device='cuda:3'), covar=tensor([0.2325, 0.1380, 0.1264, 0.2819, 0.0651, 0.0783, 0.4591, 0.1287], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0172, 0.0158, 0.0127, 0.0155, 0.0121, 0.0145, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 14:08:59,836 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.062e+02 1.754e+02 2.049e+02 2.499e+02 4.974e+02, threshold=4.098e+02, percent-clipped=3.0 2023-03-26 14:09:04,727 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.11 vs. limit=2.0 2023-03-26 14:09:11,207 INFO [finetune.py:976] (3/7) Epoch 12, batch 750, loss[loss=0.1962, simple_loss=0.2846, pruned_loss=0.05386, over 4782.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2639, pruned_loss=0.06495, over 933595.35 frames. ], batch size: 29, lr: 3.66e-03, grad_scale: 16.0 2023-03-26 14:09:15,555 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0 2023-03-26 14:09:25,545 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63776.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:09:26,758 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63778.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:09:56,448 INFO [finetune.py:976] (3/7) Epoch 12, batch 800, loss[loss=0.1933, simple_loss=0.2473, pruned_loss=0.06962, over 4892.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.2629, pruned_loss=0.06421, over 939405.60 frames. ], batch size: 32, lr: 3.66e-03, grad_scale: 16.0 2023-03-26 14:10:04,886 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63810.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:10:26,004 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.160e+02 1.587e+02 1.868e+02 2.134e+02 3.136e+02, threshold=3.736e+02, percent-clipped=1.0 2023-03-26 14:10:32,485 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63845.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:10:38,486 INFO [finetune.py:976] (3/7) Epoch 12, batch 850, loss[loss=0.1643, simple_loss=0.2226, pruned_loss=0.05299, over 4742.00 frames. ], tot_loss[loss=0.1946, simple_loss=0.2613, pruned_loss=0.06399, over 944083.02 frames. ], batch size: 23, lr: 3.66e-03, grad_scale: 16.0 2023-03-26 14:10:51,306 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63871.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:10:59,652 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63877.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:11:22,733 INFO [finetune.py:976] (3/7) Epoch 12, batch 900, loss[loss=0.1319, simple_loss=0.2019, pruned_loss=0.03098, over 4903.00 frames. ], tot_loss[loss=0.1918, simple_loss=0.258, pruned_loss=0.06279, over 945350.47 frames. ], batch size: 32, lr: 3.66e-03, grad_scale: 16.0 2023-03-26 14:11:23,452 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63906.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:11:24,605 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8517, 3.8805, 3.7567, 1.9113, 4.0033, 2.9718, 0.7948, 2.7633], device='cuda:3'), covar=tensor([0.2487, 0.1709, 0.1508, 0.3282, 0.0949, 0.0976, 0.4533, 0.1378], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0173, 0.0159, 0.0128, 0.0155, 0.0122, 0.0145, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 14:11:29,468 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63916.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:11:35,894 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=63925.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:11:35,903 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63925.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:11:44,046 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.021e+02 1.611e+02 1.873e+02 2.372e+02 4.297e+02, threshold=3.747e+02, percent-clipped=2.0 2023-03-26 14:11:47,183 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63942.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:11:56,453 INFO [finetune.py:976] (3/7) Epoch 12, batch 950, loss[loss=0.1602, simple_loss=0.2323, pruned_loss=0.04405, over 4904.00 frames. ], tot_loss[loss=0.1895, simple_loss=0.2548, pruned_loss=0.06212, over 946634.25 frames. ], batch size: 37, lr: 3.66e-03, grad_scale: 16.0 2023-03-26 14:12:10,884 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63977.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:12:15,646 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8958, 1.6547, 1.9049, 1.1039, 1.8398, 1.8901, 1.9060, 1.5137], device='cuda:3'), covar=tensor([0.0505, 0.0746, 0.0586, 0.0905, 0.0728, 0.0604, 0.0541, 0.1106], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0134, 0.0140, 0.0124, 0.0121, 0.0142, 0.0143, 0.0161], 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') 2023-03-26 14:12:19,215 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6705, 2.4915, 2.1570, 2.5316, 2.5390, 2.3690, 2.8279, 2.6647], device='cuda:3'), covar=tensor([0.1369, 0.2037, 0.2879, 0.2404, 0.2422, 0.1571, 0.2928, 0.1732], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0188, 0.0233, 0.0255, 0.0241, 0.0198, 0.0212, 0.0197], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 14:12:31,104 INFO [finetune.py:976] (3/7) Epoch 12, batch 1000, loss[loss=0.2028, simple_loss=0.2701, pruned_loss=0.06779, over 4817.00 frames. ], tot_loss[loss=0.1917, simple_loss=0.257, pruned_loss=0.06315, over 945494.32 frames. ], batch size: 38, lr: 3.66e-03, grad_scale: 16.0 2023-03-26 14:12:43,079 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64023.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:12:51,954 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.200e+02 1.649e+02 1.875e+02 2.259e+02 3.443e+02, threshold=3.751e+02, percent-clipped=0.0 2023-03-26 14:12:58,503 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9268, 4.2350, 3.9189, 2.3540, 4.3987, 3.2529, 0.8068, 3.0812], device='cuda:3'), covar=tensor([0.2299, 0.1942, 0.1522, 0.2853, 0.0793, 0.0946, 0.4632, 0.1252], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0173, 0.0159, 0.0128, 0.0155, 0.0122, 0.0146, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 14:13:02,010 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0 2023-03-26 14:13:04,256 INFO [finetune.py:976] (3/7) Epoch 12, batch 1050, loss[loss=0.1675, simple_loss=0.2368, pruned_loss=0.04904, over 4789.00 frames. ], tot_loss[loss=0.1923, simple_loss=0.2589, pruned_loss=0.06285, over 948341.56 frames. ], batch size: 23, lr: 3.66e-03, grad_scale: 16.0 2023-03-26 14:13:17,538 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64076.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:13:19,206 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64078.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:13:22,948 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64084.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:13:37,900 INFO [finetune.py:976] (3/7) Epoch 12, batch 1100, loss[loss=0.2062, simple_loss=0.2719, pruned_loss=0.07021, over 4798.00 frames. ], tot_loss[loss=0.193, simple_loss=0.2596, pruned_loss=0.06318, over 949698.82 frames. ], batch size: 45, lr: 3.66e-03, grad_scale: 16.0 2023-03-26 14:13:53,905 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=64124.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:13:55,107 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=64126.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:14:05,898 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.091e+02 1.584e+02 1.925e+02 2.329e+02 4.054e+02, threshold=3.850e+02, percent-clipped=2.0 2023-03-26 14:14:17,903 INFO [finetune.py:976] (3/7) Epoch 12, batch 1150, loss[loss=0.1754, simple_loss=0.2442, pruned_loss=0.05333, over 4786.00 frames. ], tot_loss[loss=0.1934, simple_loss=0.2603, pruned_loss=0.06323, over 950867.97 frames. ], batch size: 26, lr: 3.66e-03, grad_scale: 16.0 2023-03-26 14:14:25,686 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64166.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:14:48,855 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64201.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:14:56,321 INFO [finetune.py:976] (3/7) Epoch 12, batch 1200, loss[loss=0.1867, simple_loss=0.2591, pruned_loss=0.05713, over 4864.00 frames. ], tot_loss[loss=0.1922, simple_loss=0.2588, pruned_loss=0.06279, over 950002.95 frames. ], batch size: 34, lr: 3.66e-03, grad_scale: 16.0 2023-03-26 14:15:12,540 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9197, 1.4165, 1.9508, 1.8285, 1.6581, 1.6462, 1.8016, 1.7891], device='cuda:3'), covar=tensor([0.3630, 0.4226, 0.3444, 0.3776, 0.4904, 0.3474, 0.4359, 0.3179], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0238, 0.0255, 0.0260, 0.0256, 0.0231, 0.0274, 0.0233], 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') 2023-03-26 14:15:13,778 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7856, 0.7865, 1.7836, 1.6357, 1.5938, 1.5273, 1.5783, 1.6809], device='cuda:3'), covar=tensor([0.3378, 0.3988, 0.3320, 0.3546, 0.4489, 0.3534, 0.4114, 0.3357], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0238, 0.0255, 0.0260, 0.0256, 0.0231, 0.0274, 0.0232], 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') 2023-03-26 14:15:14,936 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64225.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:15:24,856 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64232.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:15:31,343 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.004e+02 1.575e+02 1.833e+02 2.193e+02 5.344e+02, threshold=3.667e+02, percent-clipped=2.0 2023-03-26 14:15:34,437 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64242.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:15:43,193 INFO [finetune.py:976] (3/7) Epoch 12, batch 1250, loss[loss=0.1868, simple_loss=0.2478, pruned_loss=0.06292, over 4910.00 frames. ], tot_loss[loss=0.1912, simple_loss=0.2575, pruned_loss=0.06246, over 952267.46 frames. ], batch size: 32, lr: 3.66e-03, grad_scale: 16.0 2023-03-26 14:15:55,099 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64272.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:15:55,714 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=64273.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:15:56,405 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1267, 1.9762, 1.6792, 1.9007, 2.1206, 1.7715, 2.3797, 2.1082], device='cuda:3'), covar=tensor([0.1372, 0.2470, 0.3239, 0.2902, 0.2648, 0.1808, 0.3064, 0.2022], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0189, 0.0234, 0.0256, 0.0242, 0.0199, 0.0213, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 14:15:57,638 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7746, 1.6181, 1.5049, 1.8749, 2.0226, 1.8461, 1.2227, 1.5047], device='cuda:3'), covar=tensor([0.2176, 0.2089, 0.1925, 0.1599, 0.1627, 0.1143, 0.2554, 0.1826], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0208, 0.0209, 0.0189, 0.0242, 0.0183, 0.0213, 0.0197], 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') 2023-03-26 14:16:09,181 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=64290.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:16:11,082 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64293.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:16:27,225 INFO [finetune.py:976] (3/7) Epoch 12, batch 1300, loss[loss=0.1787, simple_loss=0.2487, pruned_loss=0.05436, over 4921.00 frames. ], tot_loss[loss=0.1882, simple_loss=0.2544, pruned_loss=0.06102, over 954242.96 frames. ], batch size: 36, lr: 3.66e-03, grad_scale: 16.0 2023-03-26 14:16:39,619 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.9685, 0.8061, 0.8548, 1.0443, 1.1528, 1.0859, 0.9857, 0.9376], device='cuda:3'), covar=tensor([0.0376, 0.0353, 0.0652, 0.0329, 0.0304, 0.0494, 0.0376, 0.0402], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0108, 0.0139, 0.0113, 0.0101, 0.0103, 0.0093, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.1678e-05, 8.4111e-05, 1.1070e-04, 8.8114e-05, 7.9117e-05, 7.6637e-05, 7.0451e-05, 8.3118e-05], device='cuda:3') 2023-03-26 14:16:48,480 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.098e+02 1.610e+02 1.842e+02 2.244e+02 4.381e+02, threshold=3.684e+02, percent-clipped=1.0 2023-03-26 14:16:51,029 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9242, 1.5333, 2.2630, 1.6051, 2.0633, 2.1222, 1.5670, 2.2678], device='cuda:3'), covar=tensor([0.1230, 0.1954, 0.1287, 0.1768, 0.0728, 0.1346, 0.2456, 0.0723], device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0206, 0.0194, 0.0191, 0.0179, 0.0214, 0.0218, 0.0200], 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') 2023-03-26 14:16:53,410 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64345.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:16:59,917 INFO [finetune.py:976] (3/7) Epoch 12, batch 1350, loss[loss=0.2455, simple_loss=0.3112, pruned_loss=0.08988, over 4805.00 frames. ], tot_loss[loss=0.188, simple_loss=0.254, pruned_loss=0.06096, over 951312.78 frames. ], batch size: 51, lr: 3.66e-03, grad_scale: 32.0 2023-03-26 14:17:16,045 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64379.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:17:19,490 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-26 14:17:32,694 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-26 14:17:33,430 INFO [finetune.py:976] (3/7) Epoch 12, batch 1400, loss[loss=0.1944, simple_loss=0.268, pruned_loss=0.06045, over 4848.00 frames. ], tot_loss[loss=0.1913, simple_loss=0.2579, pruned_loss=0.06237, over 952259.80 frames. ], batch size: 49, lr: 3.66e-03, grad_scale: 32.0 2023-03-26 14:17:34,184 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64406.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:17:54,259 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.198e+02 1.617e+02 1.936e+02 2.295e+02 3.610e+02, threshold=3.872e+02, percent-clipped=0.0 2023-03-26 14:18:03,814 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2283, 1.4034, 1.5779, 1.1204, 1.1640, 1.4973, 1.3948, 1.6079], device='cuda:3'), covar=tensor([0.1236, 0.1876, 0.1236, 0.1467, 0.0996, 0.1141, 0.2598, 0.0913], device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0206, 0.0195, 0.0192, 0.0179, 0.0215, 0.0218, 0.0200], 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') 2023-03-26 14:18:06,659 INFO [finetune.py:976] (3/7) Epoch 12, batch 1450, loss[loss=0.1528, simple_loss=0.2165, pruned_loss=0.04449, over 4814.00 frames. ], tot_loss[loss=0.1943, simple_loss=0.261, pruned_loss=0.06382, over 951983.93 frames. ], batch size: 25, lr: 3.66e-03, grad_scale: 32.0 2023-03-26 14:18:13,330 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64465.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:18:13,902 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64466.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:18:37,424 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64501.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:18:39,730 INFO [finetune.py:976] (3/7) Epoch 12, batch 1500, loss[loss=0.1809, simple_loss=0.2523, pruned_loss=0.05469, over 4858.00 frames. ], tot_loss[loss=0.1936, simple_loss=0.2608, pruned_loss=0.06319, over 951742.63 frames. ], batch size: 44, lr: 3.66e-03, grad_scale: 32.0 2023-03-26 14:18:46,092 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=64514.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:18:54,917 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64526.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:19:00,154 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.58 vs. limit=5.0 2023-03-26 14:19:01,494 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.120e+02 1.738e+02 2.083e+02 2.672e+02 4.064e+02, threshold=4.165e+02, percent-clipped=1.0 2023-03-26 14:19:15,860 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=64549.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:19:19,447 INFO [finetune.py:976] (3/7) Epoch 12, batch 1550, loss[loss=0.1691, simple_loss=0.2354, pruned_loss=0.05141, over 4822.00 frames. ], tot_loss[loss=0.1953, simple_loss=0.2622, pruned_loss=0.06425, over 953459.39 frames. ], batch size: 25, lr: 3.66e-03, grad_scale: 32.0 2023-03-26 14:19:33,900 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64572.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:19:45,161 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64588.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:19:45,816 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64589.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:19:56,452 INFO [finetune.py:976] (3/7) Epoch 12, batch 1600, loss[loss=0.1838, simple_loss=0.2507, pruned_loss=0.05845, over 4895.00 frames. ], tot_loss[loss=0.193, simple_loss=0.2597, pruned_loss=0.06311, over 954548.39 frames. ], batch size: 32, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:20:08,114 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=64620.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:20:22,633 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-26 14:20:30,244 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.026e+02 1.633e+02 1.922e+02 2.431e+02 4.177e+02, threshold=3.845e+02, percent-clipped=1.0 2023-03-26 14:20:31,577 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7466, 1.2477, 0.8209, 1.5783, 2.1416, 1.0974, 1.4941, 1.6683], device='cuda:3'), covar=tensor([0.1611, 0.2219, 0.2223, 0.1301, 0.1957, 0.1973, 0.1533, 0.2025], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0095, 0.0113, 0.0092, 0.0120, 0.0094, 0.0099, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 14:20:41,557 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64648.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:20:43,266 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64650.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:20:49,612 INFO [finetune.py:976] (3/7) Epoch 12, batch 1650, loss[loss=0.1476, simple_loss=0.2199, pruned_loss=0.03758, over 4940.00 frames. ], tot_loss[loss=0.1896, simple_loss=0.2562, pruned_loss=0.06151, over 956323.26 frames. ], batch size: 38, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:21:05,207 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64679.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:21:15,752 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-03-26 14:21:19,528 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64701.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:21:19,562 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4006, 1.3898, 1.3883, 1.6567, 1.5422, 2.8695, 1.3503, 1.5200], device='cuda:3'), covar=tensor([0.0890, 0.1812, 0.1249, 0.0922, 0.1573, 0.0293, 0.1446, 0.1647], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0075, 0.0078, 0.0092, 0.0081, 0.0085, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 14:21:21,783 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64703.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:21:22,889 INFO [finetune.py:976] (3/7) Epoch 12, batch 1700, loss[loss=0.1197, simple_loss=0.1942, pruned_loss=0.02255, over 4784.00 frames. ], tot_loss[loss=0.1881, simple_loss=0.2543, pruned_loss=0.06089, over 958293.38 frames. ], batch size: 28, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:21:27,402 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64709.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:21:38,428 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9787, 1.0352, 1.8253, 1.7614, 1.6281, 1.5846, 1.6869, 1.7397], device='cuda:3'), covar=tensor([0.3526, 0.4146, 0.3665, 0.3870, 0.4954, 0.3841, 0.4602, 0.3533], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0238, 0.0256, 0.0259, 0.0257, 0.0231, 0.0274, 0.0232], 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') 2023-03-26 14:21:46,817 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=64727.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:21:50,594 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7806, 1.7375, 1.6558, 2.1083, 2.0860, 2.0100, 1.4672, 1.4956], device='cuda:3'), covar=tensor([0.2081, 0.1923, 0.1726, 0.1328, 0.1905, 0.1101, 0.2558, 0.1830], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0208, 0.0210, 0.0189, 0.0242, 0.0183, 0.0213, 0.0198], 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') 2023-03-26 14:21:53,447 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.135e+02 1.591e+02 1.930e+02 2.225e+02 5.420e+02, threshold=3.861e+02, percent-clipped=2.0 2023-03-26 14:21:58,955 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64745.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:22:03,095 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64752.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:22:05,307 INFO [finetune.py:976] (3/7) Epoch 12, batch 1750, loss[loss=0.2233, simple_loss=0.2864, pruned_loss=0.08011, over 4906.00 frames. ], tot_loss[loss=0.1905, simple_loss=0.2567, pruned_loss=0.0622, over 956534.03 frames. ], batch size: 43, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:22:10,846 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64764.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:22:32,639 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8497, 1.7305, 1.4911, 1.3838, 1.8748, 1.5626, 1.8252, 1.7675], device='cuda:3'), covar=tensor([0.1520, 0.2155, 0.3230, 0.2742, 0.2581, 0.1848, 0.2791, 0.1949], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0188, 0.0233, 0.0255, 0.0241, 0.0198, 0.0213, 0.0197], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 14:22:38,079 INFO [finetune.py:976] (3/7) Epoch 12, batch 1800, loss[loss=0.2148, simple_loss=0.2846, pruned_loss=0.07252, over 4863.00 frames. ], tot_loss[loss=0.1933, simple_loss=0.2602, pruned_loss=0.06323, over 957001.61 frames. ], batch size: 44, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:22:38,796 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64806.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:22:43,537 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64813.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:22:48,349 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64821.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:22:56,656 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6928, 1.5616, 1.5590, 1.6504, 1.1203, 3.6070, 1.4818, 1.9219], device='cuda:3'), covar=tensor([0.3442, 0.2522, 0.2107, 0.2454, 0.1914, 0.0173, 0.2447, 0.1279], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0116, 0.0120, 0.0123, 0.0116, 0.0098, 0.0098, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 14:22:58,985 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.145e+02 1.622e+02 1.968e+02 2.271e+02 4.247e+02, threshold=3.936e+02, percent-clipped=1.0 2023-03-26 14:23:11,390 INFO [finetune.py:976] (3/7) Epoch 12, batch 1850, loss[loss=0.2138, simple_loss=0.2749, pruned_loss=0.07634, over 4878.00 frames. ], tot_loss[loss=0.1938, simple_loss=0.2606, pruned_loss=0.06351, over 957292.56 frames. ], batch size: 32, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:23:25,671 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8292, 1.5942, 2.1241, 1.4337, 1.9166, 2.1158, 1.5148, 2.2179], device='cuda:3'), covar=tensor([0.1427, 0.2470, 0.1413, 0.2028, 0.0943, 0.1355, 0.3038, 0.0913], device='cuda:3'), in_proj_covar=tensor([0.0198, 0.0207, 0.0196, 0.0193, 0.0180, 0.0217, 0.0220, 0.0202], 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') 2023-03-26 14:23:33,513 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64888.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:23:45,141 INFO [finetune.py:976] (3/7) Epoch 12, batch 1900, loss[loss=0.2195, simple_loss=0.2864, pruned_loss=0.07632, over 4902.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2636, pruned_loss=0.06493, over 956320.07 frames. ], batch size: 37, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:24:06,072 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=64936.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:24:06,596 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.024e+02 1.653e+02 1.911e+02 2.364e+02 4.358e+02, threshold=3.822e+02, percent-clipped=3.0 2023-03-26 14:24:08,005 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0978, 2.0007, 1.7958, 2.1965, 1.8674, 1.9092, 1.8885, 2.7546], device='cuda:3'), covar=tensor([0.4366, 0.5445, 0.3576, 0.4732, 0.4863, 0.2715, 0.5172, 0.1839], device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0258, 0.0223, 0.0274, 0.0242, 0.0209, 0.0245, 0.0216], 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') 2023-03-26 14:24:12,012 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64945.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:24:18,921 INFO [finetune.py:976] (3/7) Epoch 12, batch 1950, loss[loss=0.1649, simple_loss=0.2345, pruned_loss=0.04766, over 4791.00 frames. ], tot_loss[loss=0.1935, simple_loss=0.2603, pruned_loss=0.06334, over 952797.79 frames. ], batch size: 51, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:24:40,798 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64976.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:24:55,680 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.0966, 1.0118, 1.0534, 0.6141, 0.8423, 1.2000, 1.2040, 1.0648], device='cuda:3'), covar=tensor([0.0791, 0.0586, 0.0518, 0.0460, 0.0594, 0.0507, 0.0367, 0.0609], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0151, 0.0120, 0.0130, 0.0128, 0.0124, 0.0141, 0.0144], device='cuda:3'), out_proj_covar=tensor([9.2134e-05, 1.1059e-04, 8.6353e-05, 9.3802e-05, 9.1271e-05, 9.0301e-05, 1.0320e-04, 1.0505e-04], device='cuda:3') 2023-03-26 14:24:58,110 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65001.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:24:58,390 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.78 vs. limit=5.0 2023-03-26 14:24:59,918 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65004.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:25:00,451 INFO [finetune.py:976] (3/7) Epoch 12, batch 2000, loss[loss=0.2027, simple_loss=0.2693, pruned_loss=0.06799, over 4832.00 frames. ], tot_loss[loss=0.191, simple_loss=0.2572, pruned_loss=0.06234, over 952016.52 frames. ], batch size: 33, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:25:03,513 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7946, 1.5089, 1.3714, 1.3435, 1.8854, 2.0093, 1.7867, 1.4478], device='cuda:3'), covar=tensor([0.0265, 0.0356, 0.0677, 0.0396, 0.0239, 0.0392, 0.0284, 0.0401], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0108, 0.0139, 0.0113, 0.0101, 0.0103, 0.0093, 0.0107], device='cuda:3'), out_proj_covar=tensor([7.1138e-05, 8.3694e-05, 1.1020e-04, 8.8169e-05, 7.9240e-05, 7.6387e-05, 7.0332e-05, 8.2609e-05], device='cuda:3') 2023-03-26 14:25:03,604 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.00 vs. limit=5.0 2023-03-26 14:25:04,715 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2020, 2.0679, 2.0436, 0.9789, 2.2446, 2.5178, 2.2031, 1.9393], device='cuda:3'), covar=tensor([0.1154, 0.0794, 0.0687, 0.0830, 0.0633, 0.0844, 0.0528, 0.0808], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0151, 0.0120, 0.0130, 0.0129, 0.0125, 0.0142, 0.0145], device='cuda:3'), out_proj_covar=tensor([9.2359e-05, 1.1076e-04, 8.6444e-05, 9.4028e-05, 9.1416e-05, 9.0521e-05, 1.0338e-04, 1.0528e-04], device='cuda:3') 2023-03-26 14:25:21,641 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.067e+02 1.609e+02 1.880e+02 2.233e+02 7.388e+02, threshold=3.760e+02, percent-clipped=1.0 2023-03-26 14:25:21,781 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65037.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 14:25:34,386 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=65049.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:25:42,374 INFO [finetune.py:976] (3/7) Epoch 12, batch 2050, loss[loss=0.1628, simple_loss=0.2254, pruned_loss=0.05007, over 4837.00 frames. ], tot_loss[loss=0.1876, simple_loss=0.2536, pruned_loss=0.06075, over 953631.49 frames. ], batch size: 49, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:25:42,494 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65055.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:25:45,414 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65059.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:25:53,067 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.82 vs. limit=5.0 2023-03-26 14:25:55,477 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-03-26 14:25:56,032 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5971, 1.5555, 2.0029, 1.2496, 1.6122, 1.8969, 1.5002, 2.0576], device='cuda:3'), covar=tensor([0.1244, 0.2221, 0.1162, 0.1853, 0.0893, 0.1395, 0.2836, 0.0759], device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0207, 0.0195, 0.0192, 0.0179, 0.0216, 0.0218, 0.0201], 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') 2023-03-26 14:26:21,862 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65101.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:26:24,691 INFO [finetune.py:976] (3/7) Epoch 12, batch 2100, loss[loss=0.1873, simple_loss=0.2564, pruned_loss=0.05905, over 4886.00 frames. ], tot_loss[loss=0.1886, simple_loss=0.2542, pruned_loss=0.06152, over 955583.13 frames. ], batch size: 32, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:26:27,107 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65108.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:26:32,491 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65116.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:26:35,497 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65121.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:26:52,600 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.054e+02 1.674e+02 1.985e+02 2.398e+02 5.597e+02, threshold=3.971e+02, percent-clipped=1.0 2023-03-26 14:27:08,171 INFO [finetune.py:976] (3/7) Epoch 12, batch 2150, loss[loss=0.1854, simple_loss=0.2635, pruned_loss=0.05363, over 4820.00 frames. ], tot_loss[loss=0.1926, simple_loss=0.2584, pruned_loss=0.06344, over 955426.32 frames. ], batch size: 51, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:27:17,728 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=65169.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:27:41,483 INFO [finetune.py:976] (3/7) Epoch 12, batch 2200, loss[loss=0.1535, simple_loss=0.2353, pruned_loss=0.03583, over 4861.00 frames. ], tot_loss[loss=0.1952, simple_loss=0.2612, pruned_loss=0.06462, over 954565.49 frames. ], batch size: 34, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:28:03,262 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.110e+02 1.670e+02 2.055e+02 2.491e+02 4.530e+02, threshold=4.111e+02, percent-clipped=2.0 2023-03-26 14:28:08,802 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65245.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:28:13,549 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1619, 2.1001, 2.2051, 0.9474, 2.3754, 2.6485, 2.2018, 1.9461], device='cuda:3'), covar=tensor([0.0846, 0.0705, 0.0511, 0.0743, 0.0607, 0.0575, 0.0563, 0.0720], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0153, 0.0122, 0.0132, 0.0130, 0.0126, 0.0143, 0.0146], device='cuda:3'), out_proj_covar=tensor([9.3512e-05, 1.1211e-04, 8.7694e-05, 9.5099e-05, 9.2718e-05, 9.1791e-05, 1.0462e-04, 1.0671e-04], device='cuda:3') 2023-03-26 14:28:15,257 INFO [finetune.py:976] (3/7) Epoch 12, batch 2250, loss[loss=0.2238, simple_loss=0.3006, pruned_loss=0.07352, over 4911.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2633, pruned_loss=0.06494, over 955217.56 frames. ], batch size: 37, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:28:35,399 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.1321, 0.9195, 0.9727, 0.4867, 0.7755, 1.0467, 1.0615, 0.9223], device='cuda:3'), covar=tensor([0.0703, 0.0474, 0.0456, 0.0469, 0.0524, 0.0525, 0.0306, 0.0463], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0154, 0.0122, 0.0132, 0.0131, 0.0126, 0.0143, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.3543e-05, 1.1239e-04, 8.7850e-05, 9.5153e-05, 9.2911e-05, 9.1788e-05, 1.0477e-04, 1.0682e-04], device='cuda:3') 2023-03-26 14:28:36,004 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65285.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:28:41,278 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=65293.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:28:48,515 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65304.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:28:49,046 INFO [finetune.py:976] (3/7) Epoch 12, batch 2300, loss[loss=0.2273, simple_loss=0.2868, pruned_loss=0.08385, over 4887.00 frames. ], tot_loss[loss=0.1945, simple_loss=0.2618, pruned_loss=0.06357, over 954776.52 frames. ], batch size: 35, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:29:07,487 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65332.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 14:29:10,444 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.717e+01 1.673e+02 1.949e+02 2.267e+02 6.743e+02, threshold=3.897e+02, percent-clipped=1.0 2023-03-26 14:29:16,552 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65346.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:29:20,082 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=65352.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:29:22,339 INFO [finetune.py:976] (3/7) Epoch 12, batch 2350, loss[loss=0.2207, simple_loss=0.2857, pruned_loss=0.07789, over 4834.00 frames. ], tot_loss[loss=0.1925, simple_loss=0.2589, pruned_loss=0.06304, over 954346.21 frames. ], batch size: 47, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:29:24,809 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65359.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:29:32,634 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2621, 1.9385, 2.0095, 0.8867, 2.1972, 2.4577, 2.0783, 1.8733], device='cuda:3'), covar=tensor([0.0977, 0.0855, 0.0566, 0.0782, 0.0503, 0.0536, 0.0564, 0.0731], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0152, 0.0121, 0.0131, 0.0130, 0.0126, 0.0142, 0.0145], device='cuda:3'), out_proj_covar=tensor([9.2809e-05, 1.1142e-04, 8.7084e-05, 9.4662e-05, 9.2455e-05, 9.1284e-05, 1.0393e-04, 1.0597e-04], device='cuda:3') 2023-03-26 14:30:02,661 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65401.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:30:04,997 INFO [finetune.py:976] (3/7) Epoch 12, batch 2400, loss[loss=0.1757, simple_loss=0.2389, pruned_loss=0.05622, over 4761.00 frames. ], tot_loss[loss=0.1891, simple_loss=0.2556, pruned_loss=0.06127, over 954266.93 frames. ], batch size: 28, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:30:06,756 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=65407.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:30:07,395 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65408.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:30:09,180 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65411.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:30:14,073 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.6308, 1.7004, 1.7104, 0.8995, 1.6943, 2.0030, 1.9244, 1.4883], device='cuda:3'), covar=tensor([0.0995, 0.0649, 0.0470, 0.0651, 0.0509, 0.0678, 0.0392, 0.0775], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0152, 0.0121, 0.0131, 0.0130, 0.0126, 0.0142, 0.0145], device='cuda:3'), out_proj_covar=tensor([9.2872e-05, 1.1132e-04, 8.6893e-05, 9.4529e-05, 9.2507e-05, 9.1293e-05, 1.0378e-04, 1.0590e-04], device='cuda:3') 2023-03-26 14:30:26,324 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.174e+02 1.534e+02 1.885e+02 2.327e+02 5.518e+02, threshold=3.771e+02, percent-clipped=1.0 2023-03-26 14:30:34,675 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=65449.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:30:38,806 INFO [finetune.py:976] (3/7) Epoch 12, batch 2450, loss[loss=0.1817, simple_loss=0.2516, pruned_loss=0.05587, over 4866.00 frames. ], tot_loss[loss=0.1878, simple_loss=0.2538, pruned_loss=0.06088, over 954731.46 frames. ], batch size: 34, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:30:39,471 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=65456.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:30:51,544 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6235, 3.3464, 3.2655, 1.4722, 3.5171, 2.7384, 0.7623, 2.3903], device='cuda:3'), covar=tensor([0.2249, 0.1906, 0.1685, 0.3683, 0.1162, 0.0942, 0.4499, 0.1654], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0173, 0.0158, 0.0127, 0.0155, 0.0121, 0.0145, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 14:31:27,821 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0412, 1.7310, 2.5197, 3.7127, 2.6446, 2.5534, 0.9851, 2.9554], device='cuda:3'), covar=tensor([0.1597, 0.1444, 0.1157, 0.0485, 0.0732, 0.1928, 0.1862, 0.0454], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0116, 0.0134, 0.0165, 0.0100, 0.0138, 0.0126, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 14:31:31,329 INFO [finetune.py:976] (3/7) Epoch 12, batch 2500, loss[loss=0.1599, simple_loss=0.2272, pruned_loss=0.04632, over 4771.00 frames. ], tot_loss[loss=0.1895, simple_loss=0.2554, pruned_loss=0.06178, over 954107.80 frames. ], batch size: 26, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:31:49,217 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65532.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 14:31:52,647 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.473e+01 1.650e+02 2.020e+02 2.341e+02 4.049e+02, threshold=4.040e+02, percent-clipped=1.0 2023-03-26 14:31:56,073 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-26 14:32:06,906 INFO [finetune.py:976] (3/7) Epoch 12, batch 2550, loss[loss=0.1976, simple_loss=0.2651, pruned_loss=0.06506, over 4918.00 frames. ], tot_loss[loss=0.1892, simple_loss=0.2557, pruned_loss=0.06134, over 950167.05 frames. ], batch size: 36, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:32:29,656 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8092, 4.3673, 4.1431, 2.7464, 4.4879, 3.4729, 0.9178, 3.0985], device='cuda:3'), covar=tensor([0.2547, 0.1405, 0.1288, 0.2471, 0.0666, 0.0786, 0.4452, 0.1183], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0174, 0.0159, 0.0128, 0.0156, 0.0122, 0.0146, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 14:32:40,934 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65593.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 14:32:42,663 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6144, 4.0449, 3.9321, 2.2446, 4.1728, 3.0793, 0.9262, 2.9251], device='cuda:3'), covar=tensor([0.2215, 0.1535, 0.1209, 0.2539, 0.0894, 0.1030, 0.4080, 0.1249], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0173, 0.0158, 0.0127, 0.0155, 0.0121, 0.0146, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 14:32:48,715 INFO [finetune.py:976] (3/7) Epoch 12, batch 2600, loss[loss=0.2303, simple_loss=0.3001, pruned_loss=0.08024, over 4797.00 frames. ], tot_loss[loss=0.1928, simple_loss=0.2587, pruned_loss=0.0634, over 948956.47 frames. ], batch size: 51, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:32:58,352 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.1672, 1.2916, 1.3923, 0.7260, 1.2529, 1.5344, 1.5937, 1.2779], device='cuda:3'), covar=tensor([0.0821, 0.0535, 0.0456, 0.0461, 0.0442, 0.0538, 0.0321, 0.0597], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0153, 0.0122, 0.0131, 0.0130, 0.0125, 0.0143, 0.0146], device='cuda:3'), out_proj_covar=tensor([9.3135e-05, 1.1160e-04, 8.7784e-05, 9.4759e-05, 9.2731e-05, 9.1188e-05, 1.0409e-04, 1.0641e-04], device='cuda:3') 2023-03-26 14:33:06,632 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65632.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:33:10,009 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.524e+01 1.645e+02 2.049e+02 2.494e+02 4.393e+02, threshold=4.097e+02, percent-clipped=1.0 2023-03-26 14:33:12,518 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65641.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:33:22,368 INFO [finetune.py:976] (3/7) Epoch 12, batch 2650, loss[loss=0.1836, simple_loss=0.2563, pruned_loss=0.05547, over 4828.00 frames. ], tot_loss[loss=0.1938, simple_loss=0.2603, pruned_loss=0.06359, over 949620.78 frames. ], batch size: 47, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:33:34,535 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.3370, 3.7417, 3.9659, 4.1652, 4.0976, 3.7722, 4.4143, 1.4648], device='cuda:3'), covar=tensor([0.0681, 0.0836, 0.0706, 0.0875, 0.1142, 0.1469, 0.0604, 0.5451], device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0242, 0.0276, 0.0291, 0.0330, 0.0282, 0.0300, 0.0296], 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') 2023-03-26 14:33:38,619 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=65680.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:33:55,609 INFO [finetune.py:976] (3/7) Epoch 12, batch 2700, loss[loss=0.2405, simple_loss=0.294, pruned_loss=0.09348, over 4272.00 frames. ], tot_loss[loss=0.1923, simple_loss=0.2588, pruned_loss=0.06293, over 947482.37 frames. ], batch size: 65, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:33:59,776 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65711.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:34:03,244 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6186, 3.3168, 3.3284, 1.9150, 3.5977, 2.7038, 1.3359, 2.5183], device='cuda:3'), covar=tensor([0.2987, 0.1913, 0.1597, 0.2815, 0.0943, 0.0987, 0.3566, 0.1339], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0174, 0.0159, 0.0128, 0.0155, 0.0122, 0.0146, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 14:34:17,011 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.115e+02 1.547e+02 1.884e+02 2.200e+02 3.210e+02, threshold=3.769e+02, percent-clipped=0.0 2023-03-26 14:34:21,323 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.92 vs. limit=5.0 2023-03-26 14:34:30,282 INFO [finetune.py:976] (3/7) Epoch 12, batch 2750, loss[loss=0.1672, simple_loss=0.2417, pruned_loss=0.0463, over 4813.00 frames. ], tot_loss[loss=0.1913, simple_loss=0.2572, pruned_loss=0.06272, over 947443.44 frames. ], batch size: 41, lr: 3.65e-03, grad_scale: 32.0 2023-03-26 14:34:38,002 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=65759.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:35:22,234 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6156, 2.2143, 2.9250, 1.7296, 2.6970, 2.6545, 2.1178, 2.8687], device='cuda:3'), covar=tensor([0.1404, 0.2146, 0.1816, 0.2447, 0.0864, 0.1771, 0.2787, 0.0950], device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0207, 0.0195, 0.0192, 0.0179, 0.0216, 0.0218, 0.0201], 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') 2023-03-26 14:35:30,779 INFO [finetune.py:976] (3/7) Epoch 12, batch 2800, loss[loss=0.1908, simple_loss=0.2567, pruned_loss=0.06243, over 4833.00 frames. ], tot_loss[loss=0.1895, simple_loss=0.255, pruned_loss=0.06196, over 950588.19 frames. ], batch size: 33, lr: 3.64e-03, grad_scale: 32.0 2023-03-26 14:35:52,227 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.914e+01 1.578e+02 1.887e+02 2.176e+02 5.167e+02, threshold=3.774e+02, percent-clipped=1.0 2023-03-26 14:35:52,957 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65838.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 14:36:04,205 INFO [finetune.py:976] (3/7) Epoch 12, batch 2850, loss[loss=0.1595, simple_loss=0.2354, pruned_loss=0.04177, over 4900.00 frames. ], tot_loss[loss=0.1879, simple_loss=0.2532, pruned_loss=0.06125, over 953024.59 frames. ], batch size: 32, lr: 3.64e-03, grad_scale: 32.0 2023-03-26 14:36:28,854 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-26 14:36:36,963 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65888.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 14:36:45,242 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65899.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 14:36:48,785 INFO [finetune.py:976] (3/7) Epoch 12, batch 2900, loss[loss=0.175, simple_loss=0.2583, pruned_loss=0.0458, over 4748.00 frames. ], tot_loss[loss=0.1903, simple_loss=0.2556, pruned_loss=0.06249, over 952347.99 frames. ], batch size: 54, lr: 3.64e-03, grad_scale: 32.0 2023-03-26 14:37:10,263 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.038e+02 1.527e+02 1.842e+02 2.377e+02 4.547e+02, threshold=3.684e+02, percent-clipped=3.0 2023-03-26 14:37:12,798 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65941.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:37:27,698 INFO [finetune.py:976] (3/7) Epoch 12, batch 2950, loss[loss=0.2089, simple_loss=0.2836, pruned_loss=0.06706, over 4818.00 frames. ], tot_loss[loss=0.1931, simple_loss=0.2595, pruned_loss=0.06338, over 953082.50 frames. ], batch size: 38, lr: 3.64e-03, grad_scale: 32.0 2023-03-26 14:37:38,804 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.52 vs. limit=5.0 2023-03-26 14:37:59,196 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-26 14:38:02,422 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=65989.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:38:26,400 INFO [finetune.py:976] (3/7) Epoch 12, batch 3000, loss[loss=0.1766, simple_loss=0.243, pruned_loss=0.05513, over 4723.00 frames. ], tot_loss[loss=0.1941, simple_loss=0.2607, pruned_loss=0.06373, over 953371.68 frames. ], batch size: 23, lr: 3.64e-03, grad_scale: 32.0 2023-03-26 14:38:26,401 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 14:38:34,436 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9264, 1.8642, 2.0215, 1.2600, 1.9958, 2.0580, 1.9322, 1.6793], device='cuda:3'), covar=tensor([0.0513, 0.0580, 0.0561, 0.0847, 0.0687, 0.0547, 0.0536, 0.1012], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0132, 0.0140, 0.0122, 0.0121, 0.0141, 0.0141, 0.0160], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 14:38:37,073 INFO [finetune.py:1010] (3/7) Epoch 12, validation: loss=0.1571, simple_loss=0.2281, pruned_loss=0.04309, over 2265189.00 frames. 2023-03-26 14:38:37,073 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 14:38:43,788 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.1575, 1.3636, 1.4794, 0.6762, 1.3658, 1.5779, 1.6448, 1.3198], device='cuda:3'), covar=tensor([0.0817, 0.0559, 0.0480, 0.0517, 0.0468, 0.0584, 0.0324, 0.0673], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0153, 0.0123, 0.0132, 0.0131, 0.0127, 0.0144, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.3576e-05, 1.1208e-04, 8.8320e-05, 9.5280e-05, 9.3043e-05, 9.1926e-05, 1.0493e-04, 1.0691e-04], device='cuda:3') 2023-03-26 14:38:52,811 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8590, 1.5928, 2.3446, 1.5172, 2.0499, 2.1658, 1.5671, 2.2731], device='cuda:3'), covar=tensor([0.1488, 0.2363, 0.1372, 0.2004, 0.0956, 0.1688, 0.2984, 0.1032], device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0206, 0.0195, 0.0191, 0.0179, 0.0215, 0.0218, 0.0201], 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') 2023-03-26 14:38:58,513 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.011e+02 1.619e+02 1.943e+02 2.343e+02 4.325e+02, threshold=3.886e+02, percent-clipped=3.0 2023-03-26 14:39:21,450 INFO [finetune.py:976] (3/7) Epoch 12, batch 3050, loss[loss=0.2382, simple_loss=0.2979, pruned_loss=0.08923, over 4871.00 frames. ], tot_loss[loss=0.1958, simple_loss=0.2623, pruned_loss=0.06466, over 951158.83 frames. ], batch size: 34, lr: 3.64e-03, grad_scale: 32.0 2023-03-26 14:39:41,791 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.77 vs. limit=2.0 2023-03-26 14:40:36,352 INFO [finetune.py:976] (3/7) Epoch 12, batch 3100, loss[loss=0.1514, simple_loss=0.2292, pruned_loss=0.03683, over 4826.00 frames. ], tot_loss[loss=0.1938, simple_loss=0.2597, pruned_loss=0.06391, over 951665.13 frames. ], batch size: 39, lr: 3.64e-03, grad_scale: 32.0 2023-03-26 14:41:19,787 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.081e+02 1.628e+02 1.972e+02 2.398e+02 4.316e+02, threshold=3.945e+02, percent-clipped=3.0 2023-03-26 14:41:26,913 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1488, 2.1358, 2.2135, 1.5754, 2.1868, 2.3557, 2.2120, 1.8930], device='cuda:3'), covar=tensor([0.0594, 0.0638, 0.0681, 0.0834, 0.0564, 0.0585, 0.0575, 0.0993], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0133, 0.0141, 0.0123, 0.0122, 0.0141, 0.0142, 0.0161], 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') 2023-03-26 14:41:27,558 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66144.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:41:40,145 INFO [finetune.py:976] (3/7) Epoch 12, batch 3150, loss[loss=0.193, simple_loss=0.2553, pruned_loss=0.06529, over 4745.00 frames. ], tot_loss[loss=0.1917, simple_loss=0.2572, pruned_loss=0.06312, over 952785.00 frames. ], batch size: 59, lr: 3.64e-03, grad_scale: 32.0 2023-03-26 14:41:43,353 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66160.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 14:42:24,576 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66188.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 14:42:32,947 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66194.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 14:42:35,881 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.0763, 1.0324, 1.0558, 0.5334, 0.9668, 1.2302, 1.2066, 1.0678], device='cuda:3'), covar=tensor([0.0911, 0.0555, 0.0535, 0.0535, 0.0540, 0.0544, 0.0429, 0.0674], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0153, 0.0122, 0.0132, 0.0130, 0.0126, 0.0144, 0.0146], device='cuda:3'), out_proj_covar=tensor([9.3816e-05, 1.1206e-04, 8.8080e-05, 9.5261e-05, 9.2672e-05, 9.1631e-05, 1.0509e-04, 1.0671e-04], device='cuda:3') 2023-03-26 14:42:40,452 INFO [finetune.py:976] (3/7) Epoch 12, batch 3200, loss[loss=0.1849, simple_loss=0.2331, pruned_loss=0.06833, over 4821.00 frames. ], tot_loss[loss=0.1881, simple_loss=0.2533, pruned_loss=0.0614, over 951363.72 frames. ], batch size: 25, lr: 3.64e-03, grad_scale: 32.0 2023-03-26 14:42:40,569 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66205.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 14:42:43,542 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8783, 2.1134, 1.7634, 1.6706, 2.3883, 2.3281, 1.9978, 1.8707], device='cuda:3'), covar=tensor([0.0401, 0.0308, 0.0563, 0.0337, 0.0313, 0.0547, 0.0341, 0.0413], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0107, 0.0138, 0.0112, 0.0101, 0.0103, 0.0092, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.1193e-05, 8.3137e-05, 1.0976e-04, 8.7553e-05, 7.9105e-05, 7.5979e-05, 6.9852e-05, 8.2799e-05], device='cuda:3') 2023-03-26 14:42:50,947 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.68 vs. limit=5.0 2023-03-26 14:42:51,173 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66221.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 14:43:01,312 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=66236.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 14:43:01,823 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.059e+02 1.602e+02 1.878e+02 2.419e+02 4.134e+02, threshold=3.755e+02, percent-clipped=1.0 2023-03-26 14:43:14,215 INFO [finetune.py:976] (3/7) Epoch 12, batch 3250, loss[loss=0.1716, simple_loss=0.2478, pruned_loss=0.04768, over 4781.00 frames. ], tot_loss[loss=0.1892, simple_loss=0.2544, pruned_loss=0.06201, over 953102.99 frames. ], batch size: 26, lr: 3.64e-03, grad_scale: 32.0 2023-03-26 14:43:16,212 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66258.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:43:18,725 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1027, 1.9555, 1.6077, 2.0152, 2.0598, 1.7347, 2.3644, 2.0476], device='cuda:3'), covar=tensor([0.1305, 0.2274, 0.3080, 0.2635, 0.2560, 0.1730, 0.3094, 0.1903], device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0187, 0.0233, 0.0255, 0.0241, 0.0198, 0.0214, 0.0197], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 14:43:48,267 INFO [finetune.py:976] (3/7) Epoch 12, batch 3300, loss[loss=0.2172, simple_loss=0.2942, pruned_loss=0.07012, over 4822.00 frames. ], tot_loss[loss=0.1931, simple_loss=0.2589, pruned_loss=0.06368, over 955341.82 frames. ], batch size: 38, lr: 3.64e-03, grad_scale: 32.0 2023-03-26 14:43:56,985 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66319.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:44:14,657 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.247e+02 1.642e+02 1.904e+02 2.370e+02 4.024e+02, threshold=3.808e+02, percent-clipped=2.0 2023-03-26 14:44:22,794 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.08 vs. limit=5.0 2023-03-26 14:44:29,722 INFO [finetune.py:976] (3/7) Epoch 12, batch 3350, loss[loss=0.2166, simple_loss=0.2881, pruned_loss=0.07257, over 4829.00 frames. ], tot_loss[loss=0.1942, simple_loss=0.2604, pruned_loss=0.06394, over 956037.20 frames. ], batch size: 47, lr: 3.64e-03, grad_scale: 64.0 2023-03-26 14:45:02,384 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-26 14:45:02,613 INFO [finetune.py:976] (3/7) Epoch 12, batch 3400, loss[loss=0.2122, simple_loss=0.2859, pruned_loss=0.06927, over 4834.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.2624, pruned_loss=0.06503, over 955932.23 frames. ], batch size: 44, lr: 3.64e-03, grad_scale: 64.0 2023-03-26 14:45:24,436 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.132e+02 1.693e+02 1.994e+02 2.432e+02 3.824e+02, threshold=3.988e+02, percent-clipped=2.0 2023-03-26 14:45:30,613 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7026, 4.1521, 4.0470, 2.0172, 4.3501, 3.3097, 0.9034, 2.9254], device='cuda:3'), covar=tensor([0.2717, 0.1537, 0.1410, 0.3153, 0.0771, 0.0872, 0.4198, 0.1376], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0177, 0.0162, 0.0130, 0.0158, 0.0123, 0.0148, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 14:45:36,091 INFO [finetune.py:976] (3/7) Epoch 12, batch 3450, loss[loss=0.207, simple_loss=0.2579, pruned_loss=0.07809, over 4824.00 frames. ], tot_loss[loss=0.1955, simple_loss=0.2621, pruned_loss=0.0645, over 955874.65 frames. ], batch size: 30, lr: 3.64e-03, grad_scale: 64.0 2023-03-26 14:46:02,767 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66494.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 14:46:06,885 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66500.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 14:46:09,848 INFO [finetune.py:976] (3/7) Epoch 12, batch 3500, loss[loss=0.2053, simple_loss=0.2745, pruned_loss=0.06802, over 4820.00 frames. ], tot_loss[loss=0.196, simple_loss=0.2615, pruned_loss=0.06525, over 956349.41 frames. ], batch size: 38, lr: 3.64e-03, grad_scale: 64.0 2023-03-26 14:46:20,797 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66516.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 14:46:23,826 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66521.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 14:46:29,337 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-03-26 14:46:36,430 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.070e+02 1.655e+02 1.937e+02 2.486e+02 6.010e+02, threshold=3.875e+02, percent-clipped=2.0 2023-03-26 14:46:39,447 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7675, 1.6810, 1.5335, 1.7061, 1.1044, 3.8306, 1.4537, 1.9490], device='cuda:3'), covar=tensor([0.3174, 0.2497, 0.2099, 0.2256, 0.1866, 0.0186, 0.2479, 0.1292], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0116, 0.0120, 0.0124, 0.0115, 0.0098, 0.0098, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 14:46:40,000 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=66542.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 14:46:57,614 INFO [finetune.py:976] (3/7) Epoch 12, batch 3550, loss[loss=0.1446, simple_loss=0.2156, pruned_loss=0.03684, over 4784.00 frames. ], tot_loss[loss=0.193, simple_loss=0.2583, pruned_loss=0.06383, over 956446.81 frames. ], batch size: 26, lr: 3.64e-03, grad_scale: 64.0 2023-03-26 14:47:23,092 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66582.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 14:47:24,902 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.5700, 3.9286, 4.2113, 4.3833, 4.3359, 4.0078, 4.6408, 1.4350], device='cuda:3'), covar=tensor([0.0699, 0.0813, 0.0933, 0.0969, 0.1125, 0.1652, 0.0589, 0.5784], device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0245, 0.0278, 0.0293, 0.0331, 0.0285, 0.0302, 0.0299], 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') 2023-03-26 14:47:43,653 INFO [finetune.py:976] (3/7) Epoch 12, batch 3600, loss[loss=0.1248, simple_loss=0.1893, pruned_loss=0.03011, over 4229.00 frames. ], tot_loss[loss=0.1894, simple_loss=0.2547, pruned_loss=0.06205, over 955307.93 frames. ], batch size: 18, lr: 3.64e-03, grad_scale: 64.0 2023-03-26 14:47:44,099 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-26 14:47:53,889 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66614.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:48:08,738 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.078e+02 1.581e+02 1.999e+02 2.430e+02 3.919e+02, threshold=3.999e+02, percent-clipped=1.0 2023-03-26 14:48:21,112 INFO [finetune.py:976] (3/7) Epoch 12, batch 3650, loss[loss=0.1861, simple_loss=0.2627, pruned_loss=0.05482, over 4818.00 frames. ], tot_loss[loss=0.1909, simple_loss=0.2567, pruned_loss=0.06256, over 953982.94 frames. ], batch size: 39, lr: 3.64e-03, grad_scale: 64.0 2023-03-26 14:48:41,057 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-26 14:48:54,888 INFO [finetune.py:976] (3/7) Epoch 12, batch 3700, loss[loss=0.1828, simple_loss=0.2583, pruned_loss=0.05358, over 4741.00 frames. ], tot_loss[loss=0.193, simple_loss=0.2597, pruned_loss=0.06314, over 952981.30 frames. ], batch size: 54, lr: 3.64e-03, grad_scale: 64.0 2023-03-26 14:48:55,601 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7492, 1.5933, 2.2619, 3.0197, 2.0869, 2.2892, 1.3849, 2.3527], device='cuda:3'), covar=tensor([0.1393, 0.1250, 0.0957, 0.0561, 0.0720, 0.1819, 0.1347, 0.0497], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0116, 0.0134, 0.0165, 0.0101, 0.0138, 0.0126, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 14:49:14,941 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66731.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 14:49:18,459 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.046e+02 1.675e+02 1.999e+02 2.466e+02 3.717e+02, threshold=3.997e+02, percent-clipped=0.0 2023-03-26 14:49:38,724 INFO [finetune.py:976] (3/7) Epoch 12, batch 3750, loss[loss=0.1787, simple_loss=0.2523, pruned_loss=0.05259, over 4781.00 frames. ], tot_loss[loss=0.1948, simple_loss=0.2618, pruned_loss=0.0639, over 953706.30 frames. ], batch size: 51, lr: 3.64e-03, grad_scale: 32.0 2023-03-26 14:49:39,243 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-26 14:49:58,396 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-26 14:50:03,076 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66792.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 14:50:08,993 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66800.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 14:50:12,444 INFO [finetune.py:976] (3/7) Epoch 12, batch 3800, loss[loss=0.16, simple_loss=0.2463, pruned_loss=0.03681, over 4814.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.2627, pruned_loss=0.06453, over 953605.93 frames. ], batch size: 38, lr: 3.64e-03, grad_scale: 32.0 2023-03-26 14:50:19,260 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66816.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 14:50:34,226 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.175e+02 1.668e+02 2.101e+02 2.666e+02 4.038e+02, threshold=4.202e+02, percent-clipped=1.0 2023-03-26 14:50:40,364 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=66848.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:50:45,466 INFO [finetune.py:976] (3/7) Epoch 12, batch 3850, loss[loss=0.2291, simple_loss=0.2921, pruned_loss=0.08306, over 4909.00 frames. ], tot_loss[loss=0.1942, simple_loss=0.2611, pruned_loss=0.06363, over 951982.02 frames. ], batch size: 46, lr: 3.64e-03, grad_scale: 32.0 2023-03-26 14:50:51,376 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=66864.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 14:51:00,116 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66877.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 14:51:18,884 INFO [finetune.py:976] (3/7) Epoch 12, batch 3900, loss[loss=0.1829, simple_loss=0.2524, pruned_loss=0.05675, over 4934.00 frames. ], tot_loss[loss=0.1913, simple_loss=0.2576, pruned_loss=0.06251, over 952050.81 frames. ], batch size: 33, lr: 3.64e-03, grad_scale: 32.0 2023-03-26 14:51:24,973 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66914.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:51:40,887 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.325e+01 1.578e+02 1.785e+02 2.294e+02 5.103e+02, threshold=3.570e+02, percent-clipped=1.0 2023-03-26 14:51:51,222 INFO [finetune.py:976] (3/7) Epoch 12, batch 3950, loss[loss=0.1797, simple_loss=0.2441, pruned_loss=0.05764, over 4834.00 frames. ], tot_loss[loss=0.1867, simple_loss=0.2528, pruned_loss=0.06032, over 953937.93 frames. ], batch size: 30, lr: 3.64e-03, grad_scale: 32.0 2023-03-26 14:51:53,022 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66957.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:51:58,480 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=66962.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:52:16,910 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1670, 2.0475, 1.6212, 2.1604, 2.1042, 1.8213, 2.4959, 2.1569], device='cuda:3'), covar=tensor([0.1316, 0.2321, 0.3128, 0.2608, 0.2424, 0.1668, 0.3090, 0.1875], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0188, 0.0234, 0.0256, 0.0243, 0.0199, 0.0214, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 14:52:47,107 INFO [finetune.py:976] (3/7) Epoch 12, batch 4000, loss[loss=0.235, simple_loss=0.2874, pruned_loss=0.09135, over 4870.00 frames. ], tot_loss[loss=0.1879, simple_loss=0.2535, pruned_loss=0.06115, over 952090.26 frames. ], batch size: 34, lr: 3.63e-03, grad_scale: 32.0 2023-03-26 14:53:04,492 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67018.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 14:53:16,866 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-03-26 14:53:18,625 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.442e+01 1.595e+02 2.014e+02 2.521e+02 4.335e+02, threshold=4.027e+02, percent-clipped=3.0 2023-03-26 14:53:28,873 INFO [finetune.py:976] (3/7) Epoch 12, batch 4050, loss[loss=0.2112, simple_loss=0.2744, pruned_loss=0.07395, over 4821.00 frames. ], tot_loss[loss=0.1902, simple_loss=0.2565, pruned_loss=0.06197, over 952554.76 frames. ], batch size: 45, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 14:53:42,360 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.95 vs. limit=5.0 2023-03-26 14:53:47,875 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67083.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:53:50,806 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67087.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 14:54:02,029 INFO [finetune.py:976] (3/7) Epoch 12, batch 4100, loss[loss=0.2406, simple_loss=0.3051, pruned_loss=0.08808, over 4827.00 frames. ], tot_loss[loss=0.192, simple_loss=0.259, pruned_loss=0.06251, over 954143.52 frames. ], batch size: 39, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 14:54:15,030 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8054, 1.7720, 1.6328, 1.7216, 1.1992, 4.2718, 1.6899, 2.2182], device='cuda:3'), covar=tensor([0.3488, 0.2486, 0.2123, 0.2484, 0.1883, 0.0143, 0.2382, 0.1271], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0116, 0.0121, 0.0124, 0.0116, 0.0099, 0.0099, 0.0098], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 14:54:29,986 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.152e+02 1.725e+02 1.998e+02 2.409e+02 3.172e+02, threshold=3.997e+02, percent-clipped=0.0 2023-03-26 14:54:34,098 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67144.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:54:44,209 INFO [finetune.py:976] (3/7) Epoch 12, batch 4150, loss[loss=0.1781, simple_loss=0.2525, pruned_loss=0.0518, over 4792.00 frames. ], tot_loss[loss=0.195, simple_loss=0.2615, pruned_loss=0.06425, over 954042.23 frames. ], batch size: 51, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 14:54:59,070 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67177.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 14:55:17,549 INFO [finetune.py:976] (3/7) Epoch 12, batch 4200, loss[loss=0.1593, simple_loss=0.2375, pruned_loss=0.04053, over 4790.00 frames. ], tot_loss[loss=0.196, simple_loss=0.2629, pruned_loss=0.06452, over 954646.15 frames. ], batch size: 51, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 14:55:18,802 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7118, 1.2969, 0.7249, 1.6726, 2.0989, 1.3872, 1.4196, 1.7285], device='cuda:3'), covar=tensor([0.1358, 0.1977, 0.2272, 0.1155, 0.1844, 0.2067, 0.1452, 0.1887], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0096, 0.0113, 0.0091, 0.0120, 0.0094, 0.0099, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 14:55:30,584 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=67225.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 14:55:36,872 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-26 14:55:39,456 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.066e+02 1.549e+02 1.852e+02 2.427e+02 4.145e+02, threshold=3.704e+02, percent-clipped=1.0 2023-03-26 14:55:41,910 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5779, 2.5482, 2.0429, 1.1051, 2.3054, 1.9711, 1.7795, 2.1966], device='cuda:3'), covar=tensor([0.0898, 0.0677, 0.1682, 0.2064, 0.1480, 0.2224, 0.2114, 0.1039], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0197, 0.0198, 0.0184, 0.0212, 0.0205, 0.0222, 0.0196], 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') 2023-03-26 14:55:50,531 INFO [finetune.py:976] (3/7) Epoch 12, batch 4250, loss[loss=0.1952, simple_loss=0.258, pruned_loss=0.06616, over 4883.00 frames. ], tot_loss[loss=0.1937, simple_loss=0.26, pruned_loss=0.06369, over 954081.35 frames. ], batch size: 35, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 14:55:55,532 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1328, 3.5927, 3.7801, 3.9932, 3.9223, 3.7119, 4.1900, 1.2988], device='cuda:3'), covar=tensor([0.0729, 0.0752, 0.0850, 0.0920, 0.1151, 0.1354, 0.0697, 0.5319], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0242, 0.0274, 0.0289, 0.0327, 0.0280, 0.0300, 0.0294], 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') 2023-03-26 14:55:58,667 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6431, 0.6730, 1.6448, 1.5175, 1.4299, 1.3375, 1.4094, 1.5319], device='cuda:3'), covar=tensor([0.3552, 0.3972, 0.3357, 0.3466, 0.4616, 0.3512, 0.4171, 0.3124], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0238, 0.0254, 0.0260, 0.0257, 0.0232, 0.0273, 0.0232], 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') 2023-03-26 14:56:21,832 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2376, 2.1792, 1.6496, 2.1853, 2.1269, 1.8577, 2.8571, 2.1773], device='cuda:3'), covar=tensor([0.1587, 0.2261, 0.3794, 0.2957, 0.2979, 0.1993, 0.2278, 0.2200], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0189, 0.0235, 0.0257, 0.0244, 0.0200, 0.0215, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 14:56:28,365 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8576, 1.8024, 1.6628, 2.0321, 2.2739, 2.1001, 1.5525, 1.5325], device='cuda:3'), covar=tensor([0.2468, 0.2199, 0.2157, 0.1813, 0.1995, 0.1171, 0.2736, 0.2164], device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0205, 0.0209, 0.0189, 0.0239, 0.0181, 0.0212, 0.0198], 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') 2023-03-26 14:56:29,019 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-03-26 14:56:32,187 INFO [finetune.py:976] (3/7) Epoch 12, batch 4300, loss[loss=0.1665, simple_loss=0.244, pruned_loss=0.04446, over 4754.00 frames. ], tot_loss[loss=0.1918, simple_loss=0.2575, pruned_loss=0.06307, over 954392.48 frames. ], batch size: 28, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 14:56:37,170 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67313.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 14:56:54,399 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.954e+01 1.550e+02 1.913e+02 2.348e+02 5.397e+02, threshold=3.825e+02, percent-clipped=3.0 2023-03-26 14:56:58,209 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.2511, 1.1855, 1.3171, 0.5528, 1.1646, 1.5063, 1.4917, 1.2572], device='cuda:3'), covar=tensor([0.0945, 0.0837, 0.0521, 0.0593, 0.0528, 0.0501, 0.0406, 0.0695], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0156, 0.0124, 0.0133, 0.0132, 0.0127, 0.0146, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.4307e-05, 1.1380e-04, 8.9485e-05, 9.6176e-05, 9.3979e-05, 9.2427e-05, 1.0630e-04, 1.0743e-04], device='cuda:3') 2023-03-26 14:57:04,553 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4348, 1.2776, 1.6491, 2.4504, 1.6983, 2.2253, 0.9829, 2.0779], device='cuda:3'), covar=tensor([0.1703, 0.1519, 0.1192, 0.0848, 0.0932, 0.1149, 0.1524, 0.0659], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0116, 0.0134, 0.0165, 0.0101, 0.0137, 0.0126, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 14:57:05,081 INFO [finetune.py:976] (3/7) Epoch 12, batch 4350, loss[loss=0.2142, simple_loss=0.2766, pruned_loss=0.07591, over 4890.00 frames. ], tot_loss[loss=0.1898, simple_loss=0.2552, pruned_loss=0.06226, over 955719.57 frames. ], batch size: 32, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 14:57:11,236 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-03-26 14:57:28,544 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67387.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 14:57:39,911 INFO [finetune.py:976] (3/7) Epoch 12, batch 4400, loss[loss=0.2139, simple_loss=0.2741, pruned_loss=0.07679, over 4850.00 frames. ], tot_loss[loss=0.1908, simple_loss=0.2555, pruned_loss=0.06302, over 954424.64 frames. ], batch size: 47, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 14:58:14,139 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=67435.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 14:58:16,974 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.120e+02 1.656e+02 1.972e+02 2.339e+02 4.406e+02, threshold=3.944e+02, percent-clipped=2.0 2023-03-26 14:58:17,065 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67439.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:58:30,809 INFO [finetune.py:976] (3/7) Epoch 12, batch 4450, loss[loss=0.2143, simple_loss=0.2841, pruned_loss=0.07228, over 4775.00 frames. ], tot_loss[loss=0.1924, simple_loss=0.2583, pruned_loss=0.06331, over 953650.59 frames. ], batch size: 59, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 14:59:03,971 INFO [finetune.py:976] (3/7) Epoch 12, batch 4500, loss[loss=0.2051, simple_loss=0.2749, pruned_loss=0.06769, over 4827.00 frames. ], tot_loss[loss=0.1938, simple_loss=0.2599, pruned_loss=0.06386, over 953147.49 frames. ], batch size: 30, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 14:59:09,848 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67513.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 14:59:21,634 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-26 14:59:26,035 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.163e+02 1.686e+02 1.980e+02 2.352e+02 4.001e+02, threshold=3.961e+02, percent-clipped=1.0 2023-03-26 14:59:37,240 INFO [finetune.py:976] (3/7) Epoch 12, batch 4550, loss[loss=0.2419, simple_loss=0.2856, pruned_loss=0.09912, over 4206.00 frames. ], tot_loss[loss=0.1958, simple_loss=0.2622, pruned_loss=0.06474, over 953592.58 frames. ], batch size: 65, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 14:59:56,262 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67574.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 15:00:19,986 INFO [finetune.py:976] (3/7) Epoch 12, batch 4600, loss[loss=0.2401, simple_loss=0.3043, pruned_loss=0.08792, over 4815.00 frames. ], tot_loss[loss=0.1944, simple_loss=0.2608, pruned_loss=0.06397, over 954507.34 frames. ], batch size: 38, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 15:00:24,951 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67613.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:00:42,116 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.172e+02 1.477e+02 1.878e+02 2.272e+02 4.960e+02, threshold=3.756e+02, percent-clipped=1.0 2023-03-26 15:00:42,251 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4216, 1.3919, 1.5102, 0.8000, 1.5336, 1.4802, 1.4591, 1.3247], device='cuda:3'), covar=tensor([0.0598, 0.0739, 0.0671, 0.0938, 0.0779, 0.0701, 0.0620, 0.1176], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0134, 0.0142, 0.0125, 0.0123, 0.0141, 0.0142, 0.0162], 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') 2023-03-26 15:00:53,236 INFO [finetune.py:976] (3/7) Epoch 12, batch 4650, loss[loss=0.1545, simple_loss=0.2236, pruned_loss=0.04265, over 4786.00 frames. ], tot_loss[loss=0.1927, simple_loss=0.2588, pruned_loss=0.06324, over 955439.85 frames. ], batch size: 25, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 15:00:56,982 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=67661.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:01:10,465 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67680.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:01:17,718 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67692.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:01:22,671 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-03-26 15:01:31,299 INFO [finetune.py:976] (3/7) Epoch 12, batch 4700, loss[loss=0.1553, simple_loss=0.2198, pruned_loss=0.04543, over 4711.00 frames. ], tot_loss[loss=0.1906, simple_loss=0.2562, pruned_loss=0.06251, over 955562.04 frames. ], batch size: 54, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 15:01:31,383 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.3685, 3.7823, 3.9979, 4.1864, 4.1332, 3.8323, 4.4373, 1.4709], device='cuda:3'), covar=tensor([0.0776, 0.0812, 0.0911, 0.0947, 0.1103, 0.1713, 0.0678, 0.5473], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0243, 0.0276, 0.0290, 0.0329, 0.0280, 0.0301, 0.0294], 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') 2023-03-26 15:01:56,970 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.008e+02 1.553e+02 1.823e+02 2.116e+02 3.808e+02, threshold=3.646e+02, percent-clipped=1.0 2023-03-26 15:01:57,075 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67739.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:01:58,312 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67741.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:02:05,940 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67753.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 15:02:07,550 INFO [finetune.py:976] (3/7) Epoch 12, batch 4750, loss[loss=0.2245, simple_loss=0.2943, pruned_loss=0.07736, over 4761.00 frames. ], tot_loss[loss=0.19, simple_loss=0.2548, pruned_loss=0.06254, over 954692.69 frames. ], batch size: 28, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 15:02:28,910 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=67787.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:02:30,802 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2686, 1.8465, 1.8845, 0.9064, 2.1455, 2.1843, 2.1158, 1.7860], device='cuda:3'), covar=tensor([0.0814, 0.0689, 0.0563, 0.0684, 0.0489, 0.0629, 0.0469, 0.0748], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0154, 0.0123, 0.0132, 0.0132, 0.0127, 0.0145, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.3863e-05, 1.1299e-04, 8.8647e-05, 9.5443e-05, 9.3915e-05, 9.2460e-05, 1.0579e-04, 1.0693e-04], device='cuda:3') 2023-03-26 15:02:40,329 INFO [finetune.py:976] (3/7) Epoch 12, batch 4800, loss[loss=0.2445, simple_loss=0.3032, pruned_loss=0.09286, over 4872.00 frames. ], tot_loss[loss=0.1955, simple_loss=0.2603, pruned_loss=0.06533, over 953803.22 frames. ], batch size: 34, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 15:03:07,509 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.086e+02 1.756e+02 1.975e+02 2.556e+02 4.813e+02, threshold=3.950e+02, percent-clipped=3.0 2023-03-26 15:03:25,929 INFO [finetune.py:976] (3/7) Epoch 12, batch 4850, loss[loss=0.2235, simple_loss=0.2856, pruned_loss=0.08072, over 4923.00 frames. ], tot_loss[loss=0.1944, simple_loss=0.2604, pruned_loss=0.06415, over 954026.17 frames. ], batch size: 38, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 15:03:39,914 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67869.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 15:03:51,113 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2101, 2.0684, 1.4635, 0.6098, 1.8308, 1.8767, 1.6718, 1.8200], device='cuda:3'), covar=tensor([0.0883, 0.0766, 0.1534, 0.2065, 0.1263, 0.2126, 0.2225, 0.0922], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0197, 0.0199, 0.0184, 0.0212, 0.0206, 0.0221, 0.0196], 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') 2023-03-26 15:04:03,159 INFO [finetune.py:976] (3/7) Epoch 12, batch 4900, loss[loss=0.2357, simple_loss=0.2958, pruned_loss=0.08778, over 4919.00 frames. ], tot_loss[loss=0.1936, simple_loss=0.2602, pruned_loss=0.0635, over 953582.79 frames. ], batch size: 33, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 15:04:26,939 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.121e+02 1.717e+02 1.971e+02 2.418e+02 4.222e+02, threshold=3.942e+02, percent-clipped=1.0 2023-03-26 15:04:31,928 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2170, 2.4512, 2.2268, 1.6984, 2.1496, 2.7407, 2.4520, 2.1894], device='cuda:3'), covar=tensor([0.0573, 0.0555, 0.0698, 0.0859, 0.0966, 0.0533, 0.0590, 0.0883], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0133, 0.0142, 0.0124, 0.0122, 0.0141, 0.0142, 0.0161], 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') 2023-03-26 15:04:36,657 INFO [finetune.py:976] (3/7) Epoch 12, batch 4950, loss[loss=0.1812, simple_loss=0.2571, pruned_loss=0.05262, over 4833.00 frames. ], tot_loss[loss=0.1926, simple_loss=0.2597, pruned_loss=0.06279, over 952538.25 frames. ], batch size: 47, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 15:04:53,973 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67981.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:05:11,609 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.0439, 4.7288, 4.5481, 2.8574, 4.8255, 3.8386, 0.9541, 3.5457], device='cuda:3'), covar=tensor([0.2393, 0.1739, 0.1298, 0.2872, 0.0734, 0.0793, 0.4682, 0.1364], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0174, 0.0160, 0.0129, 0.0156, 0.0121, 0.0146, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 15:05:20,899 INFO [finetune.py:976] (3/7) Epoch 12, batch 5000, loss[loss=0.1744, simple_loss=0.2401, pruned_loss=0.05435, over 4902.00 frames. ], tot_loss[loss=0.1901, simple_loss=0.257, pruned_loss=0.06158, over 952086.30 frames. ], batch size: 37, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 15:05:39,129 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.73 vs. limit=5.0 2023-03-26 15:05:41,210 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68036.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:05:43,409 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.964e+01 1.543e+02 1.867e+02 2.301e+02 3.447e+02, threshold=3.734e+02, percent-clipped=0.0 2023-03-26 15:05:46,813 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68042.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:05:50,384 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68048.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 15:05:54,509 INFO [finetune.py:976] (3/7) Epoch 12, batch 5050, loss[loss=0.1556, simple_loss=0.2238, pruned_loss=0.04376, over 4763.00 frames. ], tot_loss[loss=0.1901, simple_loss=0.2558, pruned_loss=0.06214, over 952985.12 frames. ], batch size: 26, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 15:06:15,478 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68087.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 15:06:27,688 INFO [finetune.py:976] (3/7) Epoch 12, batch 5100, loss[loss=0.2057, simple_loss=0.2804, pruned_loss=0.06551, over 4836.00 frames. ], tot_loss[loss=0.1868, simple_loss=0.2522, pruned_loss=0.06068, over 953122.99 frames. ], batch size: 33, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 15:06:42,367 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-03-26 15:06:59,402 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.056e+02 1.565e+02 1.837e+02 2.198e+02 4.078e+02, threshold=3.675e+02, percent-clipped=2.0 2023-03-26 15:07:05,475 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68148.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 15:07:10,950 INFO [finetune.py:976] (3/7) Epoch 12, batch 5150, loss[loss=0.1629, simple_loss=0.2277, pruned_loss=0.04902, over 4816.00 frames. ], tot_loss[loss=0.1874, simple_loss=0.2527, pruned_loss=0.06106, over 953633.86 frames. ], batch size: 25, lr: 3.63e-03, grad_scale: 16.0 2023-03-26 15:07:16,839 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-26 15:07:19,523 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68169.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:07:34,886 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9351, 1.5741, 2.2451, 3.6015, 2.4986, 2.7842, 1.3169, 2.9221], device='cuda:3'), covar=tensor([0.1789, 0.1633, 0.1439, 0.0654, 0.0865, 0.1815, 0.1783, 0.0575], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0116, 0.0135, 0.0165, 0.0101, 0.0138, 0.0127, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 15:07:43,716 INFO [finetune.py:976] (3/7) Epoch 12, batch 5200, loss[loss=0.2587, simple_loss=0.316, pruned_loss=0.1006, over 4730.00 frames. ], tot_loss[loss=0.1912, simple_loss=0.257, pruned_loss=0.0627, over 953395.17 frames. ], batch size: 59, lr: 3.62e-03, grad_scale: 16.0 2023-03-26 15:07:51,085 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=68217.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:08:05,785 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.176e+02 1.664e+02 1.889e+02 2.252e+02 3.665e+02, threshold=3.778e+02, percent-clipped=0.0 2023-03-26 15:08:16,526 INFO [finetune.py:976] (3/7) Epoch 12, batch 5250, loss[loss=0.2009, simple_loss=0.2776, pruned_loss=0.06205, over 4814.00 frames. ], tot_loss[loss=0.1937, simple_loss=0.2604, pruned_loss=0.0635, over 955126.73 frames. ], batch size: 40, lr: 3.62e-03, grad_scale: 16.0 2023-03-26 15:08:47,224 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4159, 2.2197, 1.9176, 2.3534, 2.1210, 2.1207, 2.0508, 3.0401], device='cuda:3'), covar=tensor([0.4298, 0.5483, 0.3770, 0.5072, 0.5162, 0.2801, 0.5317, 0.1700], device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0259, 0.0224, 0.0275, 0.0243, 0.0210, 0.0246, 0.0219], 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') 2023-03-26 15:08:56,199 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.36 vs. limit=5.0 2023-03-26 15:09:03,120 INFO [finetune.py:976] (3/7) Epoch 12, batch 5300, loss[loss=0.2101, simple_loss=0.2744, pruned_loss=0.07289, over 4803.00 frames. ], tot_loss[loss=0.1953, simple_loss=0.2619, pruned_loss=0.06431, over 955479.78 frames. ], batch size: 45, lr: 3.62e-03, grad_scale: 16.0 2023-03-26 15:09:18,719 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1553, 1.9568, 1.8560, 2.2261, 2.7113, 2.1799, 1.9931, 1.7310], device='cuda:3'), covar=tensor([0.2224, 0.2061, 0.1946, 0.1586, 0.1752, 0.1158, 0.2225, 0.1912], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0206, 0.0209, 0.0189, 0.0239, 0.0181, 0.0212, 0.0197], 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') 2023-03-26 15:09:24,970 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68336.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:09:25,565 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68337.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:09:26,706 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.294e+02 1.838e+02 2.123e+02 2.651e+02 4.524e+02, threshold=4.245e+02, percent-clipped=5.0 2023-03-26 15:09:32,261 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68348.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:09:32,881 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6779, 1.6818, 1.5908, 1.7590, 1.0214, 3.7001, 1.4397, 2.0300], device='cuda:3'), covar=tensor([0.3264, 0.2236, 0.2064, 0.2186, 0.1822, 0.0152, 0.2515, 0.1203], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0120, 0.0123, 0.0115, 0.0098, 0.0098, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 15:09:34,810 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-26 15:09:36,485 INFO [finetune.py:976] (3/7) Epoch 12, batch 5350, loss[loss=0.1791, simple_loss=0.254, pruned_loss=0.05208, over 4821.00 frames. ], tot_loss[loss=0.1953, simple_loss=0.2625, pruned_loss=0.06406, over 956044.16 frames. ], batch size: 38, lr: 3.62e-03, grad_scale: 16.0 2023-03-26 15:09:54,191 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5167, 1.4366, 1.8232, 1.7791, 1.6669, 3.1203, 1.3088, 1.6190], device='cuda:3'), covar=tensor([0.0866, 0.1647, 0.1271, 0.0890, 0.1392, 0.0245, 0.1426, 0.1563], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0074, 0.0077, 0.0091, 0.0081, 0.0084, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 15:09:55,984 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=68384.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:10:04,680 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=68396.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:10:06,584 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68399.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:10:10,286 INFO [finetune.py:976] (3/7) Epoch 12, batch 5400, loss[loss=0.1784, simple_loss=0.2405, pruned_loss=0.05814, over 4856.00 frames. ], tot_loss[loss=0.1946, simple_loss=0.2609, pruned_loss=0.06417, over 956729.85 frames. ], batch size: 47, lr: 3.62e-03, grad_scale: 16.0 2023-03-26 15:10:16,840 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.4945, 1.6455, 1.6765, 0.8960, 1.7165, 1.9333, 1.9459, 1.4720], device='cuda:3'), covar=tensor([0.0958, 0.0642, 0.0453, 0.0553, 0.0402, 0.0566, 0.0304, 0.0760], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0156, 0.0125, 0.0133, 0.0133, 0.0128, 0.0146, 0.0148], device='cuda:3'), out_proj_covar=tensor([9.4312e-05, 1.1404e-04, 8.9909e-05, 9.6146e-05, 9.4338e-05, 9.3181e-05, 1.0674e-04, 1.0784e-04], device='cuda:3') 2023-03-26 15:10:40,850 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.097e+02 1.541e+02 1.801e+02 2.082e+02 4.267e+02, threshold=3.602e+02, percent-clipped=1.0 2023-03-26 15:10:44,356 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68443.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 15:10:51,600 INFO [finetune.py:976] (3/7) Epoch 12, batch 5450, loss[loss=0.1317, simple_loss=0.2032, pruned_loss=0.03005, over 4777.00 frames. ], tot_loss[loss=0.1908, simple_loss=0.2566, pruned_loss=0.06249, over 957745.16 frames. ], batch size: 26, lr: 3.62e-03, grad_scale: 16.0 2023-03-26 15:10:54,741 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68460.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:11:17,404 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7114, 1.6727, 1.6639, 1.7044, 1.1259, 3.8737, 1.3935, 1.9678], device='cuda:3'), covar=tensor([0.3368, 0.2421, 0.2051, 0.2327, 0.1936, 0.0158, 0.2543, 0.1276], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0115, 0.0120, 0.0123, 0.0115, 0.0098, 0.0098, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 15:11:24,505 INFO [finetune.py:976] (3/7) Epoch 12, batch 5500, loss[loss=0.1796, simple_loss=0.2407, pruned_loss=0.05918, over 4763.00 frames. ], tot_loss[loss=0.1879, simple_loss=0.2536, pruned_loss=0.06111, over 957403.19 frames. ], batch size: 28, lr: 3.62e-03, grad_scale: 16.0 2023-03-26 15:11:47,044 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.070e+02 1.509e+02 1.942e+02 2.407e+02 6.603e+02, threshold=3.884e+02, percent-clipped=3.0 2023-03-26 15:11:59,914 INFO [finetune.py:976] (3/7) Epoch 12, batch 5550, loss[loss=0.1865, simple_loss=0.2535, pruned_loss=0.0598, over 4906.00 frames. ], tot_loss[loss=0.1897, simple_loss=0.2552, pruned_loss=0.06209, over 958697.88 frames. ], batch size: 35, lr: 3.62e-03, grad_scale: 16.0 2023-03-26 15:12:09,511 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7922, 3.8646, 3.7036, 1.8900, 4.0266, 2.8661, 0.8135, 2.6976], device='cuda:3'), covar=tensor([0.2194, 0.2211, 0.1479, 0.3100, 0.0900, 0.1078, 0.4293, 0.1483], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0172, 0.0158, 0.0127, 0.0154, 0.0120, 0.0144, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 15:12:21,950 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.64 vs. limit=2.0 2023-03-26 15:12:23,716 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68578.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:12:29,564 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.51 vs. limit=5.0 2023-03-26 15:12:39,651 INFO [finetune.py:976] (3/7) Epoch 12, batch 5600, loss[loss=0.2082, simple_loss=0.2757, pruned_loss=0.07035, over 4818.00 frames. ], tot_loss[loss=0.1913, simple_loss=0.2579, pruned_loss=0.06236, over 958329.91 frames. ], batch size: 51, lr: 3.62e-03, grad_scale: 16.0 2023-03-26 15:12:58,320 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68637.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:12:59,422 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.238e+02 1.664e+02 1.965e+02 2.319e+02 3.885e+02, threshold=3.931e+02, percent-clipped=1.0 2023-03-26 15:12:59,536 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68639.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:13:09,174 INFO [finetune.py:976] (3/7) Epoch 12, batch 5650, loss[loss=0.1948, simple_loss=0.2691, pruned_loss=0.06024, over 4927.00 frames. ], tot_loss[loss=0.1914, simple_loss=0.2583, pruned_loss=0.06225, over 955915.89 frames. ], batch size: 36, lr: 3.62e-03, grad_scale: 16.0 2023-03-26 15:13:22,626 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2958, 2.1068, 1.8321, 2.0749, 2.0103, 1.9974, 2.0370, 2.7926], device='cuda:3'), covar=tensor([0.4248, 0.4688, 0.3634, 0.4641, 0.4379, 0.2664, 0.4429, 0.1834], device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0259, 0.0224, 0.0276, 0.0244, 0.0211, 0.0247, 0.0220], 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') 2023-03-26 15:13:27,854 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=68685.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:13:27,918 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68685.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:13:41,344 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8130, 2.5244, 2.3257, 2.8685, 2.4529, 2.5654, 2.3863, 3.4823], device='cuda:3'), covar=tensor([0.4221, 0.5383, 0.3999, 0.4388, 0.4784, 0.2777, 0.4760, 0.1687], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0260, 0.0225, 0.0277, 0.0245, 0.0211, 0.0248, 0.0221], 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') 2023-03-26 15:13:41,826 INFO [finetune.py:976] (3/7) Epoch 12, batch 5700, loss[loss=0.142, simple_loss=0.2022, pruned_loss=0.04093, over 4185.00 frames. ], tot_loss[loss=0.1879, simple_loss=0.2542, pruned_loss=0.0608, over 941224.28 frames. ], batch size: 18, lr: 3.62e-03, grad_scale: 16.0 2023-03-26 15:14:27,874 INFO [finetune.py:976] (3/7) Epoch 13, batch 0, loss[loss=0.204, simple_loss=0.2615, pruned_loss=0.07322, over 4889.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2615, pruned_loss=0.07322, over 4889.00 frames. ], batch size: 35, lr: 3.62e-03, grad_scale: 16.0 2023-03-26 15:14:27,874 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 15:14:30,656 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6257, 3.5121, 3.4474, 1.5908, 3.6208, 2.7520, 0.8542, 2.3545], device='cuda:3'), covar=tensor([0.2042, 0.1710, 0.1339, 0.3010, 0.1105, 0.0988, 0.3607, 0.1478], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0173, 0.0160, 0.0127, 0.0155, 0.0121, 0.0145, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 15:14:42,136 INFO [finetune.py:1010] (3/7) Epoch 13, validation: loss=0.1598, simple_loss=0.23, pruned_loss=0.04482, over 2265189.00 frames. 2023-03-26 15:14:42,136 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 15:14:47,272 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.040e+02 1.546e+02 1.915e+02 2.253e+02 4.332e+02, threshold=3.830e+02, percent-clipped=1.0 2023-03-26 15:14:49,810 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68743.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 15:14:49,817 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8011, 1.7676, 1.8270, 1.1316, 1.8771, 1.9270, 1.7976, 1.5517], device='cuda:3'), covar=tensor([0.0586, 0.0638, 0.0754, 0.1001, 0.0719, 0.0673, 0.0676, 0.1157], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0133, 0.0142, 0.0125, 0.0122, 0.0141, 0.0142, 0.0161], 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') 2023-03-26 15:14:52,122 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68746.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:14:56,349 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.4251, 1.5784, 1.5687, 0.7672, 1.6425, 1.8404, 1.8359, 1.4280], device='cuda:3'), covar=tensor([0.1098, 0.0695, 0.0471, 0.0616, 0.0412, 0.0605, 0.0322, 0.0722], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0155, 0.0123, 0.0132, 0.0132, 0.0127, 0.0145, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.3373e-05, 1.1311e-04, 8.8848e-05, 9.5292e-05, 9.3380e-05, 9.2446e-05, 1.0547e-04, 1.0665e-04], device='cuda:3') 2023-03-26 15:14:58,498 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68755.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:15:15,977 INFO [finetune.py:976] (3/7) Epoch 13, batch 50, loss[loss=0.2235, simple_loss=0.2793, pruned_loss=0.08382, over 4908.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2659, pruned_loss=0.0659, over 217210.32 frames. ], batch size: 43, lr: 3.62e-03, grad_scale: 16.0 2023-03-26 15:15:21,852 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=68791.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 15:15:45,214 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-26 15:15:57,666 INFO [finetune.py:976] (3/7) Epoch 13, batch 100, loss[loss=0.1901, simple_loss=0.2609, pruned_loss=0.05962, over 4919.00 frames. ], tot_loss[loss=0.1924, simple_loss=0.2582, pruned_loss=0.06327, over 382082.69 frames. ], batch size: 38, lr: 3.62e-03, grad_scale: 16.0 2023-03-26 15:16:02,753 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.134e+02 1.681e+02 1.901e+02 2.429e+02 4.753e+02, threshold=3.802e+02, percent-clipped=2.0 2023-03-26 15:16:31,426 INFO [finetune.py:976] (3/7) Epoch 13, batch 150, loss[loss=0.1933, simple_loss=0.2584, pruned_loss=0.06405, over 4856.00 frames. ], tot_loss[loss=0.1876, simple_loss=0.253, pruned_loss=0.06113, over 509979.44 frames. ], batch size: 31, lr: 3.62e-03, grad_scale: 16.0 2023-03-26 15:17:05,107 INFO [finetune.py:976] (3/7) Epoch 13, batch 200, loss[loss=0.2053, simple_loss=0.2632, pruned_loss=0.07372, over 4830.00 frames. ], tot_loss[loss=0.1855, simple_loss=0.2505, pruned_loss=0.06025, over 608366.15 frames. ], batch size: 33, lr: 3.62e-03, grad_scale: 16.0 2023-03-26 15:17:05,752 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68934.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:17:09,208 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.222e+02 1.603e+02 1.930e+02 2.189e+02 8.191e+02, threshold=3.861e+02, percent-clipped=2.0 2023-03-26 15:17:27,422 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-03-26 15:17:46,315 INFO [finetune.py:976] (3/7) Epoch 13, batch 250, loss[loss=0.1862, simple_loss=0.2664, pruned_loss=0.05301, over 4811.00 frames. ], tot_loss[loss=0.1885, simple_loss=0.254, pruned_loss=0.06146, over 685837.58 frames. ], batch size: 38, lr: 3.62e-03, grad_scale: 16.0 2023-03-26 15:17:57,591 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6168, 0.6437, 1.6192, 1.4960, 1.3871, 1.3206, 1.4202, 1.5100], device='cuda:3'), covar=tensor([0.3666, 0.4187, 0.3418, 0.3537, 0.4620, 0.3535, 0.4426, 0.3212], device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0237, 0.0254, 0.0260, 0.0257, 0.0232, 0.0275, 0.0233], 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') 2023-03-26 15:18:06,115 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6808, 0.6690, 1.6806, 1.5550, 1.4762, 1.4299, 1.4825, 1.5880], device='cuda:3'), covar=tensor([0.3885, 0.4463, 0.3558, 0.3584, 0.4687, 0.3519, 0.4544, 0.3589], device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0237, 0.0254, 0.0260, 0.0257, 0.0232, 0.0275, 0.0233], 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') 2023-03-26 15:18:19,710 INFO [finetune.py:976] (3/7) Epoch 13, batch 300, loss[loss=0.1852, simple_loss=0.2568, pruned_loss=0.05676, over 4714.00 frames. ], tot_loss[loss=0.1926, simple_loss=0.2592, pruned_loss=0.06302, over 744485.92 frames. ], batch size: 23, lr: 3.62e-03, grad_scale: 32.0 2023-03-26 15:18:23,316 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.215e+02 1.585e+02 1.877e+02 2.328e+02 4.201e+02, threshold=3.755e+02, percent-clipped=2.0 2023-03-26 15:18:24,575 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69041.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:18:25,873 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69043.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:18:34,516 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69055.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:18:55,352 INFO [finetune.py:976] (3/7) Epoch 13, batch 350, loss[loss=0.1965, simple_loss=0.2702, pruned_loss=0.06144, over 4795.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.2624, pruned_loss=0.06448, over 791529.90 frames. ], batch size: 45, lr: 3.62e-03, grad_scale: 32.0 2023-03-26 15:19:18,985 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=69103.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:19:19,637 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69104.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:19:41,448 INFO [finetune.py:976] (3/7) Epoch 13, batch 400, loss[loss=0.1601, simple_loss=0.2381, pruned_loss=0.04106, over 4799.00 frames. ], tot_loss[loss=0.1938, simple_loss=0.2612, pruned_loss=0.06326, over 827333.20 frames. ], batch size: 45, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:19:50,066 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.154e+02 1.689e+02 1.999e+02 2.345e+02 4.076e+02, threshold=3.998e+02, percent-clipped=3.0 2023-03-26 15:20:09,330 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-03-26 15:20:09,756 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69162.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:20:13,409 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69168.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:20:23,375 INFO [finetune.py:976] (3/7) Epoch 13, batch 450, loss[loss=0.2493, simple_loss=0.3048, pruned_loss=0.0969, over 4850.00 frames. ], tot_loss[loss=0.1923, simple_loss=0.2593, pruned_loss=0.06262, over 854675.28 frames. ], batch size: 44, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:20:29,466 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9204, 1.8124, 1.6051, 1.8992, 2.2949, 1.9663, 1.6390, 1.5064], device='cuda:3'), covar=tensor([0.2049, 0.2035, 0.1794, 0.1647, 0.1648, 0.1095, 0.2249, 0.1814], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0206, 0.0209, 0.0190, 0.0239, 0.0181, 0.0212, 0.0197], 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') 2023-03-26 15:20:30,148 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-26 15:21:04,202 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69223.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:21:05,425 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69225.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:21:07,851 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69229.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:21:10,198 INFO [finetune.py:976] (3/7) Epoch 13, batch 500, loss[loss=0.1835, simple_loss=0.2491, pruned_loss=0.05896, over 4681.00 frames. ], tot_loss[loss=0.1895, simple_loss=0.2565, pruned_loss=0.06128, over 876970.95 frames. ], batch size: 23, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:21:10,907 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69234.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:21:14,294 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.151e+02 1.659e+02 1.928e+02 2.205e+02 4.798e+02, threshold=3.855e+02, percent-clipped=1.0 2023-03-26 15:21:18,076 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69245.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 15:21:37,033 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69273.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:21:43,321 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=69282.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:21:43,877 INFO [finetune.py:976] (3/7) Epoch 13, batch 550, loss[loss=0.2208, simple_loss=0.2801, pruned_loss=0.08075, over 4822.00 frames. ], tot_loss[loss=0.1883, simple_loss=0.2547, pruned_loss=0.06097, over 895105.48 frames. ], batch size: 51, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:21:45,826 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69286.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:21:49,042 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-26 15:21:56,639 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9796, 1.4178, 2.0313, 1.8778, 1.6597, 1.6575, 1.8198, 1.7982], device='cuda:3'), covar=tensor([0.4236, 0.4215, 0.3332, 0.4081, 0.4961, 0.3921, 0.4892, 0.3426], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0238, 0.0255, 0.0261, 0.0258, 0.0233, 0.0275, 0.0233], 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') 2023-03-26 15:21:59,414 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69306.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 15:22:08,837 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69320.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:22:17,499 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.04 vs. limit=2.0 2023-03-26 15:22:17,557 INFO [finetune.py:976] (3/7) Epoch 13, batch 600, loss[loss=0.1882, simple_loss=0.2491, pruned_loss=0.06363, over 4912.00 frames. ], tot_loss[loss=0.189, simple_loss=0.255, pruned_loss=0.06151, over 907964.83 frames. ], batch size: 32, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:22:18,303 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69334.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 15:22:21,204 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.059e+02 1.536e+02 1.861e+02 2.296e+02 3.946e+02, threshold=3.721e+02, percent-clipped=1.0 2023-03-26 15:22:22,514 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69341.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:22:50,003 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69368.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:22:58,839 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69381.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:22:59,983 INFO [finetune.py:976] (3/7) Epoch 13, batch 650, loss[loss=0.1858, simple_loss=0.2616, pruned_loss=0.05498, over 4723.00 frames. ], tot_loss[loss=0.1921, simple_loss=0.2584, pruned_loss=0.0629, over 917680.28 frames. ], batch size: 59, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:23:03,691 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=69389.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:23:06,749 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69394.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:23:10,353 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69399.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:23:18,897 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7189, 1.6861, 1.4865, 1.8248, 2.1054, 1.8434, 1.4443, 1.4533], device='cuda:3'), covar=tensor([0.2512, 0.2174, 0.2100, 0.1794, 0.1980, 0.1311, 0.2650, 0.2117], device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0207, 0.0210, 0.0190, 0.0241, 0.0182, 0.0213, 0.0198], 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') 2023-03-26 15:23:30,650 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69429.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:23:33,419 INFO [finetune.py:976] (3/7) Epoch 13, batch 700, loss[loss=0.1599, simple_loss=0.1959, pruned_loss=0.06189, over 4060.00 frames. ], tot_loss[loss=0.1932, simple_loss=0.2598, pruned_loss=0.06328, over 924080.74 frames. ], batch size: 17, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:23:37,536 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.135e+02 1.702e+02 1.957e+02 2.425e+02 4.096e+02, threshold=3.913e+02, percent-clipped=2.0 2023-03-26 15:23:47,860 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69455.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:23:54,780 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6046, 1.4888, 1.4514, 1.5591, 1.0901, 3.4951, 1.2579, 1.8096], device='cuda:3'), covar=tensor([0.3318, 0.2493, 0.2191, 0.2346, 0.1876, 0.0166, 0.2618, 0.1308], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0120, 0.0123, 0.0115, 0.0098, 0.0098, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 15:23:58,910 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.1061, 4.3441, 4.6353, 4.9296, 4.8220, 4.5122, 5.2407, 1.6038], device='cuda:3'), covar=tensor([0.0736, 0.0864, 0.0746, 0.0768, 0.1266, 0.1670, 0.0594, 0.5605], device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0243, 0.0277, 0.0292, 0.0329, 0.0283, 0.0302, 0.0296], 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') 2023-03-26 15:24:06,514 INFO [finetune.py:976] (3/7) Epoch 13, batch 750, loss[loss=0.2516, simple_loss=0.3233, pruned_loss=0.08993, over 4911.00 frames. ], tot_loss[loss=0.1935, simple_loss=0.2607, pruned_loss=0.06321, over 931517.18 frames. ], batch size: 37, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:24:40,543 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69518.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:24:44,558 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69524.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:24:50,486 INFO [finetune.py:976] (3/7) Epoch 13, batch 800, loss[loss=0.1554, simple_loss=0.2105, pruned_loss=0.05017, over 4150.00 frames. ], tot_loss[loss=0.1933, simple_loss=0.2605, pruned_loss=0.06305, over 935793.33 frames. ], batch size: 17, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:24:57,837 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.091e+02 1.694e+02 1.982e+02 2.355e+02 4.334e+02, threshold=3.964e+02, percent-clipped=1.0 2023-03-26 15:25:47,594 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.5360, 3.8516, 4.1330, 4.3482, 4.2924, 3.9378, 4.6026, 1.3837], device='cuda:3'), covar=tensor([0.0726, 0.0881, 0.0781, 0.0951, 0.1079, 0.1770, 0.0655, 0.5606], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0242, 0.0276, 0.0291, 0.0327, 0.0280, 0.0301, 0.0294], 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') 2023-03-26 15:25:47,596 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69581.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:25:48,246 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8012, 1.7457, 1.8880, 1.2930, 1.8827, 1.8612, 1.8194, 1.5771], device='cuda:3'), covar=tensor([0.0563, 0.0674, 0.0652, 0.0847, 0.0718, 0.0708, 0.0613, 0.1081], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0131, 0.0140, 0.0122, 0.0121, 0.0140, 0.0140, 0.0159], 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') 2023-03-26 15:25:48,750 INFO [finetune.py:976] (3/7) Epoch 13, batch 850, loss[loss=0.2313, simple_loss=0.2732, pruned_loss=0.09469, over 4916.00 frames. ], tot_loss[loss=0.1907, simple_loss=0.2578, pruned_loss=0.06184, over 942737.64 frames. ], batch size: 43, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:26:03,115 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69601.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 15:26:10,635 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-26 15:26:21,349 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69629.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 15:26:24,227 INFO [finetune.py:976] (3/7) Epoch 13, batch 900, loss[loss=0.1613, simple_loss=0.2317, pruned_loss=0.04548, over 4868.00 frames. ], tot_loss[loss=0.1886, simple_loss=0.2553, pruned_loss=0.06091, over 946396.38 frames. ], batch size: 31, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:26:27,890 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.145e+02 1.604e+02 1.856e+02 2.224e+02 3.601e+02, threshold=3.711e+02, percent-clipped=0.0 2023-03-26 15:26:55,526 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69674.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 15:26:56,704 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69676.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:27:06,462 INFO [finetune.py:976] (3/7) Epoch 13, batch 950, loss[loss=0.1731, simple_loss=0.2438, pruned_loss=0.05117, over 4746.00 frames. ], tot_loss[loss=0.1861, simple_loss=0.2523, pruned_loss=0.05999, over 947917.03 frames. ], batch size: 54, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:27:21,772 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-03-26 15:27:29,365 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69699.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:28:02,964 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69724.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:28:08,356 INFO [finetune.py:976] (3/7) Epoch 13, batch 1000, loss[loss=0.175, simple_loss=0.2458, pruned_loss=0.0521, over 4847.00 frames. ], tot_loss[loss=0.1887, simple_loss=0.2553, pruned_loss=0.06109, over 951445.80 frames. ], batch size: 30, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:28:10,722 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69735.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 15:28:12,517 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4058, 1.0986, 0.7656, 1.3214, 1.7834, 0.6979, 1.2156, 1.3395], device='cuda:3'), covar=tensor([0.1494, 0.2041, 0.1898, 0.1181, 0.2100, 0.2075, 0.1393, 0.1938], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0096, 0.0113, 0.0092, 0.0121, 0.0094, 0.0099, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 15:28:12,999 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.062e+02 1.598e+02 1.856e+02 2.406e+02 4.029e+02, threshold=3.712e+02, percent-clipped=2.0 2023-03-26 15:28:13,729 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1441, 2.2475, 2.1535, 1.5837, 2.2695, 2.3300, 2.2164, 1.9230], device='cuda:3'), covar=tensor([0.0669, 0.0565, 0.0700, 0.0860, 0.0583, 0.0646, 0.0625, 0.0983], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0131, 0.0140, 0.0122, 0.0121, 0.0140, 0.0140, 0.0159], 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') 2023-03-26 15:28:18,441 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=69747.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:28:20,308 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69750.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:28:51,847 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1632, 2.1006, 1.6615, 2.0786, 2.1516, 1.8384, 2.4303, 2.1499], device='cuda:3'), covar=tensor([0.1346, 0.2168, 0.3066, 0.2563, 0.2482, 0.1690, 0.2929, 0.1849], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0187, 0.0233, 0.0254, 0.0242, 0.0199, 0.0212, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 15:28:52,947 INFO [finetune.py:976] (3/7) Epoch 13, batch 1050, loss[loss=0.1918, simple_loss=0.2659, pruned_loss=0.05882, over 4927.00 frames. ], tot_loss[loss=0.1917, simple_loss=0.2587, pruned_loss=0.06232, over 953038.10 frames. ], batch size: 38, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:29:38,099 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69818.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:29:48,094 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69824.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:29:59,169 INFO [finetune.py:976] (3/7) Epoch 13, batch 1100, loss[loss=0.207, simple_loss=0.275, pruned_loss=0.06953, over 4725.00 frames. ], tot_loss[loss=0.1916, simple_loss=0.2593, pruned_loss=0.06196, over 953330.44 frames. ], batch size: 59, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:30:02,888 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.063e+02 1.609e+02 1.898e+02 2.282e+02 6.010e+02, threshold=3.795e+02, percent-clipped=2.0 2023-03-26 15:30:12,686 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-26 15:30:35,886 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=69866.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:30:43,312 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=69872.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:30:51,897 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69881.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:30:53,488 INFO [finetune.py:976] (3/7) Epoch 13, batch 1150, loss[loss=0.194, simple_loss=0.2673, pruned_loss=0.0604, over 4738.00 frames. ], tot_loss[loss=0.191, simple_loss=0.2588, pruned_loss=0.0616, over 952223.38 frames. ], batch size: 27, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:30:58,474 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8528, 1.5575, 2.1383, 1.5415, 1.9439, 2.1237, 1.5894, 2.2766], device='cuda:3'), covar=tensor([0.1239, 0.1976, 0.1355, 0.1885, 0.0814, 0.1297, 0.2429, 0.0713], device='cuda:3'), in_proj_covar=tensor([0.0198, 0.0207, 0.0196, 0.0193, 0.0180, 0.0216, 0.0219, 0.0201], 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') 2023-03-26 15:31:12,297 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69901.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 15:31:22,042 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5580, 1.4783, 2.0172, 1.9025, 1.7863, 3.7138, 1.2785, 1.6915], device='cuda:3'), covar=tensor([0.1161, 0.2135, 0.1405, 0.1144, 0.1736, 0.0241, 0.1863, 0.2223], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0073, 0.0077, 0.0091, 0.0080, 0.0084, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 15:31:42,211 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=69929.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:31:42,260 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69929.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 15:31:44,610 INFO [finetune.py:976] (3/7) Epoch 13, batch 1200, loss[loss=0.201, simple_loss=0.2721, pruned_loss=0.06496, over 4745.00 frames. ], tot_loss[loss=0.1894, simple_loss=0.2573, pruned_loss=0.06074, over 953053.19 frames. ], batch size: 27, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:31:48,755 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.031e+02 1.603e+02 1.893e+02 2.321e+02 3.158e+02, threshold=3.786e+02, percent-clipped=0.0 2023-03-26 15:31:55,833 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=69949.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 15:32:13,211 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69976.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:32:13,759 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=69977.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:32:17,362 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5687, 1.6782, 1.3819, 1.6805, 2.0298, 1.9105, 1.6635, 1.4782], device='cuda:3'), covar=tensor([0.0317, 0.0269, 0.0603, 0.0287, 0.0192, 0.0336, 0.0284, 0.0360], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0109, 0.0140, 0.0113, 0.0101, 0.0105, 0.0095, 0.0109], device='cuda:3'), out_proj_covar=tensor([7.3165e-05, 8.4740e-05, 1.1092e-04, 8.8067e-05, 7.9069e-05, 7.8170e-05, 7.1828e-05, 8.3652e-05], device='cuda:3') 2023-03-26 15:32:17,838 INFO [finetune.py:976] (3/7) Epoch 13, batch 1250, loss[loss=0.184, simple_loss=0.2449, pruned_loss=0.06155, over 4915.00 frames. ], tot_loss[loss=0.1868, simple_loss=0.2545, pruned_loss=0.0596, over 952865.71 frames. ], batch size: 37, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:32:45,360 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.0561, 1.0326, 1.0096, 0.4197, 0.9006, 1.1493, 1.1932, 1.0184], device='cuda:3'), covar=tensor([0.0923, 0.0658, 0.0572, 0.0569, 0.0551, 0.0692, 0.0389, 0.0787], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0151, 0.0120, 0.0128, 0.0129, 0.0125, 0.0141, 0.0143], device='cuda:3'), out_proj_covar=tensor([9.1747e-05, 1.1010e-04, 8.6705e-05, 9.2450e-05, 9.1264e-05, 9.0504e-05, 1.0303e-04, 1.0419e-04], device='cuda:3') 2023-03-26 15:32:46,507 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=70024.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:32:46,533 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70024.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:32:47,251 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.71 vs. limit=5.0 2023-03-26 15:32:50,168 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70030.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 15:32:52,399 INFO [finetune.py:976] (3/7) Epoch 13, batch 1300, loss[loss=0.1524, simple_loss=0.2228, pruned_loss=0.04097, over 4105.00 frames. ], tot_loss[loss=0.1844, simple_loss=0.2517, pruned_loss=0.05851, over 953224.95 frames. ], batch size: 17, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:32:52,533 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4015, 2.5498, 2.6981, 1.3307, 2.9865, 3.1026, 2.6673, 2.4001], device='cuda:3'), covar=tensor([0.0853, 0.0742, 0.0403, 0.0606, 0.0519, 0.0540, 0.0420, 0.0533], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0150, 0.0120, 0.0128, 0.0129, 0.0125, 0.0141, 0.0143], device='cuda:3'), out_proj_covar=tensor([9.1692e-05, 1.1003e-04, 8.6630e-05, 9.2386e-05, 9.1150e-05, 9.0448e-05, 1.0294e-04, 1.0406e-04], device='cuda:3') 2023-03-26 15:32:56,055 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.095e+02 1.649e+02 1.897e+02 2.309e+02 4.234e+02, threshold=3.795e+02, percent-clipped=2.0 2023-03-26 15:33:03,827 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70050.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:33:06,696 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0265, 2.2501, 2.2378, 1.5780, 2.2098, 2.3001, 2.2543, 1.9102], device='cuda:3'), covar=tensor([0.0614, 0.0530, 0.0647, 0.0841, 0.0647, 0.0679, 0.0590, 0.1000], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0130, 0.0140, 0.0123, 0.0122, 0.0140, 0.0140, 0.0160], 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') 2023-03-26 15:33:10,857 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7330, 3.7874, 3.6697, 2.1569, 3.9816, 3.0404, 0.9092, 2.6673], device='cuda:3'), covar=tensor([0.2490, 0.2510, 0.1422, 0.2950, 0.1096, 0.0965, 0.4332, 0.1571], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0175, 0.0160, 0.0129, 0.0156, 0.0121, 0.0145, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 15:33:16,379 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-26 15:33:19,133 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=70072.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:33:25,839 INFO [finetune.py:976] (3/7) Epoch 13, batch 1350, loss[loss=0.1363, simple_loss=0.203, pruned_loss=0.03479, over 4744.00 frames. ], tot_loss[loss=0.1852, simple_loss=0.2522, pruned_loss=0.0591, over 953062.72 frames. ], batch size: 23, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:33:36,436 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=70098.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:33:53,161 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70110.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:34:08,077 INFO [finetune.py:976] (3/7) Epoch 13, batch 1400, loss[loss=0.1785, simple_loss=0.2613, pruned_loss=0.04782, over 4902.00 frames. ], tot_loss[loss=0.1888, simple_loss=0.2563, pruned_loss=0.06065, over 955166.96 frames. ], batch size: 37, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:34:12,153 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.202e+02 1.588e+02 1.939e+02 2.393e+02 8.943e+02, threshold=3.877e+02, percent-clipped=1.0 2023-03-26 15:34:34,203 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70171.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:34:41,771 INFO [finetune.py:976] (3/7) Epoch 13, batch 1450, loss[loss=0.1856, simple_loss=0.2551, pruned_loss=0.05804, over 4826.00 frames. ], tot_loss[loss=0.1903, simple_loss=0.2579, pruned_loss=0.06131, over 954924.75 frames. ], batch size: 33, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:35:26,431 INFO [finetune.py:976] (3/7) Epoch 13, batch 1500, loss[loss=0.1609, simple_loss=0.2316, pruned_loss=0.0451, over 4795.00 frames. ], tot_loss[loss=0.191, simple_loss=0.2586, pruned_loss=0.06172, over 954659.80 frames. ], batch size: 29, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:35:30,134 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.195e+02 1.613e+02 1.899e+02 2.364e+02 4.350e+02, threshold=3.798e+02, percent-clipped=1.0 2023-03-26 15:35:46,608 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-03-26 15:35:46,864 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70260.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:35:53,447 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9216, 1.6173, 2.2534, 1.5852, 2.0200, 2.1936, 1.6270, 2.2268], device='cuda:3'), covar=tensor([0.1229, 0.1840, 0.1365, 0.1746, 0.0805, 0.1229, 0.2409, 0.0774], device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0205, 0.0195, 0.0192, 0.0178, 0.0214, 0.0217, 0.0200], 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') 2023-03-26 15:36:10,535 INFO [finetune.py:976] (3/7) Epoch 13, batch 1550, loss[loss=0.1989, simple_loss=0.2625, pruned_loss=0.06764, over 4912.00 frames. ], tot_loss[loss=0.1923, simple_loss=0.2599, pruned_loss=0.06238, over 954044.60 frames. ], batch size: 46, lr: 3.61e-03, grad_scale: 32.0 2023-03-26 15:36:49,035 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.5626, 1.7230, 1.8944, 1.0768, 1.8797, 2.0665, 2.0271, 1.6370], device='cuda:3'), covar=tensor([0.0899, 0.0799, 0.0483, 0.0545, 0.0504, 0.0741, 0.0405, 0.0729], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0151, 0.0122, 0.0129, 0.0130, 0.0126, 0.0142, 0.0144], device='cuda:3'), out_proj_covar=tensor([9.2455e-05, 1.1064e-04, 8.7716e-05, 9.3188e-05, 9.1858e-05, 9.1104e-05, 1.0348e-04, 1.0456e-04], device='cuda:3') 2023-03-26 15:36:49,642 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70321.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:36:58,940 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70330.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 15:37:00,630 INFO [finetune.py:976] (3/7) Epoch 13, batch 1600, loss[loss=0.1826, simple_loss=0.2567, pruned_loss=0.05425, over 4889.00 frames. ], tot_loss[loss=0.1914, simple_loss=0.2585, pruned_loss=0.06214, over 955951.78 frames. ], batch size: 32, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:37:04,738 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.057e+02 1.529e+02 1.873e+02 2.318e+02 5.550e+02, threshold=3.745e+02, percent-clipped=4.0 2023-03-26 15:37:18,746 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-03-26 15:37:30,809 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=70378.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 15:37:34,182 INFO [finetune.py:976] (3/7) Epoch 13, batch 1650, loss[loss=0.2138, simple_loss=0.2728, pruned_loss=0.07738, over 4727.00 frames. ], tot_loss[loss=0.1891, simple_loss=0.2552, pruned_loss=0.06155, over 954026.38 frames. ], batch size: 54, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:37:52,249 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-26 15:38:08,088 INFO [finetune.py:976] (3/7) Epoch 13, batch 1700, loss[loss=0.2101, simple_loss=0.2633, pruned_loss=0.07842, over 4775.00 frames. ], tot_loss[loss=0.1873, simple_loss=0.2527, pruned_loss=0.06094, over 950980.71 frames. ], batch size: 26, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:38:10,637 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6248, 2.4552, 2.0160, 1.0701, 2.1864, 1.9726, 1.8529, 2.1120], device='cuda:3'), covar=tensor([0.0771, 0.0771, 0.1606, 0.1983, 0.1584, 0.2345, 0.2133, 0.1136], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0197, 0.0200, 0.0186, 0.0215, 0.0208, 0.0224, 0.0197], 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') 2023-03-26 15:38:11,732 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.265e+02 1.610e+02 1.926e+02 2.276e+02 4.227e+02, threshold=3.852e+02, percent-clipped=1.0 2023-03-26 15:38:30,203 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70466.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:38:41,460 INFO [finetune.py:976] (3/7) Epoch 13, batch 1750, loss[loss=0.1414, simple_loss=0.2132, pruned_loss=0.03482, over 4724.00 frames. ], tot_loss[loss=0.1884, simple_loss=0.2544, pruned_loss=0.0612, over 950726.72 frames. ], batch size: 23, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:39:24,240 INFO [finetune.py:976] (3/7) Epoch 13, batch 1800, loss[loss=0.1801, simple_loss=0.2429, pruned_loss=0.05868, over 4885.00 frames. ], tot_loss[loss=0.191, simple_loss=0.2578, pruned_loss=0.06206, over 954196.92 frames. ], batch size: 32, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:39:28,350 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.316e+01 1.597e+02 2.051e+02 2.548e+02 3.844e+02, threshold=4.101e+02, percent-clipped=0.0 2023-03-26 15:39:47,466 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2653, 2.1620, 1.6799, 2.2104, 2.1524, 1.8677, 2.5109, 2.2684], device='cuda:3'), covar=tensor([0.1448, 0.2376, 0.3222, 0.2892, 0.2680, 0.1826, 0.3212, 0.1938], device='cuda:3'), in_proj_covar=tensor([0.0178, 0.0186, 0.0231, 0.0253, 0.0242, 0.0199, 0.0212, 0.0196], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 15:39:49,314 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6598, 0.7160, 1.7122, 1.5321, 1.4507, 1.3439, 1.4854, 1.5774], device='cuda:3'), covar=tensor([0.2837, 0.3253, 0.2720, 0.2909, 0.3633, 0.3067, 0.3331, 0.2574], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0239, 0.0256, 0.0263, 0.0260, 0.0234, 0.0277, 0.0234], 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') 2023-03-26 15:39:58,059 INFO [finetune.py:976] (3/7) Epoch 13, batch 1850, loss[loss=0.1545, simple_loss=0.2379, pruned_loss=0.03557, over 4897.00 frames. ], tot_loss[loss=0.1917, simple_loss=0.2594, pruned_loss=0.06194, over 955772.68 frames. ], batch size: 43, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:40:07,841 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7989, 1.7447, 1.6524, 1.8850, 1.2864, 3.8905, 1.5531, 2.1822], device='cuda:3'), covar=tensor([0.3451, 0.2416, 0.2100, 0.2346, 0.1840, 0.0209, 0.2387, 0.1205], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0116, 0.0120, 0.0124, 0.0115, 0.0098, 0.0098, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 15:40:26,915 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70616.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:40:42,679 INFO [finetune.py:976] (3/7) Epoch 13, batch 1900, loss[loss=0.2054, simple_loss=0.2693, pruned_loss=0.07073, over 4876.00 frames. ], tot_loss[loss=0.1913, simple_loss=0.2598, pruned_loss=0.06142, over 957098.85 frames. ], batch size: 35, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:40:46,779 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.033e+02 1.570e+02 1.884e+02 2.217e+02 6.026e+02, threshold=3.769e+02, percent-clipped=2.0 2023-03-26 15:41:27,336 INFO [finetune.py:976] (3/7) Epoch 13, batch 1950, loss[loss=0.2009, simple_loss=0.2362, pruned_loss=0.08278, over 4166.00 frames. ], tot_loss[loss=0.1909, simple_loss=0.259, pruned_loss=0.06138, over 955891.37 frames. ], batch size: 18, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:41:34,552 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3335, 1.2147, 1.1947, 1.3415, 1.5741, 1.3813, 1.2935, 1.1405], device='cuda:3'), covar=tensor([0.0293, 0.0289, 0.0519, 0.0245, 0.0192, 0.0439, 0.0294, 0.0409], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0109, 0.0140, 0.0113, 0.0101, 0.0105, 0.0095, 0.0109], device='cuda:3'), out_proj_covar=tensor([7.2937e-05, 8.4819e-05, 1.1104e-04, 8.8209e-05, 7.8768e-05, 7.8023e-05, 7.2183e-05, 8.4054e-05], device='cuda:3') 2023-03-26 15:41:44,814 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6952, 1.5270, 1.3168, 1.1735, 1.7656, 1.4702, 2.1025, 1.7171], device='cuda:3'), covar=tensor([0.1551, 0.2119, 0.3522, 0.2732, 0.2729, 0.1849, 0.2233, 0.2021], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0187, 0.0233, 0.0254, 0.0243, 0.0200, 0.0213, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 15:42:06,901 INFO [finetune.py:976] (3/7) Epoch 13, batch 2000, loss[loss=0.1683, simple_loss=0.2317, pruned_loss=0.05246, over 4928.00 frames. ], tot_loss[loss=0.1878, simple_loss=0.2553, pruned_loss=0.06011, over 956228.23 frames. ], batch size: 33, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:42:15,810 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.177e+02 1.535e+02 1.807e+02 2.194e+02 3.140e+02, threshold=3.615e+02, percent-clipped=0.0 2023-03-26 15:42:36,901 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70766.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:42:48,489 INFO [finetune.py:976] (3/7) Epoch 13, batch 2050, loss[loss=0.1341, simple_loss=0.2164, pruned_loss=0.02593, over 4757.00 frames. ], tot_loss[loss=0.185, simple_loss=0.2523, pruned_loss=0.05883, over 956619.59 frames. ], batch size: 27, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:43:09,365 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=70814.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:43:11,809 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0994, 2.1883, 2.1785, 1.5970, 2.2043, 2.3560, 2.1717, 1.8492], device='cuda:3'), covar=tensor([0.0667, 0.0526, 0.0682, 0.0895, 0.0606, 0.0664, 0.0619, 0.1028], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0131, 0.0140, 0.0123, 0.0122, 0.0140, 0.0140, 0.0159], 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') 2023-03-26 15:43:22,315 INFO [finetune.py:976] (3/7) Epoch 13, batch 2100, loss[loss=0.247, simple_loss=0.3111, pruned_loss=0.09148, over 4909.00 frames. ], tot_loss[loss=0.1854, simple_loss=0.2523, pruned_loss=0.05927, over 956627.72 frames. ], batch size: 43, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:43:26,461 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.827e+01 1.609e+02 1.892e+02 2.240e+02 3.187e+02, threshold=3.783e+02, percent-clipped=0.0 2023-03-26 15:43:56,100 INFO [finetune.py:976] (3/7) Epoch 13, batch 2150, loss[loss=0.2059, simple_loss=0.2902, pruned_loss=0.06081, over 4807.00 frames. ], tot_loss[loss=0.1902, simple_loss=0.2576, pruned_loss=0.06145, over 956874.36 frames. ], batch size: 41, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:44:25,463 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5496, 2.2559, 1.8681, 0.9713, 2.1168, 1.9874, 1.8340, 1.9828], device='cuda:3'), covar=tensor([0.0835, 0.0855, 0.1654, 0.2080, 0.1367, 0.2077, 0.2096, 0.1034], device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0195, 0.0197, 0.0184, 0.0211, 0.0205, 0.0221, 0.0194], 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') 2023-03-26 15:44:33,244 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2008, 1.9619, 1.5321, 0.6668, 1.7578, 1.8231, 1.6489, 1.8169], device='cuda:3'), covar=tensor([0.0793, 0.0735, 0.1260, 0.1878, 0.1156, 0.1931, 0.2116, 0.0826], device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0195, 0.0197, 0.0184, 0.0211, 0.0205, 0.0221, 0.0194], 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') 2023-03-26 15:44:35,061 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70916.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:44:46,789 INFO [finetune.py:976] (3/7) Epoch 13, batch 2200, loss[loss=0.209, simple_loss=0.2703, pruned_loss=0.07384, over 4812.00 frames. ], tot_loss[loss=0.1936, simple_loss=0.2612, pruned_loss=0.06295, over 957342.98 frames. ], batch size: 38, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:44:50,485 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.006e+02 1.701e+02 1.958e+02 2.316e+02 4.574e+02, threshold=3.916e+02, percent-clipped=1.0 2023-03-26 15:44:56,338 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.4442, 3.8753, 4.0183, 4.2691, 4.1722, 3.9213, 4.4929, 1.3562], device='cuda:3'), covar=tensor([0.0671, 0.0775, 0.0815, 0.0857, 0.1125, 0.1594, 0.0664, 0.5432], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0244, 0.0276, 0.0292, 0.0331, 0.0283, 0.0302, 0.0296], 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') 2023-03-26 15:45:07,668 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=70964.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:45:19,186 INFO [finetune.py:976] (3/7) Epoch 13, batch 2250, loss[loss=0.1991, simple_loss=0.2608, pruned_loss=0.06875, over 4772.00 frames. ], tot_loss[loss=0.1933, simple_loss=0.2611, pruned_loss=0.06274, over 956532.34 frames. ], batch size: 51, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:45:26,620 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70992.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:45:56,573 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0445, 1.2787, 0.6404, 1.9582, 2.4151, 1.8207, 1.5439, 1.7875], device='cuda:3'), covar=tensor([0.1286, 0.2079, 0.2259, 0.1122, 0.1755, 0.1859, 0.1426, 0.1900], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0095, 0.0112, 0.0092, 0.0120, 0.0094, 0.0099, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 15:46:02,085 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71030.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:46:03,002 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-03-26 15:46:03,710 INFO [finetune.py:976] (3/7) Epoch 13, batch 2300, loss[loss=0.1843, simple_loss=0.2465, pruned_loss=0.06101, over 4002.00 frames. ], tot_loss[loss=0.1927, simple_loss=0.2604, pruned_loss=0.06245, over 954248.07 frames. ], batch size: 17, lr: 3.60e-03, grad_scale: 64.0 2023-03-26 15:46:08,249 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.237e+01 1.685e+02 2.000e+02 2.324e+02 3.629e+02, threshold=3.999e+02, percent-clipped=0.0 2023-03-26 15:46:23,765 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71053.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:46:57,906 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6991, 0.7157, 1.6141, 1.5162, 1.4884, 1.4126, 1.4457, 1.6053], device='cuda:3'), covar=tensor([0.3975, 0.4350, 0.4577, 0.3869, 0.5524, 0.4028, 0.4914, 0.3907], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0239, 0.0256, 0.0263, 0.0261, 0.0235, 0.0276, 0.0234], 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') 2023-03-26 15:46:59,591 INFO [finetune.py:976] (3/7) Epoch 13, batch 2350, loss[loss=0.1687, simple_loss=0.2435, pruned_loss=0.04694, over 4749.00 frames. ], tot_loss[loss=0.1916, simple_loss=0.2588, pruned_loss=0.06219, over 952517.19 frames. ], batch size: 27, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:47:10,256 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71091.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:47:29,266 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3738, 2.1595, 1.7079, 0.8602, 1.9786, 1.9207, 1.7033, 1.9249], device='cuda:3'), covar=tensor([0.0785, 0.0757, 0.1457, 0.1863, 0.1233, 0.2298, 0.2176, 0.0929], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0196, 0.0198, 0.0185, 0.0212, 0.0206, 0.0222, 0.0195], 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') 2023-03-26 15:47:40,155 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6840, 1.2667, 0.9284, 1.5611, 2.0961, 1.2298, 1.5124, 1.6187], device='cuda:3'), covar=tensor([0.1435, 0.2058, 0.1998, 0.1204, 0.1914, 0.2000, 0.1306, 0.1860], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0095, 0.0112, 0.0092, 0.0120, 0.0094, 0.0099, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 15:47:48,274 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-26 15:48:00,809 INFO [finetune.py:976] (3/7) Epoch 13, batch 2400, loss[loss=0.1302, simple_loss=0.1989, pruned_loss=0.03076, over 4934.00 frames. ], tot_loss[loss=0.1883, simple_loss=0.2551, pruned_loss=0.06079, over 954473.55 frames. ], batch size: 33, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:48:09,285 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.076e+01 1.502e+02 1.791e+02 2.104e+02 3.987e+02, threshold=3.583e+02, percent-clipped=0.0 2023-03-26 15:49:05,622 INFO [finetune.py:976] (3/7) Epoch 13, batch 2450, loss[loss=0.1667, simple_loss=0.2404, pruned_loss=0.04655, over 4938.00 frames. ], tot_loss[loss=0.1871, simple_loss=0.2534, pruned_loss=0.0604, over 956549.43 frames. ], batch size: 38, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:49:45,804 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4294, 1.4056, 1.8624, 2.8431, 1.9448, 2.1011, 1.0834, 2.2956], device='cuda:3'), covar=tensor([0.1756, 0.1455, 0.1173, 0.0539, 0.0871, 0.1575, 0.1586, 0.0571], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0116, 0.0134, 0.0164, 0.0101, 0.0138, 0.0127, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 15:49:58,571 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6905, 3.3668, 3.1356, 1.3143, 3.4872, 2.6449, 0.8254, 2.1531], device='cuda:3'), covar=tensor([0.2657, 0.2226, 0.1787, 0.3597, 0.1192, 0.1018, 0.4065, 0.1720], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0172, 0.0158, 0.0127, 0.0155, 0.0120, 0.0144, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 15:50:04,539 INFO [finetune.py:976] (3/7) Epoch 13, batch 2500, loss[loss=0.1885, simple_loss=0.2642, pruned_loss=0.05645, over 4792.00 frames. ], tot_loss[loss=0.1893, simple_loss=0.2558, pruned_loss=0.06144, over 955997.62 frames. ], batch size: 51, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:50:08,818 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.134e+02 1.629e+02 1.890e+02 2.415e+02 4.682e+02, threshold=3.780e+02, percent-clipped=4.0 2023-03-26 15:50:27,225 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5081, 1.3928, 1.8482, 1.2406, 1.6458, 1.7972, 1.3380, 2.0232], device='cuda:3'), covar=tensor([0.1263, 0.2327, 0.1264, 0.1856, 0.0791, 0.1202, 0.2922, 0.0722], device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0206, 0.0195, 0.0192, 0.0179, 0.0215, 0.0218, 0.0201], 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') 2023-03-26 15:50:41,417 INFO [finetune.py:976] (3/7) Epoch 13, batch 2550, loss[loss=0.2048, simple_loss=0.2831, pruned_loss=0.06323, over 4809.00 frames. ], tot_loss[loss=0.1915, simple_loss=0.2587, pruned_loss=0.06218, over 955464.29 frames. ], batch size: 45, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:51:02,507 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2899, 2.1287, 1.6652, 0.8300, 1.8640, 1.8609, 1.6724, 1.8754], device='cuda:3'), covar=tensor([0.0749, 0.0693, 0.1453, 0.1784, 0.1368, 0.2048, 0.1940, 0.0827], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0195, 0.0199, 0.0185, 0.0213, 0.0206, 0.0222, 0.0194], 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') 2023-03-26 15:51:11,987 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.12 vs. limit=5.0 2023-03-26 15:51:13,784 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8796, 1.6696, 1.5291, 1.2345, 1.6785, 1.6893, 1.6459, 2.2071], device='cuda:3'), covar=tensor([0.4337, 0.4309, 0.3499, 0.4327, 0.3962, 0.2561, 0.3905, 0.1916], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0259, 0.0223, 0.0277, 0.0245, 0.0211, 0.0246, 0.0220], 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') 2023-03-26 15:51:16,164 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8851, 4.2050, 4.0306, 2.2391, 4.2898, 3.1525, 0.6370, 2.8117], device='cuda:3'), covar=tensor([0.2664, 0.1745, 0.1284, 0.3012, 0.0737, 0.0985, 0.4899, 0.1549], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0174, 0.0159, 0.0128, 0.0156, 0.0122, 0.0145, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 15:51:22,574 INFO [finetune.py:976] (3/7) Epoch 13, batch 2600, loss[loss=0.1958, simple_loss=0.2639, pruned_loss=0.0639, over 4910.00 frames. ], tot_loss[loss=0.1918, simple_loss=0.2592, pruned_loss=0.06223, over 955567.06 frames. ], batch size: 37, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:51:26,871 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.157e+02 1.678e+02 1.922e+02 2.428e+02 5.321e+02, threshold=3.843e+02, percent-clipped=3.0 2023-03-26 15:51:31,779 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71348.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:51:55,370 INFO [finetune.py:976] (3/7) Epoch 13, batch 2650, loss[loss=0.1617, simple_loss=0.2392, pruned_loss=0.04212, over 4848.00 frames. ], tot_loss[loss=0.1922, simple_loss=0.2598, pruned_loss=0.06233, over 953978.23 frames. ], batch size: 49, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:51:58,315 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71386.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:52:29,325 INFO [finetune.py:976] (3/7) Epoch 13, batch 2700, loss[loss=0.1605, simple_loss=0.2332, pruned_loss=0.0439, over 4748.00 frames. ], tot_loss[loss=0.1911, simple_loss=0.2588, pruned_loss=0.06168, over 954468.00 frames. ], batch size: 27, lr: 3.60e-03, grad_scale: 32.0 2023-03-26 15:52:34,537 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.057e+02 1.578e+02 1.884e+02 2.307e+02 4.300e+02, threshold=3.769e+02, percent-clipped=2.0 2023-03-26 15:53:02,923 INFO [finetune.py:976] (3/7) Epoch 13, batch 2750, loss[loss=0.2079, simple_loss=0.2494, pruned_loss=0.08322, over 4063.00 frames. ], tot_loss[loss=0.1884, simple_loss=0.2556, pruned_loss=0.06061, over 954285.14 frames. ], batch size: 17, lr: 3.59e-03, grad_scale: 32.0 2023-03-26 15:53:36,642 INFO [finetune.py:976] (3/7) Epoch 13, batch 2800, loss[loss=0.1701, simple_loss=0.23, pruned_loss=0.05512, over 4766.00 frames. ], tot_loss[loss=0.186, simple_loss=0.2522, pruned_loss=0.05994, over 954688.94 frames. ], batch size: 59, lr: 3.59e-03, grad_scale: 32.0 2023-03-26 15:53:38,539 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4145, 1.4124, 1.4893, 1.6488, 1.4967, 3.0518, 1.2824, 1.4690], device='cuda:3'), covar=tensor([0.1031, 0.1897, 0.1194, 0.1008, 0.1869, 0.0311, 0.1631, 0.1878], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0082, 0.0074, 0.0078, 0.0092, 0.0081, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 15:53:40,882 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.138e+02 1.564e+02 1.863e+02 2.304e+02 3.302e+02, threshold=3.726e+02, percent-clipped=0.0 2023-03-26 15:54:05,576 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.80 vs. limit=5.0 2023-03-26 15:54:08,214 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0190, 1.8686, 1.6043, 1.8408, 1.8209, 1.7928, 1.8439, 2.5770], device='cuda:3'), covar=tensor([0.4001, 0.3982, 0.3530, 0.3981, 0.4376, 0.2566, 0.3901, 0.1704], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0259, 0.0223, 0.0277, 0.0245, 0.0212, 0.0246, 0.0221], 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') 2023-03-26 15:54:23,059 INFO [finetune.py:976] (3/7) Epoch 13, batch 2850, loss[loss=0.2459, simple_loss=0.2969, pruned_loss=0.09739, over 4898.00 frames. ], tot_loss[loss=0.1855, simple_loss=0.2516, pruned_loss=0.05974, over 953774.86 frames. ], batch size: 32, lr: 3.59e-03, grad_scale: 32.0 2023-03-26 15:54:52,334 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71616.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:55:06,324 INFO [finetune.py:976] (3/7) Epoch 13, batch 2900, loss[loss=0.2221, simple_loss=0.2873, pruned_loss=0.07848, over 4822.00 frames. ], tot_loss[loss=0.1881, simple_loss=0.2546, pruned_loss=0.06085, over 954421.30 frames. ], batch size: 30, lr: 3.59e-03, grad_scale: 32.0 2023-03-26 15:55:15,494 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.635e+01 1.661e+02 1.944e+02 2.530e+02 6.475e+02, threshold=3.888e+02, percent-clipped=5.0 2023-03-26 15:55:23,276 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-26 15:55:24,624 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71648.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:55:40,324 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6616, 2.5067, 2.0644, 0.9930, 2.2432, 1.9512, 1.8566, 2.1392], device='cuda:3'), covar=tensor([0.0721, 0.0716, 0.1604, 0.2122, 0.1511, 0.2416, 0.2186, 0.1005], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0197, 0.0200, 0.0186, 0.0214, 0.0208, 0.0224, 0.0196], 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') 2023-03-26 15:55:49,605 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71677.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:55:58,581 INFO [finetune.py:976] (3/7) Epoch 13, batch 2950, loss[loss=0.207, simple_loss=0.2782, pruned_loss=0.06789, over 4814.00 frames. ], tot_loss[loss=0.1911, simple_loss=0.2583, pruned_loss=0.06198, over 955770.87 frames. ], batch size: 40, lr: 3.59e-03, grad_scale: 32.0 2023-03-26 15:56:00,470 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71686.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:56:11,125 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=71696.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:56:39,790 INFO [finetune.py:976] (3/7) Epoch 13, batch 3000, loss[loss=0.1562, simple_loss=0.2184, pruned_loss=0.047, over 4396.00 frames. ], tot_loss[loss=0.1927, simple_loss=0.2601, pruned_loss=0.06261, over 956590.57 frames. ], batch size: 19, lr: 3.59e-03, grad_scale: 32.0 2023-03-26 15:56:39,790 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 15:56:41,986 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1832, 2.0635, 1.5158, 0.6428, 1.7541, 1.8510, 1.7152, 1.9215], device='cuda:3'), covar=tensor([0.1014, 0.0711, 0.1430, 0.1884, 0.1453, 0.2441, 0.2328, 0.0775], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0197, 0.0200, 0.0186, 0.0214, 0.0208, 0.0223, 0.0196], 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') 2023-03-26 15:56:50,410 INFO [finetune.py:1010] (3/7) Epoch 13, validation: loss=0.1572, simple_loss=0.2278, pruned_loss=0.04333, over 2265189.00 frames. 2023-03-26 15:56:50,410 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6324MB 2023-03-26 15:56:51,091 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=71734.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:56:51,766 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4753, 1.0493, 0.7880, 1.3211, 1.9574, 0.7608, 1.2626, 1.4738], device='cuda:3'), covar=tensor([0.1557, 0.2161, 0.1851, 0.1283, 0.1942, 0.1951, 0.1481, 0.1807], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0096, 0.0113, 0.0093, 0.0121, 0.0095, 0.0099, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 15:56:55,668 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.204e+02 1.624e+02 1.953e+02 2.376e+02 4.887e+02, threshold=3.907e+02, percent-clipped=1.0 2023-03-26 15:57:19,174 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.2865, 2.9556, 3.0589, 3.2106, 3.0488, 2.8646, 3.3385, 0.9750], device='cuda:3'), covar=tensor([0.1134, 0.0957, 0.1090, 0.1268, 0.1739, 0.2034, 0.1171, 0.5223], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0243, 0.0276, 0.0291, 0.0331, 0.0282, 0.0302, 0.0297], 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') 2023-03-26 15:57:22,728 INFO [finetune.py:976] (3/7) Epoch 13, batch 3050, loss[loss=0.1898, simple_loss=0.2478, pruned_loss=0.06588, over 4810.00 frames. ], tot_loss[loss=0.1941, simple_loss=0.2619, pruned_loss=0.06312, over 956473.70 frames. ], batch size: 41, lr: 3.59e-03, grad_scale: 32.0 2023-03-26 15:57:28,432 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.7327, 4.1366, 4.3561, 4.4860, 4.5272, 4.2640, 4.7964, 1.9389], device='cuda:3'), covar=tensor([0.0753, 0.0739, 0.0667, 0.0979, 0.1177, 0.1298, 0.0711, 0.4484], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0243, 0.0275, 0.0291, 0.0330, 0.0282, 0.0301, 0.0296], 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') 2023-03-26 15:57:46,573 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.6474, 1.3951, 1.4718, 0.8110, 1.5621, 1.7938, 1.6433, 1.3681], device='cuda:3'), covar=tensor([0.1026, 0.1032, 0.0562, 0.0634, 0.0530, 0.0556, 0.0445, 0.0776], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0152, 0.0122, 0.0129, 0.0130, 0.0127, 0.0142, 0.0145], device='cuda:3'), out_proj_covar=tensor([9.3139e-05, 1.1116e-04, 8.7712e-05, 9.2906e-05, 9.2521e-05, 9.2126e-05, 1.0366e-04, 1.0552e-04], device='cuda:3') 2023-03-26 15:57:52,608 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4540, 1.3687, 1.4017, 1.3813, 0.7927, 2.2083, 0.7099, 1.2666], device='cuda:3'), covar=tensor([0.3287, 0.2493, 0.2172, 0.2501, 0.2143, 0.0361, 0.2812, 0.1311], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0116, 0.0120, 0.0124, 0.0116, 0.0098, 0.0098, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 15:57:54,436 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8480, 1.7943, 1.4765, 1.7576, 2.1003, 1.9572, 1.7493, 1.5038], device='cuda:3'), covar=tensor([0.0267, 0.0251, 0.0531, 0.0261, 0.0195, 0.0320, 0.0289, 0.0307], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0109, 0.0140, 0.0112, 0.0101, 0.0105, 0.0095, 0.0109], device='cuda:3'), out_proj_covar=tensor([7.2703e-05, 8.4428e-05, 1.1094e-04, 8.7142e-05, 7.9039e-05, 7.7940e-05, 7.1768e-05, 8.3334e-05], device='cuda:3') 2023-03-26 15:57:55,478 INFO [finetune.py:976] (3/7) Epoch 13, batch 3100, loss[loss=0.2124, simple_loss=0.2757, pruned_loss=0.07462, over 4751.00 frames. ], tot_loss[loss=0.1929, simple_loss=0.2602, pruned_loss=0.06281, over 955911.08 frames. ], batch size: 27, lr: 3.59e-03, grad_scale: 32.0 2023-03-26 15:58:01,085 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.937e+01 1.560e+02 1.843e+02 2.215e+02 5.565e+02, threshold=3.687e+02, percent-clipped=1.0 2023-03-26 15:58:02,312 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8443, 1.7639, 1.5725, 1.9456, 2.4399, 1.9911, 1.6449, 1.4592], device='cuda:3'), covar=tensor([0.2153, 0.2089, 0.2035, 0.1661, 0.1614, 0.1129, 0.2359, 0.1968], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0207, 0.0210, 0.0190, 0.0240, 0.0183, 0.0214, 0.0198], 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') 2023-03-26 15:58:29,154 INFO [finetune.py:976] (3/7) Epoch 13, batch 3150, loss[loss=0.2142, simple_loss=0.2836, pruned_loss=0.0724, over 4821.00 frames. ], tot_loss[loss=0.1897, simple_loss=0.2568, pruned_loss=0.0613, over 954892.82 frames. ], batch size: 55, lr: 3.59e-03, grad_scale: 32.0 2023-03-26 15:58:29,235 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.3005, 3.7105, 3.9575, 4.1136, 4.0601, 3.7654, 4.3775, 1.3788], device='cuda:3'), covar=tensor([0.0746, 0.0784, 0.0811, 0.0988, 0.1212, 0.1448, 0.0672, 0.5551], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0243, 0.0275, 0.0290, 0.0330, 0.0281, 0.0302, 0.0297], 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') 2023-03-26 15:58:53,751 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71919.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:59:03,051 INFO [finetune.py:976] (3/7) Epoch 13, batch 3200, loss[loss=0.1779, simple_loss=0.245, pruned_loss=0.05545, over 4781.00 frames. ], tot_loss[loss=0.1863, simple_loss=0.2532, pruned_loss=0.05974, over 953469.91 frames. ], batch size: 26, lr: 3.59e-03, grad_scale: 32.0 2023-03-26 15:59:07,313 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.138e+02 1.561e+02 1.912e+02 2.265e+02 3.518e+02, threshold=3.824e+02, percent-clipped=0.0 2023-03-26 15:59:31,863 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-26 15:59:40,773 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71972.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:59:46,992 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.3273, 2.9281, 3.0695, 3.2733, 3.0786, 2.9350, 3.3612, 0.9351], device='cuda:3'), covar=tensor([0.1117, 0.1011, 0.1058, 0.1082, 0.1743, 0.1677, 0.1128, 0.5816], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0243, 0.0276, 0.0290, 0.0330, 0.0282, 0.0302, 0.0297], 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') 2023-03-26 15:59:50,148 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-03-26 15:59:54,381 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71980.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 15:59:56,075 INFO [finetune.py:976] (3/7) Epoch 13, batch 3250, loss[loss=0.2108, simple_loss=0.2692, pruned_loss=0.07615, over 4902.00 frames. ], tot_loss[loss=0.1851, simple_loss=0.252, pruned_loss=0.05911, over 954305.52 frames. ], batch size: 35, lr: 3.59e-03, grad_scale: 32.0 2023-03-26 16:00:04,179 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0651, 2.0056, 1.4779, 1.8990, 2.0752, 1.7259, 2.7228, 2.0200], device='cuda:3'), covar=tensor([0.1578, 0.2378, 0.3980, 0.3338, 0.3047, 0.1999, 0.2767, 0.2237], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0187, 0.0234, 0.0254, 0.0244, 0.0199, 0.0213, 0.0197], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 16:00:39,651 INFO [finetune.py:976] (3/7) Epoch 13, batch 3300, loss[loss=0.1921, simple_loss=0.2706, pruned_loss=0.05679, over 4863.00 frames. ], tot_loss[loss=0.1869, simple_loss=0.2543, pruned_loss=0.05976, over 954609.78 frames. ], batch size: 34, lr: 3.59e-03, grad_scale: 16.0 2023-03-26 16:00:44,477 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.095e+02 1.593e+02 1.995e+02 2.341e+02 5.205e+02, threshold=3.991e+02, percent-clipped=4.0 2023-03-26 16:01:29,182 INFO [finetune.py:976] (3/7) Epoch 13, batch 3350, loss[loss=0.2385, simple_loss=0.294, pruned_loss=0.09147, over 4138.00 frames. ], tot_loss[loss=0.19, simple_loss=0.2574, pruned_loss=0.06131, over 951675.79 frames. ], batch size: 65, lr: 3.59e-03, grad_scale: 16.0 2023-03-26 16:01:45,266 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6265, 2.3531, 1.9369, 2.7247, 2.4646, 2.0500, 3.0460, 2.4341], device='cuda:3'), covar=tensor([0.1291, 0.2659, 0.3403, 0.2645, 0.2779, 0.1784, 0.3053, 0.1918], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0187, 0.0234, 0.0255, 0.0245, 0.0199, 0.0214, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 16:01:58,848 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2023-03-26 16:01:59,480 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 2023-03-26 16:02:07,932 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72129.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:02:11,038 INFO [finetune.py:976] (3/7) Epoch 13, batch 3400, loss[loss=0.2196, simple_loss=0.2805, pruned_loss=0.07932, over 4827.00 frames. ], tot_loss[loss=0.1915, simple_loss=0.2587, pruned_loss=0.0622, over 951164.93 frames. ], batch size: 49, lr: 3.59e-03, grad_scale: 16.0 2023-03-26 16:02:16,765 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.109e+02 1.708e+02 2.008e+02 2.371e+02 4.954e+02, threshold=4.015e+02, percent-clipped=4.0 2023-03-26 16:02:49,867 INFO [finetune.py:976] (3/7) Epoch 13, batch 3450, loss[loss=0.2352, simple_loss=0.2951, pruned_loss=0.08766, over 4894.00 frames. ], tot_loss[loss=0.1915, simple_loss=0.2588, pruned_loss=0.06209, over 952976.63 frames. ], batch size: 35, lr: 3.59e-03, grad_scale: 16.0 2023-03-26 16:02:53,948 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1351, 2.1835, 2.3892, 1.5873, 2.2560, 2.3656, 2.2506, 1.9825], device='cuda:3'), covar=tensor([0.0565, 0.0572, 0.0561, 0.0826, 0.0589, 0.0621, 0.0568, 0.0964], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0133, 0.0141, 0.0123, 0.0124, 0.0142, 0.0141, 0.0161], 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') 2023-03-26 16:02:55,201 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72190.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:03:06,811 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-26 16:03:23,344 INFO [finetune.py:976] (3/7) Epoch 13, batch 3500, loss[loss=0.1593, simple_loss=0.2307, pruned_loss=0.04396, over 4781.00 frames. ], tot_loss[loss=0.1897, simple_loss=0.2563, pruned_loss=0.06149, over 953728.64 frames. ], batch size: 28, lr: 3.59e-03, grad_scale: 16.0 2023-03-26 16:03:29,066 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.069e+02 1.641e+02 1.993e+02 2.438e+02 4.377e+02, threshold=3.986e+02, percent-clipped=2.0 2023-03-26 16:03:45,136 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7151, 1.7039, 1.6253, 1.7610, 1.1719, 3.7911, 1.4858, 2.1830], device='cuda:3'), covar=tensor([0.3561, 0.2520, 0.2212, 0.2608, 0.1909, 0.0223, 0.2591, 0.1150], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0115, 0.0119, 0.0123, 0.0115, 0.0097, 0.0097, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 16:03:49,882 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72272.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:03:51,649 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72275.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:03:56,423 INFO [finetune.py:976] (3/7) Epoch 13, batch 3550, loss[loss=0.1895, simple_loss=0.2488, pruned_loss=0.06516, over 4838.00 frames. ], tot_loss[loss=0.1885, simple_loss=0.2546, pruned_loss=0.06118, over 956142.21 frames. ], batch size: 47, lr: 3.59e-03, grad_scale: 16.0 2023-03-26 16:04:17,905 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8756, 1.7880, 1.7368, 1.8488, 1.4165, 3.8572, 1.5875, 2.1647], device='cuda:3'), covar=tensor([0.3296, 0.2419, 0.2048, 0.2351, 0.1677, 0.0205, 0.2493, 0.1183], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0115, 0.0119, 0.0123, 0.0115, 0.0097, 0.0097, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 16:04:37,445 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=72320.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:04:47,560 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72326.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:04:51,694 INFO [finetune.py:976] (3/7) Epoch 13, batch 3600, loss[loss=0.1597, simple_loss=0.2225, pruned_loss=0.04847, over 4797.00 frames. ], tot_loss[loss=0.1863, simple_loss=0.2522, pruned_loss=0.06016, over 957030.29 frames. ], batch size: 29, lr: 3.59e-03, grad_scale: 16.0 2023-03-26 16:04:58,281 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.845e+01 1.525e+02 1.754e+02 2.048e+02 3.586e+02, threshold=3.507e+02, percent-clipped=0.0 2023-03-26 16:05:26,919 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-26 16:05:42,445 INFO [finetune.py:976] (3/7) Epoch 13, batch 3650, loss[loss=0.1657, simple_loss=0.2473, pruned_loss=0.042, over 4941.00 frames. ], tot_loss[loss=0.188, simple_loss=0.2544, pruned_loss=0.06077, over 954956.62 frames. ], batch size: 33, lr: 3.59e-03, grad_scale: 16.0 2023-03-26 16:05:50,438 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72387.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:06:20,308 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9252, 1.4994, 0.8747, 1.7770, 2.2640, 1.5807, 1.6953, 1.6493], device='cuda:3'), covar=tensor([0.1423, 0.1903, 0.2012, 0.1080, 0.1736, 0.1760, 0.1327, 0.1915], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0094, 0.0111, 0.0091, 0.0119, 0.0093, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 16:06:22,157 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1098, 3.5429, 3.7640, 3.9450, 3.8424, 3.6352, 4.1767, 1.3188], device='cuda:3'), covar=tensor([0.0745, 0.0914, 0.0800, 0.1017, 0.1211, 0.1454, 0.0706, 0.5541], device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0241, 0.0274, 0.0289, 0.0329, 0.0279, 0.0301, 0.0294], 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') 2023-03-26 16:06:54,367 INFO [finetune.py:976] (3/7) Epoch 13, batch 3700, loss[loss=0.1769, simple_loss=0.2486, pruned_loss=0.05255, over 4737.00 frames. ], tot_loss[loss=0.1914, simple_loss=0.2588, pruned_loss=0.06202, over 955152.59 frames. ], batch size: 54, lr: 3.59e-03, grad_scale: 16.0 2023-03-26 16:06:55,114 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8758, 1.7259, 1.5361, 1.4334, 1.8669, 1.6131, 1.8912, 1.8319], device='cuda:3'), covar=tensor([0.1429, 0.2040, 0.2989, 0.2534, 0.2606, 0.1764, 0.2480, 0.1837], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0187, 0.0234, 0.0254, 0.0245, 0.0199, 0.0213, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 16:07:04,384 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.108e+02 1.616e+02 1.915e+02 2.308e+02 4.437e+02, threshold=3.829e+02, percent-clipped=1.0 2023-03-26 16:07:52,818 INFO [finetune.py:976] (3/7) Epoch 13, batch 3750, loss[loss=0.1639, simple_loss=0.2376, pruned_loss=0.04507, over 4845.00 frames. ], tot_loss[loss=0.1911, simple_loss=0.2588, pruned_loss=0.06168, over 953945.48 frames. ], batch size: 44, lr: 3.59e-03, grad_scale: 16.0 2023-03-26 16:07:54,153 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72485.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:08:29,283 INFO [finetune.py:976] (3/7) Epoch 13, batch 3800, loss[loss=0.1625, simple_loss=0.2308, pruned_loss=0.04707, over 4723.00 frames. ], tot_loss[loss=0.1919, simple_loss=0.2599, pruned_loss=0.06197, over 953464.67 frames. ], batch size: 23, lr: 3.59e-03, grad_scale: 16.0 2023-03-26 16:08:33,618 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 2023-03-26 16:08:34,665 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.445e+01 1.578e+02 1.803e+02 2.155e+02 3.901e+02, threshold=3.607e+02, percent-clipped=1.0 2023-03-26 16:08:56,827 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72575.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:09:02,586 INFO [finetune.py:976] (3/7) Epoch 13, batch 3850, loss[loss=0.1685, simple_loss=0.2311, pruned_loss=0.05293, over 4919.00 frames. ], tot_loss[loss=0.1897, simple_loss=0.2576, pruned_loss=0.06095, over 953465.95 frames. ], batch size: 37, lr: 3.59e-03, grad_scale: 16.0 2023-03-26 16:09:09,801 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.9227, 3.3577, 3.5674, 3.6706, 3.7088, 3.4572, 3.9437, 1.7122], device='cuda:3'), covar=tensor([0.0694, 0.0842, 0.0742, 0.1020, 0.1055, 0.1215, 0.0766, 0.4208], device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0241, 0.0273, 0.0288, 0.0328, 0.0278, 0.0300, 0.0293], 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') 2023-03-26 16:09:12,221 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4787, 1.0559, 0.8102, 1.3335, 1.9347, 0.7292, 1.2140, 1.3500], device='cuda:3'), covar=tensor([0.1590, 0.2180, 0.1810, 0.1294, 0.1945, 0.1920, 0.1603, 0.2005], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0094, 0.0111, 0.0091, 0.0119, 0.0093, 0.0098, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 16:09:30,598 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=72623.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:09:46,306 INFO [finetune.py:976] (3/7) Epoch 13, batch 3900, loss[loss=0.1288, simple_loss=0.2043, pruned_loss=0.02668, over 4758.00 frames. ], tot_loss[loss=0.1884, simple_loss=0.2555, pruned_loss=0.06062, over 954551.04 frames. ], batch size: 28, lr: 3.59e-03, grad_scale: 16.0 2023-03-26 16:09:51,187 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.402e+01 1.493e+02 1.751e+02 2.217e+02 3.590e+02, threshold=3.501e+02, percent-clipped=0.0 2023-03-26 16:10:05,326 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-26 16:10:18,074 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72682.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:10:18,645 INFO [finetune.py:976] (3/7) Epoch 13, batch 3950, loss[loss=0.1866, simple_loss=0.2561, pruned_loss=0.05855, over 4869.00 frames. ], tot_loss[loss=0.1848, simple_loss=0.2519, pruned_loss=0.05886, over 952925.79 frames. ], batch size: 31, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:10:25,968 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-26 16:10:44,135 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-26 16:11:01,343 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.3006, 2.9389, 3.0376, 3.1913, 3.0657, 2.9243, 3.3504, 0.9606], device='cuda:3'), covar=tensor([0.1114, 0.1045, 0.1097, 0.1272, 0.1677, 0.1716, 0.1103, 0.5027], device='cuda:3'), in_proj_covar=tensor([0.0343, 0.0241, 0.0272, 0.0287, 0.0328, 0.0277, 0.0298, 0.0292], 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') 2023-03-26 16:11:09,402 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4947, 1.4922, 1.2904, 1.5128, 1.8240, 1.6119, 1.4834, 1.2565], device='cuda:3'), covar=tensor([0.0302, 0.0247, 0.0580, 0.0251, 0.0185, 0.0439, 0.0354, 0.0379], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0109, 0.0139, 0.0112, 0.0101, 0.0105, 0.0094, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.2482e-05, 8.4265e-05, 1.1007e-04, 8.6978e-05, 7.8644e-05, 7.7558e-05, 7.1317e-05, 8.2984e-05], device='cuda:3') 2023-03-26 16:11:10,484 INFO [finetune.py:976] (3/7) Epoch 13, batch 4000, loss[loss=0.1904, simple_loss=0.2629, pruned_loss=0.05901, over 4766.00 frames. ], tot_loss[loss=0.1861, simple_loss=0.2527, pruned_loss=0.05969, over 952308.07 frames. ], batch size: 26, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:11:16,799 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.220e+02 1.596e+02 1.921e+02 2.181e+02 4.609e+02, threshold=3.842e+02, percent-clipped=3.0 2023-03-26 16:11:44,631 INFO [finetune.py:976] (3/7) Epoch 13, batch 4050, loss[loss=0.2032, simple_loss=0.2844, pruned_loss=0.06102, over 4858.00 frames. ], tot_loss[loss=0.1885, simple_loss=0.2553, pruned_loss=0.06085, over 950557.32 frames. ], batch size: 44, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:11:46,017 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72785.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:12:16,605 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8381, 1.1695, 1.6755, 1.6754, 1.5607, 1.4693, 1.6489, 1.6442], device='cuda:3'), covar=tensor([0.4431, 0.4362, 0.4376, 0.4096, 0.5696, 0.4223, 0.5245, 0.4117], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0237, 0.0255, 0.0263, 0.0260, 0.0235, 0.0275, 0.0234], 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') 2023-03-26 16:12:27,876 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6322, 1.5119, 1.3450, 1.6981, 2.0020, 1.7157, 1.1469, 1.3549], device='cuda:3'), covar=tensor([0.2321, 0.2129, 0.2042, 0.1670, 0.1681, 0.1236, 0.2741, 0.2049], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0207, 0.0210, 0.0190, 0.0240, 0.0183, 0.0213, 0.0199], 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') 2023-03-26 16:12:39,938 INFO [finetune.py:976] (3/7) Epoch 13, batch 4100, loss[loss=0.1901, simple_loss=0.2688, pruned_loss=0.0557, over 4930.00 frames. ], tot_loss[loss=0.1908, simple_loss=0.2583, pruned_loss=0.06164, over 951091.56 frames. ], batch size: 42, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:12:39,998 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=72833.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:12:45,292 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.089e+02 1.592e+02 1.875e+02 2.230e+02 3.624e+02, threshold=3.749e+02, percent-clipped=0.0 2023-03-26 16:12:59,016 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7482, 3.4496, 3.2833, 1.6607, 3.5225, 2.5420, 0.8225, 2.2705], device='cuda:3'), covar=tensor([0.2051, 0.1835, 0.1716, 0.3038, 0.1165, 0.1091, 0.4152, 0.1587], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0173, 0.0159, 0.0127, 0.0156, 0.0121, 0.0146, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 16:13:13,288 INFO [finetune.py:976] (3/7) Epoch 13, batch 4150, loss[loss=0.1531, simple_loss=0.2148, pruned_loss=0.0457, over 4462.00 frames. ], tot_loss[loss=0.1927, simple_loss=0.2598, pruned_loss=0.06278, over 952439.50 frames. ], batch size: 19, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:13:26,383 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72894.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:14:03,790 INFO [finetune.py:976] (3/7) Epoch 13, batch 4200, loss[loss=0.1615, simple_loss=0.2298, pruned_loss=0.04665, over 4782.00 frames. ], tot_loss[loss=0.1912, simple_loss=0.2589, pruned_loss=0.06171, over 950528.98 frames. ], batch size: 28, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:14:07,369 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-03-26 16:14:08,710 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.693e+01 1.496e+02 1.812e+02 2.169e+02 4.504e+02, threshold=3.624e+02, percent-clipped=2.0 2023-03-26 16:14:18,182 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 2023-03-26 16:14:18,714 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72955.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:14:35,615 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72964.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 16:14:52,550 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72982.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:14:53,055 INFO [finetune.py:976] (3/7) Epoch 13, batch 4250, loss[loss=0.1656, simple_loss=0.2369, pruned_loss=0.04714, over 4903.00 frames. ], tot_loss[loss=0.1878, simple_loss=0.2551, pruned_loss=0.06019, over 950276.46 frames. ], batch size: 36, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:15:21,376 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73025.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 16:15:24,374 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=73030.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:15:26,658 INFO [finetune.py:976] (3/7) Epoch 13, batch 4300, loss[loss=0.1883, simple_loss=0.2527, pruned_loss=0.06192, over 4776.00 frames. ], tot_loss[loss=0.186, simple_loss=0.2528, pruned_loss=0.05963, over 950391.13 frames. ], batch size: 28, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:15:31,522 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.3633, 1.5788, 1.6335, 0.9587, 1.5675, 1.8340, 1.8261, 1.4247], device='cuda:3'), covar=tensor([0.0973, 0.0596, 0.0488, 0.0531, 0.0440, 0.0556, 0.0368, 0.0761], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0151, 0.0122, 0.0128, 0.0130, 0.0125, 0.0142, 0.0145], device='cuda:3'), out_proj_covar=tensor([9.1936e-05, 1.1057e-04, 8.7616e-05, 9.2187e-05, 9.2141e-05, 9.0858e-05, 1.0333e-04, 1.0506e-04], device='cuda:3') 2023-03-26 16:15:31,991 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.052e+02 1.498e+02 1.782e+02 2.254e+02 4.055e+02, threshold=3.563e+02, percent-clipped=2.0 2023-03-26 16:15:43,855 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9113, 1.3972, 0.7774, 1.6821, 2.1346, 1.5635, 1.5642, 1.6953], device='cuda:3'), covar=tensor([0.1549, 0.2027, 0.2246, 0.1272, 0.2042, 0.1961, 0.1479, 0.2096], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0094, 0.0111, 0.0091, 0.0119, 0.0093, 0.0099, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 16:15:59,433 INFO [finetune.py:976] (3/7) Epoch 13, batch 4350, loss[loss=0.1993, simple_loss=0.2458, pruned_loss=0.07639, over 4303.00 frames. ], tot_loss[loss=0.1847, simple_loss=0.2504, pruned_loss=0.05953, over 949349.37 frames. ], batch size: 18, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:16:22,601 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-03-26 16:16:34,969 INFO [finetune.py:976] (3/7) Epoch 13, batch 4400, loss[loss=0.1869, simple_loss=0.2676, pruned_loss=0.05308, over 4808.00 frames. ], tot_loss[loss=0.1869, simple_loss=0.2525, pruned_loss=0.06061, over 949247.03 frames. ], batch size: 29, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:16:40,306 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.328e+01 1.433e+02 1.829e+02 2.142e+02 3.915e+02, threshold=3.659e+02, percent-clipped=1.0 2023-03-26 16:17:06,394 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3078, 2.0714, 1.6525, 0.7520, 1.8762, 1.8344, 1.7402, 1.8346], device='cuda:3'), covar=tensor([0.0977, 0.0934, 0.1796, 0.2460, 0.1683, 0.2567, 0.2409, 0.1197], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0196, 0.0201, 0.0186, 0.0213, 0.0207, 0.0223, 0.0197], 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') 2023-03-26 16:17:08,723 INFO [finetune.py:976] (3/7) Epoch 13, batch 4450, loss[loss=0.2, simple_loss=0.2727, pruned_loss=0.06368, over 4755.00 frames. ], tot_loss[loss=0.1889, simple_loss=0.2561, pruned_loss=0.06081, over 950692.93 frames. ], batch size: 54, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:17:53,145 INFO [finetune.py:976] (3/7) Epoch 13, batch 4500, loss[loss=0.203, simple_loss=0.2614, pruned_loss=0.07234, over 4146.00 frames. ], tot_loss[loss=0.1912, simple_loss=0.2585, pruned_loss=0.0619, over 952362.17 frames. ], batch size: 66, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:17:58,019 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.220e+02 1.732e+02 2.105e+02 2.505e+02 4.470e+02, threshold=4.210e+02, percent-clipped=3.0 2023-03-26 16:17:58,164 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6943, 1.5132, 1.3890, 1.6740, 2.2355, 1.7013, 1.5376, 1.3681], device='cuda:3'), covar=tensor([0.2159, 0.2244, 0.2051, 0.1752, 0.1588, 0.1298, 0.2384, 0.1988], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0207, 0.0209, 0.0190, 0.0240, 0.0183, 0.0212, 0.0198], 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') 2023-03-26 16:18:00,492 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7592, 1.5281, 1.3947, 1.2097, 1.4955, 1.5355, 1.4767, 2.0921], device='cuda:3'), covar=tensor([0.4008, 0.4502, 0.3362, 0.4072, 0.4083, 0.2484, 0.4113, 0.1956], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0261, 0.0224, 0.0278, 0.0247, 0.0213, 0.0248, 0.0223], 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') 2023-03-26 16:18:04,478 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73250.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:18:05,809 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.1726, 1.3261, 1.3570, 0.8123, 1.2383, 1.5339, 1.5959, 1.2548], device='cuda:3'), covar=tensor([0.0930, 0.0588, 0.0549, 0.0476, 0.0491, 0.0663, 0.0338, 0.0693], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0152, 0.0122, 0.0128, 0.0130, 0.0126, 0.0142, 0.0145], device='cuda:3'), out_proj_covar=tensor([9.2281e-05, 1.1105e-04, 8.8025e-05, 9.2471e-05, 9.2547e-05, 9.1041e-05, 1.0364e-04, 1.0541e-04], device='cuda:3') 2023-03-26 16:18:22,206 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4792, 1.4205, 1.8117, 1.6968, 1.4926, 3.3887, 1.2500, 1.5015], device='cuda:3'), covar=tensor([0.0959, 0.1846, 0.1231, 0.0991, 0.1720, 0.0233, 0.1559, 0.1740], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0073, 0.0077, 0.0091, 0.0080, 0.0084, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 16:18:26,892 INFO [finetune.py:976] (3/7) Epoch 13, batch 4550, loss[loss=0.1903, simple_loss=0.2528, pruned_loss=0.06393, over 4787.00 frames. ], tot_loss[loss=0.1928, simple_loss=0.2607, pruned_loss=0.06251, over 952663.62 frames. ], batch size: 29, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:18:34,960 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73293.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:18:47,046 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 2023-03-26 16:19:07,529 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73320.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 16:19:19,981 INFO [finetune.py:976] (3/7) Epoch 13, batch 4600, loss[loss=0.2016, simple_loss=0.2576, pruned_loss=0.07282, over 4801.00 frames. ], tot_loss[loss=0.1912, simple_loss=0.2596, pruned_loss=0.06142, over 953038.33 frames. ], batch size: 51, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:19:24,888 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.809e+01 1.600e+02 1.901e+02 2.194e+02 3.702e+02, threshold=3.803e+02, percent-clipped=0.0 2023-03-26 16:19:42,009 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73354.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:19:57,392 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-03-26 16:20:11,746 INFO [finetune.py:976] (3/7) Epoch 13, batch 4650, loss[loss=0.1967, simple_loss=0.2607, pruned_loss=0.06631, over 4731.00 frames. ], tot_loss[loss=0.1898, simple_loss=0.2573, pruned_loss=0.0612, over 952970.35 frames. ], batch size: 23, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:20:25,536 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5143, 2.0264, 2.8629, 4.3499, 3.1217, 2.9777, 1.3293, 3.5704], device='cuda:3'), covar=tensor([0.1469, 0.1362, 0.1230, 0.0453, 0.0665, 0.1325, 0.1740, 0.0392], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0116, 0.0134, 0.0166, 0.0102, 0.0139, 0.0127, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 16:20:26,192 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.5730, 1.5843, 1.5323, 0.8444, 1.5726, 1.8057, 1.7311, 1.4064], device='cuda:3'), covar=tensor([0.0947, 0.0577, 0.0471, 0.0575, 0.0451, 0.0498, 0.0324, 0.0601], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0152, 0.0123, 0.0129, 0.0131, 0.0126, 0.0143, 0.0145], device='cuda:3'), out_proj_covar=tensor([9.2657e-05, 1.1123e-04, 8.8212e-05, 9.2521e-05, 9.2908e-05, 9.1212e-05, 1.0375e-04, 1.0563e-04], device='cuda:3') 2023-03-26 16:20:45,672 INFO [finetune.py:976] (3/7) Epoch 13, batch 4700, loss[loss=0.1678, simple_loss=0.2405, pruned_loss=0.04756, over 4897.00 frames. ], tot_loss[loss=0.1873, simple_loss=0.2542, pruned_loss=0.06019, over 952885.76 frames. ], batch size: 35, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:20:50,437 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.031e+02 1.614e+02 1.909e+02 2.257e+02 3.771e+02, threshold=3.817e+02, percent-clipped=0.0 2023-03-26 16:20:53,046 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-03-26 16:20:53,391 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.0487, 3.5000, 3.7393, 3.8413, 3.8583, 3.5554, 4.1065, 1.2862], device='cuda:3'), covar=tensor([0.0709, 0.0866, 0.0746, 0.0970, 0.1026, 0.1386, 0.0752, 0.5157], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0243, 0.0276, 0.0291, 0.0333, 0.0281, 0.0302, 0.0297], 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') 2023-03-26 16:21:18,783 INFO [finetune.py:976] (3/7) Epoch 13, batch 4750, loss[loss=0.1317, simple_loss=0.2048, pruned_loss=0.0293, over 4757.00 frames. ], tot_loss[loss=0.1848, simple_loss=0.2516, pruned_loss=0.059, over 952732.00 frames. ], batch size: 28, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:21:43,679 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.40 vs. limit=5.0 2023-03-26 16:21:51,904 INFO [finetune.py:976] (3/7) Epoch 13, batch 4800, loss[loss=0.1953, simple_loss=0.2519, pruned_loss=0.06935, over 4721.00 frames. ], tot_loss[loss=0.1875, simple_loss=0.2543, pruned_loss=0.06037, over 952710.25 frames. ], batch size: 23, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:21:57,193 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.195e+02 1.637e+02 2.007e+02 2.318e+02 3.852e+02, threshold=4.014e+02, percent-clipped=1.0 2023-03-26 16:22:03,309 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73550.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:22:11,731 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.2122, 4.4504, 4.8286, 5.0887, 4.9477, 4.6410, 5.3414, 1.6290], device='cuda:3'), covar=tensor([0.0633, 0.0842, 0.0772, 0.0813, 0.1049, 0.1406, 0.0493, 0.5636], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0244, 0.0278, 0.0292, 0.0334, 0.0282, 0.0303, 0.0297], 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') 2023-03-26 16:22:24,805 INFO [finetune.py:976] (3/7) Epoch 13, batch 4850, loss[loss=0.1537, simple_loss=0.2185, pruned_loss=0.04446, over 4750.00 frames. ], tot_loss[loss=0.19, simple_loss=0.2577, pruned_loss=0.0612, over 953521.49 frames. ], batch size: 27, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:22:30,084 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73590.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:22:37,207 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=73598.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:23:00,188 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73620.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 16:23:08,517 INFO [finetune.py:976] (3/7) Epoch 13, batch 4900, loss[loss=0.2049, simple_loss=0.2661, pruned_loss=0.07179, over 4193.00 frames. ], tot_loss[loss=0.1891, simple_loss=0.257, pruned_loss=0.06055, over 952645.75 frames. ], batch size: 65, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:23:14,277 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.113e+02 1.751e+02 2.108e+02 2.596e+02 5.059e+02, threshold=4.217e+02, percent-clipped=3.0 2023-03-26 16:23:19,748 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73649.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:23:19,795 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.6328, 1.6730, 1.6602, 1.1208, 1.7400, 1.9798, 1.8982, 1.4842], device='cuda:3'), covar=tensor([0.1067, 0.0723, 0.0555, 0.0597, 0.0526, 0.0713, 0.0390, 0.0760], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0154, 0.0124, 0.0130, 0.0132, 0.0127, 0.0144, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.3814e-05, 1.1226e-04, 8.9323e-05, 9.3439e-05, 9.3561e-05, 9.1723e-05, 1.0456e-04, 1.0662e-04], device='cuda:3') 2023-03-26 16:23:21,053 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73651.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:23:31,878 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=73668.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 16:23:41,382 INFO [finetune.py:976] (3/7) Epoch 13, batch 4950, loss[loss=0.138, simple_loss=0.2174, pruned_loss=0.02927, over 4781.00 frames. ], tot_loss[loss=0.1897, simple_loss=0.2579, pruned_loss=0.06079, over 950212.88 frames. ], batch size: 29, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:24:24,438 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73732.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:24:24,937 INFO [finetune.py:976] (3/7) Epoch 13, batch 5000, loss[loss=0.1504, simple_loss=0.2211, pruned_loss=0.03979, over 4775.00 frames. ], tot_loss[loss=0.19, simple_loss=0.2577, pruned_loss=0.0611, over 950718.37 frames. ], batch size: 51, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:24:33,742 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.082e+02 1.586e+02 1.888e+02 2.371e+02 3.310e+02, threshold=3.776e+02, percent-clipped=1.0 2023-03-26 16:24:51,395 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.0202, 0.8988, 0.8250, 1.0591, 1.2312, 1.1506, 0.9979, 0.9621], device='cuda:3'), covar=tensor([0.0352, 0.0362, 0.0813, 0.0318, 0.0285, 0.0414, 0.0350, 0.0440], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0109, 0.0139, 0.0113, 0.0101, 0.0105, 0.0096, 0.0109], device='cuda:3'), out_proj_covar=tensor([7.2807e-05, 8.4209e-05, 1.1030e-04, 8.7537e-05, 7.8703e-05, 7.7410e-05, 7.2173e-05, 8.3371e-05], device='cuda:3') 2023-03-26 16:25:17,330 INFO [finetune.py:976] (3/7) Epoch 13, batch 5050, loss[loss=0.1825, simple_loss=0.2399, pruned_loss=0.06255, over 4821.00 frames. ], tot_loss[loss=0.1879, simple_loss=0.2547, pruned_loss=0.06056, over 950736.35 frames. ], batch size: 41, lr: 3.58e-03, grad_scale: 16.0 2023-03-26 16:25:26,634 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73793.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:25:40,331 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6856, 1.5840, 1.5437, 1.6746, 1.0284, 3.3360, 1.2490, 1.7912], device='cuda:3'), covar=tensor([0.3221, 0.2540, 0.2164, 0.2266, 0.1906, 0.0216, 0.2722, 0.1286], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0120, 0.0123, 0.0115, 0.0098, 0.0097, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 16:25:53,900 INFO [finetune.py:976] (3/7) Epoch 13, batch 5100, loss[loss=0.1863, simple_loss=0.2405, pruned_loss=0.0661, over 4774.00 frames. ], tot_loss[loss=0.1862, simple_loss=0.2524, pruned_loss=0.05999, over 951961.22 frames. ], batch size: 51, lr: 3.57e-03, grad_scale: 16.0 2023-03-26 16:25:59,155 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.804e+01 1.435e+02 1.752e+02 2.086e+02 3.868e+02, threshold=3.504e+02, percent-clipped=1.0 2023-03-26 16:26:27,642 INFO [finetune.py:976] (3/7) Epoch 13, batch 5150, loss[loss=0.251, simple_loss=0.305, pruned_loss=0.09849, over 4893.00 frames. ], tot_loss[loss=0.1872, simple_loss=0.2531, pruned_loss=0.0606, over 954574.38 frames. ], batch size: 35, lr: 3.57e-03, grad_scale: 16.0 2023-03-26 16:26:28,963 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4242, 1.3854, 1.9691, 2.8847, 1.8935, 2.1628, 0.8099, 2.3296], device='cuda:3'), covar=tensor([0.1805, 0.1480, 0.1155, 0.0604, 0.0888, 0.1338, 0.1898, 0.0577], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0115, 0.0133, 0.0165, 0.0101, 0.0138, 0.0126, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 16:26:35,453 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8232, 1.1665, 1.8675, 1.7289, 1.5947, 1.5421, 1.7079, 1.6811], device='cuda:3'), covar=tensor([0.3236, 0.3655, 0.2923, 0.3388, 0.4036, 0.3328, 0.3861, 0.2782], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0237, 0.0256, 0.0264, 0.0262, 0.0236, 0.0276, 0.0234], 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') 2023-03-26 16:26:44,137 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2386, 2.1520, 1.8454, 1.9422, 2.2905, 1.9636, 2.4126, 2.2093], device='cuda:3'), covar=tensor([0.1238, 0.1833, 0.2816, 0.2273, 0.2162, 0.1557, 0.2208, 0.1537], device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0189, 0.0235, 0.0256, 0.0245, 0.0201, 0.0214, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 16:26:56,499 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.48 vs. limit=5.0 2023-03-26 16:27:01,333 INFO [finetune.py:976] (3/7) Epoch 13, batch 5200, loss[loss=0.2038, simple_loss=0.2643, pruned_loss=0.07167, over 4788.00 frames. ], tot_loss[loss=0.1891, simple_loss=0.2558, pruned_loss=0.06119, over 954474.76 frames. ], batch size: 25, lr: 3.57e-03, grad_scale: 16.0 2023-03-26 16:27:06,220 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.103e+02 1.675e+02 1.952e+02 2.217e+02 3.649e+02, threshold=3.904e+02, percent-clipped=2.0 2023-03-26 16:27:09,803 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73946.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:27:11,691 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73949.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:27:21,156 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-03-26 16:27:34,704 INFO [finetune.py:976] (3/7) Epoch 13, batch 5250, loss[loss=0.2277, simple_loss=0.2894, pruned_loss=0.08297, over 4809.00 frames. ], tot_loss[loss=0.1907, simple_loss=0.2585, pruned_loss=0.06145, over 955288.28 frames. ], batch size: 40, lr: 3.57e-03, grad_scale: 16.0 2023-03-26 16:27:44,364 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=73997.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:28:11,316 INFO [finetune.py:976] (3/7) Epoch 13, batch 5300, loss[loss=0.2334, simple_loss=0.2938, pruned_loss=0.08653, over 4885.00 frames. ], tot_loss[loss=0.1909, simple_loss=0.2595, pruned_loss=0.06112, over 953313.81 frames. ], batch size: 32, lr: 3.57e-03, grad_scale: 32.0 2023-03-26 16:28:17,126 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.042e+02 1.638e+02 2.059e+02 2.515e+02 4.122e+02, threshold=4.117e+02, percent-clipped=3.0 2023-03-26 16:28:33,135 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74064.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:28:45,018 INFO [finetune.py:976] (3/7) Epoch 13, batch 5350, loss[loss=0.2019, simple_loss=0.2656, pruned_loss=0.06904, over 4870.00 frames. ], tot_loss[loss=0.19, simple_loss=0.2589, pruned_loss=0.06054, over 952679.52 frames. ], batch size: 34, lr: 3.57e-03, grad_scale: 32.0 2023-03-26 16:28:48,556 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74088.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:28:53,210 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6525, 1.5403, 1.5166, 1.6171, 1.0765, 3.3546, 1.3851, 1.9080], device='cuda:3'), covar=tensor([0.3145, 0.2365, 0.2031, 0.2226, 0.1825, 0.0189, 0.2604, 0.1218], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0120, 0.0123, 0.0115, 0.0098, 0.0097, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 16:29:12,966 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74125.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:29:18,158 INFO [finetune.py:976] (3/7) Epoch 13, batch 5400, loss[loss=0.1927, simple_loss=0.2598, pruned_loss=0.06278, over 4905.00 frames. ], tot_loss[loss=0.1887, simple_loss=0.2565, pruned_loss=0.06042, over 955407.28 frames. ], batch size: 46, lr: 3.57e-03, grad_scale: 32.0 2023-03-26 16:29:27,916 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.066e+02 1.534e+02 1.853e+02 2.261e+02 4.254e+02, threshold=3.706e+02, percent-clipped=1.0 2023-03-26 16:29:47,733 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4438, 2.2784, 1.6390, 2.3313, 2.3291, 1.9890, 3.1473, 2.3592], device='cuda:3'), covar=tensor([0.1382, 0.2286, 0.3518, 0.3164, 0.2668, 0.1690, 0.2209, 0.1990], device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0189, 0.0235, 0.0255, 0.0245, 0.0201, 0.0214, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 16:29:54,175 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=6.35 vs. limit=5.0 2023-03-26 16:30:11,889 INFO [finetune.py:976] (3/7) Epoch 13, batch 5450, loss[loss=0.1696, simple_loss=0.2296, pruned_loss=0.05477, over 4781.00 frames. ], tot_loss[loss=0.1857, simple_loss=0.2529, pruned_loss=0.05924, over 953330.53 frames. ], batch size: 26, lr: 3.57e-03, grad_scale: 32.0 2023-03-26 16:30:56,457 INFO [finetune.py:976] (3/7) Epoch 13, batch 5500, loss[loss=0.176, simple_loss=0.2423, pruned_loss=0.05488, over 4828.00 frames. ], tot_loss[loss=0.1823, simple_loss=0.2491, pruned_loss=0.05777, over 950814.98 frames. ], batch size: 30, lr: 3.57e-03, grad_scale: 32.0 2023-03-26 16:30:59,088 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-26 16:31:01,349 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74240.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:31:01,849 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.610e+01 1.534e+02 1.911e+02 2.187e+02 5.924e+02, threshold=3.822e+02, percent-clipped=2.0 2023-03-26 16:31:04,990 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74246.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:31:13,386 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4715, 1.4398, 1.5620, 1.7257, 1.4480, 3.1415, 1.3052, 1.5146], device='cuda:3'), covar=tensor([0.0921, 0.1733, 0.1204, 0.0935, 0.1653, 0.0260, 0.1492, 0.1624], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0080, 0.0073, 0.0077, 0.0091, 0.0080, 0.0084, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 16:31:30,480 INFO [finetune.py:976] (3/7) Epoch 13, batch 5550, loss[loss=0.2019, simple_loss=0.286, pruned_loss=0.05888, over 4813.00 frames. ], tot_loss[loss=0.1844, simple_loss=0.2514, pruned_loss=0.05864, over 952355.41 frames. ], batch size: 40, lr: 3.57e-03, grad_scale: 32.0 2023-03-26 16:31:34,810 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4319, 1.3626, 1.2477, 1.3866, 1.6774, 1.5757, 1.3953, 1.1925], device='cuda:3'), covar=tensor([0.0310, 0.0269, 0.0631, 0.0286, 0.0248, 0.0349, 0.0321, 0.0410], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0108, 0.0139, 0.0113, 0.0101, 0.0105, 0.0095, 0.0109], device='cuda:3'), out_proj_covar=tensor([7.2509e-05, 8.3898e-05, 1.1022e-04, 8.7547e-05, 7.8694e-05, 7.7496e-05, 7.1853e-05, 8.3581e-05], device='cuda:3') 2023-03-26 16:31:37,727 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=74294.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:31:42,502 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74301.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:32:02,275 INFO [finetune.py:976] (3/7) Epoch 13, batch 5600, loss[loss=0.2113, simple_loss=0.2831, pruned_loss=0.06977, over 4917.00 frames. ], tot_loss[loss=0.1874, simple_loss=0.2554, pruned_loss=0.05965, over 952620.77 frames. ], batch size: 42, lr: 3.57e-03, grad_scale: 32.0 2023-03-26 16:32:06,846 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.439e+01 1.528e+02 1.833e+02 2.251e+02 4.644e+02, threshold=3.666e+02, percent-clipped=1.0 2023-03-26 16:32:31,595 INFO [finetune.py:976] (3/7) Epoch 13, batch 5650, loss[loss=0.2366, simple_loss=0.2893, pruned_loss=0.09193, over 4836.00 frames. ], tot_loss[loss=0.1891, simple_loss=0.2579, pruned_loss=0.06012, over 952379.24 frames. ], batch size: 30, lr: 3.57e-03, grad_scale: 32.0 2023-03-26 16:32:35,018 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74388.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:32:54,211 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74420.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:33:01,852 INFO [finetune.py:976] (3/7) Epoch 13, batch 5700, loss[loss=0.1894, simple_loss=0.2415, pruned_loss=0.06865, over 4062.00 frames. ], tot_loss[loss=0.1865, simple_loss=0.2536, pruned_loss=0.05968, over 934746.14 frames. ], batch size: 17, lr: 3.57e-03, grad_scale: 32.0 2023-03-26 16:33:03,646 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=74436.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:33:06,489 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.627e+01 1.487e+02 1.916e+02 2.529e+02 4.839e+02, threshold=3.833e+02, percent-clipped=5.0 2023-03-26 16:33:31,117 INFO [finetune.py:976] (3/7) Epoch 14, batch 0, loss[loss=0.1807, simple_loss=0.239, pruned_loss=0.06122, over 4549.00 frames. ], tot_loss[loss=0.1807, simple_loss=0.239, pruned_loss=0.06122, over 4549.00 frames. ], batch size: 20, lr: 3.57e-03, grad_scale: 32.0 2023-03-26 16:33:31,117 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 16:33:41,691 INFO [finetune.py:1010] (3/7) Epoch 14, validation: loss=0.1582, simple_loss=0.2295, pruned_loss=0.04344, over 2265189.00 frames. 2023-03-26 16:33:41,692 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6434MB 2023-03-26 16:33:44,171 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9332, 1.9146, 1.6594, 1.8588, 1.9573, 1.6821, 2.1548, 1.9814], device='cuda:3'), covar=tensor([0.1223, 0.1904, 0.2604, 0.2397, 0.2096, 0.1506, 0.2723, 0.1570], device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0189, 0.0234, 0.0256, 0.0245, 0.0201, 0.0214, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 16:33:47,843 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.7663, 1.7202, 1.6836, 1.1487, 1.7688, 2.0186, 1.9899, 1.5686], device='cuda:3'), covar=tensor([0.0892, 0.0623, 0.0569, 0.0540, 0.0432, 0.0567, 0.0364, 0.0694], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0154, 0.0124, 0.0130, 0.0133, 0.0128, 0.0144, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.3835e-05, 1.1254e-04, 8.9182e-05, 9.3891e-05, 9.4142e-05, 9.2527e-05, 1.0451e-04, 1.0675e-04], device='cuda:3') 2023-03-26 16:34:10,272 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8811, 1.7527, 1.6793, 1.7289, 1.2820, 3.9486, 1.6052, 1.8746], device='cuda:3'), covar=tensor([0.3143, 0.2322, 0.2038, 0.2200, 0.1608, 0.0177, 0.2622, 0.1345], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0116, 0.0120, 0.0124, 0.0115, 0.0098, 0.0097, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 16:34:14,919 INFO [finetune.py:976] (3/7) Epoch 14, batch 50, loss[loss=0.2378, simple_loss=0.2861, pruned_loss=0.09479, over 4716.00 frames. ], tot_loss[loss=0.1933, simple_loss=0.2605, pruned_loss=0.06304, over 216647.91 frames. ], batch size: 59, lr: 3.57e-03, grad_scale: 32.0 2023-03-26 16:34:42,636 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.553e+01 1.578e+02 1.920e+02 2.248e+02 3.729e+02, threshold=3.841e+02, percent-clipped=1.0 2023-03-26 16:35:02,562 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3620, 2.2382, 2.0861, 2.4769, 2.7860, 2.2398, 2.0933, 1.7535], device='cuda:3'), covar=tensor([0.2166, 0.1855, 0.1826, 0.1507, 0.1627, 0.1090, 0.2170, 0.1970], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0207, 0.0209, 0.0190, 0.0239, 0.0183, 0.0213, 0.0198], 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') 2023-03-26 16:35:04,283 INFO [finetune.py:976] (3/7) Epoch 14, batch 100, loss[loss=0.1392, simple_loss=0.2154, pruned_loss=0.03153, over 4818.00 frames. ], tot_loss[loss=0.1867, simple_loss=0.2534, pruned_loss=0.05999, over 379715.00 frames. ], batch size: 51, lr: 3.57e-03, grad_scale: 32.0 2023-03-26 16:35:31,034 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74595.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:35:31,614 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74596.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:35:49,900 INFO [finetune.py:976] (3/7) Epoch 14, batch 150, loss[loss=0.1572, simple_loss=0.229, pruned_loss=0.0427, over 4826.00 frames. ], tot_loss[loss=0.1821, simple_loss=0.2479, pruned_loss=0.05812, over 509048.89 frames. ], batch size: 41, lr: 3.57e-03, grad_scale: 32.0 2023-03-26 16:36:22,702 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.045e+02 1.552e+02 1.795e+02 2.139e+02 3.747e+02, threshold=3.589e+02, percent-clipped=0.0 2023-03-26 16:36:33,032 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74656.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:36:35,947 INFO [finetune.py:976] (3/7) Epoch 14, batch 200, loss[loss=0.1756, simple_loss=0.2443, pruned_loss=0.05349, over 4903.00 frames. ], tot_loss[loss=0.1835, simple_loss=0.2493, pruned_loss=0.05888, over 608714.04 frames. ], batch size: 35, lr: 3.57e-03, grad_scale: 32.0 2023-03-26 16:37:09,842 INFO [finetune.py:976] (3/7) Epoch 14, batch 250, loss[loss=0.208, simple_loss=0.2767, pruned_loss=0.06962, over 4796.00 frames. ], tot_loss[loss=0.1881, simple_loss=0.2547, pruned_loss=0.06071, over 687073.64 frames. ], batch size: 29, lr: 3.57e-03, grad_scale: 32.0 2023-03-26 16:37:15,965 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74720.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:37:27,378 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7180, 1.6349, 1.4614, 1.8585, 2.2638, 1.7720, 1.4327, 1.3717], device='cuda:3'), covar=tensor([0.2258, 0.2073, 0.2021, 0.1686, 0.1743, 0.1226, 0.2591, 0.2065], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0207, 0.0209, 0.0190, 0.0239, 0.0183, 0.0213, 0.0198], 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') 2023-03-26 16:37:30,100 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.051e+02 1.719e+02 2.081e+02 2.468e+02 4.342e+02, threshold=4.162e+02, percent-clipped=2.0 2023-03-26 16:37:42,699 INFO [finetune.py:976] (3/7) Epoch 14, batch 300, loss[loss=0.251, simple_loss=0.3167, pruned_loss=0.09268, over 4902.00 frames. ], tot_loss[loss=0.1903, simple_loss=0.2584, pruned_loss=0.06106, over 747999.11 frames. ], batch size: 38, lr: 3.56e-03, grad_scale: 32.0 2023-03-26 16:37:48,014 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=74768.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:37:48,674 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7508, 2.5329, 2.3706, 1.4210, 2.5617, 2.1369, 1.8655, 2.3131], device='cuda:3'), covar=tensor([0.1355, 0.0741, 0.1584, 0.2047, 0.1577, 0.2164, 0.2241, 0.1111], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0196, 0.0199, 0.0185, 0.0212, 0.0208, 0.0223, 0.0197], 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') 2023-03-26 16:38:16,372 INFO [finetune.py:976] (3/7) Epoch 14, batch 350, loss[loss=0.1591, simple_loss=0.23, pruned_loss=0.04406, over 4868.00 frames. ], tot_loss[loss=0.1924, simple_loss=0.2605, pruned_loss=0.06218, over 793836.55 frames. ], batch size: 34, lr: 3.56e-03, grad_scale: 32.0 2023-03-26 16:38:28,462 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-03-26 16:38:36,814 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.618e+01 1.648e+02 1.967e+02 2.475e+02 5.107e+02, threshold=3.933e+02, percent-clipped=3.0 2023-03-26 16:38:49,813 INFO [finetune.py:976] (3/7) Epoch 14, batch 400, loss[loss=0.2063, simple_loss=0.2706, pruned_loss=0.07099, over 4866.00 frames. ], tot_loss[loss=0.193, simple_loss=0.2611, pruned_loss=0.06241, over 831033.70 frames. ], batch size: 31, lr: 3.56e-03, grad_scale: 32.0 2023-03-26 16:39:13,554 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74896.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:39:15,504 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-26 16:39:22,409 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74909.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:39:23,518 INFO [finetune.py:976] (3/7) Epoch 14, batch 450, loss[loss=0.16, simple_loss=0.2349, pruned_loss=0.04254, over 4941.00 frames. ], tot_loss[loss=0.1921, simple_loss=0.2602, pruned_loss=0.06203, over 857842.21 frames. ], batch size: 39, lr: 3.56e-03, grad_scale: 32.0 2023-03-26 16:39:26,088 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1999, 2.1210, 1.9154, 2.3237, 2.8831, 2.2667, 2.2580, 1.6454], device='cuda:3'), covar=tensor([0.2224, 0.1976, 0.1894, 0.1577, 0.1645, 0.1099, 0.1967, 0.1884], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0207, 0.0209, 0.0189, 0.0239, 0.0183, 0.0213, 0.0198], 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') 2023-03-26 16:39:43,596 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.160e+02 1.587e+02 1.868e+02 2.260e+02 4.285e+02, threshold=3.737e+02, percent-clipped=2.0 2023-03-26 16:39:45,322 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8087, 4.3443, 4.1165, 2.1977, 4.5297, 3.1661, 0.7861, 2.9586], device='cuda:3'), covar=tensor([0.2591, 0.1663, 0.1527, 0.2995, 0.0751, 0.1043, 0.4521, 0.1467], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0172, 0.0160, 0.0127, 0.0155, 0.0121, 0.0145, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 16:39:45,925 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=74944.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:39:50,155 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74951.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:39:52,297 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-26 16:39:58,628 INFO [finetune.py:976] (3/7) Epoch 14, batch 500, loss[loss=0.1664, simple_loss=0.2372, pruned_loss=0.04779, over 4829.00 frames. ], tot_loss[loss=0.1882, simple_loss=0.256, pruned_loss=0.06022, over 880392.95 frames. ], batch size: 33, lr: 3.56e-03, grad_scale: 32.0 2023-03-26 16:40:00,590 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74964.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:40:08,360 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8055, 1.5935, 1.4800, 1.8734, 2.0468, 1.8553, 1.3579, 1.5199], device='cuda:3'), covar=tensor([0.2226, 0.2103, 0.2016, 0.1684, 0.1588, 0.1189, 0.2524, 0.1981], device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0206, 0.0208, 0.0189, 0.0238, 0.0183, 0.0212, 0.0197], 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') 2023-03-26 16:40:08,964 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74970.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:40:39,057 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3261, 2.2713, 2.2979, 1.0261, 2.5979, 2.8811, 2.3772, 2.1386], device='cuda:3'), covar=tensor([0.0770, 0.0713, 0.0540, 0.0700, 0.0478, 0.0511, 0.0466, 0.0684], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0152, 0.0122, 0.0129, 0.0130, 0.0126, 0.0141, 0.0145], device='cuda:3'), out_proj_covar=tensor([9.2396e-05, 1.1075e-04, 8.7793e-05, 9.2579e-05, 9.2028e-05, 9.1241e-05, 1.0262e-04, 1.0520e-04], device='cuda:3') 2023-03-26 16:40:45,475 INFO [finetune.py:976] (3/7) Epoch 14, batch 550, loss[loss=0.1586, simple_loss=0.2175, pruned_loss=0.04988, over 4826.00 frames. ], tot_loss[loss=0.1866, simple_loss=0.2531, pruned_loss=0.06002, over 895309.88 frames. ], batch size: 51, lr: 3.56e-03, grad_scale: 32.0 2023-03-26 16:40:57,877 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75025.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:41:09,612 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.017e+02 1.564e+02 1.846e+02 2.204e+02 7.411e+02, threshold=3.691e+02, percent-clipped=3.0 2023-03-26 16:41:23,831 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 2023-03-26 16:41:32,954 INFO [finetune.py:976] (3/7) Epoch 14, batch 600, loss[loss=0.2272, simple_loss=0.2887, pruned_loss=0.0828, over 4134.00 frames. ], tot_loss[loss=0.187, simple_loss=0.2538, pruned_loss=0.06007, over 909588.26 frames. ], batch size: 65, lr: 3.56e-03, grad_scale: 16.0 2023-03-26 16:41:43,536 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-03-26 16:41:52,040 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.1161, 1.0129, 1.0398, 0.4888, 0.8873, 1.1551, 1.2114, 1.0380], device='cuda:3'), covar=tensor([0.0923, 0.0639, 0.0524, 0.0501, 0.0541, 0.0583, 0.0425, 0.0658], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0153, 0.0123, 0.0130, 0.0131, 0.0127, 0.0142, 0.0146], device='cuda:3'), out_proj_covar=tensor([9.3157e-05, 1.1175e-04, 8.8233e-05, 9.3210e-05, 9.2859e-05, 9.1982e-05, 1.0318e-04, 1.0596e-04], device='cuda:3') 2023-03-26 16:42:10,222 INFO [finetune.py:976] (3/7) Epoch 14, batch 650, loss[loss=0.1974, simple_loss=0.2736, pruned_loss=0.06058, over 4813.00 frames. ], tot_loss[loss=0.1899, simple_loss=0.2575, pruned_loss=0.06114, over 920899.90 frames. ], batch size: 39, lr: 3.56e-03, grad_scale: 16.0 2023-03-26 16:42:30,910 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.139e+02 1.620e+02 1.922e+02 2.248e+02 3.855e+02, threshold=3.845e+02, percent-clipped=1.0 2023-03-26 16:42:43,796 INFO [finetune.py:976] (3/7) Epoch 14, batch 700, loss[loss=0.1699, simple_loss=0.2291, pruned_loss=0.05533, over 4743.00 frames. ], tot_loss[loss=0.1908, simple_loss=0.2588, pruned_loss=0.06136, over 929401.80 frames. ], batch size: 27, lr: 3.56e-03, grad_scale: 16.0 2023-03-26 16:43:14,995 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 2023-03-26 16:43:16,884 INFO [finetune.py:976] (3/7) Epoch 14, batch 750, loss[loss=0.1845, simple_loss=0.2503, pruned_loss=0.05938, over 4796.00 frames. ], tot_loss[loss=0.1922, simple_loss=0.2605, pruned_loss=0.06201, over 936113.78 frames. ], batch size: 29, lr: 3.56e-03, grad_scale: 16.0 2023-03-26 16:43:28,273 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75228.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:43:37,215 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4848, 1.5966, 1.2786, 1.5181, 1.8847, 1.6445, 1.5272, 1.3501], device='cuda:3'), covar=tensor([0.0338, 0.0229, 0.0541, 0.0281, 0.0178, 0.0451, 0.0292, 0.0330], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0108, 0.0140, 0.0113, 0.0100, 0.0105, 0.0095, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.2473e-05, 8.3693e-05, 1.1062e-04, 8.7584e-05, 7.8007e-05, 7.7997e-05, 7.1546e-05, 8.2724e-05], device='cuda:3') 2023-03-26 16:43:37,703 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.931e+01 1.583e+02 1.834e+02 2.163e+02 4.783e+02, threshold=3.668e+02, percent-clipped=1.0 2023-03-26 16:43:39,066 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2833, 2.0117, 2.1411, 0.9930, 2.3980, 2.6766, 2.1266, 1.9476], device='cuda:3'), covar=tensor([0.1082, 0.0776, 0.0539, 0.0767, 0.0608, 0.0597, 0.0519, 0.0738], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0152, 0.0122, 0.0128, 0.0130, 0.0126, 0.0141, 0.0145], device='cuda:3'), out_proj_covar=tensor([9.2595e-05, 1.1078e-04, 8.7454e-05, 9.2236e-05, 9.2143e-05, 9.1101e-05, 1.0273e-04, 1.0519e-04], device='cuda:3') 2023-03-26 16:43:44,242 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75251.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:43:50,667 INFO [finetune.py:976] (3/7) Epoch 14, batch 800, loss[loss=0.1732, simple_loss=0.2416, pruned_loss=0.05246, over 4772.00 frames. ], tot_loss[loss=0.1908, simple_loss=0.2591, pruned_loss=0.06124, over 940055.24 frames. ], batch size: 28, lr: 3.56e-03, grad_scale: 16.0 2023-03-26 16:43:53,646 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75265.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:43:56,116 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6291, 1.3802, 1.3802, 1.6555, 2.0161, 1.6847, 1.2637, 1.3805], device='cuda:3'), covar=tensor([0.2430, 0.2444, 0.2227, 0.1874, 0.1669, 0.1461, 0.2734, 0.2153], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0206, 0.0209, 0.0189, 0.0238, 0.0183, 0.0212, 0.0197], 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') 2023-03-26 16:43:56,685 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75270.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:44:09,540 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75289.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:44:16,558 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=75299.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:44:24,290 INFO [finetune.py:976] (3/7) Epoch 14, batch 850, loss[loss=0.1761, simple_loss=0.2511, pruned_loss=0.05062, over 4889.00 frames. ], tot_loss[loss=0.1888, simple_loss=0.2566, pruned_loss=0.06048, over 944481.25 frames. ], batch size: 35, lr: 3.56e-03, grad_scale: 16.0 2023-03-26 16:44:30,269 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75320.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:44:37,476 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75331.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 16:44:41,109 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-26 16:44:44,944 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.704e+01 1.586e+02 1.985e+02 2.275e+02 3.825e+02, threshold=3.970e+02, percent-clipped=2.0 2023-03-26 16:44:57,417 INFO [finetune.py:976] (3/7) Epoch 14, batch 900, loss[loss=0.1922, simple_loss=0.2559, pruned_loss=0.06423, over 4922.00 frames. ], tot_loss[loss=0.1871, simple_loss=0.2545, pruned_loss=0.05991, over 947092.75 frames. ], batch size: 38, lr: 3.56e-03, grad_scale: 16.0 2023-03-26 16:45:44,929 INFO [finetune.py:976] (3/7) Epoch 14, batch 950, loss[loss=0.2072, simple_loss=0.267, pruned_loss=0.07366, over 4830.00 frames. ], tot_loss[loss=0.1863, simple_loss=0.2529, pruned_loss=0.05983, over 950108.45 frames. ], batch size: 40, lr: 3.56e-03, grad_scale: 16.0 2023-03-26 16:46:02,412 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.0563, 2.8180, 2.3568, 1.5178, 2.5048, 2.4553, 2.1701, 2.5592], device='cuda:3'), covar=tensor([0.0706, 0.0598, 0.1203, 0.1720, 0.1258, 0.1567, 0.1814, 0.0813], device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0194, 0.0197, 0.0183, 0.0211, 0.0206, 0.0221, 0.0194], 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') 2023-03-26 16:46:05,783 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.942e+01 1.566e+02 1.871e+02 2.243e+02 4.539e+02, threshold=3.743e+02, percent-clipped=1.0 2023-03-26 16:46:18,857 INFO [finetune.py:976] (3/7) Epoch 14, batch 1000, loss[loss=0.1662, simple_loss=0.2335, pruned_loss=0.04948, over 4804.00 frames. ], tot_loss[loss=0.1881, simple_loss=0.2546, pruned_loss=0.0608, over 950214.40 frames. ], batch size: 25, lr: 3.56e-03, grad_scale: 16.0 2023-03-26 16:47:07,212 INFO [finetune.py:976] (3/7) Epoch 14, batch 1050, loss[loss=0.1448, simple_loss=0.2216, pruned_loss=0.03397, over 4753.00 frames. ], tot_loss[loss=0.188, simple_loss=0.2557, pruned_loss=0.06019, over 950131.98 frames. ], batch size: 27, lr: 3.56e-03, grad_scale: 16.0 2023-03-26 16:47:29,024 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.01 vs. limit=5.0 2023-03-26 16:47:31,085 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.109e+01 1.606e+02 2.003e+02 2.356e+02 8.983e+02, threshold=4.007e+02, percent-clipped=2.0 2023-03-26 16:47:34,731 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4244, 2.3382, 1.8347, 2.5250, 2.3732, 2.1307, 2.9007, 2.4658], device='cuda:3'), covar=tensor([0.1357, 0.2449, 0.3212, 0.2795, 0.2632, 0.1614, 0.3312, 0.1859], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0188, 0.0234, 0.0255, 0.0244, 0.0200, 0.0214, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 16:47:44,012 INFO [finetune.py:976] (3/7) Epoch 14, batch 1100, loss[loss=0.2019, simple_loss=0.237, pruned_loss=0.0834, over 4417.00 frames. ], tot_loss[loss=0.1903, simple_loss=0.2579, pruned_loss=0.06137, over 951333.57 frames. ], batch size: 19, lr: 3.56e-03, grad_scale: 16.0 2023-03-26 16:47:47,113 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75565.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:47:57,021 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.37 vs. limit=5.0 2023-03-26 16:48:00,114 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75584.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:48:02,929 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-26 16:48:18,069 INFO [finetune.py:976] (3/7) Epoch 14, batch 1150, loss[loss=0.1736, simple_loss=0.2302, pruned_loss=0.05854, over 4706.00 frames. ], tot_loss[loss=0.1915, simple_loss=0.2597, pruned_loss=0.06162, over 953431.10 frames. ], batch size: 23, lr: 3.56e-03, grad_scale: 16.0 2023-03-26 16:48:19,337 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=75613.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:48:24,070 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75620.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:48:28,193 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75626.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 16:48:38,761 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.806e+01 1.586e+02 1.986e+02 2.337e+02 5.787e+02, threshold=3.972e+02, percent-clipped=2.0 2023-03-26 16:48:51,184 INFO [finetune.py:976] (3/7) Epoch 14, batch 1200, loss[loss=0.1634, simple_loss=0.2329, pruned_loss=0.04691, over 4676.00 frames. ], tot_loss[loss=0.1905, simple_loss=0.2583, pruned_loss=0.0614, over 951170.28 frames. ], batch size: 23, lr: 3.56e-03, grad_scale: 16.0 2023-03-26 16:48:54,174 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8929, 1.7827, 1.6521, 2.0567, 2.2420, 2.0658, 1.5864, 1.5259], device='cuda:3'), covar=tensor([0.2254, 0.1981, 0.2005, 0.1651, 0.1802, 0.1108, 0.2513, 0.2067], device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0206, 0.0209, 0.0189, 0.0239, 0.0183, 0.0212, 0.0197], 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') 2023-03-26 16:48:56,399 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=75668.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:48:59,478 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7887, 1.7047, 1.6128, 1.7385, 1.1773, 3.5511, 1.4059, 1.8860], device='cuda:3'), covar=tensor([0.2925, 0.2266, 0.1991, 0.2099, 0.1680, 0.0183, 0.2453, 0.1250], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0120, 0.0123, 0.0115, 0.0097, 0.0097, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 16:49:24,700 INFO [finetune.py:976] (3/7) Epoch 14, batch 1250, loss[loss=0.1985, simple_loss=0.2573, pruned_loss=0.06987, over 4802.00 frames. ], tot_loss[loss=0.1877, simple_loss=0.2553, pruned_loss=0.06002, over 953994.29 frames. ], batch size: 51, lr: 3.56e-03, grad_scale: 16.0 2023-03-26 16:49:45,240 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.316e+01 1.497e+02 1.829e+02 2.293e+02 4.240e+02, threshold=3.659e+02, percent-clipped=2.0 2023-03-26 16:49:57,807 INFO [finetune.py:976] (3/7) Epoch 14, batch 1300, loss[loss=0.1751, simple_loss=0.248, pruned_loss=0.05105, over 4828.00 frames. ], tot_loss[loss=0.1861, simple_loss=0.2531, pruned_loss=0.05957, over 955894.10 frames. ], batch size: 39, lr: 3.56e-03, grad_scale: 16.0 2023-03-26 16:50:17,888 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0983, 1.9671, 2.1163, 1.5171, 1.9859, 2.2900, 2.1385, 1.6943], device='cuda:3'), covar=tensor([0.0590, 0.0655, 0.0668, 0.0911, 0.0681, 0.0597, 0.0565, 0.1112], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0132, 0.0141, 0.0123, 0.0123, 0.0141, 0.0140, 0.0162], 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') 2023-03-26 16:50:23,255 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7244, 1.4768, 2.1856, 3.1846, 2.2103, 2.2715, 0.9621, 2.5252], device='cuda:3'), covar=tensor([0.1604, 0.1430, 0.1192, 0.0528, 0.0752, 0.2116, 0.1808, 0.0539], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0116, 0.0134, 0.0165, 0.0101, 0.0139, 0.0127, 0.0103], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 16:50:31,734 INFO [finetune.py:976] (3/7) Epoch 14, batch 1350, loss[loss=0.1688, simple_loss=0.249, pruned_loss=0.04435, over 4794.00 frames. ], tot_loss[loss=0.186, simple_loss=0.253, pruned_loss=0.05949, over 953395.54 frames. ], batch size: 29, lr: 3.56e-03, grad_scale: 16.0 2023-03-26 16:51:07,732 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.659e+01 1.636e+02 1.953e+02 2.256e+02 6.748e+02, threshold=3.906e+02, percent-clipped=1.0 2023-03-26 16:51:19,723 INFO [finetune.py:976] (3/7) Epoch 14, batch 1400, loss[loss=0.2382, simple_loss=0.3046, pruned_loss=0.08595, over 4818.00 frames. ], tot_loss[loss=0.1879, simple_loss=0.2554, pruned_loss=0.06018, over 951543.83 frames. ], batch size: 40, lr: 3.56e-03, grad_scale: 16.0 2023-03-26 16:51:35,787 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75884.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:51:53,579 INFO [finetune.py:976] (3/7) Epoch 14, batch 1450, loss[loss=0.1519, simple_loss=0.2305, pruned_loss=0.03662, over 4835.00 frames. ], tot_loss[loss=0.1875, simple_loss=0.2555, pruned_loss=0.05977, over 949414.93 frames. ], batch size: 47, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 16:52:08,068 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75926.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 16:52:17,402 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=75932.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:52:27,769 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.094e+02 1.664e+02 1.939e+02 2.609e+02 1.085e+03, threshold=3.877e+02, percent-clipped=5.0 2023-03-26 16:52:27,929 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5945, 1.4478, 1.3160, 1.5823, 2.0968, 1.5733, 1.2346, 1.3044], device='cuda:3'), covar=tensor([0.2306, 0.2209, 0.2058, 0.1792, 0.1454, 0.1360, 0.2427, 0.2027], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0207, 0.0210, 0.0190, 0.0241, 0.0184, 0.0214, 0.0199], 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') 2023-03-26 16:52:31,943 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7329, 1.4746, 1.0312, 0.2026, 1.3022, 1.4965, 1.4117, 1.4918], device='cuda:3'), covar=tensor([0.0867, 0.0814, 0.1380, 0.1985, 0.1526, 0.2264, 0.2331, 0.0878], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0195, 0.0198, 0.0183, 0.0211, 0.0207, 0.0220, 0.0195], 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') 2023-03-26 16:52:44,458 INFO [finetune.py:976] (3/7) Epoch 14, batch 1500, loss[loss=0.1615, simple_loss=0.2351, pruned_loss=0.04394, over 4696.00 frames. ], tot_loss[loss=0.1897, simple_loss=0.2578, pruned_loss=0.06084, over 951506.77 frames. ], batch size: 23, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 16:52:52,971 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=75974.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:53:19,428 INFO [finetune.py:976] (3/7) Epoch 14, batch 1550, loss[loss=0.1608, simple_loss=0.2303, pruned_loss=0.04569, over 4911.00 frames. ], tot_loss[loss=0.188, simple_loss=0.2563, pruned_loss=0.05991, over 952494.42 frames. ], batch size: 37, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 16:53:24,672 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.02 vs. limit=5.0 2023-03-26 16:53:40,208 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.569e+01 1.489e+02 1.761e+02 2.263e+02 3.823e+02, threshold=3.522e+02, percent-clipped=0.0 2023-03-26 16:53:53,247 INFO [finetune.py:976] (3/7) Epoch 14, batch 1600, loss[loss=0.1905, simple_loss=0.2553, pruned_loss=0.0628, over 4843.00 frames. ], tot_loss[loss=0.1875, simple_loss=0.2552, pruned_loss=0.05995, over 953813.38 frames. ], batch size: 47, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 16:54:26,637 INFO [finetune.py:976] (3/7) Epoch 14, batch 1650, loss[loss=0.1885, simple_loss=0.2452, pruned_loss=0.06586, over 4868.00 frames. ], tot_loss[loss=0.1843, simple_loss=0.2516, pruned_loss=0.05856, over 954502.66 frames. ], batch size: 31, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 16:54:47,811 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.271e+01 1.580e+02 1.872e+02 2.187e+02 4.946e+02, threshold=3.744e+02, percent-clipped=3.0 2023-03-26 16:55:00,266 INFO [finetune.py:976] (3/7) Epoch 14, batch 1700, loss[loss=0.2132, simple_loss=0.2679, pruned_loss=0.07925, over 4808.00 frames. ], tot_loss[loss=0.1831, simple_loss=0.2499, pruned_loss=0.0581, over 956840.46 frames. ], batch size: 40, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 16:55:34,230 INFO [finetune.py:976] (3/7) Epoch 14, batch 1750, loss[loss=0.234, simple_loss=0.2913, pruned_loss=0.0884, over 4897.00 frames. ], tot_loss[loss=0.1852, simple_loss=0.252, pruned_loss=0.05918, over 957562.33 frames. ], batch size: 32, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 16:55:55,253 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.088e+02 1.620e+02 1.973e+02 2.349e+02 4.562e+02, threshold=3.945e+02, percent-clipped=4.0 2023-03-26 16:56:17,741 INFO [finetune.py:976] (3/7) Epoch 14, batch 1800, loss[loss=0.2055, simple_loss=0.2819, pruned_loss=0.06458, over 4802.00 frames. ], tot_loss[loss=0.1876, simple_loss=0.2553, pruned_loss=0.05998, over 955727.95 frames. ], batch size: 41, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 16:56:27,629 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76269.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 16:56:47,386 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76294.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:56:58,600 INFO [finetune.py:976] (3/7) Epoch 14, batch 1850, loss[loss=0.1857, simple_loss=0.255, pruned_loss=0.05823, over 4891.00 frames. ], tot_loss[loss=0.1888, simple_loss=0.2567, pruned_loss=0.06046, over 955052.73 frames. ], batch size: 43, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 16:57:07,113 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76323.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 16:57:07,806 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.66 vs. limit=2.0 2023-03-26 16:57:11,395 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76330.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 16:57:19,099 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.107e+02 1.514e+02 1.907e+02 2.299e+02 3.483e+02, threshold=3.815e+02, percent-clipped=0.0 2023-03-26 16:57:20,087 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 2023-03-26 16:57:35,315 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76355.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 16:57:38,873 INFO [finetune.py:976] (3/7) Epoch 14, batch 1900, loss[loss=0.1825, simple_loss=0.2487, pruned_loss=0.0582, over 4790.00 frames. ], tot_loss[loss=0.1887, simple_loss=0.257, pruned_loss=0.06017, over 953973.22 frames. ], batch size: 25, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 16:57:57,151 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76384.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 16:58:15,526 INFO [finetune.py:976] (3/7) Epoch 14, batch 1950, loss[loss=0.1339, simple_loss=0.21, pruned_loss=0.02892, over 4798.00 frames. ], tot_loss[loss=0.1875, simple_loss=0.2558, pruned_loss=0.05963, over 951374.74 frames. ], batch size: 25, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 16:58:35,771 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.938e+01 1.459e+02 1.786e+02 2.082e+02 3.715e+02, threshold=3.572e+02, percent-clipped=0.0 2023-03-26 16:58:49,141 INFO [finetune.py:976] (3/7) Epoch 14, batch 2000, loss[loss=0.1733, simple_loss=0.2392, pruned_loss=0.05372, over 4900.00 frames. ], tot_loss[loss=0.1864, simple_loss=0.2538, pruned_loss=0.05953, over 952914.30 frames. ], batch size: 32, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 16:58:49,232 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.2777, 3.7413, 3.9330, 4.0972, 4.0931, 3.8002, 4.3974, 1.3955], device='cuda:3'), covar=tensor([0.0778, 0.0810, 0.0795, 0.0978, 0.1099, 0.1422, 0.0635, 0.5308], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0244, 0.0277, 0.0292, 0.0332, 0.0281, 0.0302, 0.0297], 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') 2023-03-26 16:59:22,661 INFO [finetune.py:976] (3/7) Epoch 14, batch 2050, loss[loss=0.1987, simple_loss=0.2603, pruned_loss=0.06859, over 4906.00 frames. ], tot_loss[loss=0.1841, simple_loss=0.2508, pruned_loss=0.0587, over 953346.24 frames. ], batch size: 37, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 16:59:42,964 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.833e+01 1.460e+02 1.798e+02 2.157e+02 5.136e+02, threshold=3.595e+02, percent-clipped=3.0 2023-03-26 16:59:56,038 INFO [finetune.py:976] (3/7) Epoch 14, batch 2100, loss[loss=0.1551, simple_loss=0.2295, pruned_loss=0.04037, over 4827.00 frames. ], tot_loss[loss=0.1837, simple_loss=0.2504, pruned_loss=0.05851, over 953983.72 frames. ], batch size: 25, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 17:00:24,163 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-26 17:00:29,580 INFO [finetune.py:976] (3/7) Epoch 14, batch 2150, loss[loss=0.2549, simple_loss=0.3103, pruned_loss=0.09981, over 4806.00 frames. ], tot_loss[loss=0.1876, simple_loss=0.2551, pruned_loss=0.06009, over 955755.99 frames. ], batch size: 45, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 17:00:38,762 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76625.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 17:00:50,409 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.070e+02 1.680e+02 1.855e+02 2.274e+02 3.771e+02, threshold=3.710e+02, percent-clipped=2.0 2023-03-26 17:00:55,369 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76650.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:01:02,483 INFO [finetune.py:976] (3/7) Epoch 14, batch 2200, loss[loss=0.2079, simple_loss=0.2526, pruned_loss=0.08157, over 4701.00 frames. ], tot_loss[loss=0.1897, simple_loss=0.2575, pruned_loss=0.06095, over 954866.61 frames. ], batch size: 23, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 17:01:21,648 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76679.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 17:01:25,753 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9372, 1.7331, 2.2910, 1.6989, 2.1375, 2.1696, 1.7654, 2.3448], device='cuda:3'), covar=tensor([0.1317, 0.1791, 0.1233, 0.1668, 0.0656, 0.1247, 0.2223, 0.0676], device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0206, 0.0193, 0.0192, 0.0178, 0.0216, 0.0219, 0.0200], 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') 2023-03-26 17:01:55,186 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2041, 1.9525, 2.0683, 0.8530, 2.2745, 2.5214, 2.0753, 1.8650], device='cuda:3'), covar=tensor([0.1058, 0.0853, 0.0449, 0.0825, 0.0469, 0.0687, 0.0515, 0.0718], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0152, 0.0121, 0.0129, 0.0130, 0.0126, 0.0141, 0.0145], device='cuda:3'), out_proj_covar=tensor([9.2774e-05, 1.1128e-04, 8.7103e-05, 9.2769e-05, 9.1974e-05, 9.1207e-05, 1.0244e-04, 1.0511e-04], device='cuda:3') 2023-03-26 17:01:57,530 INFO [finetune.py:976] (3/7) Epoch 14, batch 2250, loss[loss=0.1944, simple_loss=0.2574, pruned_loss=0.06572, over 4860.00 frames. ], tot_loss[loss=0.1917, simple_loss=0.2598, pruned_loss=0.06182, over 955513.44 frames. ], batch size: 31, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 17:02:02,150 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7762, 1.7096, 1.5033, 1.3887, 1.8693, 1.5613, 1.8120, 1.7998], device='cuda:3'), covar=tensor([0.1417, 0.1996, 0.2966, 0.2464, 0.2575, 0.1702, 0.2830, 0.1753], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0188, 0.0234, 0.0254, 0.0245, 0.0199, 0.0213, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 17:02:18,736 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.028e+02 1.580e+02 1.848e+02 2.151e+02 3.368e+02, threshold=3.695e+02, percent-clipped=0.0 2023-03-26 17:02:31,273 INFO [finetune.py:976] (3/7) Epoch 14, batch 2300, loss[loss=0.1503, simple_loss=0.2146, pruned_loss=0.043, over 4190.00 frames. ], tot_loss[loss=0.1906, simple_loss=0.259, pruned_loss=0.06108, over 955988.67 frames. ], batch size: 18, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 17:02:42,052 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-03-26 17:02:43,762 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9245, 4.3521, 4.1528, 2.6452, 4.4567, 3.3257, 0.7927, 3.0141], device='cuda:3'), covar=tensor([0.2510, 0.1876, 0.1273, 0.2602, 0.0732, 0.0906, 0.4466, 0.1332], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0172, 0.0159, 0.0127, 0.0155, 0.0122, 0.0144, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 17:03:06,756 INFO [finetune.py:976] (3/7) Epoch 14, batch 2350, loss[loss=0.1577, simple_loss=0.2207, pruned_loss=0.04732, over 4927.00 frames. ], tot_loss[loss=0.1886, simple_loss=0.2567, pruned_loss=0.06028, over 957917.12 frames. ], batch size: 38, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 17:03:25,943 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76839.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:03:28,178 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.075e+02 1.531e+02 1.863e+02 2.217e+02 4.521e+02, threshold=3.725e+02, percent-clipped=2.0 2023-03-26 17:03:40,659 INFO [finetune.py:976] (3/7) Epoch 14, batch 2400, loss[loss=0.1568, simple_loss=0.2179, pruned_loss=0.0478, over 4214.00 frames. ], tot_loss[loss=0.1864, simple_loss=0.254, pruned_loss=0.05942, over 958093.45 frames. ], batch size: 65, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 17:04:05,231 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.0639, 2.7179, 2.4771, 1.3624, 2.6944, 2.2356, 2.0726, 2.4205], device='cuda:3'), covar=tensor([0.1092, 0.0828, 0.1808, 0.2197, 0.1698, 0.2105, 0.2060, 0.1202], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0195, 0.0198, 0.0184, 0.0213, 0.0207, 0.0222, 0.0196], 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') 2023-03-26 17:04:05,243 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2198, 1.8952, 1.9875, 1.0340, 2.1725, 2.3579, 2.0679, 1.8609], device='cuda:3'), covar=tensor([0.0973, 0.0804, 0.0574, 0.0678, 0.0547, 0.0594, 0.0496, 0.0723], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0154, 0.0122, 0.0130, 0.0131, 0.0128, 0.0142, 0.0146], device='cuda:3'), out_proj_covar=tensor([9.3960e-05, 1.1234e-04, 8.8165e-05, 9.3582e-05, 9.3254e-05, 9.2409e-05, 1.0325e-04, 1.0580e-04], device='cuda:3') 2023-03-26 17:04:06,940 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76900.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:04:14,025 INFO [finetune.py:976] (3/7) Epoch 14, batch 2450, loss[loss=0.1672, simple_loss=0.23, pruned_loss=0.05215, over 4922.00 frames. ], tot_loss[loss=0.1844, simple_loss=0.2512, pruned_loss=0.05882, over 956736.15 frames. ], batch size: 37, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 17:04:15,952 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4992, 1.3417, 1.1247, 1.2711, 1.7466, 1.5781, 1.5638, 1.1959], device='cuda:3'), covar=tensor([0.0273, 0.0333, 0.0756, 0.0348, 0.0212, 0.0496, 0.0253, 0.0410], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0109, 0.0141, 0.0113, 0.0100, 0.0106, 0.0096, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.3034e-05, 8.4260e-05, 1.1140e-04, 8.7875e-05, 7.8307e-05, 7.8292e-05, 7.2585e-05, 8.2428e-05], device='cuda:3') 2023-03-26 17:04:23,087 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76925.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 17:04:34,661 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.650e+01 1.627e+02 1.914e+02 2.448e+02 4.488e+02, threshold=3.829e+02, percent-clipped=3.0 2023-03-26 17:04:40,094 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76950.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:04:47,674 INFO [finetune.py:976] (3/7) Epoch 14, batch 2500, loss[loss=0.2115, simple_loss=0.2725, pruned_loss=0.07522, over 4931.00 frames. ], tot_loss[loss=0.1883, simple_loss=0.2542, pruned_loss=0.06116, over 953788.64 frames. ], batch size: 33, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 17:04:55,524 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=76973.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 17:04:59,663 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76979.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 17:05:05,916 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-03-26 17:05:12,271 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=76998.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:05:21,625 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-26 17:05:21,742 INFO [finetune.py:976] (3/7) Epoch 14, batch 2550, loss[loss=0.2373, simple_loss=0.3005, pruned_loss=0.087, over 4904.00 frames. ], tot_loss[loss=0.191, simple_loss=0.2578, pruned_loss=0.06213, over 954574.20 frames. ], batch size: 36, lr: 3.55e-03, grad_scale: 16.0 2023-03-26 17:05:32,005 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=77027.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 17:05:42,423 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.113e+02 1.599e+02 1.863e+02 2.531e+02 5.028e+02, threshold=3.725e+02, percent-clipped=2.0 2023-03-26 17:05:43,154 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5961, 1.4782, 1.4364, 1.5382, 1.0229, 3.2480, 1.2271, 1.5899], device='cuda:3'), covar=tensor([0.3315, 0.2486, 0.2168, 0.2356, 0.1847, 0.0204, 0.2589, 0.1305], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0116, 0.0121, 0.0124, 0.0115, 0.0097, 0.0097, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 17:05:43,772 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.3837, 1.2175, 1.3908, 0.6695, 1.3991, 1.6556, 1.5876, 1.2927], device='cuda:3'), covar=tensor([0.0882, 0.0901, 0.0543, 0.0577, 0.0483, 0.0566, 0.0416, 0.0755], device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0154, 0.0122, 0.0130, 0.0131, 0.0128, 0.0142, 0.0146], device='cuda:3'), out_proj_covar=tensor([9.3893e-05, 1.1209e-04, 8.7921e-05, 9.3593e-05, 9.2919e-05, 9.2523e-05, 1.0312e-04, 1.0582e-04], device='cuda:3') 2023-03-26 17:05:55,386 INFO [finetune.py:976] (3/7) Epoch 14, batch 2600, loss[loss=0.2194, simple_loss=0.2932, pruned_loss=0.0728, over 4812.00 frames. ], tot_loss[loss=0.1916, simple_loss=0.259, pruned_loss=0.06211, over 955751.92 frames. ], batch size: 39, lr: 3.55e-03, grad_scale: 32.0 2023-03-26 17:06:18,038 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77095.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:06:30,487 INFO [finetune.py:976] (3/7) Epoch 14, batch 2650, loss[loss=0.1506, simple_loss=0.2372, pruned_loss=0.03195, over 4726.00 frames. ], tot_loss[loss=0.1915, simple_loss=0.2593, pruned_loss=0.06189, over 953700.25 frames. ], batch size: 54, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:07:08,843 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.161e+02 1.606e+02 1.948e+02 2.371e+02 3.624e+02, threshold=3.895e+02, percent-clipped=0.0 2023-03-26 17:07:15,170 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-03-26 17:07:22,915 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77156.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 17:07:26,328 INFO [finetune.py:976] (3/7) Epoch 14, batch 2700, loss[loss=0.2011, simple_loss=0.2681, pruned_loss=0.06712, over 4800.00 frames. ], tot_loss[loss=0.1911, simple_loss=0.259, pruned_loss=0.06161, over 951251.94 frames. ], batch size: 41, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:07:45,605 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3836, 2.1730, 2.0490, 2.4401, 2.8034, 2.2532, 2.0334, 1.7815], device='cuda:3'), covar=tensor([0.2197, 0.2055, 0.1928, 0.1528, 0.1829, 0.1178, 0.2273, 0.1906], device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0207, 0.0211, 0.0190, 0.0241, 0.0184, 0.0214, 0.0200], 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') 2023-03-26 17:07:57,717 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77195.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:08:03,470 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-26 17:08:07,956 INFO [finetune.py:976] (3/7) Epoch 14, batch 2750, loss[loss=0.1745, simple_loss=0.2354, pruned_loss=0.05679, over 4718.00 frames. ], tot_loss[loss=0.188, simple_loss=0.2556, pruned_loss=0.06018, over 953474.68 frames. ], batch size: 23, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:08:08,075 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77211.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:08:10,330 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-03-26 17:08:18,248 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7070, 2.4893, 2.0798, 1.0232, 2.2840, 2.0720, 1.8895, 2.1811], device='cuda:3'), covar=tensor([0.0703, 0.0796, 0.1441, 0.2039, 0.1373, 0.1841, 0.1939, 0.0985], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0198, 0.0201, 0.0186, 0.0216, 0.0210, 0.0225, 0.0198], 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') 2023-03-26 17:08:27,883 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.8830, 3.4233, 3.5518, 3.7425, 3.6478, 3.3700, 3.9726, 1.2237], device='cuda:3'), covar=tensor([0.0874, 0.0846, 0.0955, 0.1041, 0.1384, 0.1834, 0.0830, 0.5168], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0244, 0.0278, 0.0294, 0.0333, 0.0285, 0.0303, 0.0299], 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') 2023-03-26 17:08:28,394 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.127e+02 1.514e+02 1.871e+02 2.163e+02 3.576e+02, threshold=3.742e+02, percent-clipped=0.0 2023-03-26 17:08:40,953 INFO [finetune.py:976] (3/7) Epoch 14, batch 2800, loss[loss=0.1843, simple_loss=0.2508, pruned_loss=0.05891, over 4830.00 frames. ], tot_loss[loss=0.1843, simple_loss=0.2517, pruned_loss=0.05851, over 954367.06 frames. ], batch size: 40, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:08:45,231 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2989, 2.3624, 1.8032, 2.3927, 2.2390, 1.9349, 2.7744, 2.3810], device='cuda:3'), covar=tensor([0.1363, 0.2450, 0.3023, 0.2956, 0.2782, 0.1696, 0.3977, 0.1835], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0188, 0.0233, 0.0253, 0.0244, 0.0199, 0.0213, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 17:08:48,264 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77272.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:08:48,850 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77273.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:08:52,192 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77277.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:09:14,626 INFO [finetune.py:976] (3/7) Epoch 14, batch 2850, loss[loss=0.175, simple_loss=0.2406, pruned_loss=0.05468, over 4789.00 frames. ], tot_loss[loss=0.1832, simple_loss=0.2503, pruned_loss=0.05799, over 955760.20 frames. ], batch size: 26, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:09:14,952 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.97 vs. limit=5.0 2023-03-26 17:09:19,735 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-26 17:09:29,762 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77334.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:09:32,160 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77338.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:09:34,910 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.028e+02 1.563e+02 1.884e+02 2.320e+02 5.201e+02, threshold=3.768e+02, percent-clipped=2.0 2023-03-26 17:09:42,242 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-03-26 17:09:47,962 INFO [finetune.py:976] (3/7) Epoch 14, batch 2900, loss[loss=0.211, simple_loss=0.2728, pruned_loss=0.07457, over 4940.00 frames. ], tot_loss[loss=0.1855, simple_loss=0.2529, pruned_loss=0.05908, over 954976.65 frames. ], batch size: 33, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:10:21,778 INFO [finetune.py:976] (3/7) Epoch 14, batch 2950, loss[loss=0.1641, simple_loss=0.2298, pruned_loss=0.0492, over 4807.00 frames. ], tot_loss[loss=0.1882, simple_loss=0.2563, pruned_loss=0.06004, over 954617.34 frames. ], batch size: 51, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:10:41,986 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.681e+01 1.651e+02 1.924e+02 2.203e+02 4.754e+02, threshold=3.848e+02, percent-clipped=2.0 2023-03-26 17:10:48,021 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77451.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 17:10:54,978 INFO [finetune.py:976] (3/7) Epoch 14, batch 3000, loss[loss=0.229, simple_loss=0.2879, pruned_loss=0.085, over 4738.00 frames. ], tot_loss[loss=0.1898, simple_loss=0.2583, pruned_loss=0.06061, over 955746.13 frames. ], batch size: 54, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:10:54,978 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 17:11:00,806 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7946, 1.7848, 1.8791, 1.2168, 1.9221, 1.9526, 1.9288, 1.6387], device='cuda:3'), covar=tensor([0.0584, 0.0661, 0.0658, 0.0908, 0.0820, 0.0659, 0.0606, 0.1079], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0131, 0.0139, 0.0122, 0.0122, 0.0139, 0.0139, 0.0160], 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') 2023-03-26 17:11:04,294 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9562, 1.1553, 2.0062, 1.8810, 1.7785, 1.6560, 1.7327, 1.8694], device='cuda:3'), covar=tensor([0.3894, 0.4398, 0.3884, 0.3938, 0.5661, 0.4061, 0.5035, 0.3423], device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0237, 0.0256, 0.0265, 0.0263, 0.0236, 0.0276, 0.0235], 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') 2023-03-26 17:11:09,361 INFO [finetune.py:1010] (3/7) Epoch 14, validation: loss=0.1563, simple_loss=0.2268, pruned_loss=0.04293, over 2265189.00 frames. 2023-03-26 17:11:09,361 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6434MB 2023-03-26 17:11:10,414 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.14 vs. limit=2.0 2023-03-26 17:11:34,070 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77495.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:11:44,251 INFO [finetune.py:976] (3/7) Epoch 14, batch 3050, loss[loss=0.1968, simple_loss=0.2642, pruned_loss=0.06473, over 4818.00 frames. ], tot_loss[loss=0.1891, simple_loss=0.2582, pruned_loss=0.05998, over 957096.60 frames. ], batch size: 47, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:12:02,294 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-03-26 17:12:13,224 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.117e+02 1.567e+02 1.800e+02 2.244e+02 5.193e+02, threshold=3.600e+02, percent-clipped=2.0 2023-03-26 17:12:13,781 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-26 17:12:13,931 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=77543.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:12:29,445 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1688, 2.0113, 1.6859, 2.0579, 1.8548, 1.8329, 1.9240, 2.6888], device='cuda:3'), covar=tensor([0.3780, 0.4797, 0.3533, 0.4252, 0.4395, 0.2545, 0.4246, 0.1693], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0259, 0.0225, 0.0277, 0.0246, 0.0214, 0.0249, 0.0223], 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') 2023-03-26 17:12:35,684 INFO [finetune.py:976] (3/7) Epoch 14, batch 3100, loss[loss=0.1962, simple_loss=0.2657, pruned_loss=0.06337, over 4847.00 frames. ], tot_loss[loss=0.1876, simple_loss=0.2561, pruned_loss=0.05951, over 956743.71 frames. ], batch size: 49, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:12:39,909 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77567.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:13:08,170 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6907, 1.7127, 2.3355, 2.1704, 1.8704, 4.4421, 1.6112, 1.8929], device='cuda:3'), covar=tensor([0.0942, 0.1763, 0.0952, 0.0908, 0.1454, 0.0143, 0.1409, 0.1551], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0073, 0.0077, 0.0091, 0.0080, 0.0084, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 17:13:22,407 INFO [finetune.py:976] (3/7) Epoch 14, batch 3150, loss[loss=0.149, simple_loss=0.2157, pruned_loss=0.04113, over 4939.00 frames. ], tot_loss[loss=0.1854, simple_loss=0.253, pruned_loss=0.05889, over 954963.34 frames. ], batch size: 33, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:13:25,565 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77616.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:13:35,442 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77629.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:13:37,888 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77633.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:13:43,292 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.167e+02 1.643e+02 1.968e+02 2.398e+02 4.679e+02, threshold=3.936e+02, percent-clipped=3.0 2023-03-26 17:13:48,732 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9616, 2.6360, 3.3597, 4.6362, 3.4467, 3.2832, 2.1399, 3.9077], device='cuda:3'), covar=tensor([0.1322, 0.1211, 0.1090, 0.0514, 0.0600, 0.1252, 0.1420, 0.0437], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0116, 0.0133, 0.0164, 0.0100, 0.0137, 0.0125, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 17:13:56,367 INFO [finetune.py:976] (3/7) Epoch 14, batch 3200, loss[loss=0.2079, simple_loss=0.2693, pruned_loss=0.07329, over 4891.00 frames. ], tot_loss[loss=0.1849, simple_loss=0.2518, pruned_loss=0.05905, over 955060.63 frames. ], batch size: 36, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:14:06,663 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77677.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:14:06,676 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77677.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:14:29,512 INFO [finetune.py:976] (3/7) Epoch 14, batch 3250, loss[loss=0.2462, simple_loss=0.3082, pruned_loss=0.09212, over 4812.00 frames. ], tot_loss[loss=0.1863, simple_loss=0.2528, pruned_loss=0.05992, over 955258.72 frames. ], batch size: 38, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:14:36,698 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6842, 1.6937, 1.7871, 1.2315, 1.7435, 1.9059, 1.8640, 1.5160], device='cuda:3'), covar=tensor([0.0554, 0.0516, 0.0615, 0.0770, 0.0837, 0.0529, 0.0487, 0.0967], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0132, 0.0140, 0.0123, 0.0123, 0.0140, 0.0140, 0.0161], 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') 2023-03-26 17:14:47,322 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77738.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:14:49,602 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.063e+02 1.623e+02 1.885e+02 2.287e+02 7.301e+02, threshold=3.769e+02, percent-clipped=4.0 2023-03-26 17:14:55,592 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77751.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 17:15:02,069 INFO [finetune.py:976] (3/7) Epoch 14, batch 3300, loss[loss=0.1766, simple_loss=0.2454, pruned_loss=0.05391, over 4843.00 frames. ], tot_loss[loss=0.1901, simple_loss=0.2569, pruned_loss=0.06166, over 954118.35 frames. ], batch size: 25, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:15:14,146 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9453, 1.9998, 1.9888, 1.3245, 2.0211, 2.1422, 2.1092, 1.7613], device='cuda:3'), covar=tensor([0.0631, 0.0672, 0.0740, 0.0974, 0.0640, 0.0694, 0.0645, 0.1088], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0132, 0.0140, 0.0123, 0.0123, 0.0140, 0.0140, 0.0161], 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') 2023-03-26 17:15:27,675 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=77799.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:15:35,623 INFO [finetune.py:976] (3/7) Epoch 14, batch 3350, loss[loss=0.1649, simple_loss=0.2342, pruned_loss=0.04784, over 4761.00 frames. ], tot_loss[loss=0.1899, simple_loss=0.2573, pruned_loss=0.0613, over 953080.53 frames. ], batch size: 28, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:15:41,515 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6779, 4.4902, 4.1824, 2.1053, 4.4459, 3.3888, 0.6544, 2.9127], device='cuda:3'), covar=tensor([0.2651, 0.1475, 0.1198, 0.3058, 0.0773, 0.0800, 0.4542, 0.1411], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0172, 0.0158, 0.0127, 0.0156, 0.0121, 0.0144, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 17:15:57,263 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.144e+02 1.683e+02 1.958e+02 2.277e+02 5.309e+02, threshold=3.915e+02, percent-clipped=1.0 2023-03-26 17:16:09,330 INFO [finetune.py:976] (3/7) Epoch 14, batch 3400, loss[loss=0.2038, simple_loss=0.2736, pruned_loss=0.06702, over 4789.00 frames. ], tot_loss[loss=0.1901, simple_loss=0.2576, pruned_loss=0.06126, over 954153.98 frames. ], batch size: 29, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:16:18,552 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77867.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:16:34,153 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-03-26 17:16:51,418 INFO [finetune.py:976] (3/7) Epoch 14, batch 3450, loss[loss=0.1772, simple_loss=0.2498, pruned_loss=0.05229, over 4797.00 frames. ], tot_loss[loss=0.1892, simple_loss=0.2574, pruned_loss=0.06051, over 954544.88 frames. ], batch size: 51, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:16:53,842 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=77915.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:17:01,762 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-26 17:17:03,230 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77929.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:17:05,716 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77933.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:17:11,407 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.160e+01 1.485e+02 1.825e+02 2.093e+02 4.049e+02, threshold=3.650e+02, percent-clipped=1.0 2023-03-26 17:17:33,220 INFO [finetune.py:976] (3/7) Epoch 14, batch 3500, loss[loss=0.1956, simple_loss=0.2606, pruned_loss=0.06529, over 4799.00 frames. ], tot_loss[loss=0.1871, simple_loss=0.2551, pruned_loss=0.05954, over 954177.53 frames. ], batch size: 51, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:17:35,601 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4252, 2.3320, 2.0133, 1.0411, 2.2149, 1.9363, 1.7962, 2.1530], device='cuda:3'), covar=tensor([0.0852, 0.0694, 0.1460, 0.1817, 0.1210, 0.1948, 0.1842, 0.0866], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0195, 0.0200, 0.0184, 0.0214, 0.0208, 0.0223, 0.0196], 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') 2023-03-26 17:17:40,921 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77972.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:17:44,503 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=77977.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:17:46,959 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=77981.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:18:20,969 INFO [finetune.py:976] (3/7) Epoch 14, batch 3550, loss[loss=0.1608, simple_loss=0.2328, pruned_loss=0.04439, over 4903.00 frames. ], tot_loss[loss=0.1853, simple_loss=0.2528, pruned_loss=0.05889, over 954228.99 frames. ], batch size: 36, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:18:35,865 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78033.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:18:41,167 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.035e+02 1.505e+02 1.960e+02 2.472e+02 4.194e+02, threshold=3.920e+02, percent-clipped=4.0 2023-03-26 17:18:54,344 INFO [finetune.py:976] (3/7) Epoch 14, batch 3600, loss[loss=0.1878, simple_loss=0.2551, pruned_loss=0.06026, over 4793.00 frames. ], tot_loss[loss=0.1834, simple_loss=0.2501, pruned_loss=0.05831, over 952035.43 frames. ], batch size: 25, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:19:28,414 INFO [finetune.py:976] (3/7) Epoch 14, batch 3650, loss[loss=0.1984, simple_loss=0.2704, pruned_loss=0.06323, over 4859.00 frames. ], tot_loss[loss=0.1852, simple_loss=0.2524, pruned_loss=0.05906, over 954280.75 frames. ], batch size: 44, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:19:48,730 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.050e+02 1.711e+02 2.033e+02 2.406e+02 8.151e+02, threshold=4.067e+02, percent-clipped=4.0 2023-03-26 17:20:02,233 INFO [finetune.py:976] (3/7) Epoch 14, batch 3700, loss[loss=0.2591, simple_loss=0.3138, pruned_loss=0.1022, over 4801.00 frames. ], tot_loss[loss=0.1886, simple_loss=0.2563, pruned_loss=0.06049, over 951377.01 frames. ], batch size: 45, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:20:16,980 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.69 vs. limit=5.0 2023-03-26 17:20:35,986 INFO [finetune.py:976] (3/7) Epoch 14, batch 3750, loss[loss=0.2368, simple_loss=0.3087, pruned_loss=0.08245, over 4809.00 frames. ], tot_loss[loss=0.19, simple_loss=0.2581, pruned_loss=0.06093, over 952972.98 frames. ], batch size: 45, lr: 3.54e-03, grad_scale: 32.0 2023-03-26 17:20:49,439 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78232.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:20:55,808 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.186e+02 1.631e+02 1.949e+02 2.239e+02 4.423e+02, threshold=3.899e+02, percent-clipped=1.0 2023-03-26 17:20:56,000 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0 2023-03-26 17:21:03,992 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.70 vs. limit=5.0 2023-03-26 17:21:05,948 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1139, 1.9227, 1.7373, 2.1400, 2.6146, 2.1047, 1.8916, 1.5470], device='cuda:3'), covar=tensor([0.2074, 0.1969, 0.1817, 0.1581, 0.1773, 0.1139, 0.2242, 0.1881], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0210, 0.0212, 0.0193, 0.0244, 0.0186, 0.0218, 0.0201], 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') 2023-03-26 17:21:08,215 INFO [finetune.py:976] (3/7) Epoch 14, batch 3800, loss[loss=0.2199, simple_loss=0.2867, pruned_loss=0.07661, over 4814.00 frames. ], tot_loss[loss=0.1907, simple_loss=0.2593, pruned_loss=0.06102, over 953981.80 frames. ], batch size: 38, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:21:15,906 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78272.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:21:31,044 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78293.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 17:21:31,674 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3351, 2.3088, 1.8120, 0.9178, 2.0923, 1.8805, 1.7428, 2.0296], device='cuda:3'), covar=tensor([0.1038, 0.0687, 0.1516, 0.1956, 0.1265, 0.2011, 0.1923, 0.0902], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0195, 0.0199, 0.0183, 0.0213, 0.0207, 0.0222, 0.0196], 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') 2023-03-26 17:21:49,336 INFO [finetune.py:976] (3/7) Epoch 14, batch 3850, loss[loss=0.1976, simple_loss=0.2714, pruned_loss=0.06191, over 4802.00 frames. ], tot_loss[loss=0.1882, simple_loss=0.2569, pruned_loss=0.05972, over 952167.02 frames. ], batch size: 39, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:21:55,865 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=78320.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:22:03,827 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78333.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:22:10,189 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.042e+02 1.518e+02 1.784e+02 2.158e+02 4.566e+02, threshold=3.568e+02, percent-clipped=2.0 2023-03-26 17:22:21,568 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5824, 0.9849, 0.7569, 1.4368, 2.0049, 0.7269, 1.2528, 1.4468], device='cuda:3'), covar=tensor([0.1521, 0.2183, 0.1769, 0.1236, 0.1838, 0.1970, 0.1520, 0.1947], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0094, 0.0111, 0.0092, 0.0119, 0.0094, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 17:22:22,702 INFO [finetune.py:976] (3/7) Epoch 14, batch 3900, loss[loss=0.1812, simple_loss=0.254, pruned_loss=0.05416, over 4820.00 frames. ], tot_loss[loss=0.1857, simple_loss=0.2538, pruned_loss=0.05881, over 953620.51 frames. ], batch size: 30, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:22:45,627 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=78381.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:22:56,528 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6328, 1.5358, 1.5142, 1.6108, 0.9646, 2.8962, 1.0872, 1.5117], device='cuda:3'), covar=tensor([0.3425, 0.2557, 0.2163, 0.2396, 0.2021, 0.0255, 0.2711, 0.1294], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0115, 0.0121, 0.0124, 0.0116, 0.0098, 0.0097, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 17:23:09,952 INFO [finetune.py:976] (3/7) Epoch 14, batch 3950, loss[loss=0.2126, simple_loss=0.2666, pruned_loss=0.07926, over 4711.00 frames. ], tot_loss[loss=0.1826, simple_loss=0.2503, pruned_loss=0.05748, over 954625.78 frames. ], batch size: 54, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:23:27,597 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.2611, 3.7353, 3.9226, 4.1371, 4.0445, 3.8348, 4.3904, 1.3357], device='cuda:3'), covar=tensor([0.0705, 0.0842, 0.0777, 0.0941, 0.1052, 0.1394, 0.0675, 0.5327], device='cuda:3'), in_proj_covar=tensor([0.0343, 0.0239, 0.0272, 0.0290, 0.0329, 0.0280, 0.0297, 0.0295], 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') 2023-03-26 17:23:37,906 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.052e+02 1.587e+02 1.893e+02 2.259e+02 3.905e+02, threshold=3.786e+02, percent-clipped=1.0 2023-03-26 17:23:50,383 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78460.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:23:50,892 INFO [finetune.py:976] (3/7) Epoch 14, batch 4000, loss[loss=0.1721, simple_loss=0.2506, pruned_loss=0.04682, over 4800.00 frames. ], tot_loss[loss=0.1818, simple_loss=0.2493, pruned_loss=0.05712, over 954738.48 frames. ], batch size: 29, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:23:56,985 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6616, 1.5291, 1.4781, 1.6848, 0.9037, 3.5922, 1.4173, 1.8751], device='cuda:3'), covar=tensor([0.3409, 0.2447, 0.2158, 0.2302, 0.1984, 0.0176, 0.2612, 0.1215], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0116, 0.0121, 0.0124, 0.0116, 0.0098, 0.0097, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 17:24:02,135 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2094, 1.3010, 1.2817, 1.3591, 1.4348, 2.4604, 1.2829, 1.4127], device='cuda:3'), covar=tensor([0.0992, 0.1924, 0.1093, 0.0955, 0.1727, 0.0374, 0.1539, 0.1835], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0073, 0.0077, 0.0091, 0.0080, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 17:24:24,847 INFO [finetune.py:976] (3/7) Epoch 14, batch 4050, loss[loss=0.2068, simple_loss=0.282, pruned_loss=0.0658, over 4813.00 frames. ], tot_loss[loss=0.1849, simple_loss=0.2526, pruned_loss=0.05866, over 954497.24 frames. ], batch size: 51, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:24:31,533 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78521.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:24:45,460 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.285e+01 1.581e+02 1.919e+02 2.315e+02 3.488e+02, threshold=3.837e+02, percent-clipped=0.0 2023-03-26 17:24:54,867 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78556.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:24:57,774 INFO [finetune.py:976] (3/7) Epoch 14, batch 4100, loss[loss=0.1753, simple_loss=0.2381, pruned_loss=0.05628, over 4710.00 frames. ], tot_loss[loss=0.1875, simple_loss=0.2555, pruned_loss=0.05975, over 954091.95 frames. ], batch size: 23, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:25:01,384 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1357, 1.9827, 1.7549, 1.7601, 2.0710, 1.8766, 2.2535, 2.0858], device='cuda:3'), covar=tensor([0.1282, 0.1966, 0.2986, 0.2586, 0.2334, 0.1566, 0.2385, 0.1801], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0187, 0.0234, 0.0254, 0.0244, 0.0199, 0.0212, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 17:25:15,449 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-03-26 17:25:16,502 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78588.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 17:25:22,418 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-26 17:25:31,552 INFO [finetune.py:976] (3/7) Epoch 14, batch 4150, loss[loss=0.1958, simple_loss=0.2727, pruned_loss=0.05941, over 4821.00 frames. ], tot_loss[loss=0.1876, simple_loss=0.256, pruned_loss=0.05959, over 954133.55 frames. ], batch size: 55, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:25:35,311 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78617.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:25:52,376 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.171e+02 1.587e+02 1.869e+02 2.218e+02 3.242e+02, threshold=3.739e+02, percent-clipped=0.0 2023-03-26 17:26:04,875 INFO [finetune.py:976] (3/7) Epoch 14, batch 4200, loss[loss=0.1873, simple_loss=0.2616, pruned_loss=0.05652, over 4879.00 frames. ], tot_loss[loss=0.1869, simple_loss=0.2558, pruned_loss=0.05901, over 954753.83 frames. ], batch size: 35, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:26:06,301 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-26 17:26:37,998 INFO [finetune.py:976] (3/7) Epoch 14, batch 4250, loss[loss=0.1425, simple_loss=0.2131, pruned_loss=0.0359, over 4900.00 frames. ], tot_loss[loss=0.1856, simple_loss=0.2541, pruned_loss=0.05856, over 954950.31 frames. ], batch size: 37, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:27:05,886 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.090e+02 1.501e+02 1.722e+02 2.085e+02 5.543e+02, threshold=3.444e+02, percent-clipped=2.0 2023-03-26 17:27:21,260 INFO [finetune.py:976] (3/7) Epoch 14, batch 4300, loss[loss=0.1539, simple_loss=0.2266, pruned_loss=0.04062, over 4862.00 frames. ], tot_loss[loss=0.1833, simple_loss=0.2511, pruned_loss=0.05774, over 956059.26 frames. ], batch size: 31, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:27:59,986 INFO [finetune.py:976] (3/7) Epoch 14, batch 4350, loss[loss=0.1949, simple_loss=0.2614, pruned_loss=0.06423, over 4939.00 frames. ], tot_loss[loss=0.1815, simple_loss=0.2492, pruned_loss=0.05687, over 958822.16 frames. ], batch size: 33, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:28:06,791 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78816.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:28:34,427 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.179e+02 1.620e+02 1.919e+02 2.276e+02 4.946e+02, threshold=3.838e+02, percent-clipped=3.0 2023-03-26 17:28:54,136 INFO [finetune.py:976] (3/7) Epoch 14, batch 4400, loss[loss=0.2329, simple_loss=0.2933, pruned_loss=0.08623, over 4937.00 frames. ], tot_loss[loss=0.183, simple_loss=0.2504, pruned_loss=0.05783, over 957645.02 frames. ], batch size: 33, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:29:12,616 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78888.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:29:27,941 INFO [finetune.py:976] (3/7) Epoch 14, batch 4450, loss[loss=0.1669, simple_loss=0.2446, pruned_loss=0.04462, over 4744.00 frames. ], tot_loss[loss=0.186, simple_loss=0.2542, pruned_loss=0.05886, over 959335.25 frames. ], batch size: 27, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:29:28,616 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78912.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:29:44,126 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=78936.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:29:48,694 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.157e+02 1.623e+02 2.075e+02 2.454e+02 4.700e+02, threshold=4.150e+02, percent-clipped=3.0 2023-03-26 17:29:51,132 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-03-26 17:30:01,641 INFO [finetune.py:976] (3/7) Epoch 14, batch 4500, loss[loss=0.2035, simple_loss=0.2646, pruned_loss=0.07123, over 4140.00 frames. ], tot_loss[loss=0.1856, simple_loss=0.2541, pruned_loss=0.05851, over 955997.90 frames. ], batch size: 66, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:30:04,116 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78965.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:30:34,870 INFO [finetune.py:976] (3/7) Epoch 14, batch 4550, loss[loss=0.1933, simple_loss=0.2678, pruned_loss=0.05938, over 4901.00 frames. ], tot_loss[loss=0.1873, simple_loss=0.2558, pruned_loss=0.05942, over 957223.41 frames. ], batch size: 36, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:30:44,002 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79026.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:30:54,528 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.032e+02 1.627e+02 1.835e+02 2.182e+02 4.419e+02, threshold=3.671e+02, percent-clipped=1.0 2023-03-26 17:31:08,634 INFO [finetune.py:976] (3/7) Epoch 14, batch 4600, loss[loss=0.1803, simple_loss=0.2527, pruned_loss=0.05394, over 4883.00 frames. ], tot_loss[loss=0.1876, simple_loss=0.2566, pruned_loss=0.05928, over 958668.34 frames. ], batch size: 32, lr: 3.53e-03, grad_scale: 64.0 2023-03-26 17:31:42,443 INFO [finetune.py:976] (3/7) Epoch 14, batch 4650, loss[loss=0.1556, simple_loss=0.2301, pruned_loss=0.04052, over 4750.00 frames. ], tot_loss[loss=0.1864, simple_loss=0.2546, pruned_loss=0.05905, over 957560.24 frames. ], batch size: 26, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:31:45,566 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79116.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:32:02,946 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.138e+02 1.520e+02 1.886e+02 2.415e+02 4.094e+02, threshold=3.771e+02, percent-clipped=1.0 2023-03-26 17:32:18,240 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-03-26 17:32:22,759 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1407, 2.3620, 2.1761, 1.6937, 2.0723, 2.5383, 2.3559, 2.0444], device='cuda:3'), covar=tensor([0.0505, 0.0497, 0.0674, 0.0858, 0.1503, 0.0528, 0.0524, 0.0910], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0132, 0.0141, 0.0123, 0.0123, 0.0140, 0.0140, 0.0162], 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') 2023-03-26 17:32:24,493 INFO [finetune.py:976] (3/7) Epoch 14, batch 4700, loss[loss=0.1793, simple_loss=0.2366, pruned_loss=0.06097, over 4853.00 frames. ], tot_loss[loss=0.1832, simple_loss=0.2513, pruned_loss=0.05759, over 957923.49 frames. ], batch size: 44, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:32:26,311 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=79164.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:32:31,907 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3174, 2.1489, 1.5529, 0.7656, 1.8958, 1.8769, 1.6878, 1.8984], device='cuda:3'), covar=tensor([0.0814, 0.0695, 0.1538, 0.2025, 0.1220, 0.1992, 0.2105, 0.0916], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0195, 0.0200, 0.0184, 0.0214, 0.0207, 0.0222, 0.0197], 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') 2023-03-26 17:32:50,086 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-26 17:32:58,004 INFO [finetune.py:976] (3/7) Epoch 14, batch 4750, loss[loss=0.2095, simple_loss=0.2868, pruned_loss=0.06613, over 4860.00 frames. ], tot_loss[loss=0.1822, simple_loss=0.2498, pruned_loss=0.05732, over 958112.39 frames. ], batch size: 44, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:32:59,189 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79212.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:33:06,637 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1646, 2.1444, 1.6638, 2.2200, 2.0518, 1.8278, 2.5234, 2.2133], device='cuda:3'), covar=tensor([0.1406, 0.2081, 0.3215, 0.2632, 0.2676, 0.1757, 0.3078, 0.1703], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0186, 0.0234, 0.0254, 0.0243, 0.0199, 0.0212, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 17:33:32,019 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.558e+01 1.654e+02 1.996e+02 2.359e+02 6.861e+02, threshold=3.993e+02, percent-clipped=2.0 2023-03-26 17:33:51,147 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=79260.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:33:51,674 INFO [finetune.py:976] (3/7) Epoch 14, batch 4800, loss[loss=0.2215, simple_loss=0.2864, pruned_loss=0.07826, over 4769.00 frames. ], tot_loss[loss=0.1854, simple_loss=0.2529, pruned_loss=0.05892, over 955365.61 frames. ], batch size: 28, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:34:10,281 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79285.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 17:34:21,573 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9351, 1.7127, 1.5945, 1.6631, 2.1942, 2.1710, 1.7762, 1.5861], device='cuda:3'), covar=tensor([0.0274, 0.0320, 0.0517, 0.0339, 0.0216, 0.0418, 0.0300, 0.0400], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0109, 0.0142, 0.0114, 0.0101, 0.0107, 0.0097, 0.0109], device='cuda:3'), out_proj_covar=tensor([7.3271e-05, 8.4518e-05, 1.1243e-04, 8.8955e-05, 7.8835e-05, 7.9503e-05, 7.2902e-05, 8.3541e-05], device='cuda:3') 2023-03-26 17:34:27,365 INFO [finetune.py:976] (3/7) Epoch 14, batch 4850, loss[loss=0.182, simple_loss=0.2585, pruned_loss=0.05275, over 4811.00 frames. ], tot_loss[loss=0.1874, simple_loss=0.2552, pruned_loss=0.05978, over 954833.48 frames. ], batch size: 51, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:34:35,463 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79321.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:34:42,710 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79333.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:34:49,132 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.151e+02 1.639e+02 1.937e+02 2.312e+02 4.640e+02, threshold=3.873e+02, percent-clipped=1.0 2023-03-26 17:34:51,068 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79346.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 17:35:00,513 INFO [finetune.py:976] (3/7) Epoch 14, batch 4900, loss[loss=0.1779, simple_loss=0.252, pruned_loss=0.05187, over 4824.00 frames. ], tot_loss[loss=0.1883, simple_loss=0.2564, pruned_loss=0.06007, over 954573.65 frames. ], batch size: 33, lr: 3.53e-03, grad_scale: 32.0 2023-03-26 17:35:03,415 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79364.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:35:11,516 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79375.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 17:35:23,188 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79394.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:35:34,335 INFO [finetune.py:976] (3/7) Epoch 14, batch 4950, loss[loss=0.1872, simple_loss=0.2674, pruned_loss=0.05349, over 4758.00 frames. ], tot_loss[loss=0.1887, simple_loss=0.2574, pruned_loss=0.05996, over 955172.01 frames. ], batch size: 59, lr: 3.52e-03, grad_scale: 32.0 2023-03-26 17:35:45,274 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79425.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:35:51,956 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79436.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 17:35:55,996 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.049e+02 1.541e+02 1.877e+02 2.434e+02 3.585e+02, threshold=3.755e+02, percent-clipped=0.0 2023-03-26 17:35:57,429 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0 2023-03-26 17:36:07,907 INFO [finetune.py:976] (3/7) Epoch 14, batch 5000, loss[loss=0.1673, simple_loss=0.2261, pruned_loss=0.05421, over 4748.00 frames. ], tot_loss[loss=0.1873, simple_loss=0.2557, pruned_loss=0.05948, over 953790.03 frames. ], batch size: 26, lr: 3.52e-03, grad_scale: 32.0 2023-03-26 17:36:21,871 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9063, 1.0654, 1.8927, 1.8354, 1.6440, 1.5651, 1.7245, 1.7543], device='cuda:3'), covar=tensor([0.3764, 0.4099, 0.3345, 0.3517, 0.4586, 0.3624, 0.4249, 0.3094], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0240, 0.0257, 0.0267, 0.0265, 0.0239, 0.0279, 0.0236], 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') 2023-03-26 17:36:26,624 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.0642, 1.8147, 1.9707, 1.3423, 2.0463, 2.1213, 2.1351, 1.7633], device='cuda:3'), covar=tensor([0.0749, 0.0619, 0.0421, 0.0487, 0.0396, 0.0705, 0.0335, 0.0627], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0151, 0.0122, 0.0129, 0.0130, 0.0126, 0.0141, 0.0145], device='cuda:3'), out_proj_covar=tensor([9.2731e-05, 1.0997e-04, 8.7766e-05, 9.2387e-05, 9.1917e-05, 9.1275e-05, 1.0215e-04, 1.0487e-04], device='cuda:3') 2023-03-26 17:36:40,950 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.6432, 1.6606, 1.7644, 1.1562, 1.7691, 1.9557, 1.9974, 1.5299], device='cuda:3'), covar=tensor([0.0798, 0.0605, 0.0420, 0.0487, 0.0375, 0.0575, 0.0264, 0.0655], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0151, 0.0122, 0.0129, 0.0130, 0.0126, 0.0141, 0.0145], device='cuda:3'), out_proj_covar=tensor([9.2829e-05, 1.1009e-04, 8.7846e-05, 9.2486e-05, 9.2003e-05, 9.1458e-05, 1.0213e-04, 1.0495e-04], device='cuda:3') 2023-03-26 17:36:41,415 INFO [finetune.py:976] (3/7) Epoch 14, batch 5050, loss[loss=0.1728, simple_loss=0.2391, pruned_loss=0.05324, over 4764.00 frames. ], tot_loss[loss=0.1859, simple_loss=0.2534, pruned_loss=0.0592, over 952478.40 frames. ], batch size: 26, lr: 3.52e-03, grad_scale: 32.0 2023-03-26 17:36:53,666 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0582, 1.3544, 0.8508, 1.9319, 2.3567, 1.7538, 1.7322, 1.8513], device='cuda:3'), covar=tensor([0.1204, 0.1811, 0.1971, 0.1074, 0.1674, 0.1793, 0.1293, 0.1741], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0094, 0.0111, 0.0092, 0.0119, 0.0094, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 17:36:56,804 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 2023-03-26 17:37:02,237 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5762, 1.4872, 1.3994, 1.4733, 1.0998, 2.9981, 1.1693, 1.5614], device='cuda:3'), covar=tensor([0.3985, 0.3095, 0.2515, 0.3000, 0.1920, 0.0339, 0.2473, 0.1352], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0120, 0.0124, 0.0114, 0.0097, 0.0097, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 17:37:02,695 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.152e+02 1.527e+02 1.776e+02 2.127e+02 3.568e+02, threshold=3.553e+02, percent-clipped=0.0 2023-03-26 17:37:07,092 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.2318, 2.5469, 2.3819, 1.2585, 2.6393, 2.2191, 1.8436, 2.2687], device='cuda:3'), covar=tensor([0.0950, 0.1269, 0.2305, 0.2748, 0.2005, 0.2581, 0.3095, 0.1590], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0197, 0.0201, 0.0185, 0.0215, 0.0208, 0.0225, 0.0198], 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') 2023-03-26 17:37:08,286 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9511, 1.6111, 2.4249, 3.8317, 2.6022, 2.5484, 1.2123, 3.0817], device='cuda:3'), covar=tensor([0.1718, 0.1442, 0.1206, 0.0546, 0.0734, 0.1801, 0.1520, 0.0478], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0116, 0.0132, 0.0163, 0.0100, 0.0136, 0.0124, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 17:37:14,552 INFO [finetune.py:976] (3/7) Epoch 14, batch 5100, loss[loss=0.175, simple_loss=0.2391, pruned_loss=0.05545, over 4819.00 frames. ], tot_loss[loss=0.1823, simple_loss=0.2496, pruned_loss=0.05753, over 952793.34 frames. ], batch size: 38, lr: 3.52e-03, grad_scale: 32.0 2023-03-26 17:37:38,102 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79585.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:37:50,469 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.85 vs. limit=5.0 2023-03-26 17:37:50,848 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6393, 1.1560, 0.8708, 1.5352, 2.0392, 1.2346, 1.4532, 1.5430], device='cuda:3'), covar=tensor([0.1528, 0.2192, 0.1960, 0.1272, 0.1962, 0.2016, 0.1513, 0.1851], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0095, 0.0111, 0.0092, 0.0119, 0.0094, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 17:37:57,944 INFO [finetune.py:976] (3/7) Epoch 14, batch 5150, loss[loss=0.1754, simple_loss=0.2644, pruned_loss=0.04324, over 4827.00 frames. ], tot_loss[loss=0.1827, simple_loss=0.2498, pruned_loss=0.05782, over 953109.05 frames. ], batch size: 39, lr: 3.52e-03, grad_scale: 32.0 2023-03-26 17:38:02,831 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.3102, 2.8573, 2.5989, 1.3633, 2.6665, 2.2671, 2.2621, 2.6023], device='cuda:3'), covar=tensor([0.0754, 0.0886, 0.1956, 0.2112, 0.1894, 0.2123, 0.1975, 0.1157], device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0196, 0.0200, 0.0184, 0.0214, 0.0207, 0.0224, 0.0197], 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') 2023-03-26 17:38:04,610 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79621.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:38:20,360 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79641.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 17:38:21,527 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.081e+02 1.601e+02 1.924e+02 2.369e+02 3.228e+02, threshold=3.849e+02, percent-clipped=0.0 2023-03-26 17:38:23,473 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79646.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:38:38,696 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79657.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:38:41,520 INFO [finetune.py:976] (3/7) Epoch 14, batch 5200, loss[loss=0.2145, simple_loss=0.2603, pruned_loss=0.08435, over 4805.00 frames. ], tot_loss[loss=0.1867, simple_loss=0.2538, pruned_loss=0.05986, over 953117.24 frames. ], batch size: 25, lr: 3.52e-03, grad_scale: 32.0 2023-03-26 17:38:50,388 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1438, 1.2649, 1.1473, 1.2244, 1.3722, 2.4401, 1.1768, 1.3705], device='cuda:3'), covar=tensor([0.1080, 0.1853, 0.1198, 0.1007, 0.1717, 0.0372, 0.1646, 0.1776], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0081, 0.0074, 0.0078, 0.0092, 0.0081, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 17:38:50,978 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=79669.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:39:02,318 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9222, 1.7492, 2.2230, 1.5623, 2.0871, 2.1349, 1.7078, 2.2676], device='cuda:3'), covar=tensor([0.1128, 0.1630, 0.1430, 0.1694, 0.0603, 0.1109, 0.2164, 0.0681], device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0206, 0.0193, 0.0191, 0.0176, 0.0214, 0.0217, 0.0199], 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') 2023-03-26 17:39:10,032 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79689.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:39:27,298 INFO [finetune.py:976] (3/7) Epoch 14, batch 5250, loss[loss=0.1474, simple_loss=0.2217, pruned_loss=0.03651, over 4763.00 frames. ], tot_loss[loss=0.189, simple_loss=0.2566, pruned_loss=0.06065, over 953175.33 frames. ], batch size: 27, lr: 3.52e-03, grad_scale: 32.0 2023-03-26 17:39:32,145 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79718.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:39:33,310 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79720.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:39:40,420 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79731.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 17:39:49,074 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.058e+02 1.683e+02 1.987e+02 2.478e+02 3.642e+02, threshold=3.974e+02, percent-clipped=0.0 2023-03-26 17:39:59,973 INFO [finetune.py:976] (3/7) Epoch 14, batch 5300, loss[loss=0.2738, simple_loss=0.3175, pruned_loss=0.1151, over 4894.00 frames. ], tot_loss[loss=0.1912, simple_loss=0.2587, pruned_loss=0.06182, over 954163.05 frames. ], batch size: 32, lr: 3.52e-03, grad_scale: 32.0 2023-03-26 17:40:00,726 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79762.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:40:03,207 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-26 17:40:33,374 INFO [finetune.py:976] (3/7) Epoch 14, batch 5350, loss[loss=0.2557, simple_loss=0.3048, pruned_loss=0.1033, over 4884.00 frames. ], tot_loss[loss=0.1916, simple_loss=0.2594, pruned_loss=0.0619, over 955560.42 frames. ], batch size: 32, lr: 3.52e-03, grad_scale: 32.0 2023-03-26 17:40:41,355 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79823.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:40:55,395 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.090e+02 1.503e+02 1.776e+02 2.291e+02 4.117e+02, threshold=3.553e+02, percent-clipped=3.0 2023-03-26 17:41:06,814 INFO [finetune.py:976] (3/7) Epoch 14, batch 5400, loss[loss=0.1608, simple_loss=0.2197, pruned_loss=0.05098, over 4793.00 frames. ], tot_loss[loss=0.1894, simple_loss=0.2569, pruned_loss=0.06095, over 954956.63 frames. ], batch size: 25, lr: 3.52e-03, grad_scale: 32.0 2023-03-26 17:41:40,250 INFO [finetune.py:976] (3/7) Epoch 14, batch 5450, loss[loss=0.1663, simple_loss=0.2288, pruned_loss=0.05186, over 4902.00 frames. ], tot_loss[loss=0.1876, simple_loss=0.2543, pruned_loss=0.06041, over 954959.22 frames. ], batch size: 32, lr: 3.52e-03, grad_scale: 32.0 2023-03-26 17:41:47,692 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.1374, 0.9747, 0.9902, 0.4173, 0.8480, 1.1879, 1.1957, 1.0334], device='cuda:3'), covar=tensor([0.0958, 0.0638, 0.0519, 0.0566, 0.0559, 0.0638, 0.0391, 0.0731], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0151, 0.0122, 0.0128, 0.0129, 0.0126, 0.0140, 0.0144], device='cuda:3'), out_proj_covar=tensor([9.2438e-05, 1.0977e-04, 8.7535e-05, 9.2195e-05, 9.1750e-05, 9.1463e-05, 1.0159e-04, 1.0430e-04], device='cuda:3') 2023-03-26 17:41:55,366 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5552, 1.4981, 1.8374, 1.8590, 1.6307, 3.3704, 1.3905, 1.6838], device='cuda:3'), covar=tensor([0.0954, 0.1765, 0.1294, 0.0934, 0.1553, 0.0244, 0.1421, 0.1593], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0081, 0.0074, 0.0078, 0.0092, 0.0081, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 17:41:59,678 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79941.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:41:59,697 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79941.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 17:42:00,809 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.028e+02 1.485e+02 1.773e+02 2.175e+02 4.141e+02, threshold=3.546e+02, percent-clipped=3.0 2023-03-26 17:42:14,248 INFO [finetune.py:976] (3/7) Epoch 14, batch 5500, loss[loss=0.1751, simple_loss=0.2449, pruned_loss=0.05267, over 4850.00 frames. ], tot_loss[loss=0.1832, simple_loss=0.2496, pruned_loss=0.05841, over 955227.69 frames. ], batch size: 49, lr: 3.52e-03, grad_scale: 32.0 2023-03-26 17:42:32,376 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=79989.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 17:42:32,407 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79989.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:42:54,668 INFO [finetune.py:976] (3/7) Epoch 14, batch 5550, loss[loss=0.2653, simple_loss=0.3172, pruned_loss=0.1066, over 4762.00 frames. ], tot_loss[loss=0.1862, simple_loss=0.2524, pruned_loss=0.06002, over 955722.55 frames. ], batch size: 59, lr: 3.52e-03, grad_scale: 32.0 2023-03-26 17:42:55,988 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80013.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:43:00,251 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80020.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:43:07,432 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80031.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 17:43:07,448 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6802, 1.4062, 0.9902, 0.2994, 1.2043, 1.4140, 1.2183, 1.2986], device='cuda:3'), covar=tensor([0.0817, 0.0882, 0.1411, 0.1878, 0.1366, 0.2436, 0.2452, 0.0971], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0194, 0.0199, 0.0182, 0.0212, 0.0206, 0.0223, 0.0196], 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') 2023-03-26 17:43:11,544 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=80037.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:43:15,065 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.061e+02 1.609e+02 1.862e+02 2.441e+02 4.163e+02, threshold=3.724e+02, percent-clipped=2.0 2023-03-26 17:43:25,569 INFO [finetune.py:976] (3/7) Epoch 14, batch 5600, loss[loss=0.2023, simple_loss=0.2742, pruned_loss=0.0652, over 4715.00 frames. ], tot_loss[loss=0.1886, simple_loss=0.2554, pruned_loss=0.06086, over 954654.61 frames. ], batch size: 59, lr: 3.52e-03, grad_scale: 32.0 2023-03-26 17:43:29,673 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=80068.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:43:36,038 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=80079.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 17:44:12,670 INFO [finetune.py:976] (3/7) Epoch 14, batch 5650, loss[loss=0.2173, simple_loss=0.2746, pruned_loss=0.07997, over 4837.00 frames. ], tot_loss[loss=0.1899, simple_loss=0.2579, pruned_loss=0.06097, over 955772.99 frames. ], batch size: 49, lr: 3.52e-03, grad_scale: 32.0 2023-03-26 17:44:21,799 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80118.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:44:41,677 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.115e+02 1.565e+02 1.838e+02 2.201e+02 4.652e+02, threshold=3.676e+02, percent-clipped=2.0 2023-03-26 17:44:51,652 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4585, 1.1733, 0.7976, 1.3831, 1.6113, 1.1608, 1.3687, 1.3522], device='cuda:3'), covar=tensor([0.0735, 0.1014, 0.1139, 0.0599, 0.1055, 0.1252, 0.0724, 0.1002], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0095, 0.0111, 0.0093, 0.0120, 0.0094, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 17:44:56,310 INFO [finetune.py:976] (3/7) Epoch 14, batch 5700, loss[loss=0.1655, simple_loss=0.2195, pruned_loss=0.05574, over 4388.00 frames. ], tot_loss[loss=0.1863, simple_loss=0.2535, pruned_loss=0.05961, over 937109.55 frames. ], batch size: 19, lr: 3.52e-03, grad_scale: 32.0 2023-03-26 17:45:28,232 INFO [finetune.py:976] (3/7) Epoch 15, batch 0, loss[loss=0.1644, simple_loss=0.246, pruned_loss=0.04135, over 4775.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.246, pruned_loss=0.04135, over 4775.00 frames. ], batch size: 28, lr: 3.52e-03, grad_scale: 32.0 2023-03-26 17:45:28,232 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 17:45:33,522 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8124, 3.6041, 3.3782, 1.5598, 3.6237, 2.6525, 0.7287, 2.3067], device='cuda:3'), covar=tensor([0.1549, 0.1763, 0.1472, 0.3440, 0.0931, 0.1165, 0.3787, 0.1516], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0171, 0.0158, 0.0127, 0.0155, 0.0122, 0.0144, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 17:45:36,695 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6933, 1.5740, 1.5536, 1.5902, 0.9587, 2.9792, 1.1490, 1.6524], device='cuda:3'), covar=tensor([0.3467, 0.2540, 0.2154, 0.2396, 0.1965, 0.0276, 0.2735, 0.1272], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0116, 0.0121, 0.0124, 0.0115, 0.0098, 0.0097, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 17:45:37,289 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1130, 1.9569, 1.7220, 1.9291, 2.1362, 1.8093, 2.2699, 2.1327], device='cuda:3'), covar=tensor([0.1461, 0.2637, 0.3387, 0.2505, 0.2750, 0.1730, 0.3232, 0.1920], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0187, 0.0234, 0.0255, 0.0245, 0.0200, 0.0213, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 17:45:42,542 INFO [finetune.py:1010] (3/7) Epoch 15, validation: loss=0.1586, simple_loss=0.2288, pruned_loss=0.0442, over 2265189.00 frames. 2023-03-26 17:45:42,542 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6434MB 2023-03-26 17:45:42,685 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3829, 2.3900, 1.8142, 2.5593, 2.3969, 2.0065, 2.9131, 2.4236], device='cuda:3'), covar=tensor([0.1352, 0.2459, 0.3166, 0.2942, 0.2720, 0.1666, 0.3186, 0.1921], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0187, 0.0234, 0.0255, 0.0245, 0.0200, 0.0213, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 17:45:46,977 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8630, 1.7423, 2.4096, 1.4937, 2.1396, 2.4165, 1.6553, 2.4192], device='cuda:3'), covar=tensor([0.1500, 0.2133, 0.1593, 0.2224, 0.0992, 0.1464, 0.2887, 0.0977], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0203, 0.0191, 0.0189, 0.0175, 0.0212, 0.0216, 0.0197], 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') 2023-03-26 17:46:08,709 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9025, 3.2677, 2.8910, 2.1835, 2.8861, 3.2220, 3.0298, 2.9731], device='cuda:3'), covar=tensor([0.0557, 0.0449, 0.0647, 0.0821, 0.0700, 0.0632, 0.0625, 0.0709], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0132, 0.0141, 0.0123, 0.0124, 0.0141, 0.0141, 0.0162], 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') 2023-03-26 17:46:15,194 INFO [finetune.py:976] (3/7) Epoch 15, batch 50, loss[loss=0.1791, simple_loss=0.2533, pruned_loss=0.05249, over 4815.00 frames. ], tot_loss[loss=0.1873, simple_loss=0.2554, pruned_loss=0.05958, over 216343.22 frames. ], batch size: 39, lr: 3.52e-03, grad_scale: 32.0 2023-03-26 17:46:17,633 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80241.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:46:18,755 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.025e+02 1.470e+02 1.896e+02 2.201e+02 3.299e+02, threshold=3.792e+02, percent-clipped=0.0 2023-03-26 17:46:45,383 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80283.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:46:48,319 INFO [finetune.py:976] (3/7) Epoch 15, batch 100, loss[loss=0.2, simple_loss=0.2595, pruned_loss=0.07027, over 4873.00 frames. ], tot_loss[loss=0.184, simple_loss=0.2516, pruned_loss=0.05826, over 382158.23 frames. ], batch size: 34, lr: 3.52e-03, grad_scale: 32.0 2023-03-26 17:46:48,982 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=80289.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:46:52,592 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9349, 1.8606, 1.6509, 1.8569, 1.6779, 4.6252, 2.0040, 2.1437], device='cuda:3'), covar=tensor([0.4067, 0.2937, 0.2323, 0.2838, 0.1572, 0.0156, 0.2318, 0.1201], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0120, 0.0123, 0.0115, 0.0097, 0.0096, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 17:47:05,103 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80313.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:47:05,683 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5669, 1.1455, 0.9168, 1.4783, 1.9746, 1.0518, 1.3594, 1.5406], device='cuda:3'), covar=tensor([0.1562, 0.2302, 0.1918, 0.1346, 0.2008, 0.2061, 0.1640, 0.1999], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0095, 0.0111, 0.0093, 0.0120, 0.0094, 0.0099, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-26 17:47:20,511 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.5203, 1.5063, 1.6223, 0.8996, 1.5841, 1.8893, 1.8907, 1.4415], device='cuda:3'), covar=tensor([0.1050, 0.0738, 0.0461, 0.0550, 0.0493, 0.0562, 0.0309, 0.0775], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0151, 0.0122, 0.0128, 0.0129, 0.0126, 0.0141, 0.0144], device='cuda:3'), out_proj_covar=tensor([9.2526e-05, 1.1032e-04, 8.7830e-05, 9.2220e-05, 9.1680e-05, 9.1409e-05, 1.0192e-04, 1.0433e-04], device='cuda:3') 2023-03-26 17:47:21,597 INFO [finetune.py:976] (3/7) Epoch 15, batch 150, loss[loss=0.193, simple_loss=0.2573, pruned_loss=0.06434, over 4922.00 frames. ], tot_loss[loss=0.1805, simple_loss=0.2472, pruned_loss=0.05683, over 510292.33 frames. ], batch size: 43, lr: 3.51e-03, grad_scale: 32.0 2023-03-26 17:47:25,134 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.875e+01 1.589e+02 1.860e+02 2.189e+02 4.694e+02, threshold=3.721e+02, percent-clipped=1.0 2023-03-26 17:47:25,886 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80344.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:47:36,665 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=80361.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:47:54,530 INFO [finetune.py:976] (3/7) Epoch 15, batch 200, loss[loss=0.1727, simple_loss=0.2397, pruned_loss=0.05287, over 4867.00 frames. ], tot_loss[loss=0.1809, simple_loss=0.2476, pruned_loss=0.05707, over 609822.95 frames. ], batch size: 31, lr: 3.51e-03, grad_scale: 32.0 2023-03-26 17:47:58,245 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80393.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:48:16,699 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80418.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:48:34,172 INFO [finetune.py:976] (3/7) Epoch 15, batch 250, loss[loss=0.1918, simple_loss=0.2867, pruned_loss=0.04842, over 4821.00 frames. ], tot_loss[loss=0.1838, simple_loss=0.2509, pruned_loss=0.05836, over 688275.74 frames. ], batch size: 40, lr: 3.51e-03, grad_scale: 32.0 2023-03-26 17:48:37,160 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.156e+02 1.638e+02 2.049e+02 2.410e+02 5.367e+02, threshold=4.098e+02, percent-clipped=2.0 2023-03-26 17:48:40,478 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-26 17:48:48,465 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80454.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:48:52,841 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-03-26 17:48:58,618 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80463.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:49:00,361 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=80466.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:49:06,783 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4189, 2.1491, 1.6332, 0.7083, 1.9451, 1.8776, 1.6869, 1.8837], device='cuda:3'), covar=tensor([0.0803, 0.0911, 0.1606, 0.2278, 0.1447, 0.2614, 0.2440, 0.0948], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0195, 0.0201, 0.0183, 0.0213, 0.0207, 0.0225, 0.0197], 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') 2023-03-26 17:49:07,452 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-03-26 17:49:11,620 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2388, 1.4436, 0.8450, 2.0457, 2.3975, 1.7432, 1.8329, 1.9519], device='cuda:3'), covar=tensor([0.1515, 0.2125, 0.2230, 0.1253, 0.1945, 0.2133, 0.1517, 0.2027], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0096, 0.0112, 0.0093, 0.0121, 0.0095, 0.0100, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-26 17:49:20,044 INFO [finetune.py:976] (3/7) Epoch 15, batch 300, loss[loss=0.1506, simple_loss=0.2144, pruned_loss=0.04338, over 4720.00 frames. ], tot_loss[loss=0.1868, simple_loss=0.2547, pruned_loss=0.05949, over 747180.30 frames. ], batch size: 23, lr: 3.51e-03, grad_scale: 32.0 2023-03-26 17:49:24,788 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80494.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:50:03,935 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80524.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:50:12,917 INFO [finetune.py:976] (3/7) Epoch 15, batch 350, loss[loss=0.235, simple_loss=0.2945, pruned_loss=0.08776, over 4915.00 frames. ], tot_loss[loss=0.1891, simple_loss=0.2575, pruned_loss=0.06037, over 793066.86 frames. ], batch size: 37, lr: 3.51e-03, grad_scale: 32.0 2023-03-26 17:50:16,428 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.008e+02 1.511e+02 1.809e+02 2.185e+02 3.892e+02, threshold=3.618e+02, percent-clipped=0.0 2023-03-26 17:50:24,827 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80555.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:50:47,435 INFO [finetune.py:976] (3/7) Epoch 15, batch 400, loss[loss=0.1964, simple_loss=0.2522, pruned_loss=0.07027, over 4175.00 frames. ], tot_loss[loss=0.1881, simple_loss=0.2565, pruned_loss=0.05988, over 827703.84 frames. ], batch size: 65, lr: 3.51e-03, grad_scale: 32.0 2023-03-26 17:50:59,791 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.66 vs. limit=5.0 2023-03-26 17:51:20,767 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7851, 1.5247, 2.2982, 3.3683, 2.2854, 2.3317, 1.0074, 2.8425], device='cuda:3'), covar=tensor([0.1737, 0.1497, 0.1276, 0.0588, 0.0843, 0.1654, 0.1911, 0.0493], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0117, 0.0134, 0.0164, 0.0101, 0.0137, 0.0124, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 17:51:29,119 INFO [finetune.py:976] (3/7) Epoch 15, batch 450, loss[loss=0.1877, simple_loss=0.253, pruned_loss=0.0612, over 4853.00 frames. ], tot_loss[loss=0.1868, simple_loss=0.2553, pruned_loss=0.05913, over 855560.78 frames. ], batch size: 31, lr: 3.51e-03, grad_scale: 32.0 2023-03-26 17:51:29,783 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80639.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:51:31,564 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0947, 2.1975, 2.0560, 1.5428, 2.2450, 2.3293, 2.2697, 2.0003], device='cuda:3'), covar=tensor([0.0598, 0.0569, 0.0718, 0.0911, 0.0576, 0.0683, 0.0598, 0.0907], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0131, 0.0140, 0.0122, 0.0122, 0.0139, 0.0139, 0.0161], 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') 2023-03-26 17:51:32,682 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.016e+02 1.580e+02 1.854e+02 2.177e+02 4.594e+02, threshold=3.707e+02, percent-clipped=2.0 2023-03-26 17:52:03,114 INFO [finetune.py:976] (3/7) Epoch 15, batch 500, loss[loss=0.1631, simple_loss=0.238, pruned_loss=0.04409, over 4767.00 frames. ], tot_loss[loss=0.1849, simple_loss=0.2528, pruned_loss=0.05843, over 879206.15 frames. ], batch size: 27, lr: 3.51e-03, grad_scale: 32.0 2023-03-26 17:52:23,152 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.0357, 4.4191, 4.1987, 2.2899, 4.5016, 3.5271, 0.8954, 3.1329], device='cuda:3'), covar=tensor([0.2320, 0.1693, 0.1282, 0.3020, 0.0839, 0.0821, 0.4401, 0.1454], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0174, 0.0160, 0.0128, 0.0157, 0.0123, 0.0146, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 17:52:36,000 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80736.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:52:37,121 INFO [finetune.py:976] (3/7) Epoch 15, batch 550, loss[loss=0.2054, simple_loss=0.2656, pruned_loss=0.07265, over 4862.00 frames. ], tot_loss[loss=0.1821, simple_loss=0.2494, pruned_loss=0.05744, over 896327.52 frames. ], batch size: 34, lr: 3.51e-03, grad_scale: 32.0 2023-03-26 17:52:40,201 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.067e+02 1.496e+02 1.725e+02 2.011e+02 3.976e+02, threshold=3.451e+02, percent-clipped=1.0 2023-03-26 17:52:44,399 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80749.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:52:55,973 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8855, 1.3956, 0.8936, 1.8189, 2.1186, 1.4623, 1.6644, 1.8505], device='cuda:3'), covar=tensor([0.1323, 0.1927, 0.2008, 0.1136, 0.1918, 0.2130, 0.1316, 0.1701], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0096, 0.0113, 0.0094, 0.0121, 0.0095, 0.0100, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-26 17:53:01,300 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80773.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:53:10,748 INFO [finetune.py:976] (3/7) Epoch 15, batch 600, loss[loss=0.1484, simple_loss=0.21, pruned_loss=0.04339, over 4033.00 frames. ], tot_loss[loss=0.1828, simple_loss=0.2497, pruned_loss=0.05799, over 909450.35 frames. ], batch size: 17, lr: 3.51e-03, grad_scale: 32.0 2023-03-26 17:53:14,572 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7178, 1.7433, 1.2605, 1.5856, 1.8411, 1.4583, 2.3886, 1.7282], device='cuda:3'), covar=tensor([0.1471, 0.2049, 0.3412, 0.2919, 0.2730, 0.1766, 0.2761, 0.1996], device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0186, 0.0232, 0.0252, 0.0242, 0.0198, 0.0212, 0.0197], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 17:53:15,172 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1255, 1.9265, 1.6736, 1.6313, 1.9087, 1.8583, 1.8514, 2.6009], device='cuda:3'), covar=tensor([0.4236, 0.4636, 0.3647, 0.4170, 0.4018, 0.2636, 0.4170, 0.1775], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0259, 0.0225, 0.0276, 0.0247, 0.0214, 0.0249, 0.0224], 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') 2023-03-26 17:53:16,808 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80797.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:53:32,684 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80819.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:53:35,800 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3987, 2.8759, 2.7402, 1.3658, 2.9364, 2.1536, 0.7399, 1.8729], device='cuda:3'), covar=tensor([0.2501, 0.2408, 0.1855, 0.3542, 0.1427, 0.1253, 0.4502, 0.1836], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0173, 0.0160, 0.0128, 0.0157, 0.0122, 0.0146, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 17:53:42,411 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-26 17:53:44,659 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80834.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:53:47,023 INFO [finetune.py:976] (3/7) Epoch 15, batch 650, loss[loss=0.2163, simple_loss=0.2788, pruned_loss=0.07689, over 4917.00 frames. ], tot_loss[loss=0.1863, simple_loss=0.2537, pruned_loss=0.05952, over 918585.21 frames. ], batch size: 38, lr: 3.51e-03, grad_scale: 32.0 2023-03-26 17:53:50,575 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.196e+02 1.643e+02 1.965e+02 2.358e+02 6.399e+02, threshold=3.929e+02, percent-clipped=5.0 2023-03-26 17:53:55,341 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80850.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:53:56,549 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6343, 1.5220, 1.1124, 0.3437, 1.2600, 1.4225, 1.3331, 1.3648], device='cuda:3'), covar=tensor([0.0920, 0.0806, 0.1325, 0.1859, 0.1359, 0.2445, 0.2358, 0.0916], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0196, 0.0200, 0.0184, 0.0215, 0.0207, 0.0225, 0.0197], 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') 2023-03-26 17:54:09,216 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-03-26 17:54:29,240 INFO [finetune.py:976] (3/7) Epoch 15, batch 700, loss[loss=0.1951, simple_loss=0.2608, pruned_loss=0.06467, over 4904.00 frames. ], tot_loss[loss=0.1873, simple_loss=0.255, pruned_loss=0.05986, over 924641.55 frames. ], batch size: 32, lr: 3.51e-03, grad_scale: 32.0 2023-03-26 17:54:33,512 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3056, 2.9505, 2.7611, 1.2662, 3.0324, 2.2276, 0.7338, 1.9022], device='cuda:3'), covar=tensor([0.2336, 0.2132, 0.2040, 0.3595, 0.1379, 0.1222, 0.4391, 0.1662], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0174, 0.0159, 0.0128, 0.0157, 0.0122, 0.0146, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 17:54:42,715 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.2722, 2.9384, 3.0456, 3.2037, 3.0645, 2.8976, 3.3583, 0.9135], device='cuda:3'), covar=tensor([0.1238, 0.1053, 0.1112, 0.1189, 0.1763, 0.1941, 0.1126, 0.5672], device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0242, 0.0272, 0.0291, 0.0329, 0.0280, 0.0297, 0.0295], 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') 2023-03-26 17:55:17,874 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.0819, 2.8363, 2.3894, 1.4778, 2.5171, 2.3934, 2.2259, 2.5479], device='cuda:3'), covar=tensor([0.0674, 0.0610, 0.1261, 0.1542, 0.1125, 0.1796, 0.1760, 0.0710], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0195, 0.0200, 0.0184, 0.0214, 0.0207, 0.0225, 0.0197], 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') 2023-03-26 17:55:23,121 INFO [finetune.py:976] (3/7) Epoch 15, batch 750, loss[loss=0.2099, simple_loss=0.2884, pruned_loss=0.06563, over 4906.00 frames. ], tot_loss[loss=0.1896, simple_loss=0.2573, pruned_loss=0.06095, over 930943.69 frames. ], batch size: 36, lr: 3.51e-03, grad_scale: 32.0 2023-03-26 17:55:23,797 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80939.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:55:26,160 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.122e+02 1.628e+02 1.856e+02 2.303e+02 3.612e+02, threshold=3.712e+02, percent-clipped=0.0 2023-03-26 17:55:32,796 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6356, 2.4895, 2.1005, 2.7222, 2.6006, 2.1813, 3.0972, 2.5243], device='cuda:3'), covar=tensor([0.1305, 0.2387, 0.3024, 0.2521, 0.2450, 0.1656, 0.2803, 0.1783], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0186, 0.0233, 0.0253, 0.0243, 0.0199, 0.0212, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 17:55:56,363 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=80987.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:55:56,904 INFO [finetune.py:976] (3/7) Epoch 15, batch 800, loss[loss=0.1801, simple_loss=0.247, pruned_loss=0.05664, over 4793.00 frames. ], tot_loss[loss=0.1892, simple_loss=0.2574, pruned_loss=0.06047, over 936526.91 frames. ], batch size: 51, lr: 3.51e-03, grad_scale: 32.0 2023-03-26 17:55:59,498 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6166, 1.8375, 2.2084, 1.9319, 2.0143, 4.3228, 1.6680, 1.8702], device='cuda:3'), covar=tensor([0.1056, 0.1688, 0.1286, 0.1001, 0.1475, 0.0175, 0.1459, 0.1702], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0081, 0.0074, 0.0077, 0.0091, 0.0080, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 17:56:05,675 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1330, 1.7615, 2.1431, 2.0842, 1.8513, 1.8543, 1.9996, 1.9505], device='cuda:3'), covar=tensor([0.4299, 0.4475, 0.3333, 0.4136, 0.5087, 0.4207, 0.5149, 0.3428], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0239, 0.0256, 0.0266, 0.0266, 0.0238, 0.0280, 0.0235], 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') 2023-03-26 17:56:38,263 INFO [finetune.py:976] (3/7) Epoch 15, batch 850, loss[loss=0.1947, simple_loss=0.2623, pruned_loss=0.06359, over 4824.00 frames. ], tot_loss[loss=0.1872, simple_loss=0.2552, pruned_loss=0.0596, over 942073.30 frames. ], batch size: 39, lr: 3.51e-03, grad_scale: 32.0 2023-03-26 17:56:41,289 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.001e+02 1.679e+02 1.976e+02 2.340e+02 3.768e+02, threshold=3.952e+02, percent-clipped=2.0 2023-03-26 17:56:44,971 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81049.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:57:06,263 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-26 17:57:11,961 INFO [finetune.py:976] (3/7) Epoch 15, batch 900, loss[loss=0.1849, simple_loss=0.262, pruned_loss=0.0539, over 4751.00 frames. ], tot_loss[loss=0.1857, simple_loss=0.2533, pruned_loss=0.05911, over 946130.03 frames. ], batch size: 26, lr: 3.51e-03, grad_scale: 64.0 2023-03-26 17:57:14,452 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81092.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:57:17,458 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=81097.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:57:29,878 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81116.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:57:31,669 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81119.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:57:38,664 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81129.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:57:45,616 INFO [finetune.py:976] (3/7) Epoch 15, batch 950, loss[loss=0.1658, simple_loss=0.2357, pruned_loss=0.04794, over 4907.00 frames. ], tot_loss[loss=0.1847, simple_loss=0.2517, pruned_loss=0.05883, over 947796.80 frames. ], batch size: 37, lr: 3.51e-03, grad_scale: 64.0 2023-03-26 17:57:48,668 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.129e+01 1.456e+02 1.848e+02 2.216e+02 5.430e+02, threshold=3.695e+02, percent-clipped=2.0 2023-03-26 17:57:52,985 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81150.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:58:03,766 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=81167.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:58:11,302 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81177.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:58:19,387 INFO [finetune.py:976] (3/7) Epoch 15, batch 1000, loss[loss=0.1927, simple_loss=0.261, pruned_loss=0.06215, over 4105.00 frames. ], tot_loss[loss=0.1868, simple_loss=0.2538, pruned_loss=0.05991, over 946250.62 frames. ], batch size: 66, lr: 3.51e-03, grad_scale: 64.0 2023-03-26 17:58:25,510 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=81198.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:58:32,828 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.4416, 3.0503, 2.7471, 1.5330, 2.8591, 2.4978, 2.3312, 2.6317], device='cuda:3'), covar=tensor([0.0826, 0.0709, 0.1400, 0.2124, 0.1483, 0.2176, 0.2040, 0.1185], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0195, 0.0200, 0.0184, 0.0214, 0.0207, 0.0225, 0.0196], 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') 2023-03-26 17:58:52,894 INFO [finetune.py:976] (3/7) Epoch 15, batch 1050, loss[loss=0.1745, simple_loss=0.2401, pruned_loss=0.0545, over 4784.00 frames. ], tot_loss[loss=0.187, simple_loss=0.2552, pruned_loss=0.05938, over 949417.51 frames. ], batch size: 29, lr: 3.51e-03, grad_scale: 64.0 2023-03-26 17:58:56,385 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.183e+02 1.566e+02 1.800e+02 2.282e+02 3.514e+02, threshold=3.601e+02, percent-clipped=0.0 2023-03-26 17:59:00,081 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.3683, 3.8010, 3.9805, 4.2149, 4.1215, 3.8539, 4.4856, 1.3177], device='cuda:3'), covar=tensor([0.0837, 0.0821, 0.0784, 0.1048, 0.1347, 0.1493, 0.0683, 0.5678], device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0242, 0.0270, 0.0289, 0.0329, 0.0280, 0.0295, 0.0294], 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') 2023-03-26 17:59:13,454 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8499, 0.9747, 1.8093, 1.8052, 1.6042, 1.5401, 1.6436, 1.6877], device='cuda:3'), covar=tensor([0.3809, 0.4145, 0.3516, 0.3435, 0.4728, 0.3684, 0.4172, 0.3314], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0239, 0.0256, 0.0267, 0.0266, 0.0238, 0.0280, 0.0235], 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') 2023-03-26 17:59:31,698 INFO [finetune.py:976] (3/7) Epoch 15, batch 1100, loss[loss=0.2412, simple_loss=0.3013, pruned_loss=0.09062, over 4811.00 frames. ], tot_loss[loss=0.1873, simple_loss=0.2557, pruned_loss=0.05946, over 951669.01 frames. ], batch size: 40, lr: 3.51e-03, grad_scale: 64.0 2023-03-26 17:59:43,343 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81299.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 17:59:49,388 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81301.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:00:16,529 INFO [finetune.py:976] (3/7) Epoch 15, batch 1150, loss[loss=0.2076, simple_loss=0.2687, pruned_loss=0.0733, over 4734.00 frames. ], tot_loss[loss=0.1879, simple_loss=0.2566, pruned_loss=0.05957, over 952084.57 frames. ], batch size: 54, lr: 3.51e-03, grad_scale: 64.0 2023-03-26 18:00:23,247 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.122e+02 1.635e+02 2.084e+02 2.407e+02 3.907e+02, threshold=4.168e+02, percent-clipped=1.0 2023-03-26 18:00:39,844 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3564, 2.1541, 1.9388, 2.1932, 2.0378, 2.1368, 2.0929, 2.8337], device='cuda:3'), covar=tensor([0.3920, 0.4795, 0.3481, 0.4035, 0.4019, 0.2619, 0.3808, 0.1703], device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0258, 0.0224, 0.0275, 0.0245, 0.0213, 0.0247, 0.0224], 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') 2023-03-26 18:00:41,618 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81360.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:00:42,865 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81362.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:00:51,051 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.46 vs. limit=5.0 2023-03-26 18:00:59,899 INFO [finetune.py:976] (3/7) Epoch 15, batch 1200, loss[loss=0.1759, simple_loss=0.2564, pruned_loss=0.04766, over 4790.00 frames. ], tot_loss[loss=0.1862, simple_loss=0.2548, pruned_loss=0.05874, over 951858.08 frames. ], batch size: 51, lr: 3.51e-03, grad_scale: 64.0 2023-03-26 18:01:03,401 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81392.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:01:06,959 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2511, 2.0851, 1.8178, 2.0761, 1.9389, 2.0491, 1.9670, 2.7672], device='cuda:3'), covar=tensor([0.3960, 0.4619, 0.3507, 0.4421, 0.4472, 0.2521, 0.4077, 0.1860], device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0258, 0.0224, 0.0275, 0.0245, 0.0213, 0.0248, 0.0224], 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') 2023-03-26 18:01:13,560 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.61 vs. limit=5.0 2023-03-26 18:01:27,580 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81429.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:01:35,102 INFO [finetune.py:976] (3/7) Epoch 15, batch 1250, loss[loss=0.1591, simple_loss=0.2235, pruned_loss=0.04735, over 4683.00 frames. ], tot_loss[loss=0.1853, simple_loss=0.2529, pruned_loss=0.05886, over 952038.60 frames. ], batch size: 23, lr: 3.51e-03, grad_scale: 64.0 2023-03-26 18:01:40,099 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=81440.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:01:42,323 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.103e+01 1.546e+02 1.830e+02 2.259e+02 3.665e+02, threshold=3.660e+02, percent-clipped=0.0 2023-03-26 18:01:55,868 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6425, 1.4622, 2.0348, 3.2576, 2.0838, 2.2545, 1.0742, 2.6424], device='cuda:3'), covar=tensor([0.1690, 0.1463, 0.1316, 0.0562, 0.0844, 0.1407, 0.1732, 0.0516], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0114, 0.0131, 0.0162, 0.0099, 0.0135, 0.0123, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 18:02:05,451 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81472.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:02:08,460 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=81477.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:02:15,565 INFO [finetune.py:976] (3/7) Epoch 15, batch 1300, loss[loss=0.184, simple_loss=0.2518, pruned_loss=0.05804, over 4873.00 frames. ], tot_loss[loss=0.1842, simple_loss=0.2511, pruned_loss=0.05871, over 952194.61 frames. ], batch size: 34, lr: 3.50e-03, grad_scale: 64.0 2023-03-26 18:02:49,394 INFO [finetune.py:976] (3/7) Epoch 15, batch 1350, loss[loss=0.1773, simple_loss=0.2333, pruned_loss=0.06066, over 4703.00 frames. ], tot_loss[loss=0.1831, simple_loss=0.25, pruned_loss=0.05811, over 954892.71 frames. ], batch size: 23, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:02:53,478 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.113e+02 1.608e+02 1.859e+02 2.257e+02 3.880e+02, threshold=3.719e+02, percent-clipped=1.0 2023-03-26 18:03:04,212 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4716, 1.3607, 1.4128, 1.5458, 1.5302, 2.9618, 1.2490, 1.4539], device='cuda:3'), covar=tensor([0.1026, 0.1916, 0.1280, 0.1070, 0.1673, 0.0296, 0.1611, 0.1848], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0082, 0.0074, 0.0078, 0.0092, 0.0081, 0.0086, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 18:03:06,713 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.97 vs. limit=5.0 2023-03-26 18:03:22,723 INFO [finetune.py:976] (3/7) Epoch 15, batch 1400, loss[loss=0.2085, simple_loss=0.2825, pruned_loss=0.0673, over 4907.00 frames. ], tot_loss[loss=0.1852, simple_loss=0.2531, pruned_loss=0.05862, over 955113.14 frames. ], batch size: 43, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:03:33,292 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0121, 1.7513, 2.6951, 1.3848, 2.1986, 2.3578, 1.5837, 2.4141], device='cuda:3'), covar=tensor([0.1666, 0.2323, 0.1252, 0.2345, 0.1194, 0.1648, 0.3008, 0.1217], device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0205, 0.0193, 0.0191, 0.0178, 0.0214, 0.0218, 0.0201], 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') 2023-03-26 18:03:56,014 INFO [finetune.py:976] (3/7) Epoch 15, batch 1450, loss[loss=0.2451, simple_loss=0.2881, pruned_loss=0.101, over 4897.00 frames. ], tot_loss[loss=0.1867, simple_loss=0.2548, pruned_loss=0.05926, over 953941.14 frames. ], batch size: 32, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:04:00,098 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.106e+02 1.620e+02 1.887e+02 2.237e+02 3.719e+02, threshold=3.774e+02, percent-clipped=1.0 2023-03-26 18:04:07,944 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81655.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:04:09,566 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81657.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:04:18,095 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81670.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:04:29,493 INFO [finetune.py:976] (3/7) Epoch 15, batch 1500, loss[loss=0.1793, simple_loss=0.2527, pruned_loss=0.05295, over 4905.00 frames. ], tot_loss[loss=0.188, simple_loss=0.2567, pruned_loss=0.05959, over 956241.50 frames. ], batch size: 37, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:05:16,721 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81731.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 18:05:20,801 INFO [finetune.py:976] (3/7) Epoch 15, batch 1550, loss[loss=0.1905, simple_loss=0.2609, pruned_loss=0.06007, over 4924.00 frames. ], tot_loss[loss=0.1873, simple_loss=0.2561, pruned_loss=0.0592, over 955184.40 frames. ], batch size: 38, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:05:24,955 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.055e+02 1.530e+02 1.898e+02 2.293e+02 4.636e+02, threshold=3.795e+02, percent-clipped=1.0 2023-03-26 18:05:31,570 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6084, 1.1138, 0.8216, 1.4588, 1.9961, 0.8309, 1.2814, 1.4089], device='cuda:3'), covar=tensor([0.1483, 0.2192, 0.1726, 0.1228, 0.1993, 0.2022, 0.1514, 0.2008], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0096, 0.0112, 0.0093, 0.0121, 0.0095, 0.0100, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-26 18:05:50,091 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81772.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:06:03,756 INFO [finetune.py:976] (3/7) Epoch 15, batch 1600, loss[loss=0.1915, simple_loss=0.2509, pruned_loss=0.06602, over 4823.00 frames. ], tot_loss[loss=0.185, simple_loss=0.2535, pruned_loss=0.05826, over 956818.36 frames. ], batch size: 25, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:06:10,398 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1103, 1.8311, 2.4928, 4.1995, 2.8794, 2.7413, 0.9370, 3.4928], device='cuda:3'), covar=tensor([0.1632, 0.1375, 0.1301, 0.0399, 0.0688, 0.1374, 0.1841, 0.0309], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0115, 0.0131, 0.0162, 0.0099, 0.0136, 0.0123, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 18:06:25,746 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=81820.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:06:37,140 INFO [finetune.py:976] (3/7) Epoch 15, batch 1650, loss[loss=0.1618, simple_loss=0.2288, pruned_loss=0.04741, over 4921.00 frames. ], tot_loss[loss=0.1833, simple_loss=0.2515, pruned_loss=0.05753, over 956485.19 frames. ], batch size: 36, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:06:40,772 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.362e+01 1.564e+02 1.826e+02 2.251e+02 4.924e+02, threshold=3.651e+02, percent-clipped=3.0 2023-03-26 18:06:45,903 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-26 18:06:56,346 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2460, 1.7570, 2.3164, 2.1531, 1.9534, 1.9292, 2.0637, 2.0261], device='cuda:3'), covar=tensor([0.4173, 0.4289, 0.3153, 0.4024, 0.5199, 0.4037, 0.5134, 0.3101], device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0238, 0.0255, 0.0266, 0.0264, 0.0238, 0.0279, 0.0234], 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') 2023-03-26 18:07:18,076 INFO [finetune.py:976] (3/7) Epoch 15, batch 1700, loss[loss=0.1956, simple_loss=0.2583, pruned_loss=0.06642, over 4937.00 frames. ], tot_loss[loss=0.182, simple_loss=0.2498, pruned_loss=0.05716, over 958754.83 frames. ], batch size: 33, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:07:51,480 INFO [finetune.py:976] (3/7) Epoch 15, batch 1750, loss[loss=0.1859, simple_loss=0.2521, pruned_loss=0.05987, over 4739.00 frames. ], tot_loss[loss=0.1848, simple_loss=0.2526, pruned_loss=0.0585, over 957155.48 frames. ], batch size: 23, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:07:55,584 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.016e+02 1.510e+02 1.914e+02 2.293e+02 4.004e+02, threshold=3.828e+02, percent-clipped=1.0 2023-03-26 18:08:01,741 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9354, 1.4552, 1.0673, 1.9473, 2.3616, 1.4844, 1.6975, 1.8899], device='cuda:3'), covar=tensor([0.1172, 0.1785, 0.1618, 0.0855, 0.1391, 0.1646, 0.1145, 0.1537], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0096, 0.0112, 0.0093, 0.0121, 0.0094, 0.0100, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-26 18:08:02,960 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81955.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:08:04,188 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81957.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:08:09,554 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.2576, 3.6805, 3.9119, 4.0815, 4.0756, 3.8138, 4.3408, 1.4691], device='cuda:3'), covar=tensor([0.0723, 0.0847, 0.0792, 0.0948, 0.1028, 0.1448, 0.0623, 0.5303], device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0241, 0.0271, 0.0290, 0.0328, 0.0280, 0.0296, 0.0294], 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') 2023-03-26 18:08:24,913 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81987.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:08:25,414 INFO [finetune.py:976] (3/7) Epoch 15, batch 1800, loss[loss=0.1788, simple_loss=0.2593, pruned_loss=0.04922, over 4827.00 frames. ], tot_loss[loss=0.1858, simple_loss=0.2543, pruned_loss=0.05867, over 953164.44 frames. ], batch size: 33, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:08:36,207 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=82003.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:08:37,855 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=82005.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:08:38,483 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6803, 1.1258, 0.8620, 1.5728, 2.1174, 1.3962, 1.4154, 1.5380], device='cuda:3'), covar=tensor([0.1685, 0.2303, 0.2067, 0.1316, 0.1980, 0.2253, 0.1473, 0.2008], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0097, 0.0113, 0.0094, 0.0121, 0.0095, 0.0100, 0.0091], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-26 18:08:52,965 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82026.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 18:09:00,021 INFO [finetune.py:976] (3/7) Epoch 15, batch 1850, loss[loss=0.1814, simple_loss=0.247, pruned_loss=0.05785, over 4755.00 frames. ], tot_loss[loss=0.1873, simple_loss=0.256, pruned_loss=0.05931, over 953096.01 frames. ], batch size: 27, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:09:03,672 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.098e+02 1.664e+02 1.894e+02 2.440e+02 3.763e+02, threshold=3.787e+02, percent-clipped=0.0 2023-03-26 18:09:06,714 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82048.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:09:11,020 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82055.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:09:33,282 INFO [finetune.py:976] (3/7) Epoch 15, batch 1900, loss[loss=0.2172, simple_loss=0.2723, pruned_loss=0.08102, over 4820.00 frames. ], tot_loss[loss=0.1883, simple_loss=0.2574, pruned_loss=0.05964, over 953965.97 frames. ], batch size: 49, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:09:51,832 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82116.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:10:16,117 INFO [finetune.py:976] (3/7) Epoch 15, batch 1950, loss[loss=0.2015, simple_loss=0.2691, pruned_loss=0.06698, over 4722.00 frames. ], tot_loss[loss=0.1874, simple_loss=0.2562, pruned_loss=0.05927, over 954048.93 frames. ], batch size: 59, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:10:24,222 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.964e+01 1.525e+02 1.906e+02 2.226e+02 4.434e+02, threshold=3.812e+02, percent-clipped=2.0 2023-03-26 18:11:01,528 INFO [finetune.py:976] (3/7) Epoch 15, batch 2000, loss[loss=0.2005, simple_loss=0.2602, pruned_loss=0.07041, over 4816.00 frames. ], tot_loss[loss=0.186, simple_loss=0.2542, pruned_loss=0.0589, over 956329.11 frames. ], batch size: 41, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:11:16,200 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6752, 1.5922, 1.3456, 1.2803, 1.7684, 1.4724, 2.0685, 1.7112], device='cuda:3'), covar=tensor([0.1363, 0.1902, 0.3193, 0.2527, 0.2582, 0.1553, 0.2161, 0.1831], device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0188, 0.0236, 0.0255, 0.0247, 0.0201, 0.0214, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 18:11:38,383 INFO [finetune.py:976] (3/7) Epoch 15, batch 2050, loss[loss=0.1572, simple_loss=0.2402, pruned_loss=0.03704, over 4817.00 frames. ], tot_loss[loss=0.1843, simple_loss=0.2524, pruned_loss=0.05809, over 956823.13 frames. ], batch size: 40, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:11:42,514 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.084e+02 1.417e+02 1.727e+02 2.274e+02 4.171e+02, threshold=3.454e+02, percent-clipped=1.0 2023-03-26 18:12:11,838 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2062, 2.1486, 1.8187, 2.1730, 2.0282, 2.0281, 1.9571, 2.8444], device='cuda:3'), covar=tensor([0.3922, 0.5242, 0.3730, 0.4786, 0.5056, 0.2659, 0.5016, 0.1710], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0259, 0.0226, 0.0275, 0.0247, 0.0214, 0.0248, 0.0225], 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') 2023-03-26 18:12:24,919 INFO [finetune.py:976] (3/7) Epoch 15, batch 2100, loss[loss=0.2569, simple_loss=0.3093, pruned_loss=0.1022, over 4916.00 frames. ], tot_loss[loss=0.1838, simple_loss=0.2518, pruned_loss=0.05787, over 955784.20 frames. ], batch size: 37, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:12:31,362 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7018, 1.5937, 1.4180, 1.5607, 1.9472, 1.8661, 1.6439, 1.4639], device='cuda:3'), covar=tensor([0.0289, 0.0301, 0.0524, 0.0277, 0.0176, 0.0495, 0.0274, 0.0372], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0108, 0.0141, 0.0112, 0.0099, 0.0106, 0.0096, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.2202e-05, 8.3721e-05, 1.1166e-04, 8.7037e-05, 7.7468e-05, 7.8528e-05, 7.2343e-05, 8.2500e-05], device='cuda:3') 2023-03-26 18:12:54,233 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82326.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:13:02,490 INFO [finetune.py:976] (3/7) Epoch 15, batch 2150, loss[loss=0.183, simple_loss=0.2592, pruned_loss=0.05346, over 4835.00 frames. ], tot_loss[loss=0.1865, simple_loss=0.2546, pruned_loss=0.05918, over 951522.65 frames. ], batch size: 30, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:13:06,113 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82343.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:13:06,660 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.137e+02 1.626e+02 1.861e+02 2.291e+02 4.001e+02, threshold=3.721e+02, percent-clipped=2.0 2023-03-26 18:13:07,989 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82346.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:13:25,988 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=82374.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:13:35,325 INFO [finetune.py:976] (3/7) Epoch 15, batch 2200, loss[loss=0.1935, simple_loss=0.2602, pruned_loss=0.06343, over 4891.00 frames. ], tot_loss[loss=0.1885, simple_loss=0.2566, pruned_loss=0.06016, over 950171.92 frames. ], batch size: 35, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:13:48,067 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82407.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:13:50,412 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82411.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:14:08,110 INFO [finetune.py:976] (3/7) Epoch 15, batch 2250, loss[loss=0.1548, simple_loss=0.2288, pruned_loss=0.04046, over 4795.00 frames. ], tot_loss[loss=0.1888, simple_loss=0.2572, pruned_loss=0.06022, over 950964.79 frames. ], batch size: 26, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:14:12,178 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.061e+02 1.505e+02 1.711e+02 2.071e+02 3.892e+02, threshold=3.421e+02, percent-clipped=2.0 2023-03-26 18:14:41,716 INFO [finetune.py:976] (3/7) Epoch 15, batch 2300, loss[loss=0.1917, simple_loss=0.2554, pruned_loss=0.06397, over 4899.00 frames. ], tot_loss[loss=0.1873, simple_loss=0.2563, pruned_loss=0.05912, over 951773.96 frames. ], batch size: 35, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:14:45,270 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82493.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 18:14:53,048 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-03-26 18:15:17,447 INFO [finetune.py:976] (3/7) Epoch 15, batch 2350, loss[loss=0.1599, simple_loss=0.2171, pruned_loss=0.05132, over 4761.00 frames. ], tot_loss[loss=0.1869, simple_loss=0.255, pruned_loss=0.05934, over 954371.24 frames. ], batch size: 28, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:15:19,144 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-26 18:15:19,329 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6218, 3.5676, 3.4344, 1.7094, 3.6149, 2.7709, 1.0044, 2.5587], device='cuda:3'), covar=tensor([0.2534, 0.1864, 0.1411, 0.3040, 0.0997, 0.0982, 0.3951, 0.1293], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0173, 0.0158, 0.0127, 0.0156, 0.0122, 0.0145, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 18:15:21,101 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.035e+02 1.644e+02 1.984e+02 2.390e+02 4.799e+02, threshold=3.967e+02, percent-clipped=3.0 2023-03-26 18:15:28,161 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8980, 1.8948, 1.6630, 2.1147, 2.3104, 1.9808, 1.7618, 1.5380], device='cuda:3'), covar=tensor([0.2220, 0.1905, 0.1843, 0.1577, 0.1698, 0.1178, 0.2376, 0.1932], device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0209, 0.0212, 0.0193, 0.0243, 0.0186, 0.0216, 0.0200], 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') 2023-03-26 18:15:28,713 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82554.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 18:15:44,890 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82569.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:16:00,853 INFO [finetune.py:976] (3/7) Epoch 15, batch 2400, loss[loss=0.1632, simple_loss=0.2291, pruned_loss=0.04862, over 4913.00 frames. ], tot_loss[loss=0.1854, simple_loss=0.2531, pruned_loss=0.05886, over 956536.43 frames. ], batch size: 36, lr: 3.50e-03, grad_scale: 32.0 2023-03-26 18:16:37,329 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82630.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:16:42,547 INFO [finetune.py:976] (3/7) Epoch 15, batch 2450, loss[loss=0.1974, simple_loss=0.2698, pruned_loss=0.06246, over 4863.00 frames. ], tot_loss[loss=0.1822, simple_loss=0.2495, pruned_loss=0.05751, over 952463.45 frames. ], batch size: 31, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:16:45,693 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82643.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:16:46,161 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.277e+01 1.596e+02 1.937e+02 2.264e+02 4.235e+02, threshold=3.875e+02, percent-clipped=1.0 2023-03-26 18:16:59,670 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.21 vs. limit=5.0 2023-03-26 18:17:12,246 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5866, 1.4440, 2.0344, 3.2735, 2.2270, 2.3563, 0.7480, 2.7800], device='cuda:3'), covar=tensor([0.1761, 0.1539, 0.1319, 0.0583, 0.0747, 0.1412, 0.1947, 0.0455], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0116, 0.0132, 0.0163, 0.0099, 0.0137, 0.0124, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 18:17:18,142 INFO [finetune.py:976] (3/7) Epoch 15, batch 2500, loss[loss=0.2143, simple_loss=0.2792, pruned_loss=0.07465, over 4845.00 frames. ], tot_loss[loss=0.1833, simple_loss=0.2509, pruned_loss=0.05789, over 953083.04 frames. ], batch size: 49, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:17:20,060 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=82691.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:17:35,740 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82702.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:17:46,369 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82711.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:17:57,251 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1913, 2.0818, 1.6877, 0.8334, 1.8990, 1.7444, 1.5558, 1.8801], device='cuda:3'), covar=tensor([0.1001, 0.0840, 0.1575, 0.2009, 0.1466, 0.2237, 0.2399, 0.0941], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0194, 0.0200, 0.0184, 0.0213, 0.0207, 0.0224, 0.0196], 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') 2023-03-26 18:18:03,683 INFO [finetune.py:976] (3/7) Epoch 15, batch 2550, loss[loss=0.2603, simple_loss=0.316, pruned_loss=0.1023, over 4908.00 frames. ], tot_loss[loss=0.1872, simple_loss=0.2553, pruned_loss=0.05958, over 954805.88 frames. ], batch size: 36, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:18:06,088 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8510, 1.8177, 1.7005, 1.8880, 1.8303, 4.6843, 1.7921, 2.5702], device='cuda:3'), covar=tensor([0.3344, 0.2504, 0.2048, 0.2372, 0.1510, 0.0104, 0.2352, 0.1066], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0116, 0.0120, 0.0124, 0.0114, 0.0097, 0.0097, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 18:18:07,766 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.099e+02 1.574e+02 1.850e+02 2.269e+02 4.152e+02, threshold=3.700e+02, percent-clipped=3.0 2023-03-26 18:18:16,100 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3190, 2.1157, 1.6849, 1.9951, 2.0183, 1.9418, 2.0715, 2.7697], device='cuda:3'), covar=tensor([0.3762, 0.4649, 0.3702, 0.4169, 0.4099, 0.2663, 0.3881, 0.1841], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0259, 0.0225, 0.0274, 0.0246, 0.0213, 0.0247, 0.0224], 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') 2023-03-26 18:18:17,863 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=82759.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:18:36,856 INFO [finetune.py:976] (3/7) Epoch 15, batch 2600, loss[loss=0.2066, simple_loss=0.2716, pruned_loss=0.07078, over 4886.00 frames. ], tot_loss[loss=0.187, simple_loss=0.2553, pruned_loss=0.05937, over 951377.41 frames. ], batch size: 32, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:19:06,663 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1754, 1.7080, 2.1089, 2.0663, 1.7881, 1.8095, 1.9900, 1.9273], device='cuda:3'), covar=tensor([0.4138, 0.4388, 0.3478, 0.4386, 0.5071, 0.4026, 0.5203, 0.3313], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0238, 0.0256, 0.0268, 0.0266, 0.0239, 0.0280, 0.0235], 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') 2023-03-26 18:19:10,652 INFO [finetune.py:976] (3/7) Epoch 15, batch 2650, loss[loss=0.2041, simple_loss=0.2792, pruned_loss=0.06451, over 4787.00 frames. ], tot_loss[loss=0.1891, simple_loss=0.2573, pruned_loss=0.0604, over 949418.88 frames. ], batch size: 45, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:19:14,265 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.056e+02 1.579e+02 1.879e+02 2.251e+02 6.929e+02, threshold=3.759e+02, percent-clipped=2.0 2023-03-26 18:19:17,846 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82849.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 18:19:31,957 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6561, 1.7017, 2.0906, 1.3951, 1.8222, 2.0117, 1.6140, 2.1920], device='cuda:3'), covar=tensor([0.1636, 0.2011, 0.1460, 0.1803, 0.1119, 0.1530, 0.2685, 0.1015], device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0204, 0.0194, 0.0191, 0.0177, 0.0214, 0.0218, 0.0202], 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') 2023-03-26 18:19:32,318 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 2023-03-26 18:19:43,180 INFO [finetune.py:976] (3/7) Epoch 15, batch 2700, loss[loss=0.1612, simple_loss=0.2192, pruned_loss=0.05162, over 4345.00 frames. ], tot_loss[loss=0.1875, simple_loss=0.2564, pruned_loss=0.05926, over 951136.44 frames. ], batch size: 18, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:20:08,048 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82925.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:20:16,388 INFO [finetune.py:976] (3/7) Epoch 15, batch 2750, loss[loss=0.1749, simple_loss=0.247, pruned_loss=0.05139, over 4775.00 frames. ], tot_loss[loss=0.1842, simple_loss=0.2529, pruned_loss=0.0578, over 953306.98 frames. ], batch size: 28, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:20:20,502 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.092e+02 1.574e+02 1.758e+02 2.107e+02 4.076e+02, threshold=3.515e+02, percent-clipped=2.0 2023-03-26 18:20:33,230 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82963.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:20:49,689 INFO [finetune.py:976] (3/7) Epoch 15, batch 2800, loss[loss=0.2025, simple_loss=0.263, pruned_loss=0.07094, over 4869.00 frames. ], tot_loss[loss=0.1814, simple_loss=0.2493, pruned_loss=0.0568, over 952578.56 frames. ], batch size: 31, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:21:07,089 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83002.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:21:21,945 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83024.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:21:21,961 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3335, 2.2764, 1.8279, 2.3845, 2.3034, 1.9974, 2.6703, 2.3347], device='cuda:3'), covar=tensor([0.1432, 0.2235, 0.3380, 0.2754, 0.2697, 0.1838, 0.3054, 0.1980], device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0187, 0.0236, 0.0254, 0.0245, 0.0201, 0.0213, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 18:21:37,766 INFO [finetune.py:976] (3/7) Epoch 15, batch 2850, loss[loss=0.155, simple_loss=0.2158, pruned_loss=0.0471, over 4738.00 frames. ], tot_loss[loss=0.1811, simple_loss=0.2486, pruned_loss=0.05682, over 955346.54 frames. ], batch size: 27, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:21:41,396 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.061e+02 1.604e+02 1.866e+02 2.227e+02 4.125e+02, threshold=3.733e+02, percent-clipped=3.0 2023-03-26 18:21:48,543 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=83050.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:22:15,028 INFO [finetune.py:976] (3/7) Epoch 15, batch 2900, loss[loss=0.1407, simple_loss=0.2196, pruned_loss=0.03085, over 4748.00 frames. ], tot_loss[loss=0.1842, simple_loss=0.2524, pruned_loss=0.05799, over 956569.00 frames. ], batch size: 27, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:22:57,709 INFO [finetune.py:976] (3/7) Epoch 15, batch 2950, loss[loss=0.1873, simple_loss=0.2697, pruned_loss=0.05245, over 4841.00 frames. ], tot_loss[loss=0.1857, simple_loss=0.254, pruned_loss=0.05865, over 955141.66 frames. ], batch size: 47, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:23:01,331 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.293e+02 1.748e+02 2.030e+02 2.368e+02 3.585e+02, threshold=4.059e+02, percent-clipped=0.0 2023-03-26 18:23:07,705 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7682, 1.5022, 1.4864, 1.4984, 1.9171, 1.8526, 1.6177, 1.4186], device='cuda:3'), covar=tensor([0.0268, 0.0316, 0.0537, 0.0302, 0.0232, 0.0400, 0.0333, 0.0358], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0108, 0.0142, 0.0112, 0.0099, 0.0106, 0.0097, 0.0109], device='cuda:3'), out_proj_covar=tensor([7.2424e-05, 8.3787e-05, 1.1226e-04, 8.6779e-05, 7.7655e-05, 7.8499e-05, 7.3035e-05, 8.2826e-05], device='cuda:3') 2023-03-26 18:23:08,291 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6054, 1.1044, 0.9074, 1.5168, 1.9363, 1.1706, 1.3058, 1.6159], device='cuda:3'), covar=tensor([0.1353, 0.2121, 0.1919, 0.1142, 0.2044, 0.2166, 0.1407, 0.1693], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0095, 0.0111, 0.0093, 0.0119, 0.0094, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 18:23:08,896 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83149.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 18:23:11,382 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.35 vs. limit=5.0 2023-03-26 18:23:19,801 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.33 vs. limit=5.0 2023-03-26 18:23:24,106 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3249, 1.2977, 1.7198, 2.4538, 1.6062, 2.1902, 0.8510, 2.1122], device='cuda:3'), covar=tensor([0.1796, 0.1406, 0.1043, 0.0689, 0.0906, 0.1230, 0.1593, 0.0594], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0117, 0.0133, 0.0164, 0.0100, 0.0138, 0.0124, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 18:23:37,598 INFO [finetune.py:976] (3/7) Epoch 15, batch 3000, loss[loss=0.1168, simple_loss=0.1776, pruned_loss=0.02799, over 3994.00 frames. ], tot_loss[loss=0.1884, simple_loss=0.2563, pruned_loss=0.06026, over 954256.33 frames. ], batch size: 17, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:23:37,598 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 18:23:48,369 INFO [finetune.py:1010] (3/7) Epoch 15, validation: loss=0.1564, simple_loss=0.2269, pruned_loss=0.04296, over 2265189.00 frames. 2023-03-26 18:23:48,369 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6434MB 2023-03-26 18:23:59,740 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83196.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:24:00,272 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=83197.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 18:24:05,409 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-03-26 18:24:21,548 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83225.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:24:30,316 INFO [finetune.py:976] (3/7) Epoch 15, batch 3050, loss[loss=0.1818, simple_loss=0.2545, pruned_loss=0.05458, over 4859.00 frames. ], tot_loss[loss=0.187, simple_loss=0.2556, pruned_loss=0.0592, over 953503.84 frames. ], batch size: 31, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:24:34,914 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.082e+02 1.542e+02 1.763e+02 2.135e+02 3.801e+02, threshold=3.526e+02, percent-clipped=0.0 2023-03-26 18:24:41,670 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3474, 2.1994, 1.9095, 2.2288, 2.3601, 2.0655, 2.5360, 2.3491], device='cuda:3'), covar=tensor([0.1187, 0.2040, 0.2937, 0.2418, 0.2323, 0.1642, 0.2705, 0.1794], device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0187, 0.0235, 0.0254, 0.0245, 0.0201, 0.0212, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 18:24:41,925 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-26 18:24:44,045 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83257.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:24:54,192 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=83273.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:24:55,117 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-26 18:25:01,256 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83284.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:25:04,096 INFO [finetune.py:976] (3/7) Epoch 15, batch 3100, loss[loss=0.1836, simple_loss=0.2429, pruned_loss=0.06216, over 4889.00 frames. ], tot_loss[loss=0.1859, simple_loss=0.2545, pruned_loss=0.05862, over 954630.85 frames. ], batch size: 32, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:25:07,328 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0 2023-03-26 18:25:15,161 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6027, 3.4903, 3.2466, 1.5193, 3.5759, 2.6530, 0.9156, 2.3527], device='cuda:3'), covar=tensor([0.2379, 0.2029, 0.1656, 0.3733, 0.1186, 0.1148, 0.4532, 0.1713], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0175, 0.0160, 0.0128, 0.0158, 0.0123, 0.0146, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 18:25:16,059 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.45 vs. limit=5.0 2023-03-26 18:25:24,780 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83319.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:25:37,273 INFO [finetune.py:976] (3/7) Epoch 15, batch 3150, loss[loss=0.1706, simple_loss=0.2481, pruned_loss=0.04659, over 4924.00 frames. ], tot_loss[loss=0.1845, simple_loss=0.2527, pruned_loss=0.05812, over 954897.82 frames. ], batch size: 37, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:25:41,383 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.111e+02 1.524e+02 1.821e+02 2.258e+02 3.585e+02, threshold=3.643e+02, percent-clipped=2.0 2023-03-26 18:25:42,590 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83345.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:26:12,520 INFO [finetune.py:976] (3/7) Epoch 15, batch 3200, loss[loss=0.1713, simple_loss=0.2426, pruned_loss=0.05001, over 4776.00 frames. ], tot_loss[loss=0.1805, simple_loss=0.2485, pruned_loss=0.05624, over 954022.32 frames. ], batch size: 26, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:26:25,325 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.25 vs. limit=5.0 2023-03-26 18:26:47,906 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.65 vs. limit=5.0 2023-03-26 18:26:55,876 INFO [finetune.py:976] (3/7) Epoch 15, batch 3250, loss[loss=0.1841, simple_loss=0.2568, pruned_loss=0.05571, over 4924.00 frames. ], tot_loss[loss=0.1814, simple_loss=0.2493, pruned_loss=0.05681, over 952192.14 frames. ], batch size: 38, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:27:00,083 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.648e+01 1.539e+02 1.854e+02 2.232e+02 3.646e+02, threshold=3.708e+02, percent-clipped=1.0 2023-03-26 18:27:03,145 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2767, 2.1217, 1.6688, 2.0791, 2.1218, 1.7873, 2.4269, 2.2223], device='cuda:3'), covar=tensor([0.1291, 0.2175, 0.3295, 0.2733, 0.2678, 0.1781, 0.3182, 0.1705], device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0186, 0.0234, 0.0252, 0.0243, 0.0200, 0.0211, 0.0197], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 18:27:29,590 INFO [finetune.py:976] (3/7) Epoch 15, batch 3300, loss[loss=0.1512, simple_loss=0.2215, pruned_loss=0.04048, over 4829.00 frames. ], tot_loss[loss=0.1839, simple_loss=0.2522, pruned_loss=0.05777, over 952938.74 frames. ], batch size: 25, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:27:38,126 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-03-26 18:27:49,499 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8539, 1.4258, 1.9408, 1.8033, 1.6064, 1.5556, 1.7184, 1.7252], device='cuda:3'), covar=tensor([0.3915, 0.4112, 0.3271, 0.3849, 0.4918, 0.3750, 0.4471, 0.3088], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0239, 0.0257, 0.0269, 0.0268, 0.0240, 0.0281, 0.0236], 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') 2023-03-26 18:28:07,467 INFO [finetune.py:976] (3/7) Epoch 15, batch 3350, loss[loss=0.1907, simple_loss=0.2553, pruned_loss=0.06307, over 4806.00 frames. ], tot_loss[loss=0.1872, simple_loss=0.2554, pruned_loss=0.05951, over 953413.93 frames. ], batch size: 25, lr: 3.49e-03, grad_scale: 64.0 2023-03-26 18:28:14,626 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.071e+02 1.792e+02 2.041e+02 2.510e+02 5.102e+02, threshold=4.082e+02, percent-clipped=3.0 2023-03-26 18:28:21,870 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83552.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:28:32,676 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4637, 1.3334, 1.3793, 1.3572, 0.8604, 2.1216, 0.7888, 1.2623], device='cuda:3'), covar=tensor([0.3191, 0.2404, 0.2073, 0.2329, 0.1862, 0.0395, 0.2514, 0.1261], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0116, 0.0120, 0.0124, 0.0115, 0.0098, 0.0097, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 18:28:54,245 INFO [finetune.py:976] (3/7) Epoch 15, batch 3400, loss[loss=0.2126, simple_loss=0.2832, pruned_loss=0.07094, over 4840.00 frames. ], tot_loss[loss=0.1877, simple_loss=0.2561, pruned_loss=0.05963, over 951962.56 frames. ], batch size: 49, lr: 3.49e-03, grad_scale: 64.0 2023-03-26 18:29:24,883 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83619.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:29:37,310 INFO [finetune.py:976] (3/7) Epoch 15, batch 3450, loss[loss=0.1897, simple_loss=0.2635, pruned_loss=0.05797, over 4745.00 frames. ], tot_loss[loss=0.1881, simple_loss=0.2565, pruned_loss=0.05987, over 952784.93 frames. ], batch size: 54, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:29:39,046 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83640.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:29:41,996 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.066e+02 1.651e+02 1.928e+02 2.236e+02 3.717e+02, threshold=3.855e+02, percent-clipped=0.0 2023-03-26 18:29:52,356 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.27 vs. limit=5.0 2023-03-26 18:29:57,378 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=83667.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:30:11,001 INFO [finetune.py:976] (3/7) Epoch 15, batch 3500, loss[loss=0.2195, simple_loss=0.2729, pruned_loss=0.08303, over 4841.00 frames. ], tot_loss[loss=0.187, simple_loss=0.2549, pruned_loss=0.05958, over 953379.57 frames. ], batch size: 44, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:30:18,461 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.49 vs. limit=5.0 2023-03-26 18:30:44,676 INFO [finetune.py:976] (3/7) Epoch 15, batch 3550, loss[loss=0.1726, simple_loss=0.2425, pruned_loss=0.05136, over 4895.00 frames. ], tot_loss[loss=0.1841, simple_loss=0.2512, pruned_loss=0.05846, over 954447.12 frames. ], batch size: 43, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:30:49,418 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.222e+01 1.573e+02 1.880e+02 2.102e+02 4.250e+02, threshold=3.760e+02, percent-clipped=2.0 2023-03-26 18:30:54,371 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9153, 1.2514, 0.8895, 1.8332, 2.2444, 1.4252, 1.6486, 1.8499], device='cuda:3'), covar=tensor([0.1390, 0.2074, 0.2049, 0.1146, 0.1888, 0.2182, 0.1389, 0.1764], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0095, 0.0111, 0.0093, 0.0119, 0.0094, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 18:30:59,120 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83760.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:31:18,475 INFO [finetune.py:976] (3/7) Epoch 15, batch 3600, loss[loss=0.1831, simple_loss=0.2523, pruned_loss=0.05695, over 4714.00 frames. ], tot_loss[loss=0.1817, simple_loss=0.2489, pruned_loss=0.05726, over 955276.92 frames. ], batch size: 59, lr: 3.49e-03, grad_scale: 32.0 2023-03-26 18:31:48,186 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83821.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:31:59,980 INFO [finetune.py:976] (3/7) Epoch 15, batch 3650, loss[loss=0.2025, simple_loss=0.2767, pruned_loss=0.06411, over 4876.00 frames. ], tot_loss[loss=0.1835, simple_loss=0.251, pruned_loss=0.05797, over 956554.63 frames. ], batch size: 34, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:32:04,778 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.180e+02 1.603e+02 1.941e+02 2.306e+02 4.863e+02, threshold=3.882e+02, percent-clipped=1.0 2023-03-26 18:32:09,567 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83852.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:32:09,634 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7349, 1.0544, 1.7362, 1.7056, 1.4811, 1.4484, 1.6319, 1.5803], device='cuda:3'), covar=tensor([0.3381, 0.3663, 0.2974, 0.3328, 0.4266, 0.3523, 0.3904, 0.2879], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0239, 0.0258, 0.0269, 0.0268, 0.0240, 0.0281, 0.0237], 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') 2023-03-26 18:32:33,847 INFO [finetune.py:976] (3/7) Epoch 15, batch 3700, loss[loss=0.1767, simple_loss=0.2555, pruned_loss=0.04895, over 4923.00 frames. ], tot_loss[loss=0.1865, simple_loss=0.2552, pruned_loss=0.0589, over 954681.95 frames. ], batch size: 38, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:32:41,700 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=83900.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:33:07,588 INFO [finetune.py:976] (3/7) Epoch 15, batch 3750, loss[loss=0.2322, simple_loss=0.2935, pruned_loss=0.08546, over 4925.00 frames. ], tot_loss[loss=0.1887, simple_loss=0.2572, pruned_loss=0.06009, over 955772.79 frames. ], batch size: 42, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:33:08,906 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83940.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:33:11,799 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.005e+02 1.596e+02 1.977e+02 2.275e+02 5.079e+02, threshold=3.955e+02, percent-clipped=1.0 2023-03-26 18:33:14,835 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83949.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:33:21,000 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6465, 1.4957, 1.0188, 0.2544, 1.2230, 1.4379, 1.3839, 1.4109], device='cuda:3'), covar=tensor([0.1002, 0.0833, 0.1442, 0.2054, 0.1494, 0.2371, 0.2315, 0.0871], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0194, 0.0199, 0.0183, 0.0212, 0.0206, 0.0224, 0.0195], 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') 2023-03-26 18:33:32,223 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.5606, 1.5865, 1.6363, 0.8102, 1.7243, 1.9431, 1.7909, 1.4910], device='cuda:3'), covar=tensor([0.1044, 0.0733, 0.0501, 0.0711, 0.0561, 0.0601, 0.0431, 0.0849], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0151, 0.0124, 0.0128, 0.0130, 0.0126, 0.0141, 0.0146], device='cuda:3'), out_proj_covar=tensor([9.1829e-05, 1.0960e-04, 8.8797e-05, 9.1832e-05, 9.2236e-05, 9.0969e-05, 1.0157e-04, 1.0605e-04], device='cuda:3') 2023-03-26 18:33:47,403 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1417, 1.9597, 2.6297, 4.1553, 2.8314, 2.9138, 0.8415, 3.4421], device='cuda:3'), covar=tensor([0.1804, 0.1436, 0.1307, 0.0436, 0.0749, 0.1297, 0.2090, 0.0446], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0116, 0.0131, 0.0162, 0.0099, 0.0136, 0.0124, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 18:33:53,423 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83985.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:33:55,619 INFO [finetune.py:976] (3/7) Epoch 15, batch 3800, loss[loss=0.2667, simple_loss=0.3225, pruned_loss=0.1055, over 4318.00 frames. ], tot_loss[loss=0.1883, simple_loss=0.2575, pruned_loss=0.05958, over 955793.03 frames. ], batch size: 66, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:33:55,677 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=83988.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:34:11,215 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84010.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 18:34:11,876 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-03-26 18:34:36,934 INFO [finetune.py:976] (3/7) Epoch 15, batch 3850, loss[loss=0.1487, simple_loss=0.2178, pruned_loss=0.03982, over 4924.00 frames. ], tot_loss[loss=0.1858, simple_loss=0.2549, pruned_loss=0.0584, over 954509.58 frames. ], batch size: 46, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:34:41,717 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.557e+01 1.496e+02 1.862e+02 2.338e+02 3.560e+02, threshold=3.724e+02, percent-clipped=0.0 2023-03-26 18:34:42,461 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84046.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:34:46,561 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7178, 0.7778, 1.7403, 1.6489, 1.5361, 1.4655, 1.5529, 1.6688], device='cuda:3'), covar=tensor([0.3557, 0.3838, 0.3036, 0.3356, 0.4278, 0.3290, 0.3896, 0.2938], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0239, 0.0257, 0.0268, 0.0267, 0.0240, 0.0280, 0.0236], 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') 2023-03-26 18:34:47,134 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1694, 1.9391, 1.4050, 0.5346, 1.7336, 1.7594, 1.5389, 1.8061], device='cuda:3'), covar=tensor([0.0842, 0.0667, 0.1426, 0.1829, 0.1175, 0.2128, 0.2265, 0.0809], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0195, 0.0201, 0.0184, 0.0214, 0.0207, 0.0224, 0.0196], 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') 2023-03-26 18:35:01,361 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6331, 0.6686, 1.7261, 1.5556, 1.4728, 1.4200, 1.4943, 1.6083], device='cuda:3'), covar=tensor([0.3604, 0.3922, 0.3208, 0.3594, 0.4300, 0.3464, 0.4294, 0.2971], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0240, 0.0258, 0.0269, 0.0268, 0.0241, 0.0281, 0.0237], 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') 2023-03-26 18:35:05,777 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.79 vs. limit=5.0 2023-03-26 18:35:10,033 INFO [finetune.py:976] (3/7) Epoch 15, batch 3900, loss[loss=0.1944, simple_loss=0.2592, pruned_loss=0.0648, over 4849.00 frames. ], tot_loss[loss=0.1855, simple_loss=0.2536, pruned_loss=0.05875, over 956023.22 frames. ], batch size: 44, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:35:16,522 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84097.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:35:19,960 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8975, 3.8384, 3.6172, 1.9470, 3.9299, 2.9743, 0.8729, 2.7780], device='cuda:3'), covar=tensor([0.2293, 0.1859, 0.1607, 0.3363, 0.1079, 0.1024, 0.4622, 0.1511], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0175, 0.0160, 0.0128, 0.0159, 0.0123, 0.0146, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 18:35:28,910 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84116.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:35:36,744 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4929, 2.2679, 1.7799, 0.8619, 2.1142, 1.8658, 1.6705, 2.0269], device='cuda:3'), covar=tensor([0.0894, 0.0903, 0.1843, 0.2258, 0.1509, 0.2484, 0.2539, 0.1066], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0194, 0.0200, 0.0183, 0.0213, 0.0207, 0.0223, 0.0195], 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') 2023-03-26 18:35:43,582 INFO [finetune.py:976] (3/7) Epoch 15, batch 3950, loss[loss=0.1499, simple_loss=0.2238, pruned_loss=0.03803, over 4909.00 frames. ], tot_loss[loss=0.1831, simple_loss=0.2504, pruned_loss=0.05791, over 955218.10 frames. ], batch size: 36, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:35:47,770 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.097e+02 1.469e+02 1.885e+02 2.278e+02 4.120e+02, threshold=3.770e+02, percent-clipped=1.0 2023-03-26 18:35:48,442 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3948, 2.1804, 2.0003, 2.3990, 2.2282, 2.2070, 2.1635, 3.1733], device='cuda:3'), covar=tensor([0.4121, 0.5978, 0.3860, 0.4895, 0.4456, 0.2871, 0.5353, 0.1974], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0259, 0.0226, 0.0275, 0.0247, 0.0215, 0.0249, 0.0226], 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') 2023-03-26 18:35:53,073 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0865, 1.7673, 2.3368, 1.5954, 2.2241, 2.3190, 1.7258, 2.5610], device='cuda:3'), covar=tensor([0.1260, 0.1847, 0.1366, 0.1894, 0.0792, 0.1330, 0.2466, 0.0758], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0204, 0.0193, 0.0190, 0.0176, 0.0213, 0.0218, 0.0201], 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') 2023-03-26 18:35:57,216 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84158.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:36:03,831 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5305, 1.4363, 1.2609, 1.4609, 1.7967, 1.6483, 1.4768, 1.2648], device='cuda:3'), covar=tensor([0.0332, 0.0294, 0.0615, 0.0295, 0.0190, 0.0483, 0.0300, 0.0446], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0108, 0.0141, 0.0113, 0.0100, 0.0106, 0.0097, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.2225e-05, 8.3967e-05, 1.1197e-04, 8.7295e-05, 7.8014e-05, 7.8644e-05, 7.3136e-05, 8.2573e-05], device='cuda:3') 2023-03-26 18:36:16,806 INFO [finetune.py:976] (3/7) Epoch 15, batch 4000, loss[loss=0.2084, simple_loss=0.276, pruned_loss=0.07045, over 4926.00 frames. ], tot_loss[loss=0.1823, simple_loss=0.2497, pruned_loss=0.05739, over 954851.60 frames. ], batch size: 38, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:36:24,527 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84199.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:36:48,108 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.64 vs. limit=5.0 2023-03-26 18:36:57,639 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84236.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:36:59,240 INFO [finetune.py:976] (3/7) Epoch 15, batch 4050, loss[loss=0.1562, simple_loss=0.234, pruned_loss=0.0392, over 4795.00 frames. ], tot_loss[loss=0.1853, simple_loss=0.253, pruned_loss=0.05877, over 952108.96 frames. ], batch size: 29, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:37:07,827 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.087e+02 1.586e+02 1.909e+02 2.268e+02 5.729e+02, threshold=3.818e+02, percent-clipped=2.0 2023-03-26 18:37:21,560 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84260.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:37:23,899 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0 2023-03-26 18:37:39,953 INFO [finetune.py:976] (3/7) Epoch 15, batch 4100, loss[loss=0.2554, simple_loss=0.3108, pruned_loss=0.1, over 4808.00 frames. ], tot_loss[loss=0.1875, simple_loss=0.2558, pruned_loss=0.05953, over 954166.65 frames. ], batch size: 40, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:37:46,472 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84297.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:37:52,181 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84305.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 18:38:10,157 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-03-26 18:38:11,865 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84336.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:38:13,431 INFO [finetune.py:976] (3/7) Epoch 15, batch 4150, loss[loss=0.2281, simple_loss=0.2887, pruned_loss=0.08375, over 4897.00 frames. ], tot_loss[loss=0.1881, simple_loss=0.2569, pruned_loss=0.05966, over 953015.17 frames. ], batch size: 37, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:38:15,328 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84341.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:38:18,112 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.082e+02 1.618e+02 1.997e+02 2.307e+02 7.274e+02, threshold=3.993e+02, percent-clipped=3.0 2023-03-26 18:38:38,766 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9677, 4.7402, 4.5246, 2.5709, 4.8098, 3.7468, 1.1385, 3.4940], device='cuda:3'), covar=tensor([0.2215, 0.1418, 0.1217, 0.2922, 0.0858, 0.0814, 0.4264, 0.1222], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0174, 0.0159, 0.0128, 0.0158, 0.0122, 0.0145, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 18:38:50,308 INFO [finetune.py:976] (3/7) Epoch 15, batch 4200, loss[loss=0.1775, simple_loss=0.2535, pruned_loss=0.05073, over 4861.00 frames. ], tot_loss[loss=0.1882, simple_loss=0.2576, pruned_loss=0.05943, over 953294.91 frames. ], batch size: 31, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:39:04,824 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84397.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:39:17,875 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84416.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:39:31,951 INFO [finetune.py:976] (3/7) Epoch 15, batch 4250, loss[loss=0.1524, simple_loss=0.2191, pruned_loss=0.04284, over 4910.00 frames. ], tot_loss[loss=0.1866, simple_loss=0.2555, pruned_loss=0.05888, over 956033.18 frames. ], batch size: 36, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:39:36,673 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.324e+01 1.559e+02 1.825e+02 2.300e+02 4.289e+02, threshold=3.650e+02, percent-clipped=1.0 2023-03-26 18:39:47,481 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84453.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:39:58,789 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=84464.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:40:07,263 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6633, 1.5501, 1.5342, 1.5771, 1.0012, 3.2339, 1.2129, 1.6555], device='cuda:3'), covar=tensor([0.3313, 0.2485, 0.2096, 0.2326, 0.1885, 0.0221, 0.2580, 0.1284], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0120, 0.0123, 0.0114, 0.0097, 0.0096, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 18:40:14,295 INFO [finetune.py:976] (3/7) Epoch 15, batch 4300, loss[loss=0.1602, simple_loss=0.2314, pruned_loss=0.04452, over 4913.00 frames. ], tot_loss[loss=0.1844, simple_loss=0.2524, pruned_loss=0.05823, over 956313.81 frames. ], batch size: 37, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:40:39,120 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84525.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:40:47,841 INFO [finetune.py:976] (3/7) Epoch 15, batch 4350, loss[loss=0.1905, simple_loss=0.2499, pruned_loss=0.0655, over 4913.00 frames. ], tot_loss[loss=0.1821, simple_loss=0.2495, pruned_loss=0.05728, over 954679.80 frames. ], batch size: 43, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:40:52,222 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.160e+02 1.502e+02 1.820e+02 2.196e+02 3.984e+02, threshold=3.641e+02, percent-clipped=2.0 2023-03-26 18:40:58,846 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84555.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:41:19,645 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84586.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:41:21,101 INFO [finetune.py:976] (3/7) Epoch 15, batch 4400, loss[loss=0.2694, simple_loss=0.3316, pruned_loss=0.1036, over 4737.00 frames. ], tot_loss[loss=0.1848, simple_loss=0.2521, pruned_loss=0.05878, over 953132.38 frames. ], batch size: 59, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:41:24,131 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84592.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:41:32,781 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84605.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:41:51,839 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84633.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:41:54,826 INFO [finetune.py:976] (3/7) Epoch 15, batch 4450, loss[loss=0.2069, simple_loss=0.2876, pruned_loss=0.06313, over 4762.00 frames. ], tot_loss[loss=0.187, simple_loss=0.2549, pruned_loss=0.05951, over 952011.99 frames. ], batch size: 54, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:41:57,720 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.4573, 3.8843, 4.0854, 4.2062, 4.2258, 3.9640, 4.5044, 1.7518], device='cuda:3'), covar=tensor([0.0697, 0.0806, 0.0680, 0.0859, 0.1063, 0.1375, 0.0612, 0.4983], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0244, 0.0276, 0.0292, 0.0334, 0.0284, 0.0299, 0.0298], 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') 2023-03-26 18:41:57,740 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84641.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:42:01,982 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.076e+02 1.595e+02 1.952e+02 2.292e+02 4.719e+02, threshold=3.904e+02, percent-clipped=1.0 2023-03-26 18:42:11,296 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=84653.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:42:22,161 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3761, 2.2018, 1.8512, 2.2730, 2.2072, 1.9576, 2.5646, 2.3429], device='cuda:3'), covar=tensor([0.1400, 0.2263, 0.3133, 0.2864, 0.2841, 0.1868, 0.3157, 0.1765], device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0187, 0.0233, 0.0253, 0.0244, 0.0201, 0.0211, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 18:42:46,433 INFO [finetune.py:976] (3/7) Epoch 15, batch 4500, loss[loss=0.1846, simple_loss=0.2625, pruned_loss=0.05339, over 4798.00 frames. ], tot_loss[loss=0.1869, simple_loss=0.2557, pruned_loss=0.05898, over 953708.10 frames. ], batch size: 45, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:42:47,111 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=84689.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:42:49,432 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84692.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:42:50,724 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84694.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:43:09,612 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3098, 2.1582, 1.7268, 2.2349, 2.1464, 1.9213, 2.5523, 2.3212], device='cuda:3'), covar=tensor([0.1363, 0.2150, 0.3244, 0.2601, 0.2686, 0.1749, 0.3187, 0.1886], device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0188, 0.0234, 0.0254, 0.0244, 0.0201, 0.0212, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 18:43:20,117 INFO [finetune.py:976] (3/7) Epoch 15, batch 4550, loss[loss=0.1401, simple_loss=0.2081, pruned_loss=0.03607, over 4736.00 frames. ], tot_loss[loss=0.1873, simple_loss=0.256, pruned_loss=0.05931, over 954719.95 frames. ], batch size: 23, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:43:25,284 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 1.592e+02 2.005e+02 2.406e+02 4.528e+02, threshold=4.009e+02, percent-clipped=3.0 2023-03-26 18:43:30,253 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84753.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:43:53,702 INFO [finetune.py:976] (3/7) Epoch 15, batch 4600, loss[loss=0.1542, simple_loss=0.2282, pruned_loss=0.04011, over 4843.00 frames. ], tot_loss[loss=0.1857, simple_loss=0.2547, pruned_loss=0.05835, over 954961.06 frames. ], batch size: 49, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:44:06,849 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84800.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:44:07,398 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=84801.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:44:36,214 INFO [finetune.py:976] (3/7) Epoch 15, batch 4650, loss[loss=0.1766, simple_loss=0.2491, pruned_loss=0.05207, over 4868.00 frames. ], tot_loss[loss=0.1844, simple_loss=0.2524, pruned_loss=0.05822, over 953473.06 frames. ], batch size: 31, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:44:40,389 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.176e+02 1.584e+02 1.933e+02 2.372e+02 3.946e+02, threshold=3.865e+02, percent-clipped=0.0 2023-03-26 18:44:47,565 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84855.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:44:56,259 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84861.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:45:17,964 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84881.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:45:26,374 INFO [finetune.py:976] (3/7) Epoch 15, batch 4700, loss[loss=0.1732, simple_loss=0.2379, pruned_loss=0.05426, over 4786.00 frames. ], tot_loss[loss=0.1824, simple_loss=0.2499, pruned_loss=0.05747, over 955901.79 frames. ], batch size: 26, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:45:29,333 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84892.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:45:36,039 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=84903.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:45:56,876 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.89 vs. limit=5.0 2023-03-26 18:45:59,759 INFO [finetune.py:976] (3/7) Epoch 15, batch 4750, loss[loss=0.1832, simple_loss=0.2526, pruned_loss=0.05692, over 4808.00 frames. ], tot_loss[loss=0.1825, simple_loss=0.2493, pruned_loss=0.05783, over 957854.51 frames. ], batch size: 41, lr: 3.48e-03, grad_scale: 32.0 2023-03-26 18:46:01,519 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=84940.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:46:04,956 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.000e+02 1.651e+02 1.892e+02 2.436e+02 4.596e+02, threshold=3.784e+02, percent-clipped=1.0 2023-03-26 18:46:08,762 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84951.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:46:09,983 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-26 18:46:33,699 INFO [finetune.py:976] (3/7) Epoch 15, batch 4800, loss[loss=0.1937, simple_loss=0.2663, pruned_loss=0.06051, over 4151.00 frames. ], tot_loss[loss=0.1855, simple_loss=0.2528, pruned_loss=0.05907, over 957757.69 frames. ], batch size: 65, lr: 3.47e-03, grad_scale: 32.0 2023-03-26 18:46:34,863 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84989.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:46:36,740 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84992.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:46:50,492 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85012.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:47:07,603 INFO [finetune.py:976] (3/7) Epoch 15, batch 4850, loss[loss=0.1756, simple_loss=0.2541, pruned_loss=0.04853, over 4891.00 frames. ], tot_loss[loss=0.1871, simple_loss=0.2556, pruned_loss=0.05929, over 956995.54 frames. ], batch size: 43, lr: 3.47e-03, grad_scale: 32.0 2023-03-26 18:47:09,348 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=85040.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:47:12,315 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.155e+02 1.585e+02 1.858e+02 2.141e+02 6.123e+02, threshold=3.716e+02, percent-clipped=1.0 2023-03-26 18:47:50,165 INFO [finetune.py:976] (3/7) Epoch 15, batch 4900, loss[loss=0.2352, simple_loss=0.3089, pruned_loss=0.08082, over 4912.00 frames. ], tot_loss[loss=0.1878, simple_loss=0.2564, pruned_loss=0.05957, over 953810.85 frames. ], batch size: 38, lr: 3.47e-03, grad_scale: 32.0 2023-03-26 18:47:59,524 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3856, 1.3461, 1.4425, 1.6003, 1.5663, 2.9374, 1.2575, 1.4579], device='cuda:3'), covar=tensor([0.0987, 0.1801, 0.1199, 0.0994, 0.1549, 0.0263, 0.1525, 0.1680], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0074, 0.0078, 0.0092, 0.0081, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 18:48:06,016 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85107.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:48:11,341 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85115.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:48:26,707 INFO [finetune.py:976] (3/7) Epoch 15, batch 4950, loss[loss=0.182, simple_loss=0.249, pruned_loss=0.05748, over 4712.00 frames. ], tot_loss[loss=0.1885, simple_loss=0.2579, pruned_loss=0.05954, over 952661.70 frames. ], batch size: 54, lr: 3.47e-03, grad_scale: 32.0 2023-03-26 18:48:31,434 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.158e+02 1.624e+02 1.884e+02 2.194e+02 3.725e+02, threshold=3.769e+02, percent-clipped=1.0 2023-03-26 18:48:39,199 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85156.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:48:47,012 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85168.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 18:48:52,416 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85176.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:48:55,828 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85181.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:49:00,443 INFO [finetune.py:976] (3/7) Epoch 15, batch 5000, loss[loss=0.2213, simple_loss=0.2786, pruned_loss=0.08204, over 4842.00 frames. ], tot_loss[loss=0.1874, simple_loss=0.2569, pruned_loss=0.05891, over 955011.17 frames. ], batch size: 49, lr: 3.47e-03, grad_scale: 32.0 2023-03-26 18:49:01,167 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.1872, 2.7800, 3.3054, 2.3434, 3.0547, 3.4239, 2.7539, 3.4933], device='cuda:3'), covar=tensor([0.1093, 0.1342, 0.1141, 0.1632, 0.0841, 0.1163, 0.1827, 0.0787], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0200, 0.0190, 0.0188, 0.0174, 0.0210, 0.0215, 0.0198], 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') 2023-03-26 18:49:36,186 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=85229.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:49:42,103 INFO [finetune.py:976] (3/7) Epoch 15, batch 5050, loss[loss=0.1601, simple_loss=0.231, pruned_loss=0.04459, over 4890.00 frames. ], tot_loss[loss=0.1847, simple_loss=0.2536, pruned_loss=0.05789, over 955640.60 frames. ], batch size: 35, lr: 3.47e-03, grad_scale: 32.0 2023-03-26 18:49:46,822 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.063e+02 1.593e+02 1.872e+02 2.269e+02 5.264e+02, threshold=3.745e+02, percent-clipped=1.0 2023-03-26 18:50:13,236 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0568, 1.7889, 1.5705, 1.6450, 1.7210, 1.7809, 1.7725, 2.4574], device='cuda:3'), covar=tensor([0.3898, 0.4493, 0.3376, 0.4248, 0.4229, 0.2413, 0.3875, 0.1818], device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0259, 0.0225, 0.0274, 0.0247, 0.0214, 0.0248, 0.0226], 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') 2023-03-26 18:50:22,633 INFO [finetune.py:976] (3/7) Epoch 15, batch 5100, loss[loss=0.1691, simple_loss=0.2351, pruned_loss=0.05158, over 4820.00 frames. ], tot_loss[loss=0.1814, simple_loss=0.2501, pruned_loss=0.05637, over 956388.44 frames. ], batch size: 41, lr: 3.47e-03, grad_scale: 32.0 2023-03-26 18:50:23,308 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85289.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:50:42,799 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85307.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:50:56,528 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-26 18:50:58,764 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1488, 2.0794, 1.6485, 2.0516, 2.0960, 1.7911, 2.4917, 2.1407], device='cuda:3'), covar=tensor([0.1465, 0.2133, 0.3252, 0.2664, 0.2685, 0.1916, 0.2857, 0.1831], device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0187, 0.0234, 0.0254, 0.0245, 0.0201, 0.0212, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 18:51:02,804 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=85337.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:51:03,347 INFO [finetune.py:976] (3/7) Epoch 15, batch 5150, loss[loss=0.1872, simple_loss=0.244, pruned_loss=0.06519, over 4780.00 frames. ], tot_loss[loss=0.1823, simple_loss=0.2503, pruned_loss=0.0571, over 957536.87 frames. ], batch size: 29, lr: 3.47e-03, grad_scale: 32.0 2023-03-26 18:51:08,102 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.937e+01 1.526e+02 1.888e+02 2.256e+02 3.382e+02, threshold=3.776e+02, percent-clipped=0.0 2023-03-26 18:51:10,063 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.2664, 2.8947, 3.0227, 3.2217, 3.0212, 2.8652, 3.3313, 0.9819], device='cuda:3'), covar=tensor([0.1158, 0.1169, 0.1050, 0.1117, 0.1747, 0.1689, 0.1130, 0.5476], device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0243, 0.0272, 0.0289, 0.0331, 0.0280, 0.0296, 0.0294], 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') 2023-03-26 18:51:25,686 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.94 vs. limit=5.0 2023-03-26 18:51:37,046 INFO [finetune.py:976] (3/7) Epoch 15, batch 5200, loss[loss=0.2211, simple_loss=0.2851, pruned_loss=0.07851, over 4897.00 frames. ], tot_loss[loss=0.1866, simple_loss=0.2548, pruned_loss=0.05925, over 957313.04 frames. ], batch size: 32, lr: 3.47e-03, grad_scale: 32.0 2023-03-26 18:51:41,592 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-26 18:52:04,160 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85428.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:52:05,469 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1460, 2.0044, 1.6936, 1.9326, 1.9047, 1.9248, 1.9821, 2.6884], device='cuda:3'), covar=tensor([0.3841, 0.4373, 0.3431, 0.3802, 0.3967, 0.2297, 0.3743, 0.1710], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0259, 0.0226, 0.0275, 0.0247, 0.0214, 0.0248, 0.0226], 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') 2023-03-26 18:52:10,616 INFO [finetune.py:976] (3/7) Epoch 15, batch 5250, loss[loss=0.1692, simple_loss=0.2384, pruned_loss=0.04999, over 4812.00 frames. ], tot_loss[loss=0.1868, simple_loss=0.2559, pruned_loss=0.05884, over 957984.39 frames. ], batch size: 25, lr: 3.47e-03, grad_scale: 32.0 2023-03-26 18:52:15,820 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.083e+02 1.657e+02 1.928e+02 2.523e+02 8.274e+02, threshold=3.856e+02, percent-clipped=2.0 2023-03-26 18:52:16,632 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-26 18:52:22,853 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-26 18:52:23,086 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85456.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:52:27,687 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85463.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 18:52:33,025 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85471.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:52:39,065 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0 2023-03-26 18:52:40,790 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85483.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:52:44,235 INFO [finetune.py:976] (3/7) Epoch 15, batch 5300, loss[loss=0.1697, simple_loss=0.238, pruned_loss=0.05073, over 4815.00 frames. ], tot_loss[loss=0.1881, simple_loss=0.2571, pruned_loss=0.05961, over 955638.96 frames. ], batch size: 38, lr: 3.47e-03, grad_scale: 32.0 2023-03-26 18:52:44,952 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85489.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:52:54,981 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=85504.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:52:57,330 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4301, 1.1409, 1.6962, 2.7392, 1.7653, 2.2317, 0.8523, 2.3245], device='cuda:3'), covar=tensor([0.1985, 0.2145, 0.1598, 0.1046, 0.1103, 0.1440, 0.2149, 0.0748], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0117, 0.0133, 0.0164, 0.0100, 0.0138, 0.0125, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 18:53:19,697 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85530.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:53:24,909 INFO [finetune.py:976] (3/7) Epoch 15, batch 5350, loss[loss=0.1696, simple_loss=0.2472, pruned_loss=0.04598, over 4882.00 frames. ], tot_loss[loss=0.1871, simple_loss=0.2566, pruned_loss=0.05875, over 956203.67 frames. ], batch size: 35, lr: 3.47e-03, grad_scale: 32.0 2023-03-26 18:53:28,691 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85544.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:53:29,176 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.094e+02 1.542e+02 1.806e+02 2.197e+02 4.190e+02, threshold=3.613e+02, percent-clipped=2.0 2023-03-26 18:53:42,587 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7250, 1.7609, 2.2239, 3.5850, 2.5237, 2.4858, 1.4100, 2.9276], device='cuda:3'), covar=tensor([0.1673, 0.1291, 0.1235, 0.0529, 0.0717, 0.1408, 0.1611, 0.0472], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0116, 0.0132, 0.0163, 0.0099, 0.0138, 0.0124, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 18:53:58,026 INFO [finetune.py:976] (3/7) Epoch 15, batch 5400, loss[loss=0.1948, simple_loss=0.2569, pruned_loss=0.06635, over 4747.00 frames. ], tot_loss[loss=0.1863, simple_loss=0.2547, pruned_loss=0.05892, over 957116.53 frames. ], batch size: 54, lr: 3.47e-03, grad_scale: 32.0 2023-03-26 18:54:00,477 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85591.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:54:08,271 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3720, 1.5266, 1.9190, 1.7573, 1.6399, 3.4837, 1.4604, 1.7231], device='cuda:3'), covar=tensor([0.0960, 0.1826, 0.1044, 0.1016, 0.1567, 0.0252, 0.1432, 0.1621], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0073, 0.0078, 0.0092, 0.0080, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 18:54:11,192 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85607.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:54:23,550 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85625.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:54:31,764 INFO [finetune.py:976] (3/7) Epoch 15, batch 5450, loss[loss=0.1742, simple_loss=0.2501, pruned_loss=0.04913, over 4915.00 frames. ], tot_loss[loss=0.1841, simple_loss=0.2518, pruned_loss=0.05822, over 956796.72 frames. ], batch size: 36, lr: 3.47e-03, grad_scale: 64.0 2023-03-26 18:54:41,082 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.884e+01 1.516e+02 1.902e+02 2.390e+02 5.288e+02, threshold=3.804e+02, percent-clipped=4.0 2023-03-26 18:54:52,062 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=85655.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:55:16,864 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85686.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 18:55:17,972 INFO [finetune.py:976] (3/7) Epoch 15, batch 5500, loss[loss=0.1573, simple_loss=0.2335, pruned_loss=0.04062, over 4834.00 frames. ], tot_loss[loss=0.1813, simple_loss=0.2488, pruned_loss=0.05695, over 957973.76 frames. ], batch size: 30, lr: 3.47e-03, grad_scale: 64.0 2023-03-26 18:55:27,781 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3303, 2.2226, 1.8755, 2.3581, 2.1368, 2.1516, 2.1106, 3.2124], device='cuda:3'), covar=tensor([0.3890, 0.5138, 0.3503, 0.4538, 0.4692, 0.2479, 0.5077, 0.1541], device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0259, 0.0225, 0.0274, 0.0247, 0.0214, 0.0248, 0.0225], 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') 2023-03-26 18:55:35,002 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8264, 1.6798, 1.4288, 1.2818, 1.8389, 1.5489, 2.0413, 1.7911], device='cuda:3'), covar=tensor([0.1296, 0.1594, 0.3007, 0.2324, 0.2342, 0.1612, 0.1971, 0.1713], device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0189, 0.0235, 0.0256, 0.0246, 0.0203, 0.0214, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 18:56:02,558 INFO [finetune.py:976] (3/7) Epoch 15, batch 5550, loss[loss=0.1455, simple_loss=0.2223, pruned_loss=0.03433, over 4777.00 frames. ], tot_loss[loss=0.1827, simple_loss=0.2502, pruned_loss=0.05756, over 957179.13 frames. ], batch size: 26, lr: 3.47e-03, grad_scale: 64.0 2023-03-26 18:56:06,721 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.623e+01 1.589e+02 1.875e+02 2.150e+02 4.153e+02, threshold=3.750e+02, percent-clipped=1.0 2023-03-26 18:56:19,298 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85763.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:56:24,629 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85771.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:56:26,380 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5584, 2.4087, 2.0163, 2.4202, 2.3914, 2.1620, 2.7528, 2.4815], device='cuda:3'), covar=tensor([0.1266, 0.1954, 0.3001, 0.2447, 0.2529, 0.1669, 0.2943, 0.1684], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0189, 0.0236, 0.0256, 0.0246, 0.0204, 0.0214, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 18:56:32,631 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85784.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:56:34,953 INFO [finetune.py:976] (3/7) Epoch 15, batch 5600, loss[loss=0.2154, simple_loss=0.2897, pruned_loss=0.07052, over 4815.00 frames. ], tot_loss[loss=0.1871, simple_loss=0.2552, pruned_loss=0.05944, over 955200.95 frames. ], batch size: 38, lr: 3.47e-03, grad_scale: 64.0 2023-03-26 18:56:48,430 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=85811.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:56:53,121 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=85819.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:56:55,529 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.59 vs. limit=2.0 2023-03-26 18:57:04,151 INFO [finetune.py:976] (3/7) Epoch 15, batch 5650, loss[loss=0.2005, simple_loss=0.2761, pruned_loss=0.06245, over 4909.00 frames. ], tot_loss[loss=0.1888, simple_loss=0.2579, pruned_loss=0.05984, over 954914.72 frames. ], batch size: 37, lr: 3.47e-03, grad_scale: 64.0 2023-03-26 18:57:04,770 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85839.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:57:08,245 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.126e+02 1.563e+02 1.888e+02 2.328e+02 3.522e+02, threshold=3.776e+02, percent-clipped=0.0 2023-03-26 18:57:26,023 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85875.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:57:32,569 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85886.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:57:33,724 INFO [finetune.py:976] (3/7) Epoch 15, batch 5700, loss[loss=0.1558, simple_loss=0.2143, pruned_loss=0.04868, over 4437.00 frames. ], tot_loss[loss=0.185, simple_loss=0.2533, pruned_loss=0.05831, over 938970.17 frames. ], batch size: 19, lr: 3.47e-03, grad_scale: 64.0 2023-03-26 18:57:41,052 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.54 vs. limit=5.0 2023-03-26 18:57:43,541 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-03-26 18:58:02,782 INFO [finetune.py:976] (3/7) Epoch 16, batch 0, loss[loss=0.1796, simple_loss=0.2547, pruned_loss=0.05229, over 4792.00 frames. ], tot_loss[loss=0.1796, simple_loss=0.2547, pruned_loss=0.05229, over 4792.00 frames. ], batch size: 29, lr: 3.46e-03, grad_scale: 64.0 2023-03-26 18:58:02,782 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 18:58:15,091 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9059, 1.4327, 1.0389, 1.6758, 2.1643, 1.2132, 1.7150, 1.6950], device='cuda:3'), covar=tensor([0.1197, 0.1718, 0.1594, 0.1062, 0.1597, 0.1941, 0.1115, 0.1781], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0095, 0.0110, 0.0093, 0.0119, 0.0094, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 18:58:17,920 INFO [finetune.py:1010] (3/7) Epoch 16, validation: loss=0.1572, simple_loss=0.2278, pruned_loss=0.04329, over 2265189.00 frames. 2023-03-26 18:58:17,921 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6434MB 2023-03-26 18:58:22,807 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4530, 1.7320, 1.3873, 1.5270, 1.9229, 1.9155, 1.6689, 1.7494], device='cuda:3'), covar=tensor([0.0535, 0.0314, 0.0556, 0.0317, 0.0264, 0.0475, 0.0378, 0.0308], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0109, 0.0143, 0.0113, 0.0100, 0.0107, 0.0097, 0.0109], device='cuda:3'), out_proj_covar=tensor([7.2658e-05, 8.4242e-05, 1.1341e-04, 8.7361e-05, 7.8285e-05, 7.8926e-05, 7.3140e-05, 8.2796e-05], device='cuda:3') 2023-03-26 18:58:25,997 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-26 18:58:26,972 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85930.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 18:58:29,350 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4286, 1.3133, 1.2927, 1.3328, 0.7743, 2.1860, 0.6958, 1.1507], device='cuda:3'), covar=tensor([0.3026, 0.2295, 0.2023, 0.2240, 0.1962, 0.0396, 0.2717, 0.1296], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0114, 0.0118, 0.0123, 0.0113, 0.0096, 0.0095, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 18:58:30,589 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85936.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:58:36,423 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.062e+02 1.566e+02 1.783e+02 2.274e+02 8.459e+02, threshold=3.567e+02, percent-clipped=4.0 2023-03-26 18:58:40,712 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0041, 1.9632, 1.5819, 2.0196, 2.0863, 1.7005, 2.3682, 2.0653], device='cuda:3'), covar=tensor([0.1310, 0.1969, 0.2850, 0.2380, 0.2205, 0.1597, 0.2902, 0.1683], device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0188, 0.0234, 0.0254, 0.0244, 0.0202, 0.0212, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 18:58:44,319 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85957.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:58:49,654 INFO [finetune.py:976] (3/7) Epoch 16, batch 50, loss[loss=0.2247, simple_loss=0.2914, pruned_loss=0.07896, over 4876.00 frames. ], tot_loss[loss=0.191, simple_loss=0.259, pruned_loss=0.06148, over 215022.11 frames. ], batch size: 35, lr: 3.46e-03, grad_scale: 32.0 2023-03-26 18:58:59,305 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85981.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 18:59:05,875 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85991.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 18:59:14,744 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86002.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:59:17,318 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-03-26 18:59:23,676 INFO [finetune.py:976] (3/7) Epoch 16, batch 100, loss[loss=0.1732, simple_loss=0.2367, pruned_loss=0.05481, over 4834.00 frames. ], tot_loss[loss=0.1834, simple_loss=0.2508, pruned_loss=0.05799, over 378728.18 frames. ], batch size: 30, lr: 3.46e-03, grad_scale: 32.0 2023-03-26 18:59:24,990 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86018.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 18:59:44,127 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.132e+02 1.611e+02 1.877e+02 2.147e+02 3.763e+02, threshold=3.754e+02, percent-clipped=3.0 2023-03-26 18:59:47,924 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86052.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:00:00,118 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86063.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:00:06,620 INFO [finetune.py:976] (3/7) Epoch 16, batch 150, loss[loss=0.1423, simple_loss=0.2165, pruned_loss=0.03405, over 4733.00 frames. ], tot_loss[loss=0.179, simple_loss=0.2456, pruned_loss=0.05618, over 506763.90 frames. ], batch size: 59, lr: 3.46e-03, grad_scale: 32.0 2023-03-26 19:00:08,581 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86069.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:00:27,239 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86084.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:00:36,696 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.10 vs. limit=5.0 2023-03-26 19:00:40,853 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-03-26 19:00:50,259 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86113.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:00:51,950 INFO [finetune.py:976] (3/7) Epoch 16, batch 200, loss[loss=0.2427, simple_loss=0.2896, pruned_loss=0.09797, over 4822.00 frames. ], tot_loss[loss=0.1817, simple_loss=0.2466, pruned_loss=0.05841, over 606573.86 frames. ], batch size: 40, lr: 3.46e-03, grad_scale: 32.0 2023-03-26 19:01:04,032 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86130.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:01:05,146 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=86132.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:01:09,829 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86139.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:01:14,013 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.089e+02 1.565e+02 1.801e+02 2.285e+02 3.660e+02, threshold=3.601e+02, percent-clipped=0.0 2023-03-26 19:01:27,190 INFO [finetune.py:976] (3/7) Epoch 16, batch 250, loss[loss=0.2132, simple_loss=0.2818, pruned_loss=0.0723, over 4833.00 frames. ], tot_loss[loss=0.1851, simple_loss=0.2508, pruned_loss=0.05969, over 682410.15 frames. ], batch size: 40, lr: 3.46e-03, grad_scale: 32.0 2023-03-26 19:01:40,816 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86186.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:01:41,401 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=86187.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:01:46,193 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.05 vs. limit=5.0 2023-03-26 19:02:00,655 INFO [finetune.py:976] (3/7) Epoch 16, batch 300, loss[loss=0.2024, simple_loss=0.2793, pruned_loss=0.06278, over 4841.00 frames. ], tot_loss[loss=0.1887, simple_loss=0.2566, pruned_loss=0.06044, over 744280.90 frames. ], batch size: 49, lr: 3.46e-03, grad_scale: 32.0 2023-03-26 19:02:11,159 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86231.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:02:12,979 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=86234.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:02:13,729 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.82 vs. limit=5.0 2023-03-26 19:02:20,700 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.190e+02 1.625e+02 1.967e+02 2.251e+02 5.649e+02, threshold=3.935e+02, percent-clipped=3.0 2023-03-26 19:02:20,828 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86246.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:02:23,814 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6342, 4.0732, 3.8531, 1.9836, 4.1462, 3.1348, 0.9181, 2.9207], device='cuda:3'), covar=tensor([0.2243, 0.1506, 0.1357, 0.3065, 0.0796, 0.0850, 0.4168, 0.1250], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0173, 0.0159, 0.0127, 0.0157, 0.0122, 0.0145, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 19:02:34,278 INFO [finetune.py:976] (3/7) Epoch 16, batch 350, loss[loss=0.2132, simple_loss=0.2833, pruned_loss=0.07157, over 4869.00 frames. ], tot_loss[loss=0.19, simple_loss=0.2582, pruned_loss=0.06088, over 791256.28 frames. ], batch size: 43, lr: 3.46e-03, grad_scale: 32.0 2023-03-26 19:02:39,872 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.5975, 1.6300, 1.6687, 0.8567, 1.7824, 1.9735, 1.9260, 1.4735], device='cuda:3'), covar=tensor([0.0869, 0.0576, 0.0523, 0.0582, 0.0470, 0.0576, 0.0345, 0.0670], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0151, 0.0124, 0.0128, 0.0129, 0.0127, 0.0142, 0.0146], device='cuda:3'), out_proj_covar=tensor([9.1856e-05, 1.0970e-04, 8.8804e-05, 9.1725e-05, 9.1320e-05, 9.1600e-05, 1.0220e-04, 1.0557e-04], device='cuda:3') 2023-03-26 19:02:44,510 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86281.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:02:47,960 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86286.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 19:03:01,131 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86307.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:03:05,209 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86313.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:03:07,486 INFO [finetune.py:976] (3/7) Epoch 16, batch 400, loss[loss=0.1412, simple_loss=0.2086, pruned_loss=0.03685, over 4798.00 frames. ], tot_loss[loss=0.1884, simple_loss=0.2571, pruned_loss=0.05982, over 827450.80 frames. ], batch size: 25, lr: 3.46e-03, grad_scale: 32.0 2023-03-26 19:03:08,673 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3409, 1.3233, 1.4646, 1.6038, 1.5054, 2.9351, 1.2028, 1.4026], device='cuda:3'), covar=tensor([0.1321, 0.2550, 0.1345, 0.1226, 0.2043, 0.0442, 0.2183, 0.2461], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0074, 0.0078, 0.0091, 0.0080, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 19:03:15,903 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=86329.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:03:34,375 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.307e+01 1.538e+02 1.774e+02 2.172e+02 4.200e+02, threshold=3.548e+02, percent-clipped=1.0 2023-03-26 19:03:35,738 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.5640, 1.5727, 1.5677, 0.8637, 1.7079, 1.8735, 1.8539, 1.4506], device='cuda:3'), covar=tensor([0.0819, 0.0597, 0.0557, 0.0537, 0.0412, 0.0585, 0.0293, 0.0698], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0151, 0.0124, 0.0128, 0.0130, 0.0127, 0.0142, 0.0146], device='cuda:3'), out_proj_covar=tensor([9.2080e-05, 1.1004e-04, 8.8955e-05, 9.1800e-05, 9.1824e-05, 9.1781e-05, 1.0246e-04, 1.0591e-04], device='cuda:3') 2023-03-26 19:03:45,094 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86358.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:03:50,331 INFO [finetune.py:976] (3/7) Epoch 16, batch 450, loss[loss=0.1759, simple_loss=0.2507, pruned_loss=0.05057, over 4769.00 frames. ], tot_loss[loss=0.1858, simple_loss=0.2546, pruned_loss=0.05848, over 855715.44 frames. ], batch size: 26, lr: 3.46e-03, grad_scale: 32.0 2023-03-26 19:04:18,721 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86408.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:04:24,032 INFO [finetune.py:976] (3/7) Epoch 16, batch 500, loss[loss=0.1791, simple_loss=0.2487, pruned_loss=0.05479, over 4729.00 frames. ], tot_loss[loss=0.1839, simple_loss=0.2525, pruned_loss=0.05768, over 878190.53 frames. ], batch size: 59, lr: 3.46e-03, grad_scale: 32.0 2023-03-26 19:04:30,019 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86425.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:04:40,630 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86440.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:04:43,600 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86445.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:04:44,074 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.779e+01 1.556e+02 1.875e+02 2.226e+02 4.465e+02, threshold=3.750e+02, percent-clipped=2.0 2023-03-26 19:04:57,163 INFO [finetune.py:976] (3/7) Epoch 16, batch 550, loss[loss=0.1874, simple_loss=0.2547, pruned_loss=0.06007, over 4812.00 frames. ], tot_loss[loss=0.1811, simple_loss=0.2488, pruned_loss=0.05669, over 895646.16 frames. ], batch size: 51, lr: 3.46e-03, grad_scale: 32.0 2023-03-26 19:05:07,100 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3651, 2.2912, 1.9046, 0.9048, 2.1362, 1.9528, 1.7937, 2.0784], device='cuda:3'), covar=tensor([0.0855, 0.0642, 0.1325, 0.1871, 0.1284, 0.1776, 0.1998, 0.0827], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0196, 0.0200, 0.0184, 0.0213, 0.0207, 0.0224, 0.0196], 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') 2023-03-26 19:05:31,548 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86501.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:05:39,646 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86506.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 19:05:50,204 INFO [finetune.py:976] (3/7) Epoch 16, batch 600, loss[loss=0.2095, simple_loss=0.2692, pruned_loss=0.07491, over 4890.00 frames. ], tot_loss[loss=0.1824, simple_loss=0.25, pruned_loss=0.0574, over 910585.08 frames. ], batch size: 32, lr: 3.46e-03, grad_scale: 32.0 2023-03-26 19:06:04,170 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86531.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:06:14,606 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.101e+02 1.575e+02 1.922e+02 2.222e+02 3.111e+02, threshold=3.844e+02, percent-clipped=0.0 2023-03-26 19:06:15,333 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4178, 1.0263, 0.8003, 1.2644, 1.8528, 0.7722, 1.2059, 1.2686], device='cuda:3'), covar=tensor([0.1591, 0.2272, 0.1849, 0.1358, 0.2069, 0.2262, 0.1641, 0.2015], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0095, 0.0111, 0.0093, 0.0119, 0.0094, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 19:06:26,825 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0624, 1.9689, 1.7244, 1.9845, 1.5873, 4.7770, 1.8085, 2.4956], device='cuda:3'), covar=tensor([0.3335, 0.2452, 0.2145, 0.2321, 0.1584, 0.0105, 0.2468, 0.1228], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0120, 0.0124, 0.0115, 0.0097, 0.0096, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 19:06:27,337 INFO [finetune.py:976] (3/7) Epoch 16, batch 650, loss[loss=0.1352, simple_loss=0.2096, pruned_loss=0.03041, over 4810.00 frames. ], tot_loss[loss=0.1845, simple_loss=0.2532, pruned_loss=0.05797, over 920200.42 frames. ], batch size: 25, lr: 3.46e-03, grad_scale: 32.0 2023-03-26 19:06:36,233 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=86579.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:06:36,300 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5576, 1.6141, 1.2901, 1.6312, 1.9159, 1.7867, 1.5641, 1.3390], device='cuda:3'), covar=tensor([0.0336, 0.0293, 0.0611, 0.0253, 0.0212, 0.0415, 0.0320, 0.0401], device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0106, 0.0141, 0.0111, 0.0098, 0.0105, 0.0096, 0.0107], device='cuda:3'), out_proj_covar=tensor([7.1539e-05, 8.2477e-05, 1.1165e-04, 8.5674e-05, 7.6561e-05, 7.7482e-05, 7.2488e-05, 8.1761e-05], device='cuda:3') 2023-03-26 19:06:37,378 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86580.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:06:39,256 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6936, 1.5674, 1.4419, 1.7827, 1.9931, 1.7649, 1.3196, 1.4365], device='cuda:3'), covar=tensor([0.2210, 0.2048, 0.1905, 0.1528, 0.1654, 0.1179, 0.2315, 0.1858], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0208, 0.0211, 0.0191, 0.0242, 0.0184, 0.0214, 0.0199], 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') 2023-03-26 19:06:41,055 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86586.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 19:06:52,263 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86602.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:06:59,430 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86613.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:07:01,143 INFO [finetune.py:976] (3/7) Epoch 16, batch 700, loss[loss=0.1845, simple_loss=0.229, pruned_loss=0.07005, over 4083.00 frames. ], tot_loss[loss=0.1859, simple_loss=0.255, pruned_loss=0.05842, over 927179.64 frames. ], batch size: 17, lr: 3.46e-03, grad_scale: 32.0 2023-03-26 19:07:13,496 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=86634.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 19:07:17,803 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86641.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:07:21,646 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.125e+02 1.513e+02 1.866e+02 2.326e+02 3.823e+02, threshold=3.732e+02, percent-clipped=0.0 2023-03-26 19:07:29,574 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86658.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:07:31,328 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=86661.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:07:34,797 INFO [finetune.py:976] (3/7) Epoch 16, batch 750, loss[loss=0.1954, simple_loss=0.2624, pruned_loss=0.06415, over 4916.00 frames. ], tot_loss[loss=0.1866, simple_loss=0.2557, pruned_loss=0.05868, over 933446.48 frames. ], batch size: 38, lr: 3.46e-03, grad_scale: 32.0 2023-03-26 19:08:02,112 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=86706.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:08:03,385 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86708.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:08:08,617 INFO [finetune.py:976] (3/7) Epoch 16, batch 800, loss[loss=0.1883, simple_loss=0.2679, pruned_loss=0.05437, over 4810.00 frames. ], tot_loss[loss=0.1879, simple_loss=0.2567, pruned_loss=0.05955, over 936376.44 frames. ], batch size: 39, lr: 3.46e-03, grad_scale: 32.0 2023-03-26 19:08:14,183 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86725.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:08:28,782 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.094e+02 1.501e+02 1.840e+02 2.208e+02 4.378e+02, threshold=3.681e+02, percent-clipped=4.0 2023-03-26 19:08:40,241 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=86756.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:08:49,459 INFO [finetune.py:976] (3/7) Epoch 16, batch 850, loss[loss=0.1904, simple_loss=0.2641, pruned_loss=0.05836, over 4186.00 frames. ], tot_loss[loss=0.1856, simple_loss=0.2539, pruned_loss=0.05865, over 939930.25 frames. ], batch size: 65, lr: 3.46e-03, grad_scale: 32.0 2023-03-26 19:08:54,219 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=86773.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:08:56,763 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.75 vs. limit=5.0 2023-03-26 19:09:09,592 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86796.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:09:13,591 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86801.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 19:09:21,906 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86814.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:09:23,014 INFO [finetune.py:976] (3/7) Epoch 16, batch 900, loss[loss=0.1781, simple_loss=0.2413, pruned_loss=0.05749, over 4815.00 frames. ], tot_loss[loss=0.1822, simple_loss=0.2504, pruned_loss=0.05704, over 942255.91 frames. ], batch size: 41, lr: 3.46e-03, grad_scale: 32.0 2023-03-26 19:09:40,869 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6669, 1.5934, 2.1566, 3.3349, 2.2871, 2.2819, 1.0010, 2.7499], device='cuda:3'), covar=tensor([0.1666, 0.1376, 0.1249, 0.0664, 0.0778, 0.1552, 0.1859, 0.0518], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0117, 0.0134, 0.0164, 0.0101, 0.0139, 0.0125, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 19:09:40,896 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1254, 2.0373, 2.0981, 1.5755, 2.2021, 2.2473, 2.2489, 1.8219], device='cuda:3'), covar=tensor([0.0639, 0.0588, 0.0700, 0.0854, 0.0631, 0.0656, 0.0555, 0.1062], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0135, 0.0141, 0.0123, 0.0123, 0.0140, 0.0141, 0.0163], 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') 2023-03-26 19:09:43,091 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.048e+02 1.502e+02 1.904e+02 2.205e+02 3.944e+02, threshold=3.808e+02, percent-clipped=2.0 2023-03-26 19:09:56,619 INFO [finetune.py:976] (3/7) Epoch 16, batch 950, loss[loss=0.2225, simple_loss=0.2886, pruned_loss=0.07819, over 4855.00 frames. ], tot_loss[loss=0.1799, simple_loss=0.248, pruned_loss=0.05592, over 947066.81 frames. ], batch size: 44, lr: 3.46e-03, grad_scale: 32.0 2023-03-26 19:10:02,706 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86875.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:10:04,476 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86878.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:10:20,384 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86902.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:10:40,046 INFO [finetune.py:976] (3/7) Epoch 16, batch 1000, loss[loss=0.2489, simple_loss=0.3123, pruned_loss=0.09273, over 4763.00 frames. ], tot_loss[loss=0.1824, simple_loss=0.2507, pruned_loss=0.05705, over 950258.42 frames. ], batch size: 54, lr: 3.46e-03, grad_scale: 32.0 2023-03-26 19:11:01,989 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86936.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:11:08,296 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86939.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:11:12,979 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.150e+02 1.633e+02 1.920e+02 2.320e+02 4.350e+02, threshold=3.840e+02, percent-clipped=2.0 2023-03-26 19:11:13,131 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2758, 1.1617, 1.1143, 1.1957, 1.5220, 1.4028, 1.2539, 1.1210], device='cuda:3'), covar=tensor([0.0377, 0.0330, 0.0616, 0.0319, 0.0245, 0.0486, 0.0321, 0.0459], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0108, 0.0143, 0.0113, 0.0099, 0.0106, 0.0097, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.2572e-05, 8.3815e-05, 1.1299e-04, 8.7070e-05, 7.7397e-05, 7.8312e-05, 7.3141e-05, 8.2343e-05], device='cuda:3') 2023-03-26 19:11:20,313 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=86950.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:11:34,824 INFO [finetune.py:976] (3/7) Epoch 16, batch 1050, loss[loss=0.2143, simple_loss=0.2816, pruned_loss=0.07353, over 4819.00 frames. ], tot_loss[loss=0.1833, simple_loss=0.2531, pruned_loss=0.05674, over 953315.62 frames. ], batch size: 33, lr: 3.46e-03, grad_scale: 32.0 2023-03-26 19:11:58,517 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87002.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 19:12:05,671 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 2023-03-26 19:12:08,331 INFO [finetune.py:976] (3/7) Epoch 16, batch 1100, loss[loss=0.1854, simple_loss=0.2513, pruned_loss=0.0598, over 4920.00 frames. ], tot_loss[loss=0.1848, simple_loss=0.2544, pruned_loss=0.05755, over 953951.59 frames. ], batch size: 33, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:12:27,412 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.131e+02 1.613e+02 1.835e+02 2.273e+02 4.124e+02, threshold=3.670e+02, percent-clipped=1.0 2023-03-26 19:12:31,624 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7780, 1.7709, 1.5133, 1.9160, 2.3438, 1.9568, 1.5799, 1.4350], device='cuda:3'), covar=tensor([0.2243, 0.2019, 0.1914, 0.1554, 0.1656, 0.1176, 0.2375, 0.1991], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0208, 0.0211, 0.0191, 0.0242, 0.0185, 0.0214, 0.0199], 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') 2023-03-26 19:12:39,656 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87063.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 19:12:41,762 INFO [finetune.py:976] (3/7) Epoch 16, batch 1150, loss[loss=0.2032, simple_loss=0.2729, pruned_loss=0.06674, over 4818.00 frames. ], tot_loss[loss=0.1864, simple_loss=0.256, pruned_loss=0.05839, over 954699.67 frames. ], batch size: 38, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:13:01,429 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87096.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:13:04,424 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87101.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 19:13:15,213 INFO [finetune.py:976] (3/7) Epoch 16, batch 1200, loss[loss=0.2382, simple_loss=0.2939, pruned_loss=0.09128, over 4788.00 frames. ], tot_loss[loss=0.185, simple_loss=0.2546, pruned_loss=0.05769, over 953554.13 frames. ], batch size: 51, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:13:16,024 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-03-26 19:13:33,240 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=87144.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:13:34,371 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.093e+01 1.544e+02 1.890e+02 2.251e+02 4.242e+02, threshold=3.781e+02, percent-clipped=1.0 2023-03-26 19:13:36,663 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=87149.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:13:50,131 INFO [finetune.py:976] (3/7) Epoch 16, batch 1250, loss[loss=0.1271, simple_loss=0.209, pruned_loss=0.02265, over 4788.00 frames. ], tot_loss[loss=0.1822, simple_loss=0.2515, pruned_loss=0.05647, over 954434.77 frames. ], batch size: 51, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:13:53,123 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87170.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:13:54,365 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3712, 1.4785, 1.5214, 1.5850, 1.6371, 2.9622, 1.5309, 1.5801], device='cuda:3'), covar=tensor([0.1039, 0.1753, 0.1117, 0.0954, 0.1561, 0.0300, 0.1321, 0.1722], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0082, 0.0074, 0.0078, 0.0092, 0.0080, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 19:14:31,061 INFO [finetune.py:976] (3/7) Epoch 16, batch 1300, loss[loss=0.1782, simple_loss=0.2436, pruned_loss=0.05641, over 4818.00 frames. ], tot_loss[loss=0.1805, simple_loss=0.249, pruned_loss=0.05603, over 955918.43 frames. ], batch size: 41, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:14:41,211 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87230.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:14:43,587 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87234.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:14:43,640 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9934, 1.7374, 2.1522, 1.4082, 2.0388, 2.1831, 1.6708, 2.4179], device='cuda:3'), covar=tensor([0.1215, 0.1966, 0.1401, 0.1869, 0.0855, 0.1441, 0.2852, 0.0765], device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0207, 0.0194, 0.0192, 0.0180, 0.0217, 0.0220, 0.0203], 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') 2023-03-26 19:14:44,804 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87236.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:14:51,269 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.006e+02 1.578e+02 1.944e+02 2.250e+02 4.130e+02, threshold=3.887e+02, percent-clipped=1.0 2023-03-26 19:14:59,188 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2295, 2.0700, 2.3137, 1.7250, 2.1186, 2.3896, 2.4042, 1.7844], device='cuda:3'), covar=tensor([0.0551, 0.0587, 0.0534, 0.0787, 0.0678, 0.0594, 0.0490, 0.1052], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0134, 0.0141, 0.0123, 0.0123, 0.0140, 0.0142, 0.0163], 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') 2023-03-26 19:15:02,899 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-03-26 19:15:04,397 INFO [finetune.py:976] (3/7) Epoch 16, batch 1350, loss[loss=0.1904, simple_loss=0.2668, pruned_loss=0.057, over 4859.00 frames. ], tot_loss[loss=0.18, simple_loss=0.2482, pruned_loss=0.05587, over 957007.13 frames. ], batch size: 44, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:15:16,951 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=87284.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:15:21,344 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87291.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:15:28,552 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.41 vs. limit=5.0 2023-03-26 19:15:44,164 INFO [finetune.py:976] (3/7) Epoch 16, batch 1400, loss[loss=0.1669, simple_loss=0.2596, pruned_loss=0.03712, over 4913.00 frames. ], tot_loss[loss=0.1829, simple_loss=0.2519, pruned_loss=0.05694, over 955731.86 frames. ], batch size: 42, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:16:02,446 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9049, 1.7616, 1.3932, 1.3364, 2.3262, 2.2792, 1.9627, 1.8319], device='cuda:3'), covar=tensor([0.0424, 0.0457, 0.0852, 0.0507, 0.0253, 0.0585, 0.0344, 0.0453], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0107, 0.0141, 0.0111, 0.0098, 0.0105, 0.0096, 0.0107], device='cuda:3'), out_proj_covar=tensor([7.2006e-05, 8.2867e-05, 1.1138e-04, 8.6088e-05, 7.6562e-05, 7.7367e-05, 7.2519e-05, 8.1722e-05], device='cuda:3') 2023-03-26 19:16:19,255 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.049e+02 1.619e+02 1.873e+02 2.376e+02 3.982e+02, threshold=3.745e+02, percent-clipped=1.0 2023-03-26 19:16:31,541 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87358.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 19:16:38,512 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87362.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:16:41,377 INFO [finetune.py:976] (3/7) Epoch 16, batch 1450, loss[loss=0.2345, simple_loss=0.3016, pruned_loss=0.08371, over 4811.00 frames. ], tot_loss[loss=0.1851, simple_loss=0.2543, pruned_loss=0.05789, over 953700.65 frames. ], batch size: 38, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:16:49,635 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5039, 1.5272, 1.6066, 1.6748, 1.6831, 2.9667, 1.4737, 1.5812], device='cuda:3'), covar=tensor([0.0970, 0.1626, 0.1130, 0.0912, 0.1460, 0.0320, 0.1293, 0.1632], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0082, 0.0075, 0.0078, 0.0093, 0.0081, 0.0086, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 19:17:18,471 INFO [finetune.py:976] (3/7) Epoch 16, batch 1500, loss[loss=0.2144, simple_loss=0.2832, pruned_loss=0.07276, over 4810.00 frames. ], tot_loss[loss=0.1867, simple_loss=0.2562, pruned_loss=0.05862, over 954557.41 frames. ], batch size: 40, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:17:23,382 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87423.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:17:39,129 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.097e+02 1.635e+02 2.033e+02 2.421e+02 4.092e+02, threshold=4.066e+02, percent-clipped=1.0 2023-03-26 19:17:42,913 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.3103, 2.9507, 2.6124, 1.4990, 2.7834, 2.2482, 2.1883, 2.5649], device='cuda:3'), covar=tensor([0.1121, 0.0802, 0.1851, 0.2114, 0.1795, 0.2474, 0.2362, 0.1153], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0195, 0.0200, 0.0183, 0.0212, 0.0208, 0.0224, 0.0196], 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') 2023-03-26 19:17:48,782 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87461.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:17:52,196 INFO [finetune.py:976] (3/7) Epoch 16, batch 1550, loss[loss=0.1882, simple_loss=0.2618, pruned_loss=0.05735, over 4915.00 frames. ], tot_loss[loss=0.1863, simple_loss=0.2561, pruned_loss=0.05823, over 954250.26 frames. ], batch size: 38, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:17:55,186 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87470.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:18:09,828 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0844, 2.0724, 2.2497, 1.6040, 2.2293, 2.3479, 2.3594, 1.7909], device='cuda:3'), covar=tensor([0.0602, 0.0607, 0.0647, 0.0833, 0.0622, 0.0665, 0.0529, 0.1088], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0134, 0.0140, 0.0122, 0.0123, 0.0139, 0.0141, 0.0163], 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') 2023-03-26 19:18:18,810 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87505.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:18:25,477 INFO [finetune.py:976] (3/7) Epoch 16, batch 1600, loss[loss=0.1986, simple_loss=0.2701, pruned_loss=0.06354, over 4903.00 frames. ], tot_loss[loss=0.1843, simple_loss=0.2537, pruned_loss=0.0575, over 953874.90 frames. ], batch size: 36, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:18:27,232 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=87518.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:18:30,238 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87522.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:18:38,585 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87534.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:18:43,111 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9477, 1.8712, 2.0490, 1.3629, 2.0356, 2.1435, 2.0842, 1.5867], device='cuda:3'), covar=tensor([0.0679, 0.0718, 0.0669, 0.0938, 0.0648, 0.0661, 0.0555, 0.1281], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0133, 0.0140, 0.0122, 0.0123, 0.0139, 0.0141, 0.0162], 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') 2023-03-26 19:18:46,617 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.707e+01 1.405e+02 1.628e+02 1.991e+02 3.372e+02, threshold=3.256e+02, percent-clipped=0.0 2023-03-26 19:18:48,371 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-03-26 19:18:59,343 INFO [finetune.py:976] (3/7) Epoch 16, batch 1650, loss[loss=0.1262, simple_loss=0.2008, pruned_loss=0.02584, over 4791.00 frames. ], tot_loss[loss=0.1823, simple_loss=0.2509, pruned_loss=0.05684, over 954208.49 frames. ], batch size: 29, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:18:59,482 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87566.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 19:19:10,144 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=87582.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:19:11,441 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87584.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:19:13,583 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87586.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:19:17,414 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87589.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:19:36,456 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87605.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:19:43,483 INFO [finetune.py:976] (3/7) Epoch 16, batch 1700, loss[loss=0.1982, simple_loss=0.276, pruned_loss=0.06021, over 4861.00 frames. ], tot_loss[loss=0.1813, simple_loss=0.2495, pruned_loss=0.05656, over 954383.83 frames. ], batch size: 44, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:19:51,266 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5944, 1.5338, 1.9805, 1.9933, 1.7250, 3.6398, 1.4629, 1.6509], device='cuda:3'), covar=tensor([0.0999, 0.1809, 0.1185, 0.0943, 0.1645, 0.0229, 0.1548, 0.1756], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0074, 0.0077, 0.0091, 0.0080, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 19:20:03,773 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87645.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 19:20:04,234 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.164e+02 1.606e+02 1.879e+02 2.372e+02 9.403e+02, threshold=3.758e+02, percent-clipped=6.0 2023-03-26 19:20:06,834 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87650.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:20:11,626 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87658.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 19:20:16,843 INFO [finetune.py:976] (3/7) Epoch 16, batch 1750, loss[loss=0.1784, simple_loss=0.2658, pruned_loss=0.04545, over 4906.00 frames. ], tot_loss[loss=0.1825, simple_loss=0.251, pruned_loss=0.05698, over 955234.02 frames. ], batch size: 37, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:20:16,976 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87666.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 19:20:19,989 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9960, 1.5161, 0.7702, 1.8613, 2.2948, 1.8606, 1.6138, 1.8384], device='cuda:3'), covar=tensor([0.1380, 0.1862, 0.2089, 0.1157, 0.1885, 0.1928, 0.1295, 0.1782], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0094, 0.0109, 0.0092, 0.0118, 0.0093, 0.0098, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 19:20:44,152 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=87706.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 19:20:50,182 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7436, 1.7415, 1.9061, 1.2045, 1.9304, 1.9401, 1.8991, 1.5686], device='cuda:3'), covar=tensor([0.0542, 0.0580, 0.0554, 0.0760, 0.0567, 0.0572, 0.0511, 0.0986], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0134, 0.0140, 0.0122, 0.0123, 0.0139, 0.0141, 0.0163], 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') 2023-03-26 19:20:50,640 INFO [finetune.py:976] (3/7) Epoch 16, batch 1800, loss[loss=0.2126, simple_loss=0.2705, pruned_loss=0.07733, over 4814.00 frames. ], tot_loss[loss=0.184, simple_loss=0.2536, pruned_loss=0.05725, over 956482.84 frames. ], batch size: 51, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:20:51,922 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87718.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:21:13,299 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.163e+02 1.635e+02 1.907e+02 2.399e+02 5.758e+02, threshold=3.813e+02, percent-clipped=1.0 2023-03-26 19:21:28,369 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7066, 1.7287, 1.5039, 1.5926, 1.3299, 4.1560, 1.6741, 2.0250], device='cuda:3'), covar=tensor([0.3363, 0.2450, 0.2155, 0.2374, 0.1581, 0.0135, 0.2534, 0.1255], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0121, 0.0124, 0.0114, 0.0097, 0.0097, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 19:21:36,193 INFO [finetune.py:976] (3/7) Epoch 16, batch 1850, loss[loss=0.1668, simple_loss=0.2337, pruned_loss=0.04999, over 4912.00 frames. ], tot_loss[loss=0.1856, simple_loss=0.2552, pruned_loss=0.058, over 956552.85 frames. ], batch size: 36, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:22:22,284 INFO [finetune.py:976] (3/7) Epoch 16, batch 1900, loss[loss=0.1782, simple_loss=0.2482, pruned_loss=0.05413, over 4768.00 frames. ], tot_loss[loss=0.1855, simple_loss=0.2562, pruned_loss=0.05743, over 958787.82 frames. ], batch size: 26, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:22:23,006 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87817.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:22:34,296 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-03-26 19:22:41,851 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.250e+02 1.530e+02 1.792e+02 2.215e+02 4.706e+02, threshold=3.584e+02, percent-clipped=3.0 2023-03-26 19:22:45,716 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.59 vs. limit=5.0 2023-03-26 19:22:53,011 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87861.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 19:22:55,621 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-03-26 19:22:55,973 INFO [finetune.py:976] (3/7) Epoch 16, batch 1950, loss[loss=0.1816, simple_loss=0.2551, pruned_loss=0.054, over 4742.00 frames. ], tot_loss[loss=0.1845, simple_loss=0.2548, pruned_loss=0.05709, over 957324.48 frames. ], batch size: 54, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:22:57,320 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0716, 1.9842, 1.8494, 2.0837, 2.6052, 2.0782, 2.1751, 1.7066], device='cuda:3'), covar=tensor([0.1820, 0.1750, 0.1615, 0.1462, 0.1644, 0.1097, 0.1813, 0.1600], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0208, 0.0211, 0.0192, 0.0243, 0.0185, 0.0216, 0.0199], 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') 2023-03-26 19:23:09,182 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87886.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:23:14,527 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7053, 1.6554, 1.5770, 1.6061, 1.2377, 3.3808, 1.4469, 1.9226], device='cuda:3'), covar=tensor([0.3094, 0.2255, 0.1946, 0.2209, 0.1635, 0.0218, 0.2819, 0.1153], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0115, 0.0121, 0.0125, 0.0114, 0.0097, 0.0096, 0.0097], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 19:23:20,506 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-03-26 19:23:29,595 INFO [finetune.py:976] (3/7) Epoch 16, batch 2000, loss[loss=0.1531, simple_loss=0.2254, pruned_loss=0.04044, over 4816.00 frames. ], tot_loss[loss=0.1834, simple_loss=0.2525, pruned_loss=0.05719, over 957236.24 frames. ], batch size: 38, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:23:41,096 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=87934.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:23:45,289 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87940.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 19:23:48,824 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87945.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:23:49,302 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.076e+02 1.562e+02 1.812e+02 2.199e+02 5.123e+02, threshold=3.624e+02, percent-clipped=1.0 2023-03-26 19:23:59,898 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87961.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 19:24:01,167 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9080, 1.5136, 2.0313, 1.8578, 1.6755, 1.6179, 1.8441, 1.8450], device='cuda:3'), covar=tensor([0.3986, 0.3973, 0.3266, 0.4058, 0.5013, 0.4016, 0.4658, 0.3188], device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0238, 0.0258, 0.0270, 0.0269, 0.0242, 0.0281, 0.0238], 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') 2023-03-26 19:24:02,864 INFO [finetune.py:976] (3/7) Epoch 16, batch 2050, loss[loss=0.1944, simple_loss=0.2591, pruned_loss=0.06483, over 4753.00 frames. ], tot_loss[loss=0.1807, simple_loss=0.2487, pruned_loss=0.05635, over 956732.31 frames. ], batch size: 26, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:24:32,347 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.67 vs. limit=5.0 2023-03-26 19:24:37,649 INFO [finetune.py:976] (3/7) Epoch 16, batch 2100, loss[loss=0.1917, simple_loss=0.2576, pruned_loss=0.06289, over 4919.00 frames. ], tot_loss[loss=0.1803, simple_loss=0.2481, pruned_loss=0.0562, over 958250.01 frames. ], batch size: 37, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:24:41,371 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88018.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:24:49,724 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-26 19:24:59,877 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.339e+01 1.671e+02 1.963e+02 2.403e+02 4.597e+02, threshold=3.926e+02, percent-clipped=2.0 2023-03-26 19:25:13,434 INFO [finetune.py:976] (3/7) Epoch 16, batch 2150, loss[loss=0.2253, simple_loss=0.2885, pruned_loss=0.08106, over 4275.00 frames. ], tot_loss[loss=0.1835, simple_loss=0.2517, pruned_loss=0.05768, over 955549.86 frames. ], batch size: 66, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:25:13,503 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=88066.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:25:35,470 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88100.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:25:46,508 INFO [finetune.py:976] (3/7) Epoch 16, batch 2200, loss[loss=0.1764, simple_loss=0.2585, pruned_loss=0.04721, over 4814.00 frames. ], tot_loss[loss=0.1852, simple_loss=0.2539, pruned_loss=0.05828, over 955164.01 frames. ], batch size: 40, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:25:47,205 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88117.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:25:52,081 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.3243, 3.7703, 3.9378, 4.1604, 4.1002, 3.8100, 4.4301, 1.3320], device='cuda:3'), covar=tensor([0.0779, 0.0865, 0.0993, 0.1021, 0.1248, 0.1734, 0.0683, 0.5772], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0245, 0.0276, 0.0291, 0.0335, 0.0282, 0.0298, 0.0296], 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') 2023-03-26 19:26:05,785 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.079e+02 1.658e+02 1.940e+02 2.471e+02 6.986e+02, threshold=3.880e+02, percent-clipped=3.0 2023-03-26 19:26:15,413 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88161.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 19:26:15,432 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88161.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:26:18,733 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=88165.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:26:19,283 INFO [finetune.py:976] (3/7) Epoch 16, batch 2250, loss[loss=0.1844, simple_loss=0.2598, pruned_loss=0.05452, over 4903.00 frames. ], tot_loss[loss=0.1868, simple_loss=0.2557, pruned_loss=0.05893, over 952288.67 frames. ], batch size: 46, lr: 3.45e-03, grad_scale: 32.0 2023-03-26 19:26:31,475 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2004, 1.7745, 2.1813, 2.1383, 1.8228, 1.8423, 2.1023, 1.9771], device='cuda:3'), covar=tensor([0.3883, 0.4029, 0.3171, 0.3795, 0.5204, 0.3871, 0.4699, 0.3183], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0236, 0.0256, 0.0268, 0.0268, 0.0241, 0.0279, 0.0236], 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') 2023-03-26 19:26:56,477 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=88209.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:27:05,825 INFO [finetune.py:976] (3/7) Epoch 16, batch 2300, loss[loss=0.2137, simple_loss=0.2774, pruned_loss=0.07496, over 4886.00 frames. ], tot_loss[loss=0.1855, simple_loss=0.2549, pruned_loss=0.05801, over 951983.12 frames. ], batch size: 35, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:27:22,430 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.2561, 3.7067, 3.8768, 4.0569, 4.0354, 3.7653, 4.3304, 1.3293], device='cuda:3'), covar=tensor([0.0740, 0.0854, 0.0892, 0.1035, 0.1038, 0.1473, 0.0674, 0.5553], device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0244, 0.0276, 0.0290, 0.0335, 0.0282, 0.0297, 0.0294], 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') 2023-03-26 19:27:29,755 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88240.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:27:33,305 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88245.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:27:34,388 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.097e+02 1.462e+02 1.840e+02 2.103e+02 3.666e+02, threshold=3.679e+02, percent-clipped=0.0 2023-03-26 19:27:43,884 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88261.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 19:27:47,285 INFO [finetune.py:976] (3/7) Epoch 16, batch 2350, loss[loss=0.1494, simple_loss=0.2183, pruned_loss=0.0403, over 4765.00 frames. ], tot_loss[loss=0.1842, simple_loss=0.2527, pruned_loss=0.05785, over 953388.73 frames. ], batch size: 28, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:28:01,665 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=88288.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:28:03,544 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88291.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 19:28:04,699 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=88293.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:28:15,422 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=88309.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:28:20,097 INFO [finetune.py:976] (3/7) Epoch 16, batch 2400, loss[loss=0.1678, simple_loss=0.2334, pruned_loss=0.05108, over 4938.00 frames. ], tot_loss[loss=0.1829, simple_loss=0.2509, pruned_loss=0.05745, over 954728.96 frames. ], batch size: 33, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:28:40,400 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.006e+02 1.525e+02 1.766e+02 2.106e+02 3.774e+02, threshold=3.532e+02, percent-clipped=1.0 2023-03-26 19:28:44,076 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88352.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 19:28:52,865 INFO [finetune.py:976] (3/7) Epoch 16, batch 2450, loss[loss=0.1601, simple_loss=0.2274, pruned_loss=0.04645, over 4886.00 frames. ], tot_loss[loss=0.1807, simple_loss=0.2484, pruned_loss=0.0565, over 954700.84 frames. ], batch size: 32, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:29:26,835 INFO [finetune.py:976] (3/7) Epoch 16, batch 2500, loss[loss=0.1634, simple_loss=0.2376, pruned_loss=0.04456, over 4830.00 frames. ], tot_loss[loss=0.1823, simple_loss=0.2501, pruned_loss=0.05728, over 954833.83 frames. ], batch size: 33, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:29:48,234 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.125e+02 1.743e+02 2.000e+02 2.601e+02 5.270e+02, threshold=4.000e+02, percent-clipped=5.0 2023-03-26 19:29:54,227 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88456.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:30:00,709 INFO [finetune.py:976] (3/7) Epoch 16, batch 2550, loss[loss=0.1725, simple_loss=0.2476, pruned_loss=0.04875, over 4864.00 frames. ], tot_loss[loss=0.1833, simple_loss=0.2521, pruned_loss=0.05722, over 955560.72 frames. ], batch size: 31, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:30:33,907 INFO [finetune.py:976] (3/7) Epoch 16, batch 2600, loss[loss=0.175, simple_loss=0.2367, pruned_loss=0.05669, over 4873.00 frames. ], tot_loss[loss=0.1847, simple_loss=0.254, pruned_loss=0.05768, over 955202.26 frames. ], batch size: 34, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:30:55,464 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.436e+01 1.666e+02 1.940e+02 2.279e+02 3.712e+02, threshold=3.880e+02, percent-clipped=0.0 2023-03-26 19:30:59,284 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4642, 2.4261, 2.0334, 2.6272, 2.4886, 2.0720, 2.9863, 2.5076], device='cuda:3'), covar=tensor([0.1291, 0.2363, 0.3002, 0.2597, 0.2565, 0.1732, 0.3019, 0.1790], device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0187, 0.0234, 0.0252, 0.0244, 0.0201, 0.0212, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 19:31:07,486 INFO [finetune.py:976] (3/7) Epoch 16, batch 2650, loss[loss=0.1299, simple_loss=0.1867, pruned_loss=0.0366, over 3403.00 frames. ], tot_loss[loss=0.1854, simple_loss=0.2549, pruned_loss=0.05798, over 952724.44 frames. ], batch size: 14, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:31:12,341 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.2891, 2.9290, 3.0305, 3.2483, 3.0883, 2.8398, 3.3088, 0.9093], device='cuda:3'), covar=tensor([0.0960, 0.1002, 0.1069, 0.1046, 0.1523, 0.1821, 0.1124, 0.5272], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0245, 0.0277, 0.0292, 0.0335, 0.0283, 0.0298, 0.0295], 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') 2023-03-26 19:31:41,339 INFO [finetune.py:976] (3/7) Epoch 16, batch 2700, loss[loss=0.2053, simple_loss=0.2583, pruned_loss=0.07613, over 4241.00 frames. ], tot_loss[loss=0.1848, simple_loss=0.2545, pruned_loss=0.05757, over 953983.80 frames. ], batch size: 65, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:32:05,538 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.033e+02 1.504e+02 1.815e+02 2.296e+02 4.078e+02, threshold=3.631e+02, percent-clipped=1.0 2023-03-26 19:32:05,623 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88647.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 19:32:26,940 INFO [finetune.py:976] (3/7) Epoch 16, batch 2750, loss[loss=0.1583, simple_loss=0.2262, pruned_loss=0.0452, over 4905.00 frames. ], tot_loss[loss=0.1827, simple_loss=0.2519, pruned_loss=0.05677, over 955301.47 frames. ], batch size: 43, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:32:42,726 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-03-26 19:32:51,377 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-26 19:33:17,051 INFO [finetune.py:976] (3/7) Epoch 16, batch 2800, loss[loss=0.1566, simple_loss=0.2341, pruned_loss=0.03956, over 4911.00 frames. ], tot_loss[loss=0.1798, simple_loss=0.2485, pruned_loss=0.05558, over 954796.63 frames. ], batch size: 36, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:33:37,857 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.207e+02 1.516e+02 1.815e+02 2.286e+02 3.246e+02, threshold=3.631e+02, percent-clipped=0.0 2023-03-26 19:33:43,829 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-26 19:33:44,962 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88756.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:33:50,912 INFO [finetune.py:976] (3/7) Epoch 16, batch 2850, loss[loss=0.1981, simple_loss=0.2353, pruned_loss=0.08048, over 3962.00 frames. ], tot_loss[loss=0.1788, simple_loss=0.247, pruned_loss=0.05525, over 952920.62 frames. ], batch size: 17, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:33:58,306 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.17 vs. limit=5.0 2023-03-26 19:34:17,619 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=88804.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:34:24,821 INFO [finetune.py:976] (3/7) Epoch 16, batch 2900, loss[loss=0.2101, simple_loss=0.2891, pruned_loss=0.06554, over 4812.00 frames. ], tot_loss[loss=0.1811, simple_loss=0.2497, pruned_loss=0.0562, over 950905.41 frames. ], batch size: 40, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:34:45,213 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.025e+02 1.632e+02 1.968e+02 2.500e+02 4.348e+02, threshold=3.936e+02, percent-clipped=6.0 2023-03-26 19:34:45,351 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1231, 1.0143, 0.9185, 1.1655, 1.2346, 1.1653, 1.0249, 0.9487], device='cuda:3'), covar=tensor([0.0333, 0.0318, 0.0594, 0.0289, 0.0263, 0.0420, 0.0348, 0.0403], device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0107, 0.0142, 0.0111, 0.0098, 0.0106, 0.0097, 0.0107], device='cuda:3'), out_proj_covar=tensor([7.2315e-05, 8.3011e-05, 1.1238e-04, 8.5885e-05, 7.6785e-05, 7.8081e-05, 7.2487e-05, 8.1845e-05], device='cuda:3') 2023-03-26 19:34:58,808 INFO [finetune.py:976] (3/7) Epoch 16, batch 2950, loss[loss=0.2013, simple_loss=0.2825, pruned_loss=0.0601, over 4817.00 frames. ], tot_loss[loss=0.1828, simple_loss=0.252, pruned_loss=0.05679, over 950761.95 frames. ], batch size: 39, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:35:04,886 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6264, 1.5919, 1.8336, 1.8269, 1.6605, 2.9703, 1.4565, 1.6198], device='cuda:3'), covar=tensor([0.0870, 0.1447, 0.1200, 0.0823, 0.1303, 0.0305, 0.1272, 0.1434], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0073, 0.0077, 0.0091, 0.0080, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 19:35:32,643 INFO [finetune.py:976] (3/7) Epoch 16, batch 3000, loss[loss=0.171, simple_loss=0.2327, pruned_loss=0.05465, over 4760.00 frames. ], tot_loss[loss=0.1843, simple_loss=0.2535, pruned_loss=0.05756, over 951242.15 frames. ], batch size: 23, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:35:32,643 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 19:35:38,748 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3973, 1.3281, 1.2864, 1.4500, 1.7050, 1.5001, 1.3710, 1.2288], device='cuda:3'), covar=tensor([0.0371, 0.0310, 0.0580, 0.0289, 0.0204, 0.0511, 0.0362, 0.0431], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0108, 0.0143, 0.0112, 0.0099, 0.0107, 0.0097, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.2988e-05, 8.3872e-05, 1.1301e-04, 8.6599e-05, 7.7392e-05, 7.8828e-05, 7.3023e-05, 8.2483e-05], device='cuda:3') 2023-03-26 19:35:41,778 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0956, 1.9751, 1.7812, 1.8985, 2.1115, 1.8575, 2.3249, 2.0928], device='cuda:3'), covar=tensor([0.1516, 0.2318, 0.3087, 0.2475, 0.2622, 0.1769, 0.3213, 0.1946], device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0186, 0.0233, 0.0251, 0.0242, 0.0200, 0.0210, 0.0198], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 19:35:49,227 INFO [finetune.py:1010] (3/7) Epoch 16, validation: loss=0.1563, simple_loss=0.2263, pruned_loss=0.04316, over 2265189.00 frames. 2023-03-26 19:35:49,228 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6434MB 2023-03-26 19:35:58,849 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4238, 2.1903, 2.0216, 2.2492, 2.1019, 2.1585, 2.1453, 2.9048], device='cuda:3'), covar=tensor([0.3576, 0.4430, 0.3205, 0.3572, 0.3805, 0.2338, 0.3645, 0.1464], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0262, 0.0227, 0.0276, 0.0249, 0.0217, 0.0251, 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') 2023-03-26 19:36:10,237 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0397, 1.8929, 1.6217, 1.9050, 1.8279, 1.8080, 1.7990, 2.5396], device='cuda:3'), covar=tensor([0.4008, 0.4480, 0.3603, 0.4397, 0.4331, 0.2652, 0.4482, 0.1767], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0262, 0.0227, 0.0276, 0.0249, 0.0217, 0.0251, 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') 2023-03-26 19:36:10,682 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.136e+02 1.661e+02 1.990e+02 2.439e+02 3.546e+02, threshold=3.980e+02, percent-clipped=0.0 2023-03-26 19:36:11,251 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88947.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 19:36:23,198 INFO [finetune.py:976] (3/7) Epoch 16, batch 3050, loss[loss=0.1746, simple_loss=0.254, pruned_loss=0.04758, over 4924.00 frames. ], tot_loss[loss=0.1855, simple_loss=0.2545, pruned_loss=0.05827, over 950612.92 frames. ], batch size: 42, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:36:43,520 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=88995.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 19:36:44,768 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3243, 1.3759, 1.1288, 1.2878, 1.6628, 1.5126, 1.4015, 1.2633], device='cuda:3'), covar=tensor([0.0386, 0.0372, 0.0692, 0.0360, 0.0231, 0.0629, 0.0392, 0.0422], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0108, 0.0143, 0.0112, 0.0099, 0.0107, 0.0097, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.3022e-05, 8.3670e-05, 1.1323e-04, 8.6609e-05, 7.7337e-05, 7.8762e-05, 7.2858e-05, 8.2250e-05], device='cuda:3') 2023-03-26 19:36:57,478 INFO [finetune.py:976] (3/7) Epoch 16, batch 3100, loss[loss=0.1358, simple_loss=0.2141, pruned_loss=0.02873, over 4796.00 frames. ], tot_loss[loss=0.1846, simple_loss=0.2536, pruned_loss=0.05779, over 951030.34 frames. ], batch size: 29, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:37:06,396 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89027.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:37:07,319 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-26 19:37:20,955 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.966e+01 1.506e+02 1.838e+02 2.198e+02 3.411e+02, threshold=3.676e+02, percent-clipped=0.0 2023-03-26 19:37:33,689 INFO [finetune.py:976] (3/7) Epoch 16, batch 3150, loss[loss=0.2011, simple_loss=0.2569, pruned_loss=0.07271, over 4869.00 frames. ], tot_loss[loss=0.182, simple_loss=0.2507, pruned_loss=0.05667, over 953290.77 frames. ], batch size: 31, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:37:56,585 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89088.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:38:25,684 INFO [finetune.py:976] (3/7) Epoch 16, batch 3200, loss[loss=0.1884, simple_loss=0.259, pruned_loss=0.05889, over 4876.00 frames. ], tot_loss[loss=0.1793, simple_loss=0.2473, pruned_loss=0.05562, over 952596.12 frames. ], batch size: 34, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:38:34,576 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-26 19:38:50,100 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.129e+02 1.609e+02 1.908e+02 2.339e+02 4.086e+02, threshold=3.816e+02, percent-clipped=1.0 2023-03-26 19:38:50,250 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4030, 2.1379, 1.7236, 0.7937, 1.8812, 1.8718, 1.7303, 2.0134], device='cuda:3'), covar=tensor([0.0938, 0.0782, 0.1632, 0.1971, 0.1514, 0.2122, 0.2291, 0.0957], device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0194, 0.0198, 0.0181, 0.0210, 0.0204, 0.0221, 0.0195], 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') 2023-03-26 19:38:50,316 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-03-26 19:38:53,276 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.4518, 1.4916, 1.6242, 0.8667, 1.6033, 1.8114, 1.8634, 1.4557], device='cuda:3'), covar=tensor([0.1006, 0.0736, 0.0588, 0.0642, 0.0484, 0.0702, 0.0320, 0.0778], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0150, 0.0123, 0.0126, 0.0130, 0.0127, 0.0142, 0.0146], device='cuda:3'), out_proj_covar=tensor([9.1430e-05, 1.0925e-04, 8.7801e-05, 8.9911e-05, 9.2157e-05, 9.1760e-05, 1.0192e-04, 1.0514e-04], device='cuda:3') 2023-03-26 19:38:59,803 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5491, 1.3577, 1.2887, 1.5180, 1.7620, 1.5638, 1.0431, 1.2854], device='cuda:3'), covar=tensor([0.2104, 0.2170, 0.1970, 0.1686, 0.1584, 0.1237, 0.2464, 0.1929], device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0209, 0.0212, 0.0193, 0.0245, 0.0186, 0.0215, 0.0201], 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') 2023-03-26 19:39:02,083 INFO [finetune.py:976] (3/7) Epoch 16, batch 3250, loss[loss=0.1834, simple_loss=0.2297, pruned_loss=0.06856, over 4311.00 frames. ], tot_loss[loss=0.18, simple_loss=0.2482, pruned_loss=0.0559, over 952301.29 frames. ], batch size: 18, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:39:04,497 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89169.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:39:21,687 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7103, 1.1577, 0.8726, 1.5973, 2.0216, 1.4360, 1.5048, 1.6333], device='cuda:3'), covar=tensor([0.1481, 0.2144, 0.1954, 0.1213, 0.1933, 0.2094, 0.1398, 0.1907], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0095, 0.0111, 0.0093, 0.0119, 0.0094, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 19:39:28,308 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0414, 1.4700, 0.8359, 1.7800, 2.2108, 1.5818, 1.7624, 1.8248], device='cuda:3'), covar=tensor([0.1342, 0.2052, 0.2112, 0.1202, 0.1868, 0.1939, 0.1317, 0.1912], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0095, 0.0111, 0.0093, 0.0119, 0.0094, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 19:39:35,942 INFO [finetune.py:976] (3/7) Epoch 16, batch 3300, loss[loss=0.1701, simple_loss=0.2496, pruned_loss=0.04534, over 4765.00 frames. ], tot_loss[loss=0.1836, simple_loss=0.2523, pruned_loss=0.05744, over 950608.62 frames. ], batch size: 28, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:39:45,106 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89230.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:39:56,763 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.235e+02 1.765e+02 2.004e+02 2.308e+02 3.942e+02, threshold=4.007e+02, percent-clipped=1.0 2023-03-26 19:40:09,187 INFO [finetune.py:976] (3/7) Epoch 16, batch 3350, loss[loss=0.1717, simple_loss=0.2408, pruned_loss=0.05136, over 4754.00 frames. ], tot_loss[loss=0.1843, simple_loss=0.2531, pruned_loss=0.05774, over 949835.52 frames. ], batch size: 27, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:40:26,864 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4456, 3.3858, 3.1848, 1.4248, 3.5644, 2.5813, 0.6704, 2.2638], device='cuda:3'), covar=tensor([0.2521, 0.1940, 0.1621, 0.3520, 0.1146, 0.1115, 0.4531, 0.1580], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0176, 0.0160, 0.0129, 0.0158, 0.0123, 0.0147, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 19:40:42,691 INFO [finetune.py:976] (3/7) Epoch 16, batch 3400, loss[loss=0.1592, simple_loss=0.2302, pruned_loss=0.04406, over 4905.00 frames. ], tot_loss[loss=0.1853, simple_loss=0.2543, pruned_loss=0.05818, over 950061.94 frames. ], batch size: 36, lr: 3.44e-03, grad_scale: 32.0 2023-03-26 19:40:56,401 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7000, 1.5994, 1.4674, 1.5650, 1.3207, 4.1713, 1.7046, 1.9048], device='cuda:3'), covar=tensor([0.3882, 0.3130, 0.2427, 0.2912, 0.1816, 0.0194, 0.2392, 0.1275], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0119, 0.0123, 0.0114, 0.0097, 0.0096, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 19:41:12,552 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.032e+02 1.581e+02 1.832e+02 2.219e+02 5.301e+02, threshold=3.664e+02, percent-clipped=1.0 2023-03-26 19:41:24,424 INFO [finetune.py:976] (3/7) Epoch 16, batch 3450, loss[loss=0.1487, simple_loss=0.2185, pruned_loss=0.03947, over 4771.00 frames. ], tot_loss[loss=0.1845, simple_loss=0.2537, pruned_loss=0.05763, over 951209.43 frames. ], batch size: 27, lr: 3.43e-03, grad_scale: 32.0 2023-03-26 19:41:29,843 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89374.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:41:35,741 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89383.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:41:48,194 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7792, 3.9205, 3.6515, 1.9486, 3.9883, 2.9145, 0.7635, 2.7652], device='cuda:3'), covar=tensor([0.2313, 0.1780, 0.1489, 0.3375, 0.0927, 0.1059, 0.4650, 0.1659], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0175, 0.0159, 0.0128, 0.0158, 0.0123, 0.0147, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 19:41:58,330 INFO [finetune.py:976] (3/7) Epoch 16, batch 3500, loss[loss=0.1532, simple_loss=0.2078, pruned_loss=0.04933, over 4926.00 frames. ], tot_loss[loss=0.1831, simple_loss=0.2517, pruned_loss=0.05727, over 954719.59 frames. ], batch size: 33, lr: 3.43e-03, grad_scale: 32.0 2023-03-26 19:42:04,411 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89425.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:42:10,962 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89435.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:42:13,372 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89439.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 19:42:18,648 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.936e+01 1.517e+02 1.946e+02 2.225e+02 4.216e+02, threshold=3.891e+02, percent-clipped=3.0 2023-03-26 19:42:31,137 INFO [finetune.py:976] (3/7) Epoch 16, batch 3550, loss[loss=0.1647, simple_loss=0.2394, pruned_loss=0.04503, over 4917.00 frames. ], tot_loss[loss=0.1818, simple_loss=0.2498, pruned_loss=0.05692, over 954444.39 frames. ], batch size: 37, lr: 3.43e-03, grad_scale: 32.0 2023-03-26 19:42:40,907 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.6843, 1.6351, 1.6562, 1.0882, 1.8127, 1.9839, 1.9646, 1.4480], device='cuda:3'), covar=tensor([0.1003, 0.0712, 0.0514, 0.0627, 0.0441, 0.0699, 0.0365, 0.0895], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0153, 0.0125, 0.0128, 0.0132, 0.0130, 0.0143, 0.0148], device='cuda:3'), out_proj_covar=tensor([9.3115e-05, 1.1122e-04, 8.9509e-05, 9.1377e-05, 9.3680e-05, 9.3461e-05, 1.0323e-04, 1.0698e-04], device='cuda:3') 2023-03-26 19:42:43,957 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89486.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:42:53,368 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89500.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 19:42:59,271 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89509.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:43:06,068 INFO [finetune.py:976] (3/7) Epoch 16, batch 3600, loss[loss=0.1865, simple_loss=0.2572, pruned_loss=0.05793, over 4741.00 frames. ], tot_loss[loss=0.1798, simple_loss=0.2476, pruned_loss=0.05606, over 953623.34 frames. ], batch size: 59, lr: 3.43e-03, grad_scale: 32.0 2023-03-26 19:43:14,212 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89525.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:43:37,730 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.924e+01 1.526e+02 1.890e+02 2.215e+02 3.895e+02, threshold=3.780e+02, percent-clipped=1.0 2023-03-26 19:44:03,076 INFO [finetune.py:976] (3/7) Epoch 16, batch 3650, loss[loss=0.1852, simple_loss=0.2684, pruned_loss=0.05103, over 4799.00 frames. ], tot_loss[loss=0.183, simple_loss=0.2506, pruned_loss=0.05771, over 951547.28 frames. ], batch size: 41, lr: 3.43e-03, grad_scale: 32.0 2023-03-26 19:44:06,132 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89570.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:44:36,729 INFO [finetune.py:976] (3/7) Epoch 16, batch 3700, loss[loss=0.2022, simple_loss=0.2813, pruned_loss=0.06153, over 4929.00 frames. ], tot_loss[loss=0.1848, simple_loss=0.2533, pruned_loss=0.05816, over 952547.73 frames. ], batch size: 38, lr: 3.43e-03, grad_scale: 32.0 2023-03-26 19:44:57,082 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.548e+01 1.593e+02 1.994e+02 2.376e+02 3.738e+02, threshold=3.989e+02, percent-clipped=0.0 2023-03-26 19:45:10,208 INFO [finetune.py:976] (3/7) Epoch 16, batch 3750, loss[loss=0.1901, simple_loss=0.258, pruned_loss=0.06106, over 4788.00 frames. ], tot_loss[loss=0.1873, simple_loss=0.2558, pruned_loss=0.0594, over 953757.41 frames. ], batch size: 29, lr: 3.43e-03, grad_scale: 32.0 2023-03-26 19:45:19,315 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89680.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:45:21,101 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89683.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:45:43,307 INFO [finetune.py:976] (3/7) Epoch 16, batch 3800, loss[loss=0.1752, simple_loss=0.2486, pruned_loss=0.05092, over 4886.00 frames. ], tot_loss[loss=0.1881, simple_loss=0.257, pruned_loss=0.05954, over 954488.07 frames. ], batch size: 32, lr: 3.43e-03, grad_scale: 32.0 2023-03-26 19:45:49,894 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8148, 1.7358, 1.5766, 1.7586, 1.1684, 4.3689, 1.7056, 2.1335], device='cuda:3'), covar=tensor([0.3379, 0.2474, 0.2249, 0.2308, 0.1870, 0.0111, 0.2471, 0.1231], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0120, 0.0124, 0.0114, 0.0097, 0.0096, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 19:45:52,763 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89730.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:45:53,366 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=89731.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:46:00,040 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89741.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:46:03,526 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.091e+02 1.549e+02 1.882e+02 2.353e+02 4.344e+02, threshold=3.764e+02, percent-clipped=2.0 2023-03-26 19:46:19,024 INFO [finetune.py:976] (3/7) Epoch 16, batch 3850, loss[loss=0.178, simple_loss=0.2476, pruned_loss=0.05419, over 4899.00 frames. ], tot_loss[loss=0.1872, simple_loss=0.2562, pruned_loss=0.05915, over 955512.89 frames. ], batch size: 43, lr: 3.43e-03, grad_scale: 32.0 2023-03-26 19:46:19,937 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-26 19:46:29,104 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89781.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:46:32,269 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-03-26 19:46:35,778 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2706, 2.2550, 1.8904, 2.2912, 2.1476, 2.0919, 2.1055, 2.9703], device='cuda:3'), covar=tensor([0.4044, 0.4936, 0.3532, 0.4634, 0.4431, 0.2655, 0.4799, 0.1879], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0262, 0.0226, 0.0276, 0.0249, 0.0217, 0.0251, 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') 2023-03-26 19:46:37,142 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-03-26 19:46:38,079 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89795.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 19:46:38,755 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0019, 2.0425, 1.6921, 2.0974, 2.0898, 1.8027, 2.4400, 2.1182], device='cuda:3'), covar=tensor([0.1397, 0.2071, 0.2968, 0.2406, 0.2395, 0.1613, 0.3191, 0.1561], device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0188, 0.0234, 0.0253, 0.0245, 0.0202, 0.0212, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 19:46:52,158 INFO [finetune.py:976] (3/7) Epoch 16, batch 3900, loss[loss=0.1532, simple_loss=0.2262, pruned_loss=0.04011, over 4874.00 frames. ], tot_loss[loss=0.1856, simple_loss=0.2539, pruned_loss=0.05862, over 955875.99 frames. ], batch size: 31, lr: 3.43e-03, grad_scale: 32.0 2023-03-26 19:46:58,261 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89825.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:47:12,472 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.084e+02 1.500e+02 1.857e+02 2.274e+02 5.172e+02, threshold=3.715e+02, percent-clipped=1.0 2023-03-26 19:47:23,889 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89865.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:47:23,936 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89865.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:47:24,440 INFO [finetune.py:976] (3/7) Epoch 16, batch 3950, loss[loss=0.1575, simple_loss=0.2282, pruned_loss=0.04342, over 4908.00 frames. ], tot_loss[loss=0.1829, simple_loss=0.2507, pruned_loss=0.05751, over 956579.21 frames. ], batch size: 35, lr: 3.43e-03, grad_scale: 32.0 2023-03-26 19:47:29,790 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=89873.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:47:30,478 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8130, 1.7670, 1.5382, 1.9018, 2.1927, 1.9867, 1.4082, 1.4870], device='cuda:3'), covar=tensor([0.2148, 0.1906, 0.1948, 0.1712, 0.1755, 0.1180, 0.2600, 0.2030], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0209, 0.0211, 0.0192, 0.0243, 0.0185, 0.0216, 0.0200], 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') 2023-03-26 19:47:57,664 INFO [finetune.py:976] (3/7) Epoch 16, batch 4000, loss[loss=0.164, simple_loss=0.2252, pruned_loss=0.05145, over 4298.00 frames. ], tot_loss[loss=0.1815, simple_loss=0.2488, pruned_loss=0.05709, over 953169.64 frames. ], batch size: 18, lr: 3.43e-03, grad_scale: 32.0 2023-03-26 19:48:04,750 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89926.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 19:48:18,302 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.466e+01 1.676e+02 1.974e+02 2.469e+02 4.779e+02, threshold=3.947e+02, percent-clipped=6.0 2023-03-26 19:48:32,942 INFO [finetune.py:976] (3/7) Epoch 16, batch 4050, loss[loss=0.2008, simple_loss=0.2693, pruned_loss=0.06612, over 4789.00 frames. ], tot_loss[loss=0.1844, simple_loss=0.252, pruned_loss=0.05845, over 950289.21 frames. ], batch size: 29, lr: 3.43e-03, grad_scale: 64.0 2023-03-26 19:48:33,119 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.11 vs. limit=2.0 2023-03-26 19:48:36,745 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7763, 1.7012, 2.0244, 1.5036, 1.9087, 1.9445, 1.5747, 2.1551], device='cuda:3'), covar=tensor([0.1116, 0.1596, 0.1241, 0.1579, 0.0690, 0.1204, 0.2187, 0.0650], device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0205, 0.0192, 0.0192, 0.0177, 0.0214, 0.0219, 0.0200], 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') 2023-03-26 19:48:38,333 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5278, 1.1290, 0.7139, 1.3533, 1.8930, 0.8339, 1.2638, 1.3642], device='cuda:3'), covar=tensor([0.1585, 0.2187, 0.1867, 0.1264, 0.2040, 0.2069, 0.1548, 0.2040], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0095, 0.0111, 0.0093, 0.0119, 0.0095, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 19:48:44,184 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.54 vs. limit=5.0 2023-03-26 19:49:01,338 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.86 vs. limit=5.0 2023-03-26 19:49:02,350 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2022, 1.6946, 1.7087, 0.9916, 1.9053, 2.1074, 1.8807, 1.6329], device='cuda:3'), covar=tensor([0.0814, 0.0627, 0.0543, 0.0683, 0.0559, 0.0537, 0.0451, 0.0649], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0151, 0.0124, 0.0127, 0.0131, 0.0128, 0.0142, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.2370e-05, 1.0985e-04, 8.8768e-05, 9.0640e-05, 9.2764e-05, 9.2439e-05, 1.0221e-04, 1.0601e-04], device='cuda:3') 2023-03-26 19:49:29,803 INFO [finetune.py:976] (3/7) Epoch 16, batch 4100, loss[loss=0.2504, simple_loss=0.3043, pruned_loss=0.09823, over 4746.00 frames. ], tot_loss[loss=0.1862, simple_loss=0.2548, pruned_loss=0.05876, over 951926.42 frames. ], batch size: 59, lr: 3.43e-03, grad_scale: 64.0 2023-03-26 19:49:43,228 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90030.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:49:46,946 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90036.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:49:54,440 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.029e+02 1.559e+02 1.839e+02 2.160e+02 6.359e+02, threshold=3.678e+02, percent-clipped=1.0 2023-03-26 19:50:06,383 INFO [finetune.py:976] (3/7) Epoch 16, batch 4150, loss[loss=0.2247, simple_loss=0.2955, pruned_loss=0.077, over 4818.00 frames. ], tot_loss[loss=0.188, simple_loss=0.2564, pruned_loss=0.0598, over 951409.44 frames. ], batch size: 38, lr: 3.43e-03, grad_scale: 64.0 2023-03-26 19:50:14,191 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=90078.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:50:16,527 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90081.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:50:26,425 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90095.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 19:50:39,512 INFO [finetune.py:976] (3/7) Epoch 16, batch 4200, loss[loss=0.1687, simple_loss=0.2432, pruned_loss=0.04709, over 4767.00 frames. ], tot_loss[loss=0.1875, simple_loss=0.2563, pruned_loss=0.05934, over 950915.37 frames. ], batch size: 51, lr: 3.43e-03, grad_scale: 64.0 2023-03-26 19:50:42,246 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.81 vs. limit=5.0 2023-03-26 19:50:47,973 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=90129.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:50:48,627 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90130.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:50:57,833 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=90143.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 19:51:00,648 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.087e+02 1.547e+02 1.785e+02 2.134e+02 3.751e+02, threshold=3.570e+02, percent-clipped=1.0 2023-03-26 19:51:12,335 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90165.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:51:12,839 INFO [finetune.py:976] (3/7) Epoch 16, batch 4250, loss[loss=0.1657, simple_loss=0.2343, pruned_loss=0.04855, over 4765.00 frames. ], tot_loss[loss=0.1845, simple_loss=0.2532, pruned_loss=0.05788, over 951653.85 frames. ], batch size: 28, lr: 3.43e-03, grad_scale: 64.0 2023-03-26 19:51:29,566 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90191.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:51:43,666 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=90213.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:51:45,896 INFO [finetune.py:976] (3/7) Epoch 16, batch 4300, loss[loss=0.1721, simple_loss=0.2372, pruned_loss=0.0535, over 4833.00 frames. ], tot_loss[loss=0.1817, simple_loss=0.2502, pruned_loss=0.05657, over 953211.41 frames. ], batch size: 30, lr: 3.43e-03, grad_scale: 32.0 2023-03-26 19:51:48,954 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90221.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 19:51:56,178 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90232.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:52:07,167 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.180e+02 1.551e+02 1.797e+02 2.190e+02 3.764e+02, threshold=3.594e+02, percent-clipped=1.0 2023-03-26 19:52:18,552 INFO [finetune.py:976] (3/7) Epoch 16, batch 4350, loss[loss=0.1624, simple_loss=0.2301, pruned_loss=0.04738, over 4829.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.2463, pruned_loss=0.05502, over 955060.14 frames. ], batch size: 30, lr: 3.43e-03, grad_scale: 32.0 2023-03-26 19:52:34,167 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8141, 1.6932, 1.6467, 1.8015, 1.2207, 3.3982, 1.3815, 1.7268], device='cuda:3'), covar=tensor([0.3195, 0.2468, 0.1972, 0.2202, 0.1762, 0.0228, 0.2435, 0.1303], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0120, 0.0123, 0.0114, 0.0097, 0.0096, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 19:52:36,981 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90293.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:52:51,897 INFO [finetune.py:976] (3/7) Epoch 16, batch 4400, loss[loss=0.1902, simple_loss=0.2688, pruned_loss=0.0558, over 4849.00 frames. ], tot_loss[loss=0.1797, simple_loss=0.248, pruned_loss=0.05576, over 954361.84 frames. ], batch size: 47, lr: 3.43e-03, grad_scale: 32.0 2023-03-26 19:53:04,942 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90336.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:53:13,597 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.069e+02 1.586e+02 1.845e+02 2.241e+02 4.760e+02, threshold=3.689e+02, percent-clipped=4.0 2023-03-26 19:53:14,911 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8880, 1.2920, 1.7115, 1.7508, 1.5857, 1.5819, 1.7120, 1.6795], device='cuda:3'), covar=tensor([0.4459, 0.4403, 0.4166, 0.4411, 0.5488, 0.4289, 0.5086, 0.3868], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0238, 0.0258, 0.0270, 0.0270, 0.0243, 0.0280, 0.0238], 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') 2023-03-26 19:53:15,504 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90351.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:53:25,405 INFO [finetune.py:976] (3/7) Epoch 16, batch 4450, loss[loss=0.2432, simple_loss=0.3032, pruned_loss=0.09165, over 4806.00 frames. ], tot_loss[loss=0.1828, simple_loss=0.252, pruned_loss=0.05679, over 951917.05 frames. ], batch size: 45, lr: 3.43e-03, grad_scale: 32.0 2023-03-26 19:53:37,365 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=90384.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:54:05,577 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90412.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:54:07,882 INFO [finetune.py:976] (3/7) Epoch 16, batch 4500, loss[loss=0.1764, simple_loss=0.2505, pruned_loss=0.05111, over 4777.00 frames. ], tot_loss[loss=0.1834, simple_loss=0.2526, pruned_loss=0.05714, over 950171.78 frames. ], batch size: 51, lr: 3.43e-03, grad_scale: 32.0 2023-03-26 19:54:40,951 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.141e+02 1.659e+02 2.056e+02 2.631e+02 3.688e+02, threshold=4.111e+02, percent-clipped=0.0 2023-03-26 19:54:53,414 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90459.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:54:54,057 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.3336, 1.2682, 1.2519, 0.7163, 1.2219, 1.4308, 1.4842, 1.1948], device='cuda:3'), covar=tensor([0.0729, 0.0492, 0.0472, 0.0437, 0.0502, 0.0471, 0.0300, 0.0511], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0151, 0.0124, 0.0127, 0.0131, 0.0128, 0.0142, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.2414e-05, 1.0982e-04, 8.8616e-05, 9.0594e-05, 9.2347e-05, 9.2607e-05, 1.0236e-04, 1.0636e-04], device='cuda:3') 2023-03-26 19:54:54,230 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 2023-03-26 19:55:01,919 INFO [finetune.py:976] (3/7) Epoch 16, batch 4550, loss[loss=0.1609, simple_loss=0.2093, pruned_loss=0.0563, over 4195.00 frames. ], tot_loss[loss=0.1843, simple_loss=0.2537, pruned_loss=0.05742, over 949549.06 frames. ], batch size: 18, lr: 3.43e-03, grad_scale: 32.0 2023-03-26 19:55:18,094 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90486.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:55:24,708 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90496.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:55:38,629 INFO [finetune.py:976] (3/7) Epoch 16, batch 4600, loss[loss=0.1692, simple_loss=0.2363, pruned_loss=0.05106, over 4898.00 frames. ], tot_loss[loss=0.1842, simple_loss=0.2543, pruned_loss=0.057, over 951674.82 frames. ], batch size: 43, lr: 3.43e-03, grad_scale: 32.0 2023-03-26 19:55:41,630 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90520.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:55:42,187 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90521.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 19:55:59,291 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.556e+01 1.469e+02 1.715e+02 2.012e+02 4.010e+02, threshold=3.429e+02, percent-clipped=0.0 2023-03-26 19:56:06,282 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90557.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:56:11,542 INFO [finetune.py:976] (3/7) Epoch 16, batch 4650, loss[loss=0.2173, simple_loss=0.2664, pruned_loss=0.0841, over 4897.00 frames. ], tot_loss[loss=0.1818, simple_loss=0.2513, pruned_loss=0.05616, over 952524.95 frames. ], batch size: 32, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 19:56:13,934 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=90569.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:56:16,284 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1970, 2.0732, 2.2653, 0.9422, 2.5477, 2.7144, 2.3913, 2.0508], device='cuda:3'), covar=tensor([0.0923, 0.0848, 0.0497, 0.0829, 0.0456, 0.0814, 0.0470, 0.0696], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0151, 0.0124, 0.0127, 0.0131, 0.0129, 0.0143, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.2560e-05, 1.1012e-04, 8.8600e-05, 9.0669e-05, 9.2427e-05, 9.2681e-05, 1.0261e-04, 1.0650e-04], device='cuda:3') 2023-03-26 19:56:21,053 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7525, 0.9516, 1.6357, 1.6342, 1.5238, 1.4916, 1.5526, 1.6386], device='cuda:3'), covar=tensor([0.4595, 0.4389, 0.3775, 0.3890, 0.4939, 0.4103, 0.4659, 0.3829], device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0240, 0.0260, 0.0273, 0.0272, 0.0246, 0.0283, 0.0239], 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') 2023-03-26 19:56:26,167 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90588.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:56:30,984 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3669, 1.2573, 1.1943, 1.4140, 1.6491, 1.4920, 1.3782, 1.2471], device='cuda:3'), covar=tensor([0.0290, 0.0294, 0.0582, 0.0271, 0.0190, 0.0503, 0.0291, 0.0351], device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0108, 0.0144, 0.0112, 0.0100, 0.0107, 0.0098, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.3460e-05, 8.3558e-05, 1.1351e-04, 8.6812e-05, 7.7535e-05, 7.9331e-05, 7.3629e-05, 8.2753e-05], device='cuda:3') 2023-03-26 19:56:45,056 INFO [finetune.py:976] (3/7) Epoch 16, batch 4700, loss[loss=0.1639, simple_loss=0.2308, pruned_loss=0.04846, over 4819.00 frames. ], tot_loss[loss=0.1814, simple_loss=0.2497, pruned_loss=0.05652, over 953482.59 frames. ], batch size: 25, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 19:56:59,925 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-26 19:57:04,049 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-26 19:57:05,664 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.103e+02 1.593e+02 1.858e+02 2.163e+02 3.767e+02, threshold=3.717e+02, percent-clipped=1.0 2023-03-26 19:57:14,607 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3940, 1.3861, 1.7217, 2.4271, 1.6225, 2.1375, 1.0011, 2.0923], device='cuda:3'), covar=tensor([0.1579, 0.1347, 0.1036, 0.0698, 0.0904, 0.1247, 0.1358, 0.0588], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0115, 0.0132, 0.0163, 0.0100, 0.0137, 0.0123, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 19:57:18,492 INFO [finetune.py:976] (3/7) Epoch 16, batch 4750, loss[loss=0.1548, simple_loss=0.2107, pruned_loss=0.04943, over 4737.00 frames. ], tot_loss[loss=0.1811, simple_loss=0.2488, pruned_loss=0.05671, over 955010.73 frames. ], batch size: 23, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 19:57:37,733 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0229, 1.8341, 2.2385, 1.4238, 1.9322, 2.2962, 1.7486, 2.3335], device='cuda:3'), covar=tensor([0.1216, 0.1875, 0.1250, 0.1841, 0.0990, 0.1167, 0.2672, 0.0794], device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0206, 0.0193, 0.0192, 0.0177, 0.0214, 0.0220, 0.0201], 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') 2023-03-26 19:57:45,626 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90707.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:57:52,462 INFO [finetune.py:976] (3/7) Epoch 16, batch 4800, loss[loss=0.1596, simple_loss=0.2354, pruned_loss=0.04193, over 4895.00 frames. ], tot_loss[loss=0.183, simple_loss=0.2509, pruned_loss=0.0576, over 953131.50 frames. ], batch size: 32, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 19:57:53,864 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.92 vs. limit=5.0 2023-03-26 19:58:13,286 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.231e+02 1.629e+02 1.979e+02 2.321e+02 4.531e+02, threshold=3.957e+02, percent-clipped=1.0 2023-03-26 19:58:25,071 INFO [finetune.py:976] (3/7) Epoch 16, batch 4850, loss[loss=0.2921, simple_loss=0.3411, pruned_loss=0.1215, over 4197.00 frames. ], tot_loss[loss=0.1847, simple_loss=0.2534, pruned_loss=0.05801, over 953747.55 frames. ], batch size: 65, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 19:58:28,872 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-03-26 19:58:35,485 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6057, 1.6394, 2.1136, 1.9480, 1.8455, 3.6130, 1.4948, 1.6736], device='cuda:3'), covar=tensor([0.1015, 0.1736, 0.1073, 0.0926, 0.1506, 0.0297, 0.1505, 0.1802], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0082, 0.0074, 0.0078, 0.0092, 0.0081, 0.0086, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 19:58:39,085 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90786.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:58:57,840 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90815.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:58:58,402 INFO [finetune.py:976] (3/7) Epoch 16, batch 4900, loss[loss=0.1754, simple_loss=0.2565, pruned_loss=0.04713, over 4817.00 frames. ], tot_loss[loss=0.1867, simple_loss=0.2555, pruned_loss=0.0589, over 954946.55 frames. ], batch size: 38, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 19:59:04,114 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0954, 1.9775, 1.6038, 1.8686, 2.1200, 1.7656, 2.1910, 2.0683], device='cuda:3'), covar=tensor([0.1411, 0.2019, 0.3137, 0.2617, 0.2508, 0.1759, 0.3198, 0.1839], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0187, 0.0234, 0.0253, 0.0245, 0.0202, 0.0212, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 19:59:11,165 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=90834.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:59:11,780 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1066, 3.6071, 3.7794, 3.8890, 3.8981, 3.6732, 4.1558, 1.7746], device='cuda:3'), covar=tensor([0.0733, 0.0772, 0.0722, 0.0919, 0.0975, 0.1254, 0.0646, 0.4722], device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0242, 0.0274, 0.0291, 0.0331, 0.0278, 0.0296, 0.0293], 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') 2023-03-26 19:59:24,122 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.061e+02 1.543e+02 1.926e+02 2.205e+02 3.945e+02, threshold=3.852e+02, percent-clipped=0.0 2023-03-26 19:59:25,322 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90849.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:59:26,597 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1645, 2.0428, 1.7034, 1.9810, 1.9240, 1.9015, 1.9720, 2.6899], device='cuda:3'), covar=tensor([0.4005, 0.4280, 0.3566, 0.4013, 0.4223, 0.2710, 0.3933, 0.1842], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0260, 0.0225, 0.0275, 0.0248, 0.0216, 0.0249, 0.0228], 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') 2023-03-26 19:59:27,125 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90852.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 19:59:42,914 INFO [finetune.py:976] (3/7) Epoch 16, batch 4950, loss[loss=0.2019, simple_loss=0.2694, pruned_loss=0.06724, over 4783.00 frames. ], tot_loss[loss=0.1877, simple_loss=0.2566, pruned_loss=0.05936, over 953605.25 frames. ], batch size: 51, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 20:00:12,900 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90888.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:00:34,905 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90910.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:00:38,891 INFO [finetune.py:976] (3/7) Epoch 16, batch 5000, loss[loss=0.1562, simple_loss=0.2269, pruned_loss=0.0428, over 4717.00 frames. ], tot_loss[loss=0.1845, simple_loss=0.2536, pruned_loss=0.05775, over 954199.52 frames. ], batch size: 54, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 20:00:53,037 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=90936.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:01:00,206 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.882e+01 1.531e+02 1.843e+02 2.134e+02 5.620e+02, threshold=3.687e+02, percent-clipped=2.0 2023-03-26 20:01:11,977 INFO [finetune.py:976] (3/7) Epoch 16, batch 5050, loss[loss=0.1649, simple_loss=0.2337, pruned_loss=0.04812, over 4823.00 frames. ], tot_loss[loss=0.1823, simple_loss=0.251, pruned_loss=0.05682, over 954063.28 frames. ], batch size: 25, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 20:01:29,842 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7853, 1.6846, 1.6285, 1.7075, 1.4537, 3.5623, 1.5834, 2.0986], device='cuda:3'), covar=tensor([0.3181, 0.2554, 0.2165, 0.2399, 0.1640, 0.0232, 0.2503, 0.1225], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0115, 0.0120, 0.0123, 0.0113, 0.0097, 0.0096, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 20:01:40,193 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91007.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:01:45,573 INFO [finetune.py:976] (3/7) Epoch 16, batch 5100, loss[loss=0.1537, simple_loss=0.2216, pruned_loss=0.04295, over 4905.00 frames. ], tot_loss[loss=0.1785, simple_loss=0.2467, pruned_loss=0.05513, over 953881.56 frames. ], batch size: 32, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 20:02:07,770 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.054e+02 1.471e+02 1.778e+02 2.096e+02 3.940e+02, threshold=3.556e+02, percent-clipped=2.0 2023-03-26 20:02:12,114 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=91055.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:02:18,691 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.1011, 4.5816, 4.3473, 2.3830, 4.6953, 3.6151, 0.9981, 3.4422], device='cuda:3'), covar=tensor([0.2386, 0.1921, 0.1331, 0.3203, 0.0850, 0.0852, 0.4708, 0.1341], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0176, 0.0159, 0.0129, 0.0158, 0.0123, 0.0147, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 20:02:19,222 INFO [finetune.py:976] (3/7) Epoch 16, batch 5150, loss[loss=0.219, simple_loss=0.2832, pruned_loss=0.07741, over 4923.00 frames. ], tot_loss[loss=0.1808, simple_loss=0.249, pruned_loss=0.05633, over 954010.01 frames. ], batch size: 38, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 20:02:52,457 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91115.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:02:52,958 INFO [finetune.py:976] (3/7) Epoch 16, batch 5200, loss[loss=0.1723, simple_loss=0.2503, pruned_loss=0.04712, over 4897.00 frames. ], tot_loss[loss=0.1839, simple_loss=0.2526, pruned_loss=0.05761, over 954961.38 frames. ], batch size: 35, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 20:03:10,804 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5338, 1.4285, 1.9067, 1.7907, 1.5481, 3.3719, 1.3092, 1.4887], device='cuda:3'), covar=tensor([0.0894, 0.1730, 0.1069, 0.0903, 0.1502, 0.0248, 0.1493, 0.1756], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0074, 0.0078, 0.0092, 0.0080, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 20:03:14,339 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.009e+02 1.594e+02 1.966e+02 2.465e+02 4.658e+02, threshold=3.932e+02, percent-clipped=3.0 2023-03-26 20:03:17,844 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91152.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:03:24,390 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=91163.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:03:26,646 INFO [finetune.py:976] (3/7) Epoch 16, batch 5250, loss[loss=0.1523, simple_loss=0.2145, pruned_loss=0.0451, over 4698.00 frames. ], tot_loss[loss=0.1873, simple_loss=0.2561, pruned_loss=0.05926, over 954432.45 frames. ], batch size: 23, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 20:03:28,512 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2731, 2.8816, 2.7219, 1.2702, 3.0139, 2.1997, 0.9064, 1.8007], device='cuda:3'), covar=tensor([0.2565, 0.2066, 0.1868, 0.3390, 0.1371, 0.1159, 0.3768, 0.1587], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0176, 0.0160, 0.0129, 0.0158, 0.0123, 0.0147, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 20:03:28,564 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3187, 1.2357, 1.5736, 1.0805, 1.2137, 1.4429, 1.2509, 1.6200], device='cuda:3'), covar=tensor([0.1162, 0.2043, 0.1238, 0.1537, 0.1048, 0.1209, 0.2903, 0.0851], device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0205, 0.0192, 0.0192, 0.0177, 0.0213, 0.0219, 0.0201], 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') 2023-03-26 20:03:31,543 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9514, 1.3864, 0.7374, 1.7794, 2.2682, 1.7131, 1.4748, 1.8158], device='cuda:3'), covar=tensor([0.1877, 0.2839, 0.2664, 0.1593, 0.2285, 0.2872, 0.2036, 0.2504], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0096, 0.0112, 0.0093, 0.0120, 0.0095, 0.0099, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-26 20:03:49,259 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=91200.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:03:53,268 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91205.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:03:59,840 INFO [finetune.py:976] (3/7) Epoch 16, batch 5300, loss[loss=0.2079, simple_loss=0.2685, pruned_loss=0.07359, over 4892.00 frames. ], tot_loss[loss=0.1885, simple_loss=0.2574, pruned_loss=0.05982, over 954775.07 frames. ], batch size: 32, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 20:04:04,110 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9037, 1.6455, 2.0777, 1.3009, 1.8458, 2.0972, 1.5870, 2.2699], device='cuda:3'), covar=tensor([0.1264, 0.2266, 0.1497, 0.2097, 0.0969, 0.1355, 0.2862, 0.0715], device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0203, 0.0191, 0.0191, 0.0176, 0.0212, 0.0218, 0.0200], 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') 2023-03-26 20:04:21,113 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.091e+02 1.597e+02 1.843e+02 2.222e+02 3.769e+02, threshold=3.686e+02, percent-clipped=0.0 2023-03-26 20:04:24,259 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-26 20:04:33,500 INFO [finetune.py:976] (3/7) Epoch 16, batch 5350, loss[loss=0.186, simple_loss=0.2569, pruned_loss=0.05749, over 4814.00 frames. ], tot_loss[loss=0.1887, simple_loss=0.2574, pruned_loss=0.05994, over 954425.57 frames. ], batch size: 39, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 20:04:50,640 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91285.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:05:29,810 INFO [finetune.py:976] (3/7) Epoch 16, batch 5400, loss[loss=0.1968, simple_loss=0.2667, pruned_loss=0.06342, over 4929.00 frames. ], tot_loss[loss=0.1869, simple_loss=0.2557, pruned_loss=0.05909, over 955172.10 frames. ], batch size: 38, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 20:05:58,067 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1052, 1.8800, 2.3445, 4.0935, 2.8887, 2.7303, 1.2005, 3.3084], device='cuda:3'), covar=tensor([0.1604, 0.1403, 0.1382, 0.0483, 0.0668, 0.1542, 0.1666, 0.0448], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0116, 0.0133, 0.0165, 0.0101, 0.0138, 0.0125, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 20:06:01,742 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91346.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:06:03,346 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.781e+01 1.583e+02 1.812e+02 2.291e+02 3.767e+02, threshold=3.624e+02, percent-clipped=1.0 2023-03-26 20:06:15,750 INFO [finetune.py:976] (3/7) Epoch 16, batch 5450, loss[loss=0.2001, simple_loss=0.2635, pruned_loss=0.06838, over 4826.00 frames. ], tot_loss[loss=0.1848, simple_loss=0.2528, pruned_loss=0.05846, over 955750.77 frames. ], batch size: 40, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 20:06:49,417 INFO [finetune.py:976] (3/7) Epoch 16, batch 5500, loss[loss=0.1772, simple_loss=0.2412, pruned_loss=0.05658, over 4908.00 frames. ], tot_loss[loss=0.1821, simple_loss=0.2494, pruned_loss=0.05741, over 953149.96 frames. ], batch size: 32, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 20:07:10,223 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.789e+01 1.460e+02 1.744e+02 2.187e+02 6.443e+02, threshold=3.488e+02, percent-clipped=1.0 2023-03-26 20:07:17,453 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.1704, 2.8328, 2.5088, 1.3819, 2.6517, 2.2140, 2.2046, 2.3800], device='cuda:3'), covar=tensor([0.0858, 0.0872, 0.1814, 0.2092, 0.1799, 0.2268, 0.1991, 0.1220], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0194, 0.0198, 0.0182, 0.0211, 0.0205, 0.0222, 0.0195], 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') 2023-03-26 20:07:22,093 INFO [finetune.py:976] (3/7) Epoch 16, batch 5550, loss[loss=0.1725, simple_loss=0.2472, pruned_loss=0.04892, over 4897.00 frames. ], tot_loss[loss=0.183, simple_loss=0.2506, pruned_loss=0.05772, over 953655.95 frames. ], batch size: 43, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 20:07:38,508 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.92 vs. limit=5.0 2023-03-26 20:07:47,588 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91505.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:07:53,902 INFO [finetune.py:976] (3/7) Epoch 16, batch 5600, loss[loss=0.2103, simple_loss=0.2698, pruned_loss=0.07536, over 4934.00 frames. ], tot_loss[loss=0.185, simple_loss=0.2538, pruned_loss=0.05807, over 954111.87 frames. ], batch size: 33, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 20:08:08,983 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91542.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 20:08:13,217 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.075e+02 1.603e+02 1.969e+02 2.458e+02 5.397e+02, threshold=3.938e+02, percent-clipped=5.0 2023-03-26 20:08:16,176 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=91553.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:08:23,624 INFO [finetune.py:976] (3/7) Epoch 16, batch 5650, loss[loss=0.1611, simple_loss=0.2334, pruned_loss=0.04442, over 4775.00 frames. ], tot_loss[loss=0.1862, simple_loss=0.2554, pruned_loss=0.05853, over 952316.96 frames. ], batch size: 26, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 20:08:43,914 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7357, 2.7178, 2.5599, 2.9009, 3.4784, 2.9049, 2.9203, 2.3729], device='cuda:3'), covar=tensor([0.1948, 0.1612, 0.1583, 0.1403, 0.1386, 0.0860, 0.1588, 0.1684], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0207, 0.0211, 0.0190, 0.0241, 0.0185, 0.0215, 0.0200], 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') 2023-03-26 20:08:45,100 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2931, 1.9668, 1.3450, 0.5735, 1.5826, 1.9683, 1.7143, 1.7810], device='cuda:3'), covar=tensor([0.0977, 0.0852, 0.1553, 0.1954, 0.1448, 0.2269, 0.2256, 0.0898], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0195, 0.0199, 0.0182, 0.0212, 0.0206, 0.0223, 0.0196], 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') 2023-03-26 20:08:45,668 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91603.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 20:08:53,175 INFO [finetune.py:976] (3/7) Epoch 16, batch 5700, loss[loss=0.1629, simple_loss=0.2165, pruned_loss=0.05467, over 4191.00 frames. ], tot_loss[loss=0.1833, simple_loss=0.2514, pruned_loss=0.05757, over 933848.48 frames. ], batch size: 18, lr: 3.42e-03, grad_scale: 32.0 2023-03-26 20:08:54,299 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-03-26 20:09:07,888 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91641.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:09:21,369 INFO [finetune.py:976] (3/7) Epoch 17, batch 0, loss[loss=0.2472, simple_loss=0.3095, pruned_loss=0.09245, over 4922.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3095, pruned_loss=0.09245, over 4922.00 frames. ], batch size: 42, lr: 3.41e-03, grad_scale: 32.0 2023-03-26 20:09:21,370 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 20:09:27,352 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.1750, 1.3770, 1.3989, 0.7530, 1.3673, 1.5882, 1.6426, 1.3095], device='cuda:3'), covar=tensor([0.0856, 0.0543, 0.0521, 0.0467, 0.0465, 0.0635, 0.0320, 0.0631], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0151, 0.0124, 0.0128, 0.0131, 0.0129, 0.0143, 0.0148], device='cuda:3'), out_proj_covar=tensor([9.2894e-05, 1.0965e-04, 8.8859e-05, 9.0981e-05, 9.2436e-05, 9.2722e-05, 1.0308e-04, 1.0671e-04], device='cuda:3') 2023-03-26 20:09:29,000 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6383, 1.5549, 1.5147, 1.5008, 0.9849, 2.9971, 1.1150, 1.5969], device='cuda:3'), covar=tensor([0.3487, 0.2554, 0.2176, 0.2512, 0.1969, 0.0250, 0.2712, 0.1318], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0120, 0.0123, 0.0114, 0.0097, 0.0096, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 20:09:32,014 INFO [finetune.py:1010] (3/7) Epoch 17, validation: loss=0.1591, simple_loss=0.2283, pruned_loss=0.04492, over 2265189.00 frames. 2023-03-26 20:09:32,015 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6434MB 2023-03-26 20:09:35,486 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.774e+01 1.479e+02 1.757e+02 2.057e+02 5.096e+02, threshold=3.514e+02, percent-clipped=1.0 2023-03-26 20:10:00,080 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0 2023-03-26 20:10:07,336 INFO [finetune.py:976] (3/7) Epoch 17, batch 50, loss[loss=0.1997, simple_loss=0.2595, pruned_loss=0.06995, over 4923.00 frames. ], tot_loss[loss=0.1896, simple_loss=0.2573, pruned_loss=0.06094, over 216128.56 frames. ], batch size: 38, lr: 3.41e-03, grad_scale: 32.0 2023-03-26 20:10:28,371 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.53 vs. limit=5.0 2023-03-26 20:10:41,386 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.94 vs. limit=5.0 2023-03-26 20:10:52,796 INFO [finetune.py:976] (3/7) Epoch 17, batch 100, loss[loss=0.168, simple_loss=0.2362, pruned_loss=0.04987, over 4891.00 frames. ], tot_loss[loss=0.1827, simple_loss=0.2489, pruned_loss=0.05821, over 379846.52 frames. ], batch size: 35, lr: 3.41e-03, grad_scale: 32.0 2023-03-26 20:11:01,251 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.123e+02 1.600e+02 1.810e+02 2.096e+02 3.529e+02, threshold=3.620e+02, percent-clipped=1.0 2023-03-26 20:11:37,625 INFO [finetune.py:976] (3/7) Epoch 17, batch 150, loss[loss=0.2003, simple_loss=0.2636, pruned_loss=0.06846, over 4816.00 frames. ], tot_loss[loss=0.1783, simple_loss=0.245, pruned_loss=0.0558, over 508303.73 frames. ], batch size: 41, lr: 3.41e-03, grad_scale: 32.0 2023-03-26 20:12:11,016 INFO [finetune.py:976] (3/7) Epoch 17, batch 200, loss[loss=0.1896, simple_loss=0.2526, pruned_loss=0.06333, over 4771.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2438, pruned_loss=0.05502, over 607960.41 frames. ], batch size: 29, lr: 3.41e-03, grad_scale: 32.0 2023-03-26 20:12:11,696 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6789, 1.6259, 2.0968, 1.9821, 1.7547, 3.1066, 1.4811, 1.6675], device='cuda:3'), covar=tensor([0.0882, 0.1528, 0.1194, 0.0822, 0.1383, 0.0323, 0.1387, 0.1518], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0074, 0.0077, 0.0091, 0.0080, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 20:12:14,523 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.078e+02 1.595e+02 1.938e+02 2.273e+02 4.627e+02, threshold=3.876e+02, percent-clipped=4.0 2023-03-26 20:12:43,713 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91891.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:12:44,794 INFO [finetune.py:976] (3/7) Epoch 17, batch 250, loss[loss=0.2192, simple_loss=0.2904, pruned_loss=0.07397, over 4903.00 frames. ], tot_loss[loss=0.1803, simple_loss=0.2475, pruned_loss=0.05658, over 682419.00 frames. ], batch size: 37, lr: 3.41e-03, grad_scale: 32.0 2023-03-26 20:12:47,909 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91898.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 20:12:50,136 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4127, 1.3252, 1.2255, 1.4407, 1.7737, 1.5346, 1.4430, 1.2480], device='cuda:3'), covar=tensor([0.0334, 0.0344, 0.0582, 0.0286, 0.0185, 0.0616, 0.0309, 0.0405], device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0108, 0.0144, 0.0113, 0.0100, 0.0108, 0.0098, 0.0109], device='cuda:3'), out_proj_covar=tensor([7.4044e-05, 8.3918e-05, 1.1392e-04, 8.7083e-05, 7.7791e-05, 7.9603e-05, 7.3818e-05, 8.3000e-05], device='cuda:3') 2023-03-26 20:13:14,603 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4077, 2.4764, 2.3678, 1.8399, 2.4518, 2.7080, 2.6589, 2.0844], device='cuda:3'), covar=tensor([0.0642, 0.0604, 0.0685, 0.0859, 0.0758, 0.0631, 0.0570, 0.1174], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0134, 0.0141, 0.0124, 0.0124, 0.0140, 0.0141, 0.0165], 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') 2023-03-26 20:13:17,055 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91941.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:13:18,231 INFO [finetune.py:976] (3/7) Epoch 17, batch 300, loss[loss=0.2226, simple_loss=0.285, pruned_loss=0.08013, over 4777.00 frames. ], tot_loss[loss=0.1832, simple_loss=0.2516, pruned_loss=0.05738, over 743383.75 frames. ], batch size: 29, lr: 3.41e-03, grad_scale: 32.0 2023-03-26 20:13:21,759 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.873e+01 1.605e+02 2.003e+02 2.239e+02 3.510e+02, threshold=4.006e+02, percent-clipped=0.0 2023-03-26 20:13:24,432 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91952.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:13:27,181 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91955.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:13:30,068 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.4215, 1.5688, 1.5796, 0.9476, 1.5801, 1.8705, 1.8625, 1.4338], device='cuda:3'), covar=tensor([0.0983, 0.0567, 0.0531, 0.0538, 0.0436, 0.0527, 0.0282, 0.0704], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0151, 0.0123, 0.0128, 0.0131, 0.0128, 0.0143, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.2653e-05, 1.0950e-04, 8.8382e-05, 9.0729e-05, 9.2254e-05, 9.2388e-05, 1.0296e-04, 1.0613e-04], device='cuda:3') 2023-03-26 20:13:42,644 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0169, 1.8969, 1.6443, 1.8001, 1.7887, 1.7743, 1.8245, 2.5184], device='cuda:3'), covar=tensor([0.3740, 0.4294, 0.3381, 0.3902, 0.3976, 0.2518, 0.3935, 0.1759], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0262, 0.0228, 0.0277, 0.0251, 0.0218, 0.0250, 0.0231], 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') 2023-03-26 20:13:44,452 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91981.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 20:13:49,289 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=91989.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:13:52,142 INFO [finetune.py:976] (3/7) Epoch 17, batch 350, loss[loss=0.1978, simple_loss=0.2616, pruned_loss=0.06702, over 4808.00 frames. ], tot_loss[loss=0.1848, simple_loss=0.2533, pruned_loss=0.05818, over 791176.14 frames. ], batch size: 39, lr: 3.41e-03, grad_scale: 32.0 2023-03-26 20:13:54,094 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4781, 1.4458, 1.3582, 1.4403, 1.0056, 3.1196, 1.2668, 1.6103], device='cuda:3'), covar=tensor([0.3340, 0.2561, 0.2207, 0.2384, 0.1933, 0.0248, 0.2892, 0.1330], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0115, 0.0120, 0.0123, 0.0113, 0.0096, 0.0096, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 20:14:09,817 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92016.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:14:11,114 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-03-26 20:14:26,148 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92042.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 20:14:26,649 INFO [finetune.py:976] (3/7) Epoch 17, batch 400, loss[loss=0.1592, simple_loss=0.2371, pruned_loss=0.04066, over 4751.00 frames. ], tot_loss[loss=0.1846, simple_loss=0.2539, pruned_loss=0.05768, over 827837.18 frames. ], batch size: 54, lr: 3.41e-03, grad_scale: 32.0 2023-03-26 20:14:30,186 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.886e+01 1.544e+02 1.847e+02 2.163e+02 3.487e+02, threshold=3.695e+02, percent-clipped=0.0 2023-03-26 20:14:48,521 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-26 20:15:00,230 INFO [finetune.py:976] (3/7) Epoch 17, batch 450, loss[loss=0.1959, simple_loss=0.2624, pruned_loss=0.06468, over 4859.00 frames. ], tot_loss[loss=0.1828, simple_loss=0.2525, pruned_loss=0.05653, over 857750.14 frames. ], batch size: 31, lr: 3.41e-03, grad_scale: 32.0 2023-03-26 20:15:06,686 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0795, 1.7478, 2.3318, 4.0212, 2.7946, 2.7181, 1.1348, 3.4017], device='cuda:3'), covar=tensor([0.1619, 0.1488, 0.1402, 0.0527, 0.0714, 0.1481, 0.1697, 0.0430], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0116, 0.0133, 0.0165, 0.0101, 0.0137, 0.0124, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 20:15:10,306 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7597, 1.5869, 1.4446, 1.7465, 1.9304, 1.8270, 1.2601, 1.4840], device='cuda:3'), covar=tensor([0.2176, 0.2167, 0.1995, 0.1717, 0.1807, 0.1254, 0.2661, 0.1991], device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0209, 0.0213, 0.0192, 0.0244, 0.0186, 0.0217, 0.0202], 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') 2023-03-26 20:15:28,967 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.4567, 3.8548, 4.0517, 4.3022, 4.2537, 3.9142, 4.5455, 1.3930], device='cuda:3'), covar=tensor([0.0758, 0.0856, 0.0850, 0.0957, 0.1199, 0.1635, 0.0652, 0.5681], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0246, 0.0278, 0.0295, 0.0336, 0.0283, 0.0302, 0.0298], 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') 2023-03-26 20:15:33,735 INFO [finetune.py:976] (3/7) Epoch 17, batch 500, loss[loss=0.174, simple_loss=0.2357, pruned_loss=0.05614, over 4389.00 frames. ], tot_loss[loss=0.1819, simple_loss=0.251, pruned_loss=0.05643, over 880696.35 frames. ], batch size: 65, lr: 3.41e-03, grad_scale: 32.0 2023-03-26 20:15:37,216 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.042e+02 1.578e+02 1.863e+02 2.269e+02 4.074e+02, threshold=3.727e+02, percent-clipped=2.0 2023-03-26 20:16:30,575 INFO [finetune.py:976] (3/7) Epoch 17, batch 550, loss[loss=0.2003, simple_loss=0.2718, pruned_loss=0.06435, over 4813.00 frames. ], tot_loss[loss=0.1811, simple_loss=0.2492, pruned_loss=0.05648, over 897664.79 frames. ], batch size: 41, lr: 3.41e-03, grad_scale: 32.0 2023-03-26 20:16:33,717 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92198.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 20:16:58,530 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-03-26 20:17:13,334 INFO [finetune.py:976] (3/7) Epoch 17, batch 600, loss[loss=0.1959, simple_loss=0.2659, pruned_loss=0.06292, over 4917.00 frames. ], tot_loss[loss=0.1824, simple_loss=0.2503, pruned_loss=0.0572, over 910426.54 frames. ], batch size: 43, lr: 3.41e-03, grad_scale: 64.0 2023-03-26 20:17:15,215 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=92246.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 20:17:15,825 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92247.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:17:16,353 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.142e+02 1.709e+02 1.980e+02 2.349e+02 5.069e+02, threshold=3.960e+02, percent-clipped=5.0 2023-03-26 20:17:47,091 INFO [finetune.py:976] (3/7) Epoch 17, batch 650, loss[loss=0.1414, simple_loss=0.2133, pruned_loss=0.03471, over 4776.00 frames. ], tot_loss[loss=0.1834, simple_loss=0.2519, pruned_loss=0.05749, over 917262.33 frames. ], batch size: 26, lr: 3.41e-03, grad_scale: 64.0 2023-03-26 20:17:58,667 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92311.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:18:16,781 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92337.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 20:18:20,699 INFO [finetune.py:976] (3/7) Epoch 17, batch 700, loss[loss=0.1589, simple_loss=0.2325, pruned_loss=0.04261, over 4856.00 frames. ], tot_loss[loss=0.183, simple_loss=0.2525, pruned_loss=0.05675, over 925060.67 frames. ], batch size: 31, lr: 3.41e-03, grad_scale: 64.0 2023-03-26 20:18:23,722 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.116e+02 1.553e+02 1.844e+02 2.135e+02 3.970e+02, threshold=3.688e+02, percent-clipped=1.0 2023-03-26 20:18:54,356 INFO [finetune.py:976] (3/7) Epoch 17, batch 750, loss[loss=0.2333, simple_loss=0.3047, pruned_loss=0.08096, over 4813.00 frames. ], tot_loss[loss=0.1833, simple_loss=0.2535, pruned_loss=0.05657, over 931569.01 frames. ], batch size: 39, lr: 3.41e-03, grad_scale: 64.0 2023-03-26 20:19:28,151 INFO [finetune.py:976] (3/7) Epoch 17, batch 800, loss[loss=0.1286, simple_loss=0.1923, pruned_loss=0.0325, over 4047.00 frames. ], tot_loss[loss=0.1828, simple_loss=0.2532, pruned_loss=0.05623, over 937227.95 frames. ], batch size: 17, lr: 3.41e-03, grad_scale: 64.0 2023-03-26 20:19:31,195 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.199e+02 1.753e+02 1.963e+02 2.342e+02 4.288e+02, threshold=3.926e+02, percent-clipped=2.0 2023-03-26 20:20:01,479 INFO [finetune.py:976] (3/7) Epoch 17, batch 850, loss[loss=0.1792, simple_loss=0.2456, pruned_loss=0.05644, over 4822.00 frames. ], tot_loss[loss=0.1822, simple_loss=0.2517, pruned_loss=0.05632, over 940722.53 frames. ], batch size: 40, lr: 3.41e-03, grad_scale: 64.0 2023-03-26 20:20:35,311 INFO [finetune.py:976] (3/7) Epoch 17, batch 900, loss[loss=0.1427, simple_loss=0.218, pruned_loss=0.03368, over 4765.00 frames. ], tot_loss[loss=0.1799, simple_loss=0.2488, pruned_loss=0.05546, over 943795.98 frames. ], batch size: 26, lr: 3.41e-03, grad_scale: 64.0 2023-03-26 20:20:38,304 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92547.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:20:38,821 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.037e+02 1.480e+02 1.791e+02 2.296e+02 4.324e+02, threshold=3.582e+02, percent-clipped=2.0 2023-03-26 20:21:15,097 INFO [finetune.py:976] (3/7) Epoch 17, batch 950, loss[loss=0.1972, simple_loss=0.2755, pruned_loss=0.05945, over 4824.00 frames. ], tot_loss[loss=0.1792, simple_loss=0.248, pruned_loss=0.05516, over 946907.24 frames. ], batch size: 40, lr: 3.40e-03, grad_scale: 64.0 2023-03-26 20:21:16,913 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=92595.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:21:37,241 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92611.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:21:54,765 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92625.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:22:05,420 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92637.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 20:22:12,692 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8197, 1.9822, 1.6201, 1.7127, 2.4188, 2.4523, 1.9427, 2.0173], device='cuda:3'), covar=tensor([0.0383, 0.0316, 0.0587, 0.0351, 0.0281, 0.0427, 0.0512, 0.0382], device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0107, 0.0143, 0.0112, 0.0099, 0.0107, 0.0098, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.3705e-05, 8.3141e-05, 1.1331e-04, 8.6757e-05, 7.7183e-05, 7.8927e-05, 7.3520e-05, 8.2740e-05], device='cuda:3') 2023-03-26 20:22:13,801 INFO [finetune.py:976] (3/7) Epoch 17, batch 1000, loss[loss=0.1926, simple_loss=0.2593, pruned_loss=0.06291, over 4905.00 frames. ], tot_loss[loss=0.1806, simple_loss=0.2497, pruned_loss=0.05575, over 948566.64 frames. ], batch size: 36, lr: 3.40e-03, grad_scale: 64.0 2023-03-26 20:22:20,426 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.492e+01 1.711e+02 2.074e+02 2.603e+02 6.251e+02, threshold=4.148e+02, percent-clipped=4.0 2023-03-26 20:22:27,817 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=92659.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:22:32,798 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6874, 1.3105, 0.8752, 1.6185, 2.0535, 1.4847, 1.6444, 1.6808], device='cuda:3'), covar=tensor([0.1566, 0.2128, 0.2167, 0.1329, 0.2111, 0.2137, 0.1427, 0.1944], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0095, 0.0112, 0.0093, 0.0119, 0.0095, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 20:22:45,205 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=92685.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 20:22:45,848 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92686.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:22:50,929 INFO [finetune.py:976] (3/7) Epoch 17, batch 1050, loss[loss=0.2105, simple_loss=0.2756, pruned_loss=0.0727, over 4899.00 frames. ], tot_loss[loss=0.1832, simple_loss=0.2532, pruned_loss=0.05659, over 950636.44 frames. ], batch size: 35, lr: 3.40e-03, grad_scale: 64.0 2023-03-26 20:22:51,723 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 2023-03-26 20:23:08,350 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92720.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:23:23,714 INFO [finetune.py:976] (3/7) Epoch 17, batch 1100, loss[loss=0.1962, simple_loss=0.2683, pruned_loss=0.06207, over 4778.00 frames. ], tot_loss[loss=0.1838, simple_loss=0.254, pruned_loss=0.05681, over 951737.65 frames. ], batch size: 51, lr: 3.40e-03, grad_scale: 64.0 2023-03-26 20:23:27,193 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.157e+02 1.694e+02 2.013e+02 2.338e+02 4.806e+02, threshold=4.026e+02, percent-clipped=2.0 2023-03-26 20:23:33,137 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9108, 2.6307, 3.3909, 2.1670, 2.8489, 3.1727, 2.3921, 3.2421], device='cuda:3'), covar=tensor([0.1235, 0.1812, 0.1167, 0.1832, 0.1032, 0.1236, 0.2281, 0.0797], device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0204, 0.0190, 0.0190, 0.0177, 0.0214, 0.0217, 0.0199], 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') 2023-03-26 20:23:45,406 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4094, 2.3476, 1.8438, 2.3815, 2.2557, 1.9735, 2.6375, 2.3314], device='cuda:3'), covar=tensor([0.1300, 0.2027, 0.3076, 0.2493, 0.2567, 0.1707, 0.3228, 0.1712], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0188, 0.0235, 0.0254, 0.0245, 0.0202, 0.0214, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 20:23:48,927 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92781.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:23:57,179 INFO [finetune.py:976] (3/7) Epoch 17, batch 1150, loss[loss=0.1726, simple_loss=0.2372, pruned_loss=0.05394, over 4813.00 frames. ], tot_loss[loss=0.1842, simple_loss=0.2549, pruned_loss=0.05677, over 953577.44 frames. ], batch size: 25, lr: 3.40e-03, grad_scale: 64.0 2023-03-26 20:24:14,600 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0975, 2.0477, 1.5834, 1.9292, 1.8937, 1.8470, 1.8876, 2.6650], device='cuda:3'), covar=tensor([0.4012, 0.4158, 0.3491, 0.4029, 0.4218, 0.2474, 0.3898, 0.1739], device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0259, 0.0225, 0.0275, 0.0249, 0.0216, 0.0248, 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') 2023-03-26 20:24:22,928 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92831.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:24:31,080 INFO [finetune.py:976] (3/7) Epoch 17, batch 1200, loss[loss=0.169, simple_loss=0.2373, pruned_loss=0.05039, over 4786.00 frames. ], tot_loss[loss=0.1834, simple_loss=0.2533, pruned_loss=0.05677, over 954304.81 frames. ], batch size: 51, lr: 3.40e-03, grad_scale: 64.0 2023-03-26 20:24:34,572 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.909e+01 1.547e+02 1.742e+02 2.125e+02 5.044e+02, threshold=3.483e+02, percent-clipped=2.0 2023-03-26 20:24:52,035 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-26 20:24:54,822 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3511, 1.6485, 0.8877, 2.2318, 2.5473, 1.9053, 2.2330, 2.1432], device='cuda:3'), covar=tensor([0.1318, 0.1885, 0.2004, 0.1119, 0.1716, 0.1868, 0.1082, 0.1940], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0094, 0.0110, 0.0092, 0.0118, 0.0094, 0.0098, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 20:25:04,210 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92892.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:25:04,691 INFO [finetune.py:976] (3/7) Epoch 17, batch 1250, loss[loss=0.2324, simple_loss=0.2858, pruned_loss=0.08953, over 4918.00 frames. ], tot_loss[loss=0.1813, simple_loss=0.2508, pruned_loss=0.05591, over 954303.55 frames. ], batch size: 43, lr: 3.40e-03, grad_scale: 64.0 2023-03-26 20:25:08,402 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92899.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:25:29,727 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92931.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:25:37,457 INFO [finetune.py:976] (3/7) Epoch 17, batch 1300, loss[loss=0.1863, simple_loss=0.2427, pruned_loss=0.06499, over 4908.00 frames. ], tot_loss[loss=0.1797, simple_loss=0.2483, pruned_loss=0.0555, over 954702.37 frames. ], batch size: 43, lr: 3.40e-03, grad_scale: 64.0 2023-03-26 20:25:41,346 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.030e+02 1.503e+02 1.790e+02 2.154e+02 4.064e+02, threshold=3.581e+02, percent-clipped=1.0 2023-03-26 20:25:46,882 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-26 20:25:49,695 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92960.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:25:59,933 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92975.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:26:00,590 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6197, 1.6014, 1.4067, 1.7715, 2.2011, 1.7472, 1.5843, 1.3031], device='cuda:3'), covar=tensor([0.2345, 0.2214, 0.2121, 0.1702, 0.1928, 0.1435, 0.2622, 0.2109], device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0207, 0.0210, 0.0190, 0.0240, 0.0184, 0.0214, 0.0198], 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') 2023-03-26 20:26:01,181 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5294, 1.4169, 1.3172, 1.4839, 1.8466, 1.7141, 1.4630, 1.3302], device='cuda:3'), covar=tensor([0.0319, 0.0328, 0.0609, 0.0334, 0.0203, 0.0460, 0.0344, 0.0418], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0107, 0.0143, 0.0112, 0.0098, 0.0107, 0.0098, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.3238e-05, 8.2923e-05, 1.1289e-04, 8.6591e-05, 7.6617e-05, 7.8750e-05, 7.3339e-05, 8.2664e-05], device='cuda:3') 2023-03-26 20:26:03,543 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92981.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:26:10,762 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92992.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:26:11,264 INFO [finetune.py:976] (3/7) Epoch 17, batch 1350, loss[loss=0.1728, simple_loss=0.2336, pruned_loss=0.05595, over 4899.00 frames. ], tot_loss[loss=0.1801, simple_loss=0.2484, pruned_loss=0.05587, over 954463.85 frames. ], batch size: 32, lr: 3.40e-03, grad_scale: 64.0 2023-03-26 20:26:23,864 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93010.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:26:51,543 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93036.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:26:51,553 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93036.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:27:00,928 INFO [finetune.py:976] (3/7) Epoch 17, batch 1400, loss[loss=0.1583, simple_loss=0.219, pruned_loss=0.04881, over 4231.00 frames. ], tot_loss[loss=0.1845, simple_loss=0.2535, pruned_loss=0.0578, over 955205.17 frames. ], batch size: 18, lr: 3.40e-03, grad_scale: 32.0 2023-03-26 20:27:08,967 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.085e+02 1.549e+02 1.883e+02 2.310e+02 4.523e+02, threshold=3.767e+02, percent-clipped=3.0 2023-03-26 20:27:28,094 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0904, 1.9028, 1.6623, 1.7493, 1.8313, 1.8236, 1.8051, 2.5177], device='cuda:3'), covar=tensor([0.3816, 0.4290, 0.3335, 0.3845, 0.4061, 0.2511, 0.3709, 0.1708], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0260, 0.0225, 0.0275, 0.0249, 0.0217, 0.0248, 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') 2023-03-26 20:27:33,626 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.57 vs. limit=2.0 2023-03-26 20:27:34,726 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93071.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:27:39,783 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93076.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:27:50,103 INFO [finetune.py:976] (3/7) Epoch 17, batch 1450, loss[loss=0.1974, simple_loss=0.2595, pruned_loss=0.06767, over 4916.00 frames. ], tot_loss[loss=0.1857, simple_loss=0.2554, pruned_loss=0.05799, over 954407.49 frames. ], batch size: 37, lr: 3.40e-03, grad_scale: 32.0 2023-03-26 20:27:53,137 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93097.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 20:28:05,992 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93115.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:28:23,787 INFO [finetune.py:976] (3/7) Epoch 17, batch 1500, loss[loss=0.1933, simple_loss=0.2824, pruned_loss=0.05215, over 4901.00 frames. ], tot_loss[loss=0.1853, simple_loss=0.2557, pruned_loss=0.05748, over 955172.45 frames. ], batch size: 37, lr: 3.40e-03, grad_scale: 32.0 2023-03-26 20:28:27,863 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.132e+02 1.651e+02 1.993e+02 2.270e+02 5.642e+02, threshold=3.987e+02, percent-clipped=1.0 2023-03-26 20:28:33,237 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.37 vs. limit=5.0 2023-03-26 20:28:47,128 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93176.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:28:53,710 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93187.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:28:57,281 INFO [finetune.py:976] (3/7) Epoch 17, batch 1550, loss[loss=0.1979, simple_loss=0.2686, pruned_loss=0.0636, over 4828.00 frames. ], tot_loss[loss=0.1849, simple_loss=0.2555, pruned_loss=0.05718, over 955756.70 frames. ], batch size: 39, lr: 3.40e-03, grad_scale: 32.0 2023-03-26 20:29:23,653 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.55 vs. limit=5.0 2023-03-26 20:29:30,932 INFO [finetune.py:976] (3/7) Epoch 17, batch 1600, loss[loss=0.1664, simple_loss=0.2325, pruned_loss=0.05013, over 4860.00 frames. ], tot_loss[loss=0.1823, simple_loss=0.252, pruned_loss=0.05631, over 956100.04 frames. ], batch size: 31, lr: 3.40e-03, grad_scale: 32.0 2023-03-26 20:29:34,598 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.059e+02 1.546e+02 1.807e+02 2.216e+02 3.989e+02, threshold=3.613e+02, percent-clipped=1.0 2023-03-26 20:29:38,724 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93255.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:29:49,928 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5465, 1.5925, 1.3302, 1.6425, 1.9831, 1.8016, 1.4752, 1.4302], device='cuda:3'), covar=tensor([0.0336, 0.0268, 0.0550, 0.0269, 0.0174, 0.0500, 0.0307, 0.0360], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0108, 0.0142, 0.0112, 0.0099, 0.0106, 0.0097, 0.0108], device='cuda:3'), out_proj_covar=tensor([7.3153e-05, 8.3209e-05, 1.1241e-04, 8.6471e-05, 7.6788e-05, 7.8623e-05, 7.3010e-05, 8.2614e-05], device='cuda:3') 2023-03-26 20:29:57,387 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93281.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:30:01,028 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93287.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:30:04,615 INFO [finetune.py:976] (3/7) Epoch 17, batch 1650, loss[loss=0.1669, simple_loss=0.2352, pruned_loss=0.04933, over 4902.00 frames. ], tot_loss[loss=0.1786, simple_loss=0.248, pruned_loss=0.0546, over 957866.31 frames. ], batch size: 32, lr: 3.40e-03, grad_scale: 32.0 2023-03-26 20:30:28,979 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=93329.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:30:30,701 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93331.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:30:38,309 INFO [finetune.py:976] (3/7) Epoch 17, batch 1700, loss[loss=0.2033, simple_loss=0.2566, pruned_loss=0.075, over 4848.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2463, pruned_loss=0.05413, over 958006.16 frames. ], batch size: 49, lr: 3.40e-03, grad_scale: 32.0 2023-03-26 20:30:41,942 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.003e+02 1.487e+02 1.694e+02 2.142e+02 3.933e+02, threshold=3.388e+02, percent-clipped=2.0 2023-03-26 20:30:53,822 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93366.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:31:00,377 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93376.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:31:03,935 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.5215, 3.0087, 2.7280, 1.4452, 2.8364, 2.5239, 2.4320, 2.5116], device='cuda:3'), covar=tensor([0.0933, 0.0879, 0.1815, 0.2110, 0.1749, 0.2091, 0.1845, 0.1321], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0196, 0.0200, 0.0185, 0.0214, 0.0208, 0.0225, 0.0198], 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') 2023-03-26 20:31:11,619 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93392.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 20:31:12,182 INFO [finetune.py:976] (3/7) Epoch 17, batch 1750, loss[loss=0.2488, simple_loss=0.3144, pruned_loss=0.09159, over 4028.00 frames. ], tot_loss[loss=0.1799, simple_loss=0.2491, pruned_loss=0.05535, over 955088.40 frames. ], batch size: 65, lr: 3.40e-03, grad_scale: 32.0 2023-03-26 20:31:33,284 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=93424.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:31:47,036 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6270, 1.4326, 1.9678, 1.9242, 1.6555, 3.6338, 1.3807, 1.6237], device='cuda:3'), covar=tensor([0.0962, 0.1886, 0.1085, 0.0971, 0.1602, 0.0212, 0.1581, 0.1739], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0073, 0.0078, 0.0091, 0.0080, 0.0084, 0.0078], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 20:31:48,782 INFO [finetune.py:976] (3/7) Epoch 17, batch 1800, loss[loss=0.1919, simple_loss=0.2606, pruned_loss=0.06158, over 4789.00 frames. ], tot_loss[loss=0.1821, simple_loss=0.2519, pruned_loss=0.05618, over 954182.98 frames. ], batch size: 29, lr: 3.40e-03, grad_scale: 32.0 2023-03-26 20:31:56,894 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.045e+02 1.552e+02 1.846e+02 2.179e+02 3.576e+02, threshold=3.692e+02, percent-clipped=3.0 2023-03-26 20:32:01,222 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.0090, 4.3455, 4.5070, 4.7636, 4.7343, 4.4556, 5.1252, 1.5512], device='cuda:3'), covar=tensor([0.0703, 0.0866, 0.0746, 0.0876, 0.1258, 0.1587, 0.0480, 0.5642], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0243, 0.0275, 0.0292, 0.0331, 0.0281, 0.0300, 0.0295], 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') 2023-03-26 20:32:08,257 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8594, 1.2546, 0.8540, 1.6139, 2.1952, 1.3793, 1.4573, 1.5624], device='cuda:3'), covar=tensor([0.1381, 0.2179, 0.2143, 0.1237, 0.1888, 0.2015, 0.1523, 0.1988], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0095, 0.0111, 0.0092, 0.0119, 0.0094, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 20:32:20,921 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93471.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:32:41,058 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93487.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:32:49,592 INFO [finetune.py:976] (3/7) Epoch 17, batch 1850, loss[loss=0.1519, simple_loss=0.2125, pruned_loss=0.04569, over 4722.00 frames. ], tot_loss[loss=0.1826, simple_loss=0.2524, pruned_loss=0.05636, over 954092.84 frames. ], batch size: 23, lr: 3.40e-03, grad_scale: 32.0 2023-03-26 20:33:19,855 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.0465, 3.4706, 3.6585, 3.8208, 3.7994, 3.5984, 4.1051, 1.7291], device='cuda:3'), covar=tensor([0.0719, 0.0865, 0.0818, 0.0927, 0.1212, 0.1397, 0.0717, 0.4741], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0243, 0.0275, 0.0292, 0.0333, 0.0281, 0.0301, 0.0295], 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') 2023-03-26 20:33:20,423 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=93535.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:33:26,141 INFO [finetune.py:976] (3/7) Epoch 17, batch 1900, loss[loss=0.221, simple_loss=0.2929, pruned_loss=0.07452, over 4138.00 frames. ], tot_loss[loss=0.1833, simple_loss=0.2534, pruned_loss=0.05658, over 952816.33 frames. ], batch size: 65, lr: 3.40e-03, grad_scale: 32.0 2023-03-26 20:33:30,348 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.241e+02 1.618e+02 1.925e+02 2.327e+02 3.543e+02, threshold=3.851e+02, percent-clipped=0.0 2023-03-26 20:33:34,111 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93555.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:33:55,450 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93587.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:33:59,457 INFO [finetune.py:976] (3/7) Epoch 17, batch 1950, loss[loss=0.1474, simple_loss=0.2261, pruned_loss=0.03436, over 4786.00 frames. ], tot_loss[loss=0.1825, simple_loss=0.2526, pruned_loss=0.05626, over 953010.75 frames. ], batch size: 26, lr: 3.40e-03, grad_scale: 32.0 2023-03-26 20:34:06,617 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=93603.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:34:25,027 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93631.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:34:27,389 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=93635.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:34:30,944 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7139, 1.6516, 1.5188, 1.8758, 2.0793, 1.8404, 1.3237, 1.4764], device='cuda:3'), covar=tensor([0.2031, 0.1903, 0.1871, 0.1574, 0.1524, 0.1201, 0.2455, 0.1808], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0207, 0.0211, 0.0191, 0.0240, 0.0185, 0.0214, 0.0199], 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') 2023-03-26 20:34:32,611 INFO [finetune.py:976] (3/7) Epoch 17, batch 2000, loss[loss=0.1893, simple_loss=0.2488, pruned_loss=0.06487, over 4926.00 frames. ], tot_loss[loss=0.1804, simple_loss=0.2498, pruned_loss=0.05555, over 953848.69 frames. ], batch size: 33, lr: 3.40e-03, grad_scale: 32.0 2023-03-26 20:34:37,203 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.489e+01 1.528e+02 1.753e+02 2.103e+02 5.258e+02, threshold=3.506e+02, percent-clipped=1.0 2023-03-26 20:34:48,135 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93666.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:34:50,620 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.6392, 4.0077, 4.1698, 4.4046, 4.3489, 4.0904, 4.7072, 1.5215], device='cuda:3'), covar=tensor([0.0723, 0.0853, 0.0931, 0.1069, 0.1237, 0.1536, 0.0693, 0.5854], device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0243, 0.0274, 0.0291, 0.0333, 0.0280, 0.0300, 0.0295], 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') 2023-03-26 20:34:56,476 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=93679.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:35:05,803 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93692.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:35:06,327 INFO [finetune.py:976] (3/7) Epoch 17, batch 2050, loss[loss=0.1982, simple_loss=0.2569, pruned_loss=0.0698, over 4229.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2461, pruned_loss=0.05454, over 953805.05 frames. ], batch size: 65, lr: 3.40e-03, grad_scale: 32.0 2023-03-26 20:35:08,297 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.83 vs. limit=5.0 2023-03-26 20:35:20,564 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=93714.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:35:29,720 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-26 20:35:37,804 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=93740.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:35:39,562 INFO [finetune.py:976] (3/7) Epoch 17, batch 2100, loss[loss=0.1623, simple_loss=0.2353, pruned_loss=0.04464, over 4901.00 frames. ], tot_loss[loss=0.1785, simple_loss=0.2471, pruned_loss=0.05499, over 955884.01 frames. ], batch size: 36, lr: 3.40e-03, grad_scale: 32.0 2023-03-26 20:35:43,619 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.402e+01 1.568e+02 1.860e+02 2.232e+02 5.340e+02, threshold=3.720e+02, percent-clipped=4.0 2023-03-26 20:35:58,007 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93770.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:35:58,598 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93771.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:36:12,212 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1485, 3.5303, 3.7375, 3.9806, 3.8945, 3.6270, 4.2350, 1.3042], device='cuda:3'), covar=tensor([0.0901, 0.0982, 0.0866, 0.1132, 0.1452, 0.1734, 0.0783, 0.5773], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0244, 0.0275, 0.0292, 0.0334, 0.0282, 0.0301, 0.0296], 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') 2023-03-26 20:36:13,275 INFO [finetune.py:976] (3/7) Epoch 17, batch 2150, loss[loss=0.1961, simple_loss=0.2575, pruned_loss=0.06731, over 4896.00 frames. ], tot_loss[loss=0.183, simple_loss=0.2517, pruned_loss=0.05708, over 956251.55 frames. ], batch size: 32, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:36:24,179 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-26 20:36:26,360 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6704, 1.5573, 2.2428, 3.2990, 2.3760, 2.4950, 1.0747, 2.6908], device='cuda:3'), covar=tensor([0.1778, 0.1427, 0.1186, 0.0628, 0.0769, 0.1137, 0.1814, 0.0559], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0116, 0.0134, 0.0165, 0.0101, 0.0137, 0.0125, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 20:36:31,171 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=93819.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:36:38,624 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93831.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:36:47,327 INFO [finetune.py:976] (3/7) Epoch 17, batch 2200, loss[loss=0.2322, simple_loss=0.2803, pruned_loss=0.09203, over 4194.00 frames. ], tot_loss[loss=0.1847, simple_loss=0.2537, pruned_loss=0.05784, over 953760.62 frames. ], batch size: 66, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:36:51,477 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.198e+02 1.569e+02 1.869e+02 2.308e+02 4.137e+02, threshold=3.738e+02, percent-clipped=1.0 2023-03-26 20:37:36,156 INFO [finetune.py:976] (3/7) Epoch 17, batch 2250, loss[loss=0.1987, simple_loss=0.2642, pruned_loss=0.06657, over 4876.00 frames. ], tot_loss[loss=0.1851, simple_loss=0.2543, pruned_loss=0.05796, over 952521.48 frames. ], batch size: 35, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:38:03,965 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.32 vs. limit=5.0 2023-03-26 20:38:30,051 INFO [finetune.py:976] (3/7) Epoch 17, batch 2300, loss[loss=0.163, simple_loss=0.233, pruned_loss=0.04647, over 4850.00 frames. ], tot_loss[loss=0.1845, simple_loss=0.2544, pruned_loss=0.05728, over 953377.06 frames. ], batch size: 44, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:38:34,185 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.189e+02 1.601e+02 1.890e+02 2.328e+02 3.292e+02, threshold=3.781e+02, percent-clipped=0.0 2023-03-26 20:38:50,094 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.9479, 4.3179, 4.5056, 4.8115, 4.6918, 4.3253, 5.0580, 1.5429], device='cuda:3'), covar=tensor([0.0658, 0.0742, 0.0746, 0.0811, 0.1123, 0.1563, 0.0485, 0.5585], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0244, 0.0275, 0.0292, 0.0334, 0.0282, 0.0301, 0.0296], 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') 2023-03-26 20:38:56,114 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93981.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:39:03,825 INFO [finetune.py:976] (3/7) Epoch 17, batch 2350, loss[loss=0.1717, simple_loss=0.2393, pruned_loss=0.05203, over 4825.00 frames. ], tot_loss[loss=0.183, simple_loss=0.2523, pruned_loss=0.05688, over 952751.91 frames. ], batch size: 39, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:39:06,426 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1720, 2.1123, 1.7891, 2.0337, 1.9957, 1.9395, 1.9878, 2.7223], device='cuda:3'), covar=tensor([0.3698, 0.4306, 0.3160, 0.4027, 0.4214, 0.2367, 0.3986, 0.1610], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0262, 0.0228, 0.0277, 0.0251, 0.0219, 0.0250, 0.0232], 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') 2023-03-26 20:39:37,901 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94042.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:39:38,384 INFO [finetune.py:976] (3/7) Epoch 17, batch 2400, loss[loss=0.1835, simple_loss=0.2501, pruned_loss=0.05849, over 4832.00 frames. ], tot_loss[loss=0.1808, simple_loss=0.2496, pruned_loss=0.05598, over 953956.04 frames. ], batch size: 47, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:39:42,507 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.024e+02 1.475e+02 1.781e+02 2.110e+02 4.538e+02, threshold=3.563e+02, percent-clipped=2.0 2023-03-26 20:39:51,960 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7232, 2.4208, 2.3840, 1.4441, 2.5641, 2.0980, 1.9757, 2.2844], device='cuda:3'), covar=tensor([0.1002, 0.0728, 0.1544, 0.1814, 0.1430, 0.1600, 0.1840, 0.0941], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0195, 0.0198, 0.0183, 0.0212, 0.0206, 0.0223, 0.0197], 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') 2023-03-26 20:40:11,638 INFO [finetune.py:976] (3/7) Epoch 17, batch 2450, loss[loss=0.1589, simple_loss=0.2309, pruned_loss=0.04346, over 4764.00 frames. ], tot_loss[loss=0.1786, simple_loss=0.2471, pruned_loss=0.05508, over 952993.18 frames. ], batch size: 26, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:40:34,651 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94126.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:40:45,406 INFO [finetune.py:976] (3/7) Epoch 17, batch 2500, loss[loss=0.2182, simple_loss=0.288, pruned_loss=0.07417, over 4126.00 frames. ], tot_loss[loss=0.1798, simple_loss=0.2484, pruned_loss=0.05563, over 952889.70 frames. ], batch size: 65, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:40:49,552 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.530e+01 1.683e+02 1.909e+02 2.220e+02 4.342e+02, threshold=3.819e+02, percent-clipped=2.0 2023-03-26 20:40:53,993 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.18 vs. limit=5.0 2023-03-26 20:41:12,821 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-26 20:41:18,603 INFO [finetune.py:976] (3/7) Epoch 17, batch 2550, loss[loss=0.2156, simple_loss=0.2692, pruned_loss=0.08102, over 4227.00 frames. ], tot_loss[loss=0.1824, simple_loss=0.2516, pruned_loss=0.05665, over 949486.01 frames. ], batch size: 18, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:41:38,775 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94223.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 20:41:44,968 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-26 20:41:52,380 INFO [finetune.py:976] (3/7) Epoch 17, batch 2600, loss[loss=0.2381, simple_loss=0.3115, pruned_loss=0.08232, over 4781.00 frames. ], tot_loss[loss=0.183, simple_loss=0.2522, pruned_loss=0.05684, over 949377.71 frames. ], batch size: 51, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:41:56,018 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.038e+02 1.647e+02 1.955e+02 2.265e+02 3.573e+02, threshold=3.911e+02, percent-clipped=0.0 2023-03-26 20:42:04,802 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-26 20:42:09,670 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5889, 1.4571, 2.2391, 3.3104, 2.2963, 2.3661, 1.2799, 2.7084], device='cuda:3'), covar=tensor([0.1859, 0.1546, 0.1211, 0.0588, 0.0829, 0.1785, 0.1668, 0.0531], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0117, 0.0135, 0.0166, 0.0102, 0.0138, 0.0126, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 20:42:09,961 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.72 vs. limit=5.0 2023-03-26 20:42:17,294 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7829, 1.1102, 1.8109, 1.7330, 1.5709, 1.5225, 1.6390, 1.6774], device='cuda:3'), covar=tensor([0.3852, 0.4019, 0.3024, 0.3403, 0.4408, 0.3599, 0.4128, 0.3007], device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0239, 0.0258, 0.0272, 0.0270, 0.0246, 0.0283, 0.0239], 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') 2023-03-26 20:42:19,687 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94284.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 20:42:25,432 INFO [finetune.py:976] (3/7) Epoch 17, batch 2650, loss[loss=0.1493, simple_loss=0.2227, pruned_loss=0.0379, over 4791.00 frames. ], tot_loss[loss=0.1833, simple_loss=0.2527, pruned_loss=0.05691, over 950458.21 frames. ], batch size: 25, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:43:12,616 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94337.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:43:20,896 INFO [finetune.py:976] (3/7) Epoch 17, batch 2700, loss[loss=0.1474, simple_loss=0.224, pruned_loss=0.03539, over 4781.00 frames. ], tot_loss[loss=0.1819, simple_loss=0.2513, pruned_loss=0.05623, over 950769.65 frames. ], batch size: 29, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:43:24,084 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94348.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:43:28,204 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.025e+02 1.514e+02 1.766e+02 2.145e+02 4.618e+02, threshold=3.532e+02, percent-clipped=2.0 2023-03-26 20:44:10,179 INFO [finetune.py:976] (3/7) Epoch 17, batch 2750, loss[loss=0.1787, simple_loss=0.2376, pruned_loss=0.05991, over 4833.00 frames. ], tot_loss[loss=0.1785, simple_loss=0.2479, pruned_loss=0.05451, over 949583.70 frames. ], batch size: 33, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:44:20,189 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94409.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 20:44:22,492 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5827, 1.8056, 2.2741, 1.9118, 1.8717, 4.1440, 1.4681, 1.9744], device='cuda:3'), covar=tensor([0.0968, 0.1643, 0.1083, 0.0959, 0.1445, 0.0179, 0.1484, 0.1583], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0074, 0.0078, 0.0091, 0.0080, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 20:44:31,942 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94426.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:44:43,082 INFO [finetune.py:976] (3/7) Epoch 17, batch 2800, loss[loss=0.1738, simple_loss=0.2417, pruned_loss=0.05296, over 4907.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2458, pruned_loss=0.05367, over 950765.35 frames. ], batch size: 35, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:44:45,424 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.81 vs. limit=5.0 2023-03-26 20:44:47,185 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.577e+01 1.587e+02 1.887e+02 2.313e+02 4.372e+02, threshold=3.775e+02, percent-clipped=5.0 2023-03-26 20:44:49,107 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.2890, 2.9525, 3.0333, 3.0588, 2.9428, 2.7388, 3.2570, 1.0779], device='cuda:3'), covar=tensor([0.1375, 0.1621, 0.1857, 0.1841, 0.2143, 0.2594, 0.1517, 0.6904], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0244, 0.0277, 0.0293, 0.0335, 0.0281, 0.0302, 0.0296], 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') 2023-03-26 20:45:02,860 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=94474.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:45:11,382 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-03-26 20:45:11,791 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2920, 2.1959, 1.8001, 2.1013, 2.2325, 1.9331, 2.4668, 2.2640], device='cuda:3'), covar=tensor([0.1289, 0.2072, 0.3066, 0.2525, 0.2506, 0.1615, 0.3131, 0.1825], device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0189, 0.0235, 0.0254, 0.0246, 0.0203, 0.0214, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 20:45:16,199 INFO [finetune.py:976] (3/7) Epoch 17, batch 2850, loss[loss=0.185, simple_loss=0.2535, pruned_loss=0.05822, over 4223.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2445, pruned_loss=0.05352, over 947893.57 frames. ], batch size: 65, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:45:21,455 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-26 20:45:42,129 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7901, 1.3761, 0.8559, 1.6002, 2.3142, 1.1219, 1.5101, 1.5818], device='cuda:3'), covar=tensor([0.1379, 0.2057, 0.1891, 0.1225, 0.1712, 0.1874, 0.1517, 0.1983], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0095, 0.0110, 0.0092, 0.0119, 0.0094, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 20:45:49,604 INFO [finetune.py:976] (3/7) Epoch 17, batch 2900, loss[loss=0.174, simple_loss=0.2536, pruned_loss=0.04719, over 4841.00 frames. ], tot_loss[loss=0.1792, simple_loss=0.2482, pruned_loss=0.05511, over 949825.37 frames. ], batch size: 49, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:45:53,202 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 1.550e+02 1.801e+02 2.117e+02 3.911e+02, threshold=3.601e+02, percent-clipped=1.0 2023-03-26 20:46:12,367 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94579.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 20:46:22,371 INFO [finetune.py:976] (3/7) Epoch 17, batch 2950, loss[loss=0.2093, simple_loss=0.2815, pruned_loss=0.06858, over 4828.00 frames. ], tot_loss[loss=0.1813, simple_loss=0.2507, pruned_loss=0.05596, over 951148.54 frames. ], batch size: 51, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:46:46,964 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4650, 1.0633, 0.7987, 1.2341, 1.9699, 0.7199, 1.1559, 1.2885], device='cuda:3'), covar=tensor([0.1994, 0.2992, 0.2163, 0.1803, 0.2219, 0.2679, 0.2074, 0.2788], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0094, 0.0110, 0.0092, 0.0118, 0.0094, 0.0097, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 20:46:52,122 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94637.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:46:56,158 INFO [finetune.py:976] (3/7) Epoch 17, batch 3000, loss[loss=0.1681, simple_loss=0.2494, pruned_loss=0.04339, over 4841.00 frames. ], tot_loss[loss=0.1833, simple_loss=0.2528, pruned_loss=0.05688, over 952192.81 frames. ], batch size: 44, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:46:56,158 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 20:47:06,772 INFO [finetune.py:1010] (3/7) Epoch 17, validation: loss=0.1562, simple_loss=0.2257, pruned_loss=0.04335, over 2265189.00 frames. 2023-03-26 20:47:06,773 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6434MB 2023-03-26 20:47:10,421 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.006e+02 1.605e+02 1.916e+02 2.337e+02 3.800e+02, threshold=3.832e+02, percent-clipped=2.0 2023-03-26 20:47:28,354 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5372, 3.7820, 3.5673, 1.6848, 3.8414, 2.8788, 0.6624, 2.5538], device='cuda:3'), covar=tensor([0.2431, 0.2096, 0.1583, 0.3451, 0.0942, 0.0981, 0.4745, 0.1457], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0176, 0.0160, 0.0129, 0.0159, 0.0124, 0.0147, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 20:47:33,745 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=94685.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:47:38,998 INFO [finetune.py:976] (3/7) Epoch 17, batch 3050, loss[loss=0.1553, simple_loss=0.2394, pruned_loss=0.03561, over 4905.00 frames. ], tot_loss[loss=0.1846, simple_loss=0.2542, pruned_loss=0.05751, over 950221.25 frames. ], batch size: 37, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:47:47,350 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94704.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 20:48:19,138 INFO [finetune.py:976] (3/7) Epoch 17, batch 3100, loss[loss=0.2331, simple_loss=0.2858, pruned_loss=0.09018, over 4167.00 frames. ], tot_loss[loss=0.1842, simple_loss=0.2535, pruned_loss=0.05748, over 947282.66 frames. ], batch size: 65, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:48:27,689 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.157e+02 1.533e+02 1.793e+02 2.269e+02 8.706e+02, threshold=3.585e+02, percent-clipped=3.0 2023-03-26 20:48:38,635 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0911, 1.9540, 1.7271, 1.8031, 1.8865, 1.8777, 1.8890, 2.5101], device='cuda:3'), covar=tensor([0.3979, 0.4409, 0.3544, 0.3811, 0.3695, 0.2699, 0.3862, 0.1926], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0261, 0.0227, 0.0276, 0.0250, 0.0219, 0.0250, 0.0230], 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') 2023-03-26 20:49:07,477 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7360, 1.4415, 0.7462, 1.7151, 2.1644, 1.3481, 1.5172, 1.7092], device='cuda:3'), covar=tensor([0.1473, 0.1869, 0.2022, 0.1138, 0.1820, 0.1966, 0.1365, 0.1855], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0095, 0.0110, 0.0092, 0.0119, 0.0094, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 20:49:17,517 INFO [finetune.py:976] (3/7) Epoch 17, batch 3150, loss[loss=0.2205, simple_loss=0.2747, pruned_loss=0.08317, over 4858.00 frames. ], tot_loss[loss=0.182, simple_loss=0.2507, pruned_loss=0.05667, over 948922.18 frames. ], batch size: 47, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:49:40,142 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3828, 2.3939, 1.8547, 2.4678, 2.3514, 2.0012, 2.8685, 2.4669], device='cuda:3'), covar=tensor([0.1457, 0.2258, 0.3122, 0.2800, 0.2751, 0.1772, 0.3000, 0.1855], device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0189, 0.0234, 0.0254, 0.0245, 0.0203, 0.0213, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 20:49:50,329 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6911, 2.5783, 2.0599, 2.7822, 2.5903, 2.2103, 3.1690, 2.7296], device='cuda:3'), covar=tensor([0.1314, 0.2111, 0.3094, 0.2552, 0.2524, 0.1582, 0.2582, 0.1792], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0188, 0.0234, 0.0253, 0.0245, 0.0203, 0.0212, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 20:49:51,404 INFO [finetune.py:976] (3/7) Epoch 17, batch 3200, loss[loss=0.1963, simple_loss=0.2569, pruned_loss=0.06789, over 4121.00 frames. ], tot_loss[loss=0.1797, simple_loss=0.248, pruned_loss=0.05564, over 950221.94 frames. ], batch size: 65, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:49:54,426 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5836, 1.6040, 2.1931, 1.8650, 1.7237, 3.9101, 1.4604, 1.7046], device='cuda:3'), covar=tensor([0.0978, 0.1697, 0.1293, 0.0948, 0.1541, 0.0225, 0.1478, 0.1695], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0074, 0.0078, 0.0091, 0.0081, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 20:49:55,533 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.020e+02 1.463e+02 1.766e+02 2.027e+02 4.168e+02, threshold=3.532e+02, percent-clipped=1.0 2023-03-26 20:50:12,643 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1822, 3.6595, 3.7979, 4.0259, 3.9404, 3.6767, 4.2669, 1.3237], device='cuda:3'), covar=tensor([0.0797, 0.0881, 0.0898, 0.1010, 0.1289, 0.1745, 0.0748, 0.6085], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0245, 0.0277, 0.0293, 0.0336, 0.0281, 0.0302, 0.0296], 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') 2023-03-26 20:50:16,329 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94879.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 20:50:25,246 INFO [finetune.py:976] (3/7) Epoch 17, batch 3250, loss[loss=0.2282, simple_loss=0.3022, pruned_loss=0.07712, over 4825.00 frames. ], tot_loss[loss=0.1793, simple_loss=0.2478, pruned_loss=0.05537, over 953120.28 frames. ], batch size: 39, lr: 3.39e-03, grad_scale: 32.0 2023-03-26 20:50:48,424 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=94927.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 20:50:58,674 INFO [finetune.py:976] (3/7) Epoch 17, batch 3300, loss[loss=0.1968, simple_loss=0.2641, pruned_loss=0.06477, over 4799.00 frames. ], tot_loss[loss=0.1818, simple_loss=0.2508, pruned_loss=0.05639, over 954580.47 frames. ], batch size: 29, lr: 3.38e-03, grad_scale: 32.0 2023-03-26 20:51:02,382 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.215e+02 1.785e+02 2.188e+02 2.532e+02 5.228e+02, threshold=4.375e+02, percent-clipped=4.0 2023-03-26 20:51:31,809 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-03-26 20:51:32,709 INFO [finetune.py:976] (3/7) Epoch 17, batch 3350, loss[loss=0.2122, simple_loss=0.2828, pruned_loss=0.07081, over 4816.00 frames. ], tot_loss[loss=0.1846, simple_loss=0.2542, pruned_loss=0.05754, over 956166.64 frames. ], batch size: 39, lr: 3.38e-03, grad_scale: 32.0 2023-03-26 20:51:40,180 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95004.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:52:06,488 INFO [finetune.py:976] (3/7) Epoch 17, batch 3400, loss[loss=0.2374, simple_loss=0.2928, pruned_loss=0.091, over 4104.00 frames. ], tot_loss[loss=0.186, simple_loss=0.2555, pruned_loss=0.0582, over 956934.57 frames. ], batch size: 65, lr: 3.38e-03, grad_scale: 64.0 2023-03-26 20:52:10,132 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.060e+02 1.553e+02 1.863e+02 2.086e+02 3.757e+02, threshold=3.727e+02, percent-clipped=0.0 2023-03-26 20:52:12,046 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=95052.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:52:40,292 INFO [finetune.py:976] (3/7) Epoch 17, batch 3450, loss[loss=0.1624, simple_loss=0.2348, pruned_loss=0.04503, over 4845.00 frames. ], tot_loss[loss=0.1855, simple_loss=0.255, pruned_loss=0.05801, over 955895.82 frames. ], batch size: 49, lr: 3.38e-03, grad_scale: 64.0 2023-03-26 20:52:47,691 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95105.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 20:53:11,424 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0044, 1.8947, 1.8474, 2.1876, 2.4376, 2.0795, 2.1751, 1.5945], device='cuda:3'), covar=tensor([0.2070, 0.1973, 0.1700, 0.1414, 0.1901, 0.1270, 0.1944, 0.1725], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0209, 0.0212, 0.0191, 0.0241, 0.0186, 0.0215, 0.0200], 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') 2023-03-26 20:53:13,091 INFO [finetune.py:976] (3/7) Epoch 17, batch 3500, loss[loss=0.1603, simple_loss=0.2333, pruned_loss=0.04364, over 4794.00 frames. ], tot_loss[loss=0.1832, simple_loss=0.2524, pruned_loss=0.05702, over 955688.25 frames. ], batch size: 29, lr: 3.38e-03, grad_scale: 64.0 2023-03-26 20:53:15,726 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.72 vs. limit=5.0 2023-03-26 20:53:17,181 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.085e+02 1.619e+02 1.940e+02 2.279e+02 3.817e+02, threshold=3.880e+02, percent-clipped=1.0 2023-03-26 20:53:35,217 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95166.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 20:54:05,087 INFO [finetune.py:976] (3/7) Epoch 17, batch 3550, loss[loss=0.1902, simple_loss=0.247, pruned_loss=0.06674, over 4868.00 frames. ], tot_loss[loss=0.1808, simple_loss=0.2495, pruned_loss=0.05606, over 955582.25 frames. ], batch size: 34, lr: 3.38e-03, grad_scale: 64.0 2023-03-26 20:54:51,060 INFO [finetune.py:976] (3/7) Epoch 17, batch 3600, loss[loss=0.1903, simple_loss=0.2759, pruned_loss=0.05231, over 4851.00 frames. ], tot_loss[loss=0.1793, simple_loss=0.2472, pruned_loss=0.05567, over 954953.92 frames. ], batch size: 49, lr: 3.38e-03, grad_scale: 64.0 2023-03-26 20:54:54,645 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.130e+02 1.580e+02 1.871e+02 2.182e+02 4.206e+02, threshold=3.742e+02, percent-clipped=1.0 2023-03-26 20:55:02,009 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95260.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:55:06,895 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95268.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:55:12,160 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5969, 2.1809, 0.9351, 2.5775, 2.8489, 2.3554, 2.5615, 2.4424], device='cuda:3'), covar=tensor([0.1194, 0.1759, 0.1951, 0.0987, 0.1544, 0.1481, 0.1175, 0.1737], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0095, 0.0110, 0.0092, 0.0119, 0.0095, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 20:55:24,757 INFO [finetune.py:976] (3/7) Epoch 17, batch 3650, loss[loss=0.2047, simple_loss=0.2795, pruned_loss=0.065, over 4820.00 frames. ], tot_loss[loss=0.1806, simple_loss=0.2489, pruned_loss=0.05612, over 954170.12 frames. ], batch size: 38, lr: 3.38e-03, grad_scale: 64.0 2023-03-26 20:55:26,696 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5387, 1.1128, 0.8572, 1.4427, 1.9374, 1.0338, 1.3134, 1.5137], device='cuda:3'), covar=tensor([0.1570, 0.2191, 0.1815, 0.1220, 0.2053, 0.2020, 0.1477, 0.1898], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0096, 0.0110, 0.0092, 0.0119, 0.0095, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 20:55:42,966 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95321.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:55:48,242 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95329.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:55:55,822 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95339.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:55:58,602 INFO [finetune.py:976] (3/7) Epoch 17, batch 3700, loss[loss=0.1986, simple_loss=0.2728, pruned_loss=0.06216, over 4824.00 frames. ], tot_loss[loss=0.1845, simple_loss=0.254, pruned_loss=0.05743, over 954140.39 frames. ], batch size: 49, lr: 3.38e-03, grad_scale: 64.0 2023-03-26 20:56:02,229 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.130e+02 1.605e+02 1.907e+02 2.409e+02 3.957e+02, threshold=3.813e+02, percent-clipped=4.0 2023-03-26 20:56:31,739 INFO [finetune.py:976] (3/7) Epoch 17, batch 3750, loss[loss=0.1617, simple_loss=0.2336, pruned_loss=0.04488, over 4821.00 frames. ], tot_loss[loss=0.1861, simple_loss=0.2554, pruned_loss=0.0584, over 953133.19 frames. ], batch size: 33, lr: 3.38e-03, grad_scale: 64.0 2023-03-26 20:56:36,733 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95400.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:56:43,985 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6571, 1.6047, 2.1256, 3.1933, 2.2502, 2.2954, 1.1818, 2.7200], device='cuda:3'), covar=tensor([0.1717, 0.1306, 0.1186, 0.0554, 0.0760, 0.1224, 0.1597, 0.0533], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0133, 0.0165, 0.0101, 0.0135, 0.0124, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 20:56:46,898 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6595, 1.1366, 0.8581, 1.5068, 2.0378, 1.3071, 1.4290, 1.6286], device='cuda:3'), covar=tensor([0.1462, 0.2238, 0.1947, 0.1239, 0.1967, 0.1936, 0.1512, 0.1912], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0096, 0.0111, 0.0093, 0.0119, 0.0095, 0.0099, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 20:56:52,989 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.3870, 3.7875, 3.9996, 4.2062, 4.1607, 3.8652, 4.4441, 1.4481], device='cuda:3'), covar=tensor([0.0783, 0.0905, 0.0825, 0.0997, 0.1220, 0.1599, 0.0765, 0.5414], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0246, 0.0276, 0.0292, 0.0335, 0.0281, 0.0302, 0.0295], 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') 2023-03-26 20:57:04,529 INFO [finetune.py:976] (3/7) Epoch 17, batch 3800, loss[loss=0.1575, simple_loss=0.2274, pruned_loss=0.04377, over 4746.00 frames. ], tot_loss[loss=0.1848, simple_loss=0.2548, pruned_loss=0.05738, over 954948.96 frames. ], batch size: 27, lr: 3.38e-03, grad_scale: 64.0 2023-03-26 20:57:09,550 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.108e+02 1.488e+02 1.737e+02 2.235e+02 4.648e+02, threshold=3.475e+02, percent-clipped=3.0 2023-03-26 20:57:16,912 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95461.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 20:57:17,582 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.1749, 1.3026, 1.4567, 0.7688, 1.3837, 1.6498, 1.5918, 1.3381], device='cuda:3'), covar=tensor([0.0884, 0.0682, 0.0527, 0.0574, 0.0535, 0.0622, 0.0397, 0.0674], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0152, 0.0124, 0.0128, 0.0132, 0.0130, 0.0144, 0.0149], device='cuda:3'), out_proj_covar=tensor([9.2580e-05, 1.1050e-04, 8.8937e-05, 9.1042e-05, 9.2894e-05, 9.3223e-05, 1.0400e-04, 1.0732e-04], device='cuda:3') 2023-03-26 20:57:37,562 INFO [finetune.py:976] (3/7) Epoch 17, batch 3850, loss[loss=0.1853, simple_loss=0.2524, pruned_loss=0.0591, over 4809.00 frames. ], tot_loss[loss=0.1834, simple_loss=0.2537, pruned_loss=0.05661, over 955886.86 frames. ], batch size: 30, lr: 3.38e-03, grad_scale: 64.0 2023-03-26 20:57:45,781 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2689, 2.1800, 1.9454, 2.2516, 2.0934, 2.0674, 2.0857, 2.9887], device='cuda:3'), covar=tensor([0.3874, 0.5335, 0.3418, 0.4549, 0.4884, 0.2389, 0.4714, 0.1800], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0261, 0.0227, 0.0276, 0.0250, 0.0219, 0.0250, 0.0231], 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') 2023-03-26 20:58:10,768 INFO [finetune.py:976] (3/7) Epoch 17, batch 3900, loss[loss=0.1617, simple_loss=0.2271, pruned_loss=0.04817, over 4807.00 frames. ], tot_loss[loss=0.1819, simple_loss=0.2514, pruned_loss=0.05624, over 956658.47 frames. ], batch size: 25, lr: 3.38e-03, grad_scale: 64.0 2023-03-26 20:58:15,380 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.225e+02 1.550e+02 1.834e+02 2.229e+02 4.290e+02, threshold=3.669e+02, percent-clipped=3.0 2023-03-26 20:58:29,132 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0701, 1.8340, 2.4776, 3.7879, 2.7474, 2.5824, 1.0583, 3.1223], device='cuda:3'), covar=tensor([0.1578, 0.1341, 0.1250, 0.0601, 0.0718, 0.1864, 0.1747, 0.0439], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0116, 0.0134, 0.0165, 0.0101, 0.0136, 0.0124, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 20:58:44,611 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.1038, 4.4411, 4.6501, 4.9019, 4.8133, 4.4903, 5.2015, 1.6200], device='cuda:3'), covar=tensor([0.0644, 0.0767, 0.0722, 0.0822, 0.1095, 0.1585, 0.0482, 0.5665], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0246, 0.0277, 0.0292, 0.0335, 0.0282, 0.0303, 0.0296], 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') 2023-03-26 20:58:46,338 INFO [finetune.py:976] (3/7) Epoch 17, batch 3950, loss[loss=0.1536, simple_loss=0.2303, pruned_loss=0.03842, over 4895.00 frames. ], tot_loss[loss=0.1788, simple_loss=0.2479, pruned_loss=0.05481, over 956643.58 frames. ], batch size: 32, lr: 3.38e-03, grad_scale: 32.0 2023-03-26 20:58:47,142 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.03 vs. limit=5.0 2023-03-26 20:58:47,762 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.88 vs. limit=5.0 2023-03-26 20:59:04,368 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95616.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:59:05,019 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95617.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:59:13,756 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95624.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 20:59:34,008 INFO [finetune.py:976] (3/7) Epoch 17, batch 4000, loss[loss=0.1207, simple_loss=0.18, pruned_loss=0.03068, over 3837.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2454, pruned_loss=0.0537, over 955169.64 frames. ], batch size: 16, lr: 3.38e-03, grad_scale: 32.0 2023-03-26 20:59:42,363 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.075e+02 1.540e+02 1.979e+02 2.285e+02 3.877e+02, threshold=3.958e+02, percent-clipped=2.0 2023-03-26 20:59:54,337 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6583, 1.5852, 1.3623, 1.7641, 2.0366, 1.8035, 1.2969, 1.3589], device='cuda:3'), covar=tensor([0.2182, 0.2031, 0.2006, 0.1619, 0.1613, 0.1243, 0.2492, 0.1917], device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0207, 0.0210, 0.0190, 0.0239, 0.0185, 0.0213, 0.0199], 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') 2023-03-26 21:00:04,451 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.76 vs. limit=5.0 2023-03-26 21:00:12,653 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95678.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:00:25,668 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.59 vs. limit=5.0 2023-03-26 21:00:26,083 INFO [finetune.py:976] (3/7) Epoch 17, batch 4050, loss[loss=0.1892, simple_loss=0.2621, pruned_loss=0.05812, over 4932.00 frames. ], tot_loss[loss=0.1788, simple_loss=0.2478, pruned_loss=0.0549, over 954711.76 frames. ], batch size: 38, lr: 3.38e-03, grad_scale: 32.0 2023-03-26 21:00:27,353 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95695.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:00:39,053 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6994, 1.5646, 1.8959, 1.1530, 1.6462, 1.7644, 1.4781, 1.9709], device='cuda:3'), covar=tensor([0.1056, 0.1928, 0.1260, 0.1740, 0.0833, 0.1310, 0.2559, 0.0704], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0204, 0.0189, 0.0190, 0.0176, 0.0212, 0.0216, 0.0199], 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') 2023-03-26 21:00:59,870 INFO [finetune.py:976] (3/7) Epoch 17, batch 4100, loss[loss=0.2152, simple_loss=0.2838, pruned_loss=0.0733, over 4821.00 frames. ], tot_loss[loss=0.181, simple_loss=0.2503, pruned_loss=0.05586, over 952979.93 frames. ], batch size: 40, lr: 3.38e-03, grad_scale: 32.0 2023-03-26 21:01:04,067 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.117e+02 1.600e+02 1.864e+02 2.304e+02 4.240e+02, threshold=3.729e+02, percent-clipped=2.0 2023-03-26 21:01:10,347 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-26 21:01:12,336 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95761.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 21:01:17,611 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7939, 1.2618, 1.8725, 1.8065, 1.6268, 1.5442, 1.7726, 1.6970], device='cuda:3'), covar=tensor([0.3812, 0.3851, 0.3019, 0.3568, 0.4520, 0.3645, 0.4049, 0.3031], device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0238, 0.0257, 0.0272, 0.0270, 0.0246, 0.0281, 0.0239], 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') 2023-03-26 21:01:33,031 INFO [finetune.py:976] (3/7) Epoch 17, batch 4150, loss[loss=0.2005, simple_loss=0.2765, pruned_loss=0.06225, over 4812.00 frames. ], tot_loss[loss=0.1831, simple_loss=0.2521, pruned_loss=0.05706, over 951055.23 frames. ], batch size: 39, lr: 3.38e-03, grad_scale: 32.0 2023-03-26 21:01:33,992 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-03-26 21:01:44,397 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=95809.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 21:01:46,204 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95812.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 21:01:54,853 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2745, 1.8009, 2.1520, 2.2779, 1.9419, 1.9362, 2.1671, 2.0146], device='cuda:3'), covar=tensor([0.3904, 0.4069, 0.3357, 0.3750, 0.5011, 0.3748, 0.5178, 0.3127], device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0238, 0.0257, 0.0272, 0.0270, 0.0245, 0.0280, 0.0238], 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') 2023-03-26 21:02:06,757 INFO [finetune.py:976] (3/7) Epoch 17, batch 4200, loss[loss=0.1647, simple_loss=0.2432, pruned_loss=0.04303, over 4842.00 frames. ], tot_loss[loss=0.1845, simple_loss=0.2541, pruned_loss=0.05743, over 952685.79 frames. ], batch size: 49, lr: 3.38e-03, grad_scale: 32.0 2023-03-26 21:02:09,313 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95847.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:02:11,508 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.143e+02 1.538e+02 1.932e+02 2.354e+02 8.206e+02, threshold=3.863e+02, percent-clipped=2.0 2023-03-26 21:02:25,593 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8424, 2.0028, 2.3585, 2.0665, 2.1354, 4.5068, 2.0305, 2.1186], device='cuda:3'), covar=tensor([0.1044, 0.1774, 0.1190, 0.1094, 0.1571, 0.0358, 0.1479, 0.1716], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0074, 0.0078, 0.0092, 0.0081, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 21:02:27,896 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95873.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 21:02:37,582 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5488, 1.4013, 1.5075, 0.7522, 1.5667, 1.4537, 1.4958, 1.3586], device='cuda:3'), covar=tensor([0.0562, 0.0790, 0.0691, 0.1010, 0.0782, 0.0812, 0.0697, 0.1200], device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0136, 0.0142, 0.0124, 0.0125, 0.0142, 0.0143, 0.0165], 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') 2023-03-26 21:02:38,242 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9920, 1.8454, 1.5447, 1.6545, 1.7329, 1.6882, 1.7678, 2.4750], device='cuda:3'), covar=tensor([0.3782, 0.3855, 0.3094, 0.3587, 0.3624, 0.2411, 0.3440, 0.1583], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0261, 0.0226, 0.0276, 0.0251, 0.0218, 0.0251, 0.0230], 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') 2023-03-26 21:02:39,921 INFO [finetune.py:976] (3/7) Epoch 17, batch 4250, loss[loss=0.1835, simple_loss=0.247, pruned_loss=0.06001, over 4811.00 frames. ], tot_loss[loss=0.1824, simple_loss=0.2519, pruned_loss=0.05642, over 953197.49 frames. ], batch size: 30, lr: 3.38e-03, grad_scale: 32.0 2023-03-26 21:02:50,074 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95908.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 21:02:55,838 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95916.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:03:02,195 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95924.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:03:13,534 INFO [finetune.py:976] (3/7) Epoch 17, batch 4300, loss[loss=0.165, simple_loss=0.2291, pruned_loss=0.05045, over 4901.00 frames. ], tot_loss[loss=0.18, simple_loss=0.2489, pruned_loss=0.05549, over 952966.64 frames. ], batch size: 35, lr: 3.38e-03, grad_scale: 32.0 2023-03-26 21:03:17,531 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-26 21:03:18,258 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.024e+02 1.465e+02 1.654e+02 2.123e+02 3.225e+02, threshold=3.308e+02, percent-clipped=0.0 2023-03-26 21:03:26,497 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-03-26 21:03:27,290 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=95964.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:03:33,598 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=95972.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:03:34,216 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95973.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:03:38,603 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.33 vs. limit=5.0 2023-03-26 21:03:39,487 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95980.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:03:47,253 INFO [finetune.py:976] (3/7) Epoch 17, batch 4350, loss[loss=0.1846, simple_loss=0.2527, pruned_loss=0.05822, over 4813.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2451, pruned_loss=0.05416, over 952984.98 frames. ], batch size: 39, lr: 3.38e-03, grad_scale: 32.0 2023-03-26 21:03:48,533 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95995.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:04:20,828 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96041.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:04:21,946 INFO [finetune.py:976] (3/7) Epoch 17, batch 4400, loss[loss=0.157, simple_loss=0.2303, pruned_loss=0.04185, over 4934.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.245, pruned_loss=0.05407, over 954596.87 frames. ], batch size: 38, lr: 3.38e-03, grad_scale: 32.0 2023-03-26 21:04:22,001 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=96043.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:04:23,226 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.9999, 3.4809, 3.6267, 3.8589, 3.7564, 3.4797, 4.0573, 1.3749], device='cuda:3'), covar=tensor([0.0776, 0.0846, 0.0891, 0.0965, 0.1255, 0.1685, 0.0795, 0.5717], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0243, 0.0275, 0.0291, 0.0334, 0.0280, 0.0300, 0.0293], 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') 2023-03-26 21:04:28,706 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.398e+01 1.490e+02 1.749e+02 2.200e+02 3.209e+02, threshold=3.497e+02, percent-clipped=0.0 2023-03-26 21:04:40,098 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-03-26 21:04:59,721 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96080.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:05:13,437 INFO [finetune.py:976] (3/7) Epoch 17, batch 4450, loss[loss=0.1616, simple_loss=0.2476, pruned_loss=0.03784, over 4317.00 frames. ], tot_loss[loss=0.1809, simple_loss=0.2501, pruned_loss=0.05587, over 952674.12 frames. ], batch size: 19, lr: 3.38e-03, grad_scale: 32.0 2023-03-26 21:05:52,532 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5960, 2.3960, 2.9694, 1.8069, 2.6855, 3.0190, 2.1658, 3.0343], device='cuda:3'), covar=tensor([0.1517, 0.1925, 0.1672, 0.2679, 0.1022, 0.1471, 0.2676, 0.0894], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0204, 0.0189, 0.0190, 0.0177, 0.0212, 0.0216, 0.0199], 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') 2023-03-26 21:05:58,468 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96141.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:05:59,553 INFO [finetune.py:976] (3/7) Epoch 17, batch 4500, loss[loss=0.1728, simple_loss=0.2455, pruned_loss=0.05001, over 4889.00 frames. ], tot_loss[loss=0.1827, simple_loss=0.2526, pruned_loss=0.05641, over 955711.84 frames. ], batch size: 35, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:06:03,840 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.121e+02 1.724e+02 1.946e+02 2.358e+02 4.504e+02, threshold=3.891e+02, percent-clipped=3.0 2023-03-26 21:06:13,020 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0637, 1.0948, 1.9836, 1.9014, 1.7834, 1.7247, 1.7606, 1.8921], device='cuda:3'), covar=tensor([0.3637, 0.3935, 0.3388, 0.3611, 0.4955, 0.3828, 0.4417, 0.3103], device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0241, 0.0260, 0.0275, 0.0274, 0.0248, 0.0284, 0.0242], 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') 2023-03-26 21:06:13,982 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96165.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:06:15,788 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96168.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 21:06:28,846 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 2023-03-26 21:06:33,243 INFO [finetune.py:976] (3/7) Epoch 17, batch 4550, loss[loss=0.149, simple_loss=0.2253, pruned_loss=0.03632, over 4766.00 frames. ], tot_loss[loss=0.184, simple_loss=0.2543, pruned_loss=0.05683, over 957171.50 frames. ], batch size: 26, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:06:33,425 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-26 21:06:35,808 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2623, 2.1041, 1.8191, 2.2020, 1.9401, 1.9854, 1.9621, 2.7715], device='cuda:3'), covar=tensor([0.4256, 0.4868, 0.3624, 0.4304, 0.4677, 0.2673, 0.4601, 0.1871], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0261, 0.0227, 0.0277, 0.0252, 0.0219, 0.0252, 0.0231], 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') 2023-03-26 21:06:39,485 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96203.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 21:06:54,552 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96226.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:06:57,461 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6518, 1.5007, 2.2358, 3.3954, 2.3169, 2.4086, 1.1342, 2.7504], device='cuda:3'), covar=tensor([0.1694, 0.1449, 0.1225, 0.0512, 0.0786, 0.1741, 0.1730, 0.0521], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0115, 0.0133, 0.0164, 0.0100, 0.0135, 0.0123, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 21:07:05,026 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0 2023-03-26 21:07:07,174 INFO [finetune.py:976] (3/7) Epoch 17, batch 4600, loss[loss=0.1294, simple_loss=0.2008, pruned_loss=0.02901, over 4787.00 frames. ], tot_loss[loss=0.1835, simple_loss=0.2538, pruned_loss=0.0566, over 958528.55 frames. ], batch size: 25, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:07:11,422 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.099e+02 1.530e+02 1.886e+02 2.340e+02 4.335e+02, threshold=3.772e+02, percent-clipped=2.0 2023-03-26 21:07:26,464 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96273.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:07:40,044 INFO [finetune.py:976] (3/7) Epoch 17, batch 4650, loss[loss=0.1885, simple_loss=0.2542, pruned_loss=0.06139, over 4865.00 frames. ], tot_loss[loss=0.1815, simple_loss=0.2513, pruned_loss=0.05584, over 960127.45 frames. ], batch size: 34, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:07:42,486 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9009, 1.1075, 1.8976, 1.8497, 1.6757, 1.6355, 1.7032, 1.7749], device='cuda:3'), covar=tensor([0.3718, 0.3771, 0.3217, 0.3157, 0.4556, 0.3573, 0.4183, 0.3065], device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0241, 0.0259, 0.0274, 0.0273, 0.0248, 0.0283, 0.0241], 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') 2023-03-26 21:07:48,808 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-26 21:07:58,449 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=96321.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:08:00,269 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96324.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:08:04,541 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5730, 1.8421, 2.1821, 1.7272, 1.7618, 4.1095, 1.6138, 1.8432], device='cuda:3'), covar=tensor([0.0933, 0.1674, 0.1171, 0.0993, 0.1499, 0.0188, 0.1428, 0.1651], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0083, 0.0075, 0.0079, 0.0092, 0.0081, 0.0086, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 21:08:08,002 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96336.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:08:13,200 INFO [finetune.py:976] (3/7) Epoch 17, batch 4700, loss[loss=0.149, simple_loss=0.2219, pruned_loss=0.03802, over 4918.00 frames. ], tot_loss[loss=0.1795, simple_loss=0.2487, pruned_loss=0.05514, over 958743.96 frames. ], batch size: 37, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:08:18,315 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.127e+02 1.531e+02 1.882e+02 2.216e+02 4.319e+02, threshold=3.764e+02, percent-clipped=2.0 2023-03-26 21:08:38,658 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.2835, 3.8539, 4.0335, 3.9867, 3.8145, 3.6698, 4.4133, 1.4963], device='cuda:3'), covar=tensor([0.1300, 0.1688, 0.1320, 0.1942, 0.2274, 0.2526, 0.1221, 0.7538], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0245, 0.0277, 0.0292, 0.0335, 0.0281, 0.0301, 0.0294], 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') 2023-03-26 21:08:40,451 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96385.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:08:43,961 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96390.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:08:45,650 INFO [finetune.py:976] (3/7) Epoch 17, batch 4750, loss[loss=0.17, simple_loss=0.2398, pruned_loss=0.05007, over 4748.00 frames. ], tot_loss[loss=0.1789, simple_loss=0.2476, pruned_loss=0.05514, over 958055.22 frames. ], batch size: 27, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:09:10,012 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3738, 1.3481, 1.1856, 1.4071, 1.6871, 1.6210, 1.4084, 1.2090], device='cuda:3'), covar=tensor([0.0422, 0.0391, 0.0699, 0.0370, 0.0242, 0.0555, 0.0381, 0.0505], device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0109, 0.0144, 0.0113, 0.0101, 0.0109, 0.0099, 0.0110], device='cuda:3'), out_proj_covar=tensor([7.4880e-05, 8.3759e-05, 1.1347e-04, 8.7015e-05, 7.8374e-05, 8.0575e-05, 7.3802e-05, 8.4261e-05], device='cuda:3') 2023-03-26 21:09:14,233 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3264, 2.1857, 1.8400, 2.0858, 2.3125, 2.0569, 2.5115, 2.2990], device='cuda:3'), covar=tensor([0.1414, 0.2294, 0.3381, 0.2667, 0.2571, 0.1773, 0.2904, 0.1967], device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0188, 0.0236, 0.0255, 0.0246, 0.0204, 0.0214, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 21:09:14,756 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96436.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:09:14,900 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-03-26 21:09:19,420 INFO [finetune.py:976] (3/7) Epoch 17, batch 4800, loss[loss=0.1982, simple_loss=0.2694, pruned_loss=0.06347, over 4804.00 frames. ], tot_loss[loss=0.1807, simple_loss=0.2497, pruned_loss=0.05582, over 958185.68 frames. ], batch size: 41, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:09:25,023 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.112e+02 1.609e+02 1.875e+02 2.422e+02 6.864e+02, threshold=3.750e+02, percent-clipped=2.0 2023-03-26 21:09:25,758 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96451.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:09:36,657 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96468.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 21:09:40,277 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.0779, 0.9423, 1.0128, 0.3826, 0.9327, 1.1338, 1.1733, 0.9826], device='cuda:3'), covar=tensor([0.0954, 0.0587, 0.0521, 0.0601, 0.0561, 0.0634, 0.0393, 0.0737], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0151, 0.0123, 0.0127, 0.0129, 0.0129, 0.0143, 0.0148], device='cuda:3'), out_proj_covar=tensor([9.1699e-05, 1.0931e-04, 8.8104e-05, 9.0516e-05, 9.1084e-05, 9.2829e-05, 1.0266e-04, 1.0627e-04], device='cuda:3') 2023-03-26 21:09:45,507 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7874, 1.3222, 0.8722, 1.6201, 2.1349, 1.3457, 1.5448, 1.6223], device='cuda:3'), covar=tensor([0.1447, 0.2092, 0.2007, 0.1185, 0.1868, 0.2060, 0.1437, 0.2039], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0095, 0.0110, 0.0092, 0.0119, 0.0094, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 21:09:54,757 INFO [finetune.py:976] (3/7) Epoch 17, batch 4850, loss[loss=0.2301, simple_loss=0.2963, pruned_loss=0.08193, over 4809.00 frames. ], tot_loss[loss=0.1829, simple_loss=0.2528, pruned_loss=0.05644, over 956084.10 frames. ], batch size: 45, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:10:02,926 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96503.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:10:15,688 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=96516.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 21:10:18,746 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96521.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:10:45,274 INFO [finetune.py:976] (3/7) Epoch 17, batch 4900, loss[loss=0.3316, simple_loss=0.3604, pruned_loss=0.1514, over 4160.00 frames. ], tot_loss[loss=0.1838, simple_loss=0.2538, pruned_loss=0.05687, over 955502.01 frames. ], batch size: 65, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:10:54,621 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.014e+02 1.592e+02 1.896e+02 2.164e+02 3.347e+02, threshold=3.792e+02, percent-clipped=0.0 2023-03-26 21:10:55,804 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=96551.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:11:14,854 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6579, 2.3469, 1.8908, 0.8216, 2.0245, 2.0351, 1.8630, 2.0783], device='cuda:3'), covar=tensor([0.0744, 0.0861, 0.1580, 0.2237, 0.1489, 0.2177, 0.2226, 0.0971], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0197, 0.0200, 0.0185, 0.0214, 0.0209, 0.0224, 0.0198], 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') 2023-03-26 21:11:26,120 INFO [finetune.py:976] (3/7) Epoch 17, batch 4950, loss[loss=0.1519, simple_loss=0.2209, pruned_loss=0.04143, over 4812.00 frames. ], tot_loss[loss=0.1849, simple_loss=0.2552, pruned_loss=0.05731, over 955955.76 frames. ], batch size: 25, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:11:55,611 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96636.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:11:59,760 INFO [finetune.py:976] (3/7) Epoch 17, batch 5000, loss[loss=0.1795, simple_loss=0.2446, pruned_loss=0.05723, over 4306.00 frames. ], tot_loss[loss=0.1827, simple_loss=0.253, pruned_loss=0.05617, over 955254.34 frames. ], batch size: 66, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:12:04,406 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.088e+02 1.531e+02 1.819e+02 2.156e+02 3.437e+02, threshold=3.638e+02, percent-clipped=0.0 2023-03-26 21:12:14,901 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2099, 1.4746, 0.6294, 1.9497, 2.5105, 1.8023, 1.7338, 1.8523], device='cuda:3'), covar=tensor([0.1970, 0.3071, 0.2887, 0.1715, 0.2092, 0.2506, 0.2092, 0.3058], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0095, 0.0110, 0.0092, 0.0119, 0.0094, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 21:12:24,599 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96680.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:12:26,963 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=96684.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:12:30,398 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7808, 1.3628, 0.7804, 1.6572, 2.2818, 1.3208, 1.5325, 1.6414], device='cuda:3'), covar=tensor([0.1479, 0.2129, 0.2091, 0.1174, 0.1803, 0.1904, 0.1556, 0.2025], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0095, 0.0109, 0.0092, 0.0118, 0.0094, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 21:12:33,328 INFO [finetune.py:976] (3/7) Epoch 17, batch 5050, loss[loss=0.2043, simple_loss=0.2662, pruned_loss=0.07115, over 4900.00 frames. ], tot_loss[loss=0.1802, simple_loss=0.2498, pruned_loss=0.05524, over 956362.36 frames. ], batch size: 43, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:12:40,860 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-26 21:12:55,826 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0905, 1.8664, 2.2306, 1.6258, 2.0669, 2.3004, 2.2533, 1.4195], device='cuda:3'), covar=tensor([0.0797, 0.0906, 0.0678, 0.0964, 0.0804, 0.0686, 0.0679, 0.1862], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0134, 0.0139, 0.0122, 0.0124, 0.0139, 0.0141, 0.0162], 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') 2023-03-26 21:13:01,825 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96736.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:13:06,442 INFO [finetune.py:976] (3/7) Epoch 17, batch 5100, loss[loss=0.1814, simple_loss=0.2531, pruned_loss=0.05483, over 4815.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2463, pruned_loss=0.05422, over 956753.06 frames. ], batch size: 40, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:13:08,318 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96746.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:13:10,589 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.091e+02 1.513e+02 1.848e+02 2.246e+02 3.685e+02, threshold=3.695e+02, percent-clipped=1.0 2023-03-26 21:13:28,916 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96776.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:13:33,646 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=96784.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:13:39,088 INFO [finetune.py:976] (3/7) Epoch 17, batch 5150, loss[loss=0.2033, simple_loss=0.2716, pruned_loss=0.06753, over 4814.00 frames. ], tot_loss[loss=0.1792, simple_loss=0.2474, pruned_loss=0.05545, over 953005.87 frames. ], batch size: 41, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:13:58,216 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96821.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:14:08,361 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96837.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:14:11,882 INFO [finetune.py:976] (3/7) Epoch 17, batch 5200, loss[loss=0.2162, simple_loss=0.2988, pruned_loss=0.06685, over 4862.00 frames. ], tot_loss[loss=0.1823, simple_loss=0.2517, pruned_loss=0.0564, over 955514.42 frames. ], batch size: 34, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:14:16,597 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.064e+02 1.635e+02 1.977e+02 2.300e+02 5.939e+02, threshold=3.955e+02, percent-clipped=5.0 2023-03-26 21:14:29,996 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=96869.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:14:40,791 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7069, 3.6625, 3.4052, 1.7241, 3.7092, 2.8854, 0.7788, 2.6351], device='cuda:3'), covar=tensor([0.2450, 0.2013, 0.1743, 0.3402, 0.1179, 0.0989, 0.4684, 0.1394], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0177, 0.0160, 0.0129, 0.0160, 0.0124, 0.0148, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 21:14:44,941 INFO [finetune.py:976] (3/7) Epoch 17, batch 5250, loss[loss=0.1907, simple_loss=0.2597, pruned_loss=0.06082, over 4755.00 frames. ], tot_loss[loss=0.1836, simple_loss=0.2536, pruned_loss=0.05679, over 954467.70 frames. ], batch size: 27, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:14:53,275 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.72 vs. limit=2.0 2023-03-26 21:15:05,815 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3520, 2.8218, 2.6624, 1.2632, 2.8633, 2.1983, 0.7053, 1.9962], device='cuda:3'), covar=tensor([0.2716, 0.2120, 0.1801, 0.3189, 0.1489, 0.1167, 0.3948, 0.1473], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0178, 0.0161, 0.0129, 0.0161, 0.0125, 0.0149, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 21:15:05,891 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7264, 1.6631, 1.4972, 1.4673, 1.8241, 1.5473, 1.8220, 1.7487], device='cuda:3'), covar=tensor([0.1413, 0.1945, 0.2756, 0.2228, 0.2395, 0.1654, 0.2877, 0.1703], device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0189, 0.0236, 0.0255, 0.0247, 0.0204, 0.0215, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 21:15:19,344 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4923, 2.6662, 2.5722, 1.9858, 2.7499, 2.8083, 2.9313, 2.4042], device='cuda:3'), covar=tensor([0.0667, 0.0588, 0.0723, 0.0857, 0.0492, 0.0740, 0.0596, 0.0995], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0134, 0.0139, 0.0122, 0.0124, 0.0139, 0.0141, 0.0163], 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') 2023-03-26 21:15:21,071 INFO [finetune.py:976] (3/7) Epoch 17, batch 5300, loss[loss=0.1853, simple_loss=0.2673, pruned_loss=0.05167, over 4790.00 frames. ], tot_loss[loss=0.1848, simple_loss=0.2548, pruned_loss=0.05742, over 950879.79 frames. ], batch size: 51, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:15:30,005 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.125e+02 1.644e+02 1.959e+02 2.379e+02 3.599e+02, threshold=3.918e+02, percent-clipped=0.0 2023-03-26 21:15:49,161 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7014, 1.5139, 1.3986, 1.7295, 1.8940, 1.7210, 1.2517, 1.3865], device='cuda:3'), covar=tensor([0.2058, 0.2095, 0.1996, 0.1571, 0.1722, 0.1259, 0.2572, 0.1962], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0208, 0.0212, 0.0191, 0.0241, 0.0186, 0.0215, 0.0200], 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') 2023-03-26 21:16:00,364 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96980.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:16:17,642 INFO [finetune.py:976] (3/7) Epoch 17, batch 5350, loss[loss=0.1842, simple_loss=0.2576, pruned_loss=0.05536, over 4727.00 frames. ], tot_loss[loss=0.1845, simple_loss=0.2549, pruned_loss=0.057, over 951439.59 frames. ], batch size: 54, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:16:25,286 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97000.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:16:39,445 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5545, 1.5740, 2.0898, 1.7616, 1.7068, 3.6370, 1.4770, 1.6846], device='cuda:3'), covar=tensor([0.0957, 0.1685, 0.1067, 0.0984, 0.1483, 0.0199, 0.1440, 0.1634], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0081, 0.0073, 0.0077, 0.0091, 0.0080, 0.0084, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 21:16:44,612 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=97028.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:16:54,581 INFO [finetune.py:976] (3/7) Epoch 17, batch 5400, loss[loss=0.1477, simple_loss=0.2226, pruned_loss=0.03637, over 4752.00 frames. ], tot_loss[loss=0.1825, simple_loss=0.2522, pruned_loss=0.05638, over 951035.92 frames. ], batch size: 28, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:16:56,527 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97046.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:16:58,804 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.012e+02 1.591e+02 1.860e+02 2.348e+02 4.043e+02, threshold=3.721e+02, percent-clipped=1.0 2023-03-26 21:17:06,262 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97061.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:17:27,397 INFO [finetune.py:976] (3/7) Epoch 17, batch 5450, loss[loss=0.1384, simple_loss=0.2103, pruned_loss=0.0332, over 4761.00 frames. ], tot_loss[loss=0.1795, simple_loss=0.2487, pruned_loss=0.05511, over 951526.17 frames. ], batch size: 54, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:17:28,069 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=97094.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:17:52,425 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97132.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:17:55,970 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8416, 3.3663, 3.3628, 1.8910, 3.5528, 2.7091, 1.0462, 2.5780], device='cuda:3'), covar=tensor([0.2950, 0.2228, 0.1514, 0.2972, 0.1109, 0.1113, 0.4016, 0.1445], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0176, 0.0159, 0.0128, 0.0159, 0.0124, 0.0148, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 21:18:00,547 INFO [finetune.py:976] (3/7) Epoch 17, batch 5500, loss[loss=0.1666, simple_loss=0.2419, pruned_loss=0.04567, over 4827.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2456, pruned_loss=0.05434, over 953176.38 frames. ], batch size: 39, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:18:04,743 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.221e+01 1.533e+02 1.769e+02 2.042e+02 3.202e+02, threshold=3.539e+02, percent-clipped=0.0 2023-03-26 21:18:33,718 INFO [finetune.py:976] (3/7) Epoch 17, batch 5550, loss[loss=0.1912, simple_loss=0.2562, pruned_loss=0.0631, over 4906.00 frames. ], tot_loss[loss=0.179, simple_loss=0.2479, pruned_loss=0.05507, over 952943.63 frames. ], batch size: 32, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:18:41,651 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1309, 1.2667, 1.2486, 1.2989, 1.3530, 2.3925, 1.1924, 1.3522], device='cuda:3'), covar=tensor([0.0979, 0.1821, 0.1211, 0.0982, 0.1590, 0.0373, 0.1588, 0.1772], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0074, 0.0077, 0.0091, 0.0081, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 21:19:05,109 INFO [finetune.py:976] (3/7) Epoch 17, batch 5600, loss[loss=0.2086, simple_loss=0.2911, pruned_loss=0.06309, over 4869.00 frames. ], tot_loss[loss=0.1819, simple_loss=0.2518, pruned_loss=0.05599, over 953174.71 frames. ], batch size: 34, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:19:06,324 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9219, 3.8425, 3.7001, 1.8334, 3.9690, 2.9717, 0.7540, 2.8031], device='cuda:3'), covar=tensor([0.2089, 0.2382, 0.1416, 0.3382, 0.1050, 0.0958, 0.4624, 0.1411], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0177, 0.0159, 0.0129, 0.0159, 0.0124, 0.0148, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 21:19:09,081 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.154e+02 1.641e+02 1.914e+02 2.406e+02 4.422e+02, threshold=3.827e+02, percent-clipped=1.0 2023-03-26 21:19:14,863 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97260.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:19:27,714 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0083, 1.9154, 1.5907, 1.8093, 1.7106, 1.6578, 1.7324, 2.4692], device='cuda:3'), covar=tensor([0.3719, 0.3946, 0.3219, 0.3641, 0.3964, 0.2419, 0.3781, 0.1630], device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0262, 0.0228, 0.0277, 0.0253, 0.0220, 0.0253, 0.0232], 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') 2023-03-26 21:19:29,763 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3295, 2.8984, 2.8094, 1.1945, 3.0497, 2.1645, 0.6103, 1.9316], device='cuda:3'), covar=tensor([0.2406, 0.2720, 0.1591, 0.3769, 0.1318, 0.1248, 0.4337, 0.1796], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0176, 0.0158, 0.0128, 0.0158, 0.0123, 0.0147, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 21:19:29,845 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8682, 1.6227, 1.4951, 1.2544, 1.6048, 1.6067, 1.5651, 2.1450], device='cuda:3'), covar=tensor([0.3993, 0.4574, 0.3512, 0.4307, 0.4146, 0.2646, 0.3997, 0.2121], device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0263, 0.0228, 0.0277, 0.0253, 0.0220, 0.0253, 0.0232], 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') 2023-03-26 21:19:34,293 INFO [finetune.py:976] (3/7) Epoch 17, batch 5650, loss[loss=0.2216, simple_loss=0.304, pruned_loss=0.0696, over 4770.00 frames. ], tot_loss[loss=0.1839, simple_loss=0.2545, pruned_loss=0.05667, over 951047.49 frames. ], batch size: 54, lr: 3.37e-03, grad_scale: 32.0 2023-03-26 21:19:35,046 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.11 vs. limit=2.0 2023-03-26 21:19:39,017 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1921, 3.6420, 3.8725, 4.0268, 3.9615, 3.7208, 4.2818, 1.4743], device='cuda:3'), covar=tensor([0.0743, 0.0824, 0.0742, 0.0889, 0.1147, 0.1398, 0.0613, 0.5122], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0245, 0.0276, 0.0290, 0.0335, 0.0280, 0.0298, 0.0294], 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') 2023-03-26 21:19:47,306 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1000, 1.3131, 1.4707, 0.9726, 1.1615, 1.3531, 1.2358, 1.4805], device='cuda:3'), covar=tensor([0.1042, 0.1609, 0.1054, 0.1215, 0.0865, 0.1030, 0.2313, 0.0698], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0202, 0.0188, 0.0189, 0.0175, 0.0212, 0.0215, 0.0198], 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') 2023-03-26 21:19:51,444 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97321.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:19:56,363 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3052, 1.2505, 1.3147, 0.6518, 1.2815, 1.2800, 1.2620, 1.1914], device='cuda:3'), covar=tensor([0.0510, 0.0690, 0.0604, 0.0883, 0.0773, 0.0639, 0.0602, 0.1097], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0135, 0.0140, 0.0123, 0.0125, 0.0140, 0.0141, 0.0163], 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') 2023-03-26 21:20:04,084 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97342.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:20:04,601 INFO [finetune.py:976] (3/7) Epoch 17, batch 5700, loss[loss=0.1991, simple_loss=0.2516, pruned_loss=0.07329, over 3989.00 frames. ], tot_loss[loss=0.181, simple_loss=0.2502, pruned_loss=0.05586, over 933740.14 frames. ], batch size: 17, lr: 3.36e-03, grad_scale: 32.0 2023-03-26 21:20:07,070 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97347.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:20:08,735 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.528e+01 1.533e+02 1.739e+02 2.212e+02 3.283e+02, threshold=3.478e+02, percent-clipped=0.0 2023-03-26 21:20:12,306 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97356.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:20:35,933 INFO [finetune.py:976] (3/7) Epoch 18, batch 0, loss[loss=0.1916, simple_loss=0.2622, pruned_loss=0.06048, over 4930.00 frames. ], tot_loss[loss=0.1916, simple_loss=0.2622, pruned_loss=0.06048, over 4930.00 frames. ], batch size: 42, lr: 3.36e-03, grad_scale: 32.0 2023-03-26 21:20:35,933 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 21:20:46,783 INFO [finetune.py:1010] (3/7) Epoch 18, validation: loss=0.1584, simple_loss=0.2281, pruned_loss=0.0444, over 2265189.00 frames. 2023-03-26 21:20:46,783 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6434MB 2023-03-26 21:20:49,151 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97374.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:21:26,087 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97403.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 21:21:30,222 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97408.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:21:39,452 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1604, 1.8806, 2.2865, 3.9997, 2.7369, 2.7299, 1.0060, 3.3380], device='cuda:3'), covar=tensor([0.1550, 0.1392, 0.1402, 0.0509, 0.0743, 0.1430, 0.1857, 0.0403], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0134, 0.0164, 0.0100, 0.0136, 0.0123, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 21:21:47,679 INFO [finetune.py:976] (3/7) Epoch 18, batch 50, loss[loss=0.1954, simple_loss=0.2673, pruned_loss=0.06175, over 4832.00 frames. ], tot_loss[loss=0.1878, simple_loss=0.2571, pruned_loss=0.05923, over 216825.98 frames. ], batch size: 49, lr: 3.36e-03, grad_scale: 32.0 2023-03-26 21:21:58,963 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97432.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:22:00,769 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97435.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:22:09,812 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.163e+01 1.563e+02 1.902e+02 2.308e+02 3.615e+02, threshold=3.804e+02, percent-clipped=1.0 2023-03-26 21:22:25,131 INFO [finetune.py:976] (3/7) Epoch 18, batch 100, loss[loss=0.1699, simple_loss=0.2401, pruned_loss=0.04982, over 4888.00 frames. ], tot_loss[loss=0.1858, simple_loss=0.2528, pruned_loss=0.05935, over 381668.27 frames. ], batch size: 32, lr: 3.36e-03, grad_scale: 32.0 2023-03-26 21:22:31,665 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=97480.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:22:54,249 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0775, 2.0025, 1.9588, 1.3484, 2.1362, 2.1430, 2.0542, 1.7307], device='cuda:3'), covar=tensor([0.0589, 0.0606, 0.0739, 0.0965, 0.0601, 0.0691, 0.0633, 0.1079], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0134, 0.0139, 0.0122, 0.0123, 0.0138, 0.0140, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 21:22:58,722 INFO [finetune.py:976] (3/7) Epoch 18, batch 150, loss[loss=0.1476, simple_loss=0.2269, pruned_loss=0.03415, over 4873.00 frames. ], tot_loss[loss=0.1794, simple_loss=0.2466, pruned_loss=0.05617, over 510424.70 frames. ], batch size: 34, lr: 3.36e-03, grad_scale: 32.0 2023-03-26 21:23:17,208 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.032e+02 1.559e+02 1.792e+02 2.227e+02 6.409e+02, threshold=3.584e+02, percent-clipped=2.0 2023-03-26 21:23:19,727 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7571, 1.4261, 0.9181, 1.7540, 2.1707, 1.4714, 1.5871, 1.7825], device='cuda:3'), covar=tensor([0.1508, 0.1982, 0.1856, 0.1097, 0.1801, 0.1831, 0.1383, 0.1890], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0095, 0.0110, 0.0092, 0.0118, 0.0094, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 21:23:20,360 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97555.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:23:26,129 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9155, 1.7237, 1.5653, 1.2494, 1.6554, 1.6505, 1.6645, 2.2479], device='cuda:3'), covar=tensor([0.3832, 0.3948, 0.3107, 0.3642, 0.3779, 0.2294, 0.3530, 0.1771], device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0259, 0.0224, 0.0273, 0.0249, 0.0217, 0.0249, 0.0228], 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') 2023-03-26 21:23:32,346 INFO [finetune.py:976] (3/7) Epoch 18, batch 200, loss[loss=0.1502, simple_loss=0.229, pruned_loss=0.0357, over 4783.00 frames. ], tot_loss[loss=0.1804, simple_loss=0.2468, pruned_loss=0.05705, over 608303.55 frames. ], batch size: 29, lr: 3.36e-03, grad_scale: 32.0 2023-03-26 21:24:01,851 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97616.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:24:01,908 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97616.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:24:04,228 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7755, 1.3774, 1.8478, 1.8082, 1.5846, 1.5535, 1.7084, 1.7362], device='cuda:3'), covar=tensor([0.4009, 0.4043, 0.3211, 0.3730, 0.4530, 0.3711, 0.4654, 0.3053], device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0239, 0.0259, 0.0273, 0.0272, 0.0247, 0.0282, 0.0240], 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') 2023-03-26 21:24:05,308 INFO [finetune.py:976] (3/7) Epoch 18, batch 250, loss[loss=0.158, simple_loss=0.2315, pruned_loss=0.04225, over 4827.00 frames. ], tot_loss[loss=0.1805, simple_loss=0.2483, pruned_loss=0.05636, over 686238.20 frames. ], batch size: 30, lr: 3.36e-03, grad_scale: 64.0 2023-03-26 21:24:24,121 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.114e+02 1.615e+02 1.960e+02 2.417e+02 4.168e+02, threshold=3.921e+02, percent-clipped=3.0 2023-03-26 21:24:27,892 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97656.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:24:37,889 INFO [finetune.py:976] (3/7) Epoch 18, batch 300, loss[loss=0.2338, simple_loss=0.3033, pruned_loss=0.08213, over 4820.00 frames. ], tot_loss[loss=0.1846, simple_loss=0.2534, pruned_loss=0.0579, over 746241.92 frames. ], batch size: 38, lr: 3.36e-03, grad_scale: 64.0 2023-03-26 21:24:40,298 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97674.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 21:24:56,231 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97698.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 21:24:59,261 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97703.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:24:59,859 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=97704.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:25:10,593 INFO [finetune.py:976] (3/7) Epoch 18, batch 350, loss[loss=0.1887, simple_loss=0.2653, pruned_loss=0.05608, over 4906.00 frames. ], tot_loss[loss=0.1867, simple_loss=0.2556, pruned_loss=0.05886, over 792622.36 frames. ], batch size: 37, lr: 3.36e-03, grad_scale: 64.0 2023-03-26 21:25:17,543 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97730.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:25:20,597 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97735.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 21:25:28,926 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4177, 2.2362, 1.7848, 2.2904, 2.1702, 1.9912, 2.6547, 2.3841], device='cuda:3'), covar=tensor([0.1255, 0.2204, 0.3221, 0.2694, 0.2584, 0.1683, 0.2790, 0.1807], device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0188, 0.0235, 0.0254, 0.0246, 0.0203, 0.0214, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 21:25:30,594 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.123e+02 1.611e+02 1.882e+02 2.387e+02 3.928e+02, threshold=3.763e+02, percent-clipped=1.0 2023-03-26 21:25:43,299 INFO [finetune.py:976] (3/7) Epoch 18, batch 400, loss[loss=0.1679, simple_loss=0.2478, pruned_loss=0.04402, over 4722.00 frames. ], tot_loss[loss=0.1853, simple_loss=0.2554, pruned_loss=0.05755, over 829124.03 frames. ], batch size: 54, lr: 3.36e-03, grad_scale: 64.0 2023-03-26 21:25:46,123 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6823, 1.2029, 0.9373, 1.5608, 2.0884, 1.0946, 1.3991, 1.5749], device='cuda:3'), covar=tensor([0.1465, 0.1973, 0.1740, 0.1101, 0.1711, 0.1957, 0.1380, 0.1753], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0095, 0.0110, 0.0092, 0.0119, 0.0094, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 21:25:52,689 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.95 vs. limit=5.0 2023-03-26 21:26:07,922 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-26 21:26:22,833 INFO [finetune.py:976] (3/7) Epoch 18, batch 450, loss[loss=0.1885, simple_loss=0.2583, pruned_loss=0.05935, over 4888.00 frames. ], tot_loss[loss=0.1825, simple_loss=0.2525, pruned_loss=0.05623, over 859011.57 frames. ], batch size: 43, lr: 3.36e-03, grad_scale: 64.0 2023-03-26 21:26:52,265 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8688, 1.6923, 1.5456, 1.3791, 1.9321, 1.6677, 1.8509, 1.8608], device='cuda:3'), covar=tensor([0.1348, 0.1965, 0.2892, 0.2591, 0.2670, 0.1634, 0.2773, 0.1789], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0187, 0.0234, 0.0253, 0.0244, 0.0201, 0.0213, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 21:27:01,044 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.011e+02 1.519e+02 1.772e+02 2.128e+02 3.513e+02, threshold=3.544e+02, percent-clipped=0.0 2023-03-26 21:27:16,935 INFO [finetune.py:976] (3/7) Epoch 18, batch 500, loss[loss=0.1656, simple_loss=0.2367, pruned_loss=0.04718, over 4906.00 frames. ], tot_loss[loss=0.1807, simple_loss=0.2503, pruned_loss=0.05552, over 881414.64 frames. ], batch size: 37, lr: 3.36e-03, grad_scale: 64.0 2023-03-26 21:27:29,765 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97888.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:27:30,780 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-03-26 21:27:44,610 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97911.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:27:47,672 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97916.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:27:50,634 INFO [finetune.py:976] (3/7) Epoch 18, batch 550, loss[loss=0.1581, simple_loss=0.2492, pruned_loss=0.03351, over 4896.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2467, pruned_loss=0.05446, over 899082.97 frames. ], batch size: 35, lr: 3.36e-03, grad_scale: 64.0 2023-03-26 21:28:19,275 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97949.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:28:19,731 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.294e+01 1.530e+02 1.816e+02 2.060e+02 3.951e+02, threshold=3.633e+02, percent-clipped=3.0 2023-03-26 21:28:28,265 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=97964.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:28:32,533 INFO [finetune.py:976] (3/7) Epoch 18, batch 600, loss[loss=0.156, simple_loss=0.2268, pruned_loss=0.04262, over 4906.00 frames. ], tot_loss[loss=0.1793, simple_loss=0.2482, pruned_loss=0.05521, over 911262.11 frames. ], batch size: 37, lr: 3.36e-03, grad_scale: 64.0 2023-03-26 21:28:51,970 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97998.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:28:56,788 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98003.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:29:07,659 INFO [finetune.py:976] (3/7) Epoch 18, batch 650, loss[loss=0.2563, simple_loss=0.3308, pruned_loss=0.09085, over 4809.00 frames. ], tot_loss[loss=0.1818, simple_loss=0.2515, pruned_loss=0.05608, over 920307.66 frames. ], batch size: 51, lr: 3.36e-03, grad_scale: 64.0 2023-03-26 21:29:13,276 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98030.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 21:29:13,309 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98030.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:29:25,509 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=98046.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:29:28,302 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.027e+02 1.543e+02 1.878e+02 2.128e+02 3.672e+02, threshold=3.757e+02, percent-clipped=1.0 2023-03-26 21:29:28,989 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=98051.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:29:41,536 INFO [finetune.py:976] (3/7) Epoch 18, batch 700, loss[loss=0.1872, simple_loss=0.2633, pruned_loss=0.05553, over 4829.00 frames. ], tot_loss[loss=0.1815, simple_loss=0.2517, pruned_loss=0.05566, over 929177.67 frames. ], batch size: 47, lr: 3.36e-03, grad_scale: 64.0 2023-03-26 21:29:45,869 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=98078.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:30:05,713 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98106.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:30:11,000 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7124, 3.8054, 3.5524, 1.8775, 3.9084, 2.8269, 0.8928, 2.6529], device='cuda:3'), covar=tensor([0.2206, 0.1493, 0.1364, 0.2989, 0.0927, 0.0949, 0.4093, 0.1387], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0177, 0.0159, 0.0128, 0.0160, 0.0124, 0.0148, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 21:30:14,155 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.6121, 1.6453, 1.7225, 0.9971, 1.7935, 2.0080, 1.9595, 1.4709], device='cuda:3'), covar=tensor([0.0840, 0.0654, 0.0506, 0.0547, 0.0438, 0.0547, 0.0351, 0.0716], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0151, 0.0124, 0.0127, 0.0130, 0.0129, 0.0143, 0.0148], device='cuda:3'), out_proj_covar=tensor([9.1361e-05, 1.0967e-04, 8.8496e-05, 9.0323e-05, 9.1903e-05, 9.2905e-05, 1.0280e-04, 1.0652e-04], device='cuda:3') 2023-03-26 21:30:15,239 INFO [finetune.py:976] (3/7) Epoch 18, batch 750, loss[loss=0.1824, simple_loss=0.2532, pruned_loss=0.05586, over 4867.00 frames. ], tot_loss[loss=0.1827, simple_loss=0.2535, pruned_loss=0.05592, over 937055.27 frames. ], batch size: 31, lr: 3.36e-03, grad_scale: 64.0 2023-03-26 21:30:34,724 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.151e+02 1.501e+02 1.785e+02 2.309e+02 4.193e+02, threshold=3.569e+02, percent-clipped=2.0 2023-03-26 21:30:46,062 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98167.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:30:48,363 INFO [finetune.py:976] (3/7) Epoch 18, batch 800, loss[loss=0.1847, simple_loss=0.2475, pruned_loss=0.061, over 4921.00 frames. ], tot_loss[loss=0.182, simple_loss=0.2528, pruned_loss=0.05559, over 940659.39 frames. ], batch size: 38, lr: 3.36e-03, grad_scale: 32.0 2023-03-26 21:31:15,632 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98211.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:31:22,189 INFO [finetune.py:976] (3/7) Epoch 18, batch 850, loss[loss=0.1692, simple_loss=0.2425, pruned_loss=0.04798, over 4822.00 frames. ], tot_loss[loss=0.18, simple_loss=0.2501, pruned_loss=0.05497, over 942297.69 frames. ], batch size: 39, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:31:45,825 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98244.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:31:55,530 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.155e+02 1.511e+02 1.794e+02 2.111e+02 3.360e+02, threshold=3.589e+02, percent-clipped=0.0 2023-03-26 21:32:06,606 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=98259.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:32:16,156 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7187, 2.5095, 2.1854, 2.8622, 2.6244, 2.3196, 3.1583, 2.6493], device='cuda:3'), covar=tensor([0.1377, 0.2295, 0.3041, 0.2640, 0.2624, 0.1744, 0.2788, 0.2012], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0186, 0.0233, 0.0253, 0.0244, 0.0201, 0.0213, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 21:32:25,141 INFO [finetune.py:976] (3/7) Epoch 18, batch 900, loss[loss=0.153, simple_loss=0.2241, pruned_loss=0.04096, over 4830.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2468, pruned_loss=0.05323, over 945817.31 frames. ], batch size: 38, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:32:59,670 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-03-26 21:33:02,857 INFO [finetune.py:976] (3/7) Epoch 18, batch 950, loss[loss=0.2167, simple_loss=0.2755, pruned_loss=0.079, over 4935.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2461, pruned_loss=0.05351, over 949955.21 frames. ], batch size: 38, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:33:08,479 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98330.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 21:33:16,064 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6316, 2.0528, 1.5711, 1.6683, 2.3413, 2.2328, 1.9246, 1.9301], device='cuda:3'), covar=tensor([0.0445, 0.0351, 0.0557, 0.0350, 0.0249, 0.0545, 0.0430, 0.0416], device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0108, 0.0144, 0.0112, 0.0101, 0.0109, 0.0099, 0.0110], device='cuda:3'), out_proj_covar=tensor([7.4758e-05, 8.3073e-05, 1.1336e-04, 8.6526e-05, 7.8452e-05, 8.0564e-05, 7.3970e-05, 8.4057e-05], device='cuda:3') 2023-03-26 21:33:23,085 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.033e+02 1.511e+02 1.760e+02 2.160e+02 3.441e+02, threshold=3.521e+02, percent-clipped=0.0 2023-03-26 21:33:37,315 INFO [finetune.py:976] (3/7) Epoch 18, batch 1000, loss[loss=0.2441, simple_loss=0.3226, pruned_loss=0.08283, over 4810.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2473, pruned_loss=0.0541, over 950759.67 frames. ], batch size: 40, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:33:42,166 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=98378.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 21:34:02,765 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98410.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:34:14,798 INFO [finetune.py:976] (3/7) Epoch 18, batch 1050, loss[loss=0.2037, simple_loss=0.2807, pruned_loss=0.06334, over 4925.00 frames. ], tot_loss[loss=0.1801, simple_loss=0.2504, pruned_loss=0.05489, over 952808.43 frames. ], batch size: 38, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:34:36,631 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.123e+02 1.558e+02 1.856e+02 2.287e+02 5.753e+02, threshold=3.713e+02, percent-clipped=4.0 2023-03-26 21:34:44,665 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98462.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:34:51,512 INFO [finetune.py:976] (3/7) Epoch 18, batch 1100, loss[loss=0.1438, simple_loss=0.2163, pruned_loss=0.03564, over 4757.00 frames. ], tot_loss[loss=0.1817, simple_loss=0.2524, pruned_loss=0.0555, over 952432.09 frames. ], batch size: 27, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:34:51,637 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98471.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:35:22,275 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-26 21:35:24,231 INFO [finetune.py:976] (3/7) Epoch 18, batch 1150, loss[loss=0.2215, simple_loss=0.2876, pruned_loss=0.07775, over 4100.00 frames. ], tot_loss[loss=0.1817, simple_loss=0.2524, pruned_loss=0.05555, over 952209.85 frames. ], batch size: 66, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:35:31,531 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.59 vs. limit=5.0 2023-03-26 21:35:38,239 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-26 21:35:39,074 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98543.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:35:39,639 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98544.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:35:43,759 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.197e+02 1.665e+02 1.912e+02 2.323e+02 4.830e+02, threshold=3.825e+02, percent-clipped=3.0 2023-03-26 21:35:57,262 INFO [finetune.py:976] (3/7) Epoch 18, batch 1200, loss[loss=0.1925, simple_loss=0.2464, pruned_loss=0.06928, over 4894.00 frames. ], tot_loss[loss=0.1823, simple_loss=0.2524, pruned_loss=0.05609, over 953870.91 frames. ], batch size: 36, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:36:06,646 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1663, 2.1378, 1.9650, 2.3007, 2.6232, 2.2484, 2.1179, 1.6720], device='cuda:3'), covar=tensor([0.2086, 0.1899, 0.1731, 0.1509, 0.1772, 0.1113, 0.1977, 0.1827], device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0210, 0.0213, 0.0192, 0.0243, 0.0187, 0.0216, 0.0201], 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') 2023-03-26 21:36:12,051 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=98592.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:36:19,592 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98604.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:36:31,240 INFO [finetune.py:976] (3/7) Epoch 18, batch 1250, loss[loss=0.1639, simple_loss=0.2157, pruned_loss=0.05602, over 4000.00 frames. ], tot_loss[loss=0.1798, simple_loss=0.2493, pruned_loss=0.05513, over 953264.98 frames. ], batch size: 17, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:36:37,837 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-26 21:37:01,601 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.074e+02 1.465e+02 1.816e+02 2.333e+02 4.053e+02, threshold=3.632e+02, percent-clipped=1.0 2023-03-26 21:37:29,821 INFO [finetune.py:976] (3/7) Epoch 18, batch 1300, loss[loss=0.1714, simple_loss=0.2364, pruned_loss=0.05314, over 4760.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2463, pruned_loss=0.05403, over 955009.54 frames. ], batch size: 28, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:37:45,606 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9483, 1.0347, 1.8566, 1.7954, 1.6889, 1.6192, 1.6751, 1.8037], device='cuda:3'), covar=tensor([0.3835, 0.3965, 0.3601, 0.3589, 0.5001, 0.3904, 0.4162, 0.3290], device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0241, 0.0259, 0.0275, 0.0274, 0.0248, 0.0284, 0.0241], 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') 2023-03-26 21:37:57,837 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5868, 1.4503, 1.8273, 1.7849, 1.4816, 3.2808, 1.3047, 1.4783], device='cuda:3'), covar=tensor([0.0841, 0.1729, 0.1131, 0.0936, 0.1663, 0.0249, 0.1545, 0.1741], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0074, 0.0077, 0.0091, 0.0080, 0.0084, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 21:38:02,973 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-26 21:38:11,396 INFO [finetune.py:976] (3/7) Epoch 18, batch 1350, loss[loss=0.2204, simple_loss=0.2959, pruned_loss=0.0725, over 4872.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.247, pruned_loss=0.05492, over 954496.41 frames. ], batch size: 44, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:38:31,463 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.112e+02 1.676e+02 1.899e+02 2.271e+02 3.691e+02, threshold=3.798e+02, percent-clipped=1.0 2023-03-26 21:38:38,241 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98762.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:38:41,102 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98766.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:38:44,610 INFO [finetune.py:976] (3/7) Epoch 18, batch 1400, loss[loss=0.2083, simple_loss=0.2874, pruned_loss=0.06458, over 4934.00 frames. ], tot_loss[loss=0.1823, simple_loss=0.2513, pruned_loss=0.05664, over 955090.65 frames. ], batch size: 42, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:38:59,244 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.9958, 1.7193, 1.7659, 0.8233, 1.9279, 2.2154, 1.9461, 1.7255], device='cuda:3'), covar=tensor([0.0912, 0.0780, 0.0607, 0.0661, 0.0467, 0.0605, 0.0449, 0.0693], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0151, 0.0123, 0.0126, 0.0131, 0.0128, 0.0143, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.1204e-05, 1.0935e-04, 8.8548e-05, 9.0044e-05, 9.2325e-05, 9.2170e-05, 1.0261e-04, 1.0609e-04], device='cuda:3') 2023-03-26 21:39:06,062 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0284, 1.8834, 1.7890, 2.0556, 2.4033, 2.0685, 1.9726, 1.7736], device='cuda:3'), covar=tensor([0.1644, 0.1732, 0.1494, 0.1329, 0.1645, 0.1034, 0.1980, 0.1494], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0209, 0.0212, 0.0191, 0.0241, 0.0186, 0.0214, 0.0200], 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') 2023-03-26 21:39:10,804 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=98810.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:39:17,933 INFO [finetune.py:976] (3/7) Epoch 18, batch 1450, loss[loss=0.1653, simple_loss=0.2367, pruned_loss=0.04691, over 4741.00 frames. ], tot_loss[loss=0.182, simple_loss=0.252, pruned_loss=0.05598, over 953607.71 frames. ], batch size: 59, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:39:40,442 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.160e+02 1.618e+02 1.947e+02 2.343e+02 3.750e+02, threshold=3.895e+02, percent-clipped=0.0 2023-03-26 21:39:52,508 INFO [finetune.py:976] (3/7) Epoch 18, batch 1500, loss[loss=0.159, simple_loss=0.2398, pruned_loss=0.03912, over 4819.00 frames. ], tot_loss[loss=0.184, simple_loss=0.254, pruned_loss=0.05697, over 954697.45 frames. ], batch size: 39, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:40:06,460 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98890.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:40:08,032 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-26 21:40:10,013 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98895.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:40:12,869 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98899.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:40:26,171 INFO [finetune.py:976] (3/7) Epoch 18, batch 1550, loss[loss=0.1697, simple_loss=0.2333, pruned_loss=0.05306, over 4790.00 frames. ], tot_loss[loss=0.1835, simple_loss=0.254, pruned_loss=0.05653, over 955455.90 frames. ], batch size: 25, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:40:47,923 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.111e+02 1.516e+02 1.782e+02 2.221e+02 4.511e+02, threshold=3.564e+02, percent-clipped=1.0 2023-03-26 21:40:48,065 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98951.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:40:51,134 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98956.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:41:00,138 INFO [finetune.py:976] (3/7) Epoch 18, batch 1600, loss[loss=0.1685, simple_loss=0.2309, pruned_loss=0.05309, over 4803.00 frames. ], tot_loss[loss=0.1816, simple_loss=0.2517, pruned_loss=0.05578, over 955325.37 frames. ], batch size: 45, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:41:00,247 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98971.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:41:33,949 INFO [finetune.py:976] (3/7) Epoch 18, batch 1650, loss[loss=0.1444, simple_loss=0.2198, pruned_loss=0.03446, over 4690.00 frames. ], tot_loss[loss=0.1799, simple_loss=0.2492, pruned_loss=0.05533, over 955843.59 frames. ], batch size: 23, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:41:41,746 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99032.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:42:02,530 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.819e+01 1.633e+02 2.015e+02 2.343e+02 3.841e+02, threshold=4.030e+02, percent-clipped=3.0 2023-03-26 21:42:03,105 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99051.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:42:17,587 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99066.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:42:25,675 INFO [finetune.py:976] (3/7) Epoch 18, batch 1700, loss[loss=0.2181, simple_loss=0.2828, pruned_loss=0.07672, over 4205.00 frames. ], tot_loss[loss=0.18, simple_loss=0.2486, pruned_loss=0.05566, over 955210.85 frames. ], batch size: 65, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:43:01,000 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99112.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:43:02,158 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=99114.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:43:10,652 INFO [finetune.py:976] (3/7) Epoch 18, batch 1750, loss[loss=0.1418, simple_loss=0.2111, pruned_loss=0.03626, over 4777.00 frames. ], tot_loss[loss=0.1819, simple_loss=0.2508, pruned_loss=0.0565, over 955325.19 frames. ], batch size: 28, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:43:38,844 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.106e+01 1.643e+02 1.917e+02 2.422e+02 4.876e+02, threshold=3.835e+02, percent-clipped=2.0 2023-03-26 21:43:51,913 INFO [finetune.py:976] (3/7) Epoch 18, batch 1800, loss[loss=0.1969, simple_loss=0.2668, pruned_loss=0.06348, over 4932.00 frames. ], tot_loss[loss=0.1853, simple_loss=0.2543, pruned_loss=0.05813, over 955095.42 frames. ], batch size: 33, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:43:58,442 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-03-26 21:44:10,873 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99199.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:44:13,960 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1860, 2.0969, 1.7257, 2.1386, 2.1151, 1.8794, 2.4547, 2.1668], device='cuda:3'), covar=tensor([0.1159, 0.1828, 0.2585, 0.2289, 0.2272, 0.1482, 0.2835, 0.1546], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0188, 0.0234, 0.0253, 0.0245, 0.0202, 0.0214, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 21:44:25,712 INFO [finetune.py:976] (3/7) Epoch 18, batch 1850, loss[loss=0.1692, simple_loss=0.2549, pruned_loss=0.04178, over 4821.00 frames. ], tot_loss[loss=0.187, simple_loss=0.2559, pruned_loss=0.05902, over 954966.49 frames. ], batch size: 39, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:44:30,699 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99229.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:44:42,401 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99246.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:44:43,017 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=99247.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:44:45,838 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.071e+02 1.618e+02 1.939e+02 2.302e+02 3.831e+02, threshold=3.878e+02, percent-clipped=0.0 2023-03-26 21:44:45,940 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99251.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:44:47,180 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.2823, 2.9991, 2.6349, 1.3558, 2.8432, 2.2974, 2.2643, 2.5907], device='cuda:3'), covar=tensor([0.0791, 0.0714, 0.1739, 0.2282, 0.1716, 0.2647, 0.2085, 0.1175], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0194, 0.0201, 0.0183, 0.0214, 0.0209, 0.0222, 0.0197], 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') 2023-03-26 21:44:57,630 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99268.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:44:59,345 INFO [finetune.py:976] (3/7) Epoch 18, batch 1900, loss[loss=0.1507, simple_loss=0.2245, pruned_loss=0.03846, over 4754.00 frames. ], tot_loss[loss=0.1842, simple_loss=0.2544, pruned_loss=0.05701, over 955647.08 frames. ], batch size: 28, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:45:11,541 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99290.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:45:33,104 INFO [finetune.py:976] (3/7) Epoch 18, batch 1950, loss[loss=0.1866, simple_loss=0.2582, pruned_loss=0.05754, over 4811.00 frames. ], tot_loss[loss=0.1824, simple_loss=0.2525, pruned_loss=0.05611, over 955430.07 frames. ], batch size: 40, lr: 3.35e-03, grad_scale: 32.0 2023-03-26 21:45:35,066 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99324.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:45:36,833 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99327.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:45:38,111 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99329.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:45:52,709 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.035e+02 1.452e+02 1.660e+02 1.987e+02 4.820e+02, threshold=3.320e+02, percent-clipped=1.0 2023-03-26 21:46:06,274 INFO [finetune.py:976] (3/7) Epoch 18, batch 2000, loss[loss=0.1374, simple_loss=0.2136, pruned_loss=0.0306, over 4906.00 frames. ], tot_loss[loss=0.1791, simple_loss=0.2487, pruned_loss=0.0547, over 952197.59 frames. ], batch size: 43, lr: 3.34e-03, grad_scale: 32.0 2023-03-26 21:46:13,562 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.8742, 4.2348, 4.3778, 4.6336, 4.5871, 4.3008, 4.9583, 1.6733], device='cuda:3'), covar=tensor([0.0730, 0.0770, 0.0772, 0.0859, 0.1124, 0.1513, 0.0464, 0.5465], device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0243, 0.0278, 0.0291, 0.0332, 0.0279, 0.0300, 0.0294], 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') 2023-03-26 21:46:15,411 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99385.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:46:21,351 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3888, 1.5520, 1.2740, 1.4698, 1.6423, 1.6219, 1.5470, 1.3908], device='cuda:3'), covar=tensor([0.0362, 0.0253, 0.0542, 0.0281, 0.0216, 0.0372, 0.0267, 0.0333], device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0107, 0.0143, 0.0111, 0.0100, 0.0108, 0.0098, 0.0110], device='cuda:3'), out_proj_covar=tensor([7.4422e-05, 8.2352e-05, 1.1301e-04, 8.5279e-05, 7.8020e-05, 7.9940e-05, 7.3587e-05, 8.3770e-05], device='cuda:3') 2023-03-26 21:46:23,065 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5395, 1.3550, 1.2653, 1.5318, 1.6316, 1.5486, 1.0143, 1.3114], device='cuda:3'), covar=tensor([0.2009, 0.1977, 0.1780, 0.1484, 0.1576, 0.1198, 0.2353, 0.1758], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0208, 0.0211, 0.0191, 0.0240, 0.0186, 0.0214, 0.0199], 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') 2023-03-26 21:46:30,177 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99407.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:46:34,917 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.0179, 3.9315, 3.7891, 1.9746, 4.0890, 3.1842, 0.9975, 2.9308], device='cuda:3'), covar=tensor([0.2004, 0.1854, 0.1374, 0.3461, 0.1022, 0.0851, 0.4528, 0.1224], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0177, 0.0160, 0.0130, 0.0160, 0.0123, 0.0149, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 21:46:39,598 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99420.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:46:40,078 INFO [finetune.py:976] (3/7) Epoch 18, batch 2050, loss[loss=0.1697, simple_loss=0.237, pruned_loss=0.05119, over 4901.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2463, pruned_loss=0.0546, over 950721.06 frames. ], batch size: 35, lr: 3.34e-03, grad_scale: 32.0 2023-03-26 21:46:51,102 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.1844, 1.1844, 1.3296, 0.6146, 1.3007, 1.4551, 1.5879, 1.2301], device='cuda:3'), covar=tensor([0.0755, 0.0538, 0.0480, 0.0449, 0.0411, 0.0513, 0.0250, 0.0545], device='cuda:3'), in_proj_covar=tensor([0.0125, 0.0153, 0.0124, 0.0127, 0.0132, 0.0129, 0.0143, 0.0148], device='cuda:3'), out_proj_covar=tensor([9.1889e-05, 1.1074e-04, 8.9013e-05, 9.0297e-05, 9.2872e-05, 9.2721e-05, 1.0312e-04, 1.0672e-04], device='cuda:3') 2023-03-26 21:46:59,827 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.109e+02 1.434e+02 1.742e+02 2.145e+02 5.049e+02, threshold=3.484e+02, percent-clipped=5.0 2023-03-26 21:47:17,709 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99468.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 21:47:19,435 INFO [finetune.py:976] (3/7) Epoch 18, batch 2100, loss[loss=0.1546, simple_loss=0.223, pruned_loss=0.04312, over 4765.00 frames. ], tot_loss[loss=0.1786, simple_loss=0.2465, pruned_loss=0.05538, over 949018.19 frames. ], batch size: 27, lr: 3.34e-03, grad_scale: 32.0 2023-03-26 21:47:30,318 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.29 vs. limit=5.0 2023-03-26 21:47:31,487 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99481.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:47:58,125 INFO [finetune.py:976] (3/7) Epoch 18, batch 2150, loss[loss=0.1644, simple_loss=0.2424, pruned_loss=0.04319, over 4155.00 frames. ], tot_loss[loss=0.1822, simple_loss=0.2509, pruned_loss=0.05674, over 949472.27 frames. ], batch size: 65, lr: 3.34e-03, grad_scale: 32.0 2023-03-26 21:48:08,424 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99529.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 21:48:15,508 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9169, 4.3629, 4.2408, 2.1982, 4.4949, 3.4759, 0.8555, 3.0800], device='cuda:3'), covar=tensor([0.2489, 0.1989, 0.1162, 0.3265, 0.0784, 0.0827, 0.4344, 0.1339], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0178, 0.0160, 0.0131, 0.0161, 0.0125, 0.0150, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 21:48:28,287 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99546.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:48:35,644 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.054e+02 1.546e+02 1.925e+02 2.366e+02 3.688e+02, threshold=3.850e+02, percent-clipped=2.0 2023-03-26 21:48:35,735 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99551.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:48:51,485 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99568.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:48:53,717 INFO [finetune.py:976] (3/7) Epoch 18, batch 2200, loss[loss=0.1708, simple_loss=0.2342, pruned_loss=0.05376, over 4788.00 frames. ], tot_loss[loss=0.1842, simple_loss=0.2535, pruned_loss=0.05742, over 952565.32 frames. ], batch size: 26, lr: 3.34e-03, grad_scale: 32.0 2023-03-26 21:49:03,261 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99585.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:49:05,755 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99589.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:49:06,385 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0003, 1.8612, 1.5920, 1.8208, 1.7297, 1.7489, 1.8450, 2.5563], device='cuda:3'), covar=tensor([0.3932, 0.4473, 0.3255, 0.4010, 0.4302, 0.2492, 0.3778, 0.1608], device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0261, 0.0228, 0.0275, 0.0251, 0.0219, 0.0251, 0.0232], 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') 2023-03-26 21:49:08,744 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=99594.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:49:08,768 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.2704, 2.9384, 3.0185, 3.2001, 3.0520, 2.8829, 3.3582, 0.8150], device='cuda:3'), covar=tensor([0.1126, 0.0992, 0.0962, 0.1096, 0.1697, 0.1777, 0.1090, 0.5757], device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0242, 0.0276, 0.0290, 0.0331, 0.0278, 0.0298, 0.0292], 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') 2023-03-26 21:49:11,758 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=99599.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:49:27,186 INFO [finetune.py:976] (3/7) Epoch 18, batch 2250, loss[loss=0.2193, simple_loss=0.2734, pruned_loss=0.08266, over 4806.00 frames. ], tot_loss[loss=0.1851, simple_loss=0.2546, pruned_loss=0.05783, over 952460.43 frames. ], batch size: 45, lr: 3.34e-03, grad_scale: 32.0 2023-03-26 21:49:29,588 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99624.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:49:31,395 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99627.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:49:33,058 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99629.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:49:46,350 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99650.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:49:46,817 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.914e+01 1.516e+02 1.824e+02 2.092e+02 3.162e+02, threshold=3.647e+02, percent-clipped=0.0 2023-03-26 21:50:00,008 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 2023-03-26 21:50:00,832 INFO [finetune.py:976] (3/7) Epoch 18, batch 2300, loss[loss=0.2175, simple_loss=0.2825, pruned_loss=0.0763, over 4769.00 frames. ], tot_loss[loss=0.1845, simple_loss=0.2543, pruned_loss=0.05734, over 953913.66 frames. ], batch size: 51, lr: 3.34e-03, grad_scale: 32.0 2023-03-26 21:50:03,812 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=99675.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:50:06,805 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99680.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:50:06,831 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99680.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:50:18,912 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.87 vs. limit=5.0 2023-03-26 21:50:24,116 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99707.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:50:34,079 INFO [finetune.py:976] (3/7) Epoch 18, batch 2350, loss[loss=0.2219, simple_loss=0.2848, pruned_loss=0.07947, over 4812.00 frames. ], tot_loss[loss=0.1819, simple_loss=0.2518, pruned_loss=0.05601, over 953986.03 frames. ], batch size: 40, lr: 3.34e-03, grad_scale: 32.0 2023-03-26 21:50:36,755 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-26 21:50:47,885 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99741.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:50:54,365 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.082e+02 1.551e+02 1.845e+02 2.144e+02 4.060e+02, threshold=3.690e+02, percent-clipped=1.0 2023-03-26 21:50:56,921 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=99755.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:50:57,575 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4408, 2.2726, 2.0547, 1.1189, 2.2574, 1.9690, 1.7506, 2.1593], device='cuda:3'), covar=tensor([0.0804, 0.0785, 0.1383, 0.1952, 0.1243, 0.2040, 0.2143, 0.0918], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0195, 0.0200, 0.0184, 0.0213, 0.0209, 0.0224, 0.0197], 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') 2023-03-26 21:51:00,201 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-03-26 21:51:08,063 INFO [finetune.py:976] (3/7) Epoch 18, batch 2400, loss[loss=0.1832, simple_loss=0.2517, pruned_loss=0.05739, over 4931.00 frames. ], tot_loss[loss=0.1791, simple_loss=0.2483, pruned_loss=0.05491, over 954046.14 frames. ], batch size: 33, lr: 3.34e-03, grad_scale: 32.0 2023-03-26 21:51:11,644 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99776.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:51:41,396 INFO [finetune.py:976] (3/7) Epoch 18, batch 2450, loss[loss=0.1622, simple_loss=0.2347, pruned_loss=0.04483, over 4872.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2456, pruned_loss=0.05408, over 955756.57 frames. ], batch size: 34, lr: 3.34e-03, grad_scale: 32.0 2023-03-26 21:51:43,691 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99824.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 21:52:01,732 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.091e+01 1.409e+02 1.705e+02 2.080e+02 4.896e+02, threshold=3.409e+02, percent-clipped=2.0 2023-03-26 21:52:14,325 INFO [finetune.py:976] (3/7) Epoch 18, batch 2500, loss[loss=0.1755, simple_loss=0.2512, pruned_loss=0.04989, over 4755.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2464, pruned_loss=0.05439, over 955408.28 frames. ], batch size: 54, lr: 3.34e-03, grad_scale: 32.0 2023-03-26 21:52:26,294 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99885.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:52:32,713 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5757, 3.6158, 3.2901, 1.5037, 3.6949, 2.9314, 1.2397, 2.4904], device='cuda:3'), covar=tensor([0.2554, 0.1828, 0.1551, 0.3605, 0.0994, 0.0964, 0.3808, 0.1494], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0176, 0.0159, 0.0129, 0.0159, 0.0124, 0.0148, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 21:52:47,226 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1103, 1.4005, 0.8669, 1.9243, 2.3210, 1.5256, 1.8050, 1.7282], device='cuda:3'), covar=tensor([0.1350, 0.2119, 0.1963, 0.1111, 0.1691, 0.1901, 0.1392, 0.2005], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0096, 0.0110, 0.0092, 0.0120, 0.0094, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 21:52:50,111 INFO [finetune.py:976] (3/7) Epoch 18, batch 2550, loss[loss=0.2089, simple_loss=0.275, pruned_loss=0.07141, over 4105.00 frames. ], tot_loss[loss=0.181, simple_loss=0.2504, pruned_loss=0.05587, over 954189.73 frames. ], batch size: 65, lr: 3.34e-03, grad_scale: 32.0 2023-03-26 21:52:52,502 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99924.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:52:52,519 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99924.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:52:58,897 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=99933.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:52:58,935 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1032, 1.2276, 0.7293, 2.0140, 2.3499, 1.8240, 1.6012, 1.8861], device='cuda:3'), covar=tensor([0.1448, 0.2144, 0.2102, 0.1168, 0.1815, 0.1958, 0.1432, 0.2022], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0095, 0.0110, 0.0092, 0.0119, 0.0094, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 21:53:06,646 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99945.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:53:10,619 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.138e+02 1.647e+02 1.933e+02 2.438e+02 4.501e+02, threshold=3.867e+02, percent-clipped=7.0 2023-03-26 21:53:29,758 INFO [finetune.py:976] (3/7) Epoch 18, batch 2600, loss[loss=0.2206, simple_loss=0.2924, pruned_loss=0.07435, over 4830.00 frames. ], tot_loss[loss=0.181, simple_loss=0.2512, pruned_loss=0.05534, over 956170.79 frames. ], batch size: 39, lr: 3.34e-03, grad_scale: 32.0 2023-03-26 21:53:30,434 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=99972.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:53:32,304 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-26 21:53:40,400 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99980.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:53:48,219 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7021, 1.6165, 1.4781, 1.8143, 2.1806, 1.9311, 1.6600, 1.4276], device='cuda:3'), covar=tensor([0.2074, 0.1912, 0.1810, 0.1585, 0.1574, 0.1115, 0.2132, 0.1841], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0207, 0.0211, 0.0191, 0.0240, 0.0185, 0.0214, 0.0200], 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') 2023-03-26 21:54:12,886 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100007.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 21:54:24,615 INFO [finetune.py:976] (3/7) Epoch 18, batch 2650, loss[loss=0.1766, simple_loss=0.2574, pruned_loss=0.04793, over 4849.00 frames. ], tot_loss[loss=0.1801, simple_loss=0.2505, pruned_loss=0.05486, over 952696.39 frames. ], batch size: 44, lr: 3.34e-03, grad_scale: 16.0 2023-03-26 21:54:29,385 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=100028.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:54:35,206 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100036.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:54:46,210 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.050e+02 1.530e+02 1.912e+02 2.295e+02 4.144e+02, threshold=3.823e+02, percent-clipped=1.0 2023-03-26 21:54:56,620 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100068.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 21:54:58,304 INFO [finetune.py:976] (3/7) Epoch 18, batch 2700, loss[loss=0.2001, simple_loss=0.266, pruned_loss=0.06709, over 4816.00 frames. ], tot_loss[loss=0.1789, simple_loss=0.2498, pruned_loss=0.05404, over 953199.31 frames. ], batch size: 40, lr: 3.34e-03, grad_scale: 16.0 2023-03-26 21:55:01,423 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100076.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:55:12,540 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5255, 2.4411, 2.1242, 2.6110, 2.3498, 2.2883, 2.3551, 3.2584], device='cuda:3'), covar=tensor([0.3876, 0.4870, 0.3475, 0.4229, 0.4767, 0.2610, 0.4431, 0.1737], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0260, 0.0227, 0.0273, 0.0250, 0.0219, 0.0250, 0.0231], 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') 2023-03-26 21:55:31,863 INFO [finetune.py:976] (3/7) Epoch 18, batch 2750, loss[loss=0.1409, simple_loss=0.214, pruned_loss=0.03393, over 4803.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2475, pruned_loss=0.0536, over 953604.08 frames. ], batch size: 25, lr: 3.34e-03, grad_scale: 16.0 2023-03-26 21:55:33,700 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=100124.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:55:33,746 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100124.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 21:55:52,996 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.121e+02 1.510e+02 1.766e+02 2.096e+02 4.575e+02, threshold=3.532e+02, percent-clipped=1.0 2023-03-26 21:56:05,347 INFO [finetune.py:976] (3/7) Epoch 18, batch 2800, loss[loss=0.1917, simple_loss=0.2412, pruned_loss=0.07116, over 4915.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2448, pruned_loss=0.05291, over 953071.83 frames. ], batch size: 37, lr: 3.34e-03, grad_scale: 16.0 2023-03-26 21:56:06,012 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=100172.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 21:56:11,430 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3870, 1.9055, 2.2544, 2.2708, 1.9990, 1.9763, 2.1273, 2.0969], device='cuda:3'), covar=tensor([0.4413, 0.4342, 0.3764, 0.4345, 0.5743, 0.4502, 0.5484, 0.3559], device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0239, 0.0259, 0.0274, 0.0273, 0.0248, 0.0284, 0.0241], 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') 2023-03-26 21:56:20,206 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-26 21:56:38,931 INFO [finetune.py:976] (3/7) Epoch 18, batch 2850, loss[loss=0.2529, simple_loss=0.3052, pruned_loss=0.1003, over 4844.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2453, pruned_loss=0.05399, over 954571.84 frames. ], batch size: 44, lr: 3.34e-03, grad_scale: 16.0 2023-03-26 21:56:40,884 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100224.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:56:49,834 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100238.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:56:54,569 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100245.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:56:59,193 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.104e+02 1.635e+02 1.899e+02 2.309e+02 4.393e+02, threshold=3.799e+02, percent-clipped=4.0 2023-03-26 21:57:11,714 INFO [finetune.py:976] (3/7) Epoch 18, batch 2900, loss[loss=0.2065, simple_loss=0.2724, pruned_loss=0.07034, over 4915.00 frames. ], tot_loss[loss=0.1801, simple_loss=0.2496, pruned_loss=0.05533, over 953323.71 frames. ], batch size: 36, lr: 3.34e-03, grad_scale: 16.0 2023-03-26 21:57:12,866 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=100272.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:57:26,157 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=100293.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:57:30,863 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100299.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:57:35,921 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.58 vs. limit=5.0 2023-03-26 21:57:39,120 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3332, 2.2808, 1.9858, 2.5103, 3.0795, 2.4204, 2.4202, 1.8280], device='cuda:3'), covar=tensor([0.2149, 0.1895, 0.1884, 0.1546, 0.1532, 0.1113, 0.1899, 0.1873], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0208, 0.0211, 0.0190, 0.0240, 0.0185, 0.0215, 0.0199], 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') 2023-03-26 21:57:45,539 INFO [finetune.py:976] (3/7) Epoch 18, batch 2950, loss[loss=0.1633, simple_loss=0.2364, pruned_loss=0.04514, over 4832.00 frames. ], tot_loss[loss=0.1819, simple_loss=0.2523, pruned_loss=0.05576, over 953644.42 frames. ], batch size: 47, lr: 3.34e-03, grad_scale: 16.0 2023-03-26 21:57:55,263 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100336.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:58:06,323 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.110e+02 1.490e+02 1.822e+02 2.174e+02 4.072e+02, threshold=3.643e+02, percent-clipped=2.0 2023-03-26 21:58:13,575 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100363.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 21:58:18,481 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-26 21:58:18,817 INFO [finetune.py:976] (3/7) Epoch 18, batch 3000, loss[loss=0.2106, simple_loss=0.2793, pruned_loss=0.0709, over 4908.00 frames. ], tot_loss[loss=0.1817, simple_loss=0.2519, pruned_loss=0.05578, over 950698.68 frames. ], batch size: 36, lr: 3.34e-03, grad_scale: 16.0 2023-03-26 21:58:18,818 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 21:58:31,198 INFO [finetune.py:1010] (3/7) Epoch 18, validation: loss=0.1568, simple_loss=0.2261, pruned_loss=0.04375, over 2265189.00 frames. 2023-03-26 21:58:31,198 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6434MB 2023-03-26 21:58:44,385 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=100384.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:59:06,996 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100402.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 21:59:11,384 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-26 21:59:29,743 INFO [finetune.py:976] (3/7) Epoch 18, batch 3050, loss[loss=0.1837, simple_loss=0.2255, pruned_loss=0.07095, over 4101.00 frames. ], tot_loss[loss=0.1816, simple_loss=0.2523, pruned_loss=0.05546, over 949248.43 frames. ], batch size: 17, lr: 3.34e-03, grad_scale: 16.0 2023-03-26 21:59:53,774 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.049e+02 1.572e+02 1.917e+02 2.276e+02 3.597e+02, threshold=3.833e+02, percent-clipped=0.0 2023-03-26 22:00:01,162 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100463.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 22:00:07,399 INFO [finetune.py:976] (3/7) Epoch 18, batch 3100, loss[loss=0.1587, simple_loss=0.2248, pruned_loss=0.04628, over 4245.00 frames. ], tot_loss[loss=0.1808, simple_loss=0.2512, pruned_loss=0.05518, over 948285.59 frames. ], batch size: 18, lr: 3.34e-03, grad_scale: 16.0 2023-03-26 22:00:25,945 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0 2023-03-26 22:00:40,538 INFO [finetune.py:976] (3/7) Epoch 18, batch 3150, loss[loss=0.1514, simple_loss=0.214, pruned_loss=0.04439, over 4803.00 frames. ], tot_loss[loss=0.1807, simple_loss=0.2497, pruned_loss=0.05582, over 950124.40 frames. ], batch size: 25, lr: 3.34e-03, grad_scale: 16.0 2023-03-26 22:00:53,540 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4447, 2.3910, 1.8732, 2.3998, 2.2603, 2.1567, 2.2169, 3.2736], device='cuda:3'), covar=tensor([0.3921, 0.4937, 0.3717, 0.4262, 0.4572, 0.2522, 0.4389, 0.1572], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0259, 0.0226, 0.0272, 0.0249, 0.0217, 0.0249, 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') 2023-03-26 22:00:59,252 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8386, 1.6660, 2.3194, 2.0867, 1.9987, 4.4536, 1.6765, 1.9751], device='cuda:3'), covar=tensor([0.0897, 0.1760, 0.1099, 0.0982, 0.1450, 0.0164, 0.1493, 0.1703], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0074, 0.0077, 0.0090, 0.0080, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 22:01:00,942 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.466e+01 1.523e+02 1.835e+02 2.195e+02 4.344e+02, threshold=3.670e+02, percent-clipped=3.0 2023-03-26 22:01:12,895 INFO [finetune.py:976] (3/7) Epoch 18, batch 3200, loss[loss=0.1665, simple_loss=0.2329, pruned_loss=0.0501, over 4823.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2452, pruned_loss=0.05399, over 950754.67 frames. ], batch size: 38, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:01:28,940 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100594.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:01:46,316 INFO [finetune.py:976] (3/7) Epoch 18, batch 3250, loss[loss=0.2167, simple_loss=0.2653, pruned_loss=0.08404, over 4766.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2466, pruned_loss=0.05437, over 952830.90 frames. ], batch size: 28, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:01:48,764 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3145, 2.1928, 2.2281, 1.6673, 2.1917, 2.3052, 2.3556, 1.8874], device='cuda:3'), covar=tensor([0.0582, 0.0657, 0.0741, 0.0873, 0.0690, 0.0686, 0.0623, 0.1118], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0133, 0.0139, 0.0120, 0.0122, 0.0137, 0.0139, 0.0160], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 22:02:08,114 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.084e+02 1.502e+02 1.862e+02 2.235e+02 4.464e+02, threshold=3.723e+02, percent-clipped=3.0 2023-03-26 22:02:14,917 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100663.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 22:02:20,154 INFO [finetune.py:976] (3/7) Epoch 18, batch 3300, loss[loss=0.1759, simple_loss=0.2514, pruned_loss=0.05019, over 4877.00 frames. ], tot_loss[loss=0.1796, simple_loss=0.2494, pruned_loss=0.05486, over 953263.70 frames. ], batch size: 31, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:02:47,026 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=100711.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 22:02:53,405 INFO [finetune.py:976] (3/7) Epoch 18, batch 3350, loss[loss=0.1855, simple_loss=0.2587, pruned_loss=0.05613, over 4801.00 frames. ], tot_loss[loss=0.1825, simple_loss=0.2524, pruned_loss=0.05626, over 952992.49 frames. ], batch size: 45, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:02:55,514 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-26 22:02:59,510 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100730.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 22:03:14,083 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.901e+01 1.577e+02 1.865e+02 2.249e+02 4.268e+02, threshold=3.731e+02, percent-clipped=3.0 2023-03-26 22:03:18,262 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100758.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 22:03:26,646 INFO [finetune.py:976] (3/7) Epoch 18, batch 3400, loss[loss=0.1658, simple_loss=0.2487, pruned_loss=0.04142, over 4806.00 frames. ], tot_loss[loss=0.1843, simple_loss=0.2544, pruned_loss=0.05712, over 953629.65 frames. ], batch size: 29, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:03:40,309 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100791.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 22:04:13,286 INFO [finetune.py:976] (3/7) Epoch 18, batch 3450, loss[loss=0.1589, simple_loss=0.2284, pruned_loss=0.04467, over 4763.00 frames. ], tot_loss[loss=0.1834, simple_loss=0.2537, pruned_loss=0.05656, over 954594.88 frames. ], batch size: 27, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:04:14,015 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4133, 1.4544, 1.3005, 1.4728, 1.7124, 1.6060, 1.4255, 1.2589], device='cuda:3'), covar=tensor([0.0351, 0.0279, 0.0560, 0.0277, 0.0231, 0.0492, 0.0353, 0.0397], device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0107, 0.0143, 0.0110, 0.0100, 0.0108, 0.0098, 0.0109], device='cuda:3'), out_proj_covar=tensor([7.4024e-05, 8.2353e-05, 1.1261e-04, 8.4594e-05, 7.8234e-05, 7.9916e-05, 7.3467e-05, 8.3214e-05], device='cuda:3') 2023-03-26 22:04:49,164 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.055e+02 1.551e+02 1.762e+02 2.136e+02 3.810e+02, threshold=3.524e+02, percent-clipped=1.0 2023-03-26 22:05:04,971 INFO [finetune.py:976] (3/7) Epoch 18, batch 3500, loss[loss=0.1832, simple_loss=0.2523, pruned_loss=0.05703, over 4895.00 frames. ], tot_loss[loss=0.1814, simple_loss=0.2513, pruned_loss=0.05576, over 955989.51 frames. ], batch size: 32, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:05:08,312 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-26 22:05:21,050 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100894.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:05:38,764 INFO [finetune.py:976] (3/7) Epoch 18, batch 3550, loss[loss=0.1543, simple_loss=0.2188, pruned_loss=0.04492, over 4745.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2474, pruned_loss=0.05411, over 956008.97 frames. ], batch size: 27, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:05:52,400 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=100942.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:05:56,786 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.17 vs. limit=5.0 2023-03-26 22:05:59,322 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.557e+01 1.553e+02 1.835e+02 2.348e+02 4.609e+02, threshold=3.670e+02, percent-clipped=4.0 2023-03-26 22:06:12,132 INFO [finetune.py:976] (3/7) Epoch 18, batch 3600, loss[loss=0.1861, simple_loss=0.2563, pruned_loss=0.05794, over 4886.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2443, pruned_loss=0.05319, over 954161.80 frames. ], batch size: 35, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:06:41,482 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 2023-03-26 22:06:46,076 INFO [finetune.py:976] (3/7) Epoch 18, batch 3650, loss[loss=0.1104, simple_loss=0.1786, pruned_loss=0.02105, over 4678.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2473, pruned_loss=0.05446, over 953085.21 frames. ], batch size: 23, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:07:06,792 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.062e+02 1.559e+02 1.860e+02 2.177e+02 4.070e+02, threshold=3.719e+02, percent-clipped=1.0 2023-03-26 22:07:11,005 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101058.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:07:18,907 INFO [finetune.py:976] (3/7) Epoch 18, batch 3700, loss[loss=0.2089, simple_loss=0.2768, pruned_loss=0.07053, over 4868.00 frames. ], tot_loss[loss=0.1807, simple_loss=0.2503, pruned_loss=0.05554, over 952483.49 frames. ], batch size: 31, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:07:26,093 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3435, 2.9522, 2.7592, 1.1437, 3.1059, 2.3299, 0.5162, 1.9149], device='cuda:3'), covar=tensor([0.2352, 0.2228, 0.1835, 0.3778, 0.1311, 0.1210, 0.4364, 0.1682], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0177, 0.0161, 0.0130, 0.0160, 0.0125, 0.0148, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 22:07:28,499 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101086.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 22:07:36,841 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4572, 2.3524, 2.0158, 0.9627, 2.2717, 1.8146, 1.7198, 2.1322], device='cuda:3'), covar=tensor([0.1015, 0.0810, 0.1752, 0.2270, 0.1510, 0.2594, 0.2466, 0.1041], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0194, 0.0200, 0.0184, 0.0214, 0.0208, 0.0224, 0.0197], 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') 2023-03-26 22:07:42,148 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=101106.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:07:43,259 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101107.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:07:52,641 INFO [finetune.py:976] (3/7) Epoch 18, batch 3750, loss[loss=0.1691, simple_loss=0.2264, pruned_loss=0.05588, over 4746.00 frames. ], tot_loss[loss=0.1819, simple_loss=0.2513, pruned_loss=0.05624, over 953035.67 frames. ], batch size: 23, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:08:07,081 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2746, 2.0905, 2.2022, 0.8823, 2.5414, 2.6885, 2.3572, 1.9779], device='cuda:3'), covar=tensor([0.0993, 0.0937, 0.0569, 0.0878, 0.0524, 0.0711, 0.0576, 0.0822], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0151, 0.0123, 0.0126, 0.0130, 0.0129, 0.0141, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.1060e-05, 1.0976e-04, 8.8153e-05, 8.9480e-05, 9.1400e-05, 9.2239e-05, 1.0157e-04, 1.0590e-04], device='cuda:3') 2023-03-26 22:08:12,837 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.051e+02 1.641e+02 1.814e+02 2.131e+02 4.110e+02, threshold=3.627e+02, percent-clipped=2.0 2023-03-26 22:08:24,435 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101168.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:08:26,160 INFO [finetune.py:976] (3/7) Epoch 18, batch 3800, loss[loss=0.1755, simple_loss=0.2495, pruned_loss=0.05069, over 4847.00 frames. ], tot_loss[loss=0.1831, simple_loss=0.2527, pruned_loss=0.05677, over 952122.64 frames. ], batch size: 44, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:08:59,896 INFO [finetune.py:976] (3/7) Epoch 18, batch 3850, loss[loss=0.1847, simple_loss=0.2473, pruned_loss=0.06106, over 4882.00 frames. ], tot_loss[loss=0.1819, simple_loss=0.2517, pruned_loss=0.05601, over 952725.40 frames. ], batch size: 32, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:09:02,660 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.53 vs. limit=5.0 2023-03-26 22:09:28,093 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-03-26 22:09:28,671 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3618, 2.1993, 1.7886, 2.0690, 2.2765, 1.9738, 2.5349, 2.2547], device='cuda:3'), covar=tensor([0.1271, 0.1917, 0.2958, 0.2415, 0.2600, 0.1719, 0.2518, 0.1835], device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0189, 0.0236, 0.0254, 0.0246, 0.0204, 0.0215, 0.0203], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 22:09:30,624 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.64 vs. limit=5.0 2023-03-26 22:09:30,992 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.034e+02 1.525e+02 1.862e+02 2.180e+02 4.556e+02, threshold=3.724e+02, percent-clipped=2.0 2023-03-26 22:09:57,716 INFO [finetune.py:976] (3/7) Epoch 18, batch 3900, loss[loss=0.1633, simple_loss=0.2424, pruned_loss=0.04205, over 4737.00 frames. ], tot_loss[loss=0.1805, simple_loss=0.2494, pruned_loss=0.05584, over 952396.71 frames. ], batch size: 23, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:10:41,970 INFO [finetune.py:976] (3/7) Epoch 18, batch 3950, loss[loss=0.1572, simple_loss=0.2261, pruned_loss=0.04413, over 4823.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2462, pruned_loss=0.0548, over 951071.34 frames. ], batch size: 38, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:11:02,251 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.024e+02 1.470e+02 1.754e+02 2.083e+02 3.090e+02, threshold=3.508e+02, percent-clipped=0.0 2023-03-26 22:11:15,364 INFO [finetune.py:976] (3/7) Epoch 18, batch 4000, loss[loss=0.2391, simple_loss=0.3023, pruned_loss=0.08795, over 4928.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2458, pruned_loss=0.05486, over 951322.16 frames. ], batch size: 38, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:11:25,995 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101386.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 22:11:49,438 INFO [finetune.py:976] (3/7) Epoch 18, batch 4050, loss[loss=0.1649, simple_loss=0.2429, pruned_loss=0.04342, over 4236.00 frames. ], tot_loss[loss=0.1816, simple_loss=0.2502, pruned_loss=0.05654, over 950123.17 frames. ], batch size: 65, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:11:58,807 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=101434.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 22:12:03,153 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2083, 2.1524, 1.8246, 2.1374, 2.0667, 2.0338, 2.0810, 3.0143], device='cuda:3'), covar=tensor([0.3807, 0.4765, 0.3359, 0.4559, 0.4633, 0.2599, 0.4638, 0.1603], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0262, 0.0229, 0.0275, 0.0252, 0.0220, 0.0251, 0.0233], 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') 2023-03-26 22:12:10,024 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.148e+02 1.670e+02 2.040e+02 2.363e+02 9.256e+02, threshold=4.080e+02, percent-clipped=2.0 2023-03-26 22:12:17,270 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101463.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:12:19,399 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-26 22:12:22,997 INFO [finetune.py:976] (3/7) Epoch 18, batch 4100, loss[loss=0.1774, simple_loss=0.2517, pruned_loss=0.05153, over 4826.00 frames. ], tot_loss[loss=0.184, simple_loss=0.2533, pruned_loss=0.05735, over 951497.00 frames. ], batch size: 33, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:12:24,843 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1223, 1.7577, 2.5492, 4.2736, 2.8167, 2.7905, 0.7928, 3.6241], device='cuda:3'), covar=tensor([0.1615, 0.1453, 0.1326, 0.0455, 0.0725, 0.1492, 0.2053, 0.0318], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0133, 0.0164, 0.0100, 0.0135, 0.0124, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 22:12:31,392 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-26 22:12:56,255 INFO [finetune.py:976] (3/7) Epoch 18, batch 4150, loss[loss=0.1784, simple_loss=0.248, pruned_loss=0.0544, over 4775.00 frames. ], tot_loss[loss=0.1851, simple_loss=0.2549, pruned_loss=0.05769, over 952981.98 frames. ], batch size: 28, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:13:16,877 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.507e+01 1.508e+02 1.851e+02 2.208e+02 3.984e+02, threshold=3.702e+02, percent-clipped=0.0 2023-03-26 22:13:29,472 INFO [finetune.py:976] (3/7) Epoch 18, batch 4200, loss[loss=0.1805, simple_loss=0.2503, pruned_loss=0.05533, over 4851.00 frames. ], tot_loss[loss=0.1833, simple_loss=0.2539, pruned_loss=0.05638, over 953463.45 frames. ], batch size: 44, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:13:38,506 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-26 22:14:00,218 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-26 22:14:03,051 INFO [finetune.py:976] (3/7) Epoch 18, batch 4250, loss[loss=0.1337, simple_loss=0.2111, pruned_loss=0.02818, over 4867.00 frames. ], tot_loss[loss=0.1807, simple_loss=0.2509, pruned_loss=0.05528, over 951902.73 frames. ], batch size: 31, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:14:24,253 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.150e+02 1.580e+02 1.774e+02 2.219e+02 3.425e+02, threshold=3.547e+02, percent-clipped=0.0 2023-03-26 22:14:38,480 INFO [finetune.py:976] (3/7) Epoch 18, batch 4300, loss[loss=0.1443, simple_loss=0.2133, pruned_loss=0.03769, over 4796.00 frames. ], tot_loss[loss=0.1801, simple_loss=0.2491, pruned_loss=0.0555, over 951989.75 frames. ], batch size: 25, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:14:54,343 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101684.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:14:54,431 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-03-26 22:15:33,152 INFO [finetune.py:976] (3/7) Epoch 18, batch 4350, loss[loss=0.1454, simple_loss=0.219, pruned_loss=0.03585, over 4748.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2451, pruned_loss=0.05367, over 952464.41 frames. ], batch size: 23, lr: 3.33e-03, grad_scale: 16.0 2023-03-26 22:16:05,164 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101745.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:16:10,239 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.272e+01 1.483e+02 1.686e+02 2.087e+02 3.591e+02, threshold=3.373e+02, percent-clipped=1.0 2023-03-26 22:16:16,955 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101763.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:16:21,723 INFO [finetune.py:976] (3/7) Epoch 18, batch 4400, loss[loss=0.2006, simple_loss=0.2679, pruned_loss=0.06661, over 4751.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2465, pruned_loss=0.05434, over 953194.73 frames. ], batch size: 27, lr: 3.32e-03, grad_scale: 16.0 2023-03-26 22:16:27,120 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.1622, 2.6711, 2.5304, 1.2567, 2.6478, 2.1610, 2.0558, 2.3803], device='cuda:3'), covar=tensor([0.0883, 0.0874, 0.1569, 0.2073, 0.1632, 0.2076, 0.2069, 0.1134], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0193, 0.0199, 0.0183, 0.0212, 0.0208, 0.0223, 0.0196], 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') 2023-03-26 22:16:44,861 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4524, 2.3142, 2.0097, 0.9602, 2.1383, 1.8163, 1.7593, 2.1049], device='cuda:3'), covar=tensor([0.0878, 0.0742, 0.1578, 0.2052, 0.1470, 0.2135, 0.2071, 0.0940], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0193, 0.0199, 0.0183, 0.0213, 0.0208, 0.0223, 0.0197], 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') 2023-03-26 22:16:49,621 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=101811.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:16:55,630 INFO [finetune.py:976] (3/7) Epoch 18, batch 4450, loss[loss=0.17, simple_loss=0.2546, pruned_loss=0.0427, over 4819.00 frames. ], tot_loss[loss=0.1803, simple_loss=0.2501, pruned_loss=0.05528, over 954388.57 frames. ], batch size: 51, lr: 3.32e-03, grad_scale: 16.0 2023-03-26 22:17:16,755 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.107e+02 1.606e+02 1.893e+02 2.313e+02 4.401e+02, threshold=3.785e+02, percent-clipped=7.0 2023-03-26 22:17:17,552 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.21 vs. limit=5.0 2023-03-26 22:17:29,385 INFO [finetune.py:976] (3/7) Epoch 18, batch 4500, loss[loss=0.1955, simple_loss=0.2828, pruned_loss=0.05413, over 4815.00 frames. ], tot_loss[loss=0.1811, simple_loss=0.2515, pruned_loss=0.05536, over 954734.30 frames. ], batch size: 45, lr: 3.32e-03, grad_scale: 16.0 2023-03-26 22:18:02,646 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101920.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 22:18:03,135 INFO [finetune.py:976] (3/7) Epoch 18, batch 4550, loss[loss=0.1679, simple_loss=0.2335, pruned_loss=0.05118, over 4907.00 frames. ], tot_loss[loss=0.183, simple_loss=0.2535, pruned_loss=0.05626, over 954393.70 frames. ], batch size: 37, lr: 3.32e-03, grad_scale: 16.0 2023-03-26 22:18:24,137 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 1.522e+02 1.794e+02 2.336e+02 4.256e+02, threshold=3.587e+02, percent-clipped=2.0 2023-03-26 22:18:36,749 INFO [finetune.py:976] (3/7) Epoch 18, batch 4600, loss[loss=0.1578, simple_loss=0.2338, pruned_loss=0.0409, over 4755.00 frames. ], tot_loss[loss=0.1826, simple_loss=0.2532, pruned_loss=0.05597, over 954473.69 frames. ], batch size: 26, lr: 3.32e-03, grad_scale: 16.0 2023-03-26 22:18:42,895 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101981.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 22:18:47,464 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101987.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:18:58,375 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2199, 2.1880, 2.1957, 1.4877, 2.1749, 2.3232, 2.2808, 1.9004], device='cuda:3'), covar=tensor([0.0619, 0.0598, 0.0682, 0.0890, 0.0620, 0.0680, 0.0654, 0.0984], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0136, 0.0141, 0.0120, 0.0125, 0.0139, 0.0141, 0.0164], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 22:19:11,219 INFO [finetune.py:976] (3/7) Epoch 18, batch 4650, loss[loss=0.137, simple_loss=0.2143, pruned_loss=0.02983, over 4794.00 frames. ], tot_loss[loss=0.1804, simple_loss=0.2505, pruned_loss=0.05512, over 955785.09 frames. ], batch size: 26, lr: 3.32e-03, grad_scale: 32.0 2023-03-26 22:19:23,878 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102040.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:19:29,227 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102048.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:19:29,911 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-03-26 22:19:31,551 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.060e+02 1.564e+02 1.867e+02 2.217e+02 4.281e+02, threshold=3.734e+02, percent-clipped=4.0 2023-03-26 22:19:45,055 INFO [finetune.py:976] (3/7) Epoch 18, batch 4700, loss[loss=0.1366, simple_loss=0.2156, pruned_loss=0.02875, over 4755.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2457, pruned_loss=0.05287, over 956717.35 frames. ], batch size: 26, lr: 3.32e-03, grad_scale: 32.0 2023-03-26 22:19:57,266 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.2349, 3.6598, 3.8861, 4.0889, 4.0105, 3.7632, 4.3455, 1.4016], device='cuda:3'), covar=tensor([0.0836, 0.0942, 0.0894, 0.0971, 0.1295, 0.1598, 0.0731, 0.5566], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0243, 0.0278, 0.0290, 0.0331, 0.0281, 0.0301, 0.0295], 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') 2023-03-26 22:20:31,466 INFO [finetune.py:976] (3/7) Epoch 18, batch 4750, loss[loss=0.2381, simple_loss=0.2807, pruned_loss=0.09774, over 4026.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2441, pruned_loss=0.05234, over 954415.51 frames. ], batch size: 66, lr: 3.32e-03, grad_scale: 32.0 2023-03-26 22:20:32,876 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1510, 1.5258, 2.1280, 2.0329, 1.9510, 1.8987, 1.9884, 1.9999], device='cuda:3'), covar=tensor([0.3721, 0.4153, 0.3553, 0.3704, 0.4858, 0.3695, 0.4660, 0.3311], device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0240, 0.0259, 0.0275, 0.0275, 0.0248, 0.0283, 0.0241], 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') 2023-03-26 22:20:56,162 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.053e+02 1.538e+02 1.904e+02 2.326e+02 4.380e+02, threshold=3.807e+02, percent-clipped=2.0 2023-03-26 22:21:23,317 INFO [finetune.py:976] (3/7) Epoch 18, batch 4800, loss[loss=0.2402, simple_loss=0.3077, pruned_loss=0.08632, over 4818.00 frames. ], tot_loss[loss=0.177, simple_loss=0.247, pruned_loss=0.05351, over 954090.73 frames. ], batch size: 38, lr: 3.32e-03, grad_scale: 32.0 2023-03-26 22:22:00,412 INFO [finetune.py:976] (3/7) Epoch 18, batch 4850, loss[loss=0.1998, simple_loss=0.2737, pruned_loss=0.06289, over 4832.00 frames. ], tot_loss[loss=0.1798, simple_loss=0.2507, pruned_loss=0.05444, over 955075.39 frames. ], batch size: 30, lr: 3.32e-03, grad_scale: 32.0 2023-03-26 22:22:20,219 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.897e+01 1.478e+02 1.803e+02 2.234e+02 3.533e+02, threshold=3.606e+02, percent-clipped=0.0 2023-03-26 22:22:23,457 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-03-26 22:22:26,321 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-26 22:22:29,380 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-26 22:22:33,608 INFO [finetune.py:976] (3/7) Epoch 18, batch 4900, loss[loss=0.1859, simple_loss=0.2656, pruned_loss=0.05315, over 4860.00 frames. ], tot_loss[loss=0.1813, simple_loss=0.2519, pruned_loss=0.05534, over 952971.73 frames. ], batch size: 31, lr: 3.32e-03, grad_scale: 32.0 2023-03-26 22:22:37,663 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102276.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 22:22:56,975 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.2320, 3.6817, 3.8602, 4.0717, 3.9402, 3.6715, 4.3560, 1.2677], device='cuda:3'), covar=tensor([0.0874, 0.0853, 0.0896, 0.1102, 0.1421, 0.1783, 0.0700, 0.6237], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0244, 0.0279, 0.0292, 0.0334, 0.0283, 0.0302, 0.0297], 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') 2023-03-26 22:23:01,932 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.54 vs. limit=2.0 2023-03-26 22:23:06,930 INFO [finetune.py:976] (3/7) Epoch 18, batch 4950, loss[loss=0.1533, simple_loss=0.2209, pruned_loss=0.04281, over 4756.00 frames. ], tot_loss[loss=0.1822, simple_loss=0.2532, pruned_loss=0.05556, over 954091.88 frames. ], batch size: 28, lr: 3.32e-03, grad_scale: 32.0 2023-03-26 22:23:19,553 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102340.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:23:21,348 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102343.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:23:26,705 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.970e+01 1.449e+02 1.847e+02 2.188e+02 4.191e+02, threshold=3.694e+02, percent-clipped=1.0 2023-03-26 22:23:40,084 INFO [finetune.py:976] (3/7) Epoch 18, batch 5000, loss[loss=0.1562, simple_loss=0.2222, pruned_loss=0.04508, over 4793.00 frames. ], tot_loss[loss=0.181, simple_loss=0.2514, pruned_loss=0.0553, over 954470.61 frames. ], batch size: 29, lr: 3.32e-03, grad_scale: 32.0 2023-03-26 22:23:51,807 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=102388.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:24:01,734 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.7224, 4.1557, 4.2926, 4.4357, 4.4850, 4.2050, 4.8485, 1.8969], device='cuda:3'), covar=tensor([0.0824, 0.0986, 0.0793, 0.1131, 0.1180, 0.1689, 0.0610, 0.4933], device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0242, 0.0277, 0.0289, 0.0330, 0.0281, 0.0300, 0.0295], 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') 2023-03-26 22:24:13,295 INFO [finetune.py:976] (3/7) Epoch 18, batch 5050, loss[loss=0.1372, simple_loss=0.2153, pruned_loss=0.02957, over 4829.00 frames. ], tot_loss[loss=0.179, simple_loss=0.2486, pruned_loss=0.05474, over 953677.10 frames. ], batch size: 39, lr: 3.32e-03, grad_scale: 32.0 2023-03-26 22:24:21,525 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6629, 3.7143, 3.4718, 1.7350, 3.7729, 2.9631, 0.7244, 2.5946], device='cuda:3'), covar=tensor([0.2228, 0.2268, 0.1624, 0.3467, 0.1273, 0.0986, 0.4795, 0.1684], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0179, 0.0162, 0.0130, 0.0163, 0.0125, 0.0150, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 22:24:33,970 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.223e+01 1.582e+02 1.966e+02 2.354e+02 3.513e+02, threshold=3.932e+02, percent-clipped=0.0 2023-03-26 22:24:46,889 INFO [finetune.py:976] (3/7) Epoch 18, batch 5100, loss[loss=0.1717, simple_loss=0.2387, pruned_loss=0.0524, over 4832.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2456, pruned_loss=0.0538, over 955041.03 frames. ], batch size: 47, lr: 3.32e-03, grad_scale: 32.0 2023-03-26 22:24:56,972 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7899, 3.7137, 3.5766, 1.9607, 3.8202, 2.8730, 0.9830, 2.6710], device='cuda:3'), covar=tensor([0.2136, 0.1889, 0.1531, 0.3350, 0.1083, 0.0959, 0.4494, 0.1566], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0180, 0.0163, 0.0131, 0.0164, 0.0126, 0.0151, 0.0126], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 22:25:06,481 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7281, 1.5925, 1.6746, 1.6742, 1.1590, 3.7858, 1.4451, 1.8266], device='cuda:3'), covar=tensor([0.3264, 0.2539, 0.2052, 0.2344, 0.1754, 0.0188, 0.2513, 0.1289], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0114, 0.0119, 0.0123, 0.0113, 0.0095, 0.0096, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 22:25:11,400 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2470, 1.8249, 2.2620, 2.1265, 1.8612, 1.8882, 2.0908, 1.9581], device='cuda:3'), covar=tensor([0.3714, 0.4077, 0.3177, 0.4072, 0.5272, 0.4198, 0.4892, 0.3098], device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0240, 0.0258, 0.0275, 0.0274, 0.0247, 0.0282, 0.0239], 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') 2023-03-26 22:25:20,568 INFO [finetune.py:976] (3/7) Epoch 18, batch 5150, loss[loss=0.2319, simple_loss=0.2945, pruned_loss=0.08467, over 4926.00 frames. ], tot_loss[loss=0.1788, simple_loss=0.2474, pruned_loss=0.05509, over 955968.70 frames. ], batch size: 42, lr: 3.32e-03, grad_scale: 32.0 2023-03-26 22:25:53,892 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.053e+02 1.631e+02 1.989e+02 2.441e+02 4.766e+02, threshold=3.977e+02, percent-clipped=3.0 2023-03-26 22:26:14,502 INFO [finetune.py:976] (3/7) Epoch 18, batch 5200, loss[loss=0.1637, simple_loss=0.2514, pruned_loss=0.03795, over 4820.00 frames. ], tot_loss[loss=0.1816, simple_loss=0.2508, pruned_loss=0.05618, over 954246.13 frames. ], batch size: 39, lr: 3.32e-03, grad_scale: 32.0 2023-03-26 22:26:22,140 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102576.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 22:26:53,507 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.76 vs. limit=5.0 2023-03-26 22:26:56,624 INFO [finetune.py:976] (3/7) Epoch 18, batch 5250, loss[loss=0.1841, simple_loss=0.2427, pruned_loss=0.06275, over 4801.00 frames. ], tot_loss[loss=0.1829, simple_loss=0.2524, pruned_loss=0.05675, over 952366.03 frames. ], batch size: 25, lr: 3.32e-03, grad_scale: 32.0 2023-03-26 22:26:56,756 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8858, 1.7827, 2.3526, 1.5056, 2.1509, 2.2337, 1.7197, 2.4260], device='cuda:3'), covar=tensor([0.1350, 0.1902, 0.1372, 0.2075, 0.0885, 0.1489, 0.2620, 0.0898], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0204, 0.0190, 0.0190, 0.0176, 0.0214, 0.0218, 0.0199], 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') 2023-03-26 22:26:58,546 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=102624.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 22:27:11,987 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102643.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:27:17,760 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.182e+02 1.670e+02 1.981e+02 2.217e+02 4.217e+02, threshold=3.962e+02, percent-clipped=1.0 2023-03-26 22:27:27,029 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4062, 1.5740, 1.2891, 1.4939, 1.7400, 1.6654, 1.5075, 1.3613], device='cuda:3'), covar=tensor([0.0371, 0.0276, 0.0616, 0.0285, 0.0251, 0.0457, 0.0287, 0.0383], device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0107, 0.0143, 0.0110, 0.0099, 0.0108, 0.0098, 0.0110], device='cuda:3'), out_proj_covar=tensor([7.3927e-05, 8.2310e-05, 1.1271e-04, 8.4591e-05, 7.7309e-05, 7.9578e-05, 7.3443e-05, 8.3837e-05], device='cuda:3') 2023-03-26 22:27:29,389 INFO [finetune.py:976] (3/7) Epoch 18, batch 5300, loss[loss=0.1902, simple_loss=0.2604, pruned_loss=0.06002, over 4807.00 frames. ], tot_loss[loss=0.1836, simple_loss=0.2535, pruned_loss=0.05682, over 951768.68 frames. ], batch size: 39, lr: 3.32e-03, grad_scale: 32.0 2023-03-26 22:27:44,024 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=102691.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:28:03,083 INFO [finetune.py:976] (3/7) Epoch 18, batch 5350, loss[loss=0.156, simple_loss=0.2311, pruned_loss=0.04041, over 4811.00 frames. ], tot_loss[loss=0.1825, simple_loss=0.2529, pruned_loss=0.05604, over 952884.07 frames. ], batch size: 41, lr: 3.32e-03, grad_scale: 32.0 2023-03-26 22:28:10,284 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.6601, 3.9856, 4.2103, 4.5096, 4.3589, 4.1745, 4.7628, 1.6601], device='cuda:3'), covar=tensor([0.0797, 0.0856, 0.0746, 0.0971, 0.1264, 0.1523, 0.0700, 0.5476], device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0246, 0.0281, 0.0293, 0.0335, 0.0285, 0.0305, 0.0299], 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') 2023-03-26 22:28:12,000 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1373, 2.0895, 1.8660, 2.1952, 2.0206, 2.0582, 1.9721, 2.8819], device='cuda:3'), covar=tensor([0.3939, 0.5050, 0.3400, 0.4382, 0.4781, 0.2600, 0.4742, 0.1646], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0261, 0.0229, 0.0275, 0.0251, 0.0220, 0.0251, 0.0233], 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') 2023-03-26 22:28:19,656 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4280, 2.3824, 2.4095, 1.6423, 2.3518, 2.4693, 2.6093, 1.9748], device='cuda:3'), covar=tensor([0.0538, 0.0601, 0.0638, 0.0838, 0.0611, 0.0585, 0.0495, 0.1065], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0134, 0.0140, 0.0120, 0.0124, 0.0138, 0.0140, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 22:28:25,303 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.195e+02 1.536e+02 1.872e+02 2.228e+02 4.473e+02, threshold=3.745e+02, percent-clipped=1.0 2023-03-26 22:28:36,909 INFO [finetune.py:976] (3/7) Epoch 18, batch 5400, loss[loss=0.1665, simple_loss=0.2232, pruned_loss=0.05489, over 4717.00 frames. ], tot_loss[loss=0.1796, simple_loss=0.2494, pruned_loss=0.05494, over 953144.59 frames. ], batch size: 23, lr: 3.32e-03, grad_scale: 32.0 2023-03-26 22:29:10,821 INFO [finetune.py:976] (3/7) Epoch 18, batch 5450, loss[loss=0.1904, simple_loss=0.247, pruned_loss=0.06694, over 3977.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2457, pruned_loss=0.05393, over 952972.49 frames. ], batch size: 17, lr: 3.32e-03, grad_scale: 32.0 2023-03-26 22:29:30,978 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102851.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:29:31,447 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.551e+01 1.511e+02 1.793e+02 2.102e+02 5.113e+02, threshold=3.586e+02, percent-clipped=1.0 2023-03-26 22:29:44,409 INFO [finetune.py:976] (3/7) Epoch 18, batch 5500, loss[loss=0.161, simple_loss=0.2345, pruned_loss=0.04378, over 4895.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2434, pruned_loss=0.053, over 954137.25 frames. ], batch size: 35, lr: 3.32e-03, grad_scale: 32.0 2023-03-26 22:29:54,065 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102887.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:30:12,696 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102912.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:30:18,102 INFO [finetune.py:976] (3/7) Epoch 18, batch 5550, loss[loss=0.1549, simple_loss=0.2272, pruned_loss=0.04135, over 4729.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2446, pruned_loss=0.0533, over 954996.79 frames. ], batch size: 23, lr: 3.32e-03, grad_scale: 32.0 2023-03-26 22:30:36,192 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1640, 2.0879, 2.1024, 1.5616, 2.1262, 2.1907, 2.3258, 1.7364], device='cuda:3'), covar=tensor([0.0550, 0.0670, 0.0687, 0.0907, 0.0654, 0.0707, 0.0545, 0.1161], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0134, 0.0139, 0.0119, 0.0123, 0.0138, 0.0139, 0.0160], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 22:30:36,211 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102948.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:30:39,505 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.093e+02 1.559e+02 1.902e+02 2.213e+02 4.520e+02, threshold=3.805e+02, percent-clipped=3.0 2023-03-26 22:30:48,334 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.7940, 3.2801, 3.4821, 3.6731, 3.5587, 3.3469, 3.8526, 1.2113], device='cuda:3'), covar=tensor([0.0837, 0.0944, 0.0923, 0.1022, 0.1316, 0.1773, 0.0946, 0.5680], device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0246, 0.0281, 0.0293, 0.0335, 0.0285, 0.0305, 0.0298], 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') 2023-03-26 22:30:50,054 INFO [finetune.py:976] (3/7) Epoch 18, batch 5600, loss[loss=0.1974, simple_loss=0.2677, pruned_loss=0.0636, over 4861.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2473, pruned_loss=0.0542, over 951740.98 frames. ], batch size: 31, lr: 3.32e-03, grad_scale: 16.0 2023-03-26 22:31:42,255 INFO [finetune.py:976] (3/7) Epoch 18, batch 5650, loss[loss=0.1979, simple_loss=0.2742, pruned_loss=0.06082, over 4845.00 frames. ], tot_loss[loss=0.1805, simple_loss=0.2511, pruned_loss=0.0549, over 951598.58 frames. ], batch size: 44, lr: 3.31e-03, grad_scale: 16.0 2023-03-26 22:31:52,782 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5449, 1.0365, 0.7651, 1.3338, 2.0246, 0.6982, 1.2265, 1.2390], device='cuda:3'), covar=tensor([0.1629, 0.2387, 0.1855, 0.1341, 0.1929, 0.2101, 0.1723, 0.2312], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0095, 0.0110, 0.0092, 0.0120, 0.0094, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 22:32:09,152 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.800e+01 1.506e+02 1.835e+02 2.191e+02 3.638e+02, threshold=3.670e+02, percent-clipped=0.0 2023-03-26 22:32:19,851 INFO [finetune.py:976] (3/7) Epoch 18, batch 5700, loss[loss=0.1291, simple_loss=0.2072, pruned_loss=0.02555, over 3956.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.2474, pruned_loss=0.05448, over 931617.52 frames. ], batch size: 17, lr: 3.31e-03, grad_scale: 16.0 2023-03-26 22:32:48,079 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103098.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:32:48,576 INFO [finetune.py:976] (3/7) Epoch 19, batch 0, loss[loss=0.2075, simple_loss=0.2802, pruned_loss=0.06736, over 4898.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2802, pruned_loss=0.06736, over 4898.00 frames. ], batch size: 37, lr: 3.31e-03, grad_scale: 16.0 2023-03-26 22:32:48,576 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 22:33:03,100 INFO [finetune.py:1010] (3/7) Epoch 19, validation: loss=0.1586, simple_loss=0.2282, pruned_loss=0.04454, over 2265189.00 frames. 2023-03-26 22:33:03,100 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6434MB 2023-03-26 22:33:25,837 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103132.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:33:33,035 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-03-26 22:33:38,061 INFO [finetune.py:976] (3/7) Epoch 19, batch 50, loss[loss=0.1683, simple_loss=0.2444, pruned_loss=0.04609, over 4879.00 frames. ], tot_loss[loss=0.1829, simple_loss=0.2551, pruned_loss=0.05533, over 217614.26 frames. ], batch size: 43, lr: 3.31e-03, grad_scale: 16.0 2023-03-26 22:33:40,495 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.779e+01 1.456e+02 1.782e+02 2.150e+02 3.860e+02, threshold=3.565e+02, percent-clipped=1.0 2023-03-26 22:33:41,943 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-03-26 22:33:44,737 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103159.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:34:07,713 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103193.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:34:11,665 INFO [finetune.py:976] (3/7) Epoch 19, batch 100, loss[loss=0.1863, simple_loss=0.2517, pruned_loss=0.06046, over 4860.00 frames. ], tot_loss[loss=0.1803, simple_loss=0.2497, pruned_loss=0.05545, over 382635.24 frames. ], batch size: 31, lr: 3.31e-03, grad_scale: 16.0 2023-03-26 22:34:17,537 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103207.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:34:40,720 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103243.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:34:45,882 INFO [finetune.py:976] (3/7) Epoch 19, batch 150, loss[loss=0.1513, simple_loss=0.226, pruned_loss=0.03829, over 4819.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2443, pruned_loss=0.054, over 509968.43 frames. ], batch size: 40, lr: 3.31e-03, grad_scale: 16.0 2023-03-26 22:34:48,707 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.032e+02 1.463e+02 1.787e+02 2.269e+02 3.542e+02, threshold=3.573e+02, percent-clipped=0.0 2023-03-26 22:35:00,958 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7111, 1.4298, 2.1124, 3.3475, 2.2368, 2.4346, 1.2538, 2.7482], device='cuda:3'), covar=tensor([0.1711, 0.1634, 0.1434, 0.0795, 0.0821, 0.1488, 0.1772, 0.0565], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0115, 0.0133, 0.0163, 0.0099, 0.0134, 0.0123, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 22:35:08,285 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5847, 1.5226, 1.7968, 1.7824, 1.6471, 3.4836, 1.4494, 1.5422], device='cuda:3'), covar=tensor([0.0944, 0.1854, 0.1056, 0.0976, 0.1545, 0.0229, 0.1525, 0.1835], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0074, 0.0077, 0.0091, 0.0080, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 22:35:19,727 INFO [finetune.py:976] (3/7) Epoch 19, batch 200, loss[loss=0.1846, simple_loss=0.2478, pruned_loss=0.06068, over 4822.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2443, pruned_loss=0.05498, over 609505.19 frames. ], batch size: 41, lr: 3.31e-03, grad_scale: 16.0 2023-03-26 22:35:29,196 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7180, 1.5986, 1.5592, 1.6387, 1.1564, 3.7600, 1.4842, 2.0524], device='cuda:3'), covar=tensor([0.3231, 0.2374, 0.2170, 0.2404, 0.1821, 0.0178, 0.2485, 0.1166], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0114, 0.0120, 0.0122, 0.0113, 0.0095, 0.0095, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 22:35:30,404 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.4353, 3.8531, 4.0722, 4.2760, 4.2075, 3.9524, 4.5272, 1.4577], device='cuda:3'), covar=tensor([0.0829, 0.0876, 0.1002, 0.1179, 0.1188, 0.1834, 0.0707, 0.5485], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0246, 0.0283, 0.0294, 0.0336, 0.0286, 0.0305, 0.0298], 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') 2023-03-26 22:35:41,677 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-03-26 22:35:53,176 INFO [finetune.py:976] (3/7) Epoch 19, batch 250, loss[loss=0.1292, simple_loss=0.2066, pruned_loss=0.02593, over 4767.00 frames. ], tot_loss[loss=0.1793, simple_loss=0.2483, pruned_loss=0.05517, over 687441.29 frames. ], batch size: 28, lr: 3.31e-03, grad_scale: 16.0 2023-03-26 22:35:56,517 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.389e+01 1.572e+02 1.886e+02 2.263e+02 4.128e+02, threshold=3.772e+02, percent-clipped=1.0 2023-03-26 22:36:05,016 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103366.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 22:36:11,935 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.77 vs. limit=2.0 2023-03-26 22:36:25,724 INFO [finetune.py:976] (3/7) Epoch 19, batch 300, loss[loss=0.1448, simple_loss=0.2263, pruned_loss=0.03171, over 4926.00 frames. ], tot_loss[loss=0.1798, simple_loss=0.25, pruned_loss=0.05484, over 748316.33 frames. ], batch size: 33, lr: 3.31e-03, grad_scale: 16.0 2023-03-26 22:36:29,829 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7331, 3.4653, 3.3260, 1.6259, 3.6333, 2.6674, 0.9769, 2.4601], device='cuda:3'), covar=tensor([0.2759, 0.2141, 0.1754, 0.3314, 0.1168, 0.1060, 0.4183, 0.1467], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0175, 0.0160, 0.0129, 0.0160, 0.0123, 0.0147, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 22:37:02,161 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103427.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 22:37:21,697 INFO [finetune.py:976] (3/7) Epoch 19, batch 350, loss[loss=0.1639, simple_loss=0.2323, pruned_loss=0.04773, over 4923.00 frames. ], tot_loss[loss=0.1815, simple_loss=0.2521, pruned_loss=0.05545, over 794524.13 frames. ], batch size: 33, lr: 3.31e-03, grad_scale: 16.0 2023-03-26 22:37:28,083 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.109e+02 1.554e+02 1.898e+02 2.403e+02 5.343e+02, threshold=3.796e+02, percent-clipped=4.0 2023-03-26 22:37:29,796 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103454.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:37:57,689 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-26 22:38:00,658 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6626, 2.5578, 1.9848, 2.6368, 2.6154, 2.2545, 3.1129, 2.6760], device='cuda:3'), covar=tensor([0.1193, 0.2253, 0.3271, 0.2714, 0.2509, 0.1600, 0.2749, 0.1738], device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0187, 0.0233, 0.0252, 0.0244, 0.0202, 0.0213, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 22:38:03,041 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103488.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:38:10,178 INFO [finetune.py:976] (3/7) Epoch 19, batch 400, loss[loss=0.1593, simple_loss=0.23, pruned_loss=0.04436, over 4757.00 frames. ], tot_loss[loss=0.1819, simple_loss=0.2529, pruned_loss=0.05548, over 830812.57 frames. ], batch size: 27, lr: 3.31e-03, grad_scale: 16.0 2023-03-26 22:38:11,833 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-26 22:38:16,130 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103507.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:38:28,027 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.65 vs. limit=5.0 2023-03-26 22:38:37,761 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8073, 1.8233, 1.6111, 1.9423, 2.1922, 1.9538, 1.4928, 1.5355], device='cuda:3'), covar=tensor([0.2062, 0.1819, 0.1855, 0.1408, 0.1567, 0.1137, 0.2347, 0.1911], device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0208, 0.0213, 0.0192, 0.0241, 0.0186, 0.0215, 0.0200], 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') 2023-03-26 22:38:39,541 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103543.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:38:43,099 INFO [finetune.py:976] (3/7) Epoch 19, batch 450, loss[loss=0.1576, simple_loss=0.2236, pruned_loss=0.04586, over 3730.00 frames. ], tot_loss[loss=0.1811, simple_loss=0.2513, pruned_loss=0.05547, over 858009.07 frames. ], batch size: 16, lr: 3.31e-03, grad_scale: 16.0 2023-03-26 22:38:45,990 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.312e+01 1.502e+02 1.696e+02 2.061e+02 2.854e+02, threshold=3.392e+02, percent-clipped=0.0 2023-03-26 22:38:47,238 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=103555.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:39:17,152 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7046, 3.3512, 3.2191, 1.5191, 3.4860, 2.5430, 1.0296, 2.3603], device='cuda:3'), covar=tensor([0.2409, 0.1954, 0.1641, 0.3310, 0.1179, 0.1040, 0.3958, 0.1529], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0175, 0.0159, 0.0128, 0.0159, 0.0123, 0.0147, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 22:39:19,609 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=103591.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:39:24,374 INFO [finetune.py:976] (3/7) Epoch 19, batch 500, loss[loss=0.1757, simple_loss=0.2342, pruned_loss=0.05863, over 4942.00 frames. ], tot_loss[loss=0.1799, simple_loss=0.2493, pruned_loss=0.05521, over 880184.13 frames. ], batch size: 33, lr: 3.31e-03, grad_scale: 16.0 2023-03-26 22:39:43,300 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4240, 2.2640, 2.0070, 2.4395, 2.1409, 2.1714, 2.1867, 3.1319], device='cuda:3'), covar=tensor([0.3607, 0.4554, 0.3231, 0.4124, 0.4504, 0.2344, 0.4184, 0.1585], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0260, 0.0229, 0.0274, 0.0250, 0.0219, 0.0251, 0.0232], 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') 2023-03-26 22:39:57,690 INFO [finetune.py:976] (3/7) Epoch 19, batch 550, loss[loss=0.1667, simple_loss=0.2327, pruned_loss=0.05036, over 4861.00 frames. ], tot_loss[loss=0.178, simple_loss=0.247, pruned_loss=0.05451, over 898630.20 frames. ], batch size: 47, lr: 3.31e-03, grad_scale: 16.0 2023-03-26 22:39:57,777 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.5392, 3.8863, 4.0961, 4.3850, 4.2915, 4.0149, 4.6182, 1.4174], device='cuda:3'), covar=tensor([0.0682, 0.0906, 0.0822, 0.0907, 0.1059, 0.1490, 0.0604, 0.5449], device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0240, 0.0277, 0.0289, 0.0330, 0.0280, 0.0298, 0.0291], 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') 2023-03-26 22:40:00,582 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.093e+02 1.543e+02 1.834e+02 2.179e+02 4.966e+02, threshold=3.668e+02, percent-clipped=2.0 2023-03-26 22:40:05,912 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2813, 2.7154, 2.6117, 1.3730, 2.8272, 2.1419, 0.9224, 1.9322], device='cuda:3'), covar=tensor([0.2822, 0.2534, 0.1854, 0.3182, 0.1440, 0.1094, 0.3710, 0.1596], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0175, 0.0160, 0.0128, 0.0160, 0.0123, 0.0148, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 22:40:28,959 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8075, 1.3904, 0.8294, 1.7136, 2.0919, 1.4149, 1.7527, 1.8099], device='cuda:3'), covar=tensor([0.1366, 0.1880, 0.1853, 0.1070, 0.1904, 0.1852, 0.1213, 0.1717], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0095, 0.0110, 0.0091, 0.0119, 0.0093, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-26 22:40:29,607 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0451, 2.1021, 1.8086, 1.9153, 2.6107, 2.7665, 2.0942, 2.1602], device='cuda:3'), covar=tensor([0.0377, 0.0400, 0.0592, 0.0340, 0.0219, 0.0439, 0.0428, 0.0370], device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0106, 0.0141, 0.0109, 0.0098, 0.0106, 0.0097, 0.0109], device='cuda:3'), out_proj_covar=tensor([7.3207e-05, 8.1867e-05, 1.1083e-04, 8.3438e-05, 7.6154e-05, 7.8548e-05, 7.2842e-05, 8.3044e-05], device='cuda:3') 2023-03-26 22:40:31,342 INFO [finetune.py:976] (3/7) Epoch 19, batch 600, loss[loss=0.1606, simple_loss=0.2244, pruned_loss=0.04836, over 4708.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2464, pruned_loss=0.05457, over 909172.80 frames. ], batch size: 23, lr: 3.31e-03, grad_scale: 16.0 2023-03-26 22:40:32,654 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5391, 1.7122, 2.1740, 1.9345, 2.0483, 4.3169, 1.7817, 1.9334], device='cuda:3'), covar=tensor([0.0970, 0.1777, 0.1113, 0.0933, 0.1381, 0.0195, 0.1357, 0.1609], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0074, 0.0077, 0.0091, 0.0080, 0.0084, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 22:40:36,192 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-03-26 22:40:47,254 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103722.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 22:40:55,076 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9058, 1.4676, 1.9857, 1.9394, 1.7551, 1.7148, 1.8552, 1.7986], device='cuda:3'), covar=tensor([0.3950, 0.4028, 0.3345, 0.3528, 0.4804, 0.3697, 0.4377, 0.3155], device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0241, 0.0260, 0.0276, 0.0275, 0.0249, 0.0284, 0.0241], 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') 2023-03-26 22:40:58,704 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103740.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:41:04,043 INFO [finetune.py:976] (3/7) Epoch 19, batch 650, loss[loss=0.1935, simple_loss=0.2504, pruned_loss=0.06831, over 4886.00 frames. ], tot_loss[loss=0.1809, simple_loss=0.2499, pruned_loss=0.05598, over 919642.64 frames. ], batch size: 32, lr: 3.31e-03, grad_scale: 16.0 2023-03-26 22:41:06,467 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.131e+02 1.539e+02 1.795e+02 2.169e+02 3.837e+02, threshold=3.591e+02, percent-clipped=1.0 2023-03-26 22:41:07,217 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103754.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:41:15,353 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103765.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:41:20,066 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.87 vs. limit=5.0 2023-03-26 22:41:30,857 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103788.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:41:37,407 INFO [finetune.py:976] (3/7) Epoch 19, batch 700, loss[loss=0.1672, simple_loss=0.2414, pruned_loss=0.04652, over 4726.00 frames. ], tot_loss[loss=0.1828, simple_loss=0.2521, pruned_loss=0.05673, over 927554.13 frames. ], batch size: 59, lr: 3.31e-03, grad_scale: 16.0 2023-03-26 22:41:38,895 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103801.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:41:39,435 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=103802.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:42:02,009 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103826.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:42:03,718 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4606, 2.5013, 2.4433, 1.7344, 2.4770, 2.4893, 2.6217, 2.1072], device='cuda:3'), covar=tensor([0.0592, 0.0604, 0.0629, 0.0857, 0.0762, 0.0612, 0.0528, 0.1013], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0134, 0.0140, 0.0120, 0.0124, 0.0137, 0.0139, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 22:42:13,879 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=103836.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:42:26,321 INFO [finetune.py:976] (3/7) Epoch 19, batch 750, loss[loss=0.2413, simple_loss=0.2958, pruned_loss=0.09343, over 4813.00 frames. ], tot_loss[loss=0.1824, simple_loss=0.2517, pruned_loss=0.0565, over 932141.55 frames. ], batch size: 33, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:42:33,273 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 1.573e+02 1.871e+02 2.192e+02 5.260e+02, threshold=3.742e+02, percent-clipped=2.0 2023-03-26 22:42:40,717 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1393, 2.0472, 1.6117, 2.0840, 2.0816, 1.8118, 2.3974, 2.1990], device='cuda:3'), covar=tensor([0.1378, 0.2182, 0.2956, 0.2677, 0.2599, 0.1685, 0.3288, 0.1821], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0189, 0.0236, 0.0254, 0.0246, 0.0204, 0.0215, 0.0202], 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') 2023-03-26 22:43:03,184 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103876.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:43:05,577 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9606, 1.7510, 1.4077, 1.4199, 2.2844, 2.5542, 2.0573, 1.8467], device='cuda:3'), covar=tensor([0.0387, 0.0460, 0.0887, 0.0463, 0.0269, 0.0408, 0.0391, 0.0484], device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0107, 0.0141, 0.0109, 0.0098, 0.0107, 0.0098, 0.0109], device='cuda:3'), out_proj_covar=tensor([7.3675e-05, 8.2263e-05, 1.1118e-04, 8.3867e-05, 7.6729e-05, 7.9277e-05, 7.3135e-05, 8.3460e-05], device='cuda:3') 2023-03-26 22:43:16,807 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3978, 1.3912, 1.2584, 1.4518, 1.7198, 1.5802, 1.3903, 1.2637], device='cuda:3'), covar=tensor([0.0309, 0.0282, 0.0537, 0.0249, 0.0194, 0.0418, 0.0326, 0.0389], device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0106, 0.0141, 0.0109, 0.0098, 0.0107, 0.0098, 0.0109], device='cuda:3'), out_proj_covar=tensor([7.3538e-05, 8.2127e-05, 1.1092e-04, 8.3723e-05, 7.6606e-05, 7.9147e-05, 7.2963e-05, 8.3301e-05], device='cuda:3') 2023-03-26 22:43:22,083 INFO [finetune.py:976] (3/7) Epoch 19, batch 800, loss[loss=0.2015, simple_loss=0.2652, pruned_loss=0.06888, over 4826.00 frames. ], tot_loss[loss=0.1821, simple_loss=0.2516, pruned_loss=0.05631, over 937409.77 frames. ], batch size: 30, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:43:48,639 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103937.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 22:43:55,832 INFO [finetune.py:976] (3/7) Epoch 19, batch 850, loss[loss=0.1303, simple_loss=0.1993, pruned_loss=0.03066, over 4780.00 frames. ], tot_loss[loss=0.1786, simple_loss=0.248, pruned_loss=0.05459, over 943910.52 frames. ], batch size: 28, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:43:58,235 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.100e+02 1.499e+02 1.798e+02 2.103e+02 3.961e+02, threshold=3.597e+02, percent-clipped=1.0 2023-03-26 22:44:25,966 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103990.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:44:30,231 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103997.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:44:31,321 INFO [finetune.py:976] (3/7) Epoch 19, batch 900, loss[loss=0.1935, simple_loss=0.2625, pruned_loss=0.06219, over 4891.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2457, pruned_loss=0.05397, over 946054.54 frames. ], batch size: 35, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:44:46,755 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104022.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 22:44:46,763 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0597, 1.7877, 2.0578, 1.4975, 1.9682, 1.9733, 2.0793, 1.3427], device='cuda:3'), covar=tensor([0.0684, 0.0878, 0.0706, 0.0987, 0.0774, 0.0809, 0.0777, 0.1885], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0134, 0.0139, 0.0120, 0.0124, 0.0137, 0.0139, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 22:44:48,044 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-03-26 22:44:53,670 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1555, 1.7673, 2.1852, 2.1920, 1.9000, 1.8896, 2.0845, 2.0543], device='cuda:3'), covar=tensor([0.3973, 0.4032, 0.3036, 0.3633, 0.4734, 0.3903, 0.4932, 0.3011], device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0241, 0.0260, 0.0277, 0.0275, 0.0250, 0.0285, 0.0241], 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') 2023-03-26 22:45:05,988 INFO [finetune.py:976] (3/7) Epoch 19, batch 950, loss[loss=0.1369, simple_loss=0.225, pruned_loss=0.0244, over 4899.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2441, pruned_loss=0.05332, over 948238.87 frames. ], batch size: 43, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:45:06,108 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104049.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:45:07,318 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104051.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:45:08,387 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.542e+01 1.532e+02 1.748e+02 2.079e+02 4.067e+02, threshold=3.497e+02, percent-clipped=1.0 2023-03-26 22:45:11,535 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104058.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:45:18,740 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=104070.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 22:45:36,942 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104096.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:45:36,987 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4928, 1.3954, 1.4267, 1.4648, 0.8156, 2.9928, 1.0123, 1.4462], device='cuda:3'), covar=tensor([0.3520, 0.2651, 0.2268, 0.2506, 0.2094, 0.0254, 0.2869, 0.1409], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0120, 0.0123, 0.0114, 0.0096, 0.0096, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 22:45:38,694 INFO [finetune.py:976] (3/7) Epoch 19, batch 1000, loss[loss=0.2321, simple_loss=0.3149, pruned_loss=0.0747, over 4852.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2473, pruned_loss=0.05434, over 951649.11 frames. ], batch size: 44, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:45:45,437 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104110.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:45:52,086 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104121.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:46:12,345 INFO [finetune.py:976] (3/7) Epoch 19, batch 1050, loss[loss=0.1653, simple_loss=0.2472, pruned_loss=0.04173, over 4906.00 frames. ], tot_loss[loss=0.1793, simple_loss=0.2491, pruned_loss=0.05473, over 953062.55 frames. ], batch size: 46, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:46:12,738 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-26 22:46:14,761 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.111e+02 1.591e+02 1.940e+02 2.273e+02 3.456e+02, threshold=3.881e+02, percent-clipped=0.0 2023-03-26 22:46:53,935 INFO [finetune.py:976] (3/7) Epoch 19, batch 1100, loss[loss=0.1844, simple_loss=0.2527, pruned_loss=0.05811, over 4883.00 frames. ], tot_loss[loss=0.1805, simple_loss=0.2507, pruned_loss=0.05514, over 953802.80 frames. ], batch size: 32, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:47:16,903 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104232.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 22:47:29,940 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0298, 1.9457, 1.9498, 1.2141, 1.9289, 1.9287, 1.9904, 1.5975], device='cuda:3'), covar=tensor([0.0521, 0.0605, 0.0661, 0.0937, 0.0708, 0.0662, 0.0566, 0.1118], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0133, 0.0138, 0.0119, 0.0123, 0.0136, 0.0139, 0.0160], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 22:47:36,259 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5554, 1.4808, 1.4857, 1.5578, 1.1474, 3.3463, 1.3873, 1.8175], device='cuda:3'), covar=tensor([0.3354, 0.2620, 0.2177, 0.2408, 0.1786, 0.0174, 0.2622, 0.1214], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0120, 0.0123, 0.0114, 0.0096, 0.0096, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 22:47:39,152 INFO [finetune.py:976] (3/7) Epoch 19, batch 1150, loss[loss=0.1811, simple_loss=0.2516, pruned_loss=0.05534, over 4916.00 frames. ], tot_loss[loss=0.1816, simple_loss=0.252, pruned_loss=0.05557, over 954916.61 frames. ], batch size: 38, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:47:40,960 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0569, 1.9749, 2.8239, 4.3252, 2.8736, 2.7416, 1.3070, 3.6121], device='cuda:3'), covar=tensor([0.1786, 0.1409, 0.1258, 0.0450, 0.0771, 0.1496, 0.1866, 0.0406], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0117, 0.0133, 0.0164, 0.0099, 0.0136, 0.0124, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 22:47:47,027 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.055e+02 1.710e+02 1.992e+02 2.366e+02 4.129e+02, threshold=3.984e+02, percent-clipped=1.0 2023-03-26 22:48:25,532 INFO [finetune.py:976] (3/7) Epoch 19, batch 1200, loss[loss=0.1488, simple_loss=0.2206, pruned_loss=0.03849, over 4763.00 frames. ], tot_loss[loss=0.1807, simple_loss=0.2512, pruned_loss=0.05509, over 955885.01 frames. ], batch size: 28, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:49:05,635 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104346.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:49:07,373 INFO [finetune.py:976] (3/7) Epoch 19, batch 1250, loss[loss=0.1643, simple_loss=0.2321, pruned_loss=0.0482, over 4869.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2482, pruned_loss=0.05361, over 957133.64 frames. ], batch size: 34, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:49:09,876 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1781, 1.5917, 2.0087, 1.9987, 1.8202, 1.7829, 1.9263, 1.9081], device='cuda:3'), covar=tensor([0.4144, 0.4099, 0.3443, 0.3959, 0.5120, 0.4000, 0.4707, 0.3276], device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0242, 0.0261, 0.0278, 0.0276, 0.0251, 0.0286, 0.0242], 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') 2023-03-26 22:49:10,322 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.979e+01 1.472e+02 1.754e+02 2.218e+02 4.171e+02, threshold=3.509e+02, percent-clipped=1.0 2023-03-26 22:49:10,409 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104353.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:49:29,614 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-26 22:49:39,161 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104396.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:49:41,418 INFO [finetune.py:976] (3/7) Epoch 19, batch 1300, loss[loss=0.1804, simple_loss=0.2507, pruned_loss=0.05504, over 4773.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2451, pruned_loss=0.0529, over 955717.66 frames. ], batch size: 28, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:49:45,744 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104405.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:49:56,435 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104421.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:50:06,064 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.0540, 4.6468, 4.3420, 2.6386, 4.7555, 3.5855, 0.7215, 3.3682], device='cuda:3'), covar=tensor([0.2131, 0.2084, 0.1418, 0.2867, 0.0728, 0.0855, 0.4803, 0.1313], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0177, 0.0161, 0.0129, 0.0161, 0.0124, 0.0149, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 22:50:10,346 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=104444.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:50:14,758 INFO [finetune.py:976] (3/7) Epoch 19, batch 1350, loss[loss=0.1888, simple_loss=0.2629, pruned_loss=0.05738, over 4865.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2437, pruned_loss=0.05219, over 956068.73 frames. ], batch size: 44, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:50:17,642 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.092e+02 1.617e+02 1.931e+02 2.310e+02 3.973e+02, threshold=3.863e+02, percent-clipped=4.0 2023-03-26 22:50:29,050 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=104469.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:50:48,508 INFO [finetune.py:976] (3/7) Epoch 19, batch 1400, loss[loss=0.2032, simple_loss=0.2907, pruned_loss=0.05783, over 4848.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2476, pruned_loss=0.0541, over 954652.80 frames. ], batch size: 49, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:50:57,983 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6865, 2.7966, 2.6220, 1.9308, 2.6500, 2.7442, 2.8673, 2.2097], device='cuda:3'), covar=tensor([0.0593, 0.0555, 0.0630, 0.0884, 0.0586, 0.0695, 0.0620, 0.1038], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0133, 0.0138, 0.0118, 0.0123, 0.0136, 0.0138, 0.0159], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 22:51:10,554 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104532.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 22:51:12,481 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-26 22:51:21,267 INFO [finetune.py:976] (3/7) Epoch 19, batch 1450, loss[loss=0.2021, simple_loss=0.2684, pruned_loss=0.06794, over 4823.00 frames. ], tot_loss[loss=0.1785, simple_loss=0.2487, pruned_loss=0.05409, over 954735.68 frames. ], batch size: 33, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:51:24,645 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.038e+02 1.650e+02 1.913e+02 2.290e+02 4.485e+02, threshold=3.826e+02, percent-clipped=3.0 2023-03-26 22:51:42,313 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.3059, 2.9438, 3.0482, 3.2564, 3.0637, 2.9110, 3.3460, 0.9429], device='cuda:3'), covar=tensor([0.1119, 0.1091, 0.1112, 0.1156, 0.1836, 0.1852, 0.1214, 0.5838], device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0241, 0.0279, 0.0289, 0.0331, 0.0280, 0.0301, 0.0294], 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') 2023-03-26 22:51:42,883 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=104580.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:52:02,825 INFO [finetune.py:976] (3/7) Epoch 19, batch 1500, loss[loss=0.2123, simple_loss=0.2862, pruned_loss=0.06916, over 4844.00 frames. ], tot_loss[loss=0.1795, simple_loss=0.2501, pruned_loss=0.05446, over 955944.16 frames. ], batch size: 44, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:52:09,491 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5740, 1.4914, 2.0999, 3.3484, 2.1179, 2.3127, 0.8146, 2.7190], device='cuda:3'), covar=tensor([0.1685, 0.1485, 0.1281, 0.0600, 0.0819, 0.1686, 0.1904, 0.0497], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0115, 0.0132, 0.0162, 0.0098, 0.0135, 0.0122, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 22:52:22,433 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2377, 1.9582, 2.6433, 4.1998, 2.9260, 2.7501, 0.7556, 3.5430], device='cuda:3'), covar=tensor([0.1552, 0.1461, 0.1316, 0.0522, 0.0675, 0.1580, 0.2156, 0.0390], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0115, 0.0131, 0.0162, 0.0098, 0.0134, 0.0122, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 22:52:33,829 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104646.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:52:35,531 INFO [finetune.py:976] (3/7) Epoch 19, batch 1550, loss[loss=0.188, simple_loss=0.2558, pruned_loss=0.06014, over 4813.00 frames. ], tot_loss[loss=0.1807, simple_loss=0.2516, pruned_loss=0.05485, over 955918.71 frames. ], batch size: 40, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:52:40,198 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.878e+01 1.494e+02 1.864e+02 2.283e+02 3.386e+02, threshold=3.728e+02, percent-clipped=0.0 2023-03-26 22:52:40,298 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104653.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:53:16,283 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7768, 1.6953, 1.5128, 1.3373, 1.7731, 1.5757, 1.6054, 1.7319], device='cuda:3'), covar=tensor([0.1242, 0.1623, 0.2667, 0.2136, 0.2284, 0.1590, 0.1985, 0.1679], device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0187, 0.0234, 0.0253, 0.0245, 0.0203, 0.0213, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 22:53:26,171 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=104694.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:53:29,186 INFO [finetune.py:976] (3/7) Epoch 19, batch 1600, loss[loss=0.2114, simple_loss=0.2728, pruned_loss=0.075, over 4894.00 frames. ], tot_loss[loss=0.1787, simple_loss=0.2496, pruned_loss=0.05386, over 954291.10 frames. ], batch size: 32, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:53:35,260 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=104701.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:53:38,264 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104705.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:54:01,132 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7721, 1.6684, 1.5192, 1.8657, 2.4348, 1.9676, 1.6901, 1.4703], device='cuda:3'), covar=tensor([0.2161, 0.2029, 0.1983, 0.1621, 0.1496, 0.1244, 0.2360, 0.2046], device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0209, 0.0214, 0.0193, 0.0243, 0.0187, 0.0215, 0.0201], 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') 2023-03-26 22:54:11,339 INFO [finetune.py:976] (3/7) Epoch 19, batch 1650, loss[loss=0.1318, simple_loss=0.207, pruned_loss=0.02832, over 4830.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.247, pruned_loss=0.05309, over 955655.36 frames. ], batch size: 40, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:54:13,774 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.017e+02 1.513e+02 1.754e+02 2.121e+02 3.523e+02, threshold=3.508e+02, percent-clipped=0.0 2023-03-26 22:54:13,852 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=104753.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 22:54:36,283 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.0550, 4.8285, 4.6060, 2.8198, 4.9256, 3.8153, 0.9300, 3.5867], device='cuda:3'), covar=tensor([0.2316, 0.1884, 0.1326, 0.2776, 0.0676, 0.0767, 0.4675, 0.1289], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0177, 0.0160, 0.0128, 0.0160, 0.0123, 0.0148, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 22:54:44,698 INFO [finetune.py:976] (3/7) Epoch 19, batch 1700, loss[loss=0.1642, simple_loss=0.2492, pruned_loss=0.03961, over 4932.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2446, pruned_loss=0.05232, over 957220.44 frames. ], batch size: 33, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:54:50,350 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7527, 1.2234, 0.9019, 1.6169, 2.0579, 1.4743, 1.4778, 1.6124], device='cuda:3'), covar=tensor([0.1504, 0.2143, 0.1908, 0.1137, 0.1888, 0.1931, 0.1398, 0.1950], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0096, 0.0110, 0.0092, 0.0120, 0.0094, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-26 22:54:55,348 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.8956, 3.4400, 3.5904, 3.7638, 3.6345, 3.4717, 3.9517, 1.4019], device='cuda:3'), covar=tensor([0.0849, 0.0828, 0.0960, 0.0960, 0.1423, 0.1578, 0.0845, 0.5173], device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0241, 0.0279, 0.0290, 0.0331, 0.0282, 0.0301, 0.0295], 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') 2023-03-26 22:55:00,575 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9835, 1.6126, 2.3365, 1.4336, 2.0721, 2.2900, 1.6685, 2.4436], device='cuda:3'), covar=tensor([0.1368, 0.2211, 0.1351, 0.2011, 0.0913, 0.1350, 0.2701, 0.0789], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0204, 0.0190, 0.0189, 0.0174, 0.0213, 0.0217, 0.0200], 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') 2023-03-26 22:55:17,899 INFO [finetune.py:976] (3/7) Epoch 19, batch 1750, loss[loss=0.1787, simple_loss=0.2401, pruned_loss=0.05867, over 4893.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2466, pruned_loss=0.05341, over 958418.41 frames. ], batch size: 32, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:55:20,304 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.085e+02 1.630e+02 1.888e+02 2.368e+02 5.925e+02, threshold=3.776e+02, percent-clipped=5.0 2023-03-26 22:55:51,603 INFO [finetune.py:976] (3/7) Epoch 19, batch 1800, loss[loss=0.1549, simple_loss=0.2193, pruned_loss=0.04522, over 4429.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.248, pruned_loss=0.05348, over 958446.63 frames. ], batch size: 19, lr: 3.30e-03, grad_scale: 16.0 2023-03-26 22:56:25,140 INFO [finetune.py:976] (3/7) Epoch 19, batch 1850, loss[loss=0.1776, simple_loss=0.2465, pruned_loss=0.05435, over 4910.00 frames. ], tot_loss[loss=0.1798, simple_loss=0.2502, pruned_loss=0.05472, over 959207.32 frames. ], batch size: 37, lr: 3.30e-03, grad_scale: 32.0 2023-03-26 22:56:27,539 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.864e+01 1.560e+02 1.785e+02 2.312e+02 4.235e+02, threshold=3.569e+02, percent-clipped=1.0 2023-03-26 22:57:00,526 INFO [finetune.py:976] (3/7) Epoch 19, batch 1900, loss[loss=0.204, simple_loss=0.2574, pruned_loss=0.07532, over 4733.00 frames. ], tot_loss[loss=0.179, simple_loss=0.2496, pruned_loss=0.05421, over 958049.17 frames. ], batch size: 23, lr: 3.30e-03, grad_scale: 32.0 2023-03-26 22:57:42,259 INFO [finetune.py:976] (3/7) Epoch 19, batch 1950, loss[loss=0.1441, simple_loss=0.2251, pruned_loss=0.03153, over 4761.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.2487, pruned_loss=0.05389, over 956495.08 frames. ], batch size: 28, lr: 3.30e-03, grad_scale: 32.0 2023-03-26 22:57:44,666 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.372e+01 1.430e+02 1.759e+02 2.099e+02 5.293e+02, threshold=3.517e+02, percent-clipped=3.0 2023-03-26 22:58:31,188 INFO [finetune.py:976] (3/7) Epoch 19, batch 2000, loss[loss=0.1743, simple_loss=0.245, pruned_loss=0.05177, over 4889.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2461, pruned_loss=0.05331, over 956182.36 frames. ], batch size: 35, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 22:58:57,796 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-26 22:59:17,045 INFO [finetune.py:976] (3/7) Epoch 19, batch 2050, loss[loss=0.1619, simple_loss=0.2254, pruned_loss=0.04927, over 4827.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2442, pruned_loss=0.05303, over 956676.22 frames. ], batch size: 30, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 22:59:19,896 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.050e+02 1.557e+02 1.803e+02 2.317e+02 4.729e+02, threshold=3.605e+02, percent-clipped=3.0 2023-03-26 22:59:50,002 INFO [finetune.py:976] (3/7) Epoch 19, batch 2100, loss[loss=0.2693, simple_loss=0.3186, pruned_loss=0.1099, over 4354.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2457, pruned_loss=0.05446, over 955142.18 frames. ], batch size: 65, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:00:23,799 INFO [finetune.py:976] (3/7) Epoch 19, batch 2150, loss[loss=0.2715, simple_loss=0.3322, pruned_loss=0.1054, over 4911.00 frames. ], tot_loss[loss=0.1816, simple_loss=0.2507, pruned_loss=0.05624, over 955285.85 frames. ], batch size: 42, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:00:26,648 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.009e+02 1.573e+02 1.937e+02 2.406e+02 5.182e+02, threshold=3.875e+02, percent-clipped=4.0 2023-03-26 23:00:27,442 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-26 23:00:38,774 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1168, 1.5891, 0.9639, 1.9917, 2.2534, 1.8576, 1.7608, 1.8480], device='cuda:3'), covar=tensor([0.1289, 0.1846, 0.2121, 0.0975, 0.1938, 0.2190, 0.1253, 0.1723], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0095, 0.0110, 0.0091, 0.0119, 0.0093, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 23:00:57,387 INFO [finetune.py:976] (3/7) Epoch 19, batch 2200, loss[loss=0.1569, simple_loss=0.234, pruned_loss=0.03989, over 4791.00 frames. ], tot_loss[loss=0.1837, simple_loss=0.2534, pruned_loss=0.05702, over 955604.11 frames. ], batch size: 26, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:01:30,610 INFO [finetune.py:976] (3/7) Epoch 19, batch 2250, loss[loss=0.2007, simple_loss=0.2726, pruned_loss=0.06442, over 4886.00 frames. ], tot_loss[loss=0.1851, simple_loss=0.2549, pruned_loss=0.0577, over 955931.76 frames. ], batch size: 43, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:01:33,460 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.177e+02 1.629e+02 1.918e+02 2.372e+02 6.301e+02, threshold=3.835e+02, percent-clipped=3.0 2023-03-26 23:01:56,210 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105388.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:02:03,217 INFO [finetune.py:976] (3/7) Epoch 19, batch 2300, loss[loss=0.1286, simple_loss=0.2052, pruned_loss=0.02603, over 4762.00 frames. ], tot_loss[loss=0.1836, simple_loss=0.2541, pruned_loss=0.05659, over 955905.51 frames. ], batch size: 26, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:02:45,812 INFO [finetune.py:976] (3/7) Epoch 19, batch 2350, loss[loss=0.1616, simple_loss=0.2225, pruned_loss=0.0504, over 4756.00 frames. ], tot_loss[loss=0.1808, simple_loss=0.2512, pruned_loss=0.05526, over 954438.69 frames. ], batch size: 27, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:02:45,946 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105449.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 23:02:48,224 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.033e+02 1.493e+02 1.723e+02 2.054e+02 4.367e+02, threshold=3.447e+02, percent-clipped=1.0 2023-03-26 23:03:10,221 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105486.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:03:19,408 INFO [finetune.py:976] (3/7) Epoch 19, batch 2400, loss[loss=0.1792, simple_loss=0.2507, pruned_loss=0.05386, over 4890.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.248, pruned_loss=0.05408, over 955860.46 frames. ], batch size: 32, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:03:23,346 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105502.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:03:46,602 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8800, 1.7788, 1.6118, 1.4271, 1.8942, 1.6729, 1.7390, 1.8445], device='cuda:3'), covar=tensor([0.1360, 0.1793, 0.2869, 0.2282, 0.2436, 0.1589, 0.2413, 0.1778], device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0188, 0.0236, 0.0253, 0.0247, 0.0203, 0.0215, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 23:04:14,403 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105547.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:04:15,984 INFO [finetune.py:976] (3/7) Epoch 19, batch 2450, loss[loss=0.1303, simple_loss=0.1972, pruned_loss=0.03174, over 4763.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2446, pruned_loss=0.05291, over 956147.57 frames. ], batch size: 26, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:04:18,404 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.415e+01 1.457e+02 1.742e+02 2.171e+02 6.143e+02, threshold=3.484e+02, percent-clipped=3.0 2023-03-26 23:04:22,674 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8040, 1.3944, 0.8686, 1.6491, 2.1007, 1.3961, 1.6362, 1.6609], device='cuda:3'), covar=tensor([0.1450, 0.2065, 0.2032, 0.1174, 0.1985, 0.1976, 0.1446, 0.1932], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0094, 0.0110, 0.0091, 0.0118, 0.0092, 0.0098, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 23:04:25,634 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105563.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:04:28,743 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-26 23:04:49,912 INFO [finetune.py:976] (3/7) Epoch 19, batch 2500, loss[loss=0.1997, simple_loss=0.2639, pruned_loss=0.06779, over 4802.00 frames. ], tot_loss[loss=0.1783, simple_loss=0.2476, pruned_loss=0.05454, over 954062.08 frames. ], batch size: 29, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:04:52,016 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7480, 1.6413, 1.4542, 1.3892, 1.8400, 1.5419, 1.8461, 1.7510], device='cuda:3'), covar=tensor([0.1462, 0.1939, 0.3040, 0.2607, 0.2611, 0.1748, 0.2958, 0.1862], device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0188, 0.0236, 0.0254, 0.0247, 0.0203, 0.0216, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 23:05:06,246 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.1395, 1.2842, 1.3881, 0.7273, 1.2787, 1.5650, 1.6160, 1.2620], device='cuda:3'), covar=tensor([0.0941, 0.0596, 0.0479, 0.0481, 0.0475, 0.0554, 0.0337, 0.0641], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0152, 0.0126, 0.0127, 0.0133, 0.0130, 0.0143, 0.0149], device='cuda:3'), out_proj_covar=tensor([9.1987e-05, 1.1045e-04, 9.0062e-05, 9.0342e-05, 9.3511e-05, 9.3213e-05, 1.0254e-04, 1.0727e-04], device='cuda:3') 2023-03-26 23:05:23,461 INFO [finetune.py:976] (3/7) Epoch 19, batch 2550, loss[loss=0.1748, simple_loss=0.2503, pruned_loss=0.04965, over 4803.00 frames. ], tot_loss[loss=0.18, simple_loss=0.2499, pruned_loss=0.05502, over 956005.03 frames. ], batch size: 40, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:05:26,383 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.162e+02 1.575e+02 1.924e+02 2.412e+02 4.379e+02, threshold=3.848e+02, percent-clipped=4.0 2023-03-26 23:05:31,642 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-26 23:05:56,905 INFO [finetune.py:976] (3/7) Epoch 19, batch 2600, loss[loss=0.1838, simple_loss=0.2603, pruned_loss=0.05367, over 4864.00 frames. ], tot_loss[loss=0.1813, simple_loss=0.2521, pruned_loss=0.05528, over 957663.74 frames. ], batch size: 31, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:06:17,244 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.59 vs. limit=2.0 2023-03-26 23:06:26,488 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7231, 3.9992, 3.7146, 1.8592, 4.1970, 3.0964, 0.7520, 2.9013], device='cuda:3'), covar=tensor([0.2429, 0.1799, 0.1439, 0.3452, 0.0977, 0.0969, 0.4780, 0.1341], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0178, 0.0161, 0.0130, 0.0161, 0.0123, 0.0147, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 23:06:27,125 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105744.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 23:06:30,086 INFO [finetune.py:976] (3/7) Epoch 19, batch 2650, loss[loss=0.1752, simple_loss=0.2448, pruned_loss=0.05277, over 4752.00 frames. ], tot_loss[loss=0.1814, simple_loss=0.2522, pruned_loss=0.05529, over 956521.92 frames. ], batch size: 54, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:06:32,907 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.372e+01 1.545e+02 1.903e+02 2.181e+02 3.189e+02, threshold=3.806e+02, percent-clipped=0.0 2023-03-26 23:06:59,550 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105792.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:07:03,678 INFO [finetune.py:976] (3/7) Epoch 19, batch 2700, loss[loss=0.1499, simple_loss=0.223, pruned_loss=0.03839, over 4824.00 frames. ], tot_loss[loss=0.1797, simple_loss=0.2506, pruned_loss=0.05442, over 956975.74 frames. ], batch size: 30, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:07:03,794 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=105799.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:07:05,132 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5214, 1.4185, 1.1409, 1.3471, 1.7273, 1.7637, 1.5306, 1.3260], device='cuda:3'), covar=tensor([0.0333, 0.0329, 0.0833, 0.0363, 0.0245, 0.0459, 0.0364, 0.0407], device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0107, 0.0143, 0.0111, 0.0100, 0.0110, 0.0100, 0.0111], device='cuda:3'), out_proj_covar=tensor([7.4556e-05, 8.2727e-05, 1.1270e-04, 8.4859e-05, 7.7622e-05, 8.1366e-05, 7.4922e-05, 8.4701e-05], device='cuda:3') 2023-03-26 23:07:32,862 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2555, 1.3003, 1.5614, 1.0111, 1.3927, 1.4794, 1.3068, 1.6296], device='cuda:3'), covar=tensor([0.1360, 0.2436, 0.1387, 0.1656, 0.0969, 0.1303, 0.3040, 0.0877], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0205, 0.0191, 0.0188, 0.0174, 0.0214, 0.0217, 0.0201], 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') 2023-03-26 23:07:33,450 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105842.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:07:37,642 INFO [finetune.py:976] (3/7) Epoch 19, batch 2750, loss[loss=0.1803, simple_loss=0.2455, pruned_loss=0.05758, over 4744.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.2486, pruned_loss=0.05411, over 956461.85 frames. ], batch size: 54, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:07:40,095 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.803e+01 1.395e+02 1.671e+02 1.966e+02 3.086e+02, threshold=3.343e+02, percent-clipped=0.0 2023-03-26 23:07:40,224 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105853.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:07:43,750 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=105858.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:07:45,039 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=105860.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:07:57,402 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.4251, 1.5834, 1.6315, 0.8901, 1.7348, 1.8848, 1.9241, 1.4341], device='cuda:3'), covar=tensor([0.0899, 0.0645, 0.0544, 0.0590, 0.0442, 0.0650, 0.0349, 0.0746], device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0153, 0.0126, 0.0127, 0.0133, 0.0131, 0.0143, 0.0149], device='cuda:3'), out_proj_covar=tensor([9.2089e-05, 1.1099e-04, 9.0297e-05, 9.0301e-05, 9.4075e-05, 9.3566e-05, 1.0298e-04, 1.0719e-04], device='cuda:3') 2023-03-26 23:08:14,885 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0 2023-03-26 23:08:22,330 INFO [finetune.py:976] (3/7) Epoch 19, batch 2800, loss[loss=0.1378, simple_loss=0.2033, pruned_loss=0.03616, over 4723.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2449, pruned_loss=0.0526, over 957600.08 frames. ], batch size: 23, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:08:24,291 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9186, 1.6806, 1.5402, 1.3176, 1.6445, 1.6866, 1.6589, 2.2313], device='cuda:3'), covar=tensor([0.3560, 0.3759, 0.3079, 0.3587, 0.3790, 0.2219, 0.3473, 0.1777], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0260, 0.0229, 0.0275, 0.0251, 0.0220, 0.0251, 0.0231], 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') 2023-03-26 23:09:03,614 INFO [finetune.py:976] (3/7) Epoch 19, batch 2850, loss[loss=0.1935, simple_loss=0.2517, pruned_loss=0.06759, over 4820.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2433, pruned_loss=0.05217, over 955642.27 frames. ], batch size: 39, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:09:10,695 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.025e+01 1.444e+02 1.775e+02 2.176e+02 4.047e+02, threshold=3.549e+02, percent-clipped=5.0 2023-03-26 23:09:13,909 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1411, 2.0794, 1.7773, 2.1325, 1.9794, 2.0141, 2.0051, 2.8041], device='cuda:3'), covar=tensor([0.3744, 0.5280, 0.3422, 0.4231, 0.4606, 0.2569, 0.4434, 0.1720], device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0259, 0.0229, 0.0274, 0.0251, 0.0220, 0.0251, 0.0231], 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') 2023-03-26 23:09:36,416 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.67 vs. limit=5.0 2023-03-26 23:09:49,339 INFO [finetune.py:976] (3/7) Epoch 19, batch 2900, loss[loss=0.2145, simple_loss=0.2859, pruned_loss=0.07155, over 4805.00 frames. ], tot_loss[loss=0.177, simple_loss=0.247, pruned_loss=0.05353, over 954629.10 frames. ], batch size: 45, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:10:21,005 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106044.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:10:24,381 INFO [finetune.py:976] (3/7) Epoch 19, batch 2950, loss[loss=0.17, simple_loss=0.2299, pruned_loss=0.05503, over 4736.00 frames. ], tot_loss[loss=0.1794, simple_loss=0.25, pruned_loss=0.05439, over 954711.99 frames. ], batch size: 23, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:10:27,319 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.008e+02 1.599e+02 1.865e+02 2.251e+02 4.962e+02, threshold=3.729e+02, percent-clipped=1.0 2023-03-26 23:10:32,860 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0887, 2.0000, 1.6725, 2.0184, 2.0697, 1.7947, 2.4048, 2.1017], device='cuda:3'), covar=tensor([0.1448, 0.2075, 0.3090, 0.2607, 0.2702, 0.1834, 0.2991, 0.1750], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0189, 0.0238, 0.0256, 0.0249, 0.0205, 0.0217, 0.0203], 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') 2023-03-26 23:10:35,697 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.58 vs. limit=5.0 2023-03-26 23:10:43,289 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106077.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:10:53,299 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=106092.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:10:57,528 INFO [finetune.py:976] (3/7) Epoch 19, batch 3000, loss[loss=0.183, simple_loss=0.2549, pruned_loss=0.05555, over 4927.00 frames. ], tot_loss[loss=0.1821, simple_loss=0.2527, pruned_loss=0.05576, over 953788.22 frames. ], batch size: 33, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:10:57,528 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 23:11:02,667 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3422, 2.1608, 1.5985, 0.6194, 1.7300, 1.9950, 1.8426, 1.9847], device='cuda:3'), covar=tensor([0.0938, 0.0724, 0.1454, 0.1792, 0.1272, 0.2215, 0.2081, 0.0706], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0192, 0.0199, 0.0182, 0.0210, 0.0207, 0.0222, 0.0196], 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') 2023-03-26 23:11:08,360 INFO [finetune.py:1010] (3/7) Epoch 19, validation: loss=0.1576, simple_loss=0.2259, pruned_loss=0.04462, over 2265189.00 frames. 2023-03-26 23:11:08,360 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6434MB 2023-03-26 23:11:43,382 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106138.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:11:46,255 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106142.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:11:50,369 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106148.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:11:50,870 INFO [finetune.py:976] (3/7) Epoch 19, batch 3050, loss[loss=0.2192, simple_loss=0.2882, pruned_loss=0.07514, over 4908.00 frames. ], tot_loss[loss=0.1836, simple_loss=0.2545, pruned_loss=0.05632, over 953848.28 frames. ], batch size: 37, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:11:53,808 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.036e+02 1.577e+02 1.927e+02 2.196e+02 3.458e+02, threshold=3.854e+02, percent-clipped=0.0 2023-03-26 23:11:55,078 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106155.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:11:56,160 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5942, 1.5863, 2.1475, 3.5064, 2.2767, 2.3637, 1.0774, 2.8366], device='cuda:3'), covar=tensor([0.1796, 0.1363, 0.1214, 0.0507, 0.0777, 0.1347, 0.1819, 0.0471], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0116, 0.0133, 0.0163, 0.0099, 0.0135, 0.0123, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 23:11:57,424 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106158.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:12:18,599 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=106190.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:12:24,007 INFO [finetune.py:976] (3/7) Epoch 19, batch 3100, loss[loss=0.1544, simple_loss=0.2238, pruned_loss=0.04246, over 4860.00 frames. ], tot_loss[loss=0.1816, simple_loss=0.2521, pruned_loss=0.05554, over 954100.80 frames. ], batch size: 34, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:12:29,244 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=106206.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:12:57,684 INFO [finetune.py:976] (3/7) Epoch 19, batch 3150, loss[loss=0.1766, simple_loss=0.2406, pruned_loss=0.05626, over 4899.00 frames. ], tot_loss[loss=0.1785, simple_loss=0.2484, pruned_loss=0.05437, over 956338.14 frames. ], batch size: 35, lr: 3.29e-03, grad_scale: 32.0 2023-03-26 23:13:00,115 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.828e+01 1.675e+02 1.879e+02 2.192e+02 3.916e+02, threshold=3.758e+02, percent-clipped=1.0 2023-03-26 23:13:20,404 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106268.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:13:41,053 INFO [finetune.py:976] (3/7) Epoch 19, batch 3200, loss[loss=0.1647, simple_loss=0.2317, pruned_loss=0.04888, over 4868.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2448, pruned_loss=0.05321, over 956979.34 frames. ], batch size: 31, lr: 3.28e-03, grad_scale: 32.0 2023-03-26 23:13:58,436 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3065, 1.5254, 0.7383, 2.0137, 2.5023, 1.6010, 1.6795, 1.9576], device='cuda:3'), covar=tensor([0.1260, 0.1949, 0.2051, 0.1055, 0.1718, 0.2167, 0.1438, 0.1805], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0094, 0.0110, 0.0091, 0.0119, 0.0092, 0.0098, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 23:14:00,220 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9954, 1.9810, 1.9062, 2.3240, 2.4148, 2.2660, 1.8008, 1.6962], device='cuda:3'), covar=tensor([0.2049, 0.1762, 0.1710, 0.1357, 0.1575, 0.0962, 0.2109, 0.1943], device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0209, 0.0211, 0.0192, 0.0241, 0.0186, 0.0214, 0.0201], 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') 2023-03-26 23:14:01,443 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106329.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:14:05,722 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106336.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:14:16,902 INFO [finetune.py:976] (3/7) Epoch 19, batch 3250, loss[loss=0.1688, simple_loss=0.2513, pruned_loss=0.04309, over 4738.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2444, pruned_loss=0.05238, over 956547.90 frames. ], batch size: 59, lr: 3.28e-03, grad_scale: 32.0 2023-03-26 23:14:24,769 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.028e+02 1.520e+02 1.839e+02 2.222e+02 4.428e+02, threshold=3.677e+02, percent-clipped=2.0 2023-03-26 23:15:08,375 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106397.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:15:10,069 INFO [finetune.py:976] (3/7) Epoch 19, batch 3300, loss[loss=0.206, simple_loss=0.305, pruned_loss=0.05347, over 4801.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2482, pruned_loss=0.05341, over 955670.84 frames. ], batch size: 45, lr: 3.28e-03, grad_scale: 32.0 2023-03-26 23:15:41,459 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106433.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:15:51,017 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106448.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:15:51,553 INFO [finetune.py:976] (3/7) Epoch 19, batch 3350, loss[loss=0.1699, simple_loss=0.2313, pruned_loss=0.05422, over 4716.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.2492, pruned_loss=0.05379, over 953643.48 frames. ], batch size: 23, lr: 3.28e-03, grad_scale: 32.0 2023-03-26 23:15:54,457 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.049e+02 1.616e+02 1.883e+02 2.222e+02 4.657e+02, threshold=3.766e+02, percent-clipped=2.0 2023-03-26 23:15:55,182 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106454.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:15:55,757 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106455.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:16:31,242 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=106496.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:16:33,551 INFO [finetune.py:976] (3/7) Epoch 19, batch 3400, loss[loss=0.2362, simple_loss=0.3078, pruned_loss=0.08224, over 4912.00 frames. ], tot_loss[loss=0.1804, simple_loss=0.2513, pruned_loss=0.05481, over 952314.18 frames. ], batch size: 38, lr: 3.28e-03, grad_scale: 32.0 2023-03-26 23:16:36,073 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=106503.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:16:44,459 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106515.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:16:59,091 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5825, 3.3938, 3.3459, 1.6658, 3.6547, 2.7398, 1.1197, 2.5362], device='cuda:3'), covar=tensor([0.2943, 0.2117, 0.1504, 0.3227, 0.0993, 0.0976, 0.3902, 0.1333], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0177, 0.0160, 0.0129, 0.0160, 0.0122, 0.0146, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 23:17:04,526 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8915, 1.2278, 1.9250, 1.8684, 1.7041, 1.6528, 1.8163, 1.8379], device='cuda:3'), covar=tensor([0.3604, 0.3631, 0.2780, 0.3176, 0.4172, 0.3330, 0.3888, 0.2795], device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0241, 0.0261, 0.0279, 0.0277, 0.0251, 0.0286, 0.0243], 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') 2023-03-26 23:17:06,744 INFO [finetune.py:976] (3/7) Epoch 19, batch 3450, loss[loss=0.1813, simple_loss=0.2472, pruned_loss=0.05772, over 4816.00 frames. ], tot_loss[loss=0.1803, simple_loss=0.2511, pruned_loss=0.05475, over 952681.89 frames. ], batch size: 25, lr: 3.28e-03, grad_scale: 32.0 2023-03-26 23:17:09,626 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.792e+01 1.525e+02 1.781e+02 2.060e+02 3.433e+02, threshold=3.562e+02, percent-clipped=0.0 2023-03-26 23:17:14,659 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-03-26 23:17:40,388 INFO [finetune.py:976] (3/7) Epoch 19, batch 3500, loss[loss=0.1608, simple_loss=0.2365, pruned_loss=0.04256, over 4797.00 frames. ], tot_loss[loss=0.1798, simple_loss=0.2496, pruned_loss=0.05501, over 954678.80 frames. ], batch size: 29, lr: 3.28e-03, grad_scale: 32.0 2023-03-26 23:17:41,296 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2514, 2.1329, 1.7670, 2.1606, 2.1987, 1.9360, 2.4573, 2.2514], device='cuda:3'), covar=tensor([0.1355, 0.2001, 0.2871, 0.2454, 0.2397, 0.1691, 0.2836, 0.1723], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0187, 0.0234, 0.0252, 0.0246, 0.0202, 0.0215, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 23:17:57,682 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106624.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:18:14,143 INFO [finetune.py:976] (3/7) Epoch 19, batch 3550, loss[loss=0.1787, simple_loss=0.2474, pruned_loss=0.05503, over 4809.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2471, pruned_loss=0.05437, over 954670.54 frames. ], batch size: 45, lr: 3.28e-03, grad_scale: 32.0 2023-03-26 23:18:16,540 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.026e+02 1.567e+02 1.861e+02 2.307e+02 3.604e+02, threshold=3.722e+02, percent-clipped=2.0 2023-03-26 23:18:23,831 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.89 vs. limit=5.0 2023-03-26 23:18:31,074 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9499, 1.8734, 1.6985, 2.0935, 2.7039, 2.1471, 1.8676, 1.5632], device='cuda:3'), covar=tensor([0.2273, 0.2003, 0.1938, 0.1646, 0.1513, 0.1136, 0.2201, 0.1919], device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0210, 0.0212, 0.0193, 0.0242, 0.0188, 0.0215, 0.0202], 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') 2023-03-26 23:18:50,806 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1478, 1.8556, 1.9736, 0.9689, 2.3012, 2.3835, 2.1013, 1.7762], device='cuda:3'), covar=tensor([0.0947, 0.0823, 0.0493, 0.0717, 0.0471, 0.0794, 0.0496, 0.0764], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0150, 0.0124, 0.0124, 0.0130, 0.0128, 0.0140, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.0505e-05, 1.0859e-04, 8.8717e-05, 8.8009e-05, 9.1508e-05, 9.1716e-05, 1.0065e-04, 1.0551e-04], device='cuda:3') 2023-03-26 23:18:51,973 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106692.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:18:56,155 INFO [finetune.py:976] (3/7) Epoch 19, batch 3600, loss[loss=0.2596, simple_loss=0.2989, pruned_loss=0.1101, over 4127.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2449, pruned_loss=0.05368, over 954706.92 frames. ], batch size: 65, lr: 3.28e-03, grad_scale: 32.0 2023-03-26 23:19:19,202 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106733.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:19:29,856 INFO [finetune.py:976] (3/7) Epoch 19, batch 3650, loss[loss=0.1986, simple_loss=0.277, pruned_loss=0.06006, over 4865.00 frames. ], tot_loss[loss=0.1788, simple_loss=0.2479, pruned_loss=0.05481, over 954306.58 frames. ], batch size: 44, lr: 3.28e-03, grad_scale: 32.0 2023-03-26 23:19:34,983 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.178e+02 1.609e+02 2.013e+02 2.438e+02 4.457e+02, threshold=4.025e+02, percent-clipped=1.0 2023-03-26 23:19:47,788 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6956, 1.5114, 1.8364, 1.2525, 1.8163, 1.8697, 1.4186, 1.9428], device='cuda:3'), covar=tensor([0.0937, 0.1853, 0.1336, 0.1556, 0.0654, 0.1257, 0.2580, 0.0697], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0204, 0.0190, 0.0188, 0.0174, 0.0212, 0.0217, 0.0199], 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') 2023-03-26 23:20:03,526 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=106781.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:20:23,369 INFO [finetune.py:976] (3/7) Epoch 19, batch 3700, loss[loss=0.1621, simple_loss=0.2361, pruned_loss=0.04402, over 4936.00 frames. ], tot_loss[loss=0.18, simple_loss=0.2502, pruned_loss=0.05489, over 952565.61 frames. ], batch size: 33, lr: 3.28e-03, grad_scale: 32.0 2023-03-26 23:20:32,840 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106810.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:20:59,510 INFO [finetune.py:976] (3/7) Epoch 19, batch 3750, loss[loss=0.193, simple_loss=0.2677, pruned_loss=0.05911, over 4812.00 frames. ], tot_loss[loss=0.1811, simple_loss=0.2512, pruned_loss=0.05552, over 952466.08 frames. ], batch size: 39, lr: 3.28e-03, grad_scale: 32.0 2023-03-26 23:21:06,526 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.083e+02 1.604e+02 1.833e+02 2.350e+02 4.465e+02, threshold=3.666e+02, percent-clipped=2.0 2023-03-26 23:21:10,852 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3635, 1.2300, 1.1573, 1.2744, 1.5430, 1.4751, 1.2777, 1.1433], device='cuda:3'), covar=tensor([0.0369, 0.0379, 0.0763, 0.0397, 0.0282, 0.0515, 0.0460, 0.0482], device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0108, 0.0145, 0.0112, 0.0100, 0.0111, 0.0100, 0.0112], device='cuda:3'), out_proj_covar=tensor([7.5636e-05, 8.3124e-05, 1.1395e-04, 8.5939e-05, 7.8099e-05, 8.2224e-05, 7.4869e-05, 8.5458e-05], device='cuda:3') 2023-03-26 23:21:29,678 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-03-26 23:21:48,086 INFO [finetune.py:976] (3/7) Epoch 19, batch 3800, loss[loss=0.1962, simple_loss=0.2602, pruned_loss=0.06614, over 4812.00 frames. ], tot_loss[loss=0.1814, simple_loss=0.2514, pruned_loss=0.05566, over 950872.45 frames. ], batch size: 25, lr: 3.28e-03, grad_scale: 32.0 2023-03-26 23:22:07,680 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106924.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:22:19,082 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-03-26 23:22:24,703 INFO [finetune.py:976] (3/7) Epoch 19, batch 3850, loss[loss=0.1676, simple_loss=0.2315, pruned_loss=0.05186, over 4895.00 frames. ], tot_loss[loss=0.1802, simple_loss=0.2501, pruned_loss=0.05517, over 951341.50 frames. ], batch size: 35, lr: 3.28e-03, grad_scale: 64.0 2023-03-26 23:22:27,157 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.119e+02 1.623e+02 1.818e+02 2.255e+02 6.115e+02, threshold=3.637e+02, percent-clipped=1.0 2023-03-26 23:22:39,182 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=106972.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:22:52,754 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106992.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:22:57,319 INFO [finetune.py:976] (3/7) Epoch 19, batch 3900, loss[loss=0.2222, simple_loss=0.289, pruned_loss=0.07773, over 4925.00 frames. ], tot_loss[loss=0.1787, simple_loss=0.2477, pruned_loss=0.05487, over 950621.82 frames. ], batch size: 38, lr: 3.28e-03, grad_scale: 64.0 2023-03-26 23:23:23,589 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3252, 2.2380, 2.0667, 2.3988, 2.1746, 2.2270, 2.1030, 2.9846], device='cuda:3'), covar=tensor([0.3554, 0.4549, 0.3184, 0.4143, 0.4354, 0.2469, 0.4594, 0.1529], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0260, 0.0229, 0.0275, 0.0251, 0.0221, 0.0251, 0.0232], 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') 2023-03-26 23:23:24,074 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=107040.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:23:25,437 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-03-26 23:23:29,944 INFO [finetune.py:976] (3/7) Epoch 19, batch 3950, loss[loss=0.1553, simple_loss=0.2206, pruned_loss=0.04498, over 4804.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2459, pruned_loss=0.05456, over 950199.66 frames. ], batch size: 26, lr: 3.28e-03, grad_scale: 64.0 2023-03-26 23:23:35,026 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.301e+01 1.519e+02 1.802e+02 2.250e+02 5.271e+02, threshold=3.605e+02, percent-clipped=1.0 2023-03-26 23:24:07,127 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5315, 2.4421, 2.0188, 2.7591, 2.5236, 2.3181, 3.0543, 2.6705], device='cuda:3'), covar=tensor([0.1285, 0.2510, 0.3020, 0.2468, 0.2624, 0.1521, 0.3513, 0.1723], device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0188, 0.0236, 0.0253, 0.0248, 0.0203, 0.0215, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 23:24:12,974 INFO [finetune.py:976] (3/7) Epoch 19, batch 4000, loss[loss=0.1557, simple_loss=0.235, pruned_loss=0.03819, over 4895.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2459, pruned_loss=0.05497, over 949283.06 frames. ], batch size: 32, lr: 3.28e-03, grad_scale: 64.0 2023-03-26 23:24:21,378 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107110.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:24:21,451 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2143, 1.8780, 2.2027, 2.2264, 1.9398, 1.8815, 2.1410, 2.0002], device='cuda:3'), covar=tensor([0.3937, 0.3817, 0.3095, 0.3601, 0.4647, 0.3782, 0.4613, 0.2958], device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0239, 0.0259, 0.0277, 0.0274, 0.0250, 0.0284, 0.0241], 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') 2023-03-26 23:24:41,095 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107140.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 23:24:46,976 INFO [finetune.py:976] (3/7) Epoch 19, batch 4050, loss[loss=0.1451, simple_loss=0.209, pruned_loss=0.04066, over 4741.00 frames. ], tot_loss[loss=0.1801, simple_loss=0.2488, pruned_loss=0.05565, over 949545.90 frames. ], batch size: 23, lr: 3.28e-03, grad_scale: 32.0 2023-03-26 23:24:48,814 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107152.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:24:49,870 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.013e+02 1.579e+02 1.895e+02 2.231e+02 3.900e+02, threshold=3.790e+02, percent-clipped=1.0 2023-03-26 23:24:52,893 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=107158.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:24:56,938 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1199, 1.9765, 1.5389, 0.5966, 1.5921, 1.6907, 1.5794, 1.7346], device='cuda:3'), covar=tensor([0.0910, 0.0800, 0.1566, 0.2080, 0.1416, 0.2455, 0.2500, 0.0939], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0192, 0.0199, 0.0181, 0.0209, 0.0207, 0.0222, 0.0196], 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') 2023-03-26 23:25:40,275 INFO [finetune.py:976] (3/7) Epoch 19, batch 4100, loss[loss=0.1443, simple_loss=0.2217, pruned_loss=0.03349, over 4859.00 frames. ], tot_loss[loss=0.1832, simple_loss=0.2532, pruned_loss=0.05664, over 951107.93 frames. ], batch size: 44, lr: 3.28e-03, grad_scale: 32.0 2023-03-26 23:25:41,785 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107201.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 23:25:50,476 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107213.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 23:26:13,432 INFO [finetune.py:976] (3/7) Epoch 19, batch 4150, loss[loss=0.2221, simple_loss=0.2814, pruned_loss=0.08139, over 4846.00 frames. ], tot_loss[loss=0.1841, simple_loss=0.2543, pruned_loss=0.05691, over 950440.09 frames. ], batch size: 31, lr: 3.28e-03, grad_scale: 32.0 2023-03-26 23:26:21,813 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.077e+02 1.570e+02 1.970e+02 2.461e+02 5.293e+02, threshold=3.939e+02, percent-clipped=1.0 2023-03-26 23:26:23,787 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1717, 1.2000, 1.4332, 0.9742, 1.1882, 1.3535, 1.1920, 1.4308], device='cuda:3'), covar=tensor([0.0928, 0.1650, 0.1010, 0.1189, 0.0766, 0.0918, 0.2379, 0.0663], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0205, 0.0191, 0.0189, 0.0174, 0.0213, 0.0218, 0.0200], 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') 2023-03-26 23:26:56,741 INFO [finetune.py:976] (3/7) Epoch 19, batch 4200, loss[loss=0.1677, simple_loss=0.2351, pruned_loss=0.05016, over 4763.00 frames. ], tot_loss[loss=0.1818, simple_loss=0.2525, pruned_loss=0.05549, over 948939.04 frames. ], batch size: 28, lr: 3.28e-03, grad_scale: 32.0 2023-03-26 23:27:15,672 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6050, 1.5130, 1.9438, 1.1905, 1.7466, 1.8792, 1.4749, 2.0585], device='cuda:3'), covar=tensor([0.1209, 0.2244, 0.1430, 0.1819, 0.0898, 0.1331, 0.2743, 0.0807], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0204, 0.0190, 0.0189, 0.0173, 0.0212, 0.0216, 0.0200], 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') 2023-03-26 23:27:17,809 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-03-26 23:27:29,947 INFO [finetune.py:976] (3/7) Epoch 19, batch 4250, loss[loss=0.1841, simple_loss=0.2532, pruned_loss=0.0575, over 4738.00 frames. ], tot_loss[loss=0.1801, simple_loss=0.2504, pruned_loss=0.05489, over 950365.54 frames. ], batch size: 59, lr: 3.28e-03, grad_scale: 32.0 2023-03-26 23:27:33,460 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.100e+02 1.503e+02 1.795e+02 2.146e+02 3.676e+02, threshold=3.590e+02, percent-clipped=0.0 2023-03-26 23:28:02,846 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.2933, 4.8965, 4.5995, 2.8848, 4.8990, 3.7444, 0.6716, 3.5380], device='cuda:3'), covar=tensor([0.2039, 0.2125, 0.1318, 0.2969, 0.0863, 0.0921, 0.5320, 0.1438], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0175, 0.0159, 0.0128, 0.0159, 0.0122, 0.0147, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 23:28:03,393 INFO [finetune.py:976] (3/7) Epoch 19, batch 4300, loss[loss=0.1252, simple_loss=0.2008, pruned_loss=0.02474, over 4765.00 frames. ], tot_loss[loss=0.1785, simple_loss=0.2478, pruned_loss=0.05456, over 949319.01 frames. ], batch size: 28, lr: 3.28e-03, grad_scale: 32.0 2023-03-26 23:28:09,582 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4536, 2.4673, 1.9178, 2.7760, 2.4316, 2.1559, 2.9823, 2.5305], device='cuda:3'), covar=tensor([0.1146, 0.2108, 0.2885, 0.2344, 0.2306, 0.1512, 0.3061, 0.1681], device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0188, 0.0235, 0.0253, 0.0247, 0.0202, 0.0215, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 23:28:15,966 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7038, 3.6696, 3.5079, 1.7703, 3.7245, 2.7561, 0.9683, 2.6404], device='cuda:3'), covar=tensor([0.2418, 0.1868, 0.1512, 0.3269, 0.1085, 0.1069, 0.4299, 0.1414], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0175, 0.0159, 0.0129, 0.0159, 0.0122, 0.0147, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 23:28:17,712 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1541, 2.1195, 2.1716, 1.3275, 2.2338, 2.1689, 2.1539, 1.7627], device='cuda:3'), covar=tensor([0.0519, 0.0589, 0.0609, 0.0954, 0.0692, 0.0651, 0.0577, 0.1137], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0135, 0.0140, 0.0121, 0.0125, 0.0139, 0.0140, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 23:28:24,667 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.0763, 4.4386, 4.6358, 4.8911, 4.7792, 4.4928, 5.1618, 1.6938], device='cuda:3'), covar=tensor([0.0658, 0.0734, 0.0704, 0.0835, 0.1082, 0.1455, 0.0555, 0.5452], device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0243, 0.0277, 0.0290, 0.0332, 0.0281, 0.0301, 0.0293], 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') 2023-03-26 23:28:35,893 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.07 vs. limit=5.0 2023-03-26 23:28:36,201 INFO [finetune.py:976] (3/7) Epoch 19, batch 4350, loss[loss=0.153, simple_loss=0.2303, pruned_loss=0.0379, over 4689.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2447, pruned_loss=0.05325, over 950534.20 frames. ], batch size: 23, lr: 3.28e-03, grad_scale: 32.0 2023-03-26 23:28:40,182 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.110e+02 1.532e+02 1.813e+02 2.231e+02 3.395e+02, threshold=3.625e+02, percent-clipped=1.0 2023-03-26 23:29:21,286 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107496.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 23:29:23,018 INFO [finetune.py:976] (3/7) Epoch 19, batch 4400, loss[loss=0.1806, simple_loss=0.2374, pruned_loss=0.06184, over 4778.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2451, pruned_loss=0.05372, over 951915.09 frames. ], batch size: 25, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:29:29,600 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107508.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 23:29:31,353 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107510.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:29:56,824 INFO [finetune.py:976] (3/7) Epoch 19, batch 4450, loss[loss=0.1565, simple_loss=0.222, pruned_loss=0.04551, over 4750.00 frames. ], tot_loss[loss=0.1785, simple_loss=0.2481, pruned_loss=0.05444, over 952254.98 frames. ], batch size: 23, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:29:59,907 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.983e+01 1.601e+02 1.972e+02 2.467e+02 3.942e+02, threshold=3.944e+02, percent-clipped=4.0 2023-03-26 23:30:01,733 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.58 vs. limit=2.0 2023-03-26 23:30:10,187 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 2023-03-26 23:30:12,267 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107571.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:30:42,902 INFO [finetune.py:976] (3/7) Epoch 19, batch 4500, loss[loss=0.1851, simple_loss=0.2592, pruned_loss=0.0555, over 4926.00 frames. ], tot_loss[loss=0.1789, simple_loss=0.2489, pruned_loss=0.05445, over 954638.16 frames. ], batch size: 33, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:31:25,200 INFO [finetune.py:976] (3/7) Epoch 19, batch 4550, loss[loss=0.1947, simple_loss=0.2646, pruned_loss=0.06242, over 4910.00 frames. ], tot_loss[loss=0.1798, simple_loss=0.2501, pruned_loss=0.05475, over 955765.55 frames. ], batch size: 37, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:31:28,195 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.023e+02 1.552e+02 1.832e+02 2.186e+02 5.352e+02, threshold=3.664e+02, percent-clipped=1.0 2023-03-26 23:31:28,334 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1402, 2.1066, 2.1366, 1.4090, 2.2057, 2.1380, 2.1420, 1.7332], device='cuda:3'), covar=tensor([0.0572, 0.0614, 0.0606, 0.0849, 0.0658, 0.0636, 0.0591, 0.1087], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0134, 0.0140, 0.0120, 0.0125, 0.0139, 0.0140, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 23:31:31,344 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107659.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:32:03,429 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2119, 2.1011, 2.7023, 4.2459, 3.0541, 2.9302, 1.0719, 3.6686], device='cuda:3'), covar=tensor([0.1531, 0.1297, 0.1259, 0.0471, 0.0637, 0.1426, 0.1941, 0.0302], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0115, 0.0132, 0.0163, 0.0100, 0.0135, 0.0123, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 23:32:07,544 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6036, 1.5218, 1.4951, 1.6424, 1.2199, 3.5665, 1.4582, 1.9117], device='cuda:3'), covar=tensor([0.3200, 0.2519, 0.2162, 0.2234, 0.1758, 0.0185, 0.2448, 0.1188], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0114, 0.0119, 0.0122, 0.0113, 0.0096, 0.0095, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 23:32:12,109 INFO [finetune.py:976] (3/7) Epoch 19, batch 4600, loss[loss=0.1848, simple_loss=0.2467, pruned_loss=0.06141, over 4818.00 frames. ], tot_loss[loss=0.1796, simple_loss=0.2499, pruned_loss=0.05462, over 956699.66 frames. ], batch size: 33, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:32:12,420 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-03-26 23:32:26,322 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107720.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:32:37,572 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107737.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:32:45,689 INFO [finetune.py:976] (3/7) Epoch 19, batch 4650, loss[loss=0.1687, simple_loss=0.2368, pruned_loss=0.05027, over 4820.00 frames. ], tot_loss[loss=0.1783, simple_loss=0.248, pruned_loss=0.05427, over 958355.54 frames. ], batch size: 40, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:32:48,738 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.855e+01 1.504e+02 1.713e+02 2.086e+02 4.043e+02, threshold=3.426e+02, percent-clipped=2.0 2023-03-26 23:33:17,159 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107796.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 23:33:18,393 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107798.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 23:33:19,352 INFO [finetune.py:976] (3/7) Epoch 19, batch 4700, loss[loss=0.1783, simple_loss=0.2552, pruned_loss=0.05067, over 4818.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2457, pruned_loss=0.05386, over 958872.78 frames. ], batch size: 39, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:33:25,041 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107808.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:33:49,094 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=107844.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 23:33:54,086 INFO [finetune.py:976] (3/7) Epoch 19, batch 4750, loss[loss=0.1688, simple_loss=0.2459, pruned_loss=0.04581, over 4917.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2444, pruned_loss=0.05334, over 956470.01 frames. ], batch size: 37, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:33:57,616 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.048e+02 1.438e+02 1.688e+02 2.143e+02 3.806e+02, threshold=3.376e+02, percent-clipped=2.0 2023-03-26 23:33:58,879 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=107856.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:34:04,995 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107866.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:34:07,447 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2836, 1.3554, 1.5286, 1.4301, 1.4587, 2.8681, 1.2355, 1.4005], device='cuda:3'), covar=tensor([0.1012, 0.1795, 0.1242, 0.0966, 0.1625, 0.0308, 0.1494, 0.1731], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0074, 0.0077, 0.0091, 0.0080, 0.0084, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 23:34:16,437 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8019, 1.2855, 1.9017, 1.8652, 1.6804, 1.6036, 1.7766, 1.7451], device='cuda:3'), covar=tensor([0.4325, 0.4070, 0.3490, 0.3835, 0.4788, 0.3669, 0.4442, 0.3202], device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0239, 0.0259, 0.0278, 0.0275, 0.0250, 0.0285, 0.0241], 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') 2023-03-26 23:34:37,227 INFO [finetune.py:976] (3/7) Epoch 19, batch 4800, loss[loss=0.1774, simple_loss=0.2555, pruned_loss=0.04961, over 4936.00 frames. ], tot_loss[loss=0.1787, simple_loss=0.2477, pruned_loss=0.05485, over 956144.10 frames. ], batch size: 38, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:35:10,747 INFO [finetune.py:976] (3/7) Epoch 19, batch 4850, loss[loss=0.1894, simple_loss=0.2562, pruned_loss=0.06126, over 4323.00 frames. ], tot_loss[loss=0.1798, simple_loss=0.2499, pruned_loss=0.05481, over 954894.43 frames. ], batch size: 66, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:35:13,742 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.163e+02 1.610e+02 1.895e+02 2.225e+02 4.035e+02, threshold=3.790e+02, percent-clipped=2.0 2023-03-26 23:35:19,087 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6849, 1.6125, 1.6189, 1.6511, 1.3213, 3.6630, 1.4102, 1.8784], device='cuda:3'), covar=tensor([0.3399, 0.2563, 0.2103, 0.2312, 0.1682, 0.0198, 0.2633, 0.1280], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0115, 0.0120, 0.0123, 0.0114, 0.0097, 0.0095, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 23:35:29,653 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.75 vs. limit=5.0 2023-03-26 23:35:45,814 INFO [finetune.py:976] (3/7) Epoch 19, batch 4900, loss[loss=0.2172, simple_loss=0.2886, pruned_loss=0.07294, over 4811.00 frames. ], tot_loss[loss=0.1814, simple_loss=0.2516, pruned_loss=0.05563, over 957836.78 frames. ], batch size: 38, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:35:56,296 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-03-26 23:35:57,376 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108015.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:35:58,138 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-03-26 23:36:28,166 INFO [finetune.py:976] (3/7) Epoch 19, batch 4950, loss[loss=0.176, simple_loss=0.2418, pruned_loss=0.05511, over 4804.00 frames. ], tot_loss[loss=0.1807, simple_loss=0.2513, pruned_loss=0.05499, over 957824.17 frames. ], batch size: 25, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:36:31,632 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.136e+02 1.572e+02 1.807e+02 2.323e+02 4.539e+02, threshold=3.614e+02, percent-clipped=1.0 2023-03-26 23:37:04,014 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108093.0, num_to_drop=1, layers_to_drop={2} 2023-03-26 23:37:10,957 INFO [finetune.py:976] (3/7) Epoch 19, batch 5000, loss[loss=0.2173, simple_loss=0.2783, pruned_loss=0.07818, over 4900.00 frames. ], tot_loss[loss=0.1791, simple_loss=0.2493, pruned_loss=0.05447, over 956276.29 frames. ], batch size: 37, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:37:26,859 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9217, 1.4052, 1.9661, 1.9029, 1.7321, 1.7111, 1.8490, 1.8463], device='cuda:3'), covar=tensor([0.4160, 0.4071, 0.3242, 0.3751, 0.4551, 0.3524, 0.4333, 0.2990], device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0241, 0.0260, 0.0280, 0.0277, 0.0252, 0.0286, 0.0243], 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') 2023-03-26 23:37:48,306 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1118, 1.6270, 2.2128, 2.0719, 1.8619, 1.8222, 2.0764, 2.0108], device='cuda:3'), covar=tensor([0.4188, 0.4386, 0.3212, 0.4171, 0.4810, 0.4041, 0.4817, 0.3083], device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0240, 0.0259, 0.0279, 0.0276, 0.0251, 0.0286, 0.0242], 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') 2023-03-26 23:37:54,036 INFO [finetune.py:976] (3/7) Epoch 19, batch 5050, loss[loss=0.2182, simple_loss=0.2693, pruned_loss=0.08361, over 4908.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2469, pruned_loss=0.05372, over 957663.11 frames. ], batch size: 43, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:37:57,572 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.150e+02 1.577e+02 1.863e+02 2.132e+02 3.762e+02, threshold=3.725e+02, percent-clipped=1.0 2023-03-26 23:38:05,857 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108166.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:38:27,778 INFO [finetune.py:976] (3/7) Epoch 19, batch 5100, loss[loss=0.1891, simple_loss=0.2467, pruned_loss=0.06571, over 4874.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.242, pruned_loss=0.05176, over 956776.35 frames. ], batch size: 31, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:38:38,021 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=108214.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:38:44,664 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.50 vs. limit=5.0 2023-03-26 23:39:00,758 INFO [finetune.py:976] (3/7) Epoch 19, batch 5150, loss[loss=0.1335, simple_loss=0.1976, pruned_loss=0.03468, over 3959.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2428, pruned_loss=0.05225, over 955637.12 frames. ], batch size: 17, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:39:04,794 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.278e+01 1.451e+02 1.873e+02 2.231e+02 4.201e+02, threshold=3.747e+02, percent-clipped=0.0 2023-03-26 23:39:15,520 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2557, 2.9328, 2.6917, 1.1843, 2.9749, 2.2620, 0.5978, 1.9133], device='cuda:3'), covar=tensor([0.2599, 0.2630, 0.1940, 0.3910, 0.1440, 0.1231, 0.4368, 0.1867], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0173, 0.0157, 0.0127, 0.0157, 0.0120, 0.0144, 0.0120], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 23:39:39,544 INFO [finetune.py:976] (3/7) Epoch 19, batch 5200, loss[loss=0.2238, simple_loss=0.2946, pruned_loss=0.07649, over 4894.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2476, pruned_loss=0.05361, over 957028.74 frames. ], batch size: 35, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:39:49,778 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0901, 1.7300, 2.1022, 2.0472, 1.8315, 1.8355, 2.0400, 1.9330], device='cuda:3'), covar=tensor([0.3762, 0.4126, 0.2951, 0.3900, 0.4795, 0.3687, 0.4578, 0.2965], device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0241, 0.0261, 0.0280, 0.0277, 0.0252, 0.0287, 0.0243], 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') 2023-03-26 23:39:54,426 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108315.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:40:00,160 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-03-26 23:40:00,996 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108325.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:40:16,476 INFO [finetune.py:976] (3/7) Epoch 19, batch 5250, loss[loss=0.1797, simple_loss=0.2496, pruned_loss=0.05495, over 4064.00 frames. ], tot_loss[loss=0.1795, simple_loss=0.2501, pruned_loss=0.0544, over 955880.21 frames. ], batch size: 65, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:40:19,992 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.128e+02 1.566e+02 1.984e+02 2.332e+02 4.295e+02, threshold=3.968e+02, percent-clipped=2.0 2023-03-26 23:40:26,551 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=108363.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:40:42,061 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108386.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:40:46,337 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108393.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:40:49,957 INFO [finetune.py:976] (3/7) Epoch 19, batch 5300, loss[loss=0.1848, simple_loss=0.2554, pruned_loss=0.05709, over 4892.00 frames. ], tot_loss[loss=0.1815, simple_loss=0.2524, pruned_loss=0.05527, over 957247.65 frames. ], batch size: 36, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:40:53,138 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3773, 2.2044, 1.8174, 0.7896, 1.8898, 1.8805, 1.7290, 2.0216], device='cuda:3'), covar=tensor([0.0896, 0.0724, 0.1423, 0.2033, 0.1282, 0.2281, 0.2122, 0.0861], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0194, 0.0200, 0.0183, 0.0211, 0.0208, 0.0223, 0.0197], 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') 2023-03-26 23:41:22,544 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108432.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:41:22,577 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3002, 2.1913, 1.8916, 2.4154, 2.1161, 2.1278, 2.0573, 3.0641], device='cuda:3'), covar=tensor([0.4012, 0.5378, 0.3611, 0.4417, 0.4788, 0.2504, 0.4808, 0.1631], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0260, 0.0230, 0.0275, 0.0251, 0.0221, 0.0253, 0.0233], 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') 2023-03-26 23:41:32,019 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=108441.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:41:36,858 INFO [finetune.py:976] (3/7) Epoch 19, batch 5350, loss[loss=0.1404, simple_loss=0.209, pruned_loss=0.03594, over 4717.00 frames. ], tot_loss[loss=0.18, simple_loss=0.2516, pruned_loss=0.05415, over 956005.70 frames. ], batch size: 23, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:41:39,875 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.546e+01 1.453e+02 1.815e+02 2.266e+02 3.194e+02, threshold=3.630e+02, percent-clipped=0.0 2023-03-26 23:42:09,097 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108493.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 23:42:12,595 INFO [finetune.py:976] (3/7) Epoch 19, batch 5400, loss[loss=0.1403, simple_loss=0.2085, pruned_loss=0.03604, over 4867.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2485, pruned_loss=0.05325, over 956678.53 frames. ], batch size: 31, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:42:53,840 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5982, 2.4044, 1.9333, 2.7060, 2.5172, 2.2022, 3.0319, 2.6290], device='cuda:3'), covar=tensor([0.1340, 0.2399, 0.3160, 0.2667, 0.2573, 0.1675, 0.3077, 0.1794], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0187, 0.0233, 0.0252, 0.0245, 0.0202, 0.0213, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 23:42:58,627 INFO [finetune.py:976] (3/7) Epoch 19, batch 5450, loss[loss=0.1572, simple_loss=0.2304, pruned_loss=0.04195, over 4866.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2456, pruned_loss=0.05305, over 954922.95 frames. ], batch size: 31, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:43:01,646 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.000e+02 1.562e+02 1.816e+02 2.216e+02 4.232e+02, threshold=3.632e+02, percent-clipped=2.0 2023-03-26 23:43:04,524 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.79 vs. limit=5.0 2023-03-26 23:43:11,305 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-03-26 23:43:17,235 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108577.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:43:31,855 INFO [finetune.py:976] (3/7) Epoch 19, batch 5500, loss[loss=0.1359, simple_loss=0.2056, pruned_loss=0.03309, over 4783.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2421, pruned_loss=0.05206, over 954513.73 frames. ], batch size: 26, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:43:42,703 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.46 vs. limit=5.0 2023-03-26 23:43:58,834 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108638.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:44:05,685 INFO [finetune.py:976] (3/7) Epoch 19, batch 5550, loss[loss=0.189, simple_loss=0.2588, pruned_loss=0.05955, over 4832.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2443, pruned_loss=0.05289, over 951386.95 frames. ], batch size: 30, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:44:08,703 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.053e+02 1.553e+02 1.822e+02 2.201e+02 3.552e+02, threshold=3.643e+02, percent-clipped=0.0 2023-03-26 23:44:27,076 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108681.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:44:37,477 INFO [finetune.py:976] (3/7) Epoch 19, batch 5600, loss[loss=0.1548, simple_loss=0.2448, pruned_loss=0.03233, over 4805.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2479, pruned_loss=0.05376, over 953046.32 frames. ], batch size: 51, lr: 3.27e-03, grad_scale: 32.0 2023-03-26 23:45:09,468 INFO [finetune.py:976] (3/7) Epoch 19, batch 5650, loss[loss=0.2192, simple_loss=0.2876, pruned_loss=0.07541, over 4758.00 frames. ], tot_loss[loss=0.1783, simple_loss=0.2494, pruned_loss=0.05361, over 953419.00 frames. ], batch size: 27, lr: 3.26e-03, grad_scale: 32.0 2023-03-26 23:45:12,316 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.055e+02 1.584e+02 1.878e+02 2.184e+02 3.636e+02, threshold=3.756e+02, percent-clipped=0.0 2023-03-26 23:45:32,949 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108788.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 23:45:34,766 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108791.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:45:39,456 INFO [finetune.py:976] (3/7) Epoch 19, batch 5700, loss[loss=0.2025, simple_loss=0.2529, pruned_loss=0.07607, over 3545.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2458, pruned_loss=0.05334, over 934590.70 frames. ], batch size: 15, lr: 3.26e-03, grad_scale: 32.0 2023-03-26 23:45:49,155 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8836, 1.8798, 2.3788, 3.4628, 2.4224, 2.7011, 1.4646, 2.7514], device='cuda:3'), covar=tensor([0.1530, 0.1273, 0.1252, 0.0691, 0.0814, 0.1139, 0.1603, 0.0516], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0115, 0.0133, 0.0163, 0.0100, 0.0135, 0.0123, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-26 23:46:08,104 INFO [finetune.py:976] (3/7) Epoch 20, batch 0, loss[loss=0.1162, simple_loss=0.1841, pruned_loss=0.0241, over 4714.00 frames. ], tot_loss[loss=0.1162, simple_loss=0.1841, pruned_loss=0.0241, over 4714.00 frames. ], batch size: 23, lr: 3.26e-03, grad_scale: 32.0 2023-03-26 23:46:08,104 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-26 23:46:15,495 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8428, 1.3169, 1.0030, 1.6471, 2.0795, 1.2907, 1.6423, 1.5698], device='cuda:3'), covar=tensor([0.1281, 0.1778, 0.1663, 0.1044, 0.1742, 0.1793, 0.1184, 0.1780], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0095, 0.0111, 0.0091, 0.0120, 0.0093, 0.0098, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 23:46:16,862 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2110, 1.9621, 1.8468, 1.7773, 1.9378, 2.0045, 1.9747, 2.5987], device='cuda:3'), covar=tensor([0.4160, 0.5265, 0.3625, 0.4156, 0.4327, 0.2772, 0.4064, 0.2014], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0261, 0.0230, 0.0275, 0.0251, 0.0221, 0.0252, 0.0232], 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') 2023-03-26 23:46:24,541 INFO [finetune.py:1010] (3/7) Epoch 20, validation: loss=0.158, simple_loss=0.2276, pruned_loss=0.04423, over 2265189.00 frames. 2023-03-26 23:46:24,542 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6434MB 2023-03-26 23:46:28,018 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.6075, 1.5504, 1.4814, 0.9113, 1.6634, 1.8352, 1.7894, 1.3773], device='cuda:3'), covar=tensor([0.1213, 0.0701, 0.0576, 0.0636, 0.0456, 0.0662, 0.0419, 0.0784], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0149, 0.0124, 0.0125, 0.0131, 0.0128, 0.0142, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.0560e-05, 1.0817e-04, 8.8728e-05, 8.8289e-05, 9.1966e-05, 9.1449e-05, 1.0205e-04, 1.0559e-04], device='cuda:3') 2023-03-26 23:46:32,153 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5956, 1.0828, 0.7551, 1.3644, 1.9868, 0.7554, 1.3254, 1.4407], device='cuda:3'), covar=tensor([0.1516, 0.2095, 0.1753, 0.1173, 0.1932, 0.1858, 0.1435, 0.1804], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0095, 0.0111, 0.0091, 0.0120, 0.0093, 0.0098, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 23:46:52,590 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108852.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:46:58,208 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.806e+01 1.424e+02 1.737e+02 2.098e+02 5.389e+02, threshold=3.475e+02, percent-clipped=2.0 2023-03-26 23:47:17,655 INFO [finetune.py:976] (3/7) Epoch 20, batch 50, loss[loss=0.2088, simple_loss=0.2574, pruned_loss=0.08008, over 4889.00 frames. ], tot_loss[loss=0.1829, simple_loss=0.2524, pruned_loss=0.05671, over 217497.35 frames. ], batch size: 43, lr: 3.26e-03, grad_scale: 32.0 2023-03-26 23:47:31,876 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3175, 2.8878, 2.7726, 1.2250, 3.0020, 2.2148, 0.5995, 1.9455], device='cuda:3'), covar=tensor([0.2335, 0.2061, 0.1802, 0.3474, 0.1321, 0.1108, 0.4141, 0.1552], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0173, 0.0158, 0.0128, 0.0157, 0.0121, 0.0145, 0.0121], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 23:47:57,360 INFO [finetune.py:976] (3/7) Epoch 20, batch 100, loss[loss=0.1828, simple_loss=0.2525, pruned_loss=0.05658, over 4912.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2452, pruned_loss=0.05317, over 382304.01 frames. ], batch size: 46, lr: 3.26e-03, grad_scale: 32.0 2023-03-26 23:48:06,350 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108933.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:48:23,147 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.009e+02 1.362e+02 1.754e+02 2.070e+02 5.157e+02, threshold=3.508e+02, percent-clipped=1.0 2023-03-26 23:48:23,914 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6192, 1.5607, 1.9404, 1.2754, 1.6225, 1.8430, 1.5109, 2.0278], device='cuda:3'), covar=tensor([0.1309, 0.2042, 0.1291, 0.1727, 0.0998, 0.1394, 0.2981, 0.0827], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0203, 0.0190, 0.0188, 0.0173, 0.0211, 0.0216, 0.0198], 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') 2023-03-26 23:48:38,581 INFO [finetune.py:976] (3/7) Epoch 20, batch 150, loss[loss=0.1443, simple_loss=0.214, pruned_loss=0.03735, over 4904.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2404, pruned_loss=0.05227, over 507986.99 frames. ], batch size: 43, lr: 3.26e-03, grad_scale: 32.0 2023-03-26 23:48:41,558 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108981.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:48:42,196 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5790, 1.7047, 1.4241, 1.6719, 1.9347, 1.8983, 1.5564, 1.4931], device='cuda:3'), covar=tensor([0.0329, 0.0315, 0.0572, 0.0300, 0.0209, 0.0423, 0.0436, 0.0396], device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0106, 0.0142, 0.0110, 0.0098, 0.0109, 0.0098, 0.0110], device='cuda:3'), out_proj_covar=tensor([7.4134e-05, 8.1646e-05, 1.1191e-04, 8.4489e-05, 7.6606e-05, 8.0729e-05, 7.3179e-05, 8.3893e-05], device='cuda:3') 2023-03-26 23:49:11,416 INFO [finetune.py:976] (3/7) Epoch 20, batch 200, loss[loss=0.1654, simple_loss=0.2365, pruned_loss=0.04719, over 4860.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2384, pruned_loss=0.05158, over 606922.26 frames. ], batch size: 44, lr: 3.26e-03, grad_scale: 32.0 2023-03-26 23:49:11,523 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.1410, 2.2748, 1.2263, 2.8156, 3.0437, 2.4534, 2.8091, 2.5391], device='cuda:3'), covar=tensor([0.0977, 0.1545, 0.1821, 0.0817, 0.1409, 0.1423, 0.0925, 0.1675], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0095, 0.0111, 0.0091, 0.0120, 0.0093, 0.0098, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 23:49:13,181 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=109029.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:49:16,079 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5265, 1.0719, 0.7864, 1.3550, 1.9533, 0.7157, 1.2782, 1.3075], device='cuda:3'), covar=tensor([0.1545, 0.2239, 0.1821, 0.1259, 0.2058, 0.1930, 0.1537, 0.2083], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0095, 0.0111, 0.0091, 0.0120, 0.0093, 0.0098, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-26 23:49:19,012 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109037.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 23:49:29,131 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.011e+02 1.519e+02 1.780e+02 2.129e+02 3.450e+02, threshold=3.561e+02, percent-clipped=0.0 2023-03-26 23:49:44,471 INFO [finetune.py:976] (3/7) Epoch 20, batch 250, loss[loss=0.1508, simple_loss=0.2223, pruned_loss=0.03971, over 4764.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2424, pruned_loss=0.05253, over 684987.99 frames. ], batch size: 27, lr: 3.26e-03, grad_scale: 32.0 2023-03-26 23:49:52,141 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109088.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:49:58,702 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109098.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 23:50:17,243 INFO [finetune.py:976] (3/7) Epoch 20, batch 300, loss[loss=0.2013, simple_loss=0.285, pruned_loss=0.05881, over 4812.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2473, pruned_loss=0.05372, over 746560.90 frames. ], batch size: 40, lr: 3.26e-03, grad_scale: 32.0 2023-03-26 23:50:23,613 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=109136.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:50:31,196 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109147.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:50:35,398 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.121e+02 1.518e+02 1.856e+02 2.256e+02 3.204e+02, threshold=3.712e+02, percent-clipped=0.0 2023-03-26 23:50:50,164 INFO [finetune.py:976] (3/7) Epoch 20, batch 350, loss[loss=0.1858, simple_loss=0.2628, pruned_loss=0.05446, over 4816.00 frames. ], tot_loss[loss=0.1799, simple_loss=0.25, pruned_loss=0.05489, over 793152.24 frames. ], batch size: 39, lr: 3.26e-03, grad_scale: 64.0 2023-03-26 23:51:01,445 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109194.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:51:10,832 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8055, 1.6652, 1.4084, 1.2776, 1.5774, 1.5542, 1.5756, 2.1644], device='cuda:3'), covar=tensor([0.3520, 0.3585, 0.3103, 0.3439, 0.3524, 0.2235, 0.3248, 0.1595], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0261, 0.0230, 0.0275, 0.0252, 0.0222, 0.0253, 0.0232], 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') 2023-03-26 23:51:19,298 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.56 vs. limit=2.0 2023-03-26 23:51:25,273 INFO [finetune.py:976] (3/7) Epoch 20, batch 400, loss[loss=0.1691, simple_loss=0.2423, pruned_loss=0.04798, over 4773.00 frames. ], tot_loss[loss=0.1819, simple_loss=0.2524, pruned_loss=0.0557, over 831327.02 frames. ], batch size: 26, lr: 3.26e-03, grad_scale: 64.0 2023-03-26 23:51:32,571 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8289, 1.3029, 1.7569, 1.7841, 1.6100, 1.6097, 1.6887, 1.6954], device='cuda:3'), covar=tensor([0.5665, 0.5114, 0.4431, 0.4662, 0.6237, 0.5109, 0.6005, 0.4176], device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0241, 0.0261, 0.0280, 0.0278, 0.0252, 0.0288, 0.0244], 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') 2023-03-26 23:51:34,310 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109233.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:51:45,834 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-03-26 23:51:46,989 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3029, 1.2811, 1.2420, 1.2481, 0.7524, 2.0642, 0.6944, 1.0972], device='cuda:3'), covar=tensor([0.3064, 0.2292, 0.1938, 0.2278, 0.1770, 0.0334, 0.2312, 0.1091], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0115, 0.0120, 0.0123, 0.0114, 0.0096, 0.0095, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-26 23:52:03,146 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5824, 1.5957, 1.3946, 1.6029, 1.8373, 1.8049, 1.5368, 1.4313], device='cuda:3'), covar=tensor([0.0310, 0.0264, 0.0595, 0.0272, 0.0204, 0.0448, 0.0330, 0.0336], device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0106, 0.0142, 0.0110, 0.0098, 0.0109, 0.0098, 0.0110], device='cuda:3'), out_proj_covar=tensor([7.4279e-05, 8.1651e-05, 1.1219e-04, 8.4524e-05, 7.6647e-05, 8.0771e-05, 7.3356e-05, 8.3887e-05], device='cuda:3') 2023-03-26 23:52:03,601 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.047e+02 1.657e+02 1.900e+02 2.185e+02 4.941e+02, threshold=3.801e+02, percent-clipped=3.0 2023-03-26 23:52:04,349 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109255.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:52:26,280 INFO [finetune.py:976] (3/7) Epoch 20, batch 450, loss[loss=0.1674, simple_loss=0.2401, pruned_loss=0.04735, over 4815.00 frames. ], tot_loss[loss=0.1797, simple_loss=0.2504, pruned_loss=0.05448, over 858911.91 frames. ], batch size: 38, lr: 3.26e-03, grad_scale: 64.0 2023-03-26 23:52:29,294 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=109281.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:52:38,177 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109294.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:52:53,488 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0624, 2.0152, 1.6050, 1.9517, 2.0319, 1.7668, 2.4509, 2.0517], device='cuda:3'), covar=tensor([0.1405, 0.2169, 0.3043, 0.2621, 0.2499, 0.1671, 0.3089, 0.1743], device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0187, 0.0234, 0.0251, 0.0245, 0.0202, 0.0214, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-26 23:52:54,097 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1525, 1.7483, 2.1519, 2.1615, 1.8369, 1.8285, 2.0528, 1.9936], device='cuda:3'), covar=tensor([0.4127, 0.4299, 0.3302, 0.3847, 0.5122, 0.4195, 0.5013, 0.3161], device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0241, 0.0262, 0.0281, 0.0278, 0.0253, 0.0289, 0.0244], 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') 2023-03-26 23:52:54,665 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109318.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:53:02,101 INFO [finetune.py:976] (3/7) Epoch 20, batch 500, loss[loss=0.2424, simple_loss=0.2972, pruned_loss=0.09384, over 4766.00 frames. ], tot_loss[loss=0.1795, simple_loss=0.2494, pruned_loss=0.05485, over 880181.59 frames. ], batch size: 26, lr: 3.26e-03, grad_scale: 64.0 2023-03-26 23:53:34,821 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.087e+02 1.492e+02 1.802e+02 2.178e+02 4.247e+02, threshold=3.605e+02, percent-clipped=3.0 2023-03-26 23:53:39,382 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109355.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:53:52,982 INFO [finetune.py:976] (3/7) Epoch 20, batch 550, loss[loss=0.1368, simple_loss=0.2026, pruned_loss=0.03547, over 4742.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2451, pruned_loss=0.05348, over 895327.02 frames. ], batch size: 23, lr: 3.26e-03, grad_scale: 64.0 2023-03-26 23:53:54,337 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109379.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 23:54:03,683 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109393.0, num_to_drop=1, layers_to_drop={3} 2023-03-26 23:54:26,236 INFO [finetune.py:976] (3/7) Epoch 20, batch 600, loss[loss=0.1655, simple_loss=0.2429, pruned_loss=0.04405, over 4865.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2453, pruned_loss=0.05329, over 909334.28 frames. ], batch size: 34, lr: 3.26e-03, grad_scale: 32.0 2023-03-26 23:54:39,786 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109447.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:54:44,549 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.417e+01 1.548e+02 1.729e+02 2.159e+02 3.434e+02, threshold=3.458e+02, percent-clipped=0.0 2023-03-26 23:54:59,324 INFO [finetune.py:976] (3/7) Epoch 20, batch 650, loss[loss=0.1757, simple_loss=0.249, pruned_loss=0.05121, over 4875.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2472, pruned_loss=0.05356, over 919723.86 frames. ], batch size: 34, lr: 3.26e-03, grad_scale: 32.0 2023-03-26 23:55:05,496 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7169, 3.9053, 3.6494, 1.8335, 4.0403, 2.9321, 0.8381, 2.6976], device='cuda:3'), covar=tensor([0.2392, 0.2075, 0.1608, 0.3757, 0.0966, 0.1117, 0.4797, 0.1702], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0175, 0.0159, 0.0129, 0.0159, 0.0123, 0.0146, 0.0122], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-26 23:55:10,856 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=109495.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:55:33,023 INFO [finetune.py:976] (3/7) Epoch 20, batch 700, loss[loss=0.1631, simple_loss=0.2443, pruned_loss=0.04088, over 4807.00 frames. ], tot_loss[loss=0.1797, simple_loss=0.25, pruned_loss=0.05473, over 927279.27 frames. ], batch size: 40, lr: 3.26e-03, grad_scale: 32.0 2023-03-26 23:55:37,967 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.8721, 4.2918, 4.4778, 4.6285, 4.6001, 4.3489, 4.9775, 1.6080], device='cuda:3'), covar=tensor([0.0709, 0.0756, 0.0772, 0.1179, 0.1082, 0.1284, 0.0508, 0.5403], device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0244, 0.0277, 0.0292, 0.0331, 0.0281, 0.0301, 0.0295], 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') 2023-03-26 23:55:47,502 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109550.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:55:51,348 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.095e+02 1.485e+02 1.783e+02 2.085e+02 4.380e+02, threshold=3.566e+02, percent-clipped=3.0 2023-03-26 23:56:06,061 INFO [finetune.py:976] (3/7) Epoch 20, batch 750, loss[loss=0.2409, simple_loss=0.3035, pruned_loss=0.08917, over 4808.00 frames. ], tot_loss[loss=0.1808, simple_loss=0.2511, pruned_loss=0.05525, over 931643.44 frames. ], batch size: 40, lr: 3.26e-03, grad_scale: 32.0 2023-03-26 23:56:07,744 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109579.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:56:12,638 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1717, 1.7681, 1.9065, 0.8574, 2.1587, 2.2876, 1.9481, 1.7226], device='cuda:3'), covar=tensor([0.0994, 0.0881, 0.0646, 0.0815, 0.0740, 0.0921, 0.0570, 0.0921], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0150, 0.0124, 0.0124, 0.0131, 0.0128, 0.0142, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.0524e-05, 1.0863e-04, 8.8719e-05, 8.8048e-05, 9.1909e-05, 9.1738e-05, 1.0175e-04, 1.0555e-04], device='cuda:3') 2023-03-26 23:56:39,568 INFO [finetune.py:976] (3/7) Epoch 20, batch 800, loss[loss=0.1506, simple_loss=0.2192, pruned_loss=0.041, over 4865.00 frames. ], tot_loss[loss=0.1794, simple_loss=0.2502, pruned_loss=0.05428, over 937832.06 frames. ], batch size: 34, lr: 3.25e-03, grad_scale: 32.0 2023-03-26 23:56:50,330 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109640.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:56:56,818 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109650.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:56:59,763 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.018e+02 1.497e+02 1.774e+02 2.103e+02 3.199e+02, threshold=3.548e+02, percent-clipped=0.0 2023-03-26 23:57:03,306 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109659.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:57:23,372 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109674.0, num_to_drop=1, layers_to_drop={0} 2023-03-26 23:57:23,451 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2455, 2.2965, 1.9213, 2.3618, 2.1766, 2.1894, 2.2404, 3.0831], device='cuda:3'), covar=tensor([0.3587, 0.4458, 0.3198, 0.4260, 0.4291, 0.2331, 0.3902, 0.1485], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0260, 0.0230, 0.0274, 0.0251, 0.0220, 0.0252, 0.0231], 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') 2023-03-26 23:57:25,125 INFO [finetune.py:976] (3/7) Epoch 20, batch 850, loss[loss=0.1985, simple_loss=0.2653, pruned_loss=0.06581, over 4863.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.248, pruned_loss=0.05364, over 942565.09 frames. ], batch size: 34, lr: 3.25e-03, grad_scale: 32.0 2023-03-26 23:57:40,034 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109693.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 23:58:06,031 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109720.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:58:10,730 INFO [finetune.py:976] (3/7) Epoch 20, batch 900, loss[loss=0.1475, simple_loss=0.2107, pruned_loss=0.04214, over 4372.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2461, pruned_loss=0.05344, over 945143.19 frames. ], batch size: 19, lr: 3.25e-03, grad_scale: 32.0 2023-03-26 23:58:22,309 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=109741.0, num_to_drop=1, layers_to_drop={1} 2023-03-26 23:58:40,323 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.017e+02 1.530e+02 1.823e+02 2.181e+02 3.809e+02, threshold=3.647e+02, percent-clipped=1.0 2023-03-26 23:59:03,843 INFO [finetune.py:976] (3/7) Epoch 20, batch 950, loss[loss=0.166, simple_loss=0.2246, pruned_loss=0.05373, over 4811.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2441, pruned_loss=0.05315, over 949418.14 frames. ], batch size: 25, lr: 3.25e-03, grad_scale: 32.0 2023-03-26 23:59:08,301 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-26 23:59:36,852 INFO [finetune.py:976] (3/7) Epoch 20, batch 1000, loss[loss=0.2329, simple_loss=0.2967, pruned_loss=0.08454, over 4819.00 frames. ], tot_loss[loss=0.1793, simple_loss=0.2488, pruned_loss=0.05494, over 951021.28 frames. ], batch size: 40, lr: 3.25e-03, grad_scale: 32.0 2023-03-26 23:59:39,841 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9170, 2.7353, 2.4807, 2.9939, 2.6363, 2.6131, 2.6497, 3.5095], device='cuda:3'), covar=tensor([0.3301, 0.4474, 0.2932, 0.3638, 0.3812, 0.2188, 0.3815, 0.1499], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0261, 0.0231, 0.0276, 0.0252, 0.0221, 0.0252, 0.0232], 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') 2023-03-26 23:59:44,333 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-03-26 23:59:52,405 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109850.0, num_to_drop=0, layers_to_drop=set() 2023-03-26 23:59:54,872 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5367, 2.3049, 1.9836, 0.9439, 2.1171, 1.8715, 1.8105, 2.1741], device='cuda:3'), covar=tensor([0.0910, 0.0915, 0.1798, 0.2279, 0.1561, 0.2560, 0.2361, 0.1138], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0193, 0.0201, 0.0183, 0.0211, 0.0208, 0.0224, 0.0197], 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') 2023-03-26 23:59:55,343 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.032e+02 1.630e+02 1.951e+02 2.313e+02 5.473e+02, threshold=3.903e+02, percent-clipped=2.0 2023-03-26 23:59:55,527 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.34 vs. limit=5.0 2023-03-27 00:00:07,837 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5917, 1.4522, 1.3298, 1.5194, 1.7905, 1.7478, 1.4697, 1.3495], device='cuda:3'), covar=tensor([0.0327, 0.0326, 0.0636, 0.0324, 0.0221, 0.0489, 0.0351, 0.0417], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0108, 0.0145, 0.0112, 0.0100, 0.0111, 0.0100, 0.0111], device='cuda:3'), out_proj_covar=tensor([7.5595e-05, 8.2912e-05, 1.1420e-04, 8.5891e-05, 7.7957e-05, 8.2067e-05, 7.4726e-05, 8.5024e-05], device='cuda:3') 2023-03-27 00:00:10,591 INFO [finetune.py:976] (3/7) Epoch 20, batch 1050, loss[loss=0.1435, simple_loss=0.2064, pruned_loss=0.04035, over 4715.00 frames. ], tot_loss[loss=0.1808, simple_loss=0.2514, pruned_loss=0.05509, over 953780.02 frames. ], batch size: 23, lr: 3.25e-03, grad_scale: 32.0 2023-03-27 00:00:16,385 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4550, 3.3364, 3.2101, 1.2707, 3.4450, 2.6014, 0.8587, 2.2539], device='cuda:3'), covar=tensor([0.2613, 0.2258, 0.1509, 0.3795, 0.1228, 0.1054, 0.4104, 0.1613], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0177, 0.0160, 0.0131, 0.0161, 0.0123, 0.0147, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 00:00:20,356 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.36 vs. limit=5.0 2023-03-27 00:00:24,789 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=109898.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:00:36,721 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3211, 1.3595, 1.6164, 1.1402, 1.3210, 1.4569, 1.3396, 1.6388], device='cuda:3'), covar=tensor([0.1231, 0.2229, 0.1268, 0.1430, 0.0941, 0.1246, 0.3125, 0.0865], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0204, 0.0189, 0.0189, 0.0174, 0.0213, 0.0217, 0.0199], 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') 2023-03-27 00:00:43,714 INFO [finetune.py:976] (3/7) Epoch 20, batch 1100, loss[loss=0.1749, simple_loss=0.2254, pruned_loss=0.06217, over 4284.00 frames. ], tot_loss[loss=0.1814, simple_loss=0.252, pruned_loss=0.05541, over 954902.42 frames. ], batch size: 18, lr: 3.25e-03, grad_scale: 32.0 2023-03-27 00:00:49,667 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109935.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:00:59,755 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109950.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:01:02,698 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.100e+02 1.532e+02 1.890e+02 2.271e+02 3.423e+02, threshold=3.780e+02, percent-clipped=0.0 2023-03-27 00:01:07,338 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-27 00:01:15,501 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109974.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:01:17,221 INFO [finetune.py:976] (3/7) Epoch 20, batch 1150, loss[loss=0.1708, simple_loss=0.2461, pruned_loss=0.04781, over 4888.00 frames. ], tot_loss[loss=0.181, simple_loss=0.252, pruned_loss=0.055, over 955067.29 frames. ], batch size: 35, lr: 3.25e-03, grad_scale: 32.0 2023-03-27 00:01:31,933 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=109998.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:01:37,935 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3556, 2.2825, 1.8001, 2.4049, 2.2303, 2.0488, 2.6566, 2.3935], device='cuda:3'), covar=tensor([0.1248, 0.2008, 0.2904, 0.2406, 0.2494, 0.1495, 0.2717, 0.1693], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0186, 0.0232, 0.0251, 0.0245, 0.0202, 0.0213, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 00:01:44,406 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110015.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:01:45,679 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110017.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:01:48,679 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=110022.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:01:52,168 INFO [finetune.py:976] (3/7) Epoch 20, batch 1200, loss[loss=0.1614, simple_loss=0.2324, pruned_loss=0.04514, over 4818.00 frames. ], tot_loss[loss=0.1785, simple_loss=0.2492, pruned_loss=0.05391, over 954235.81 frames. ], batch size: 39, lr: 3.25e-03, grad_scale: 32.0 2023-03-27 00:02:04,559 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110044.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:02:11,656 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.682e+01 1.527e+02 1.776e+02 2.166e+02 4.163e+02, threshold=3.552e+02, percent-clipped=2.0 2023-03-27 00:02:32,490 INFO [finetune.py:976] (3/7) Epoch 20, batch 1250, loss[loss=0.1991, simple_loss=0.2564, pruned_loss=0.07097, over 4810.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2469, pruned_loss=0.05367, over 955724.66 frames. ], batch size: 39, lr: 3.25e-03, grad_scale: 32.0 2023-03-27 00:02:33,257 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110078.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:02:45,716 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.8286, 4.1764, 4.4438, 4.6303, 4.5973, 4.3205, 4.9797, 1.5331], device='cuda:3'), covar=tensor([0.0732, 0.0839, 0.0736, 0.0849, 0.1058, 0.1438, 0.0506, 0.5426], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0245, 0.0280, 0.0292, 0.0333, 0.0284, 0.0302, 0.0298], 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') 2023-03-27 00:03:02,222 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110105.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:03:09,990 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110110.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:03:13,058 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110115.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:03:23,865 INFO [finetune.py:976] (3/7) Epoch 20, batch 1300, loss[loss=0.1758, simple_loss=0.2338, pruned_loss=0.05884, over 4935.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2435, pruned_loss=0.05259, over 954418.98 frames. ], batch size: 38, lr: 3.25e-03, grad_scale: 32.0 2023-03-27 00:03:45,427 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.072e+02 1.532e+02 1.884e+02 2.318e+02 4.682e+02, threshold=3.767e+02, percent-clipped=2.0 2023-03-27 00:03:52,957 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-27 00:04:04,748 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110171.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:04:12,699 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110176.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:04:13,182 INFO [finetune.py:976] (3/7) Epoch 20, batch 1350, loss[loss=0.1599, simple_loss=0.2281, pruned_loss=0.04579, over 4289.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2454, pruned_loss=0.05343, over 953415.03 frames. ], batch size: 18, lr: 3.25e-03, grad_scale: 32.0 2023-03-27 00:05:15,848 INFO [finetune.py:976] (3/7) Epoch 20, batch 1400, loss[loss=0.2121, simple_loss=0.2764, pruned_loss=0.07387, over 4905.00 frames. ], tot_loss[loss=0.1796, simple_loss=0.2493, pruned_loss=0.05498, over 955775.68 frames. ], batch size: 36, lr: 3.25e-03, grad_scale: 32.0 2023-03-27 00:05:26,714 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110235.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:05:48,219 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.073e+02 1.545e+02 1.824e+02 2.176e+02 3.637e+02, threshold=3.648e+02, percent-clipped=0.0 2023-03-27 00:06:02,046 INFO [finetune.py:976] (3/7) Epoch 20, batch 1450, loss[loss=0.1559, simple_loss=0.2294, pruned_loss=0.04118, over 4715.00 frames. ], tot_loss[loss=0.18, simple_loss=0.2503, pruned_loss=0.05481, over 955585.73 frames. ], batch size: 23, lr: 3.25e-03, grad_scale: 32.0 2023-03-27 00:06:06,234 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=110283.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:06:06,282 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110283.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:06:16,220 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6860, 3.8503, 3.7048, 1.9344, 3.9644, 2.8651, 0.7308, 2.7108], device='cuda:3'), covar=tensor([0.2231, 0.1656, 0.1389, 0.3316, 0.0860, 0.1032, 0.4690, 0.1369], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0178, 0.0161, 0.0131, 0.0162, 0.0124, 0.0147, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 00:06:20,401 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110303.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:06:28,395 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-27 00:06:28,599 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110315.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:06:30,458 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110318.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:06:35,827 INFO [finetune.py:976] (3/7) Epoch 20, batch 1500, loss[loss=0.2069, simple_loss=0.2735, pruned_loss=0.07014, over 4821.00 frames. ], tot_loss[loss=0.1806, simple_loss=0.2512, pruned_loss=0.05504, over 955996.15 frames. ], batch size: 47, lr: 3.25e-03, grad_scale: 32.0 2023-03-27 00:06:47,640 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110344.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:06:55,109 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.117e+02 1.590e+02 1.939e+02 2.244e+02 3.777e+02, threshold=3.878e+02, percent-clipped=2.0 2023-03-27 00:07:00,486 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=110363.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:07:01,164 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110364.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:07:07,074 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110373.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:07:09,448 INFO [finetune.py:976] (3/7) Epoch 20, batch 1550, loss[loss=0.2067, simple_loss=0.2735, pruned_loss=0.06999, over 4751.00 frames. ], tot_loss[loss=0.1818, simple_loss=0.2523, pruned_loss=0.05561, over 954892.35 frames. ], batch size: 54, lr: 3.25e-03, grad_scale: 32.0 2023-03-27 00:07:10,804 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110379.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:07:25,521 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110400.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:07:43,128 INFO [finetune.py:976] (3/7) Epoch 20, batch 1600, loss[loss=0.1816, simple_loss=0.2539, pruned_loss=0.05469, over 4897.00 frames. ], tot_loss[loss=0.1808, simple_loss=0.2506, pruned_loss=0.0555, over 954453.43 frames. ], batch size: 35, lr: 3.25e-03, grad_scale: 32.0 2023-03-27 00:08:13,511 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.758e+01 1.502e+02 1.787e+02 2.306e+02 4.709e+02, threshold=3.574e+02, percent-clipped=2.0 2023-03-27 00:08:29,698 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110466.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:08:33,171 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110471.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:08:40,393 INFO [finetune.py:976] (3/7) Epoch 20, batch 1650, loss[loss=0.156, simple_loss=0.2263, pruned_loss=0.04288, over 4909.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2472, pruned_loss=0.05439, over 953494.33 frames. ], batch size: 32, lr: 3.25e-03, grad_scale: 32.0 2023-03-27 00:08:55,547 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.8226, 4.1893, 4.4085, 4.6400, 4.5667, 4.2740, 4.9290, 1.6752], device='cuda:3'), covar=tensor([0.0656, 0.0750, 0.0697, 0.0777, 0.1094, 0.1463, 0.0473, 0.5358], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0244, 0.0278, 0.0291, 0.0331, 0.0282, 0.0302, 0.0296], 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') 2023-03-27 00:09:05,830 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.44 vs. limit=5.0 2023-03-27 00:09:22,347 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7171, 2.5374, 1.9941, 1.0434, 2.2679, 2.1806, 1.8087, 2.3561], device='cuda:3'), covar=tensor([0.0695, 0.0717, 0.1275, 0.1853, 0.1111, 0.1883, 0.1949, 0.0739], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0191, 0.0199, 0.0181, 0.0208, 0.0206, 0.0221, 0.0195], 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') 2023-03-27 00:09:24,056 INFO [finetune.py:976] (3/7) Epoch 20, batch 1700, loss[loss=0.1826, simple_loss=0.2416, pruned_loss=0.0618, over 4208.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2443, pruned_loss=0.05313, over 953892.04 frames. ], batch size: 18, lr: 3.25e-03, grad_scale: 32.0 2023-03-27 00:09:40,333 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2824, 2.8667, 2.7414, 1.2375, 3.0224, 2.2062, 0.7694, 1.8864], device='cuda:3'), covar=tensor([0.2421, 0.2371, 0.1814, 0.3739, 0.1413, 0.1175, 0.4085, 0.1678], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0178, 0.0161, 0.0131, 0.0163, 0.0124, 0.0147, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 00:09:42,518 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.071e+02 1.500e+02 1.773e+02 2.246e+02 3.830e+02, threshold=3.546e+02, percent-clipped=2.0 2023-03-27 00:09:56,513 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4960, 1.5382, 2.2040, 1.6789, 1.7728, 3.9653, 1.4641, 1.7173], device='cuda:3'), covar=tensor([0.0932, 0.1743, 0.1298, 0.1048, 0.1550, 0.0182, 0.1531, 0.1749], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0075, 0.0077, 0.0091, 0.0080, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 00:09:57,623 INFO [finetune.py:976] (3/7) Epoch 20, batch 1750, loss[loss=0.1723, simple_loss=0.2478, pruned_loss=0.04844, over 4861.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2464, pruned_loss=0.05347, over 952581.75 frames. ], batch size: 34, lr: 3.25e-03, grad_scale: 32.0 2023-03-27 00:10:20,396 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110611.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:10:30,986 INFO [finetune.py:976] (3/7) Epoch 20, batch 1800, loss[loss=0.2082, simple_loss=0.2834, pruned_loss=0.06656, over 4717.00 frames. ], tot_loss[loss=0.1789, simple_loss=0.2493, pruned_loss=0.05428, over 953747.75 frames. ], batch size: 59, lr: 3.25e-03, grad_scale: 32.0 2023-03-27 00:10:31,102 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6828, 2.6467, 2.4240, 2.5898, 2.5836, 5.0110, 2.5475, 3.0739], device='cuda:3'), covar=tensor([0.2428, 0.1888, 0.1654, 0.1693, 0.1088, 0.0117, 0.1833, 0.0852], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0115, 0.0119, 0.0122, 0.0113, 0.0095, 0.0095, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 00:10:38,897 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110639.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:10:55,666 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.008e+02 1.630e+02 1.887e+02 2.220e+02 3.285e+02, threshold=3.774e+02, percent-clipped=0.0 2023-03-27 00:11:01,585 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110659.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:11:11,044 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110672.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:11:11,645 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110673.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:11:12,218 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110674.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:11:14,002 INFO [finetune.py:976] (3/7) Epoch 20, batch 1850, loss[loss=0.1565, simple_loss=0.2326, pruned_loss=0.0402, over 4742.00 frames. ], tot_loss[loss=0.1801, simple_loss=0.2508, pruned_loss=0.05474, over 953288.14 frames. ], batch size: 54, lr: 3.25e-03, grad_scale: 32.0 2023-03-27 00:11:29,619 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110700.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:11:35,583 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7613, 1.1933, 0.8748, 1.5567, 2.0552, 1.4005, 1.4783, 1.6274], device='cuda:3'), covar=tensor([0.1471, 0.2103, 0.1926, 0.1219, 0.1940, 0.1967, 0.1457, 0.1950], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0095, 0.0111, 0.0092, 0.0120, 0.0093, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 00:11:44,149 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=110721.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:11:47,772 INFO [finetune.py:976] (3/7) Epoch 20, batch 1900, loss[loss=0.2017, simple_loss=0.2739, pruned_loss=0.06474, over 4827.00 frames. ], tot_loss[loss=0.1817, simple_loss=0.2525, pruned_loss=0.05549, over 954278.05 frames. ], batch size: 47, lr: 3.25e-03, grad_scale: 32.0 2023-03-27 00:11:53,711 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4213, 1.3910, 1.7891, 1.6733, 1.5506, 3.0809, 1.3379, 1.5248], device='cuda:3'), covar=tensor([0.0946, 0.1657, 0.1170, 0.0893, 0.1440, 0.0220, 0.1405, 0.1730], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0075, 0.0077, 0.0091, 0.0080, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 00:11:54,339 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110736.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:12:01,609 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=110748.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:12:06,290 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.024e+02 1.472e+02 1.826e+02 2.198e+02 3.929e+02, threshold=3.651e+02, percent-clipped=1.0 2023-03-27 00:12:14,048 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110766.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:12:17,567 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110771.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:12:21,622 INFO [finetune.py:976] (3/7) Epoch 20, batch 1950, loss[loss=0.1813, simple_loss=0.2548, pruned_loss=0.05388, over 4816.00 frames. ], tot_loss[loss=0.18, simple_loss=0.2504, pruned_loss=0.05475, over 952431.76 frames. ], batch size: 41, lr: 3.25e-03, grad_scale: 32.0 2023-03-27 00:12:34,814 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110797.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 00:12:46,184 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=110814.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:12:49,734 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=110819.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:12:55,561 INFO [finetune.py:976] (3/7) Epoch 20, batch 2000, loss[loss=0.1673, simple_loss=0.2296, pruned_loss=0.05248, over 4755.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2463, pruned_loss=0.05317, over 954988.27 frames. ], batch size: 28, lr: 3.25e-03, grad_scale: 32.0 2023-03-27 00:13:17,582 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.033e+02 1.482e+02 1.740e+02 2.016e+02 2.901e+02, threshold=3.480e+02, percent-clipped=0.0 2023-03-27 00:13:18,648 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-03-27 00:13:30,407 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110868.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:13:38,676 INFO [finetune.py:976] (3/7) Epoch 20, batch 2050, loss[loss=0.1308, simple_loss=0.2051, pruned_loss=0.0283, over 4754.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2439, pruned_loss=0.05242, over 956481.26 frames. ], batch size: 28, lr: 3.24e-03, grad_scale: 32.0 2023-03-27 00:14:04,560 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.0009, 4.3190, 4.5863, 4.7946, 4.7038, 4.5139, 5.1194, 1.5361], device='cuda:3'), covar=tensor([0.0860, 0.0933, 0.0723, 0.1039, 0.1432, 0.1734, 0.0619, 0.6184], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0244, 0.0277, 0.0290, 0.0332, 0.0282, 0.0301, 0.0296], 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') 2023-03-27 00:14:28,243 INFO [finetune.py:976] (3/7) Epoch 20, batch 2100, loss[loss=0.224, simple_loss=0.2798, pruned_loss=0.08408, over 4754.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2439, pruned_loss=0.05304, over 954414.46 frames. ], batch size: 26, lr: 3.24e-03, grad_scale: 32.0 2023-03-27 00:14:29,618 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110929.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:14:39,994 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110939.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:14:44,156 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1461, 2.0611, 1.6128, 2.1412, 1.9929, 1.7950, 2.4316, 2.1779], device='cuda:3'), covar=tensor([0.1370, 0.2217, 0.3235, 0.2702, 0.2778, 0.1723, 0.2973, 0.1771], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0190, 0.0236, 0.0255, 0.0249, 0.0205, 0.0216, 0.0202], 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') 2023-03-27 00:14:50,629 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.458e+01 1.520e+02 1.862e+02 2.217e+02 3.516e+02, threshold=3.725e+02, percent-clipped=1.0 2023-03-27 00:14:52,572 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110958.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:14:53,163 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110959.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:14:56,326 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-03-27 00:14:58,416 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110967.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:15:03,704 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110974.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:15:05,920 INFO [finetune.py:976] (3/7) Epoch 20, batch 2150, loss[loss=0.2622, simple_loss=0.3239, pruned_loss=0.1002, over 4783.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.247, pruned_loss=0.05397, over 954134.92 frames. ], batch size: 54, lr: 3.24e-03, grad_scale: 32.0 2023-03-27 00:15:12,590 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=110987.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:15:25,802 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=111007.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:15:30,206 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-03-27 00:15:33,651 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111019.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:15:35,880 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=111022.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:15:38,862 INFO [finetune.py:976] (3/7) Epoch 20, batch 2200, loss[loss=0.1367, simple_loss=0.2054, pruned_loss=0.03404, over 3973.00 frames. ], tot_loss[loss=0.1803, simple_loss=0.2504, pruned_loss=0.05511, over 954199.85 frames. ], batch size: 17, lr: 3.24e-03, grad_scale: 32.0 2023-03-27 00:15:52,453 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1803, 3.6161, 3.8160, 3.9539, 3.9625, 3.7410, 4.2592, 1.5222], device='cuda:3'), covar=tensor([0.0773, 0.0853, 0.0839, 0.1029, 0.1225, 0.1490, 0.0671, 0.5455], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0245, 0.0279, 0.0292, 0.0334, 0.0283, 0.0302, 0.0298], 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') 2023-03-27 00:15:59,013 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-03-27 00:16:00,234 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.032e+02 1.523e+02 1.854e+02 2.195e+02 4.707e+02, threshold=3.708e+02, percent-clipped=2.0 2023-03-27 00:16:20,835 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6294, 1.6256, 2.3494, 1.9425, 1.8251, 3.6524, 1.5431, 1.8605], device='cuda:3'), covar=tensor([0.0821, 0.1491, 0.1281, 0.0831, 0.1294, 0.0236, 0.1289, 0.1372], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0075, 0.0077, 0.0091, 0.0081, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 00:16:23,022 INFO [finetune.py:976] (3/7) Epoch 20, batch 2250, loss[loss=0.219, simple_loss=0.2842, pruned_loss=0.07695, over 4773.00 frames. ], tot_loss[loss=0.1811, simple_loss=0.2512, pruned_loss=0.05547, over 953916.24 frames. ], batch size: 51, lr: 3.24e-03, grad_scale: 32.0 2023-03-27 00:16:33,691 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111092.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 00:16:37,458 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2216, 1.9142, 1.9820, 0.8933, 2.1939, 2.4042, 2.1314, 1.7853], device='cuda:3'), covar=tensor([0.0901, 0.0764, 0.0482, 0.0701, 0.0487, 0.0642, 0.0459, 0.0738], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0150, 0.0125, 0.0124, 0.0130, 0.0129, 0.0140, 0.0147], device='cuda:3'), out_proj_covar=tensor([9.0343e-05, 1.0816e-04, 8.9122e-05, 8.7727e-05, 9.1555e-05, 9.1968e-05, 1.0078e-04, 1.0550e-04], device='cuda:3') 2023-03-27 00:16:56,461 INFO [finetune.py:976] (3/7) Epoch 20, batch 2300, loss[loss=0.1714, simple_loss=0.2497, pruned_loss=0.04649, over 4900.00 frames. ], tot_loss[loss=0.1799, simple_loss=0.2512, pruned_loss=0.05434, over 955337.34 frames. ], batch size: 37, lr: 3.24e-03, grad_scale: 32.0 2023-03-27 00:17:15,906 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.035e+02 1.510e+02 1.791e+02 2.077e+02 4.254e+02, threshold=3.582e+02, percent-clipped=2.0 2023-03-27 00:17:25,612 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.3594, 3.7921, 3.9739, 4.1589, 4.1245, 3.8271, 4.4712, 1.4067], device='cuda:3'), covar=tensor([0.0812, 0.0921, 0.0791, 0.1121, 0.1345, 0.1793, 0.0737, 0.5856], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0244, 0.0277, 0.0291, 0.0333, 0.0282, 0.0301, 0.0297], 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') 2023-03-27 00:17:30,237 INFO [finetune.py:976] (3/7) Epoch 20, batch 2350, loss[loss=0.2015, simple_loss=0.2663, pruned_loss=0.0683, over 4894.00 frames. ], tot_loss[loss=0.1787, simple_loss=0.2495, pruned_loss=0.05392, over 956559.90 frames. ], batch size: 32, lr: 3.24e-03, grad_scale: 32.0 2023-03-27 00:17:35,527 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4515, 2.6537, 2.3917, 1.8953, 2.4135, 2.6715, 2.8562, 2.2606], device='cuda:3'), covar=tensor([0.0638, 0.0601, 0.0775, 0.0891, 0.0968, 0.0751, 0.0583, 0.1045], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0135, 0.0140, 0.0120, 0.0124, 0.0139, 0.0140, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 00:17:38,387 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7302, 1.5256, 1.9569, 2.9587, 2.0401, 2.2706, 1.1856, 2.5384], device='cuda:3'), covar=tensor([0.1581, 0.1487, 0.1225, 0.0674, 0.0860, 0.1172, 0.1640, 0.0514], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0133, 0.0163, 0.0100, 0.0135, 0.0124, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 00:18:01,642 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111224.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:18:02,311 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111225.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:18:03,430 INFO [finetune.py:976] (3/7) Epoch 20, batch 2400, loss[loss=0.162, simple_loss=0.225, pruned_loss=0.04946, over 4279.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2469, pruned_loss=0.05362, over 954535.30 frames. ], batch size: 65, lr: 3.24e-03, grad_scale: 32.0 2023-03-27 00:18:19,334 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-03-27 00:18:22,742 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.826e+01 1.548e+02 1.775e+02 2.063e+02 3.363e+02, threshold=3.550e+02, percent-clipped=0.0 2023-03-27 00:18:30,965 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111267.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:18:36,922 INFO [finetune.py:976] (3/7) Epoch 20, batch 2450, loss[loss=0.2355, simple_loss=0.2943, pruned_loss=0.08837, over 4827.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2446, pruned_loss=0.05312, over 955325.66 frames. ], batch size: 40, lr: 3.24e-03, grad_scale: 32.0 2023-03-27 00:18:42,667 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7033, 1.3315, 0.8960, 1.6882, 2.1277, 1.4440, 1.5800, 1.7916], device='cuda:3'), covar=tensor([0.1592, 0.2201, 0.2025, 0.1228, 0.1970, 0.1825, 0.1492, 0.1973], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0095, 0.0111, 0.0092, 0.0120, 0.0093, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-27 00:18:50,048 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111286.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:18:52,507 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-27 00:19:21,763 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111314.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:19:22,367 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=111315.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:19:33,327 INFO [finetune.py:976] (3/7) Epoch 20, batch 2500, loss[loss=0.1898, simple_loss=0.2619, pruned_loss=0.05886, over 4845.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2449, pruned_loss=0.0531, over 956038.64 frames. ], batch size: 44, lr: 3.24e-03, grad_scale: 32.0 2023-03-27 00:19:57,313 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3582, 1.3727, 2.1362, 1.8186, 1.6709, 4.0059, 1.2340, 1.5033], device='cuda:3'), covar=tensor([0.1248, 0.2374, 0.1319, 0.1207, 0.1894, 0.0257, 0.2122, 0.2520], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0082, 0.0075, 0.0077, 0.0091, 0.0081, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 00:20:03,161 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.143e+02 1.591e+02 1.841e+02 2.119e+02 4.112e+02, threshold=3.683e+02, percent-clipped=1.0 2023-03-27 00:20:07,553 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6977, 1.5905, 1.9813, 1.3074, 1.8190, 2.0865, 1.5047, 2.1601], device='cuda:3'), covar=tensor([0.1352, 0.2071, 0.1577, 0.2102, 0.1049, 0.1428, 0.2921, 0.0889], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0203, 0.0189, 0.0188, 0.0173, 0.0211, 0.0217, 0.0199], 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') 2023-03-27 00:20:17,414 INFO [finetune.py:976] (3/7) Epoch 20, batch 2550, loss[loss=0.1764, simple_loss=0.2401, pruned_loss=0.05641, over 4809.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2479, pruned_loss=0.05352, over 953528.51 frames. ], batch size: 40, lr: 3.24e-03, grad_scale: 32.0 2023-03-27 00:20:27,554 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111392.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 00:20:36,460 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7370, 1.2378, 1.8285, 1.7955, 1.6233, 1.5534, 1.7437, 1.7103], device='cuda:3'), covar=tensor([0.3624, 0.3910, 0.3141, 0.3433, 0.4589, 0.3643, 0.4102, 0.2931], device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0243, 0.0264, 0.0283, 0.0280, 0.0254, 0.0290, 0.0245], 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') 2023-03-27 00:20:50,732 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7150, 1.5735, 2.1503, 3.5640, 2.3886, 2.5463, 0.9983, 2.9190], device='cuda:3'), covar=tensor([0.1707, 0.1446, 0.1421, 0.0498, 0.0789, 0.1331, 0.1953, 0.0427], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0116, 0.0134, 0.0164, 0.0101, 0.0135, 0.0124, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 00:20:51,266 INFO [finetune.py:976] (3/7) Epoch 20, batch 2600, loss[loss=0.1429, simple_loss=0.2193, pruned_loss=0.03322, over 4780.00 frames. ], tot_loss[loss=0.1797, simple_loss=0.2507, pruned_loss=0.0543, over 951982.70 frames. ], batch size: 26, lr: 3.24e-03, grad_scale: 64.0 2023-03-27 00:20:59,723 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=111440.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:21:10,162 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.233e+02 1.591e+02 1.878e+02 2.220e+02 5.233e+02, threshold=3.757e+02, percent-clipped=1.0 2023-03-27 00:21:20,906 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111468.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:21:31,397 INFO [finetune.py:976] (3/7) Epoch 20, batch 2650, loss[loss=0.193, simple_loss=0.2672, pruned_loss=0.05944, over 4821.00 frames. ], tot_loss[loss=0.1806, simple_loss=0.2517, pruned_loss=0.05479, over 951413.68 frames. ], batch size: 30, lr: 3.24e-03, grad_scale: 64.0 2023-03-27 00:21:37,310 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6938, 1.2676, 0.8431, 1.5880, 2.0062, 1.3934, 1.3950, 1.6302], device='cuda:3'), covar=tensor([0.1437, 0.2014, 0.2027, 0.1136, 0.1943, 0.2093, 0.1485, 0.1855], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0095, 0.0111, 0.0091, 0.0120, 0.0093, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-27 00:22:04,032 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-03-27 00:22:06,271 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111524.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:22:07,976 INFO [finetune.py:976] (3/7) Epoch 20, batch 2700, loss[loss=0.1932, simple_loss=0.2615, pruned_loss=0.06244, over 4869.00 frames. ], tot_loss[loss=0.181, simple_loss=0.2518, pruned_loss=0.05513, over 951031.01 frames. ], batch size: 34, lr: 3.24e-03, grad_scale: 64.0 2023-03-27 00:22:10,258 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111529.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:22:25,112 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6353, 1.5394, 1.4625, 1.6080, 1.2609, 3.3256, 1.3064, 1.7520], device='cuda:3'), covar=tensor([0.3786, 0.2688, 0.2361, 0.2615, 0.1680, 0.0222, 0.2539, 0.1221], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0120, 0.0123, 0.0113, 0.0096, 0.0096, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 00:22:25,720 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.2543, 2.8219, 2.6705, 1.3232, 2.8049, 2.3314, 2.2733, 2.6021], device='cuda:3'), covar=tensor([0.0854, 0.0807, 0.1468, 0.2030, 0.1509, 0.1883, 0.1804, 0.1019], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0191, 0.0200, 0.0182, 0.0211, 0.0206, 0.0222, 0.0195], 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') 2023-03-27 00:22:27,296 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.466e+01 1.581e+02 1.832e+02 2.307e+02 3.346e+02, threshold=3.664e+02, percent-clipped=0.0 2023-03-27 00:22:38,120 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=111572.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:22:41,150 INFO [finetune.py:976] (3/7) Epoch 20, batch 2750, loss[loss=0.1548, simple_loss=0.229, pruned_loss=0.04035, over 4797.00 frames. ], tot_loss[loss=0.1783, simple_loss=0.2489, pruned_loss=0.05383, over 952510.74 frames. ], batch size: 29, lr: 3.24e-03, grad_scale: 64.0 2023-03-27 00:22:44,075 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111581.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:23:06,496 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111614.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:23:14,366 INFO [finetune.py:976] (3/7) Epoch 20, batch 2800, loss[loss=0.1101, simple_loss=0.1865, pruned_loss=0.01683, over 4844.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2459, pruned_loss=0.05288, over 952479.69 frames. ], batch size: 47, lr: 3.24e-03, grad_scale: 64.0 2023-03-27 00:23:32,757 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.029e+01 1.565e+02 1.748e+02 2.177e+02 3.583e+02, threshold=3.496e+02, percent-clipped=0.0 2023-03-27 00:23:38,053 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=111662.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:23:47,153 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-03-27 00:23:48,042 INFO [finetune.py:976] (3/7) Epoch 20, batch 2850, loss[loss=0.1851, simple_loss=0.259, pruned_loss=0.05557, over 4855.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2453, pruned_loss=0.05319, over 954158.42 frames. ], batch size: 44, lr: 3.24e-03, grad_scale: 64.0 2023-03-27 00:23:49,736 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-27 00:23:53,596 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5031, 1.5618, 2.2628, 1.7730, 1.7521, 4.0050, 1.4214, 1.7898], device='cuda:3'), covar=tensor([0.0950, 0.1757, 0.1080, 0.0958, 0.1533, 0.0188, 0.1500, 0.1656], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0082, 0.0074, 0.0077, 0.0091, 0.0081, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 00:24:41,629 INFO [finetune.py:976] (3/7) Epoch 20, batch 2900, loss[loss=0.2291, simple_loss=0.3, pruned_loss=0.07914, over 4905.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2466, pruned_loss=0.05299, over 954352.39 frames. ], batch size: 36, lr: 3.24e-03, grad_scale: 64.0 2023-03-27 00:25:12,980 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.101e+02 1.542e+02 1.820e+02 2.254e+02 5.949e+02, threshold=3.641e+02, percent-clipped=1.0 2023-03-27 00:25:31,432 INFO [finetune.py:976] (3/7) Epoch 20, batch 2950, loss[loss=0.1399, simple_loss=0.1997, pruned_loss=0.04002, over 3975.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.248, pruned_loss=0.05308, over 951833.38 frames. ], batch size: 17, lr: 3.24e-03, grad_scale: 64.0 2023-03-27 00:25:40,060 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111791.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 00:26:03,393 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111824.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:26:05,121 INFO [finetune.py:976] (3/7) Epoch 20, batch 3000, loss[loss=0.1737, simple_loss=0.2445, pruned_loss=0.05143, over 4841.00 frames. ], tot_loss[loss=0.1794, simple_loss=0.25, pruned_loss=0.0544, over 951403.64 frames. ], batch size: 49, lr: 3.24e-03, grad_scale: 64.0 2023-03-27 00:26:05,121 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-27 00:26:20,305 INFO [finetune.py:1010] (3/7) Epoch 20, validation: loss=0.1563, simple_loss=0.2257, pruned_loss=0.04344, over 2265189.00 frames. 2023-03-27 00:26:20,306 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6434MB 2023-03-27 00:26:40,762 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111852.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 00:26:48,439 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.202e+02 1.618e+02 1.875e+02 2.230e+02 3.575e+02, threshold=3.749e+02, percent-clipped=0.0 2023-03-27 00:27:11,149 INFO [finetune.py:976] (3/7) Epoch 20, batch 3050, loss[loss=0.1965, simple_loss=0.2688, pruned_loss=0.06211, over 4892.00 frames. ], tot_loss[loss=0.1794, simple_loss=0.2503, pruned_loss=0.05424, over 952302.29 frames. ], batch size: 43, lr: 3.24e-03, grad_scale: 32.0 2023-03-27 00:27:18,694 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111881.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:27:29,014 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-03-27 00:28:10,120 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5652, 1.4034, 1.4501, 1.4886, 1.1849, 2.9653, 1.1696, 1.5697], device='cuda:3'), covar=tensor([0.3344, 0.2483, 0.2110, 0.2427, 0.1606, 0.0257, 0.2630, 0.1186], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0115, 0.0120, 0.0123, 0.0113, 0.0096, 0.0095, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 00:28:14,418 INFO [finetune.py:976] (3/7) Epoch 20, batch 3100, loss[loss=0.1649, simple_loss=0.2359, pruned_loss=0.04701, over 4810.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2487, pruned_loss=0.0534, over 954484.83 frames. ], batch size: 39, lr: 3.24e-03, grad_scale: 32.0 2023-03-27 00:28:15,440 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-27 00:28:15,707 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=111929.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:28:26,162 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-03-27 00:28:33,001 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.135e+02 1.507e+02 1.737e+02 2.301e+02 5.151e+02, threshold=3.474e+02, percent-clipped=3.0 2023-03-27 00:28:39,691 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3597, 2.4593, 1.9694, 2.6854, 2.4056, 2.0434, 2.9758, 2.5299], device='cuda:3'), covar=tensor([0.1239, 0.1990, 0.2513, 0.2066, 0.2034, 0.1456, 0.2760, 0.1499], device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0187, 0.0232, 0.0250, 0.0244, 0.0202, 0.0212, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 00:28:40,236 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.0066, 4.3702, 4.5486, 4.8290, 4.7564, 4.4264, 5.1528, 1.7597], device='cuda:3'), covar=tensor([0.0720, 0.0779, 0.0671, 0.0795, 0.1173, 0.1570, 0.0493, 0.5300], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0243, 0.0276, 0.0290, 0.0332, 0.0282, 0.0302, 0.0297], 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') 2023-03-27 00:28:47,144 INFO [finetune.py:976] (3/7) Epoch 20, batch 3150, loss[loss=0.1829, simple_loss=0.2477, pruned_loss=0.05904, over 4822.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2471, pruned_loss=0.05316, over 956431.22 frames. ], batch size: 30, lr: 3.24e-03, grad_scale: 32.0 2023-03-27 00:28:54,905 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6150, 1.5105, 1.9720, 3.1375, 2.0633, 2.3546, 1.0195, 2.6765], device='cuda:3'), covar=tensor([0.1923, 0.1588, 0.1553, 0.0820, 0.0926, 0.1418, 0.2075, 0.0542], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0133, 0.0163, 0.0100, 0.0134, 0.0123, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 00:29:14,186 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6477, 1.5131, 1.0688, 0.2222, 1.1917, 1.4785, 1.4878, 1.4746], device='cuda:3'), covar=tensor([0.0947, 0.0828, 0.1453, 0.2024, 0.1468, 0.2680, 0.2251, 0.0912], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0192, 0.0199, 0.0183, 0.0211, 0.0207, 0.0222, 0.0195], 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') 2023-03-27 00:29:23,069 INFO [finetune.py:976] (3/7) Epoch 20, batch 3200, loss[loss=0.2086, simple_loss=0.2659, pruned_loss=0.07566, over 4905.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2452, pruned_loss=0.05283, over 956686.36 frames. ], batch size: 35, lr: 3.24e-03, grad_scale: 32.0 2023-03-27 00:29:56,972 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.705e+01 1.565e+02 1.739e+02 2.228e+02 3.922e+02, threshold=3.479e+02, percent-clipped=1.0 2023-03-27 00:30:06,304 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112066.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:30:13,836 INFO [finetune.py:976] (3/7) Epoch 20, batch 3250, loss[loss=0.1605, simple_loss=0.2267, pruned_loss=0.04714, over 4737.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2465, pruned_loss=0.05402, over 956330.59 frames. ], batch size: 23, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:30:54,495 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112124.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:30:56,192 INFO [finetune.py:976] (3/7) Epoch 20, batch 3300, loss[loss=0.1815, simple_loss=0.2588, pruned_loss=0.05209, over 4770.00 frames. ], tot_loss[loss=0.18, simple_loss=0.2503, pruned_loss=0.05481, over 956780.60 frames. ], batch size: 28, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:30:56,305 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112127.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:31:10,260 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112147.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 00:31:16,114 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.319e+01 1.653e+02 1.982e+02 2.472e+02 3.934e+02, threshold=3.965e+02, percent-clipped=2.0 2023-03-27 00:31:24,583 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-27 00:31:26,940 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=112172.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:31:29,978 INFO [finetune.py:976] (3/7) Epoch 20, batch 3350, loss[loss=0.2278, simple_loss=0.2872, pruned_loss=0.08423, over 4760.00 frames. ], tot_loss[loss=0.1821, simple_loss=0.2529, pruned_loss=0.05568, over 955858.19 frames. ], batch size: 54, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:32:11,815 INFO [finetune.py:976] (3/7) Epoch 20, batch 3400, loss[loss=0.182, simple_loss=0.2644, pruned_loss=0.04977, over 4811.00 frames. ], tot_loss[loss=0.1828, simple_loss=0.2537, pruned_loss=0.05598, over 955728.29 frames. ], batch size: 40, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:32:19,200 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6564, 0.6720, 1.7076, 1.5929, 1.5066, 1.4130, 1.4969, 1.6247], device='cuda:3'), covar=tensor([0.3194, 0.3289, 0.2718, 0.2961, 0.3864, 0.3093, 0.3608, 0.2549], device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0242, 0.0264, 0.0282, 0.0279, 0.0254, 0.0289, 0.0244], 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') 2023-03-27 00:32:31,198 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.110e+02 1.676e+02 1.964e+02 2.392e+02 4.564e+02, threshold=3.928e+02, percent-clipped=2.0 2023-03-27 00:32:44,377 INFO [finetune.py:976] (3/7) Epoch 20, batch 3450, loss[loss=0.153, simple_loss=0.2291, pruned_loss=0.03842, over 4830.00 frames. ], tot_loss[loss=0.1827, simple_loss=0.2532, pruned_loss=0.0561, over 955199.59 frames. ], batch size: 33, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:33:08,054 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.8500, 3.3527, 3.5158, 3.7263, 3.6356, 3.4238, 3.9367, 1.2449], device='cuda:3'), covar=tensor([0.0873, 0.0901, 0.0969, 0.0980, 0.1252, 0.1618, 0.0841, 0.5144], device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0245, 0.0280, 0.0294, 0.0335, 0.0285, 0.0307, 0.0300], 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') 2023-03-27 00:33:19,433 INFO [finetune.py:976] (3/7) Epoch 20, batch 3500, loss[loss=0.1336, simple_loss=0.2103, pruned_loss=0.02845, over 4838.00 frames. ], tot_loss[loss=0.1803, simple_loss=0.2498, pruned_loss=0.05536, over 953933.58 frames. ], batch size: 47, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:33:19,537 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9768, 1.6307, 2.2994, 1.5271, 2.1200, 2.2168, 1.6100, 2.3534], device='cuda:3'), covar=tensor([0.1415, 0.2070, 0.1619, 0.2024, 0.0956, 0.1524, 0.2716, 0.0924], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0205, 0.0192, 0.0190, 0.0175, 0.0213, 0.0220, 0.0202], 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') 2023-03-27 00:33:56,444 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.856e+01 1.454e+02 1.799e+02 2.150e+02 5.052e+02, threshold=3.598e+02, percent-clipped=2.0 2023-03-27 00:33:57,834 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1550, 1.9391, 1.7027, 1.9531, 1.8844, 1.8805, 1.8977, 2.6419], device='cuda:3'), covar=tensor([0.3482, 0.4232, 0.3243, 0.4076, 0.4192, 0.2441, 0.3990, 0.1681], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0262, 0.0231, 0.0276, 0.0251, 0.0221, 0.0251, 0.0232], 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') 2023-03-27 00:34:18,955 INFO [finetune.py:976] (3/7) Epoch 20, batch 3550, loss[loss=0.1598, simple_loss=0.2289, pruned_loss=0.04535, over 4816.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2463, pruned_loss=0.0546, over 955031.03 frames. ], batch size: 41, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:34:36,592 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112389.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:34:46,949 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-27 00:34:57,719 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.62 vs. limit=2.0 2023-03-27 00:35:17,935 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112422.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:35:20,889 INFO [finetune.py:976] (3/7) Epoch 20, batch 3600, loss[loss=0.1365, simple_loss=0.2068, pruned_loss=0.03314, over 4899.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2432, pruned_loss=0.05319, over 956962.75 frames. ], batch size: 32, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:35:37,947 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112447.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 00:35:38,570 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7082, 2.7697, 2.5755, 1.9694, 2.5863, 2.8969, 2.9933, 2.3332], device='cuda:3'), covar=tensor([0.0597, 0.0582, 0.0718, 0.0825, 0.0705, 0.0615, 0.0552, 0.0984], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0136, 0.0140, 0.0120, 0.0124, 0.0138, 0.0140, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 00:35:39,813 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112450.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:35:44,342 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.087e+02 1.527e+02 1.833e+02 2.137e+02 4.874e+02, threshold=3.667e+02, percent-clipped=1.0 2023-03-27 00:36:08,331 INFO [finetune.py:976] (3/7) Epoch 20, batch 3650, loss[loss=0.2037, simple_loss=0.2832, pruned_loss=0.06217, over 4829.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2449, pruned_loss=0.05353, over 954387.06 frames. ], batch size: 39, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:36:20,411 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=112495.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 00:36:41,250 INFO [finetune.py:976] (3/7) Epoch 20, batch 3700, loss[loss=0.231, simple_loss=0.2967, pruned_loss=0.08265, over 4917.00 frames. ], tot_loss[loss=0.1783, simple_loss=0.2479, pruned_loss=0.0544, over 955233.77 frames. ], batch size: 38, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:36:47,473 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3219, 2.2981, 2.1382, 2.5090, 2.8294, 2.5421, 2.3091, 2.0320], device='cuda:3'), covar=tensor([0.1896, 0.1842, 0.1623, 0.1366, 0.1319, 0.0929, 0.1753, 0.1635], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0210, 0.0212, 0.0194, 0.0243, 0.0188, 0.0217, 0.0203], 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') 2023-03-27 00:36:56,150 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5805, 1.6301, 1.3172, 1.5143, 1.9285, 1.7935, 1.6290, 1.4367], device='cuda:3'), covar=tensor([0.0327, 0.0274, 0.0573, 0.0294, 0.0187, 0.0413, 0.0314, 0.0389], device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0107, 0.0144, 0.0111, 0.0100, 0.0111, 0.0100, 0.0112], device='cuda:3'), out_proj_covar=tensor([7.5111e-05, 8.2412e-05, 1.1350e-04, 8.5323e-05, 7.8004e-05, 8.2423e-05, 7.4177e-05, 8.5790e-05], device='cuda:3') 2023-03-27 00:37:01,295 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.006e+02 1.568e+02 1.941e+02 2.323e+02 3.962e+02, threshold=3.882e+02, percent-clipped=1.0 2023-03-27 00:37:05,579 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112562.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:37:09,275 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-27 00:37:12,585 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1136, 1.8131, 2.2354, 2.1476, 1.8990, 1.9129, 2.1289, 2.1100], device='cuda:3'), covar=tensor([0.4654, 0.4433, 0.3342, 0.3946, 0.4939, 0.3978, 0.5137, 0.3309], device='cuda:3'), in_proj_covar=tensor([0.0254, 0.0243, 0.0263, 0.0282, 0.0279, 0.0255, 0.0290, 0.0245], 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') 2023-03-27 00:37:15,914 INFO [finetune.py:976] (3/7) Epoch 20, batch 3750, loss[loss=0.2004, simple_loss=0.2662, pruned_loss=0.06727, over 4920.00 frames. ], tot_loss[loss=0.1805, simple_loss=0.2505, pruned_loss=0.05525, over 954684.08 frames. ], batch size: 38, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:37:16,030 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112577.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:37:50,112 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112617.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:37:54,220 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112623.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:37:57,418 INFO [finetune.py:976] (3/7) Epoch 20, batch 3800, loss[loss=0.1734, simple_loss=0.2583, pruned_loss=0.04425, over 4861.00 frames. ], tot_loss[loss=0.1808, simple_loss=0.251, pruned_loss=0.05533, over 953038.25 frames. ], batch size: 31, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:38:04,247 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112638.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 00:38:16,043 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.507e+01 1.515e+02 1.800e+02 2.233e+02 3.828e+02, threshold=3.600e+02, percent-clipped=0.0 2023-03-27 00:38:17,474 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.62 vs. limit=5.0 2023-03-27 00:38:28,503 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0205, 1.9404, 1.9643, 1.3119, 1.9723, 2.0723, 2.0773, 1.5632], device='cuda:3'), covar=tensor([0.0591, 0.0686, 0.0763, 0.0900, 0.0782, 0.0684, 0.0606, 0.1222], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0136, 0.0140, 0.0120, 0.0125, 0.0139, 0.0140, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 00:38:30,666 INFO [finetune.py:976] (3/7) Epoch 20, batch 3850, loss[loss=0.1901, simple_loss=0.2703, pruned_loss=0.05491, over 4816.00 frames. ], tot_loss[loss=0.1799, simple_loss=0.25, pruned_loss=0.05484, over 954115.63 frames. ], batch size: 39, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:38:31,366 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112678.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 00:38:59,742 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112722.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:39:03,155 INFO [finetune.py:976] (3/7) Epoch 20, batch 3900, loss[loss=0.2048, simple_loss=0.2717, pruned_loss=0.06897, over 4829.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.2479, pruned_loss=0.05423, over 954473.01 frames. ], batch size: 39, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:39:15,046 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112745.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:39:21,551 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.403e+01 1.612e+02 1.959e+02 2.410e+02 4.123e+02, threshold=3.918e+02, percent-clipped=1.0 2023-03-27 00:39:32,077 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=112770.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:39:36,282 INFO [finetune.py:976] (3/7) Epoch 20, batch 3950, loss[loss=0.1481, simple_loss=0.2156, pruned_loss=0.04033, over 4830.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2444, pruned_loss=0.05268, over 956060.08 frames. ], batch size: 30, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:39:52,592 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8350, 1.6573, 2.0324, 1.4280, 1.9587, 2.1611, 1.5119, 2.2874], device='cuda:3'), covar=tensor([0.1165, 0.1991, 0.1257, 0.1764, 0.0898, 0.1229, 0.2850, 0.0719], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0205, 0.0191, 0.0190, 0.0175, 0.0213, 0.0219, 0.0201], 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') 2023-03-27 00:40:19,040 INFO [finetune.py:976] (3/7) Epoch 20, batch 4000, loss[loss=0.2064, simple_loss=0.2856, pruned_loss=0.06359, over 4915.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2444, pruned_loss=0.05267, over 956630.84 frames. ], batch size: 37, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:40:48,438 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.088e+02 1.638e+02 1.904e+02 2.320e+02 3.891e+02, threshold=3.808e+02, percent-clipped=0.0 2023-03-27 00:41:04,948 INFO [finetune.py:976] (3/7) Epoch 20, batch 4050, loss[loss=0.1768, simple_loss=0.2554, pruned_loss=0.04912, over 4754.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2472, pruned_loss=0.05359, over 955537.12 frames. ], batch size: 27, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:41:33,625 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-27 00:41:41,243 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112918.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:41:43,126 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112921.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:41:47,212 INFO [finetune.py:976] (3/7) Epoch 20, batch 4100, loss[loss=0.1823, simple_loss=0.2549, pruned_loss=0.05487, over 4747.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2492, pruned_loss=0.05328, over 954417.94 frames. ], batch size: 54, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:41:51,359 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112933.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 00:42:06,078 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.082e+02 1.570e+02 1.869e+02 2.355e+02 4.214e+02, threshold=3.739e+02, percent-clipped=0.0 2023-03-27 00:42:17,473 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112973.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 00:42:19,819 INFO [finetune.py:976] (3/7) Epoch 20, batch 4150, loss[loss=0.1866, simple_loss=0.2424, pruned_loss=0.06537, over 4383.00 frames. ], tot_loss[loss=0.1809, simple_loss=0.2521, pruned_loss=0.05486, over 953807.30 frames. ], batch size: 19, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:42:23,987 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112982.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:43:03,966 INFO [finetune.py:976] (3/7) Epoch 20, batch 4200, loss[loss=0.1615, simple_loss=0.2459, pruned_loss=0.03859, over 4846.00 frames. ], tot_loss[loss=0.1805, simple_loss=0.2524, pruned_loss=0.05435, over 954809.02 frames. ], batch size: 47, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:43:16,384 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113045.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:43:23,432 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.027e+02 1.578e+02 1.900e+02 2.258e+02 5.235e+02, threshold=3.800e+02, percent-clipped=4.0 2023-03-27 00:43:35,249 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.57 vs. limit=5.0 2023-03-27 00:43:36,994 INFO [finetune.py:976] (3/7) Epoch 20, batch 4250, loss[loss=0.1839, simple_loss=0.2583, pruned_loss=0.05479, over 4924.00 frames. ], tot_loss[loss=0.1792, simple_loss=0.2505, pruned_loss=0.05396, over 954206.17 frames. ], batch size: 33, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:43:47,717 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=113093.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:44:10,286 INFO [finetune.py:976] (3/7) Epoch 20, batch 4300, loss[loss=0.1823, simple_loss=0.2557, pruned_loss=0.05446, over 4911.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2477, pruned_loss=0.05368, over 955104.65 frames. ], batch size: 43, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:44:15,148 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7662, 1.3585, 0.9244, 1.6094, 2.1459, 1.5926, 1.5195, 1.8171], device='cuda:3'), covar=tensor([0.1466, 0.1970, 0.1963, 0.1228, 0.1906, 0.1819, 0.1464, 0.1873], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0094, 0.0110, 0.0091, 0.0119, 0.0093, 0.0097, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-27 00:44:30,859 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.679e+01 1.396e+02 1.741e+02 2.023e+02 4.034e+02, threshold=3.482e+02, percent-clipped=1.0 2023-03-27 00:44:43,636 INFO [finetune.py:976] (3/7) Epoch 20, batch 4350, loss[loss=0.2054, simple_loss=0.2549, pruned_loss=0.07792, over 4903.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2444, pruned_loss=0.05278, over 955966.47 frames. ], batch size: 43, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:44:54,751 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.0893, 4.4189, 4.6275, 4.9243, 4.8162, 4.5258, 5.2030, 1.5029], device='cuda:3'), covar=tensor([0.0691, 0.0751, 0.0682, 0.0845, 0.1164, 0.1579, 0.0540, 0.6093], device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0242, 0.0277, 0.0291, 0.0330, 0.0283, 0.0303, 0.0297], 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') 2023-03-27 00:45:06,172 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.1781, 3.2023, 2.9283, 2.2751, 2.9897, 3.3394, 3.4436, 2.7296], device='cuda:3'), covar=tensor([0.0424, 0.0421, 0.0561, 0.0766, 0.0514, 0.0466, 0.0414, 0.0773], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0134, 0.0138, 0.0119, 0.0123, 0.0137, 0.0139, 0.0159], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 00:45:12,209 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113218.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:45:17,616 INFO [finetune.py:976] (3/7) Epoch 20, batch 4400, loss[loss=0.1651, simple_loss=0.2377, pruned_loss=0.0463, over 4853.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.245, pruned_loss=0.0532, over 955561.02 frames. ], batch size: 31, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:45:21,832 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113233.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:45:22,468 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2005, 2.0435, 2.1270, 1.5310, 2.1265, 2.2688, 2.2444, 1.6983], device='cuda:3'), covar=tensor([0.0547, 0.0667, 0.0602, 0.0845, 0.0706, 0.0648, 0.0572, 0.1177], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0134, 0.0138, 0.0119, 0.0123, 0.0137, 0.0139, 0.0159], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 00:45:22,476 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113234.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:45:49,232 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.729e+01 1.603e+02 1.846e+02 2.173e+02 5.642e+02, threshold=3.692e+02, percent-clipped=4.0 2023-03-27 00:45:49,372 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0925, 1.7879, 2.2647, 1.5132, 2.1385, 2.2516, 1.6867, 2.4579], device='cuda:3'), covar=tensor([0.1322, 0.2022, 0.1406, 0.1999, 0.0945, 0.1359, 0.2836, 0.0841], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0204, 0.0189, 0.0188, 0.0174, 0.0211, 0.0218, 0.0199], 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') 2023-03-27 00:46:00,754 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=113266.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:46:08,335 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113273.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 00:46:10,701 INFO [finetune.py:976] (3/7) Epoch 20, batch 4450, loss[loss=0.1862, simple_loss=0.2652, pruned_loss=0.05357, over 4854.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2477, pruned_loss=0.05345, over 955447.17 frames. ], batch size: 44, lr: 3.23e-03, grad_scale: 32.0 2023-03-27 00:46:10,769 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113277.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:46:13,203 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=113281.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:46:23,177 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113295.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:46:50,259 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=113321.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:46:53,837 INFO [finetune.py:976] (3/7) Epoch 20, batch 4500, loss[loss=0.2205, simple_loss=0.2761, pruned_loss=0.08249, over 4886.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2487, pruned_loss=0.0535, over 956429.58 frames. ], batch size: 35, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 00:46:58,924 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6863, 1.7016, 1.4525, 1.8945, 2.1839, 1.8729, 1.5523, 1.3907], device='cuda:3'), covar=tensor([0.2585, 0.2237, 0.2253, 0.1765, 0.1821, 0.1368, 0.2697, 0.2273], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0210, 0.0212, 0.0194, 0.0242, 0.0187, 0.0217, 0.0203], 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') 2023-03-27 00:47:13,422 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.100e+02 1.617e+02 2.027e+02 2.372e+02 5.258e+02, threshold=4.055e+02, percent-clipped=2.0 2023-03-27 00:47:27,583 INFO [finetune.py:976] (3/7) Epoch 20, batch 4550, loss[loss=0.2166, simple_loss=0.2886, pruned_loss=0.07233, over 4819.00 frames. ], tot_loss[loss=0.1792, simple_loss=0.2502, pruned_loss=0.05409, over 954940.19 frames. ], batch size: 38, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 00:47:38,622 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.6508, 3.9547, 4.2305, 4.5097, 4.3544, 4.1117, 4.7390, 1.5695], device='cuda:3'), covar=tensor([0.0691, 0.0768, 0.0819, 0.0952, 0.1164, 0.1520, 0.0544, 0.5463], device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0245, 0.0280, 0.0294, 0.0335, 0.0287, 0.0307, 0.0301], 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') 2023-03-27 00:47:38,898 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.21 vs. limit=5.0 2023-03-27 00:48:03,324 INFO [finetune.py:976] (3/7) Epoch 20, batch 4600, loss[loss=0.1787, simple_loss=0.2621, pruned_loss=0.04762, over 4713.00 frames. ], tot_loss[loss=0.1797, simple_loss=0.2506, pruned_loss=0.0544, over 955522.16 frames. ], batch size: 54, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 00:48:31,339 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.757e+01 1.614e+02 1.832e+02 2.145e+02 3.668e+02, threshold=3.664e+02, percent-clipped=0.0 2023-03-27 00:48:45,584 INFO [finetune.py:976] (3/7) Epoch 20, batch 4650, loss[loss=0.1594, simple_loss=0.233, pruned_loss=0.04292, over 4918.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2478, pruned_loss=0.05325, over 956119.04 frames. ], batch size: 37, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 00:49:10,269 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.9769, 4.2939, 4.4562, 4.7929, 4.6639, 4.3514, 5.0911, 1.5596], device='cuda:3'), covar=tensor([0.0795, 0.0816, 0.0867, 0.0980, 0.1259, 0.1661, 0.0525, 0.5994], device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0245, 0.0281, 0.0294, 0.0335, 0.0287, 0.0306, 0.0300], 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') 2023-03-27 00:49:19,010 INFO [finetune.py:976] (3/7) Epoch 20, batch 4700, loss[loss=0.1171, simple_loss=0.1886, pruned_loss=0.02278, over 4783.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2443, pruned_loss=0.05189, over 955604.75 frames. ], batch size: 26, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 00:49:37,205 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.713e+01 1.518e+02 1.819e+02 2.172e+02 5.123e+02, threshold=3.639e+02, percent-clipped=2.0 2023-03-27 00:49:51,489 INFO [finetune.py:976] (3/7) Epoch 20, batch 4750, loss[loss=0.1942, simple_loss=0.2531, pruned_loss=0.06762, over 4861.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2429, pruned_loss=0.0519, over 955493.60 frames. ], batch size: 44, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 00:49:52,071 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113577.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:50:00,075 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113590.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:50:18,341 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113617.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:50:23,691 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=113625.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:50:24,847 INFO [finetune.py:976] (3/7) Epoch 20, batch 4800, loss[loss=0.1498, simple_loss=0.2072, pruned_loss=0.04623, over 4176.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2446, pruned_loss=0.05273, over 952887.56 frames. ], batch size: 18, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 00:50:37,471 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113646.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:50:43,817 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.144e+02 1.583e+02 1.918e+02 2.188e+02 4.674e+02, threshold=3.836e+02, percent-clipped=2.0 2023-03-27 00:50:48,782 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.41 vs. limit=5.0 2023-03-27 00:51:01,327 INFO [finetune.py:976] (3/7) Epoch 20, batch 4850, loss[loss=0.1838, simple_loss=0.2629, pruned_loss=0.05235, over 4766.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2482, pruned_loss=0.05345, over 953434.78 frames. ], batch size: 26, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 00:51:02,064 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113678.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:51:13,556 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.3378, 2.9332, 3.1163, 3.2860, 3.1577, 2.9112, 3.3725, 1.0280], device='cuda:3'), covar=tensor([0.1077, 0.1039, 0.1237, 0.1196, 0.1554, 0.1999, 0.1071, 0.5584], device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0245, 0.0280, 0.0294, 0.0335, 0.0287, 0.0306, 0.0300], 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') 2023-03-27 00:51:34,436 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113707.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:51:43,390 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-27 00:51:57,945 INFO [finetune.py:976] (3/7) Epoch 20, batch 4900, loss[loss=0.23, simple_loss=0.291, pruned_loss=0.08451, over 4880.00 frames. ], tot_loss[loss=0.1788, simple_loss=0.2497, pruned_loss=0.05394, over 952945.03 frames. ], batch size: 35, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 00:52:20,123 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.029e+02 1.649e+02 2.053e+02 2.498e+02 4.513e+02, threshold=4.105e+02, percent-clipped=3.0 2023-03-27 00:52:34,812 INFO [finetune.py:976] (3/7) Epoch 20, batch 4950, loss[loss=0.1844, simple_loss=0.2564, pruned_loss=0.05617, over 4816.00 frames. ], tot_loss[loss=0.1825, simple_loss=0.2531, pruned_loss=0.05594, over 953497.33 frames. ], batch size: 38, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 00:53:07,553 INFO [finetune.py:976] (3/7) Epoch 20, batch 5000, loss[loss=0.1687, simple_loss=0.2341, pruned_loss=0.05159, over 4822.00 frames. ], tot_loss[loss=0.1797, simple_loss=0.2501, pruned_loss=0.05461, over 953080.40 frames. ], batch size: 30, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 00:53:10,755 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0 2023-03-27 00:53:26,537 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.026e+02 1.516e+02 1.779e+02 2.023e+02 5.156e+02, threshold=3.559e+02, percent-clipped=2.0 2023-03-27 00:53:36,246 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7756, 1.2370, 0.9141, 1.6542, 2.3503, 1.5124, 1.4784, 1.7769], device='cuda:3'), covar=tensor([0.1484, 0.2142, 0.1888, 0.1218, 0.1618, 0.1776, 0.1503, 0.1924], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0094, 0.0109, 0.0090, 0.0118, 0.0092, 0.0096, 0.0087], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-27 00:53:42,068 INFO [finetune.py:976] (3/7) Epoch 20, batch 5050, loss[loss=0.1985, simple_loss=0.2648, pruned_loss=0.06612, over 4897.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2471, pruned_loss=0.05362, over 952304.40 frames. ], batch size: 43, lr: 3.22e-03, grad_scale: 64.0 2023-03-27 00:53:46,364 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1588, 2.0589, 1.7506, 1.9598, 1.9675, 1.9509, 2.0113, 2.6586], device='cuda:3'), covar=tensor([0.3603, 0.4228, 0.3269, 0.3586, 0.3796, 0.2322, 0.3672, 0.1678], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0261, 0.0230, 0.0276, 0.0251, 0.0221, 0.0251, 0.0233], 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') 2023-03-27 00:53:51,006 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113890.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:54:14,769 INFO [finetune.py:976] (3/7) Epoch 20, batch 5100, loss[loss=0.1459, simple_loss=0.2155, pruned_loss=0.03818, over 4935.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2436, pruned_loss=0.05238, over 953945.42 frames. ], batch size: 38, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 00:54:23,010 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=113938.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:54:35,763 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.073e+02 1.567e+02 1.826e+02 2.193e+02 3.507e+02, threshold=3.652e+02, percent-clipped=0.0 2023-03-27 00:54:45,928 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.47 vs. limit=5.0 2023-03-27 00:54:46,072 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113973.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:54:48,407 INFO [finetune.py:976] (3/7) Epoch 20, batch 5150, loss[loss=0.1284, simple_loss=0.212, pruned_loss=0.02244, over 4742.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2424, pruned_loss=0.05174, over 954363.18 frames. ], batch size: 27, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 00:55:07,779 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114002.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:55:15,380 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.55 vs. limit=5.0 2023-03-27 00:55:16,319 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6927, 1.7165, 1.4162, 1.8759, 2.3964, 1.8761, 1.6247, 1.3925], device='cuda:3'), covar=tensor([0.2335, 0.2099, 0.2171, 0.1706, 0.1616, 0.1255, 0.2339, 0.2046], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0209, 0.0211, 0.0193, 0.0242, 0.0186, 0.0216, 0.0201], 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') 2023-03-27 00:55:23,296 INFO [finetune.py:976] (3/7) Epoch 20, batch 5200, loss[loss=0.1834, simple_loss=0.2588, pruned_loss=0.05397, over 4896.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2453, pruned_loss=0.05272, over 953215.23 frames. ], batch size: 36, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 00:55:43,820 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.106e+02 1.581e+02 1.814e+02 2.243e+02 3.815e+02, threshold=3.628e+02, percent-clipped=1.0 2023-03-27 00:55:56,417 INFO [finetune.py:976] (3/7) Epoch 20, batch 5250, loss[loss=0.2156, simple_loss=0.2843, pruned_loss=0.07342, over 4886.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2481, pruned_loss=0.05361, over 954622.25 frames. ], batch size: 35, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 00:56:34,553 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6229, 1.6770, 1.3575, 1.6016, 1.9733, 1.7878, 1.6816, 1.4392], device='cuda:3'), covar=tensor([0.0335, 0.0327, 0.0586, 0.0321, 0.0208, 0.0598, 0.0295, 0.0411], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0108, 0.0144, 0.0111, 0.0100, 0.0111, 0.0100, 0.0113], device='cuda:3'), out_proj_covar=tensor([7.5510e-05, 8.2786e-05, 1.1338e-04, 8.5452e-05, 7.7896e-05, 8.2561e-05, 7.4434e-05, 8.6087e-05], device='cuda:3') 2023-03-27 00:56:36,234 INFO [finetune.py:976] (3/7) Epoch 20, batch 5300, loss[loss=0.2039, simple_loss=0.2701, pruned_loss=0.06884, over 4821.00 frames. ], tot_loss[loss=0.1788, simple_loss=0.2496, pruned_loss=0.05401, over 953932.99 frames. ], batch size: 39, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 00:56:36,348 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114127.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:56:46,561 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-27 00:57:13,340 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.589e+01 1.501e+02 1.834e+02 2.338e+02 4.244e+02, threshold=3.669e+02, percent-clipped=1.0 2023-03-27 00:57:30,109 INFO [finetune.py:976] (3/7) Epoch 20, batch 5350, loss[loss=0.1567, simple_loss=0.2299, pruned_loss=0.04177, over 4901.00 frames. ], tot_loss[loss=0.1791, simple_loss=0.2504, pruned_loss=0.05386, over 953077.86 frames. ], batch size: 36, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 00:57:37,328 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5420, 1.4557, 1.6681, 1.8003, 1.5106, 3.1795, 1.3948, 1.5486], device='cuda:3'), covar=tensor([0.0929, 0.1863, 0.1094, 0.0865, 0.1576, 0.0222, 0.1494, 0.1793], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0075, 0.0077, 0.0092, 0.0081, 0.0086, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 00:57:37,354 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114188.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:57:48,096 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6902, 1.5652, 2.0064, 1.9246, 1.7297, 3.6674, 1.5047, 1.6899], device='cuda:3'), covar=tensor([0.0892, 0.1659, 0.0975, 0.0858, 0.1455, 0.0201, 0.1466, 0.1666], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0074, 0.0076, 0.0091, 0.0080, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 00:57:48,131 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4072, 2.2529, 1.8561, 2.4695, 2.3311, 2.0493, 2.6764, 2.3858], device='cuda:3'), covar=tensor([0.1380, 0.2032, 0.3051, 0.2318, 0.2510, 0.1722, 0.2948, 0.1761], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0188, 0.0235, 0.0255, 0.0248, 0.0205, 0.0216, 0.0203], 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') 2023-03-27 00:58:03,283 INFO [finetune.py:976] (3/7) Epoch 20, batch 5400, loss[loss=0.1764, simple_loss=0.2503, pruned_loss=0.05129, over 4888.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2485, pruned_loss=0.05354, over 954597.73 frames. ], batch size: 35, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 00:58:18,259 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5809, 1.5183, 1.3346, 1.5035, 1.9250, 1.8013, 1.5338, 1.3632], device='cuda:3'), covar=tensor([0.0336, 0.0328, 0.0557, 0.0320, 0.0193, 0.0521, 0.0341, 0.0439], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0108, 0.0145, 0.0112, 0.0101, 0.0112, 0.0101, 0.0113], device='cuda:3'), out_proj_covar=tensor([7.5936e-05, 8.3250e-05, 1.1383e-04, 8.5974e-05, 7.8271e-05, 8.2763e-05, 7.4917e-05, 8.6561e-05], device='cuda:3') 2023-03-27 00:58:23,337 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.275e+01 1.521e+02 1.786e+02 2.103e+02 5.074e+02, threshold=3.573e+02, percent-clipped=1.0 2023-03-27 00:58:30,952 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 2023-03-27 00:58:33,643 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114273.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:58:35,980 INFO [finetune.py:976] (3/7) Epoch 20, batch 5450, loss[loss=0.1647, simple_loss=0.2342, pruned_loss=0.04765, over 4837.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2465, pruned_loss=0.05299, over 955896.29 frames. ], batch size: 33, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 00:58:51,606 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114302.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:58:51,656 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2893, 2.2355, 2.3955, 1.1168, 2.7622, 2.8849, 2.5214, 2.1201], device='cuda:3'), covar=tensor([0.1090, 0.0973, 0.0486, 0.0753, 0.0487, 0.0765, 0.0463, 0.0805], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0149, 0.0124, 0.0122, 0.0129, 0.0128, 0.0141, 0.0147], device='cuda:3'), out_proj_covar=tensor([8.9403e-05, 1.0756e-04, 8.8727e-05, 8.6433e-05, 9.0699e-05, 9.1326e-05, 1.0095e-04, 1.0575e-04], device='cuda:3') 2023-03-27 00:58:52,213 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1789, 1.8309, 2.4218, 4.1829, 2.8292, 2.8099, 0.8215, 3.4985], device='cuda:3'), covar=tensor([0.1652, 0.1494, 0.1534, 0.0499, 0.0754, 0.1550, 0.2195, 0.0357], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0116, 0.0134, 0.0165, 0.0101, 0.0136, 0.0125, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 00:58:52,273 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3440, 2.3261, 1.8917, 2.4112, 2.1896, 2.1476, 2.1185, 3.1518], device='cuda:3'), covar=tensor([0.3925, 0.4906, 0.3632, 0.4363, 0.4147, 0.2511, 0.4246, 0.1571], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0262, 0.0232, 0.0277, 0.0253, 0.0223, 0.0252, 0.0233], 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') 2023-03-27 00:59:05,064 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=114321.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:59:08,649 INFO [finetune.py:976] (3/7) Epoch 20, batch 5500, loss[loss=0.2145, simple_loss=0.2754, pruned_loss=0.07678, over 4923.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2434, pruned_loss=0.05185, over 957274.84 frames. ], batch size: 38, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 00:59:23,096 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=114350.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 00:59:27,726 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.653e+01 1.524e+02 1.755e+02 2.254e+02 4.886e+02, threshold=3.510e+02, percent-clipped=3.0 2023-03-27 00:59:36,021 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7099, 1.7155, 1.4161, 1.5756, 2.0463, 1.9505, 1.6456, 1.5270], device='cuda:3'), covar=tensor([0.0290, 0.0313, 0.0536, 0.0316, 0.0175, 0.0409, 0.0301, 0.0355], device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0107, 0.0144, 0.0111, 0.0100, 0.0111, 0.0100, 0.0113], device='cuda:3'), out_proj_covar=tensor([7.5502e-05, 8.2570e-05, 1.1316e-04, 8.5413e-05, 7.7769e-05, 8.2152e-05, 7.4340e-05, 8.6040e-05], device='cuda:3') 2023-03-27 00:59:42,371 INFO [finetune.py:976] (3/7) Epoch 20, batch 5550, loss[loss=0.1234, simple_loss=0.2024, pruned_loss=0.02215, over 4746.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2448, pruned_loss=0.05203, over 955546.30 frames. ], batch size: 28, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 01:00:10,659 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8115, 1.8439, 1.5423, 1.9985, 2.4498, 2.0155, 1.7512, 1.5001], device='cuda:3'), covar=tensor([0.2060, 0.2002, 0.1840, 0.1534, 0.1549, 0.1138, 0.2228, 0.1864], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0211, 0.0213, 0.0195, 0.0243, 0.0187, 0.0217, 0.0202], 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') 2023-03-27 01:00:14,062 INFO [finetune.py:976] (3/7) Epoch 20, batch 5600, loss[loss=0.1482, simple_loss=0.2259, pruned_loss=0.03524, over 4754.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2478, pruned_loss=0.05219, over 955912.57 frames. ], batch size: 26, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 01:00:31,890 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.005e+02 1.587e+02 1.911e+02 2.256e+02 4.682e+02, threshold=3.822e+02, percent-clipped=4.0 2023-03-27 01:00:34,375 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-27 01:00:43,493 INFO [finetune.py:976] (3/7) Epoch 20, batch 5650, loss[loss=0.2162, simple_loss=0.2834, pruned_loss=0.07454, over 4838.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2488, pruned_loss=0.0522, over 955307.42 frames. ], batch size: 47, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 01:00:46,991 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114483.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:01:02,761 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3382, 2.2490, 2.0151, 2.3753, 2.0185, 4.9000, 2.0674, 2.8295], device='cuda:3'), covar=tensor([0.3005, 0.2187, 0.1891, 0.1997, 0.1376, 0.0174, 0.2142, 0.0977], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0120, 0.0123, 0.0114, 0.0096, 0.0095, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 01:01:13,265 INFO [finetune.py:976] (3/7) Epoch 20, batch 5700, loss[loss=0.1511, simple_loss=0.2047, pruned_loss=0.04869, over 4108.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2453, pruned_loss=0.05204, over 942721.42 frames. ], batch size: 18, lr: 3.22e-03, grad_scale: 32.0 2023-03-27 01:01:39,096 INFO [finetune.py:976] (3/7) Epoch 21, batch 0, loss[loss=0.1808, simple_loss=0.2452, pruned_loss=0.05817, over 4746.00 frames. ], tot_loss[loss=0.1808, simple_loss=0.2452, pruned_loss=0.05817, over 4746.00 frames. ], batch size: 27, lr: 3.21e-03, grad_scale: 32.0 2023-03-27 01:01:39,096 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-27 01:01:48,857 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.6677, 3.2819, 3.4109, 3.5744, 3.4349, 3.3141, 3.7338, 1.5442], device='cuda:3'), covar=tensor([0.0833, 0.0714, 0.0759, 0.0828, 0.1249, 0.1394, 0.0744, 0.4529], device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0241, 0.0277, 0.0289, 0.0330, 0.0283, 0.0301, 0.0296], 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') 2023-03-27 01:01:52,343 INFO [finetune.py:1010] (3/7) Epoch 21, validation: loss=0.1598, simple_loss=0.2277, pruned_loss=0.0459, over 2265189.00 frames. 2023-03-27 01:01:52,343 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6434MB 2023-03-27 01:01:56,941 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.573e+01 1.356e+02 1.658e+02 2.014e+02 3.472e+02, threshold=3.316e+02, percent-clipped=0.0 2023-03-27 01:02:29,073 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8873, 1.3727, 1.9354, 1.8990, 1.7026, 1.6552, 1.8784, 1.7986], device='cuda:3'), covar=tensor([0.3607, 0.3739, 0.2988, 0.3326, 0.4205, 0.3631, 0.3871, 0.2737], device='cuda:3'), in_proj_covar=tensor([0.0254, 0.0243, 0.0263, 0.0281, 0.0280, 0.0255, 0.0290, 0.0245], 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') 2023-03-27 01:02:47,786 INFO [finetune.py:976] (3/7) Epoch 21, batch 50, loss[loss=0.1481, simple_loss=0.2328, pruned_loss=0.03172, over 4767.00 frames. ], tot_loss[loss=0.182, simple_loss=0.2522, pruned_loss=0.0559, over 215579.17 frames. ], batch size: 28, lr: 3.21e-03, grad_scale: 32.0 2023-03-27 01:03:04,160 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4374, 2.3026, 2.0253, 2.2469, 2.1979, 2.2225, 2.2032, 2.8987], device='cuda:3'), covar=tensor([0.3294, 0.4183, 0.3067, 0.3428, 0.3438, 0.2224, 0.3459, 0.1519], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0262, 0.0231, 0.0276, 0.0252, 0.0222, 0.0252, 0.0234], 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') 2023-03-27 01:03:12,803 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0 2023-03-27 01:03:21,579 INFO [finetune.py:976] (3/7) Epoch 21, batch 100, loss[loss=0.2004, simple_loss=0.2711, pruned_loss=0.06481, over 4835.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2426, pruned_loss=0.05162, over 378827.42 frames. ], batch size: 47, lr: 3.21e-03, grad_scale: 32.0 2023-03-27 01:03:23,372 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.865e+01 1.560e+02 1.971e+02 2.354e+02 5.080e+02, threshold=3.943e+02, percent-clipped=2.0 2023-03-27 01:03:23,568 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-27 01:03:54,252 INFO [finetune.py:976] (3/7) Epoch 21, batch 150, loss[loss=0.1051, simple_loss=0.178, pruned_loss=0.01608, over 4753.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2384, pruned_loss=0.05018, over 506444.18 frames. ], batch size: 27, lr: 3.21e-03, grad_scale: 32.0 2023-03-27 01:04:02,528 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114716.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:04:03,534 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 2023-03-27 01:04:26,908 INFO [finetune.py:976] (3/7) Epoch 21, batch 200, loss[loss=0.1653, simple_loss=0.2383, pruned_loss=0.04613, over 4826.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2372, pruned_loss=0.05017, over 604364.51 frames. ], batch size: 38, lr: 3.21e-03, grad_scale: 32.0 2023-03-27 01:04:29,188 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.002e+02 1.562e+02 1.886e+02 2.300e+02 5.249e+02, threshold=3.772e+02, percent-clipped=1.0 2023-03-27 01:04:42,932 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114777.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:04:44,173 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0928, 2.0071, 1.6116, 1.9378, 2.0611, 1.7667, 2.3464, 2.1161], device='cuda:3'), covar=tensor([0.1353, 0.1937, 0.3015, 0.2657, 0.2564, 0.1753, 0.2714, 0.1737], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0188, 0.0236, 0.0254, 0.0248, 0.0204, 0.0215, 0.0202], 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') 2023-03-27 01:04:46,595 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114783.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:05:00,781 INFO [finetune.py:976] (3/7) Epoch 21, batch 250, loss[loss=0.1434, simple_loss=0.2175, pruned_loss=0.0347, over 4771.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2409, pruned_loss=0.05158, over 682719.31 frames. ], batch size: 26, lr: 3.21e-03, grad_scale: 32.0 2023-03-27 01:05:19,224 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=114831.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:05:33,143 INFO [finetune.py:976] (3/7) Epoch 21, batch 300, loss[loss=0.2349, simple_loss=0.3033, pruned_loss=0.08327, over 4162.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2455, pruned_loss=0.05251, over 744089.60 frames. ], batch size: 65, lr: 3.21e-03, grad_scale: 32.0 2023-03-27 01:05:36,362 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.093e+02 1.500e+02 1.787e+02 2.137e+02 3.935e+02, threshold=3.575e+02, percent-clipped=3.0 2023-03-27 01:05:45,144 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3892, 1.2911, 1.2754, 1.4011, 0.7792, 2.9729, 1.0842, 1.4875], device='cuda:3'), covar=tensor([0.3534, 0.2765, 0.2408, 0.2567, 0.2165, 0.0231, 0.2904, 0.1393], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0116, 0.0120, 0.0123, 0.0113, 0.0096, 0.0095, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 01:05:46,341 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114871.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:05:51,808 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5592, 1.5800, 1.3221, 1.4532, 1.8837, 1.7490, 1.5198, 1.3872], device='cuda:3'), covar=tensor([0.0303, 0.0322, 0.0600, 0.0333, 0.0213, 0.0460, 0.0366, 0.0367], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0108, 0.0145, 0.0112, 0.0100, 0.0111, 0.0100, 0.0113], device='cuda:3'), out_proj_covar=tensor([7.5750e-05, 8.2828e-05, 1.1374e-04, 8.5712e-05, 7.8034e-05, 8.2437e-05, 7.4557e-05, 8.5994e-05], device='cuda:3') 2023-03-27 01:06:06,529 INFO [finetune.py:976] (3/7) Epoch 21, batch 350, loss[loss=0.1881, simple_loss=0.2612, pruned_loss=0.05753, over 4786.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2482, pruned_loss=0.05342, over 791295.56 frames. ], batch size: 45, lr: 3.21e-03, grad_scale: 32.0 2023-03-27 01:06:26,482 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114932.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:06:40,072 INFO [finetune.py:976] (3/7) Epoch 21, batch 400, loss[loss=0.2017, simple_loss=0.2669, pruned_loss=0.06823, over 4821.00 frames. ], tot_loss[loss=0.1796, simple_loss=0.2509, pruned_loss=0.05415, over 827012.11 frames. ], batch size: 30, lr: 3.21e-03, grad_scale: 32.0 2023-03-27 01:06:41,873 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.016e+02 1.627e+02 1.950e+02 2.383e+02 4.205e+02, threshold=3.900e+02, percent-clipped=3.0 2023-03-27 01:07:20,664 INFO [finetune.py:976] (3/7) Epoch 21, batch 450, loss[loss=0.1539, simple_loss=0.2211, pruned_loss=0.04338, over 4819.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2484, pruned_loss=0.05287, over 854400.20 frames. ], batch size: 25, lr: 3.21e-03, grad_scale: 32.0 2023-03-27 01:08:02,752 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.2053, 3.5968, 3.8019, 3.9846, 3.9889, 3.7564, 4.3107, 1.4150], device='cuda:3'), covar=tensor([0.0782, 0.0871, 0.0881, 0.0939, 0.1214, 0.1450, 0.0690, 0.5615], device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0242, 0.0277, 0.0291, 0.0330, 0.0282, 0.0302, 0.0297], 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') 2023-03-27 01:08:11,192 INFO [finetune.py:976] (3/7) Epoch 21, batch 500, loss[loss=0.152, simple_loss=0.2245, pruned_loss=0.03978, over 4894.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2464, pruned_loss=0.0527, over 875405.46 frames. ], batch size: 32, lr: 3.21e-03, grad_scale: 32.0 2023-03-27 01:08:13,013 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.154e+02 1.445e+02 1.728e+02 2.124e+02 2.919e+02, threshold=3.456e+02, percent-clipped=0.0 2023-03-27 01:08:24,215 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115072.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:08:33,769 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115086.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:08:45,012 INFO [finetune.py:976] (3/7) Epoch 21, batch 550, loss[loss=0.1572, simple_loss=0.2219, pruned_loss=0.04621, over 4891.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2443, pruned_loss=0.05216, over 895175.49 frames. ], batch size: 35, lr: 3.21e-03, grad_scale: 32.0 2023-03-27 01:09:10,585 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8744, 1.7486, 1.5178, 1.3618, 1.8990, 1.6337, 1.8424, 1.8774], device='cuda:3'), covar=tensor([0.1430, 0.1912, 0.3079, 0.2617, 0.2656, 0.1755, 0.2794, 0.1780], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0188, 0.0236, 0.0253, 0.0247, 0.0204, 0.0215, 0.0202], 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') 2023-03-27 01:09:14,155 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115147.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:09:18,137 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-27 01:09:18,263 INFO [finetune.py:976] (3/7) Epoch 21, batch 600, loss[loss=0.1405, simple_loss=0.2057, pruned_loss=0.03767, over 4766.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2451, pruned_loss=0.05285, over 908438.59 frames. ], batch size: 27, lr: 3.21e-03, grad_scale: 32.0 2023-03-27 01:09:20,110 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.082e+02 1.558e+02 1.835e+02 2.263e+02 4.639e+02, threshold=3.670e+02, percent-clipped=5.0 2023-03-27 01:09:41,576 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2102, 2.1944, 2.3036, 1.5740, 2.2707, 2.3540, 2.4446, 1.8326], device='cuda:3'), covar=tensor([0.0642, 0.0683, 0.0682, 0.0921, 0.0774, 0.0754, 0.0579, 0.1225], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0137, 0.0141, 0.0121, 0.0125, 0.0140, 0.0141, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 01:09:51,867 INFO [finetune.py:976] (3/7) Epoch 21, batch 650, loss[loss=0.128, simple_loss=0.1998, pruned_loss=0.02808, over 4769.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2481, pruned_loss=0.05373, over 919960.65 frames. ], batch size: 26, lr: 3.21e-03, grad_scale: 32.0 2023-03-27 01:10:01,555 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1703, 2.1887, 2.2316, 1.6269, 2.1310, 2.2723, 2.3750, 1.7887], device='cuda:3'), covar=tensor([0.0704, 0.0649, 0.0725, 0.0946, 0.0692, 0.0784, 0.0624, 0.1180], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0137, 0.0141, 0.0121, 0.0125, 0.0140, 0.0141, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 01:10:07,426 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115227.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:10:25,138 INFO [finetune.py:976] (3/7) Epoch 21, batch 700, loss[loss=0.1603, simple_loss=0.2391, pruned_loss=0.04079, over 4815.00 frames. ], tot_loss[loss=0.1783, simple_loss=0.2498, pruned_loss=0.05341, over 929360.74 frames. ], batch size: 51, lr: 3.21e-03, grad_scale: 32.0 2023-03-27 01:10:26,911 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.090e+02 1.664e+02 1.957e+02 2.283e+02 3.730e+02, threshold=3.914e+02, percent-clipped=1.0 2023-03-27 01:10:58,917 INFO [finetune.py:976] (3/7) Epoch 21, batch 750, loss[loss=0.1496, simple_loss=0.2387, pruned_loss=0.03029, over 4751.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2503, pruned_loss=0.05296, over 936147.79 frames. ], batch size: 27, lr: 3.21e-03, grad_scale: 32.0 2023-03-27 01:11:31,742 INFO [finetune.py:976] (3/7) Epoch 21, batch 800, loss[loss=0.2276, simple_loss=0.292, pruned_loss=0.08163, over 4818.00 frames. ], tot_loss[loss=0.1783, simple_loss=0.2504, pruned_loss=0.05313, over 941039.42 frames. ], batch size: 38, lr: 3.21e-03, grad_scale: 32.0 2023-03-27 01:11:33,561 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.112e+02 1.426e+02 1.720e+02 2.044e+02 3.360e+02, threshold=3.441e+02, percent-clipped=0.0 2023-03-27 01:11:34,896 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5830, 1.5566, 1.3215, 1.5228, 1.9353, 1.8277, 1.5805, 1.3773], device='cuda:3'), covar=tensor([0.0347, 0.0330, 0.0613, 0.0299, 0.0208, 0.0372, 0.0292, 0.0408], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0109, 0.0146, 0.0113, 0.0101, 0.0112, 0.0101, 0.0113], device='cuda:3'), out_proj_covar=tensor([7.6548e-05, 8.3424e-05, 1.1491e-04, 8.6424e-05, 7.8313e-05, 8.2820e-05, 7.5108e-05, 8.6654e-05], device='cuda:3') 2023-03-27 01:11:39,715 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.6733, 1.7123, 1.6051, 0.8511, 1.8393, 2.0472, 1.9290, 1.4855], device='cuda:3'), covar=tensor([0.0875, 0.0565, 0.0512, 0.0623, 0.0378, 0.0530, 0.0277, 0.0701], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0148, 0.0125, 0.0122, 0.0129, 0.0127, 0.0141, 0.0147], device='cuda:3'), out_proj_covar=tensor([8.8930e-05, 1.0708e-04, 8.9053e-05, 8.6509e-05, 9.0402e-05, 9.0892e-05, 1.0082e-04, 1.0522e-04], device='cuda:3') 2023-03-27 01:11:41,518 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115370.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:11:42,117 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115371.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:11:42,698 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115372.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:11:49,181 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6555, 2.3889, 1.9409, 1.0242, 2.0985, 2.0923, 1.9253, 2.1686], device='cuda:3'), covar=tensor([0.0719, 0.0921, 0.1725, 0.2118, 0.1543, 0.2103, 0.2180, 0.0988], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0191, 0.0198, 0.0181, 0.0209, 0.0208, 0.0222, 0.0195], 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') 2023-03-27 01:12:04,583 INFO [finetune.py:976] (3/7) Epoch 21, batch 850, loss[loss=0.163, simple_loss=0.2332, pruned_loss=0.04636, over 4823.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2486, pruned_loss=0.0527, over 944262.69 frames. ], batch size: 40, lr: 3.21e-03, grad_scale: 32.0 2023-03-27 01:12:14,852 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=115420.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:12:24,424 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115431.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:12:25,045 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115432.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:12:41,966 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115442.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:12:54,666 INFO [finetune.py:976] (3/7) Epoch 21, batch 900, loss[loss=0.1117, simple_loss=0.1867, pruned_loss=0.01835, over 4934.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2455, pruned_loss=0.05167, over 945179.50 frames. ], batch size: 33, lr: 3.20e-03, grad_scale: 32.0 2023-03-27 01:13:00,793 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.046e+02 1.494e+02 1.776e+02 2.140e+02 4.219e+02, threshold=3.551e+02, percent-clipped=3.0 2023-03-27 01:13:01,549 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4127, 2.1788, 1.8180, 0.8748, 1.9318, 1.9216, 1.8109, 2.0326], device='cuda:3'), covar=tensor([0.0854, 0.0777, 0.1474, 0.2012, 0.1337, 0.2185, 0.2122, 0.0870], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0190, 0.0197, 0.0181, 0.0208, 0.0208, 0.0221, 0.0194], 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') 2023-03-27 01:13:33,844 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4312, 1.4498, 1.9090, 1.8128, 1.7728, 3.3260, 1.4789, 1.5784], device='cuda:3'), covar=tensor([0.1107, 0.2100, 0.1262, 0.1078, 0.1675, 0.0365, 0.1777, 0.2064], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0074, 0.0077, 0.0092, 0.0081, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 01:13:36,233 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1232, 2.1012, 2.1156, 1.6642, 2.1382, 2.3416, 2.2151, 1.6706], device='cuda:3'), covar=tensor([0.0611, 0.0643, 0.0760, 0.0855, 0.0710, 0.0626, 0.0582, 0.1279], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0138, 0.0142, 0.0122, 0.0127, 0.0141, 0.0143, 0.0164], 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') 2023-03-27 01:13:37,347 INFO [finetune.py:976] (3/7) Epoch 21, batch 950, loss[loss=0.2123, simple_loss=0.2764, pruned_loss=0.07406, over 4808.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2445, pruned_loss=0.05188, over 948940.11 frames. ], batch size: 39, lr: 3.20e-03, grad_scale: 32.0 2023-03-27 01:13:51,938 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115527.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:14:11,292 INFO [finetune.py:976] (3/7) Epoch 21, batch 1000, loss[loss=0.2128, simple_loss=0.2888, pruned_loss=0.06844, over 4712.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2467, pruned_loss=0.0526, over 950059.41 frames. ], batch size: 59, lr: 3.20e-03, grad_scale: 32.0 2023-03-27 01:14:13,113 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.049e+02 1.550e+02 1.849e+02 2.159e+02 3.452e+02, threshold=3.698e+02, percent-clipped=0.0 2023-03-27 01:14:24,464 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=115575.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:14:40,338 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.43 vs. limit=5.0 2023-03-27 01:14:44,024 INFO [finetune.py:976] (3/7) Epoch 21, batch 1050, loss[loss=0.1719, simple_loss=0.2419, pruned_loss=0.05092, over 4761.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.2496, pruned_loss=0.05358, over 950643.02 frames. ], batch size: 26, lr: 3.20e-03, grad_scale: 32.0 2023-03-27 01:15:16,672 INFO [finetune.py:976] (3/7) Epoch 21, batch 1100, loss[loss=0.1698, simple_loss=0.2385, pruned_loss=0.05053, over 4784.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.2496, pruned_loss=0.05364, over 950416.20 frames. ], batch size: 29, lr: 3.20e-03, grad_scale: 32.0 2023-03-27 01:15:19,448 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.043e+02 1.582e+02 1.822e+02 2.328e+02 4.675e+02, threshold=3.643e+02, percent-clipped=4.0 2023-03-27 01:15:50,437 INFO [finetune.py:976] (3/7) Epoch 21, batch 1150, loss[loss=0.1803, simple_loss=0.2569, pruned_loss=0.05181, over 4839.00 frames. ], tot_loss[loss=0.1787, simple_loss=0.2505, pruned_loss=0.05343, over 950019.38 frames. ], batch size: 44, lr: 3.20e-03, grad_scale: 32.0 2023-03-27 01:16:05,312 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115726.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:16:05,921 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115727.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:16:14,927 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115742.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:16:21,876 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.17 vs. limit=5.0 2023-03-27 01:16:24,029 INFO [finetune.py:976] (3/7) Epoch 21, batch 1200, loss[loss=0.1447, simple_loss=0.2219, pruned_loss=0.03374, over 4712.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2495, pruned_loss=0.05332, over 951738.01 frames. ], batch size: 23, lr: 3.20e-03, grad_scale: 32.0 2023-03-27 01:16:25,823 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.936e+01 1.465e+02 1.737e+02 2.048e+02 4.574e+02, threshold=3.475e+02, percent-clipped=2.0 2023-03-27 01:16:26,236 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.64 vs. limit=5.0 2023-03-27 01:16:47,357 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=115790.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:16:54,936 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-27 01:16:56,822 INFO [finetune.py:976] (3/7) Epoch 21, batch 1250, loss[loss=0.1452, simple_loss=0.2122, pruned_loss=0.03912, over 4826.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.247, pruned_loss=0.05281, over 949865.11 frames. ], batch size: 41, lr: 3.20e-03, grad_scale: 32.0 2023-03-27 01:17:22,454 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7269, 1.2011, 0.7404, 1.5276, 2.1341, 1.0639, 1.5652, 1.5419], device='cuda:3'), covar=tensor([0.1446, 0.2136, 0.2058, 0.1242, 0.1903, 0.1978, 0.1399, 0.1928], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0095, 0.0111, 0.0092, 0.0119, 0.0094, 0.0097, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-27 01:17:23,099 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.8515, 1.5709, 1.4323, 0.7979, 1.6338, 1.8103, 1.7097, 1.4770], device='cuda:3'), covar=tensor([0.0800, 0.0548, 0.0545, 0.0586, 0.0559, 0.0480, 0.0359, 0.0561], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0149, 0.0125, 0.0123, 0.0130, 0.0128, 0.0141, 0.0148], device='cuda:3'), out_proj_covar=tensor([8.9710e-05, 1.0761e-04, 8.9502e-05, 8.7100e-05, 9.1219e-05, 9.1650e-05, 1.0141e-04, 1.0597e-04], device='cuda:3') 2023-03-27 01:17:29,532 INFO [finetune.py:976] (3/7) Epoch 21, batch 1300, loss[loss=0.1753, simple_loss=0.2459, pruned_loss=0.05232, over 4805.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2448, pruned_loss=0.0522, over 950527.52 frames. ], batch size: 51, lr: 3.20e-03, grad_scale: 32.0 2023-03-27 01:17:32,370 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.105e+02 1.563e+02 1.759e+02 2.181e+02 4.124e+02, threshold=3.519e+02, percent-clipped=1.0 2023-03-27 01:17:52,856 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0823, 1.9511, 1.7002, 1.7382, 1.8809, 1.8398, 1.8901, 2.5395], device='cuda:3'), covar=tensor([0.3664, 0.4071, 0.3122, 0.3327, 0.3583, 0.2369, 0.3560, 0.1638], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0262, 0.0233, 0.0277, 0.0253, 0.0223, 0.0253, 0.0234], 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') 2023-03-27 01:18:12,452 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115892.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 01:18:23,975 INFO [finetune.py:976] (3/7) Epoch 21, batch 1350, loss[loss=0.1766, simple_loss=0.2572, pruned_loss=0.04801, over 4765.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2448, pruned_loss=0.05253, over 951019.23 frames. ], batch size: 28, lr: 3.20e-03, grad_scale: 64.0 2023-03-27 01:18:33,800 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-03-27 01:19:00,573 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115953.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 01:19:01,043 INFO [finetune.py:976] (3/7) Epoch 21, batch 1400, loss[loss=0.1696, simple_loss=0.243, pruned_loss=0.04805, over 4896.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2479, pruned_loss=0.05323, over 952966.82 frames. ], batch size: 35, lr: 3.20e-03, grad_scale: 64.0 2023-03-27 01:19:02,862 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.014e+02 1.541e+02 1.818e+02 2.071e+02 3.575e+02, threshold=3.635e+02, percent-clipped=1.0 2023-03-27 01:19:11,475 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-27 01:19:17,428 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1276, 1.9780, 2.6485, 4.3727, 3.0145, 2.7734, 0.8737, 3.6765], device='cuda:3'), covar=tensor([0.1728, 0.1411, 0.1435, 0.0544, 0.0727, 0.1594, 0.2224, 0.0383], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0117, 0.0134, 0.0165, 0.0100, 0.0137, 0.0125, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 01:19:35,506 INFO [finetune.py:976] (3/7) Epoch 21, batch 1450, loss[loss=0.1749, simple_loss=0.2396, pruned_loss=0.05511, over 4931.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2491, pruned_loss=0.05359, over 952490.91 frames. ], batch size: 33, lr: 3.20e-03, grad_scale: 64.0 2023-03-27 01:19:43,011 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116014.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:19:49,432 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6454, 1.5376, 2.1820, 3.3674, 2.3197, 2.4139, 1.3356, 2.7931], device='cuda:3'), covar=tensor([0.1673, 0.1418, 0.1302, 0.0519, 0.0721, 0.1789, 0.1546, 0.0440], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0116, 0.0134, 0.0165, 0.0100, 0.0137, 0.0125, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 01:19:51,744 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116026.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:19:52,338 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116027.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:20:06,197 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-03-27 01:20:09,068 INFO [finetune.py:976] (3/7) Epoch 21, batch 1500, loss[loss=0.1736, simple_loss=0.2467, pruned_loss=0.05024, over 4890.00 frames. ], tot_loss[loss=0.1805, simple_loss=0.2515, pruned_loss=0.05474, over 952333.41 frames. ], batch size: 43, lr: 3.20e-03, grad_scale: 64.0 2023-03-27 01:20:10,888 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.231e+02 1.719e+02 2.058e+02 2.312e+02 4.180e+02, threshold=4.116e+02, percent-clipped=2.0 2023-03-27 01:20:23,168 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=116074.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:20:23,791 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=116075.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:20:23,856 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116075.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 01:20:31,472 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.8442, 3.3779, 3.0824, 1.8149, 3.3714, 2.8261, 2.7185, 3.2280], device='cuda:3'), covar=tensor([0.0548, 0.0774, 0.1423, 0.1813, 0.1070, 0.1534, 0.1585, 0.0720], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0191, 0.0197, 0.0181, 0.0209, 0.0208, 0.0222, 0.0194], 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') 2023-03-27 01:20:36,704 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3556, 2.3069, 1.8599, 2.5218, 2.3633, 2.0659, 2.8825, 2.4968], device='cuda:3'), covar=tensor([0.1307, 0.2397, 0.2733, 0.2498, 0.2280, 0.1534, 0.2873, 0.1649], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0188, 0.0235, 0.0253, 0.0247, 0.0203, 0.0214, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 01:20:42,654 INFO [finetune.py:976] (3/7) Epoch 21, batch 1550, loss[loss=0.1503, simple_loss=0.2268, pruned_loss=0.03686, over 4906.00 frames. ], tot_loss[loss=0.1797, simple_loss=0.251, pruned_loss=0.05419, over 954790.27 frames. ], batch size: 37, lr: 3.20e-03, grad_scale: 64.0 2023-03-27 01:20:43,385 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116105.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:20:50,868 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116116.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:21:15,952 INFO [finetune.py:976] (3/7) Epoch 21, batch 1600, loss[loss=0.1611, simple_loss=0.2306, pruned_loss=0.04575, over 4145.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2488, pruned_loss=0.05366, over 953629.59 frames. ], batch size: 18, lr: 3.20e-03, grad_scale: 64.0 2023-03-27 01:21:17,776 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.061e+01 1.555e+02 1.841e+02 2.223e+02 4.654e+02, threshold=3.683e+02, percent-clipped=1.0 2023-03-27 01:21:23,358 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116166.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:21:29,792 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-27 01:21:31,955 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116177.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:21:49,852 INFO [finetune.py:976] (3/7) Epoch 21, batch 1650, loss[loss=0.1762, simple_loss=0.2394, pruned_loss=0.05644, over 4750.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2461, pruned_loss=0.05264, over 954197.92 frames. ], batch size: 27, lr: 3.20e-03, grad_scale: 64.0 2023-03-27 01:22:19,469 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116248.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 01:22:23,481 INFO [finetune.py:976] (3/7) Epoch 21, batch 1700, loss[loss=0.2142, simple_loss=0.2871, pruned_loss=0.07071, over 4738.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2447, pruned_loss=0.05251, over 956074.63 frames. ], batch size: 54, lr: 3.20e-03, grad_scale: 64.0 2023-03-27 01:22:25,327 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.524e+01 1.502e+02 1.774e+02 2.116e+02 3.203e+02, threshold=3.548e+02, percent-clipped=0.0 2023-03-27 01:22:29,774 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-27 01:22:59,246 INFO [finetune.py:976] (3/7) Epoch 21, batch 1750, loss[loss=0.1955, simple_loss=0.271, pruned_loss=0.06001, over 4738.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2454, pruned_loss=0.05233, over 957262.59 frames. ], batch size: 54, lr: 3.20e-03, grad_scale: 64.0 2023-03-27 01:23:15,877 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6812, 2.4994, 2.0608, 2.7654, 2.5645, 2.3576, 3.1418, 2.6647], device='cuda:3'), covar=tensor([0.1240, 0.2165, 0.3023, 0.2508, 0.2576, 0.1575, 0.2941, 0.1647], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0188, 0.0236, 0.0255, 0.0249, 0.0204, 0.0215, 0.0203], 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') 2023-03-27 01:23:19,653 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116323.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:23:58,747 INFO [finetune.py:976] (3/7) Epoch 21, batch 1800, loss[loss=0.2061, simple_loss=0.2715, pruned_loss=0.07032, over 4810.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2488, pruned_loss=0.05344, over 956826.62 frames. ], batch size: 38, lr: 3.20e-03, grad_scale: 64.0 2023-03-27 01:24:00,588 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.248e+02 1.627e+02 1.938e+02 2.363e+02 5.057e+02, threshold=3.876e+02, percent-clipped=3.0 2023-03-27 01:24:08,540 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116370.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 01:24:10,825 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-03-27 01:24:18,594 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116384.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:24:31,788 INFO [finetune.py:976] (3/7) Epoch 21, batch 1850, loss[loss=0.1681, simple_loss=0.2458, pruned_loss=0.04521, over 4823.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.249, pruned_loss=0.05322, over 956065.93 frames. ], batch size: 33, lr: 3.20e-03, grad_scale: 64.0 2023-03-27 01:24:46,137 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.18 vs. limit=5.0 2023-03-27 01:25:00,263 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116446.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:25:05,406 INFO [finetune.py:976] (3/7) Epoch 21, batch 1900, loss[loss=0.123, simple_loss=0.1976, pruned_loss=0.02418, over 4696.00 frames. ], tot_loss[loss=0.1795, simple_loss=0.251, pruned_loss=0.05397, over 957153.74 frames. ], batch size: 23, lr: 3.20e-03, grad_scale: 64.0 2023-03-27 01:25:07,228 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.037e+02 1.572e+02 1.880e+02 2.123e+02 3.861e+02, threshold=3.760e+02, percent-clipped=0.0 2023-03-27 01:25:10,209 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116461.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:25:16,980 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116472.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:25:38,738 INFO [finetune.py:976] (3/7) Epoch 21, batch 1950, loss[loss=0.1573, simple_loss=0.2331, pruned_loss=0.04072, over 4897.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.249, pruned_loss=0.05297, over 955689.26 frames. ], batch size: 36, lr: 3.20e-03, grad_scale: 64.0 2023-03-27 01:25:39,478 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116505.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:25:40,700 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116507.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 01:25:41,245 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6559, 3.7674, 3.6545, 1.5547, 3.8784, 2.9697, 1.1048, 2.6424], device='cuda:3'), covar=tensor([0.2220, 0.2049, 0.1358, 0.3415, 0.0994, 0.0868, 0.3990, 0.1417], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0178, 0.0159, 0.0129, 0.0161, 0.0123, 0.0148, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 01:26:07,961 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116548.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 01:26:11,459 INFO [finetune.py:976] (3/7) Epoch 21, batch 2000, loss[loss=0.1383, simple_loss=0.2037, pruned_loss=0.03649, over 4685.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2454, pruned_loss=0.05171, over 955144.73 frames. ], batch size: 23, lr: 3.20e-03, grad_scale: 64.0 2023-03-27 01:26:13,785 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 7.123e+01 1.433e+02 1.690e+02 2.067e+02 3.885e+02, threshold=3.380e+02, percent-clipped=2.0 2023-03-27 01:26:19,735 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116566.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:26:39,353 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=116596.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 01:26:44,684 INFO [finetune.py:976] (3/7) Epoch 21, batch 2050, loss[loss=0.1759, simple_loss=0.2366, pruned_loss=0.05762, over 4834.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2424, pruned_loss=0.0508, over 956877.93 frames. ], batch size: 33, lr: 3.20e-03, grad_scale: 64.0 2023-03-27 01:26:54,227 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116618.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:27:18,449 INFO [finetune.py:976] (3/7) Epoch 21, batch 2100, loss[loss=0.168, simple_loss=0.2405, pruned_loss=0.04778, over 4856.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2418, pruned_loss=0.05039, over 956538.11 frames. ], batch size: 44, lr: 3.20e-03, grad_scale: 32.0 2023-03-27 01:27:20,848 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.015e+02 1.617e+02 1.783e+02 2.198e+02 6.495e+02, threshold=3.567e+02, percent-clipped=4.0 2023-03-27 01:27:29,071 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116670.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 01:27:34,426 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116679.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:27:34,483 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116679.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:27:51,927 INFO [finetune.py:976] (3/7) Epoch 21, batch 2150, loss[loss=0.1455, simple_loss=0.216, pruned_loss=0.0375, over 4683.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2458, pruned_loss=0.05197, over 955691.04 frames. ], batch size: 23, lr: 3.20e-03, grad_scale: 32.0 2023-03-27 01:27:52,033 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116704.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:28:00,962 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=116718.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:28:03,960 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1653, 1.4609, 1.1543, 1.3428, 1.5620, 1.5446, 1.4071, 1.3054], device='cuda:3'), covar=tensor([0.0475, 0.0250, 0.0613, 0.0307, 0.0265, 0.0511, 0.0382, 0.0398], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0108, 0.0145, 0.0113, 0.0100, 0.0111, 0.0100, 0.0113], device='cuda:3'), out_proj_covar=tensor([7.6104e-05, 8.2543e-05, 1.1412e-04, 8.6567e-05, 7.8102e-05, 8.2313e-05, 7.4597e-05, 8.6572e-05], device='cuda:3') 2023-03-27 01:28:26,697 INFO [finetune.py:976] (3/7) Epoch 21, batch 2200, loss[loss=0.1224, simple_loss=0.2011, pruned_loss=0.02181, over 4711.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2483, pruned_loss=0.05301, over 951054.23 frames. ], batch size: 23, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:28:30,714 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.107e+02 1.636e+02 2.054e+02 2.505e+02 6.138e+02, threshold=4.108e+02, percent-clipped=5.0 2023-03-27 01:28:32,688 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116761.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:28:34,561 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-27 01:28:39,675 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116765.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:28:44,486 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116772.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:29:19,243 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116802.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 01:29:19,274 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116802.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:29:20,383 INFO [finetune.py:976] (3/7) Epoch 21, batch 2250, loss[loss=0.19, simple_loss=0.2636, pruned_loss=0.05822, over 4811.00 frames. ], tot_loss[loss=0.179, simple_loss=0.25, pruned_loss=0.05394, over 951225.23 frames. ], batch size: 39, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:29:28,969 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=116809.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:29:39,990 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=116820.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:29:52,920 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.88 vs. limit=5.0 2023-03-27 01:29:53,302 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4019, 2.3092, 1.9759, 2.4102, 2.1856, 2.2236, 2.2033, 3.1359], device='cuda:3'), covar=tensor([0.3830, 0.5005, 0.3571, 0.4257, 0.4556, 0.2642, 0.4494, 0.1652], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0263, 0.0232, 0.0277, 0.0253, 0.0223, 0.0253, 0.0234], 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') 2023-03-27 01:30:01,468 INFO [finetune.py:976] (3/7) Epoch 21, batch 2300, loss[loss=0.1671, simple_loss=0.2411, pruned_loss=0.04653, over 4834.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2493, pruned_loss=0.05316, over 951917.29 frames. ], batch size: 30, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:30:04,886 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.160e+02 1.477e+02 1.740e+02 2.172e+02 4.454e+02, threshold=3.479e+02, percent-clipped=1.0 2023-03-27 01:30:06,811 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116861.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:30:08,065 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116863.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:30:34,138 INFO [finetune.py:976] (3/7) Epoch 21, batch 2350, loss[loss=0.1473, simple_loss=0.2245, pruned_loss=0.0351, over 4762.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2473, pruned_loss=0.05246, over 951294.70 frames. ], batch size: 27, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:31:07,402 INFO [finetune.py:976] (3/7) Epoch 21, batch 2400, loss[loss=0.156, simple_loss=0.2276, pruned_loss=0.0422, over 4899.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2447, pruned_loss=0.0517, over 951628.59 frames. ], batch size: 35, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:31:08,655 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6766, 1.5678, 2.0767, 3.4231, 2.3359, 2.3966, 0.7960, 2.9438], device='cuda:3'), covar=tensor([0.1721, 0.1465, 0.1423, 0.0552, 0.0796, 0.1479, 0.2043, 0.0431], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0116, 0.0133, 0.0163, 0.0100, 0.0137, 0.0124, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 01:31:09,769 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.097e+02 1.566e+02 1.863e+02 2.219e+02 3.648e+02, threshold=3.726e+02, percent-clipped=1.0 2023-03-27 01:31:21,634 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116974.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:31:25,169 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116979.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:31:40,995 INFO [finetune.py:976] (3/7) Epoch 21, batch 2450, loss[loss=0.1938, simple_loss=0.2688, pruned_loss=0.0594, over 4815.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2417, pruned_loss=0.05131, over 951639.40 frames. ], batch size: 40, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:31:50,915 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7163, 3.7257, 3.4491, 1.4641, 3.8783, 2.7906, 0.9077, 2.5983], device='cuda:3'), covar=tensor([0.2086, 0.1810, 0.1482, 0.3659, 0.0858, 0.0999, 0.4236, 0.1510], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0178, 0.0158, 0.0130, 0.0161, 0.0123, 0.0149, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 01:31:57,423 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=117027.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:32:14,703 INFO [finetune.py:976] (3/7) Epoch 21, batch 2500, loss[loss=0.1912, simple_loss=0.2674, pruned_loss=0.05755, over 4795.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.243, pruned_loss=0.05204, over 950710.34 frames. ], batch size: 45, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:32:17,112 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.025e+02 1.510e+02 1.797e+02 2.340e+02 3.968e+02, threshold=3.593e+02, percent-clipped=1.0 2023-03-27 01:32:18,390 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117060.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:32:46,782 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117102.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:32:47,649 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-27 01:32:47,902 INFO [finetune.py:976] (3/7) Epoch 21, batch 2550, loss[loss=0.1784, simple_loss=0.2615, pruned_loss=0.04772, over 4803.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2467, pruned_loss=0.05262, over 951036.96 frames. ], batch size: 41, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:33:02,842 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8925, 1.7716, 1.8657, 1.1033, 1.8876, 1.9234, 1.7951, 1.5545], device='cuda:3'), covar=tensor([0.0562, 0.0685, 0.0601, 0.0893, 0.0715, 0.0630, 0.0614, 0.1098], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0134, 0.0138, 0.0119, 0.0124, 0.0138, 0.0140, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 01:33:19,396 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=117150.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:33:21,775 INFO [finetune.py:976] (3/7) Epoch 21, batch 2600, loss[loss=0.1778, simple_loss=0.2602, pruned_loss=0.04764, over 4818.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2485, pruned_loss=0.05298, over 950236.75 frames. ], batch size: 33, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:33:24,221 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.102e+02 1.533e+02 1.830e+02 2.226e+02 4.351e+02, threshold=3.661e+02, percent-clipped=3.0 2023-03-27 01:33:24,302 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117158.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:33:26,115 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117161.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:33:38,582 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 2023-03-27 01:34:06,271 INFO [finetune.py:976] (3/7) Epoch 21, batch 2650, loss[loss=0.1547, simple_loss=0.2452, pruned_loss=0.03213, over 4816.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2493, pruned_loss=0.05286, over 952077.86 frames. ], batch size: 39, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:34:09,385 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=117209.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:34:26,089 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117221.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:35:03,766 INFO [finetune.py:976] (3/7) Epoch 21, batch 2700, loss[loss=0.1929, simple_loss=0.2679, pruned_loss=0.05895, over 4835.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2483, pruned_loss=0.05211, over 954686.20 frames. ], batch size: 47, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:35:06,200 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.072e+02 1.522e+02 1.732e+02 2.127e+02 4.053e+02, threshold=3.464e+02, percent-clipped=3.0 2023-03-27 01:35:23,781 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117274.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:35:30,663 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117282.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:35:41,276 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0833, 1.9358, 1.6884, 1.7175, 1.8097, 1.7915, 1.8502, 2.4992], device='cuda:3'), covar=tensor([0.3348, 0.3602, 0.2901, 0.3362, 0.3598, 0.2214, 0.3277, 0.1499], device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0261, 0.0232, 0.0276, 0.0252, 0.0222, 0.0251, 0.0234], 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') 2023-03-27 01:35:45,278 INFO [finetune.py:976] (3/7) Epoch 21, batch 2750, loss[loss=0.1705, simple_loss=0.2438, pruned_loss=0.04862, over 4898.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.246, pruned_loss=0.0515, over 956486.34 frames. ], batch size: 32, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:35:54,485 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2131, 2.0061, 2.0942, 1.5358, 1.9847, 2.1924, 2.1576, 1.6560], device='cuda:3'), covar=tensor([0.0415, 0.0611, 0.0561, 0.0787, 0.0684, 0.0589, 0.0468, 0.1125], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0136, 0.0140, 0.0121, 0.0125, 0.0139, 0.0141, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 01:35:56,249 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=117322.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:36:01,563 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117329.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:36:18,643 INFO [finetune.py:976] (3/7) Epoch 21, batch 2800, loss[loss=0.1412, simple_loss=0.2125, pruned_loss=0.035, over 4767.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2436, pruned_loss=0.05097, over 956555.76 frames. ], batch size: 28, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:36:21,562 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.020e+02 1.495e+02 1.752e+02 2.115e+02 2.888e+02, threshold=3.503e+02, percent-clipped=0.0 2023-03-27 01:36:22,887 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117360.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:36:32,672 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-03-27 01:36:42,944 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117390.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:36:44,800 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9183, 2.6923, 2.5064, 1.4855, 2.6416, 2.1544, 2.0790, 2.4796], device='cuda:3'), covar=tensor([0.0947, 0.0672, 0.1558, 0.1892, 0.1586, 0.2055, 0.1999, 0.0952], device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0189, 0.0196, 0.0181, 0.0207, 0.0206, 0.0219, 0.0193], 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') 2023-03-27 01:36:49,644 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4909, 1.4718, 2.0364, 2.8405, 1.9847, 2.0761, 1.2104, 2.4389], device='cuda:3'), covar=tensor([0.1530, 0.1373, 0.1055, 0.0625, 0.0793, 0.1546, 0.1491, 0.0492], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0116, 0.0133, 0.0163, 0.0101, 0.0136, 0.0124, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 01:36:52,478 INFO [finetune.py:976] (3/7) Epoch 21, batch 2850, loss[loss=0.1574, simple_loss=0.2275, pruned_loss=0.04359, over 4817.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2422, pruned_loss=0.05103, over 956900.75 frames. ], batch size: 38, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:36:52,587 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6201, 1.5246, 1.4761, 1.5905, 1.2267, 3.1106, 1.3222, 1.7048], device='cuda:3'), covar=tensor([0.3024, 0.2199, 0.1976, 0.2192, 0.1588, 0.0246, 0.2700, 0.1169], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0115, 0.0120, 0.0122, 0.0113, 0.0096, 0.0094, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 01:36:54,972 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=117408.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:37:25,545 INFO [finetune.py:976] (3/7) Epoch 21, batch 2900, loss[loss=0.1242, simple_loss=0.2027, pruned_loss=0.02282, over 4721.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2456, pruned_loss=0.0526, over 955364.81 frames. ], batch size: 23, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:37:28,391 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.016e+02 1.554e+02 1.875e+02 2.295e+02 6.888e+02, threshold=3.749e+02, percent-clipped=2.0 2023-03-27 01:37:28,495 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117458.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:37:52,814 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9055, 1.8422, 1.9274, 1.1823, 1.9379, 1.9116, 1.8880, 1.6011], device='cuda:3'), covar=tensor([0.0571, 0.0665, 0.0648, 0.0881, 0.0681, 0.0715, 0.0616, 0.1184], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0136, 0.0140, 0.0121, 0.0126, 0.0139, 0.0141, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 01:37:59,201 INFO [finetune.py:976] (3/7) Epoch 21, batch 2950, loss[loss=0.1832, simple_loss=0.2577, pruned_loss=0.0544, over 4826.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2487, pruned_loss=0.05348, over 955497.24 frames. ], batch size: 33, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:38:00,493 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=117506.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:38:32,257 INFO [finetune.py:976] (3/7) Epoch 21, batch 3000, loss[loss=0.1756, simple_loss=0.2403, pruned_loss=0.05547, over 4923.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2488, pruned_loss=0.0537, over 953907.11 frames. ], batch size: 42, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:38:32,257 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-27 01:38:40,540 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5776, 3.4859, 3.3589, 1.4347, 3.6250, 2.8026, 0.8165, 2.3930], device='cuda:3'), covar=tensor([0.2075, 0.1722, 0.1483, 0.3309, 0.1049, 0.0996, 0.3929, 0.1494], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0180, 0.0159, 0.0131, 0.0162, 0.0124, 0.0149, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 01:38:42,800 INFO [finetune.py:1010] (3/7) Epoch 21, validation: loss=0.1567, simple_loss=0.2253, pruned_loss=0.04408, over 2265189.00 frames. 2023-03-27 01:38:42,800 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6434MB 2023-03-27 01:38:45,673 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.002e+02 1.534e+02 1.924e+02 2.362e+02 3.621e+02, threshold=3.849e+02, percent-clipped=0.0 2023-03-27 01:39:00,154 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117577.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:39:17,642 INFO [finetune.py:976] (3/7) Epoch 21, batch 3050, loss[loss=0.1871, simple_loss=0.2636, pruned_loss=0.05531, over 4893.00 frames. ], tot_loss[loss=0.1787, simple_loss=0.2499, pruned_loss=0.05376, over 954291.81 frames. ], batch size: 43, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:39:55,169 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.0092, 4.3737, 4.5385, 4.8190, 4.7336, 4.3951, 5.0782, 1.4973], device='cuda:3'), covar=tensor([0.0655, 0.0739, 0.0774, 0.0782, 0.1108, 0.1549, 0.0509, 0.5651], device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0239, 0.0277, 0.0289, 0.0330, 0.0281, 0.0300, 0.0296], 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') 2023-03-27 01:40:13,667 INFO [finetune.py:976] (3/7) Epoch 21, batch 3100, loss[loss=0.1708, simple_loss=0.2476, pruned_loss=0.04696, over 4813.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2475, pruned_loss=0.05258, over 954191.12 frames. ], batch size: 41, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:40:19,669 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.886e+01 1.482e+02 1.759e+02 2.208e+02 4.258e+02, threshold=3.518e+02, percent-clipped=1.0 2023-03-27 01:40:46,800 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117685.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:40:58,311 INFO [finetune.py:976] (3/7) Epoch 21, batch 3150, loss[loss=0.1866, simple_loss=0.2483, pruned_loss=0.06241, over 4822.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2444, pruned_loss=0.05139, over 953569.25 frames. ], batch size: 30, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:41:29,883 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8591, 1.2706, 0.8707, 1.7209, 2.2030, 1.5959, 1.5675, 1.7159], device='cuda:3'), covar=tensor([0.1379, 0.1976, 0.1916, 0.1200, 0.1880, 0.1972, 0.1398, 0.1904], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0095, 0.0110, 0.0092, 0.0119, 0.0094, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-27 01:41:31,633 INFO [finetune.py:976] (3/7) Epoch 21, batch 3200, loss[loss=0.1558, simple_loss=0.2289, pruned_loss=0.04136, over 4834.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2417, pruned_loss=0.05083, over 954010.69 frames. ], batch size: 47, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:41:34,036 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.069e+02 1.569e+02 1.801e+02 2.101e+02 4.822e+02, threshold=3.602e+02, percent-clipped=2.0 2023-03-27 01:42:02,035 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8129, 3.7163, 3.4488, 1.5626, 3.8542, 3.0028, 0.9036, 2.5629], device='cuda:3'), covar=tensor([0.2058, 0.2217, 0.1442, 0.3796, 0.1006, 0.0920, 0.4549, 0.1671], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0178, 0.0157, 0.0130, 0.0160, 0.0122, 0.0147, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 01:42:05,172 INFO [finetune.py:976] (3/7) Epoch 21, batch 3250, loss[loss=0.1607, simple_loss=0.2342, pruned_loss=0.04359, over 4783.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2431, pruned_loss=0.05172, over 954821.82 frames. ], batch size: 29, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:42:23,385 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.0636, 2.7407, 2.8523, 2.8240, 2.7355, 2.5830, 3.0781, 1.0251], device='cuda:3'), covar=tensor([0.1732, 0.2123, 0.2081, 0.2312, 0.2561, 0.2928, 0.1786, 0.7196], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0241, 0.0279, 0.0291, 0.0332, 0.0282, 0.0302, 0.0297], 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') 2023-03-27 01:42:38,402 INFO [finetune.py:976] (3/7) Epoch 21, batch 3300, loss[loss=0.2407, simple_loss=0.3084, pruned_loss=0.08651, over 4828.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2465, pruned_loss=0.05296, over 952853.40 frames. ], batch size: 49, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:42:40,847 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.148e+02 1.638e+02 1.917e+02 2.241e+02 9.038e+02, threshold=3.833e+02, percent-clipped=2.0 2023-03-27 01:42:54,834 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117877.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:43:11,570 INFO [finetune.py:976] (3/7) Epoch 21, batch 3350, loss[loss=0.158, simple_loss=0.2431, pruned_loss=0.03647, over 4857.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2482, pruned_loss=0.05327, over 953889.69 frames. ], batch size: 44, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:43:25,775 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=117925.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:43:45,043 INFO [finetune.py:976] (3/7) Epoch 21, batch 3400, loss[loss=0.2096, simple_loss=0.2765, pruned_loss=0.07137, over 4922.00 frames. ], tot_loss[loss=0.1797, simple_loss=0.2503, pruned_loss=0.05449, over 954276.34 frames. ], batch size: 42, lr: 3.19e-03, grad_scale: 32.0 2023-03-27 01:43:47,450 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.189e+02 1.619e+02 1.880e+02 2.233e+02 5.629e+02, threshold=3.761e+02, percent-clipped=2.0 2023-03-27 01:44:05,868 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117985.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:44:18,631 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1421, 1.2396, 1.0478, 1.1879, 1.4205, 1.3721, 1.2213, 1.1205], device='cuda:3'), covar=tensor([0.0458, 0.0260, 0.0653, 0.0302, 0.0223, 0.0433, 0.0334, 0.0370], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0107, 0.0144, 0.0112, 0.0099, 0.0110, 0.0101, 0.0113], device='cuda:3'), out_proj_covar=tensor([7.5933e-05, 8.2125e-05, 1.1337e-04, 8.5943e-05, 7.7362e-05, 8.1266e-05, 7.4969e-05, 8.6007e-05], device='cuda:3') 2023-03-27 01:44:19,699 INFO [finetune.py:976] (3/7) Epoch 21, batch 3450, loss[loss=0.1438, simple_loss=0.212, pruned_loss=0.03778, over 4760.00 frames. ], tot_loss[loss=0.1797, simple_loss=0.2509, pruned_loss=0.05423, over 955907.90 frames. ], batch size: 27, lr: 3.18e-03, grad_scale: 32.0 2023-03-27 01:44:21,251 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-27 01:44:39,324 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=118033.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:44:39,409 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4811, 2.4525, 1.9179, 2.5506, 2.4208, 2.1413, 2.9405, 2.5623], device='cuda:3'), covar=tensor([0.1251, 0.2069, 0.2763, 0.2415, 0.2451, 0.1513, 0.2907, 0.1597], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0188, 0.0235, 0.0253, 0.0247, 0.0204, 0.0215, 0.0202], 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') 2023-03-27 01:44:42,942 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7370, 2.5439, 2.1621, 1.0720, 2.3086, 2.0434, 1.9221, 2.2864], device='cuda:3'), covar=tensor([0.0993, 0.0851, 0.1915, 0.2353, 0.1473, 0.2525, 0.2475, 0.1089], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0190, 0.0197, 0.0182, 0.0208, 0.0208, 0.0220, 0.0194], 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') 2023-03-27 01:44:44,127 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118040.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:44:59,763 INFO [finetune.py:976] (3/7) Epoch 21, batch 3500, loss[loss=0.2033, simple_loss=0.275, pruned_loss=0.06577, over 4818.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2485, pruned_loss=0.05327, over 956477.95 frames. ], batch size: 40, lr: 3.18e-03, grad_scale: 32.0 2023-03-27 01:45:02,217 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.127e+01 1.501e+02 1.833e+02 2.184e+02 3.839e+02, threshold=3.666e+02, percent-clipped=2.0 2023-03-27 01:45:33,202 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2083, 1.7998, 2.1721, 2.1100, 1.8686, 1.9483, 2.0744, 2.0891], device='cuda:3'), covar=tensor([0.4762, 0.4446, 0.3415, 0.4055, 0.5419, 0.4434, 0.5033, 0.3134], device='cuda:3'), in_proj_covar=tensor([0.0256, 0.0243, 0.0264, 0.0283, 0.0282, 0.0257, 0.0291, 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') 2023-03-27 01:45:52,333 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118101.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:45:57,919 INFO [finetune.py:976] (3/7) Epoch 21, batch 3550, loss[loss=0.1788, simple_loss=0.2487, pruned_loss=0.05448, over 4822.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2449, pruned_loss=0.05197, over 957130.02 frames. ], batch size: 40, lr: 3.18e-03, grad_scale: 32.0 2023-03-27 01:46:27,276 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-03-27 01:46:30,091 INFO [finetune.py:976] (3/7) Epoch 21, batch 3600, loss[loss=0.1635, simple_loss=0.2348, pruned_loss=0.04607, over 4823.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2424, pruned_loss=0.05119, over 956021.77 frames. ], batch size: 33, lr: 3.18e-03, grad_scale: 32.0 2023-03-27 01:46:33,041 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.102e+02 1.558e+02 1.902e+02 2.180e+02 3.976e+02, threshold=3.804e+02, percent-clipped=2.0 2023-03-27 01:46:39,826 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118168.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:46:51,068 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7699, 1.3205, 1.0487, 1.6690, 1.9665, 1.4096, 1.4803, 1.7211], device='cuda:3'), covar=tensor([0.1194, 0.1736, 0.1618, 0.1004, 0.1791, 0.1812, 0.1176, 0.1447], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0095, 0.0111, 0.0093, 0.0121, 0.0094, 0.0099, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 01:46:54,036 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9360, 1.8395, 1.7100, 2.0281, 2.3682, 2.0784, 1.8564, 1.6704], device='cuda:3'), covar=tensor([0.1843, 0.1822, 0.1703, 0.1410, 0.1755, 0.1137, 0.2395, 0.1756], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0210, 0.0213, 0.0195, 0.0244, 0.0189, 0.0218, 0.0204], 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') 2023-03-27 01:47:03,625 INFO [finetune.py:976] (3/7) Epoch 21, batch 3650, loss[loss=0.1944, simple_loss=0.2523, pruned_loss=0.06819, over 4787.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.245, pruned_loss=0.05253, over 954984.27 frames. ], batch size: 29, lr: 3.18e-03, grad_scale: 32.0 2023-03-27 01:47:19,812 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7915, 2.6050, 2.0506, 1.2361, 2.2270, 2.2986, 2.0791, 2.2325], device='cuda:3'), covar=tensor([0.0671, 0.0724, 0.1635, 0.1852, 0.1294, 0.1720, 0.1841, 0.0841], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0191, 0.0197, 0.0182, 0.0208, 0.0208, 0.0221, 0.0194], 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') 2023-03-27 01:47:19,822 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118229.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 01:47:36,721 INFO [finetune.py:976] (3/7) Epoch 21, batch 3700, loss[loss=0.175, simple_loss=0.2554, pruned_loss=0.04727, over 4849.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.2488, pruned_loss=0.05384, over 953241.71 frames. ], batch size: 44, lr: 3.18e-03, grad_scale: 32.0 2023-03-27 01:47:37,983 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4383, 2.4034, 2.3561, 1.6266, 2.3293, 2.4895, 2.4263, 1.9949], device='cuda:3'), covar=tensor([0.0567, 0.0540, 0.0634, 0.0848, 0.0645, 0.0646, 0.0671, 0.1035], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0135, 0.0139, 0.0120, 0.0125, 0.0138, 0.0139, 0.0160], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 01:47:39,056 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.015e+02 1.607e+02 1.941e+02 2.377e+02 3.454e+02, threshold=3.882e+02, percent-clipped=0.0 2023-03-27 01:47:46,805 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118269.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:48:00,444 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118289.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:48:10,297 INFO [finetune.py:976] (3/7) Epoch 21, batch 3750, loss[loss=0.1256, simple_loss=0.207, pruned_loss=0.0221, over 4774.00 frames. ], tot_loss[loss=0.1799, simple_loss=0.2502, pruned_loss=0.05483, over 953895.77 frames. ], batch size: 26, lr: 3.18e-03, grad_scale: 32.0 2023-03-27 01:48:26,328 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118329.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:48:26,956 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118330.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:48:40,955 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118350.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:48:40,975 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2510, 2.2661, 1.8866, 2.3040, 2.1239, 2.1000, 2.0335, 2.9747], device='cuda:3'), covar=tensor([0.3749, 0.4557, 0.3327, 0.4123, 0.4647, 0.2464, 0.4491, 0.1612], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0262, 0.0233, 0.0277, 0.0253, 0.0223, 0.0252, 0.0235], 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') 2023-03-27 01:48:43,714 INFO [finetune.py:976] (3/7) Epoch 21, batch 3800, loss[loss=0.183, simple_loss=0.2555, pruned_loss=0.05523, over 4899.00 frames. ], tot_loss[loss=0.1807, simple_loss=0.2512, pruned_loss=0.05506, over 953563.44 frames. ], batch size: 36, lr: 3.18e-03, grad_scale: 32.0 2023-03-27 01:48:46,090 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.364e+01 1.561e+02 1.815e+02 2.293e+02 4.441e+02, threshold=3.631e+02, percent-clipped=1.0 2023-03-27 01:49:06,637 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118390.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:49:08,519 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-03-27 01:49:11,211 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118396.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:49:17,071 INFO [finetune.py:976] (3/7) Epoch 21, batch 3850, loss[loss=0.1862, simple_loss=0.2576, pruned_loss=0.05736, over 4815.00 frames. ], tot_loss[loss=0.1792, simple_loss=0.2499, pruned_loss=0.05426, over 954597.30 frames. ], batch size: 39, lr: 3.18e-03, grad_scale: 32.0 2023-03-27 01:49:41,134 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6953, 1.5948, 1.4209, 1.7939, 1.7135, 1.7430, 1.0552, 1.4800], device='cuda:3'), covar=tensor([0.2099, 0.1967, 0.1880, 0.1506, 0.1378, 0.1171, 0.2375, 0.1806], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0210, 0.0212, 0.0194, 0.0243, 0.0188, 0.0217, 0.0202], 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') 2023-03-27 01:49:50,287 INFO [finetune.py:976] (3/7) Epoch 21, batch 3900, loss[loss=0.159, simple_loss=0.2284, pruned_loss=0.04479, over 4888.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2465, pruned_loss=0.05297, over 954084.75 frames. ], batch size: 32, lr: 3.18e-03, grad_scale: 32.0 2023-03-27 01:49:52,686 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.028e+02 1.530e+02 1.819e+02 2.327e+02 4.856e+02, threshold=3.639e+02, percent-clipped=2.0 2023-03-27 01:50:25,016 INFO [finetune.py:976] (3/7) Epoch 21, batch 3950, loss[loss=0.1513, simple_loss=0.2315, pruned_loss=0.03555, over 4786.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2449, pruned_loss=0.05281, over 954847.02 frames. ], batch size: 28, lr: 3.18e-03, grad_scale: 32.0 2023-03-27 01:50:43,584 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118524.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 01:50:50,448 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5038, 1.3982, 1.5009, 0.7440, 1.5027, 1.5427, 1.4790, 1.2940], device='cuda:3'), covar=tensor([0.0594, 0.0767, 0.0685, 0.0972, 0.0833, 0.0709, 0.0637, 0.1264], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0134, 0.0138, 0.0119, 0.0124, 0.0137, 0.0138, 0.0159], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 01:50:53,084 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-27 01:51:19,654 INFO [finetune.py:976] (3/7) Epoch 21, batch 4000, loss[loss=0.1725, simple_loss=0.2381, pruned_loss=0.05342, over 4889.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2449, pruned_loss=0.05305, over 957085.01 frames. ], batch size: 32, lr: 3.18e-03, grad_scale: 32.0 2023-03-27 01:51:26,595 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.742e+01 1.543e+02 1.809e+02 2.201e+02 4.154e+02, threshold=3.618e+02, percent-clipped=2.0 2023-03-27 01:51:27,082 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-27 01:51:56,640 INFO [finetune.py:976] (3/7) Epoch 21, batch 4050, loss[loss=0.1908, simple_loss=0.2748, pruned_loss=0.05343, over 4836.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2478, pruned_loss=0.05401, over 956630.00 frames. ], batch size: 49, lr: 3.18e-03, grad_scale: 32.0 2023-03-27 01:52:11,486 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118625.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:52:15,691 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118631.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:52:24,639 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118645.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:52:30,074 INFO [finetune.py:976] (3/7) Epoch 21, batch 4100, loss[loss=0.1643, simple_loss=0.2389, pruned_loss=0.04481, over 4807.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.2491, pruned_loss=0.05391, over 956765.01 frames. ], batch size: 45, lr: 3.18e-03, grad_scale: 64.0 2023-03-27 01:52:33,510 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.828e+01 1.601e+02 1.824e+02 2.338e+02 3.980e+02, threshold=3.647e+02, percent-clipped=1.0 2023-03-27 01:52:51,552 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118685.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:52:56,337 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118692.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:52:58,684 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118696.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:53:03,518 INFO [finetune.py:976] (3/7) Epoch 21, batch 4150, loss[loss=0.1924, simple_loss=0.2644, pruned_loss=0.06017, over 4800.00 frames. ], tot_loss[loss=0.1821, simple_loss=0.2523, pruned_loss=0.05595, over 954384.61 frames. ], batch size: 45, lr: 3.18e-03, grad_scale: 64.0 2023-03-27 01:53:30,072 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6199, 1.6105, 1.4218, 1.7985, 1.9289, 1.8021, 1.3739, 1.3751], device='cuda:3'), covar=tensor([0.2475, 0.2069, 0.2044, 0.1636, 0.1651, 0.1270, 0.2419, 0.2016], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0209, 0.0212, 0.0194, 0.0242, 0.0188, 0.0216, 0.0202], 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') 2023-03-27 01:53:31,224 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=118744.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:53:35,899 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-27 01:53:37,408 INFO [finetune.py:976] (3/7) Epoch 21, batch 4200, loss[loss=0.1318, simple_loss=0.1952, pruned_loss=0.03425, over 3948.00 frames. ], tot_loss[loss=0.1803, simple_loss=0.2514, pruned_loss=0.05455, over 954207.80 frames. ], batch size: 17, lr: 3.18e-03, grad_scale: 64.0 2023-03-27 01:53:39,820 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.932e+01 1.492e+02 1.761e+02 2.066e+02 3.902e+02, threshold=3.521e+02, percent-clipped=1.0 2023-03-27 01:53:45,702 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3826, 2.1856, 1.6766, 0.7926, 1.9607, 1.9162, 1.7079, 2.0432], device='cuda:3'), covar=tensor([0.0856, 0.0811, 0.1588, 0.2001, 0.1265, 0.2351, 0.2510, 0.0911], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0192, 0.0199, 0.0183, 0.0209, 0.0209, 0.0223, 0.0195], 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') 2023-03-27 01:54:11,363 INFO [finetune.py:976] (3/7) Epoch 21, batch 4250, loss[loss=0.1256, simple_loss=0.1949, pruned_loss=0.02813, over 4838.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2486, pruned_loss=0.05313, over 954558.20 frames. ], batch size: 25, lr: 3.18e-03, grad_scale: 64.0 2023-03-27 01:54:25,975 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118824.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:54:45,141 INFO [finetune.py:976] (3/7) Epoch 21, batch 4300, loss[loss=0.1407, simple_loss=0.2166, pruned_loss=0.03245, over 4799.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2459, pruned_loss=0.05226, over 955819.66 frames. ], batch size: 29, lr: 3.18e-03, grad_scale: 64.0 2023-03-27 01:54:47,576 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.108e+02 1.563e+02 1.855e+02 2.179e+02 3.656e+02, threshold=3.709e+02, percent-clipped=1.0 2023-03-27 01:54:57,553 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=118872.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:55:18,872 INFO [finetune.py:976] (3/7) Epoch 21, batch 4350, loss[loss=0.1653, simple_loss=0.2356, pruned_loss=0.04751, over 4934.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.243, pruned_loss=0.05121, over 956623.51 frames. ], batch size: 33, lr: 3.18e-03, grad_scale: 64.0 2023-03-27 01:55:33,225 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118925.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:55:48,468 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118945.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:55:59,298 INFO [finetune.py:976] (3/7) Epoch 21, batch 4400, loss[loss=0.156, simple_loss=0.2317, pruned_loss=0.04019, over 4788.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2449, pruned_loss=0.05236, over 955911.23 frames. ], batch size: 29, lr: 3.18e-03, grad_scale: 64.0 2023-03-27 01:56:01,709 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.753e+01 1.466e+02 1.745e+02 2.136e+02 3.634e+02, threshold=3.490e+02, percent-clipped=0.0 2023-03-27 01:56:18,415 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=118973.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:56:31,289 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118985.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:56:36,934 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118987.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:56:39,994 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9642, 1.5868, 2.1132, 1.5233, 1.9841, 2.1828, 1.4855, 2.2725], device='cuda:3'), covar=tensor([0.1253, 0.2363, 0.1409, 0.1830, 0.0955, 0.1315, 0.3183, 0.0903], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0205, 0.0191, 0.0189, 0.0174, 0.0213, 0.0216, 0.0199], 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') 2023-03-27 01:56:40,530 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=118993.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:56:51,290 INFO [finetune.py:976] (3/7) Epoch 21, batch 4450, loss[loss=0.1616, simple_loss=0.2353, pruned_loss=0.04397, over 4748.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2476, pruned_loss=0.05345, over 955122.70 frames. ], batch size: 23, lr: 3.18e-03, grad_scale: 64.0 2023-03-27 01:57:11,423 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=119033.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:57:16,763 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.1547, 1.2984, 1.3399, 0.6850, 1.3274, 1.5270, 1.6009, 1.2403], device='cuda:3'), covar=tensor([0.0923, 0.0567, 0.0551, 0.0507, 0.0525, 0.0609, 0.0309, 0.0692], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0149, 0.0126, 0.0123, 0.0130, 0.0128, 0.0141, 0.0147], device='cuda:3'), out_proj_covar=tensor([8.9306e-05, 1.0747e-04, 8.9880e-05, 8.6632e-05, 9.1639e-05, 9.1562e-05, 1.0119e-04, 1.0565e-04], device='cuda:3') 2023-03-27 01:57:20,387 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-03-27 01:57:25,005 INFO [finetune.py:976] (3/7) Epoch 21, batch 4500, loss[loss=0.2004, simple_loss=0.2702, pruned_loss=0.06526, over 4866.00 frames. ], tot_loss[loss=0.1798, simple_loss=0.2502, pruned_loss=0.05467, over 956232.91 frames. ], batch size: 34, lr: 3.18e-03, grad_scale: 64.0 2023-03-27 01:57:27,416 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.079e+02 1.548e+02 1.910e+02 2.429e+02 4.520e+02, threshold=3.820e+02, percent-clipped=3.0 2023-03-27 01:57:37,163 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8117, 1.0783, 1.7819, 1.7820, 1.6022, 1.5709, 1.7101, 1.7391], device='cuda:3'), covar=tensor([0.3580, 0.3616, 0.3006, 0.3474, 0.4547, 0.3600, 0.4000, 0.2888], device='cuda:3'), in_proj_covar=tensor([0.0254, 0.0241, 0.0263, 0.0281, 0.0279, 0.0256, 0.0289, 0.0244], 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') 2023-03-27 01:57:38,337 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119075.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:57:49,672 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8818, 2.6079, 2.1396, 1.2377, 2.3772, 2.2234, 1.9620, 2.3193], device='cuda:3'), covar=tensor([0.0698, 0.0735, 0.1407, 0.1795, 0.1092, 0.1713, 0.1911, 0.0834], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0191, 0.0199, 0.0183, 0.0210, 0.0209, 0.0223, 0.0196], 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') 2023-03-27 01:57:51,474 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119093.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:57:58,451 INFO [finetune.py:976] (3/7) Epoch 21, batch 4550, loss[loss=0.1757, simple_loss=0.2586, pruned_loss=0.04643, over 4906.00 frames. ], tot_loss[loss=0.1797, simple_loss=0.2506, pruned_loss=0.05441, over 955743.16 frames. ], batch size: 46, lr: 3.18e-03, grad_scale: 64.0 2023-03-27 01:58:19,747 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119136.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 01:58:31,551 INFO [finetune.py:976] (3/7) Epoch 21, batch 4600, loss[loss=0.1558, simple_loss=0.2133, pruned_loss=0.04918, over 4704.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2482, pruned_loss=0.0531, over 953416.12 frames. ], batch size: 23, lr: 3.18e-03, grad_scale: 64.0 2023-03-27 01:58:31,667 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119154.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 01:58:34,459 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.081e+02 1.653e+02 1.869e+02 2.317e+02 3.451e+02, threshold=3.738e+02, percent-clipped=0.0 2023-03-27 01:58:59,855 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1254, 1.9439, 1.7392, 1.6925, 1.8413, 1.8475, 1.8913, 2.5317], device='cuda:3'), covar=tensor([0.3775, 0.3970, 0.3171, 0.3462, 0.3844, 0.2293, 0.3423, 0.1785], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0262, 0.0232, 0.0277, 0.0254, 0.0224, 0.0253, 0.0236], 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') 2023-03-27 01:59:05,262 INFO [finetune.py:976] (3/7) Epoch 21, batch 4650, loss[loss=0.1842, simple_loss=0.254, pruned_loss=0.05722, over 4808.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2466, pruned_loss=0.05256, over 954118.62 frames. ], batch size: 25, lr: 3.18e-03, grad_scale: 64.0 2023-03-27 01:59:05,380 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4895, 2.4781, 2.1094, 1.0627, 2.2516, 1.9737, 1.8109, 2.2003], device='cuda:3'), covar=tensor([0.1039, 0.0648, 0.1632, 0.1986, 0.1420, 0.2209, 0.2070, 0.1022], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0192, 0.0200, 0.0183, 0.0211, 0.0210, 0.0224, 0.0196], 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') 2023-03-27 01:59:37,810 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7075, 1.2224, 0.8971, 1.6618, 2.1514, 1.5420, 1.3175, 1.5474], device='cuda:3'), covar=tensor([0.1519, 0.2283, 0.2125, 0.1280, 0.1886, 0.2087, 0.1636, 0.2127], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0095, 0.0111, 0.0092, 0.0120, 0.0093, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 01:59:38,067 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.19 vs. limit=5.0 2023-03-27 01:59:38,315 INFO [finetune.py:976] (3/7) Epoch 21, batch 4700, loss[loss=0.1497, simple_loss=0.2147, pruned_loss=0.04234, over 4818.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2437, pruned_loss=0.05151, over 956168.78 frames. ], batch size: 51, lr: 3.18e-03, grad_scale: 64.0 2023-03-27 01:59:40,727 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.102e+02 1.521e+02 1.791e+02 2.382e+02 6.096e+02, threshold=3.583e+02, percent-clipped=7.0 2023-03-27 01:59:59,487 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119287.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:00:05,288 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1285, 3.6221, 3.7838, 3.9832, 3.9044, 3.6519, 4.1948, 1.2531], device='cuda:3'), covar=tensor([0.0820, 0.0867, 0.0852, 0.0949, 0.1313, 0.1714, 0.0819, 0.6045], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0244, 0.0280, 0.0293, 0.0335, 0.0285, 0.0304, 0.0300], 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') 2023-03-27 02:00:11,568 INFO [finetune.py:976] (3/7) Epoch 21, batch 4750, loss[loss=0.2044, simple_loss=0.2759, pruned_loss=0.0665, over 4749.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2425, pruned_loss=0.05154, over 958244.18 frames. ], batch size: 59, lr: 3.17e-03, grad_scale: 64.0 2023-03-27 02:00:31,335 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=119335.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:00:36,544 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3114, 1.4754, 1.1984, 1.4787, 1.7489, 1.6526, 1.4149, 1.3062], device='cuda:3'), covar=tensor([0.0433, 0.0349, 0.0645, 0.0304, 0.0222, 0.0530, 0.0356, 0.0461], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0106, 0.0143, 0.0111, 0.0099, 0.0110, 0.0100, 0.0112], device='cuda:3'), out_proj_covar=tensor([7.5983e-05, 8.1465e-05, 1.1248e-04, 8.5469e-05, 7.6758e-05, 8.1520e-05, 7.4604e-05, 8.5532e-05], device='cuda:3') 2023-03-27 02:00:44,657 INFO [finetune.py:976] (3/7) Epoch 21, batch 4800, loss[loss=0.1704, simple_loss=0.2543, pruned_loss=0.04326, over 4804.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2458, pruned_loss=0.05277, over 956325.82 frames. ], batch size: 45, lr: 3.17e-03, grad_scale: 64.0 2023-03-27 02:00:47,498 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.052e+02 1.488e+02 1.781e+02 2.195e+02 3.360e+02, threshold=3.562e+02, percent-clipped=0.0 2023-03-27 02:01:22,405 INFO [finetune.py:976] (3/7) Epoch 21, batch 4850, loss[loss=0.1582, simple_loss=0.238, pruned_loss=0.03923, over 4917.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2487, pruned_loss=0.05323, over 957033.73 frames. ], batch size: 38, lr: 3.17e-03, grad_scale: 64.0 2023-03-27 02:01:55,525 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119431.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:02:15,264 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119449.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 02:02:19,177 INFO [finetune.py:976] (3/7) Epoch 21, batch 4900, loss[loss=0.2762, simple_loss=0.3253, pruned_loss=0.1136, over 4200.00 frames. ], tot_loss[loss=0.1805, simple_loss=0.2516, pruned_loss=0.05467, over 955993.29 frames. ], batch size: 66, lr: 3.17e-03, grad_scale: 32.0 2023-03-27 02:02:25,216 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.094e+02 1.684e+02 1.937e+02 2.365e+02 4.201e+02, threshold=3.874e+02, percent-clipped=2.0 2023-03-27 02:02:55,581 INFO [finetune.py:976] (3/7) Epoch 21, batch 4950, loss[loss=0.1286, simple_loss=0.2108, pruned_loss=0.02324, over 4753.00 frames. ], tot_loss[loss=0.1812, simple_loss=0.2527, pruned_loss=0.05485, over 957114.53 frames. ], batch size: 26, lr: 3.17e-03, grad_scale: 32.0 2023-03-27 02:03:02,767 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119514.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:03:29,005 INFO [finetune.py:976] (3/7) Epoch 21, batch 5000, loss[loss=0.1729, simple_loss=0.2537, pruned_loss=0.04601, over 4807.00 frames. ], tot_loss[loss=0.1795, simple_loss=0.2501, pruned_loss=0.0544, over 954484.28 frames. ], batch size: 45, lr: 3.17e-03, grad_scale: 32.0 2023-03-27 02:03:30,188 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8657, 1.7284, 1.9550, 1.4001, 1.8089, 2.0685, 1.9533, 1.4809], device='cuda:3'), covar=tensor([0.0441, 0.0613, 0.0506, 0.0743, 0.1041, 0.0416, 0.0472, 0.1125], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0136, 0.0139, 0.0120, 0.0126, 0.0138, 0.0140, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 02:03:32,983 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.061e+02 1.554e+02 1.853e+02 2.138e+02 3.358e+02, threshold=3.705e+02, percent-clipped=0.0 2023-03-27 02:03:43,303 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119575.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:04:02,226 INFO [finetune.py:976] (3/7) Epoch 21, batch 5050, loss[loss=0.1665, simple_loss=0.2483, pruned_loss=0.04233, over 4857.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2479, pruned_loss=0.05376, over 953333.77 frames. ], batch size: 49, lr: 3.17e-03, grad_scale: 32.0 2023-03-27 02:04:31,737 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.9141, 3.4059, 3.6156, 3.7946, 3.6965, 3.4641, 3.9790, 1.3983], device='cuda:3'), covar=tensor([0.0977, 0.0956, 0.0922, 0.1005, 0.1496, 0.1543, 0.0807, 0.5599], device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0245, 0.0283, 0.0295, 0.0337, 0.0286, 0.0306, 0.0302], 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') 2023-03-27 02:04:35,257 INFO [finetune.py:976] (3/7) Epoch 21, batch 5100, loss[loss=0.1665, simple_loss=0.2257, pruned_loss=0.05366, over 4916.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2435, pruned_loss=0.05202, over 952945.12 frames. ], batch size: 37, lr: 3.17e-03, grad_scale: 32.0 2023-03-27 02:04:39,201 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.052e+02 1.479e+02 1.750e+02 2.173e+02 3.976e+02, threshold=3.500e+02, percent-clipped=1.0 2023-03-27 02:04:43,509 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119665.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:05:08,856 INFO [finetune.py:976] (3/7) Epoch 21, batch 5150, loss[loss=0.1719, simple_loss=0.2434, pruned_loss=0.05022, over 4858.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2441, pruned_loss=0.05257, over 950413.13 frames. ], batch size: 31, lr: 3.17e-03, grad_scale: 32.0 2023-03-27 02:05:24,652 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119726.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:05:27,583 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119731.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:05:38,882 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119749.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 02:05:42,228 INFO [finetune.py:976] (3/7) Epoch 21, batch 5200, loss[loss=0.2307, simple_loss=0.3012, pruned_loss=0.08012, over 4728.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.2481, pruned_loss=0.0541, over 950263.98 frames. ], batch size: 54, lr: 3.17e-03, grad_scale: 32.0 2023-03-27 02:05:45,723 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.011e+02 1.656e+02 1.877e+02 2.270e+02 4.720e+02, threshold=3.754e+02, percent-clipped=1.0 2023-03-27 02:05:59,329 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=119779.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:06:10,789 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=119797.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:06:15,131 INFO [finetune.py:976] (3/7) Epoch 21, batch 5250, loss[loss=0.1867, simple_loss=0.2516, pruned_loss=0.06091, over 4826.00 frames. ], tot_loss[loss=0.179, simple_loss=0.2499, pruned_loss=0.05408, over 948474.36 frames. ], batch size: 33, lr: 3.17e-03, grad_scale: 32.0 2023-03-27 02:06:34,250 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-03-27 02:06:43,662 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0242, 1.9785, 1.7763, 2.2445, 2.6496, 2.0855, 2.2570, 1.5465], device='cuda:3'), covar=tensor([0.2135, 0.2005, 0.1920, 0.1481, 0.1751, 0.1203, 0.1876, 0.2019], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0211, 0.0214, 0.0195, 0.0245, 0.0189, 0.0218, 0.0203], 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') 2023-03-27 02:06:59,457 INFO [finetune.py:976] (3/7) Epoch 21, batch 5300, loss[loss=0.1925, simple_loss=0.2584, pruned_loss=0.06332, over 4760.00 frames. ], tot_loss[loss=0.1804, simple_loss=0.2516, pruned_loss=0.05463, over 950286.65 frames. ], batch size: 26, lr: 3.17e-03, grad_scale: 32.0 2023-03-27 02:07:07,168 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.978e+01 1.512e+02 1.747e+02 2.069e+02 4.039e+02, threshold=3.495e+02, percent-clipped=1.0 2023-03-27 02:07:10,434 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.1729, 2.9266, 2.5260, 3.4909, 3.0450, 2.8012, 3.5071, 3.1133], device='cuda:3'), covar=tensor([0.1014, 0.1779, 0.2522, 0.1872, 0.2090, 0.1382, 0.1931, 0.1502], device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0187, 0.0233, 0.0251, 0.0244, 0.0202, 0.0212, 0.0199], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 02:07:19,029 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119870.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:07:19,080 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9612, 1.6962, 2.1870, 1.5642, 2.0495, 2.2044, 1.6197, 2.3346], device='cuda:3'), covar=tensor([0.1245, 0.1881, 0.1450, 0.1799, 0.0921, 0.1251, 0.2708, 0.0732], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0205, 0.0191, 0.0189, 0.0174, 0.0214, 0.0216, 0.0199], 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') 2023-03-27 02:07:54,828 INFO [finetune.py:976] (3/7) Epoch 21, batch 5350, loss[loss=0.1775, simple_loss=0.2599, pruned_loss=0.04749, over 4837.00 frames. ], tot_loss[loss=0.1797, simple_loss=0.2512, pruned_loss=0.05409, over 950700.91 frames. ], batch size: 49, lr: 3.17e-03, grad_scale: 32.0 2023-03-27 02:08:28,088 INFO [finetune.py:976] (3/7) Epoch 21, batch 5400, loss[loss=0.1516, simple_loss=0.2271, pruned_loss=0.03802, over 4823.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.2486, pruned_loss=0.05384, over 950995.33 frames. ], batch size: 25, lr: 3.17e-03, grad_scale: 32.0 2023-03-27 02:08:31,173 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.038e+02 1.474e+02 1.662e+02 2.015e+02 3.492e+02, threshold=3.324e+02, percent-clipped=0.0 2023-03-27 02:09:02,906 INFO [finetune.py:976] (3/7) Epoch 21, batch 5450, loss[loss=0.1494, simple_loss=0.2301, pruned_loss=0.03429, over 4870.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2455, pruned_loss=0.05259, over 952923.39 frames. ], batch size: 31, lr: 3.17e-03, grad_scale: 32.0 2023-03-27 02:09:13,779 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=120021.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:09:34,559 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-27 02:09:36,195 INFO [finetune.py:976] (3/7) Epoch 21, batch 5500, loss[loss=0.1658, simple_loss=0.2466, pruned_loss=0.04253, over 4816.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2427, pruned_loss=0.05166, over 954022.24 frames. ], batch size: 40, lr: 3.17e-03, grad_scale: 32.0 2023-03-27 02:09:39,679 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.953e+01 1.422e+02 1.683e+02 2.099e+02 3.794e+02, threshold=3.366e+02, percent-clipped=2.0 2023-03-27 02:10:09,942 INFO [finetune.py:976] (3/7) Epoch 21, batch 5550, loss[loss=0.1846, simple_loss=0.2744, pruned_loss=0.04743, over 4815.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2435, pruned_loss=0.05163, over 952804.25 frames. ], batch size: 39, lr: 3.17e-03, grad_scale: 32.0 2023-03-27 02:10:10,047 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3990, 1.3818, 1.9412, 1.6818, 1.5659, 3.1460, 1.2402, 1.4669], device='cuda:3'), covar=tensor([0.1005, 0.1845, 0.1120, 0.0992, 0.1563, 0.0264, 0.1595, 0.1820], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0081, 0.0074, 0.0076, 0.0091, 0.0080, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 02:10:35,885 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2282, 1.2372, 1.4215, 2.1360, 1.4725, 1.9859, 0.8257, 1.8164], device='cuda:3'), covar=tensor([0.1562, 0.1155, 0.1014, 0.0702, 0.0834, 0.1007, 0.1407, 0.0613], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0117, 0.0134, 0.0166, 0.0102, 0.0139, 0.0126, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 02:10:42,282 INFO [finetune.py:976] (3/7) Epoch 21, batch 5600, loss[loss=0.1598, simple_loss=0.2282, pruned_loss=0.04569, over 4836.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2473, pruned_loss=0.05257, over 953845.55 frames. ], batch size: 30, lr: 3.17e-03, grad_scale: 32.0 2023-03-27 02:10:45,197 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.030e+02 1.550e+02 1.831e+02 2.203e+02 3.727e+02, threshold=3.662e+02, percent-clipped=1.0 2023-03-27 02:10:51,651 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-03-27 02:10:52,152 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=120170.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:11:12,065 INFO [finetune.py:976] (3/7) Epoch 21, batch 5650, loss[loss=0.188, simple_loss=0.2742, pruned_loss=0.05087, over 4807.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.25, pruned_loss=0.05318, over 952389.25 frames. ], batch size: 45, lr: 3.17e-03, grad_scale: 32.0 2023-03-27 02:11:12,771 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9050, 1.6810, 1.5484, 1.2942, 1.6679, 1.6809, 1.6475, 2.2234], device='cuda:3'), covar=tensor([0.3693, 0.3828, 0.3048, 0.3546, 0.3625, 0.2306, 0.3466, 0.1772], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0262, 0.0232, 0.0277, 0.0253, 0.0222, 0.0252, 0.0234], 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') 2023-03-27 02:11:20,664 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=120218.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:11:30,710 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0539, 1.5862, 1.0729, 1.8617, 2.3855, 1.6338, 1.7998, 1.8955], device='cuda:3'), covar=tensor([0.1362, 0.1915, 0.1900, 0.1165, 0.1745, 0.1643, 0.1467, 0.1989], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0095, 0.0110, 0.0092, 0.0120, 0.0093, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 02:11:41,961 INFO [finetune.py:976] (3/7) Epoch 21, batch 5700, loss[loss=0.141, simple_loss=0.2038, pruned_loss=0.03908, over 4354.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2446, pruned_loss=0.05171, over 932707.07 frames. ], batch size: 19, lr: 3.17e-03, grad_scale: 32.0 2023-03-27 02:11:44,944 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.012e+02 1.458e+02 1.705e+02 2.128e+02 3.595e+02, threshold=3.409e+02, percent-clipped=0.0 2023-03-27 02:12:12,191 INFO [finetune.py:976] (3/7) Epoch 22, batch 0, loss[loss=0.1803, simple_loss=0.2603, pruned_loss=0.05013, over 4910.00 frames. ], tot_loss[loss=0.1803, simple_loss=0.2603, pruned_loss=0.05013, over 4910.00 frames. ], batch size: 38, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:12:12,192 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-27 02:12:27,778 INFO [finetune.py:1010] (3/7) Epoch 22, validation: loss=0.1597, simple_loss=0.228, pruned_loss=0.04574, over 2265189.00 frames. 2023-03-27 02:12:27,779 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6434MB 2023-03-27 02:12:35,246 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1267, 2.0169, 2.1083, 1.4677, 2.1095, 2.2812, 2.3155, 1.6076], device='cuda:3'), covar=tensor([0.0620, 0.0698, 0.0769, 0.0941, 0.0795, 0.0674, 0.0600, 0.1408], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0137, 0.0139, 0.0121, 0.0125, 0.0139, 0.0140, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 02:13:15,375 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=120321.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:13:26,049 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=2.10 vs. limit=2.0 2023-03-27 02:13:27,846 INFO [finetune.py:976] (3/7) Epoch 22, batch 50, loss[loss=0.2132, simple_loss=0.272, pruned_loss=0.07721, over 4874.00 frames. ], tot_loss[loss=0.1793, simple_loss=0.2494, pruned_loss=0.05463, over 213879.03 frames. ], batch size: 34, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:13:45,815 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0770, 1.7850, 2.1439, 2.1214, 1.8800, 1.8679, 2.1116, 1.9989], device='cuda:3'), covar=tensor([0.4287, 0.4176, 0.3170, 0.4193, 0.4960, 0.3957, 0.4810, 0.3128], device='cuda:3'), in_proj_covar=tensor([0.0256, 0.0243, 0.0264, 0.0284, 0.0282, 0.0259, 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, 0.0002], device='cuda:3') 2023-03-27 02:13:47,643 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0336, 1.9575, 1.6095, 1.7693, 1.8804, 1.7796, 1.8643, 2.4939], device='cuda:3'), covar=tensor([0.3697, 0.3730, 0.3394, 0.3635, 0.3924, 0.2563, 0.3743, 0.2003], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0262, 0.0232, 0.0276, 0.0253, 0.0223, 0.0252, 0.0234], 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') 2023-03-27 02:13:48,210 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6310, 1.2410, 0.7990, 1.5362, 2.1072, 1.4322, 1.3409, 1.7922], device='cuda:3'), covar=tensor([0.2032, 0.2879, 0.2636, 0.1733, 0.2382, 0.2653, 0.2197, 0.2732], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0095, 0.0111, 0.0092, 0.0120, 0.0094, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 02:13:48,723 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.043e+02 1.631e+02 1.972e+02 2.363e+02 4.295e+02, threshold=3.943e+02, percent-clipped=3.0 2023-03-27 02:13:55,424 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=120369.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:14:04,333 INFO [finetune.py:976] (3/7) Epoch 22, batch 100, loss[loss=0.178, simple_loss=0.2514, pruned_loss=0.05228, over 4828.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2434, pruned_loss=0.05183, over 380991.03 frames. ], batch size: 40, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:14:34,407 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-03-27 02:14:36,543 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5269, 1.4339, 1.4279, 1.4079, 0.9071, 2.3285, 0.8007, 1.1927], device='cuda:3'), covar=tensor([0.3399, 0.2727, 0.2270, 0.2524, 0.1840, 0.0332, 0.2696, 0.1349], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0116, 0.0121, 0.0124, 0.0114, 0.0096, 0.0095, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 02:14:37,021 INFO [finetune.py:976] (3/7) Epoch 22, batch 150, loss[loss=0.1968, simple_loss=0.2569, pruned_loss=0.06832, over 4874.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2394, pruned_loss=0.0513, over 509114.74 frames. ], batch size: 31, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:14:54,538 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-27 02:14:55,338 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.054e+02 1.445e+02 1.737e+02 2.098e+02 4.550e+02, threshold=3.473e+02, percent-clipped=2.0 2023-03-27 02:15:10,227 INFO [finetune.py:976] (3/7) Epoch 22, batch 200, loss[loss=0.1938, simple_loss=0.2414, pruned_loss=0.07312, over 4324.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2378, pruned_loss=0.05085, over 604420.79 frames. ], batch size: 19, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:15:42,765 INFO [finetune.py:976] (3/7) Epoch 22, batch 250, loss[loss=0.1752, simple_loss=0.249, pruned_loss=0.05069, over 4746.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2407, pruned_loss=0.05184, over 682491.75 frames. ], batch size: 59, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:15:58,122 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-27 02:16:01,906 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.094e+02 1.519e+02 1.843e+02 2.180e+02 3.548e+02, threshold=3.686e+02, percent-clipped=1.0 2023-03-27 02:16:16,409 INFO [finetune.py:976] (3/7) Epoch 22, batch 300, loss[loss=0.1865, simple_loss=0.2709, pruned_loss=0.05109, over 4905.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2455, pruned_loss=0.05317, over 742054.64 frames. ], batch size: 37, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:16:42,745 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9482, 1.8750, 1.4768, 0.7216, 1.6008, 1.6394, 1.6007, 1.7817], device='cuda:3'), covar=tensor([0.0966, 0.0647, 0.1316, 0.1738, 0.1179, 0.1956, 0.1774, 0.0752], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0191, 0.0198, 0.0183, 0.0209, 0.0207, 0.0222, 0.0195], 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') 2023-03-27 02:16:50,477 INFO [finetune.py:976] (3/7) Epoch 22, batch 350, loss[loss=0.196, simple_loss=0.2713, pruned_loss=0.06039, over 4794.00 frames. ], tot_loss[loss=0.1792, simple_loss=0.2491, pruned_loss=0.05461, over 788902.34 frames. ], batch size: 51, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:17:09,295 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.085e+02 1.647e+02 1.881e+02 2.280e+02 4.594e+02, threshold=3.762e+02, percent-clipped=3.0 2023-03-27 02:17:19,815 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=120676.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:17:23,372 INFO [finetune.py:976] (3/7) Epoch 22, batch 400, loss[loss=0.1787, simple_loss=0.2584, pruned_loss=0.04951, over 4746.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2494, pruned_loss=0.0534, over 826712.68 frames. ], batch size: 27, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:17:41,249 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5385, 1.4880, 1.2500, 1.2834, 1.6942, 1.3579, 2.0570, 1.5922], device='cuda:3'), covar=tensor([0.1649, 0.1918, 0.3518, 0.2637, 0.2725, 0.1913, 0.2016, 0.2085], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0188, 0.0234, 0.0253, 0.0247, 0.0204, 0.0214, 0.0202], 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') 2023-03-27 02:18:12,314 INFO [finetune.py:976] (3/7) Epoch 22, batch 450, loss[loss=0.2116, simple_loss=0.2893, pruned_loss=0.06696, over 4890.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2492, pruned_loss=0.05308, over 854919.65 frames. ], batch size: 35, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:18:20,755 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=120737.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:18:39,345 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3177, 2.3190, 2.3394, 1.7135, 2.3031, 2.4170, 2.4522, 1.9455], device='cuda:3'), covar=tensor([0.0529, 0.0558, 0.0635, 0.0786, 0.0593, 0.0635, 0.0522, 0.1071], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0137, 0.0141, 0.0121, 0.0126, 0.0140, 0.0140, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 02:18:43,318 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.117e+02 1.581e+02 1.866e+02 2.243e+02 3.725e+02, threshold=3.731e+02, percent-clipped=0.0 2023-03-27 02:18:47,368 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0057, 1.8132, 1.6472, 1.5685, 1.7281, 1.7580, 1.7960, 2.4796], device='cuda:3'), covar=tensor([0.3487, 0.3642, 0.2940, 0.3480, 0.3852, 0.2276, 0.3224, 0.1522], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0264, 0.0234, 0.0278, 0.0256, 0.0225, 0.0254, 0.0235], 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') 2023-03-27 02:19:07,340 INFO [finetune.py:976] (3/7) Epoch 22, batch 500, loss[loss=0.1314, simple_loss=0.1994, pruned_loss=0.03169, over 4741.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2453, pruned_loss=0.0515, over 877152.90 frames. ], batch size: 54, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:19:14,676 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.0593, 0.9630, 0.9467, 0.3964, 0.8678, 1.1386, 1.1290, 1.0009], device='cuda:3'), covar=tensor([0.0956, 0.0796, 0.0582, 0.0575, 0.0673, 0.0647, 0.0443, 0.0780], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0149, 0.0125, 0.0123, 0.0130, 0.0128, 0.0140, 0.0147], device='cuda:3'), out_proj_covar=tensor([8.9041e-05, 1.0732e-04, 8.9627e-05, 8.6422e-05, 9.1631e-05, 9.1369e-05, 1.0067e-04, 1.0563e-04], device='cuda:3') 2023-03-27 02:19:20,817 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-03-27 02:19:40,280 INFO [finetune.py:976] (3/7) Epoch 22, batch 550, loss[loss=0.1588, simple_loss=0.2406, pruned_loss=0.0385, over 4917.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2425, pruned_loss=0.05062, over 893483.64 frames. ], batch size: 36, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:19:49,971 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.0510, 3.5070, 3.7182, 3.8388, 3.8528, 3.6376, 4.1023, 1.6854], device='cuda:3'), covar=tensor([0.0657, 0.0757, 0.0799, 0.0821, 0.0959, 0.1152, 0.0638, 0.4726], device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0241, 0.0277, 0.0290, 0.0330, 0.0281, 0.0301, 0.0297], 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') 2023-03-27 02:19:58,119 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.063e+02 1.544e+02 1.778e+02 2.172e+02 3.720e+02, threshold=3.555e+02, percent-clipped=0.0 2023-03-27 02:20:10,830 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8081, 1.6975, 1.5046, 1.3980, 1.8594, 1.5591, 1.7687, 1.8111], device='cuda:3'), covar=tensor([0.1458, 0.1950, 0.3089, 0.2486, 0.2489, 0.1797, 0.2794, 0.1790], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0188, 0.0235, 0.0253, 0.0247, 0.0204, 0.0214, 0.0202], 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') 2023-03-27 02:20:13,118 INFO [finetune.py:976] (3/7) Epoch 22, batch 600, loss[loss=0.1807, simple_loss=0.247, pruned_loss=0.05721, over 4790.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2436, pruned_loss=0.05155, over 905478.72 frames. ], batch size: 29, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:20:29,859 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6399, 2.5220, 2.1124, 2.6456, 2.5809, 2.3006, 2.8538, 2.5728], device='cuda:3'), covar=tensor([0.1241, 0.1985, 0.2972, 0.2298, 0.2210, 0.1677, 0.2283, 0.1766], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0188, 0.0235, 0.0252, 0.0246, 0.0203, 0.0213, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 02:20:46,528 INFO [finetune.py:976] (3/7) Epoch 22, batch 650, loss[loss=0.1619, simple_loss=0.232, pruned_loss=0.04586, over 4706.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2456, pruned_loss=0.05218, over 915846.56 frames. ], batch size: 23, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:20:50,802 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1882, 1.4270, 0.7185, 1.9567, 2.4484, 1.7366, 1.6178, 1.8590], device='cuda:3'), covar=tensor([0.1338, 0.2066, 0.2195, 0.1128, 0.1869, 0.1909, 0.1455, 0.1924], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0095, 0.0111, 0.0092, 0.0121, 0.0093, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 02:21:01,496 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6127, 1.5772, 2.3235, 1.9945, 1.8996, 4.2463, 1.5114, 1.7721], device='cuda:3'), covar=tensor([0.0990, 0.1830, 0.1120, 0.0948, 0.1518, 0.0216, 0.1553, 0.1813], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0081, 0.0074, 0.0076, 0.0091, 0.0080, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 02:21:04,771 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.113e+02 1.580e+02 1.877e+02 2.237e+02 3.344e+02, threshold=3.754e+02, percent-clipped=0.0 2023-03-27 02:21:20,031 INFO [finetune.py:976] (3/7) Epoch 22, batch 700, loss[loss=0.2195, simple_loss=0.2857, pruned_loss=0.07663, over 4037.00 frames. ], tot_loss[loss=0.1786, simple_loss=0.2492, pruned_loss=0.05406, over 923846.98 frames. ], batch size: 65, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:21:32,897 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121003.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:21:43,179 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3564, 1.1702, 1.5351, 2.3214, 1.5875, 2.2977, 0.9376, 2.0762], device='cuda:3'), covar=tensor([0.2227, 0.2081, 0.1534, 0.1160, 0.1114, 0.1289, 0.1883, 0.0871], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0117, 0.0135, 0.0166, 0.0102, 0.0138, 0.0126, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 02:21:53,221 INFO [finetune.py:976] (3/7) Epoch 22, batch 750, loss[loss=0.2013, simple_loss=0.273, pruned_loss=0.06477, over 4892.00 frames. ], tot_loss[loss=0.1797, simple_loss=0.2506, pruned_loss=0.05442, over 932339.12 frames. ], batch size: 43, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:21:53,298 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121032.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:22:09,808 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.017e+02 1.617e+02 1.906e+02 2.410e+02 4.829e+02, threshold=3.812e+02, percent-clipped=5.0 2023-03-27 02:22:14,819 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121064.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:22:26,765 INFO [finetune.py:976] (3/7) Epoch 22, batch 800, loss[loss=0.2023, simple_loss=0.2705, pruned_loss=0.06698, over 4871.00 frames. ], tot_loss[loss=0.1783, simple_loss=0.2496, pruned_loss=0.05351, over 936590.72 frames. ], batch size: 34, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:22:28,702 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8490, 4.6127, 4.4112, 2.3342, 4.7216, 3.4273, 0.8372, 3.1870], device='cuda:3'), covar=tensor([0.2338, 0.1443, 0.1206, 0.2949, 0.0634, 0.0948, 0.4483, 0.1318], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0177, 0.0158, 0.0130, 0.0159, 0.0122, 0.0148, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 02:22:45,137 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-27 02:22:47,257 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.0242, 1.3707, 1.4371, 1.3386, 1.4233, 2.3673, 1.2625, 1.4564], device='cuda:3'), covar=tensor([0.0926, 0.1542, 0.1004, 0.0850, 0.1381, 0.0412, 0.1291, 0.1481], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0081, 0.0073, 0.0076, 0.0091, 0.0080, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 02:23:10,262 INFO [finetune.py:976] (3/7) Epoch 22, batch 850, loss[loss=0.2108, simple_loss=0.271, pruned_loss=0.07533, over 4903.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2487, pruned_loss=0.05363, over 940725.46 frames. ], batch size: 35, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:23:28,701 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.269e+01 1.509e+02 1.830e+02 2.168e+02 4.982e+02, threshold=3.659e+02, percent-clipped=1.0 2023-03-27 02:23:37,198 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121164.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:23:48,697 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4778, 1.3861, 1.3895, 1.3623, 0.8082, 2.2319, 0.6677, 1.0942], device='cuda:3'), covar=tensor([0.3332, 0.2498, 0.2182, 0.2459, 0.1872, 0.0354, 0.2772, 0.1358], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0121, 0.0124, 0.0114, 0.0096, 0.0095, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 02:23:58,239 INFO [finetune.py:976] (3/7) Epoch 22, batch 900, loss[loss=0.1695, simple_loss=0.2502, pruned_loss=0.04443, over 4907.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2455, pruned_loss=0.05234, over 944663.62 frames. ], batch size: 37, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:24:09,537 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.3184, 1.2943, 1.4430, 0.8070, 1.4300, 1.6189, 1.6560, 1.3163], device='cuda:3'), covar=tensor([0.0859, 0.0777, 0.0505, 0.0526, 0.0482, 0.0542, 0.0321, 0.0668], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0150, 0.0125, 0.0123, 0.0130, 0.0129, 0.0141, 0.0147], device='cuda:3'), out_proj_covar=tensor([8.9136e-05, 1.0803e-04, 8.9656e-05, 8.6466e-05, 9.1737e-05, 9.1740e-05, 1.0085e-04, 1.0542e-04], device='cuda:3') 2023-03-27 02:24:24,201 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-03-27 02:24:38,323 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121225.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:24:42,941 INFO [finetune.py:976] (3/7) Epoch 22, batch 950, loss[loss=0.2362, simple_loss=0.3094, pruned_loss=0.08154, over 4931.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2442, pruned_loss=0.0522, over 946474.66 frames. ], batch size: 38, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:24:59,302 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.960e+01 1.496e+02 1.804e+02 2.225e+02 4.174e+02, threshold=3.608e+02, percent-clipped=1.0 2023-03-27 02:25:09,447 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7287, 1.6316, 1.4830, 1.3963, 1.8338, 1.5244, 1.8687, 1.7769], device='cuda:3'), covar=tensor([0.1336, 0.1949, 0.2820, 0.2440, 0.2441, 0.1593, 0.2791, 0.1641], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0187, 0.0234, 0.0253, 0.0246, 0.0203, 0.0213, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 02:25:16,226 INFO [finetune.py:976] (3/7) Epoch 22, batch 1000, loss[loss=0.2078, simple_loss=0.2774, pruned_loss=0.06907, over 4817.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2481, pruned_loss=0.05361, over 948644.83 frames. ], batch size: 38, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:25:33,741 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121311.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 02:25:42,364 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3550, 2.4101, 1.7984, 2.5317, 2.4111, 1.9765, 2.8436, 2.4737], device='cuda:3'), covar=tensor([0.1318, 0.2009, 0.3000, 0.2530, 0.2424, 0.1667, 0.3276, 0.1646], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0188, 0.0235, 0.0254, 0.0247, 0.0204, 0.0214, 0.0202], 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') 2023-03-27 02:25:49,276 INFO [finetune.py:976] (3/7) Epoch 22, batch 1050, loss[loss=0.1991, simple_loss=0.2722, pruned_loss=0.06298, over 4828.00 frames. ], tot_loss[loss=0.1799, simple_loss=0.2511, pruned_loss=0.0544, over 951134.34 frames. ], batch size: 30, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:25:49,382 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121332.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:25:51,812 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.2085, 1.4281, 1.4757, 0.7540, 1.4606, 1.6381, 1.6673, 1.3986], device='cuda:3'), covar=tensor([0.0896, 0.0787, 0.0606, 0.0537, 0.0667, 0.0742, 0.0485, 0.0668], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0150, 0.0127, 0.0124, 0.0132, 0.0130, 0.0142, 0.0149], device='cuda:3'), out_proj_covar=tensor([8.9814e-05, 1.0860e-04, 9.0406e-05, 8.7095e-05, 9.2787e-05, 9.2661e-05, 1.0186e-04, 1.0648e-04], device='cuda:3') 2023-03-27 02:26:05,531 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.073e+02 1.613e+02 2.056e+02 2.633e+02 6.948e+02, threshold=4.113e+02, percent-clipped=5.0 2023-03-27 02:26:05,613 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121359.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:26:09,274 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7433, 2.5829, 2.0709, 1.0911, 2.1649, 2.1242, 1.9233, 2.2429], device='cuda:3'), covar=tensor([0.0830, 0.0746, 0.1637, 0.2033, 0.1519, 0.1992, 0.2115, 0.1048], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0192, 0.0199, 0.0183, 0.0210, 0.0209, 0.0223, 0.0195], 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') 2023-03-27 02:26:12,287 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6599, 3.6930, 3.4162, 1.5560, 3.8204, 2.8699, 0.7099, 2.5303], device='cuda:3'), covar=tensor([0.2332, 0.1985, 0.1623, 0.3671, 0.0974, 0.1068, 0.4621, 0.1585], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0177, 0.0158, 0.0129, 0.0159, 0.0122, 0.0147, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 02:26:13,507 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121372.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 02:26:19,242 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=121380.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:26:20,878 INFO [finetune.py:976] (3/7) Epoch 22, batch 1100, loss[loss=0.1625, simple_loss=0.2462, pruned_loss=0.03943, over 4918.00 frames. ], tot_loss[loss=0.1813, simple_loss=0.2522, pruned_loss=0.05516, over 951545.62 frames. ], batch size: 38, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:26:31,161 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0444, 1.7031, 2.4592, 3.8524, 2.5749, 2.6913, 1.0607, 3.2740], device='cuda:3'), covar=tensor([0.1727, 0.1510, 0.1372, 0.0534, 0.0866, 0.1890, 0.1806, 0.0404], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0118, 0.0136, 0.0167, 0.0102, 0.0139, 0.0127, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 02:26:36,783 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121407.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:26:53,106 INFO [finetune.py:976] (3/7) Epoch 22, batch 1150, loss[loss=0.2106, simple_loss=0.2776, pruned_loss=0.07183, over 4806.00 frames. ], tot_loss[loss=0.182, simple_loss=0.2531, pruned_loss=0.05544, over 952532.03 frames. ], batch size: 40, lr: 3.16e-03, grad_scale: 32.0 2023-03-27 02:27:10,452 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.055e+01 1.683e+02 1.964e+02 2.424e+02 3.625e+02, threshold=3.928e+02, percent-clipped=0.0 2023-03-27 02:27:15,973 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121468.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:27:25,671 INFO [finetune.py:976] (3/7) Epoch 22, batch 1200, loss[loss=0.1473, simple_loss=0.2185, pruned_loss=0.03802, over 4229.00 frames. ], tot_loss[loss=0.179, simple_loss=0.2502, pruned_loss=0.05388, over 952771.56 frames. ], batch size: 66, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:27:30,869 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3003, 1.4374, 1.6797, 1.5332, 1.5792, 2.8444, 1.3472, 1.4916], device='cuda:3'), covar=tensor([0.0977, 0.1684, 0.1108, 0.0950, 0.1421, 0.0309, 0.1394, 0.1641], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0081, 0.0073, 0.0076, 0.0091, 0.0081, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 02:27:49,434 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-27 02:27:49,694 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121520.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:27:57,417 INFO [finetune.py:976] (3/7) Epoch 22, batch 1250, loss[loss=0.1209, simple_loss=0.1942, pruned_loss=0.02381, over 4831.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2469, pruned_loss=0.05279, over 953151.52 frames. ], batch size: 30, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:28:26,678 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.015e+02 1.487e+02 1.790e+02 2.203e+02 4.512e+02, threshold=3.581e+02, percent-clipped=1.0 2023-03-27 02:28:35,233 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.3010, 1.4208, 1.3644, 0.8745, 1.3667, 1.6564, 1.6757, 1.2793], device='cuda:3'), covar=tensor([0.0971, 0.0616, 0.0584, 0.0541, 0.0551, 0.0582, 0.0290, 0.0774], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0152, 0.0128, 0.0125, 0.0134, 0.0131, 0.0143, 0.0150], device='cuda:3'), out_proj_covar=tensor([9.0836e-05, 1.0956e-04, 9.1345e-05, 8.7931e-05, 9.4020e-05, 9.3492e-05, 1.0273e-04, 1.0759e-04], device='cuda:3') 2023-03-27 02:28:40,602 INFO [finetune.py:976] (3/7) Epoch 22, batch 1300, loss[loss=0.2034, simple_loss=0.2726, pruned_loss=0.06707, over 4903.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2436, pruned_loss=0.05145, over 952115.91 frames. ], batch size: 37, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:29:39,003 INFO [finetune.py:976] (3/7) Epoch 22, batch 1350, loss[loss=0.1841, simple_loss=0.2547, pruned_loss=0.05677, over 4901.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2438, pruned_loss=0.05185, over 951767.23 frames. ], batch size: 35, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:29:56,728 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5194, 1.5096, 1.2827, 1.5086, 1.8852, 1.7972, 1.5218, 1.3320], device='cuda:3'), covar=tensor([0.0378, 0.0290, 0.0608, 0.0298, 0.0203, 0.0459, 0.0324, 0.0413], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0106, 0.0142, 0.0110, 0.0098, 0.0110, 0.0100, 0.0111], device='cuda:3'), out_proj_covar=tensor([7.5812e-05, 8.1470e-05, 1.1154e-04, 8.4802e-05, 7.6510e-05, 8.1612e-05, 7.4242e-05, 8.5099e-05], device='cuda:3') 2023-03-27 02:30:01,605 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3861, 1.6806, 1.3513, 1.4945, 1.8998, 1.8982, 1.6076, 1.7391], device='cuda:3'), covar=tensor([0.0544, 0.0289, 0.0572, 0.0304, 0.0315, 0.0517, 0.0425, 0.0329], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0106, 0.0142, 0.0110, 0.0098, 0.0110, 0.0100, 0.0111], device='cuda:3'), out_proj_covar=tensor([7.5816e-05, 8.1459e-05, 1.1150e-04, 8.4799e-05, 7.6530e-05, 8.1637e-05, 7.4276e-05, 8.5098e-05], device='cuda:3') 2023-03-27 02:30:02,099 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.135e+02 1.546e+02 1.913e+02 2.289e+02 6.231e+02, threshold=3.826e+02, percent-clipped=1.0 2023-03-27 02:30:02,193 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121659.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:30:07,017 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121667.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 02:30:10,059 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4807, 1.0103, 0.7761, 1.3107, 1.8835, 0.8114, 1.1433, 1.2665], device='cuda:3'), covar=tensor([0.1756, 0.2641, 0.1947, 0.1494, 0.2136, 0.2191, 0.1756, 0.2284], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0094, 0.0111, 0.0092, 0.0120, 0.0093, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 02:30:16,103 INFO [finetune.py:976] (3/7) Epoch 22, batch 1400, loss[loss=0.1504, simple_loss=0.2261, pruned_loss=0.03735, over 4783.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2462, pruned_loss=0.05255, over 951950.90 frames. ], batch size: 26, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:30:34,338 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=121707.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:30:40,177 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-27 02:30:49,371 INFO [finetune.py:976] (3/7) Epoch 22, batch 1450, loss[loss=0.1869, simple_loss=0.2636, pruned_loss=0.05513, over 4915.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2484, pruned_loss=0.0532, over 953677.28 frames. ], batch size: 42, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:31:08,590 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.158e+02 1.585e+02 1.838e+02 2.176e+02 4.319e+02, threshold=3.676e+02, percent-clipped=1.0 2023-03-27 02:31:11,092 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121763.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:31:14,159 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5535, 1.4199, 1.5786, 0.9374, 1.5025, 1.5937, 1.5471, 1.3453], device='cuda:3'), covar=tensor([0.0581, 0.0803, 0.0651, 0.0885, 0.0933, 0.0665, 0.0626, 0.1260], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0137, 0.0141, 0.0121, 0.0128, 0.0140, 0.0141, 0.0164], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 02:31:22,464 INFO [finetune.py:976] (3/7) Epoch 22, batch 1500, loss[loss=0.1731, simple_loss=0.2488, pruned_loss=0.04868, over 4776.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2492, pruned_loss=0.0533, over 954719.31 frames. ], batch size: 29, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:31:33,766 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2405, 1.9133, 2.2434, 2.2270, 1.9853, 1.9886, 2.2064, 2.1623], device='cuda:3'), covar=tensor([0.4431, 0.4351, 0.3417, 0.4036, 0.5413, 0.4232, 0.5000, 0.3221], device='cuda:3'), in_proj_covar=tensor([0.0257, 0.0243, 0.0263, 0.0285, 0.0283, 0.0259, 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, 0.0002], device='cuda:3') 2023-03-27 02:31:43,154 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0397, 2.0038, 1.6602, 2.0079, 1.8585, 1.8073, 1.8488, 2.6317], device='cuda:3'), covar=tensor([0.3620, 0.3766, 0.3259, 0.3439, 0.3976, 0.2464, 0.3560, 0.1531], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0262, 0.0232, 0.0277, 0.0255, 0.0225, 0.0253, 0.0234], 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') 2023-03-27 02:31:49,156 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121820.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:31:56,350 INFO [finetune.py:976] (3/7) Epoch 22, batch 1550, loss[loss=0.1699, simple_loss=0.2555, pruned_loss=0.04215, over 4888.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.249, pruned_loss=0.05265, over 954848.62 frames. ], batch size: 43, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:32:15,624 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.105e+02 1.581e+02 1.856e+02 2.152e+02 3.350e+02, threshold=3.712e+02, percent-clipped=0.0 2023-03-27 02:32:21,171 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=121868.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:32:29,536 INFO [finetune.py:976] (3/7) Epoch 22, batch 1600, loss[loss=0.1781, simple_loss=0.2451, pruned_loss=0.05561, over 4842.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2472, pruned_loss=0.05235, over 956437.33 frames. ], batch size: 44, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:32:50,757 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-27 02:33:02,674 INFO [finetune.py:976] (3/7) Epoch 22, batch 1650, loss[loss=0.1483, simple_loss=0.2201, pruned_loss=0.03825, over 4782.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2446, pruned_loss=0.05206, over 957743.23 frames. ], batch size: 29, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:33:13,713 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121950.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:33:22,466 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.023e+02 1.553e+02 1.837e+02 2.107e+02 3.976e+02, threshold=3.675e+02, percent-clipped=1.0 2023-03-27 02:33:33,841 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121967.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 02:33:46,460 INFO [finetune.py:976] (3/7) Epoch 22, batch 1700, loss[loss=0.1472, simple_loss=0.2118, pruned_loss=0.04132, over 4764.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2418, pruned_loss=0.05101, over 957306.35 frames. ], batch size: 27, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:34:13,899 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122011.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:34:16,847 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=122015.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 02:34:36,870 INFO [finetune.py:976] (3/7) Epoch 22, batch 1750, loss[loss=0.2025, simple_loss=0.2778, pruned_loss=0.06357, over 4909.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2447, pruned_loss=0.052, over 958213.22 frames. ], batch size: 37, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:34:45,627 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4898, 2.6106, 2.3181, 1.7559, 2.3685, 2.6600, 2.7037, 2.1988], device='cuda:3'), covar=tensor([0.0626, 0.0564, 0.0828, 0.0898, 0.0749, 0.0774, 0.0666, 0.1038], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0137, 0.0140, 0.0121, 0.0127, 0.0139, 0.0140, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 02:35:06,506 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.061e+02 1.629e+02 1.889e+02 2.189e+02 5.095e+02, threshold=3.778e+02, percent-clipped=2.0 2023-03-27 02:35:10,494 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122063.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:35:22,845 INFO [finetune.py:976] (3/7) Epoch 22, batch 1800, loss[loss=0.186, simple_loss=0.2557, pruned_loss=0.05811, over 4903.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2473, pruned_loss=0.05264, over 956791.08 frames. ], batch size: 36, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:35:41,159 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=122111.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:35:46,356 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122117.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:35:54,888 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-27 02:35:56,274 INFO [finetune.py:976] (3/7) Epoch 22, batch 1850, loss[loss=0.1931, simple_loss=0.2679, pruned_loss=0.05914, over 4928.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2491, pruned_loss=0.05337, over 957826.66 frames. ], batch size: 38, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:36:08,736 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-03-27 02:36:11,042 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4310, 1.3496, 1.9137, 1.6516, 1.5773, 3.1113, 1.3530, 1.5341], device='cuda:3'), covar=tensor([0.0912, 0.1846, 0.0970, 0.0950, 0.1483, 0.0261, 0.1471, 0.1681], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0074, 0.0077, 0.0092, 0.0081, 0.0086, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 02:36:12,757 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.044e+01 1.577e+02 1.909e+02 2.256e+02 5.766e+02, threshold=3.818e+02, percent-clipped=1.0 2023-03-27 02:36:27,278 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122178.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 02:36:29,614 INFO [finetune.py:976] (3/7) Epoch 22, batch 1900, loss[loss=0.1602, simple_loss=0.2253, pruned_loss=0.04752, over 4168.00 frames. ], tot_loss[loss=0.1796, simple_loss=0.2509, pruned_loss=0.05418, over 955990.70 frames. ], batch size: 18, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:36:37,634 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122195.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:36:43,860 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9187, 1.5026, 2.1416, 1.5626, 1.9937, 2.2034, 1.4974, 2.2583], device='cuda:3'), covar=tensor([0.1294, 0.2367, 0.1357, 0.1668, 0.1004, 0.1347, 0.3136, 0.0886], device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0207, 0.0193, 0.0191, 0.0176, 0.0214, 0.0219, 0.0200], 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') 2023-03-27 02:36:51,574 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.11 vs. limit=2.0 2023-03-27 02:37:03,562 INFO [finetune.py:976] (3/7) Epoch 22, batch 1950, loss[loss=0.1976, simple_loss=0.2701, pruned_loss=0.0625, over 4840.00 frames. ], tot_loss[loss=0.1785, simple_loss=0.2495, pruned_loss=0.05373, over 957305.01 frames. ], batch size: 49, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:37:12,544 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-03-27 02:37:18,273 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122256.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:37:19,977 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.844e+01 1.567e+02 1.786e+02 2.225e+02 4.203e+02, threshold=3.573e+02, percent-clipped=2.0 2023-03-27 02:37:21,869 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5430, 1.4332, 2.0099, 2.9730, 1.9488, 2.2259, 0.8895, 2.5205], device='cuda:3'), covar=tensor([0.1654, 0.1397, 0.1163, 0.0501, 0.0887, 0.1199, 0.1781, 0.0469], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0114, 0.0132, 0.0163, 0.0100, 0.0136, 0.0124, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 02:37:27,133 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2978, 1.5359, 0.8519, 2.0631, 2.5708, 1.9562, 1.7660, 1.9890], device='cuda:3'), covar=tensor([0.1233, 0.1834, 0.1934, 0.1041, 0.1604, 0.1711, 0.1293, 0.1763], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0093, 0.0109, 0.0091, 0.0118, 0.0093, 0.0098, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-27 02:37:36,888 INFO [finetune.py:976] (3/7) Epoch 22, batch 2000, loss[loss=0.1588, simple_loss=0.2276, pruned_loss=0.04498, over 4783.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2463, pruned_loss=0.05284, over 957585.20 frames. ], batch size: 29, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:37:51,522 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122306.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:37:54,670 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.63 vs. limit=2.0 2023-03-27 02:37:59,454 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122319.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:38:10,018 INFO [finetune.py:976] (3/7) Epoch 22, batch 2050, loss[loss=0.191, simple_loss=0.2659, pruned_loss=0.05801, over 4868.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2426, pruned_loss=0.0514, over 956829.16 frames. ], batch size: 31, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:38:15,503 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9032, 1.7933, 1.9326, 1.0493, 1.9531, 1.9513, 1.9351, 1.5881], device='cuda:3'), covar=tensor([0.0579, 0.0624, 0.0591, 0.0835, 0.0645, 0.0646, 0.0579, 0.1071], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0136, 0.0141, 0.0121, 0.0126, 0.0139, 0.0140, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 02:38:25,168 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6203, 2.7184, 2.5654, 1.8368, 2.5926, 2.8187, 2.9781, 2.2919], device='cuda:3'), covar=tensor([0.0605, 0.0575, 0.0635, 0.0789, 0.0645, 0.0688, 0.0514, 0.0981], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0136, 0.0141, 0.0121, 0.0126, 0.0139, 0.0140, 0.0164], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 02:38:26,819 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.990e+01 1.410e+02 1.787e+02 2.088e+02 3.673e+02, threshold=3.574e+02, percent-clipped=1.0 2023-03-27 02:38:33,639 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3687, 1.3090, 1.7155, 1.5525, 1.5438, 2.8993, 1.3003, 1.4528], device='cuda:3'), covar=tensor([0.0996, 0.1940, 0.1131, 0.0991, 0.1640, 0.0334, 0.1597, 0.1880], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0074, 0.0076, 0.0092, 0.0081, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 02:38:35,323 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8824, 1.7834, 2.2385, 1.9964, 1.8890, 3.4780, 1.7058, 1.8271], device='cuda:3'), covar=tensor([0.0788, 0.1487, 0.0900, 0.0802, 0.1334, 0.0266, 0.1292, 0.1449], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0081, 0.0073, 0.0076, 0.0092, 0.0081, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 02:38:43,005 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122380.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:38:44,605 INFO [finetune.py:976] (3/7) Epoch 22, batch 2100, loss[loss=0.1777, simple_loss=0.2418, pruned_loss=0.05673, over 4872.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2429, pruned_loss=0.05149, over 955379.62 frames. ], batch size: 31, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:39:17,012 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7701, 3.6852, 3.4974, 1.7738, 3.7931, 2.8724, 1.1492, 2.5667], device='cuda:3'), covar=tensor([0.2585, 0.1803, 0.1482, 0.3495, 0.1001, 0.0994, 0.4217, 0.1506], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0178, 0.0159, 0.0131, 0.0161, 0.0123, 0.0148, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 02:39:28,194 INFO [finetune.py:976] (3/7) Epoch 22, batch 2150, loss[loss=0.1976, simple_loss=0.2623, pruned_loss=0.06639, over 4770.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2467, pruned_loss=0.05323, over 956099.65 frames. ], batch size: 54, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:39:52,242 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1634, 1.5226, 0.6625, 1.9020, 2.4956, 1.7597, 1.5997, 1.9221], device='cuda:3'), covar=tensor([0.1498, 0.2020, 0.2246, 0.1257, 0.1809, 0.1949, 0.1485, 0.1953], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0094, 0.0110, 0.0091, 0.0119, 0.0093, 0.0098, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 02:40:03,768 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.048e+02 1.591e+02 1.910e+02 2.352e+02 5.051e+02, threshold=3.819e+02, percent-clipped=3.0 2023-03-27 02:40:13,675 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122468.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:40:16,631 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122473.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 02:40:26,431 INFO [finetune.py:976] (3/7) Epoch 22, batch 2200, loss[loss=0.1886, simple_loss=0.2757, pruned_loss=0.05075, over 4813.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2475, pruned_loss=0.05251, over 956399.84 frames. ], batch size: 40, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:40:41,751 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2020, 2.1052, 1.7066, 1.9734, 1.9596, 1.8516, 1.9684, 2.7686], device='cuda:3'), covar=tensor([0.4029, 0.4353, 0.3512, 0.4007, 0.4122, 0.2717, 0.4000, 0.1775], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0263, 0.0233, 0.0276, 0.0255, 0.0225, 0.0253, 0.0234], 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') 2023-03-27 02:40:51,371 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9413, 1.8431, 1.5654, 1.5717, 1.7261, 1.6493, 1.7281, 2.4595], device='cuda:3'), covar=tensor([0.3830, 0.3706, 0.3143, 0.3726, 0.3608, 0.2421, 0.3537, 0.1696], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0262, 0.0233, 0.0276, 0.0254, 0.0225, 0.0253, 0.0234], 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') 2023-03-27 02:40:53,163 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7743, 1.7344, 1.4627, 1.8692, 2.1985, 1.9104, 1.5699, 1.4440], device='cuda:3'), covar=tensor([0.2159, 0.1976, 0.1974, 0.1590, 0.1511, 0.1148, 0.2314, 0.1903], device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0212, 0.0216, 0.0197, 0.0245, 0.0191, 0.0220, 0.0205], 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') 2023-03-27 02:40:56,768 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122529.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:40:58,944 INFO [finetune.py:976] (3/7) Epoch 22, batch 2250, loss[loss=0.1616, simple_loss=0.2509, pruned_loss=0.03617, over 4796.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2497, pruned_loss=0.05327, over 957376.46 frames. ], batch size: 29, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:41:12,977 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122551.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:41:17,776 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.977e+01 1.468e+02 1.830e+02 2.095e+02 3.153e+02, threshold=3.659e+02, percent-clipped=0.0 2023-03-27 02:41:28,135 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.0381, 4.7268, 4.5625, 2.7332, 4.8851, 3.8580, 1.3958, 3.2897], device='cuda:3'), covar=tensor([0.2084, 0.1489, 0.1130, 0.2724, 0.0691, 0.0743, 0.3913, 0.1257], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0178, 0.0159, 0.0130, 0.0161, 0.0122, 0.0148, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 02:41:29,415 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9667, 1.4310, 2.0124, 1.9962, 1.7837, 1.7549, 1.9302, 1.8788], device='cuda:3'), covar=tensor([0.3778, 0.3848, 0.3116, 0.3390, 0.4686, 0.3387, 0.4170, 0.3054], device='cuda:3'), in_proj_covar=tensor([0.0258, 0.0243, 0.0264, 0.0284, 0.0283, 0.0259, 0.0293, 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') 2023-03-27 02:41:31,700 INFO [finetune.py:976] (3/7) Epoch 22, batch 2300, loss[loss=0.1662, simple_loss=0.2428, pruned_loss=0.04482, over 4803.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2494, pruned_loss=0.05275, over 957569.79 frames. ], batch size: 41, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:41:49,475 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122606.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:41:50,744 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.5736, 1.5446, 1.4166, 0.7743, 1.5463, 1.7301, 1.6695, 1.4077], device='cuda:3'), covar=tensor([0.0950, 0.0561, 0.0562, 0.0665, 0.0516, 0.0545, 0.0332, 0.0612], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0151, 0.0128, 0.0125, 0.0133, 0.0132, 0.0143, 0.0150], device='cuda:3'), out_proj_covar=tensor([9.0720e-05, 1.0926e-04, 9.1657e-05, 8.7785e-05, 9.3697e-05, 9.4022e-05, 1.0274e-04, 1.0728e-04], device='cuda:3') 2023-03-27 02:42:05,213 INFO [finetune.py:976] (3/7) Epoch 22, batch 2350, loss[loss=0.1964, simple_loss=0.2563, pruned_loss=0.06827, over 4310.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2488, pruned_loss=0.05307, over 955693.49 frames. ], batch size: 65, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:42:21,479 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=122654.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:42:24,441 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.047e+02 1.417e+02 1.625e+02 2.018e+02 3.172e+02, threshold=3.250e+02, percent-clipped=0.0 2023-03-27 02:42:34,177 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122675.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:42:38,344 INFO [finetune.py:976] (3/7) Epoch 22, batch 2400, loss[loss=0.1833, simple_loss=0.2443, pruned_loss=0.06121, over 4828.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2463, pruned_loss=0.05219, over 955519.72 frames. ], batch size: 33, lr: 3.15e-03, grad_scale: 64.0 2023-03-27 02:42:40,340 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2231, 1.9905, 1.7951, 1.8709, 1.8966, 1.9099, 1.9498, 2.7152], device='cuda:3'), covar=tensor([0.3399, 0.4046, 0.2991, 0.3671, 0.3665, 0.2298, 0.3498, 0.1602], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0262, 0.0233, 0.0276, 0.0254, 0.0224, 0.0252, 0.0234], 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') 2023-03-27 02:42:58,920 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7412, 1.6873, 1.4470, 1.6363, 2.0094, 1.9072, 1.7359, 1.5438], device='cuda:3'), covar=tensor([0.0329, 0.0316, 0.0589, 0.0326, 0.0199, 0.0522, 0.0272, 0.0370], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0106, 0.0142, 0.0111, 0.0099, 0.0111, 0.0100, 0.0112], device='cuda:3'), out_proj_covar=tensor([7.6245e-05, 8.1624e-05, 1.1173e-04, 8.4855e-05, 7.6679e-05, 8.2208e-05, 7.4327e-05, 8.5174e-05], device='cuda:3') 2023-03-27 02:43:11,470 INFO [finetune.py:976] (3/7) Epoch 22, batch 2450, loss[loss=0.1615, simple_loss=0.2311, pruned_loss=0.04588, over 4777.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.243, pruned_loss=0.0512, over 954900.74 frames. ], batch size: 28, lr: 3.14e-03, grad_scale: 64.0 2023-03-27 02:43:31,105 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.081e+02 1.588e+02 1.841e+02 2.130e+02 2.968e+02, threshold=3.682e+02, percent-clipped=0.0 2023-03-27 02:43:39,661 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122773.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 02:43:45,020 INFO [finetune.py:976] (3/7) Epoch 22, batch 2500, loss[loss=0.1754, simple_loss=0.2456, pruned_loss=0.05257, over 4768.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2453, pruned_loss=0.05269, over 955614.56 frames. ], batch size: 28, lr: 3.14e-03, grad_scale: 64.0 2023-03-27 02:43:49,404 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8193, 1.6744, 1.5455, 1.8634, 1.9809, 1.9071, 1.3491, 1.5391], device='cuda:3'), covar=tensor([0.2459, 0.2179, 0.2167, 0.1670, 0.1785, 0.1338, 0.2609, 0.2091], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0210, 0.0214, 0.0196, 0.0243, 0.0189, 0.0218, 0.0204], 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') 2023-03-27 02:44:21,539 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=122821.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:44:23,363 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122824.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:44:28,112 INFO [finetune.py:976] (3/7) Epoch 22, batch 2550, loss[loss=0.173, simple_loss=0.2448, pruned_loss=0.05063, over 4787.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2488, pruned_loss=0.05323, over 954689.92 frames. ], batch size: 29, lr: 3.14e-03, grad_scale: 64.0 2023-03-27 02:44:42,556 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122851.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:44:53,793 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.164e+02 1.566e+02 1.889e+02 2.331e+02 3.878e+02, threshold=3.777e+02, percent-clipped=1.0 2023-03-27 02:45:07,320 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122872.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:45:22,083 INFO [finetune.py:976] (3/7) Epoch 22, batch 2600, loss[loss=0.173, simple_loss=0.2451, pruned_loss=0.05044, over 4896.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2493, pruned_loss=0.05331, over 953866.53 frames. ], batch size: 43, lr: 3.14e-03, grad_scale: 64.0 2023-03-27 02:45:40,403 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=122899.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:46:02,139 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5250, 1.4377, 2.1366, 3.2099, 2.1332, 2.1798, 0.8966, 2.6719], device='cuda:3'), covar=tensor([0.1856, 0.1501, 0.1253, 0.0674, 0.0914, 0.1523, 0.1898, 0.0555], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0134, 0.0165, 0.0100, 0.0137, 0.0125, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 02:46:02,678 INFO [finetune.py:976] (3/7) Epoch 22, batch 2650, loss[loss=0.2077, simple_loss=0.2714, pruned_loss=0.07197, over 4825.00 frames. ], tot_loss[loss=0.1805, simple_loss=0.2522, pruned_loss=0.05437, over 954676.83 frames. ], batch size: 30, lr: 3.14e-03, grad_scale: 64.0 2023-03-27 02:46:03,434 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122933.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:46:19,588 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.898e+01 1.493e+02 1.785e+02 2.135e+02 3.458e+02, threshold=3.570e+02, percent-clipped=0.0 2023-03-27 02:46:31,751 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122975.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:46:35,902 INFO [finetune.py:976] (3/7) Epoch 22, batch 2700, loss[loss=0.1665, simple_loss=0.2481, pruned_loss=0.04245, over 4758.00 frames. ], tot_loss[loss=0.1801, simple_loss=0.2516, pruned_loss=0.05433, over 952556.12 frames. ], batch size: 54, lr: 3.14e-03, grad_scale: 64.0 2023-03-27 02:46:39,065 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6188, 1.5440, 1.5044, 1.5737, 1.0997, 3.0113, 1.1604, 1.5315], device='cuda:3'), covar=tensor([0.3234, 0.2394, 0.2116, 0.2323, 0.1728, 0.0231, 0.2534, 0.1258], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0116, 0.0121, 0.0123, 0.0114, 0.0095, 0.0094, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 02:47:04,317 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=123023.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:47:09,737 INFO [finetune.py:976] (3/7) Epoch 22, batch 2750, loss[loss=0.1972, simple_loss=0.2506, pruned_loss=0.07193, over 4312.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2487, pruned_loss=0.05364, over 953514.97 frames. ], batch size: 18, lr: 3.14e-03, grad_scale: 64.0 2023-03-27 02:47:20,821 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-27 02:47:26,635 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.062e+02 1.465e+02 1.704e+02 2.045e+02 3.535e+02, threshold=3.409e+02, percent-clipped=0.0 2023-03-27 02:47:42,466 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5557, 1.5685, 2.1874, 1.9106, 1.8342, 4.1516, 1.7026, 1.7984], device='cuda:3'), covar=tensor([0.0999, 0.1789, 0.1255, 0.0991, 0.1540, 0.0217, 0.1414, 0.1768], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0074, 0.0077, 0.0092, 0.0081, 0.0086, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 02:47:42,976 INFO [finetune.py:976] (3/7) Epoch 22, batch 2800, loss[loss=0.1379, simple_loss=0.2102, pruned_loss=0.03277, over 4822.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2448, pruned_loss=0.05201, over 955410.17 frames. ], batch size: 38, lr: 3.14e-03, grad_scale: 64.0 2023-03-27 02:48:11,370 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123124.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:48:16,590 INFO [finetune.py:976] (3/7) Epoch 22, batch 2850, loss[loss=0.2011, simple_loss=0.2553, pruned_loss=0.07351, over 4491.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2429, pruned_loss=0.0515, over 956395.40 frames. ], batch size: 20, lr: 3.14e-03, grad_scale: 32.0 2023-03-27 02:48:33,488 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.071e+02 1.606e+02 1.919e+02 2.375e+02 6.875e+02, threshold=3.839e+02, percent-clipped=7.0 2023-03-27 02:48:42,228 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=123172.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:48:49,677 INFO [finetune.py:976] (3/7) Epoch 22, batch 2900, loss[loss=0.1683, simple_loss=0.2346, pruned_loss=0.05105, over 4837.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2461, pruned_loss=0.05313, over 955402.04 frames. ], batch size: 25, lr: 3.14e-03, grad_scale: 32.0 2023-03-27 02:49:10,010 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0 2023-03-27 02:49:22,635 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123228.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:49:25,029 INFO [finetune.py:976] (3/7) Epoch 22, batch 2950, loss[loss=0.2035, simple_loss=0.2766, pruned_loss=0.0652, over 4824.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2484, pruned_loss=0.05312, over 955829.78 frames. ], batch size: 40, lr: 3.14e-03, grad_scale: 32.0 2023-03-27 02:49:32,912 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9009, 1.9995, 1.7805, 2.2187, 2.4376, 2.1851, 1.7830, 1.5578], device='cuda:3'), covar=tensor([0.2319, 0.1991, 0.1912, 0.1592, 0.1721, 0.1184, 0.2391, 0.2153], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0211, 0.0214, 0.0196, 0.0243, 0.0190, 0.0217, 0.0204], 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') 2023-03-27 02:49:42,387 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.515e+01 1.556e+02 1.822e+02 2.269e+02 3.192e+02, threshold=3.643e+02, percent-clipped=0.0 2023-03-27 02:49:59,813 INFO [finetune.py:976] (3/7) Epoch 22, batch 3000, loss[loss=0.2303, simple_loss=0.2886, pruned_loss=0.08604, over 4871.00 frames. ], tot_loss[loss=0.1787, simple_loss=0.2501, pruned_loss=0.05368, over 955597.62 frames. ], batch size: 31, lr: 3.14e-03, grad_scale: 32.0 2023-03-27 02:49:59,814 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-27 02:50:15,179 INFO [finetune.py:1010] (3/7) Epoch 22, validation: loss=0.1575, simple_loss=0.2256, pruned_loss=0.04471, over 2265189.00 frames. 2023-03-27 02:50:15,179 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6434MB 2023-03-27 02:50:47,931 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123314.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:51:07,175 INFO [finetune.py:976] (3/7) Epoch 22, batch 3050, loss[loss=0.1733, simple_loss=0.2502, pruned_loss=0.04822, over 4860.00 frames. ], tot_loss[loss=0.1797, simple_loss=0.252, pruned_loss=0.05372, over 957228.15 frames. ], batch size: 31, lr: 3.14e-03, grad_scale: 32.0 2023-03-27 02:51:15,251 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8244, 1.6900, 2.4172, 3.6800, 2.5570, 2.4745, 1.3306, 3.0766], device='cuda:3'), covar=tensor([0.1726, 0.1336, 0.1240, 0.0577, 0.0763, 0.1423, 0.1660, 0.0445], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0115, 0.0134, 0.0164, 0.0100, 0.0137, 0.0124, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 02:51:27,199 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.922e+01 1.647e+02 1.960e+02 2.377e+02 4.726e+02, threshold=3.920e+02, percent-clipped=6.0 2023-03-27 02:51:36,563 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123375.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:51:41,107 INFO [finetune.py:976] (3/7) Epoch 22, batch 3100, loss[loss=0.1762, simple_loss=0.2454, pruned_loss=0.05347, over 4801.00 frames. ], tot_loss[loss=0.1788, simple_loss=0.2503, pruned_loss=0.05367, over 957798.38 frames. ], batch size: 41, lr: 3.14e-03, grad_scale: 32.0 2023-03-27 02:52:08,645 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7572, 1.6791, 1.5037, 1.8364, 2.1956, 1.8180, 1.4722, 1.5024], device='cuda:3'), covar=tensor([0.1949, 0.1752, 0.1705, 0.1434, 0.1331, 0.1142, 0.2129, 0.1669], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0210, 0.0213, 0.0196, 0.0242, 0.0189, 0.0216, 0.0203], 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') 2023-03-27 02:52:12,219 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123428.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:52:14,576 INFO [finetune.py:976] (3/7) Epoch 22, batch 3150, loss[loss=0.1729, simple_loss=0.2422, pruned_loss=0.05175, over 4906.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2469, pruned_loss=0.0528, over 957449.53 frames. ], batch size: 35, lr: 3.14e-03, grad_scale: 32.0 2023-03-27 02:52:34,405 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.134e+02 1.485e+02 1.808e+02 2.093e+02 3.688e+02, threshold=3.617e+02, percent-clipped=0.0 2023-03-27 02:52:47,863 INFO [finetune.py:976] (3/7) Epoch 22, batch 3200, loss[loss=0.1594, simple_loss=0.2323, pruned_loss=0.04329, over 4900.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2429, pruned_loss=0.05119, over 957075.53 frames. ], batch size: 35, lr: 3.14e-03, grad_scale: 32.0 2023-03-27 02:52:52,380 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123489.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:52:54,662 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6516, 3.3846, 3.2752, 1.4534, 3.5758, 2.6180, 0.8188, 2.4328], device='cuda:3'), covar=tensor([0.2265, 0.2164, 0.1569, 0.3308, 0.1126, 0.1066, 0.4002, 0.1404], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0176, 0.0157, 0.0129, 0.0159, 0.0121, 0.0146, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 02:53:18,963 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123528.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:53:21,318 INFO [finetune.py:976] (3/7) Epoch 22, batch 3250, loss[loss=0.1597, simple_loss=0.2411, pruned_loss=0.03917, over 4866.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2438, pruned_loss=0.05216, over 956907.62 frames. ], batch size: 34, lr: 3.14e-03, grad_scale: 32.0 2023-03-27 02:53:41,211 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.285e+01 1.537e+02 1.740e+02 2.121e+02 3.629e+02, threshold=3.481e+02, percent-clipped=1.0 2023-03-27 02:53:51,147 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=123576.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:53:53,635 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.0880, 2.3033, 2.4357, 1.0694, 2.6563, 2.8723, 2.5338, 2.1470], device='cuda:3'), covar=tensor([0.0975, 0.0743, 0.0420, 0.0728, 0.0495, 0.0575, 0.0344, 0.0823], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0149, 0.0127, 0.0122, 0.0132, 0.0130, 0.0141, 0.0148], device='cuda:3'), out_proj_covar=tensor([8.9467e-05, 1.0772e-04, 9.0758e-05, 8.6086e-05, 9.2548e-05, 9.2523e-05, 1.0072e-04, 1.0617e-04], device='cuda:3') 2023-03-27 02:53:54,732 INFO [finetune.py:976] (3/7) Epoch 22, batch 3300, loss[loss=0.1547, simple_loss=0.2471, pruned_loss=0.0312, over 4898.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2485, pruned_loss=0.05387, over 957088.36 frames. ], batch size: 35, lr: 3.14e-03, grad_scale: 32.0 2023-03-27 02:54:16,680 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8399, 3.7656, 3.5540, 1.5920, 3.9306, 3.0487, 1.1548, 2.6509], device='cuda:3'), covar=tensor([0.2369, 0.1595, 0.1405, 0.3632, 0.0893, 0.0927, 0.3931, 0.1406], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0175, 0.0156, 0.0128, 0.0159, 0.0121, 0.0146, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 02:54:28,302 INFO [finetune.py:976] (3/7) Epoch 22, batch 3350, loss[loss=0.2004, simple_loss=0.2501, pruned_loss=0.07537, over 4132.00 frames. ], tot_loss[loss=0.178, simple_loss=0.249, pruned_loss=0.05351, over 952896.38 frames. ], batch size: 65, lr: 3.14e-03, grad_scale: 32.0 2023-03-27 02:54:47,671 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.106e+02 1.542e+02 1.809e+02 2.055e+02 5.285e+02, threshold=3.617e+02, percent-clipped=2.0 2023-03-27 02:54:54,269 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123670.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:55:01,512 INFO [finetune.py:976] (3/7) Epoch 22, batch 3400, loss[loss=0.1874, simple_loss=0.2633, pruned_loss=0.05574, over 4809.00 frames. ], tot_loss[loss=0.1792, simple_loss=0.2502, pruned_loss=0.05409, over 954426.54 frames. ], batch size: 41, lr: 3.14e-03, grad_scale: 32.0 2023-03-27 02:55:54,859 INFO [finetune.py:976] (3/7) Epoch 22, batch 3450, loss[loss=0.1855, simple_loss=0.2554, pruned_loss=0.05781, over 4923.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2495, pruned_loss=0.05321, over 954951.73 frames. ], batch size: 33, lr: 3.14e-03, grad_scale: 32.0 2023-03-27 02:56:26,818 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.145e+02 1.545e+02 1.898e+02 2.347e+02 3.548e+02, threshold=3.797e+02, percent-clipped=0.0 2023-03-27 02:56:44,493 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-03-27 02:56:45,230 INFO [finetune.py:976] (3/7) Epoch 22, batch 3500, loss[loss=0.1471, simple_loss=0.2125, pruned_loss=0.04083, over 4831.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2462, pruned_loss=0.05216, over 954751.19 frames. ], batch size: 38, lr: 3.14e-03, grad_scale: 32.0 2023-03-27 02:56:46,496 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123784.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:57:18,476 INFO [finetune.py:976] (3/7) Epoch 22, batch 3550, loss[loss=0.1653, simple_loss=0.2401, pruned_loss=0.04526, over 4857.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.245, pruned_loss=0.05222, over 955619.81 frames. ], batch size: 47, lr: 3.14e-03, grad_scale: 32.0 2023-03-27 02:57:36,086 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.884e+01 1.429e+02 1.766e+02 2.143e+02 3.754e+02, threshold=3.531e+02, percent-clipped=0.0 2023-03-27 02:57:45,443 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123872.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:57:52,353 INFO [finetune.py:976] (3/7) Epoch 22, batch 3600, loss[loss=0.1904, simple_loss=0.2581, pruned_loss=0.06132, over 4855.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2414, pruned_loss=0.05067, over 955329.21 frames. ], batch size: 44, lr: 3.14e-03, grad_scale: 32.0 2023-03-27 02:57:53,397 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-27 02:58:17,004 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6185, 1.5103, 1.0820, 0.3493, 1.3197, 1.4205, 1.4197, 1.5011], device='cuda:3'), covar=tensor([0.0821, 0.0774, 0.1139, 0.1747, 0.1236, 0.2276, 0.2273, 0.0720], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0191, 0.0198, 0.0182, 0.0209, 0.0206, 0.0222, 0.0195], 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') 2023-03-27 02:58:25,719 INFO [finetune.py:976] (3/7) Epoch 22, batch 3650, loss[loss=0.1934, simple_loss=0.2742, pruned_loss=0.05627, over 4807.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2439, pruned_loss=0.05164, over 953822.54 frames. ], batch size: 45, lr: 3.14e-03, grad_scale: 32.0 2023-03-27 02:58:26,480 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123933.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:58:30,725 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-03-27 02:58:43,172 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.106e+01 1.598e+02 1.925e+02 2.525e+02 5.508e+02, threshold=3.851e+02, percent-clipped=5.0 2023-03-27 02:58:49,854 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123970.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:58:54,488 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.59 vs. limit=2.0 2023-03-27 02:58:59,577 INFO [finetune.py:976] (3/7) Epoch 22, batch 3700, loss[loss=0.2131, simple_loss=0.2842, pruned_loss=0.07097, over 4765.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2471, pruned_loss=0.05245, over 953666.35 frames. ], batch size: 59, lr: 3.14e-03, grad_scale: 32.0 2023-03-27 02:59:06,393 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7036, 1.4959, 2.1187, 3.5588, 2.4249, 2.4112, 1.0310, 2.8251], device='cuda:3'), covar=tensor([0.1823, 0.1562, 0.1447, 0.0564, 0.0823, 0.1600, 0.1922, 0.0524], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0133, 0.0164, 0.0100, 0.0136, 0.0124, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 02:59:23,314 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=124018.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:59:34,574 INFO [finetune.py:976] (3/7) Epoch 22, batch 3750, loss[loss=0.2166, simple_loss=0.2785, pruned_loss=0.07732, over 4732.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2483, pruned_loss=0.05328, over 951460.15 frames. ], batch size: 54, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 02:59:37,123 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124036.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 02:59:51,701 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.170e+02 1.541e+02 1.786e+02 2.095e+02 2.976e+02, threshold=3.572e+02, percent-clipped=0.0 2023-03-27 02:59:53,035 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124062.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:00:01,829 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4812, 1.3178, 1.5425, 2.4612, 1.7000, 2.1883, 0.7475, 2.0569], device='cuda:3'), covar=tensor([0.1534, 0.1353, 0.1138, 0.0689, 0.0843, 0.1169, 0.1567, 0.0610], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0116, 0.0134, 0.0165, 0.0101, 0.0137, 0.0126, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 03:00:06,947 INFO [finetune.py:976] (3/7) Epoch 22, batch 3800, loss[loss=0.1823, simple_loss=0.2398, pruned_loss=0.06247, over 4723.00 frames. ], tot_loss[loss=0.179, simple_loss=0.2499, pruned_loss=0.05399, over 950079.70 frames. ], batch size: 23, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:00:08,694 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124084.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:00:17,185 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124097.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:00:28,852 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.85 vs. limit=5.0 2023-03-27 03:00:39,633 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124123.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:00:51,383 INFO [finetune.py:976] (3/7) Epoch 22, batch 3850, loss[loss=0.1848, simple_loss=0.2473, pruned_loss=0.06108, over 4806.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2486, pruned_loss=0.0531, over 951525.21 frames. ], batch size: 41, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:00:51,452 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=124132.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:01:20,790 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.085e+02 1.486e+02 1.746e+02 2.060e+02 3.550e+02, threshold=3.491e+02, percent-clipped=0.0 2023-03-27 03:01:46,863 INFO [finetune.py:976] (3/7) Epoch 22, batch 3900, loss[loss=0.1572, simple_loss=0.2208, pruned_loss=0.04676, over 4680.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2457, pruned_loss=0.05256, over 952906.98 frames. ], batch size: 23, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:01:49,299 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4769, 1.3583, 1.9296, 1.6681, 1.6081, 3.4237, 1.3135, 1.4893], device='cuda:3'), covar=tensor([0.0987, 0.1882, 0.1138, 0.1039, 0.1621, 0.0252, 0.1546, 0.1848], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0074, 0.0077, 0.0091, 0.0081, 0.0086, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 03:02:11,448 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.0884, 1.6847, 1.8481, 0.8511, 2.1480, 2.4724, 1.9510, 1.8590], device='cuda:3'), covar=tensor([0.1004, 0.1340, 0.0623, 0.0785, 0.0605, 0.0684, 0.0540, 0.0797], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0150, 0.0128, 0.0123, 0.0132, 0.0131, 0.0142, 0.0149], device='cuda:3'), out_proj_covar=tensor([9.0332e-05, 1.0868e-04, 9.1738e-05, 8.6948e-05, 9.2872e-05, 9.3453e-05, 1.0186e-04, 1.0699e-04], device='cuda:3') 2023-03-27 03:02:17,914 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124228.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:02:20,289 INFO [finetune.py:976] (3/7) Epoch 22, batch 3950, loss[loss=0.1828, simple_loss=0.2462, pruned_loss=0.05974, over 4728.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.244, pruned_loss=0.05191, over 953995.30 frames. ], batch size: 54, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:02:22,620 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.3242, 2.9349, 3.0684, 3.2699, 3.1298, 2.8957, 3.3669, 0.9511], device='cuda:3'), covar=tensor([0.1052, 0.1015, 0.1134, 0.1042, 0.1603, 0.1719, 0.1016, 0.5751], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0244, 0.0279, 0.0290, 0.0333, 0.0284, 0.0305, 0.0299], 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') 2023-03-27 03:02:39,960 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.148e+02 1.559e+02 1.897e+02 2.284e+02 3.853e+02, threshold=3.794e+02, percent-clipped=3.0 2023-03-27 03:02:40,694 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5320, 1.3526, 1.5511, 0.8746, 1.5441, 1.5420, 1.5152, 1.3365], device='cuda:3'), covar=tensor([0.0538, 0.0780, 0.0622, 0.0872, 0.0803, 0.0660, 0.0617, 0.1234], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0134, 0.0138, 0.0119, 0.0124, 0.0136, 0.0137, 0.0159], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 03:02:47,336 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124272.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:02:53,670 INFO [finetune.py:976] (3/7) Epoch 22, batch 4000, loss[loss=0.1686, simple_loss=0.233, pruned_loss=0.05217, over 4673.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2416, pruned_loss=0.05113, over 954199.65 frames. ], batch size: 23, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:02:58,634 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 2023-03-27 03:03:26,976 INFO [finetune.py:976] (3/7) Epoch 22, batch 4050, loss[loss=0.2199, simple_loss=0.2937, pruned_loss=0.07304, over 4804.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.245, pruned_loss=0.05191, over 955890.81 frames. ], batch size: 45, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:03:27,712 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124333.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:03:39,246 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124348.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 03:03:43,425 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.6455, 1.6999, 1.7574, 0.9286, 1.8926, 2.0368, 2.0459, 1.6046], device='cuda:3'), covar=tensor([0.1000, 0.0784, 0.0587, 0.0619, 0.0462, 0.0603, 0.0357, 0.0731], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0150, 0.0127, 0.0123, 0.0132, 0.0130, 0.0142, 0.0149], device='cuda:3'), out_proj_covar=tensor([8.9733e-05, 1.0831e-04, 9.1030e-05, 8.6453e-05, 9.2531e-05, 9.2928e-05, 1.0124e-04, 1.0663e-04], device='cuda:3') 2023-03-27 03:03:46,865 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.053e+02 1.585e+02 1.909e+02 2.213e+02 3.315e+02, threshold=3.818e+02, percent-clipped=0.0 2023-03-27 03:04:00,187 INFO [finetune.py:976] (3/7) Epoch 22, batch 4100, loss[loss=0.1531, simple_loss=0.2287, pruned_loss=0.03881, over 4825.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2475, pruned_loss=0.05258, over 954534.45 frames. ], batch size: 47, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:04:07,302 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124392.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:04:19,527 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124409.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 03:04:22,743 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.93 vs. limit=5.0 2023-03-27 03:04:24,895 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124418.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:04:33,281 INFO [finetune.py:976] (3/7) Epoch 22, batch 4150, loss[loss=0.1678, simple_loss=0.2458, pruned_loss=0.04489, over 4924.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.247, pruned_loss=0.05174, over 955302.80 frames. ], batch size: 42, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:04:33,382 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124432.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:04:44,123 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.52 vs. limit=5.0 2023-03-27 03:04:53,595 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.819e+01 1.555e+02 1.746e+02 2.178e+02 5.076e+02, threshold=3.492e+02, percent-clipped=2.0 2023-03-27 03:05:06,935 INFO [finetune.py:976] (3/7) Epoch 22, batch 4200, loss[loss=0.1964, simple_loss=0.2799, pruned_loss=0.05651, over 4889.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2481, pruned_loss=0.05177, over 956094.75 frames. ], batch size: 36, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:05:12,187 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.68 vs. limit=5.0 2023-03-27 03:05:14,708 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124493.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:05:14,751 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8305, 1.3262, 1.9171, 1.8409, 1.6641, 1.6165, 1.8002, 1.7617], device='cuda:3'), covar=tensor([0.4015, 0.4027, 0.3246, 0.3757, 0.4648, 0.3785, 0.4412, 0.3231], device='cuda:3'), in_proj_covar=tensor([0.0260, 0.0245, 0.0265, 0.0286, 0.0285, 0.0261, 0.0295, 0.0248], 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') 2023-03-27 03:05:28,359 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3460, 2.9704, 2.7212, 1.3276, 3.0839, 2.2866, 0.7702, 1.9114], device='cuda:3'), covar=tensor([0.2214, 0.2113, 0.1801, 0.3665, 0.1219, 0.1154, 0.4167, 0.1791], device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0181, 0.0162, 0.0131, 0.0163, 0.0125, 0.0151, 0.0126], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 03:05:40,295 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124528.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:05:42,631 INFO [finetune.py:976] (3/7) Epoch 22, batch 4250, loss[loss=0.1984, simple_loss=0.2636, pruned_loss=0.06659, over 4708.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2466, pruned_loss=0.05171, over 956335.95 frames. ], batch size: 54, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:06:02,064 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.200e+01 1.490e+02 1.704e+02 2.072e+02 3.423e+02, threshold=3.408e+02, percent-clipped=0.0 2023-03-27 03:06:12,353 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=124576.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:06:17,859 INFO [finetune.py:976] (3/7) Epoch 22, batch 4300, loss[loss=0.1952, simple_loss=0.2523, pruned_loss=0.06901, over 4899.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.244, pruned_loss=0.05136, over 955948.04 frames. ], batch size: 32, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:06:41,348 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.3758, 4.6674, 4.9335, 5.1395, 5.1026, 4.7531, 5.5497, 1.5310], device='cuda:3'), covar=tensor([0.0663, 0.0828, 0.0796, 0.0891, 0.1093, 0.1639, 0.0424, 0.6013], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0245, 0.0280, 0.0291, 0.0335, 0.0284, 0.0305, 0.0301], 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') 2023-03-27 03:07:10,854 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124628.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:07:17,351 INFO [finetune.py:976] (3/7) Epoch 22, batch 4350, loss[loss=0.2032, simple_loss=0.2698, pruned_loss=0.06827, over 4933.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2412, pruned_loss=0.05034, over 957820.90 frames. ], batch size: 38, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:07:21,126 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124638.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:07:48,792 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.554e+01 1.414e+02 1.739e+02 2.148e+02 3.782e+02, threshold=3.479e+02, percent-clipped=2.0 2023-03-27 03:07:50,754 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124663.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:08:01,991 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-27 03:08:06,298 INFO [finetune.py:976] (3/7) Epoch 22, batch 4400, loss[loss=0.1594, simple_loss=0.236, pruned_loss=0.0414, over 4805.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.244, pruned_loss=0.05143, over 959694.10 frames. ], batch size: 41, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:08:11,265 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.3298, 3.7893, 3.9515, 4.1346, 4.0867, 3.8787, 4.4583, 1.4696], device='cuda:3'), covar=tensor([0.0862, 0.0827, 0.0843, 0.1181, 0.1341, 0.1637, 0.0665, 0.5672], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0244, 0.0279, 0.0290, 0.0334, 0.0283, 0.0304, 0.0299], 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') 2023-03-27 03:08:12,515 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124692.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:08:17,309 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124699.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:08:20,764 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124704.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 03:08:31,138 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124718.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:08:31,854 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-27 03:08:35,324 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124724.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:08:40,102 INFO [finetune.py:976] (3/7) Epoch 22, batch 4450, loss[loss=0.2072, simple_loss=0.2879, pruned_loss=0.06331, over 4914.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2473, pruned_loss=0.05198, over 957128.61 frames. ], batch size: 37, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:08:45,009 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=124740.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:08:55,872 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-03-27 03:08:58,178 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.229e+02 1.653e+02 1.910e+02 2.206e+02 5.425e+02, threshold=3.820e+02, percent-clipped=3.0 2023-03-27 03:09:02,350 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=124766.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:09:13,853 INFO [finetune.py:976] (3/7) Epoch 22, batch 4500, loss[loss=0.187, simple_loss=0.2742, pruned_loss=0.04987, over 4858.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2485, pruned_loss=0.05229, over 955698.79 frames. ], batch size: 31, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:09:17,602 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124788.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:09:21,284 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9547, 1.8451, 1.9937, 1.3177, 1.9808, 2.0576, 2.0673, 1.6241], device='cuda:3'), covar=tensor([0.0525, 0.0636, 0.0669, 0.0883, 0.0743, 0.0617, 0.0551, 0.1122], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0135, 0.0139, 0.0120, 0.0126, 0.0138, 0.0138, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 03:09:38,645 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124820.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:09:47,292 INFO [finetune.py:976] (3/7) Epoch 22, batch 4550, loss[loss=0.2289, simple_loss=0.2965, pruned_loss=0.08062, over 4915.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2501, pruned_loss=0.05297, over 956934.94 frames. ], batch size: 38, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:09:50,485 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0953, 2.0339, 1.6895, 1.8773, 1.9774, 1.9389, 1.9378, 2.6244], device='cuda:3'), covar=tensor([0.3698, 0.3572, 0.3234, 0.3473, 0.3768, 0.2426, 0.3401, 0.1731], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0261, 0.0233, 0.0276, 0.0254, 0.0224, 0.0252, 0.0234], 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') 2023-03-27 03:10:04,841 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.024e+02 1.469e+02 1.742e+02 1.998e+02 4.285e+02, threshold=3.484e+02, percent-clipped=1.0 2023-03-27 03:10:11,338 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1583, 2.0092, 2.1505, 1.5285, 2.0370, 2.1729, 2.2107, 1.7553], device='cuda:3'), covar=tensor([0.0561, 0.0667, 0.0646, 0.0857, 0.0713, 0.0657, 0.0581, 0.1103], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0135, 0.0139, 0.0120, 0.0125, 0.0138, 0.0138, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 03:10:17,474 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-27 03:10:19,605 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124881.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:10:20,090 INFO [finetune.py:976] (3/7) Epoch 22, batch 4600, loss[loss=0.1533, simple_loss=0.2236, pruned_loss=0.04152, over 4781.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.249, pruned_loss=0.05219, over 956477.78 frames. ], batch size: 29, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:10:29,777 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1138, 1.8488, 1.8669, 0.8842, 2.0948, 2.2862, 2.0115, 1.7990], device='cuda:3'), covar=tensor([0.0927, 0.0774, 0.0555, 0.0694, 0.0578, 0.0841, 0.0593, 0.0735], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0151, 0.0128, 0.0123, 0.0132, 0.0131, 0.0142, 0.0150], device='cuda:3'), out_proj_covar=tensor([8.9979e-05, 1.0889e-04, 9.1383e-05, 8.6720e-05, 9.2515e-05, 9.3099e-05, 1.0195e-04, 1.0729e-04], device='cuda:3') 2023-03-27 03:10:47,662 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5567, 2.3941, 2.0402, 2.5662, 2.4388, 2.1737, 2.7795, 2.5290], device='cuda:3'), covar=tensor([0.1228, 0.2025, 0.2720, 0.2321, 0.2487, 0.1687, 0.3046, 0.1708], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0188, 0.0234, 0.0253, 0.0247, 0.0204, 0.0214, 0.0201], 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') 2023-03-27 03:10:48,206 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6824, 1.1874, 0.8868, 1.5670, 2.1906, 1.0441, 1.4252, 1.6110], device='cuda:3'), covar=tensor([0.1371, 0.2093, 0.1871, 0.1109, 0.1757, 0.1918, 0.1429, 0.1882], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0095, 0.0110, 0.0092, 0.0120, 0.0094, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 03:10:50,952 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124928.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:10:53,306 INFO [finetune.py:976] (3/7) Epoch 22, batch 4650, loss[loss=0.1431, simple_loss=0.2144, pruned_loss=0.03588, over 4903.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2455, pruned_loss=0.05093, over 955921.58 frames. ], batch size: 35, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:11:10,703 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.046e+02 1.467e+02 1.675e+02 2.008e+02 3.769e+02, threshold=3.350e+02, percent-clipped=1.0 2023-03-27 03:11:21,849 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=124976.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:11:26,384 INFO [finetune.py:976] (3/7) Epoch 22, batch 4700, loss[loss=0.1903, simple_loss=0.2576, pruned_loss=0.06147, over 4831.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2427, pruned_loss=0.05025, over 956984.43 frames. ], batch size: 39, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:11:36,917 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124994.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:11:43,165 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125004.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 03:11:52,758 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125019.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:12:07,771 INFO [finetune.py:976] (3/7) Epoch 22, batch 4750, loss[loss=0.1667, simple_loss=0.2341, pruned_loss=0.04964, over 4849.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2405, pruned_loss=0.04958, over 957111.08 frames. ], batch size: 44, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:12:11,848 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4151, 1.9737, 2.5104, 2.3123, 2.1283, 2.0968, 2.3692, 2.3583], device='cuda:3'), covar=tensor([0.4019, 0.4235, 0.3302, 0.3871, 0.4941, 0.4209, 0.4812, 0.3036], device='cuda:3'), in_proj_covar=tensor([0.0258, 0.0243, 0.0263, 0.0284, 0.0283, 0.0260, 0.0293, 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') 2023-03-27 03:12:30,809 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=125052.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 03:12:39,937 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.109e+02 1.467e+02 1.710e+02 2.084e+02 3.277e+02, threshold=3.420e+02, percent-clipped=0.0 2023-03-27 03:13:08,206 INFO [finetune.py:976] (3/7) Epoch 22, batch 4800, loss[loss=0.1792, simple_loss=0.2339, pruned_loss=0.06229, over 4108.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.244, pruned_loss=0.05096, over 956024.21 frames. ], batch size: 17, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:13:10,129 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5222, 1.4334, 1.4565, 1.5286, 1.5456, 3.6551, 1.3849, 1.8429], device='cuda:3'), covar=tensor([0.3550, 0.2678, 0.2337, 0.2508, 0.1491, 0.0166, 0.2633, 0.1293], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0115, 0.0121, 0.0123, 0.0113, 0.0095, 0.0094, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 03:13:16,547 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125088.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:13:42,364 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125128.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:13:45,185 INFO [finetune.py:976] (3/7) Epoch 22, batch 4850, loss[loss=0.1641, simple_loss=0.2501, pruned_loss=0.0391, over 4815.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2458, pruned_loss=0.05068, over 956550.67 frames. ], batch size: 51, lr: 3.13e-03, grad_scale: 64.0 2023-03-27 03:13:47,687 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=125136.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:13:50,684 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8254, 1.3086, 1.8919, 1.7605, 1.6114, 1.5714, 1.7977, 1.8186], device='cuda:3'), covar=tensor([0.3996, 0.3802, 0.2998, 0.3615, 0.4392, 0.3613, 0.4105, 0.2819], device='cuda:3'), in_proj_covar=tensor([0.0258, 0.0243, 0.0263, 0.0284, 0.0283, 0.0260, 0.0293, 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') 2023-03-27 03:14:04,216 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.151e+02 1.665e+02 1.915e+02 2.224e+02 3.844e+02, threshold=3.831e+02, percent-clipped=1.0 2023-03-27 03:14:12,813 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3424, 1.2336, 1.6246, 2.4085, 1.5715, 2.1980, 1.0747, 2.0810], device='cuda:3'), covar=tensor([0.1663, 0.1398, 0.1087, 0.0727, 0.0960, 0.1354, 0.1348, 0.0563], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0133, 0.0164, 0.0101, 0.0136, 0.0124, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 03:14:14,554 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125176.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:14:18,599 INFO [finetune.py:976] (3/7) Epoch 22, batch 4900, loss[loss=0.1338, simple_loss=0.217, pruned_loss=0.02524, over 4830.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2481, pruned_loss=0.05171, over 957728.85 frames. ], batch size: 30, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:14:23,443 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125189.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:14:39,898 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-27 03:14:52,065 INFO [finetune.py:976] (3/7) Epoch 22, batch 4950, loss[loss=0.1899, simple_loss=0.2611, pruned_loss=0.05931, over 4762.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2489, pruned_loss=0.05188, over 957760.39 frames. ], batch size: 28, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:15:12,022 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.098e+02 1.500e+02 1.887e+02 2.169e+02 5.445e+02, threshold=3.774e+02, percent-clipped=5.0 2023-03-27 03:15:25,140 INFO [finetune.py:976] (3/7) Epoch 22, batch 5000, loss[loss=0.1228, simple_loss=0.1974, pruned_loss=0.0241, over 4740.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2473, pruned_loss=0.05148, over 958127.66 frames. ], batch size: 27, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:15:33,514 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125294.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:15:33,566 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.0327, 1.0193, 0.9923, 0.3679, 0.9525, 1.1912, 1.2051, 0.9293], device='cuda:3'), covar=tensor([0.0893, 0.0655, 0.0633, 0.0575, 0.0580, 0.0668, 0.0460, 0.0700], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0150, 0.0127, 0.0123, 0.0131, 0.0130, 0.0142, 0.0149], device='cuda:3'), out_proj_covar=tensor([8.9511e-05, 1.0858e-04, 9.1033e-05, 8.6614e-05, 9.1988e-05, 9.2691e-05, 1.0159e-04, 1.0708e-04], device='cuda:3') 2023-03-27 03:15:50,215 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125319.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:15:51,465 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6474, 1.4857, 1.4757, 1.5245, 1.1335, 3.4864, 1.4005, 1.7465], device='cuda:3'), covar=tensor([0.3315, 0.2515, 0.2284, 0.2441, 0.1818, 0.0189, 0.2527, 0.1300], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0116, 0.0121, 0.0123, 0.0113, 0.0096, 0.0094, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 03:15:58,104 INFO [finetune.py:976] (3/7) Epoch 22, batch 5050, loss[loss=0.1825, simple_loss=0.2548, pruned_loss=0.05507, over 4832.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2435, pruned_loss=0.05043, over 956282.17 frames. ], batch size: 30, lr: 3.13e-03, grad_scale: 32.0 2023-03-27 03:16:03,205 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-27 03:16:05,256 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=125342.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:16:15,324 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125356.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:16:18,273 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.708e+01 1.493e+02 1.877e+02 2.229e+02 3.838e+02, threshold=3.755e+02, percent-clipped=1.0 2023-03-27 03:16:22,415 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=125367.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:16:27,962 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125376.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:16:31,270 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.14 vs. limit=5.0 2023-03-27 03:16:31,544 INFO [finetune.py:976] (3/7) Epoch 22, batch 5100, loss[loss=0.1475, simple_loss=0.2077, pruned_loss=0.04365, over 4730.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2413, pruned_loss=0.04967, over 957257.88 frames. ], batch size: 23, lr: 3.12e-03, grad_scale: 32.0 2023-03-27 03:16:32,373 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-27 03:16:55,563 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125417.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 03:16:58,444 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125421.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:17:01,492 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8760, 1.6652, 1.5888, 1.9563, 2.2796, 1.9584, 1.5213, 1.5412], device='cuda:3'), covar=tensor([0.2230, 0.2080, 0.1970, 0.1614, 0.1604, 0.1146, 0.2375, 0.1951], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0210, 0.0212, 0.0196, 0.0242, 0.0188, 0.0215, 0.0203], 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') 2023-03-27 03:17:05,018 INFO [finetune.py:976] (3/7) Epoch 22, batch 5150, loss[loss=0.216, simple_loss=0.2784, pruned_loss=0.07682, over 4932.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2414, pruned_loss=0.0503, over 954987.35 frames. ], batch size: 38, lr: 3.12e-03, grad_scale: 32.0 2023-03-27 03:17:13,017 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125437.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:17:33,161 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.242e+02 1.608e+02 1.768e+02 2.290e+02 4.207e+02, threshold=3.536e+02, percent-clipped=1.0 2023-03-27 03:17:45,311 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125476.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:17:48,791 INFO [finetune.py:976] (3/7) Epoch 22, batch 5200, loss[loss=0.194, simple_loss=0.2746, pruned_loss=0.05664, over 4901.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2451, pruned_loss=0.0515, over 954349.91 frames. ], batch size: 37, lr: 3.12e-03, grad_scale: 32.0 2023-03-27 03:17:48,917 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125482.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:17:54,886 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125484.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:18:14,790 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.73 vs. limit=5.0 2023-03-27 03:18:39,395 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=125524.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:18:44,719 INFO [finetune.py:976] (3/7) Epoch 22, batch 5250, loss[loss=0.1887, simple_loss=0.2671, pruned_loss=0.05517, over 4821.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2489, pruned_loss=0.05253, over 954282.01 frames. ], batch size: 39, lr: 3.12e-03, grad_scale: 32.0 2023-03-27 03:19:03,922 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.075e+02 1.572e+02 1.864e+02 2.118e+02 3.675e+02, threshold=3.728e+02, percent-clipped=1.0 2023-03-27 03:19:05,131 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0590, 1.6343, 2.3390, 1.5880, 2.0663, 2.2501, 1.5646, 2.3756], device='cuda:3'), covar=tensor([0.1179, 0.2090, 0.1393, 0.1995, 0.0813, 0.1261, 0.2791, 0.0754], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0206, 0.0191, 0.0190, 0.0174, 0.0214, 0.0216, 0.0200], 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') 2023-03-27 03:19:18,600 INFO [finetune.py:976] (3/7) Epoch 22, batch 5300, loss[loss=0.1575, simple_loss=0.2268, pruned_loss=0.04414, over 4762.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2495, pruned_loss=0.05291, over 955631.49 frames. ], batch size: 28, lr: 3.12e-03, grad_scale: 32.0 2023-03-27 03:19:31,428 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-27 03:19:32,533 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125603.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:19:52,427 INFO [finetune.py:976] (3/7) Epoch 22, batch 5350, loss[loss=0.1614, simple_loss=0.235, pruned_loss=0.04392, over 4904.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2502, pruned_loss=0.05294, over 957477.28 frames. ], batch size: 36, lr: 3.12e-03, grad_scale: 32.0 2023-03-27 03:19:53,862 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-03-27 03:20:10,862 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.759e+01 1.403e+02 1.743e+02 2.180e+02 4.274e+02, threshold=3.486e+02, percent-clipped=1.0 2023-03-27 03:20:13,344 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125664.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:20:25,088 INFO [finetune.py:976] (3/7) Epoch 22, batch 5400, loss[loss=0.1599, simple_loss=0.224, pruned_loss=0.04792, over 4907.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2461, pruned_loss=0.05188, over 955549.03 frames. ], batch size: 36, lr: 3.12e-03, grad_scale: 32.0 2023-03-27 03:20:35,925 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.5527, 1.7223, 1.7981, 1.0001, 1.8142, 1.9553, 1.9848, 1.5610], device='cuda:3'), covar=tensor([0.1012, 0.0742, 0.0537, 0.0604, 0.0512, 0.0792, 0.0480, 0.0770], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0150, 0.0127, 0.0123, 0.0131, 0.0131, 0.0141, 0.0149], device='cuda:3'), out_proj_covar=tensor([8.9985e-05, 1.0857e-04, 9.1165e-05, 8.6441e-05, 9.2034e-05, 9.3171e-05, 1.0120e-04, 1.0689e-04], device='cuda:3') 2023-03-27 03:20:44,662 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125712.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 03:20:53,958 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125725.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:20:58,061 INFO [finetune.py:976] (3/7) Epoch 22, batch 5450, loss[loss=0.1446, simple_loss=0.2122, pruned_loss=0.03847, over 4943.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2428, pruned_loss=0.05093, over 954597.90 frames. ], batch size: 38, lr: 3.12e-03, grad_scale: 32.0 2023-03-27 03:20:58,129 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125732.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:21:02,299 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8018, 1.7194, 1.6275, 1.7154, 2.1633, 2.1544, 1.7806, 1.6026], device='cuda:3'), covar=tensor([0.0288, 0.0298, 0.0515, 0.0294, 0.0176, 0.0416, 0.0319, 0.0369], device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0105, 0.0141, 0.0110, 0.0098, 0.0110, 0.0100, 0.0111], device='cuda:3'), out_proj_covar=tensor([7.5639e-05, 8.1036e-05, 1.1072e-04, 8.4741e-05, 7.6009e-05, 8.1127e-05, 7.4527e-05, 8.4261e-05], device='cuda:3') 2023-03-27 03:21:14,492 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.44 vs. limit=5.0 2023-03-27 03:21:17,002 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.003e+02 1.470e+02 1.717e+02 2.000e+02 3.602e+02, threshold=3.434e+02, percent-clipped=1.0 2023-03-27 03:21:27,701 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125777.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:21:28,836 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2619, 1.7675, 1.7502, 0.9271, 1.9028, 1.9708, 1.9890, 1.6768], device='cuda:3'), covar=tensor([0.0823, 0.0729, 0.0464, 0.0659, 0.0562, 0.0690, 0.0434, 0.0682], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0151, 0.0128, 0.0123, 0.0131, 0.0131, 0.0142, 0.0150], device='cuda:3'), out_proj_covar=tensor([9.0255e-05, 1.0891e-04, 9.1415e-05, 8.6637e-05, 9.2302e-05, 9.3558e-05, 1.0141e-04, 1.0724e-04], device='cuda:3') 2023-03-27 03:21:31,106 INFO [finetune.py:976] (3/7) Epoch 22, batch 5500, loss[loss=0.171, simple_loss=0.2373, pruned_loss=0.0523, over 4906.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2393, pruned_loss=0.04955, over 953101.04 frames. ], batch size: 32, lr: 3.12e-03, grad_scale: 32.0 2023-03-27 03:21:32,442 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125784.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:21:32,453 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3291, 1.3650, 1.5561, 1.0780, 1.2970, 1.4824, 1.3479, 1.6925], device='cuda:3'), covar=tensor([0.1198, 0.2251, 0.1338, 0.1599, 0.0946, 0.1283, 0.3055, 0.0841], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0207, 0.0192, 0.0190, 0.0174, 0.0216, 0.0217, 0.0200], 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') 2023-03-27 03:21:33,652 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125786.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:22:04,441 INFO [finetune.py:976] (3/7) Epoch 22, batch 5550, loss[loss=0.1911, simple_loss=0.2654, pruned_loss=0.05842, over 4857.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2416, pruned_loss=0.0506, over 953450.08 frames. ], batch size: 31, lr: 3.12e-03, grad_scale: 32.0 2023-03-27 03:22:04,492 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=125832.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:22:05,261 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-27 03:22:32,015 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.163e+02 1.619e+02 1.949e+02 2.379e+02 4.295e+02, threshold=3.899e+02, percent-clipped=6.0 2023-03-27 03:22:48,803 INFO [finetune.py:976] (3/7) Epoch 22, batch 5600, loss[loss=0.1657, simple_loss=0.2503, pruned_loss=0.04057, over 4783.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2453, pruned_loss=0.05168, over 953704.79 frames. ], batch size: 29, lr: 3.12e-03, grad_scale: 32.0 2023-03-27 03:22:51,826 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0320, 1.9634, 1.6613, 1.8685, 1.7900, 1.8762, 1.8105, 2.5660], device='cuda:3'), covar=tensor([0.3557, 0.4133, 0.3269, 0.3932, 0.4112, 0.2442, 0.3971, 0.1605], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0262, 0.0233, 0.0275, 0.0254, 0.0224, 0.0252, 0.0234], 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') 2023-03-27 03:23:14,964 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 2023-03-27 03:23:20,374 INFO [finetune.py:976] (3/7) Epoch 22, batch 5650, loss[loss=0.179, simple_loss=0.2623, pruned_loss=0.04785, over 4831.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2481, pruned_loss=0.05212, over 954240.20 frames. ], batch size: 47, lr: 3.12e-03, grad_scale: 32.0 2023-03-27 03:23:47,402 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125959.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:23:48,544 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.919e+01 1.591e+02 1.865e+02 2.221e+02 3.612e+02, threshold=3.730e+02, percent-clipped=0.0 2023-03-27 03:24:11,014 INFO [finetune.py:976] (3/7) Epoch 22, batch 5700, loss[loss=0.1372, simple_loss=0.2051, pruned_loss=0.03463, over 4207.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2436, pruned_loss=0.05099, over 936322.20 frames. ], batch size: 18, lr: 3.12e-03, grad_scale: 32.0 2023-03-27 03:24:18,360 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.90 vs. limit=2.0 2023-03-27 03:24:20,513 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125998.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:24:37,871 INFO [finetune.py:976] (3/7) Epoch 23, batch 0, loss[loss=0.2457, simple_loss=0.3012, pruned_loss=0.09511, over 4093.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3012, pruned_loss=0.09511, over 4093.00 frames. ], batch size: 66, lr: 3.12e-03, grad_scale: 32.0 2023-03-27 03:24:37,871 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-27 03:24:44,103 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6956, 1.6378, 2.0836, 2.9500, 1.9534, 2.2945, 1.0595, 2.4933], device='cuda:3'), covar=tensor([0.1503, 0.1188, 0.1022, 0.0555, 0.0860, 0.1139, 0.1627, 0.0469], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0133, 0.0163, 0.0100, 0.0135, 0.0123, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 03:24:45,130 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0212, 1.7972, 2.0400, 1.3155, 1.9698, 2.0350, 2.0716, 1.6315], device='cuda:3'), covar=tensor([0.0498, 0.0688, 0.0556, 0.0844, 0.0739, 0.0630, 0.0522, 0.1157], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0134, 0.0138, 0.0119, 0.0123, 0.0137, 0.0137, 0.0160], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 03:24:53,039 INFO [finetune.py:1010] (3/7) Epoch 23, validation: loss=0.1587, simple_loss=0.2268, pruned_loss=0.04533, over 2265189.00 frames. 2023-03-27 03:24:53,039 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6469MB 2023-03-27 03:24:59,385 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126012.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 03:25:12,069 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126032.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:25:30,342 INFO [finetune.py:976] (3/7) Epoch 23, batch 50, loss[loss=0.1652, simple_loss=0.2408, pruned_loss=0.04482, over 4825.00 frames. ], tot_loss[loss=0.1817, simple_loss=0.2531, pruned_loss=0.05516, over 216711.54 frames. ], batch size: 33, lr: 3.12e-03, grad_scale: 32.0 2023-03-27 03:25:30,933 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126059.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:25:31,916 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=126060.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:25:32,453 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.266e+01 1.545e+02 1.923e+02 2.325e+02 3.929e+02, threshold=3.846e+02, percent-clipped=1.0 2023-03-27 03:25:42,850 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126077.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:25:44,663 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=126080.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:25:45,301 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126081.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:26:03,748 INFO [finetune.py:976] (3/7) Epoch 23, batch 100, loss[loss=0.1777, simple_loss=0.2405, pruned_loss=0.05745, over 4822.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2457, pruned_loss=0.05253, over 382491.36 frames. ], batch size: 33, lr: 3.12e-03, grad_scale: 32.0 2023-03-27 03:26:11,477 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5536, 1.1004, 0.7580, 1.4904, 1.9993, 1.1279, 1.3613, 1.5462], device='cuda:3'), covar=tensor([0.1407, 0.2095, 0.1875, 0.1138, 0.2028, 0.1942, 0.1460, 0.1930], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0095, 0.0110, 0.0092, 0.0120, 0.0094, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 03:26:14,892 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=126125.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:26:19,894 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6713, 2.5622, 2.1307, 1.1743, 2.2583, 2.0886, 1.9915, 2.2432], device='cuda:3'), covar=tensor([0.0781, 0.0769, 0.1635, 0.1993, 0.1532, 0.2103, 0.1957, 0.0957], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0194, 0.0201, 0.0184, 0.0211, 0.0210, 0.0226, 0.0198], 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') 2023-03-27 03:26:37,115 INFO [finetune.py:976] (3/7) Epoch 23, batch 150, loss[loss=0.158, simple_loss=0.233, pruned_loss=0.04146, over 4905.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2401, pruned_loss=0.05162, over 508054.80 frames. ], batch size: 36, lr: 3.12e-03, grad_scale: 32.0 2023-03-27 03:26:38,297 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.035e+02 1.556e+02 1.791e+02 2.255e+02 5.687e+02, threshold=3.583e+02, percent-clipped=3.0 2023-03-27 03:27:10,672 INFO [finetune.py:976] (3/7) Epoch 23, batch 200, loss[loss=0.1825, simple_loss=0.251, pruned_loss=0.05697, over 4908.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2379, pruned_loss=0.05011, over 609193.80 frames. ], batch size: 35, lr: 3.12e-03, grad_scale: 32.0 2023-03-27 03:28:05,167 INFO [finetune.py:976] (3/7) Epoch 23, batch 250, loss[loss=0.23, simple_loss=0.3021, pruned_loss=0.07895, over 4806.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2439, pruned_loss=0.05256, over 686773.65 frames. ], batch size: 51, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:28:05,285 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126259.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:28:06,383 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.143e+02 1.625e+02 1.849e+02 2.180e+02 4.181e+02, threshold=3.697e+02, percent-clipped=1.0 2023-03-27 03:28:37,017 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=126307.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:28:40,580 INFO [finetune.py:976] (3/7) Epoch 23, batch 300, loss[loss=0.1777, simple_loss=0.2596, pruned_loss=0.04794, over 4740.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.248, pruned_loss=0.0532, over 745488.85 frames. ], batch size: 54, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:29:00,162 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.53 vs. limit=5.0 2023-03-27 03:29:20,739 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126354.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:29:24,666 INFO [finetune.py:976] (3/7) Epoch 23, batch 350, loss[loss=0.1765, simple_loss=0.2603, pruned_loss=0.04629, over 4827.00 frames. ], tot_loss[loss=0.1791, simple_loss=0.2504, pruned_loss=0.05393, over 791067.77 frames. ], batch size: 49, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:29:25,830 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.521e+01 1.514e+02 1.805e+02 2.248e+02 3.946e+02, threshold=3.610e+02, percent-clipped=1.0 2023-03-27 03:29:39,004 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126380.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:29:39,584 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126381.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:29:45,203 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.96 vs. limit=5.0 2023-03-27 03:29:56,093 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1125, 1.7400, 2.1680, 2.1555, 1.8980, 1.9293, 2.1085, 2.0522], device='cuda:3'), covar=tensor([0.4494, 0.4607, 0.3601, 0.4342, 0.5579, 0.4232, 0.5360, 0.3252], device='cuda:3'), in_proj_covar=tensor([0.0259, 0.0244, 0.0264, 0.0287, 0.0284, 0.0262, 0.0293, 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') 2023-03-27 03:29:59,417 INFO [finetune.py:976] (3/7) Epoch 23, batch 400, loss[loss=0.141, simple_loss=0.2161, pruned_loss=0.033, over 4781.00 frames. ], tot_loss[loss=0.1787, simple_loss=0.2505, pruned_loss=0.05344, over 828362.02 frames. ], batch size: 26, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:30:21,835 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=126429.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:30:25,131 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-27 03:30:29,122 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7761, 1.6822, 1.4507, 1.6813, 2.1319, 2.0571, 1.6336, 1.4974], device='cuda:3'), covar=tensor([0.0273, 0.0297, 0.0593, 0.0303, 0.0218, 0.0385, 0.0370, 0.0422], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0106, 0.0142, 0.0111, 0.0098, 0.0111, 0.0101, 0.0111], device='cuda:3'), out_proj_covar=tensor([7.6345e-05, 8.1182e-05, 1.1140e-04, 8.5128e-05, 7.6349e-05, 8.1906e-05, 7.5054e-05, 8.4533e-05], device='cuda:3') 2023-03-27 03:30:29,721 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126441.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:30:35,179 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126450.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:30:40,549 INFO [finetune.py:976] (3/7) Epoch 23, batch 450, loss[loss=0.2508, simple_loss=0.3026, pruned_loss=0.09947, over 4825.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2483, pruned_loss=0.05229, over 858755.70 frames. ], batch size: 38, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:30:42,258 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.347e+01 1.414e+02 1.639e+02 2.017e+02 3.767e+02, threshold=3.277e+02, percent-clipped=3.0 2023-03-27 03:30:42,668 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-27 03:30:45,273 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7772, 1.7364, 1.5620, 1.9330, 2.2187, 1.9472, 1.5626, 1.4583], device='cuda:3'), covar=tensor([0.2296, 0.2082, 0.1995, 0.1761, 0.1681, 0.1190, 0.2434, 0.2065], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0211, 0.0213, 0.0197, 0.0244, 0.0189, 0.0217, 0.0205], 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') 2023-03-27 03:31:13,831 INFO [finetune.py:976] (3/7) Epoch 23, batch 500, loss[loss=0.1509, simple_loss=0.2225, pruned_loss=0.03967, over 4788.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2463, pruned_loss=0.05206, over 882903.66 frames. ], batch size: 28, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:31:15,651 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126511.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:31:47,411 INFO [finetune.py:976] (3/7) Epoch 23, batch 550, loss[loss=0.146, simple_loss=0.2132, pruned_loss=0.03945, over 4937.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2417, pruned_loss=0.05015, over 899358.26 frames. ], batch size: 33, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:31:48,615 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.316e+01 1.588e+02 1.845e+02 2.407e+02 4.330e+02, threshold=3.691e+02, percent-clipped=4.0 2023-03-27 03:32:12,919 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-03-27 03:32:21,215 INFO [finetune.py:976] (3/7) Epoch 23, batch 600, loss[loss=0.1917, simple_loss=0.269, pruned_loss=0.05724, over 4826.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.243, pruned_loss=0.05097, over 911653.03 frames. ], batch size: 40, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:32:31,610 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-27 03:32:44,173 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.57 vs. limit=5.0 2023-03-27 03:33:02,508 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126654.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:33:03,818 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3103, 2.1656, 1.7017, 2.2233, 2.1955, 1.9275, 2.5271, 2.2934], device='cuda:3'), covar=tensor([0.1398, 0.2232, 0.3206, 0.2591, 0.2583, 0.1761, 0.2813, 0.1688], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0190, 0.0237, 0.0256, 0.0251, 0.0207, 0.0217, 0.0203], 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') 2023-03-27 03:33:05,531 INFO [finetune.py:976] (3/7) Epoch 23, batch 650, loss[loss=0.1414, simple_loss=0.1999, pruned_loss=0.04141, over 4297.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2447, pruned_loss=0.05116, over 921568.68 frames. ], batch size: 18, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:33:06,760 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.083e+02 1.579e+02 1.902e+02 2.270e+02 1.001e+03, threshold=3.804e+02, percent-clipped=1.0 2023-03-27 03:33:06,998 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-27 03:33:13,495 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-03-27 03:33:42,517 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=126702.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:33:47,146 INFO [finetune.py:976] (3/7) Epoch 23, batch 700, loss[loss=0.1712, simple_loss=0.24, pruned_loss=0.05118, over 4759.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2461, pruned_loss=0.05154, over 929326.45 frames. ], batch size: 27, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:34:11,542 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126736.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:34:18,456 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3111, 1.3273, 1.5555, 1.0154, 1.2196, 1.4198, 1.3078, 1.6042], device='cuda:3'), covar=tensor([0.1260, 0.2331, 0.1308, 0.1579, 0.1058, 0.1382, 0.2938, 0.0914], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0206, 0.0192, 0.0190, 0.0174, 0.0215, 0.0217, 0.0200], 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') 2023-03-27 03:34:29,143 INFO [finetune.py:976] (3/7) Epoch 23, batch 750, loss[loss=0.1689, simple_loss=0.2522, pruned_loss=0.0428, over 4878.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2467, pruned_loss=0.05159, over 934593.92 frames. ], batch size: 35, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:34:30,821 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.063e+01 1.442e+02 1.710e+02 2.057e+02 3.398e+02, threshold=3.419e+02, percent-clipped=0.0 2023-03-27 03:35:00,637 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126806.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:35:02,394 INFO [finetune.py:976] (3/7) Epoch 23, batch 800, loss[loss=0.1474, simple_loss=0.2196, pruned_loss=0.0376, over 4857.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.245, pruned_loss=0.0504, over 939912.96 frames. ], batch size: 44, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:35:07,275 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126816.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:35:07,882 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5205, 1.8306, 2.3011, 1.8183, 1.9050, 4.2532, 1.5796, 1.8800], device='cuda:3'), covar=tensor([0.1022, 0.1615, 0.1148, 0.1019, 0.1488, 0.0211, 0.1449, 0.1637], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0073, 0.0076, 0.0091, 0.0081, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 03:35:44,510 INFO [finetune.py:976] (3/7) Epoch 23, batch 850, loss[loss=0.1974, simple_loss=0.2592, pruned_loss=0.06784, over 4820.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2431, pruned_loss=0.04952, over 942268.31 frames. ], batch size: 40, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:35:45,682 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.329e+01 1.478e+02 1.768e+02 2.024e+02 3.574e+02, threshold=3.536e+02, percent-clipped=1.0 2023-03-27 03:35:47,057 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1681, 1.9760, 1.7527, 1.8908, 1.8836, 1.8807, 1.9438, 2.6497], device='cuda:3'), covar=tensor([0.3940, 0.4418, 0.3360, 0.3836, 0.4021, 0.2509, 0.3786, 0.1672], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0262, 0.0233, 0.0277, 0.0256, 0.0226, 0.0253, 0.0235], 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') 2023-03-27 03:35:56,164 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126877.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:36:18,420 INFO [finetune.py:976] (3/7) Epoch 23, batch 900, loss[loss=0.2093, simple_loss=0.2666, pruned_loss=0.076, over 4827.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2425, pruned_loss=0.05018, over 942571.97 frames. ], batch size: 33, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:36:19,720 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1683, 3.6228, 3.8003, 4.0191, 3.9855, 3.6814, 4.2206, 1.3248], device='cuda:3'), covar=tensor([0.0812, 0.0922, 0.0863, 0.0935, 0.1165, 0.1541, 0.0777, 0.5752], device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0245, 0.0279, 0.0290, 0.0334, 0.0283, 0.0304, 0.0299], 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') 2023-03-27 03:36:52,033 INFO [finetune.py:976] (3/7) Epoch 23, batch 950, loss[loss=0.1779, simple_loss=0.2533, pruned_loss=0.05129, over 4816.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.241, pruned_loss=0.05038, over 946140.36 frames. ], batch size: 33, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:36:53,225 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.057e+02 1.486e+02 1.775e+02 2.132e+02 3.351e+02, threshold=3.551e+02, percent-clipped=0.0 2023-03-27 03:37:06,070 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9178, 4.1266, 3.9115, 1.9085, 4.2822, 3.2760, 0.8144, 2.8186], device='cuda:3'), covar=tensor([0.2597, 0.1606, 0.1502, 0.3300, 0.0785, 0.0869, 0.4565, 0.1485], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0177, 0.0159, 0.0128, 0.0160, 0.0123, 0.0148, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 03:37:22,012 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7476, 1.6415, 1.4709, 1.8352, 2.0292, 1.8030, 1.2839, 1.4432], device='cuda:3'), covar=tensor([0.2165, 0.1988, 0.1970, 0.1614, 0.1570, 0.1236, 0.2566, 0.1956], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0211, 0.0214, 0.0198, 0.0245, 0.0190, 0.0218, 0.0205], 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') 2023-03-27 03:37:26,096 INFO [finetune.py:976] (3/7) Epoch 23, batch 1000, loss[loss=0.2306, simple_loss=0.2959, pruned_loss=0.08268, over 4160.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.245, pruned_loss=0.05211, over 947697.38 frames. ], batch size: 65, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:37:43,480 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127036.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:37:51,217 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8880, 1.9355, 1.6643, 2.1403, 2.5616, 2.0328, 1.8638, 1.4826], device='cuda:3'), covar=tensor([0.2231, 0.1913, 0.1903, 0.1652, 0.1558, 0.1224, 0.2119, 0.2007], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0211, 0.0214, 0.0198, 0.0245, 0.0190, 0.0217, 0.0205], 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') 2023-03-27 03:38:01,446 INFO [finetune.py:976] (3/7) Epoch 23, batch 1050, loss[loss=0.2268, simple_loss=0.2959, pruned_loss=0.07886, over 4903.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2487, pruned_loss=0.05306, over 950081.25 frames. ], batch size: 43, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:38:02,653 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.005e+02 1.597e+02 1.830e+02 2.248e+02 5.450e+02, threshold=3.660e+02, percent-clipped=4.0 2023-03-27 03:38:27,251 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=127084.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:38:42,639 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127106.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:38:44,896 INFO [finetune.py:976] (3/7) Epoch 23, batch 1100, loss[loss=0.1795, simple_loss=0.2531, pruned_loss=0.05292, over 4855.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2492, pruned_loss=0.05304, over 952928.61 frames. ], batch size: 31, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:39:08,747 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.02 vs. limit=5.0 2023-03-27 03:39:14,544 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=127154.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:39:21,916 INFO [finetune.py:976] (3/7) Epoch 23, batch 1150, loss[loss=0.2231, simple_loss=0.2839, pruned_loss=0.08118, over 4370.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2488, pruned_loss=0.05271, over 950684.93 frames. ], batch size: 66, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:39:23,614 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.031e+02 1.547e+02 1.781e+02 2.032e+02 3.739e+02, threshold=3.562e+02, percent-clipped=1.0 2023-03-27 03:39:33,497 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.10 vs. limit=5.0 2023-03-27 03:39:35,833 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127172.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:40:04,147 INFO [finetune.py:976] (3/7) Epoch 23, batch 1200, loss[loss=0.1766, simple_loss=0.2452, pruned_loss=0.05405, over 4893.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2485, pruned_loss=0.05334, over 952581.43 frames. ], batch size: 43, lr: 3.11e-03, grad_scale: 64.0 2023-03-27 03:40:10,142 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1120, 2.0007, 2.1393, 0.9434, 2.4569, 2.6745, 2.3611, 1.8898], device='cuda:3'), covar=tensor([0.0941, 0.0780, 0.0539, 0.0736, 0.0496, 0.0614, 0.0450, 0.0786], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0148, 0.0127, 0.0122, 0.0130, 0.0129, 0.0141, 0.0148], device='cuda:3'), out_proj_covar=tensor([8.9516e-05, 1.0709e-04, 9.0898e-05, 8.5929e-05, 9.1560e-05, 9.1630e-05, 1.0076e-04, 1.0597e-04], device='cuda:3') 2023-03-27 03:40:33,028 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127249.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:40:42,288 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7937, 1.3141, 0.7455, 1.6420, 2.1863, 1.4609, 1.5915, 1.7153], device='cuda:3'), covar=tensor([0.1416, 0.2151, 0.2185, 0.1202, 0.1877, 0.1947, 0.1501, 0.1894], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0094, 0.0110, 0.0092, 0.0119, 0.0093, 0.0098, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 03:40:44,045 INFO [finetune.py:976] (3/7) Epoch 23, batch 1250, loss[loss=0.1555, simple_loss=0.2107, pruned_loss=0.05013, over 4385.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2459, pruned_loss=0.05261, over 953453.61 frames. ], batch size: 19, lr: 3.11e-03, grad_scale: 64.0 2023-03-27 03:40:45,213 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.051e+02 1.489e+02 1.742e+02 2.248e+02 3.707e+02, threshold=3.484e+02, percent-clipped=1.0 2023-03-27 03:41:21,265 INFO [finetune.py:976] (3/7) Epoch 23, batch 1300, loss[loss=0.1803, simple_loss=0.2496, pruned_loss=0.05552, over 4904.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2425, pruned_loss=0.05113, over 952980.48 frames. ], batch size: 35, lr: 3.11e-03, grad_scale: 64.0 2023-03-27 03:41:22,020 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127310.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:41:23,224 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9007, 1.9211, 1.7332, 1.9675, 1.6000, 4.5682, 1.7549, 2.2350], device='cuda:3'), covar=tensor([0.2978, 0.2375, 0.1993, 0.2110, 0.1448, 0.0121, 0.2324, 0.1119], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0116, 0.0120, 0.0123, 0.0113, 0.0095, 0.0094, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 03:41:54,395 INFO [finetune.py:976] (3/7) Epoch 23, batch 1350, loss[loss=0.1608, simple_loss=0.2359, pruned_loss=0.04291, over 4835.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2432, pruned_loss=0.05151, over 953827.43 frames. ], batch size: 30, lr: 3.11e-03, grad_scale: 64.0 2023-03-27 03:41:55,607 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.051e+02 1.500e+02 1.797e+02 2.250e+02 4.549e+02, threshold=3.594e+02, percent-clipped=1.0 2023-03-27 03:41:56,839 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6892, 1.6671, 1.5977, 1.6505, 1.2567, 4.2339, 1.5213, 1.9779], device='cuda:3'), covar=tensor([0.3228, 0.2524, 0.2107, 0.2353, 0.1729, 0.0128, 0.2543, 0.1213], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0117, 0.0121, 0.0124, 0.0113, 0.0096, 0.0094, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 03:42:27,769 INFO [finetune.py:976] (3/7) Epoch 23, batch 1400, loss[loss=0.2226, simple_loss=0.2809, pruned_loss=0.08213, over 4891.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2456, pruned_loss=0.05211, over 953621.79 frames. ], batch size: 32, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:42:28,508 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.3047, 1.4872, 1.6028, 0.8171, 1.5793, 1.8286, 1.8655, 1.4578], device='cuda:3'), covar=tensor([0.0952, 0.0801, 0.0534, 0.0528, 0.0461, 0.0592, 0.0341, 0.0687], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0149, 0.0128, 0.0122, 0.0131, 0.0130, 0.0142, 0.0149], device='cuda:3'), out_proj_covar=tensor([9.0077e-05, 1.0739e-04, 9.1468e-05, 8.6158e-05, 9.1887e-05, 9.2330e-05, 1.0113e-04, 1.0634e-04], device='cuda:3') 2023-03-27 03:42:40,734 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127427.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 03:43:01,001 INFO [finetune.py:976] (3/7) Epoch 23, batch 1450, loss[loss=0.2021, simple_loss=0.2793, pruned_loss=0.06245, over 4813.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2462, pruned_loss=0.05172, over 953452.08 frames. ], batch size: 40, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:43:03,312 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.817e+01 1.596e+02 1.913e+02 2.194e+02 3.811e+02, threshold=3.827e+02, percent-clipped=1.0 2023-03-27 03:43:09,877 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127472.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:43:25,066 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127488.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 03:43:26,254 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6580, 3.7472, 3.4875, 1.4450, 3.8267, 2.8261, 0.7075, 2.5561], device='cuda:3'), covar=tensor([0.2342, 0.1776, 0.1530, 0.3651, 0.0953, 0.1017, 0.4755, 0.1499], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0178, 0.0160, 0.0129, 0.0161, 0.0124, 0.0150, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 03:43:46,709 INFO [finetune.py:976] (3/7) Epoch 23, batch 1500, loss[loss=0.1872, simple_loss=0.263, pruned_loss=0.05569, over 4912.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2487, pruned_loss=0.05246, over 955674.22 frames. ], batch size: 38, lr: 3.11e-03, grad_scale: 32.0 2023-03-27 03:43:54,391 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=127520.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:44:00,482 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127529.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:44:20,425 INFO [finetune.py:976] (3/7) Epoch 23, batch 1550, loss[loss=0.1946, simple_loss=0.2691, pruned_loss=0.06, over 4909.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2484, pruned_loss=0.05238, over 954003.56 frames. ], batch size: 46, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:44:22,241 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.110e+02 1.462e+02 1.755e+02 2.123e+02 3.197e+02, threshold=3.511e+02, percent-clipped=0.0 2023-03-27 03:44:48,014 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127590.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:45:04,792 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127605.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:45:07,683 INFO [finetune.py:976] (3/7) Epoch 23, batch 1600, loss[loss=0.1846, simple_loss=0.2534, pruned_loss=0.05796, over 4757.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.245, pruned_loss=0.05129, over 954597.31 frames. ], batch size: 27, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:45:34,727 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 2023-03-27 03:45:40,960 INFO [finetune.py:976] (3/7) Epoch 23, batch 1650, loss[loss=0.1526, simple_loss=0.2277, pruned_loss=0.03872, over 4928.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2433, pruned_loss=0.05117, over 956442.95 frames. ], batch size: 38, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:45:43,333 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.066e+02 1.567e+02 1.812e+02 2.212e+02 4.212e+02, threshold=3.624e+02, percent-clipped=4.0 2023-03-27 03:45:43,448 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127662.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:46:17,915 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6751, 3.6420, 3.4337, 1.6400, 3.6855, 2.7352, 0.9887, 2.5784], device='cuda:3'), covar=tensor([0.2504, 0.2178, 0.1475, 0.3509, 0.1118, 0.1090, 0.4273, 0.1527], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0178, 0.0160, 0.0129, 0.0161, 0.0124, 0.0149, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 03:46:24,653 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8453, 1.8248, 1.5502, 2.0249, 2.5010, 2.0088, 1.7810, 1.5049], device='cuda:3'), covar=tensor([0.2143, 0.1862, 0.1901, 0.1531, 0.1508, 0.1168, 0.2061, 0.1852], device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0209, 0.0212, 0.0196, 0.0243, 0.0188, 0.0216, 0.0203], 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') 2023-03-27 03:46:26,937 INFO [finetune.py:976] (3/7) Epoch 23, batch 1700, loss[loss=0.145, simple_loss=0.2189, pruned_loss=0.03555, over 4762.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.241, pruned_loss=0.0503, over 957105.07 frames. ], batch size: 27, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:46:27,681 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5519, 2.4225, 2.0445, 1.0173, 2.1469, 1.8978, 1.8452, 2.1024], device='cuda:3'), covar=tensor([0.0915, 0.0819, 0.1702, 0.2188, 0.1518, 0.2541, 0.2122, 0.1098], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0193, 0.0199, 0.0183, 0.0210, 0.0210, 0.0225, 0.0197], 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') 2023-03-27 03:46:31,266 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-27 03:46:36,928 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127723.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:46:38,680 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.9955, 4.2898, 4.5632, 4.7867, 4.7366, 4.5002, 5.1133, 1.4866], device='cuda:3'), covar=tensor([0.0677, 0.0908, 0.0689, 0.0907, 0.1088, 0.1500, 0.0533, 0.6236], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0247, 0.0281, 0.0293, 0.0336, 0.0287, 0.0306, 0.0302], 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') 2023-03-27 03:46:53,103 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2023-03-27 03:47:00,095 INFO [finetune.py:976] (3/7) Epoch 23, batch 1750, loss[loss=0.2281, simple_loss=0.2946, pruned_loss=0.08077, over 4802.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2414, pruned_loss=0.04993, over 957450.41 frames. ], batch size: 41, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:47:01,902 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.127e+02 1.511e+02 1.783e+02 2.157e+02 3.427e+02, threshold=3.565e+02, percent-clipped=0.0 2023-03-27 03:47:03,267 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4194, 1.3858, 1.1795, 1.3975, 1.7122, 1.6130, 1.4111, 1.2237], device='cuda:3'), covar=tensor([0.0375, 0.0304, 0.0634, 0.0302, 0.0222, 0.0456, 0.0300, 0.0406], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0106, 0.0143, 0.0111, 0.0098, 0.0111, 0.0100, 0.0111], device='cuda:3'), out_proj_covar=tensor([7.6546e-05, 8.1052e-05, 1.1198e-04, 8.5014e-05, 7.6502e-05, 8.2071e-05, 7.4633e-05, 8.4679e-05], device='cuda:3') 2023-03-27 03:47:13,901 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127779.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:47:16,781 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127783.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 03:47:33,985 INFO [finetune.py:976] (3/7) Epoch 23, batch 1800, loss[loss=0.1679, simple_loss=0.2298, pruned_loss=0.05295, over 4755.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2445, pruned_loss=0.04992, over 958633.33 frames. ], batch size: 26, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:47:54,757 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127840.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:48:01,843 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127850.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:48:07,744 INFO [finetune.py:976] (3/7) Epoch 23, batch 1850, loss[loss=0.2053, simple_loss=0.28, pruned_loss=0.06533, over 4746.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2467, pruned_loss=0.05062, over 958307.23 frames. ], batch size: 54, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:48:09,578 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.256e+01 1.480e+02 1.674e+02 2.095e+02 4.093e+02, threshold=3.347e+02, percent-clipped=1.0 2023-03-27 03:48:16,151 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127872.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:48:24,942 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127885.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:48:40,232 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127905.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:48:42,624 INFO [finetune.py:976] (3/7) Epoch 23, batch 1900, loss[loss=0.1727, simple_loss=0.242, pruned_loss=0.05171, over 4786.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2484, pruned_loss=0.05142, over 957850.75 frames. ], batch size: 25, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:48:43,930 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127911.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:48:58,799 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127933.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:49:03,523 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5278, 1.3901, 2.2602, 3.4285, 2.2475, 2.4941, 1.1569, 2.8976], device='cuda:3'), covar=tensor([0.2110, 0.2052, 0.1506, 0.1070, 0.0979, 0.1547, 0.2088, 0.0644], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0117, 0.0136, 0.0166, 0.0102, 0.0138, 0.0126, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 03:49:11,956 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=127953.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:49:16,014 INFO [finetune.py:976] (3/7) Epoch 23, batch 1950, loss[loss=0.1662, simple_loss=0.2419, pruned_loss=0.04527, over 4259.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2467, pruned_loss=0.05071, over 957002.48 frames. ], batch size: 66, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:49:16,109 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3931, 1.3730, 1.6235, 1.6058, 1.5599, 2.9388, 1.2157, 1.4484], device='cuda:3'), covar=tensor([0.0958, 0.1748, 0.1056, 0.0951, 0.1586, 0.0314, 0.1513, 0.1852], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0074, 0.0076, 0.0092, 0.0081, 0.0086, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 03:49:17,848 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.653e+01 1.391e+02 1.764e+02 2.084e+02 3.552e+02, threshold=3.528e+02, percent-clipped=1.0 2023-03-27 03:49:53,065 INFO [finetune.py:976] (3/7) Epoch 23, batch 2000, loss[loss=0.1436, simple_loss=0.2188, pruned_loss=0.03421, over 4828.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2445, pruned_loss=0.05022, over 956733.50 frames. ], batch size: 33, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:50:07,584 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128018.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:50:27,225 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8169, 1.6773, 1.5320, 1.9598, 2.1026, 1.9039, 1.3738, 1.5083], device='cuda:3'), covar=tensor([0.2367, 0.2045, 0.1976, 0.1731, 0.1596, 0.1232, 0.2538, 0.2010], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0211, 0.0214, 0.0197, 0.0245, 0.0190, 0.0217, 0.0205], 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') 2023-03-27 03:50:29,544 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0184, 1.4527, 2.0705, 2.0127, 1.8439, 1.7916, 1.9594, 1.9764], device='cuda:3'), covar=tensor([0.3872, 0.3719, 0.3217, 0.3480, 0.4552, 0.3533, 0.4144, 0.2954], device='cuda:3'), in_proj_covar=tensor([0.0259, 0.0243, 0.0264, 0.0288, 0.0286, 0.0262, 0.0295, 0.0248], 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') 2023-03-27 03:50:38,973 INFO [finetune.py:976] (3/7) Epoch 23, batch 2050, loss[loss=0.1733, simple_loss=0.2385, pruned_loss=0.05405, over 4901.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2417, pruned_loss=0.04945, over 957652.89 frames. ], batch size: 32, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:50:41,272 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.050e+02 1.483e+02 1.749e+02 2.101e+02 3.191e+02, threshold=3.498e+02, percent-clipped=0.0 2023-03-27 03:50:42,885 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-03-27 03:50:54,693 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128083.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 03:50:56,357 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9065, 1.9394, 1.7671, 1.9096, 1.4787, 4.5161, 1.5908, 2.1086], device='cuda:3'), covar=tensor([0.3290, 0.2484, 0.2078, 0.2265, 0.1594, 0.0122, 0.2447, 0.1234], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0117, 0.0121, 0.0124, 0.0114, 0.0096, 0.0095, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 03:51:01,038 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9488, 1.8965, 2.3058, 1.4533, 2.0562, 2.2957, 1.6803, 2.4082], device='cuda:3'), covar=tensor([0.1442, 0.1956, 0.1295, 0.1893, 0.0963, 0.1553, 0.2871, 0.0932], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0208, 0.0194, 0.0191, 0.0175, 0.0215, 0.0218, 0.0200], 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') 2023-03-27 03:51:13,973 INFO [finetune.py:976] (3/7) Epoch 23, batch 2100, loss[loss=0.1482, simple_loss=0.2328, pruned_loss=0.03179, over 4812.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2422, pruned_loss=0.05003, over 956396.93 frames. ], batch size: 51, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:51:19,703 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-27 03:51:32,383 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8554, 1.7247, 2.3774, 1.5107, 1.9820, 2.2330, 1.5647, 2.2826], device='cuda:3'), covar=tensor([0.1465, 0.2168, 0.1673, 0.2099, 0.1041, 0.1586, 0.2827, 0.0985], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0207, 0.0193, 0.0190, 0.0174, 0.0215, 0.0217, 0.0199], 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') 2023-03-27 03:51:40,712 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=128131.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 03:51:43,612 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128135.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:52:00,525 INFO [finetune.py:976] (3/7) Epoch 23, batch 2150, loss[loss=0.211, simple_loss=0.2889, pruned_loss=0.06658, over 4823.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2456, pruned_loss=0.05127, over 954538.43 frames. ], batch size: 39, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:52:02,378 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.009e+02 1.558e+02 1.811e+02 2.178e+02 3.611e+02, threshold=3.622e+02, percent-clipped=2.0 2023-03-27 03:52:17,001 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128185.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:52:17,259 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-03-27 03:52:21,299 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-27 03:52:31,847 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128206.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:52:33,599 INFO [finetune.py:976] (3/7) Epoch 23, batch 2200, loss[loss=0.1797, simple_loss=0.2554, pruned_loss=0.05203, over 4891.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2475, pruned_loss=0.05207, over 955023.28 frames. ], batch size: 35, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:52:43,042 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8575, 2.6666, 2.1879, 1.2028, 2.3634, 2.2015, 2.0774, 2.3915], device='cuda:3'), covar=tensor([0.0684, 0.0801, 0.1583, 0.1970, 0.1356, 0.1995, 0.1909, 0.0891], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0191, 0.0199, 0.0182, 0.0209, 0.0207, 0.0224, 0.0194], 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') 2023-03-27 03:52:46,646 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128228.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:52:49,640 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=128233.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:53:07,264 INFO [finetune.py:976] (3/7) Epoch 23, batch 2250, loss[loss=0.2038, simple_loss=0.2777, pruned_loss=0.0649, over 4750.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2483, pruned_loss=0.05199, over 953637.12 frames. ], batch size: 54, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:53:09,084 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.465e+01 1.491e+02 1.772e+02 2.217e+02 3.841e+02, threshold=3.544e+02, percent-clipped=1.0 2023-03-27 03:53:13,170 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3041, 1.4580, 1.6790, 1.4343, 1.5898, 2.9385, 1.3733, 1.5393], device='cuda:3'), covar=tensor([0.1081, 0.1873, 0.1017, 0.1032, 0.1586, 0.0317, 0.1508, 0.1886], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0073, 0.0076, 0.0091, 0.0081, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 03:53:37,122 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-03-27 03:53:40,844 INFO [finetune.py:976] (3/7) Epoch 23, batch 2300, loss[loss=0.187, simple_loss=0.2669, pruned_loss=0.05356, over 4194.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2482, pruned_loss=0.0517, over 953121.68 frames. ], batch size: 65, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:53:47,288 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128318.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:53:48,504 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128320.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:54:13,557 INFO [finetune.py:976] (3/7) Epoch 23, batch 2350, loss[loss=0.1927, simple_loss=0.253, pruned_loss=0.0662, over 4925.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2464, pruned_loss=0.05127, over 952754.90 frames. ], batch size: 38, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:54:13,661 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4675, 1.5217, 1.9868, 1.6941, 1.6825, 3.4830, 1.3957, 1.6146], device='cuda:3'), covar=tensor([0.1011, 0.1796, 0.1102, 0.0994, 0.1547, 0.0238, 0.1489, 0.1817], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0074, 0.0076, 0.0092, 0.0081, 0.0086, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 03:54:15,915 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.030e+02 1.510e+02 1.720e+02 2.103e+02 3.385e+02, threshold=3.440e+02, percent-clipped=0.0 2023-03-27 03:54:18,448 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=128366.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:54:28,982 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128381.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:54:46,896 INFO [finetune.py:976] (3/7) Epoch 23, batch 2400, loss[loss=0.174, simple_loss=0.2403, pruned_loss=0.0538, over 4921.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2434, pruned_loss=0.05057, over 954396.97 frames. ], batch size: 37, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:54:49,469 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128413.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:55:06,817 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128435.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:55:35,543 INFO [finetune.py:976] (3/7) Epoch 23, batch 2450, loss[loss=0.2125, simple_loss=0.2715, pruned_loss=0.07677, over 4832.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2408, pruned_loss=0.04953, over 956077.62 frames. ], batch size: 40, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:55:41,390 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.054e+01 1.492e+02 1.779e+02 2.209e+02 4.084e+02, threshold=3.557e+02, percent-clipped=2.0 2023-03-27 03:55:49,229 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5598, 1.7100, 1.4271, 1.4553, 2.0905, 1.9694, 1.8535, 1.7241], device='cuda:3'), covar=tensor([0.0459, 0.0351, 0.0643, 0.0372, 0.0331, 0.0604, 0.0333, 0.0424], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0107, 0.0144, 0.0112, 0.0100, 0.0111, 0.0102, 0.0112], device='cuda:3'), out_proj_covar=tensor([7.7386e-05, 8.1911e-05, 1.1311e-04, 8.5685e-05, 7.7412e-05, 8.2172e-05, 7.5657e-05, 8.5574e-05], device='cuda:3') 2023-03-27 03:55:49,814 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128474.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:55:53,373 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0990, 1.7454, 2.4077, 3.7081, 2.6002, 2.6903, 0.8766, 2.9984], device='cuda:3'), covar=tensor([0.1594, 0.1349, 0.1291, 0.0499, 0.0716, 0.1807, 0.1867, 0.0414], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0117, 0.0134, 0.0165, 0.0100, 0.0138, 0.0125, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 03:55:55,750 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=128483.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:56:10,250 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128506.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:56:12,479 INFO [finetune.py:976] (3/7) Epoch 23, batch 2500, loss[loss=0.1472, simple_loss=0.2159, pruned_loss=0.03927, over 4689.00 frames. ], tot_loss[loss=0.171, simple_loss=0.242, pruned_loss=0.05001, over 956563.53 frames. ], batch size: 23, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:56:25,970 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128528.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:56:30,169 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8122, 1.6953, 1.5154, 1.3341, 1.6305, 1.5885, 1.6378, 2.2086], device='cuda:3'), covar=tensor([0.3487, 0.3589, 0.2937, 0.3487, 0.3588, 0.2237, 0.3145, 0.1692], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0264, 0.0236, 0.0278, 0.0257, 0.0228, 0.0255, 0.0237], 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') 2023-03-27 03:56:34,907 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8380, 1.2688, 0.7469, 1.6998, 2.0718, 1.3904, 1.5094, 1.5347], device='cuda:3'), covar=tensor([0.1427, 0.2127, 0.2127, 0.1168, 0.1928, 0.1973, 0.1496, 0.1946], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0095, 0.0111, 0.0093, 0.0120, 0.0095, 0.0100, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 03:56:48,370 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=128554.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:56:55,785 INFO [finetune.py:976] (3/7) Epoch 23, batch 2550, loss[loss=0.2104, simple_loss=0.2949, pruned_loss=0.06297, over 4179.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2464, pruned_loss=0.0515, over 958367.81 frames. ], batch size: 65, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:56:58,589 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.110e+02 1.563e+02 1.837e+02 2.202e+02 4.665e+02, threshold=3.674e+02, percent-clipped=3.0 2023-03-27 03:57:11,629 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=128576.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:57:23,667 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5447, 1.4761, 1.4811, 1.4664, 1.0417, 2.4610, 0.8870, 1.3595], device='cuda:3'), covar=tensor([0.3887, 0.3044, 0.2321, 0.2838, 0.1692, 0.0502, 0.2474, 0.1194], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0117, 0.0122, 0.0125, 0.0114, 0.0096, 0.0095, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 03:57:26,683 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6935, 1.5136, 2.1958, 3.2274, 2.1651, 2.3852, 1.0641, 2.6812], device='cuda:3'), covar=tensor([0.1547, 0.1285, 0.1062, 0.0472, 0.0739, 0.1367, 0.1616, 0.0426], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0116, 0.0133, 0.0164, 0.0100, 0.0137, 0.0124, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 03:57:33,477 INFO [finetune.py:976] (3/7) Epoch 23, batch 2600, loss[loss=0.2063, simple_loss=0.2693, pruned_loss=0.07163, over 4742.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2471, pruned_loss=0.05144, over 957623.54 frames. ], batch size: 59, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:57:50,503 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2599, 2.1478, 1.9719, 2.3907, 2.8124, 2.2950, 2.0461, 1.7148], device='cuda:3'), covar=tensor([0.2233, 0.1877, 0.1904, 0.1656, 0.1700, 0.1091, 0.2181, 0.1976], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0209, 0.0213, 0.0196, 0.0244, 0.0189, 0.0216, 0.0205], 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') 2023-03-27 03:57:55,200 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3316, 1.8673, 2.3503, 2.4012, 2.1502, 2.1242, 2.3106, 2.2779], device='cuda:3'), covar=tensor([0.3349, 0.3718, 0.3277, 0.3530, 0.4991, 0.3463, 0.4533, 0.2954], device='cuda:3'), in_proj_covar=tensor([0.0258, 0.0243, 0.0263, 0.0287, 0.0286, 0.0262, 0.0294, 0.0247], 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') 2023-03-27 03:58:07,055 INFO [finetune.py:976] (3/7) Epoch 23, batch 2650, loss[loss=0.1724, simple_loss=0.2425, pruned_loss=0.05116, over 4800.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2469, pruned_loss=0.05115, over 954292.84 frames. ], batch size: 51, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:58:08,891 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.089e+02 1.579e+02 1.820e+02 2.183e+02 3.562e+02, threshold=3.640e+02, percent-clipped=0.0 2023-03-27 03:58:18,916 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128676.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:58:40,902 INFO [finetune.py:976] (3/7) Epoch 23, batch 2700, loss[loss=0.1452, simple_loss=0.2269, pruned_loss=0.03179, over 4758.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2466, pruned_loss=0.05065, over 955858.65 frames. ], batch size: 26, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:59:04,713 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0 2023-03-27 03:59:14,675 INFO [finetune.py:976] (3/7) Epoch 23, batch 2750, loss[loss=0.1311, simple_loss=0.1884, pruned_loss=0.03692, over 4298.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.245, pruned_loss=0.05094, over 955461.29 frames. ], batch size: 18, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:59:16,465 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.314e+01 1.516e+02 1.813e+02 2.146e+02 3.615e+02, threshold=3.627e+02, percent-clipped=0.0 2023-03-27 03:59:21,319 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128769.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:59:30,014 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128781.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 03:59:48,512 INFO [finetune.py:976] (3/7) Epoch 23, batch 2800, loss[loss=0.1603, simple_loss=0.2406, pruned_loss=0.04004, over 4908.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2421, pruned_loss=0.04963, over 957713.08 frames. ], batch size: 46, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 03:59:50,431 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.5243, 1.5596, 1.5655, 0.8004, 1.6616, 1.8577, 1.8502, 1.4444], device='cuda:3'), covar=tensor([0.1111, 0.0796, 0.0538, 0.0664, 0.0525, 0.0706, 0.0327, 0.0803], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0150, 0.0127, 0.0123, 0.0131, 0.0129, 0.0141, 0.0149], device='cuda:3'), out_proj_covar=tensor([8.9726e-05, 1.0813e-04, 9.0911e-05, 8.6399e-05, 9.1923e-05, 9.2006e-05, 1.0099e-04, 1.0646e-04], device='cuda:3') 2023-03-27 04:00:10,901 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128842.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:00:22,070 INFO [finetune.py:976] (3/7) Epoch 23, batch 2850, loss[loss=0.1615, simple_loss=0.2397, pruned_loss=0.04163, over 4743.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2399, pruned_loss=0.04916, over 956480.01 frames. ], batch size: 54, lr: 3.10e-03, grad_scale: 32.0 2023-03-27 04:00:23,886 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.021e+01 1.432e+02 1.754e+02 2.169e+02 4.729e+02, threshold=3.508e+02, percent-clipped=1.0 2023-03-27 04:00:55,187 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128890.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 04:01:08,142 INFO [finetune.py:976] (3/7) Epoch 23, batch 2900, loss[loss=0.1788, simple_loss=0.252, pruned_loss=0.05277, over 4893.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2425, pruned_loss=0.04971, over 956819.36 frames. ], batch size: 43, lr: 3.09e-03, grad_scale: 32.0 2023-03-27 04:01:12,279 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-27 04:01:36,747 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128951.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 04:01:41,988 INFO [finetune.py:976] (3/7) Epoch 23, batch 2950, loss[loss=0.2055, simple_loss=0.2906, pruned_loss=0.06022, over 4803.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2457, pruned_loss=0.05048, over 957372.19 frames. ], batch size: 41, lr: 3.09e-03, grad_scale: 32.0 2023-03-27 04:01:43,786 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.806e+01 1.599e+02 2.012e+02 2.314e+02 4.261e+02, threshold=4.024e+02, percent-clipped=1.0 2023-03-27 04:01:52,340 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128976.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:02:26,892 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0045, 1.7896, 2.2428, 1.5213, 1.9050, 2.2150, 1.7102, 2.3458], device='cuda:3'), covar=tensor([0.0966, 0.1733, 0.1058, 0.1485, 0.0869, 0.1021, 0.2405, 0.0734], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0205, 0.0191, 0.0188, 0.0172, 0.0213, 0.0215, 0.0197], 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') 2023-03-27 04:02:31,486 INFO [finetune.py:976] (3/7) Epoch 23, batch 3000, loss[loss=0.2179, simple_loss=0.284, pruned_loss=0.07596, over 4011.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2476, pruned_loss=0.05144, over 955006.93 frames. ], batch size: 65, lr: 3.09e-03, grad_scale: 32.0 2023-03-27 04:02:31,486 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-27 04:02:38,288 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6303, 3.5271, 3.3378, 1.4866, 3.6123, 2.9112, 0.7518, 2.4153], device='cuda:3'), covar=tensor([0.1899, 0.1601, 0.1453, 0.3261, 0.0950, 0.0881, 0.3732, 0.1511], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0178, 0.0160, 0.0129, 0.0162, 0.0124, 0.0149, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 04:02:38,542 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4230, 1.5989, 1.7629, 1.6670, 1.8009, 3.0835, 1.5063, 1.6784], device='cuda:3'), covar=tensor([0.0921, 0.1582, 0.0858, 0.0808, 0.1321, 0.0339, 0.1285, 0.1563], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0074, 0.0076, 0.0092, 0.0081, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 04:02:40,570 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8880, 1.7007, 1.6456, 2.0063, 2.1238, 1.9369, 1.4375, 1.6197], device='cuda:3'), covar=tensor([0.2025, 0.1997, 0.1861, 0.1615, 0.1550, 0.1153, 0.2338, 0.1802], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0210, 0.0213, 0.0197, 0.0244, 0.0189, 0.0216, 0.0205], 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') 2023-03-27 04:02:42,305 INFO [finetune.py:1010] (3/7) Epoch 23, validation: loss=0.1567, simple_loss=0.225, pruned_loss=0.04424, over 2265189.00 frames. 2023-03-27 04:02:42,305 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6469MB 2023-03-27 04:02:44,227 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8095, 1.0535, 1.8216, 1.8488, 1.6429, 1.5354, 1.7128, 1.7312], device='cuda:3'), covar=tensor([0.3458, 0.3627, 0.2741, 0.3009, 0.3958, 0.3309, 0.3699, 0.2692], device='cuda:3'), in_proj_covar=tensor([0.0258, 0.0242, 0.0262, 0.0287, 0.0285, 0.0261, 0.0293, 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') 2023-03-27 04:02:51,934 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=129024.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:03:05,476 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 2023-03-27 04:03:14,563 INFO [finetune.py:976] (3/7) Epoch 23, batch 3050, loss[loss=0.1752, simple_loss=0.251, pruned_loss=0.04966, over 4928.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2473, pruned_loss=0.05115, over 952806.41 frames. ], batch size: 42, lr: 3.09e-03, grad_scale: 32.0 2023-03-27 04:03:16,830 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.176e+02 1.606e+02 1.899e+02 2.236e+02 5.313e+02, threshold=3.798e+02, percent-clipped=3.0 2023-03-27 04:03:20,538 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2023-03-27 04:03:22,120 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129069.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:03:23,920 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8276, 1.8250, 1.7585, 1.8154, 1.7563, 3.7349, 1.9727, 2.2841], device='cuda:3'), covar=tensor([0.2979, 0.2119, 0.1816, 0.2042, 0.1324, 0.0222, 0.2307, 0.0948], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0117, 0.0121, 0.0124, 0.0113, 0.0096, 0.0095, 0.0096], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 04:03:47,838 INFO [finetune.py:976] (3/7) Epoch 23, batch 3100, loss[loss=0.1468, simple_loss=0.2192, pruned_loss=0.03723, over 4821.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2461, pruned_loss=0.05105, over 954132.66 frames. ], batch size: 41, lr: 3.09e-03, grad_scale: 32.0 2023-03-27 04:03:53,281 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=129117.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:04:06,397 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129137.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:04:07,686 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1241, 1.8469, 1.9535, 1.3025, 2.0728, 2.0794, 2.0380, 1.6304], device='cuda:3'), covar=tensor([0.0581, 0.0759, 0.0740, 0.0901, 0.0699, 0.0647, 0.0628, 0.1231], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0136, 0.0139, 0.0120, 0.0125, 0.0138, 0.0138, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 04:04:20,604 INFO [finetune.py:976] (3/7) Epoch 23, batch 3150, loss[loss=0.1301, simple_loss=0.2062, pruned_loss=0.027, over 4797.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2423, pruned_loss=0.05015, over 953551.57 frames. ], batch size: 29, lr: 3.09e-03, grad_scale: 32.0 2023-03-27 04:04:22,456 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.200e+02 1.503e+02 1.751e+02 2.296e+02 3.694e+02, threshold=3.502e+02, percent-clipped=0.0 2023-03-27 04:05:01,881 INFO [finetune.py:976] (3/7) Epoch 23, batch 3200, loss[loss=0.1629, simple_loss=0.231, pruned_loss=0.04738, over 4908.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2398, pruned_loss=0.04962, over 954359.12 frames. ], batch size: 37, lr: 3.09e-03, grad_scale: 32.0 2023-03-27 04:05:26,232 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7903, 1.7385, 1.7213, 1.7273, 1.1936, 3.5572, 1.4239, 1.8824], device='cuda:3'), covar=tensor([0.3254, 0.2398, 0.1979, 0.2466, 0.1791, 0.0202, 0.2560, 0.1251], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0117, 0.0121, 0.0124, 0.0113, 0.0096, 0.0095, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 04:05:26,799 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129246.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 04:05:37,441 INFO [finetune.py:976] (3/7) Epoch 23, batch 3250, loss[loss=0.207, simple_loss=0.2733, pruned_loss=0.07041, over 4818.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2413, pruned_loss=0.05022, over 954338.53 frames. ], batch size: 33, lr: 3.09e-03, grad_scale: 32.0 2023-03-27 04:05:39,768 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.902e+01 1.487e+02 1.735e+02 2.015e+02 4.622e+02, threshold=3.470e+02, percent-clipped=1.0 2023-03-27 04:06:20,057 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.10 vs. limit=5.0 2023-03-27 04:06:22,331 INFO [finetune.py:976] (3/7) Epoch 23, batch 3300, loss[loss=0.1704, simple_loss=0.2642, pruned_loss=0.03835, over 4828.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2454, pruned_loss=0.05117, over 954428.68 frames. ], batch size: 39, lr: 3.09e-03, grad_scale: 32.0 2023-03-27 04:06:56,063 INFO [finetune.py:976] (3/7) Epoch 23, batch 3350, loss[loss=0.1881, simple_loss=0.2652, pruned_loss=0.05548, over 4825.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2477, pruned_loss=0.05193, over 955366.64 frames. ], batch size: 30, lr: 3.09e-03, grad_scale: 32.0 2023-03-27 04:06:57,831 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.264e+01 1.616e+02 1.807e+02 2.144e+02 4.365e+02, threshold=3.613e+02, percent-clipped=1.0 2023-03-27 04:07:41,549 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-27 04:07:47,796 INFO [finetune.py:976] (3/7) Epoch 23, batch 3400, loss[loss=0.1895, simple_loss=0.2571, pruned_loss=0.06093, over 4877.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2485, pruned_loss=0.05253, over 954121.12 frames. ], batch size: 31, lr: 3.09e-03, grad_scale: 64.0 2023-03-27 04:07:50,371 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129413.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 04:08:07,356 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129437.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:08:21,223 INFO [finetune.py:976] (3/7) Epoch 23, batch 3450, loss[loss=0.1741, simple_loss=0.2315, pruned_loss=0.05833, over 4912.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2484, pruned_loss=0.05193, over 954850.27 frames. ], batch size: 33, lr: 3.09e-03, grad_scale: 64.0 2023-03-27 04:08:23,471 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.062e+02 1.549e+02 1.981e+02 2.372e+02 4.494e+02, threshold=3.962e+02, percent-clipped=6.0 2023-03-27 04:08:31,793 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129474.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 04:08:39,421 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=129485.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:08:46,486 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4451, 1.3459, 1.3312, 1.3513, 0.7577, 2.2618, 0.7312, 1.2181], device='cuda:3'), covar=tensor([0.3205, 0.2494, 0.2174, 0.2556, 0.1978, 0.0342, 0.2702, 0.1295], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0116, 0.0121, 0.0124, 0.0113, 0.0096, 0.0095, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 04:08:48,665 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-27 04:08:54,911 INFO [finetune.py:976] (3/7) Epoch 23, batch 3500, loss[loss=0.1705, simple_loss=0.2331, pruned_loss=0.05392, over 4816.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2457, pruned_loss=0.05126, over 952602.94 frames. ], batch size: 41, lr: 3.09e-03, grad_scale: 64.0 2023-03-27 04:09:20,977 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129546.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 04:09:21,058 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-27 04:09:28,787 INFO [finetune.py:976] (3/7) Epoch 23, batch 3550, loss[loss=0.1352, simple_loss=0.1938, pruned_loss=0.03824, over 3979.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2428, pruned_loss=0.05023, over 953424.56 frames. ], batch size: 17, lr: 3.09e-03, grad_scale: 64.0 2023-03-27 04:09:30,571 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.427e+01 1.468e+02 1.701e+02 2.000e+02 4.470e+02, threshold=3.402e+02, percent-clipped=1.0 2023-03-27 04:09:36,081 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0 2023-03-27 04:09:36,650 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1588, 2.0074, 1.7807, 2.0122, 1.8676, 1.8947, 1.9397, 2.7368], device='cuda:3'), covar=tensor([0.3538, 0.4235, 0.3050, 0.3807, 0.4348, 0.2338, 0.3825, 0.1456], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0263, 0.0235, 0.0277, 0.0256, 0.0228, 0.0254, 0.0237], 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') 2023-03-27 04:09:44,860 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9557, 1.7228, 1.8716, 1.2177, 1.9199, 1.9297, 1.9499, 1.5308], device='cuda:3'), covar=tensor([0.0511, 0.0789, 0.0652, 0.0862, 0.0750, 0.0746, 0.0620, 0.1229], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0135, 0.0138, 0.0119, 0.0124, 0.0137, 0.0137, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 04:09:52,426 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=129594.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 04:10:10,276 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-27 04:10:11,048 INFO [finetune.py:976] (3/7) Epoch 23, batch 3600, loss[loss=0.1462, simple_loss=0.2126, pruned_loss=0.03986, over 4745.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2404, pruned_loss=0.04967, over 954956.75 frames. ], batch size: 23, lr: 3.09e-03, grad_scale: 64.0 2023-03-27 04:10:12,978 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129612.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:10:15,359 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3657, 3.3580, 3.1477, 1.5619, 3.5099, 2.6476, 0.7832, 2.3234], device='cuda:3'), covar=tensor([0.2856, 0.2379, 0.1766, 0.3427, 0.1165, 0.1103, 0.4457, 0.1611], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0178, 0.0160, 0.0129, 0.0161, 0.0123, 0.0148, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 04:10:44,782 INFO [finetune.py:976] (3/7) Epoch 23, batch 3650, loss[loss=0.1995, simple_loss=0.2656, pruned_loss=0.06667, over 4813.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2439, pruned_loss=0.05122, over 954857.88 frames. ], batch size: 38, lr: 3.09e-03, grad_scale: 64.0 2023-03-27 04:10:46,578 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.022e+02 1.535e+02 1.802e+02 2.202e+02 3.404e+02, threshold=3.605e+02, percent-clipped=1.0 2023-03-27 04:10:53,466 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129673.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:11:01,739 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4597, 1.3421, 1.4633, 0.7504, 1.5106, 1.5102, 1.4899, 1.2683], device='cuda:3'), covar=tensor([0.0619, 0.0841, 0.0746, 0.1022, 0.0918, 0.0735, 0.0665, 0.1324], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0135, 0.0139, 0.0120, 0.0125, 0.0138, 0.0138, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 04:11:13,119 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-27 04:11:23,354 INFO [finetune.py:976] (3/7) Epoch 23, batch 3700, loss[loss=0.2027, simple_loss=0.262, pruned_loss=0.07173, over 4842.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2469, pruned_loss=0.0521, over 955428.74 frames. ], batch size: 30, lr: 3.09e-03, grad_scale: 64.0 2023-03-27 04:11:40,024 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0835, 2.0695, 1.7528, 2.1264, 1.9587, 2.0169, 1.9568, 2.8704], device='cuda:3'), covar=tensor([0.3800, 0.4867, 0.3323, 0.4371, 0.4601, 0.2389, 0.4591, 0.1551], device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0263, 0.0234, 0.0275, 0.0255, 0.0227, 0.0253, 0.0236], 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') 2023-03-27 04:12:00,259 INFO [finetune.py:976] (3/7) Epoch 23, batch 3750, loss[loss=0.1312, simple_loss=0.1972, pruned_loss=0.03261, over 4734.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2487, pruned_loss=0.0528, over 953633.37 frames. ], batch size: 23, lr: 3.09e-03, grad_scale: 64.0 2023-03-27 04:12:02,071 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.237e+01 1.564e+02 1.874e+02 2.284e+02 3.839e+02, threshold=3.748e+02, percent-clipped=2.0 2023-03-27 04:12:06,380 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129769.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 04:12:20,307 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129791.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:12:35,164 INFO [finetune.py:976] (3/7) Epoch 23, batch 3800, loss[loss=0.1538, simple_loss=0.2337, pruned_loss=0.03699, over 4755.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2495, pruned_loss=0.05306, over 954323.05 frames. ], batch size: 27, lr: 3.09e-03, grad_scale: 64.0 2023-03-27 04:13:16,854 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129852.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:13:21,830 INFO [finetune.py:976] (3/7) Epoch 23, batch 3850, loss[loss=0.1649, simple_loss=0.2362, pruned_loss=0.04679, over 4932.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2473, pruned_loss=0.05162, over 955132.94 frames. ], batch size: 38, lr: 3.09e-03, grad_scale: 64.0 2023-03-27 04:13:24,152 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.288e+01 1.597e+02 1.881e+02 2.160e+02 3.613e+02, threshold=3.763e+02, percent-clipped=0.0 2023-03-27 04:13:42,890 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.4232, 3.8522, 3.9938, 4.2383, 4.1832, 3.9681, 4.5416, 1.4452], device='cuda:3'), covar=tensor([0.0817, 0.0882, 0.0940, 0.1062, 0.1426, 0.1572, 0.0771, 0.5959], device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0248, 0.0281, 0.0294, 0.0339, 0.0287, 0.0306, 0.0303], 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') 2023-03-27 04:13:55,057 INFO [finetune.py:976] (3/7) Epoch 23, batch 3900, loss[loss=0.1869, simple_loss=0.2661, pruned_loss=0.05382, over 4928.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2448, pruned_loss=0.05097, over 956296.38 frames. ], batch size: 38, lr: 3.09e-03, grad_scale: 64.0 2023-03-27 04:14:27,719 INFO [finetune.py:976] (3/7) Epoch 23, batch 3950, loss[loss=0.202, simple_loss=0.2732, pruned_loss=0.06544, over 4940.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2411, pruned_loss=0.04963, over 956039.58 frames. ], batch size: 38, lr: 3.09e-03, grad_scale: 64.0 2023-03-27 04:14:29,947 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.082e+02 1.481e+02 1.820e+02 2.101e+02 4.779e+02, threshold=3.640e+02, percent-clipped=1.0 2023-03-27 04:14:34,546 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129968.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:14:56,370 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130000.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:14:58,100 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130002.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:15:02,811 INFO [finetune.py:976] (3/7) Epoch 23, batch 4000, loss[loss=0.1555, simple_loss=0.217, pruned_loss=0.04698, over 4772.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2407, pruned_loss=0.0501, over 954965.17 frames. ], batch size: 54, lr: 3.09e-03, grad_scale: 64.0 2023-03-27 04:15:45,415 INFO [finetune.py:976] (3/7) Epoch 23, batch 4050, loss[loss=0.1835, simple_loss=0.2635, pruned_loss=0.05176, over 4914.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2455, pruned_loss=0.05249, over 954885.25 frames. ], batch size: 37, lr: 3.09e-03, grad_scale: 64.0 2023-03-27 04:15:47,217 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130061.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:15:47,690 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.063e+02 1.665e+02 1.960e+02 2.479e+02 5.275e+02, threshold=3.921e+02, percent-clipped=4.0 2023-03-27 04:15:48,449 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130063.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 04:15:53,612 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130069.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 04:16:15,194 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.6822, 1.5403, 1.4018, 0.7590, 1.5876, 1.7743, 1.7441, 1.3764], device='cuda:3'), covar=tensor([0.0898, 0.0574, 0.0511, 0.0588, 0.0428, 0.0568, 0.0343, 0.0698], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0149, 0.0127, 0.0123, 0.0131, 0.0130, 0.0141, 0.0149], device='cuda:3'), out_proj_covar=tensor([8.9573e-05, 1.0743e-04, 9.1045e-05, 8.6298e-05, 9.2306e-05, 9.2321e-05, 1.0090e-04, 1.0642e-04], device='cuda:3') 2023-03-27 04:16:19,200 INFO [finetune.py:976] (3/7) Epoch 23, batch 4100, loss[loss=0.193, simple_loss=0.2757, pruned_loss=0.05515, over 4815.00 frames. ], tot_loss[loss=0.1783, simple_loss=0.2493, pruned_loss=0.05365, over 955074.74 frames. ], batch size: 38, lr: 3.09e-03, grad_scale: 64.0 2023-03-27 04:16:26,577 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=130117.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 04:16:36,921 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6152, 1.5649, 2.3685, 2.1813, 1.7602, 4.4051, 1.4958, 1.6201], device='cuda:3'), covar=tensor([0.1265, 0.2424, 0.1435, 0.1101, 0.1990, 0.0239, 0.1988, 0.2638], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0073, 0.0076, 0.0092, 0.0081, 0.0086, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 04:16:48,386 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6742, 1.6418, 2.2134, 3.4604, 2.3424, 2.4004, 1.1908, 2.8546], device='cuda:3'), covar=tensor([0.1722, 0.1271, 0.1194, 0.0478, 0.0766, 0.1464, 0.1693, 0.0461], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0133, 0.0163, 0.0099, 0.0136, 0.0123, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 04:16:54,987 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130147.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:17:02,632 INFO [finetune.py:976] (3/7) Epoch 23, batch 4150, loss[loss=0.1624, simple_loss=0.2319, pruned_loss=0.04648, over 4745.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.249, pruned_loss=0.05331, over 953864.57 frames. ], batch size: 26, lr: 3.09e-03, grad_scale: 64.0 2023-03-27 04:17:04,905 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.036e+02 1.505e+02 1.863e+02 2.291e+02 4.324e+02, threshold=3.726e+02, percent-clipped=3.0 2023-03-27 04:17:12,127 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3974, 1.3755, 1.6014, 2.4609, 1.6488, 2.2527, 1.0149, 2.0919], device='cuda:3'), covar=tensor([0.1730, 0.1279, 0.1093, 0.0655, 0.0891, 0.1040, 0.1440, 0.0584], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0133, 0.0163, 0.0099, 0.0136, 0.0123, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 04:17:14,153 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-27 04:17:33,230 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1328, 1.9717, 2.3654, 4.0158, 2.7870, 2.8179, 1.0034, 3.3704], device='cuda:3'), covar=tensor([0.1634, 0.1297, 0.1376, 0.0558, 0.0695, 0.1512, 0.1966, 0.0411], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0133, 0.0163, 0.0099, 0.0136, 0.0123, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 04:17:36,688 INFO [finetune.py:976] (3/7) Epoch 23, batch 4200, loss[loss=0.1483, simple_loss=0.2195, pruned_loss=0.03857, over 4833.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2493, pruned_loss=0.05238, over 954954.92 frames. ], batch size: 47, lr: 3.09e-03, grad_scale: 64.0 2023-03-27 04:18:24,044 INFO [finetune.py:976] (3/7) Epoch 23, batch 4250, loss[loss=0.1741, simple_loss=0.23, pruned_loss=0.05912, over 4207.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2471, pruned_loss=0.05183, over 954896.65 frames. ], batch size: 65, lr: 3.08e-03, grad_scale: 64.0 2023-03-27 04:18:25,853 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.027e+02 1.516e+02 1.759e+02 2.094e+02 3.793e+02, threshold=3.518e+02, percent-clipped=1.0 2023-03-27 04:18:30,100 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130268.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:18:57,491 INFO [finetune.py:976] (3/7) Epoch 23, batch 4300, loss[loss=0.1477, simple_loss=0.2231, pruned_loss=0.03615, over 4897.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2437, pruned_loss=0.0507, over 954924.50 frames. ], batch size: 32, lr: 3.08e-03, grad_scale: 64.0 2023-03-27 04:19:02,812 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=130316.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:19:29,232 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130356.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:19:30,472 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130358.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 04:19:31,026 INFO [finetune.py:976] (3/7) Epoch 23, batch 4350, loss[loss=0.1539, simple_loss=0.2271, pruned_loss=0.04031, over 4834.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2401, pruned_loss=0.04932, over 953848.41 frames. ], batch size: 25, lr: 3.08e-03, grad_scale: 32.0 2023-03-27 04:19:33,425 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.012e+02 1.417e+02 1.746e+02 2.112e+02 4.412e+02, threshold=3.492e+02, percent-clipped=1.0 2023-03-27 04:19:48,221 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130384.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:20:04,340 INFO [finetune.py:976] (3/7) Epoch 23, batch 4400, loss[loss=0.1937, simple_loss=0.2642, pruned_loss=0.0616, over 4825.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2423, pruned_loss=0.05048, over 953918.82 frames. ], batch size: 40, lr: 3.08e-03, grad_scale: 32.0 2023-03-27 04:20:19,485 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130432.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:20:24,186 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130438.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:20:30,584 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130445.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:20:31,781 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130447.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:20:39,744 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.2525, 2.8707, 3.0080, 3.1909, 3.0124, 2.8544, 3.3069, 0.9858], device='cuda:3'), covar=tensor([0.1156, 0.1167, 0.1073, 0.1194, 0.1685, 0.1855, 0.1266, 0.5928], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0249, 0.0282, 0.0297, 0.0342, 0.0290, 0.0308, 0.0304], 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') 2023-03-27 04:20:46,864 INFO [finetune.py:976] (3/7) Epoch 23, batch 4450, loss[loss=0.1595, simple_loss=0.2281, pruned_loss=0.04546, over 4749.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2449, pruned_loss=0.05059, over 954537.59 frames. ], batch size: 26, lr: 3.08e-03, grad_scale: 32.0 2023-03-27 04:20:49,239 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.124e+02 1.491e+02 1.813e+02 2.246e+02 3.707e+02, threshold=3.626e+02, percent-clipped=3.0 2023-03-27 04:20:58,322 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4082, 1.3597, 1.7487, 1.6109, 1.4591, 3.1760, 1.3020, 1.4598], device='cuda:3'), covar=tensor([0.0947, 0.1801, 0.1193, 0.1052, 0.1713, 0.0218, 0.1532, 0.1899], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0081, 0.0073, 0.0076, 0.0091, 0.0081, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 04:21:00,118 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6338, 3.6165, 3.4981, 1.8725, 3.6649, 2.8479, 0.9764, 2.4947], device='cuda:3'), covar=tensor([0.2993, 0.2247, 0.1612, 0.3441, 0.1230, 0.1000, 0.4451, 0.1607], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0178, 0.0160, 0.0129, 0.0161, 0.0123, 0.0148, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 04:21:10,585 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130493.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:21:11,732 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=130495.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:21:14,694 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130499.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:21:20,673 INFO [finetune.py:976] (3/7) Epoch 23, batch 4500, loss[loss=0.1653, simple_loss=0.2452, pruned_loss=0.04268, over 4864.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2463, pruned_loss=0.05106, over 953805.77 frames. ], batch size: 34, lr: 3.08e-03, grad_scale: 32.0 2023-03-27 04:21:31,700 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-27 04:21:53,305 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130548.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:22:04,071 INFO [finetune.py:976] (3/7) Epoch 23, batch 4550, loss[loss=0.173, simple_loss=0.2459, pruned_loss=0.05009, over 4844.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2479, pruned_loss=0.05205, over 954386.85 frames. ], batch size: 44, lr: 3.08e-03, grad_scale: 32.0 2023-03-27 04:22:06,499 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.768e+01 1.546e+02 1.777e+02 2.233e+02 3.779e+02, threshold=3.553e+02, percent-clipped=2.0 2023-03-27 04:22:07,794 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130565.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:22:37,460 INFO [finetune.py:976] (3/7) Epoch 23, batch 4600, loss[loss=0.146, simple_loss=0.2205, pruned_loss=0.03569, over 4753.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2467, pruned_loss=0.05082, over 955068.39 frames. ], batch size: 27, lr: 3.08e-03, grad_scale: 32.0 2023-03-27 04:22:37,567 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130609.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 04:22:47,773 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130626.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:22:51,855 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2783, 2.1202, 1.8827, 2.2525, 2.0855, 2.0882, 2.1050, 2.8857], device='cuda:3'), covar=tensor([0.3596, 0.4276, 0.3149, 0.3482, 0.3532, 0.2428, 0.3453, 0.1536], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0263, 0.0234, 0.0275, 0.0256, 0.0226, 0.0254, 0.0236], 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') 2023-03-27 04:23:10,853 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130656.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:23:17,054 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130658.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 04:23:17,586 INFO [finetune.py:976] (3/7) Epoch 23, batch 4650, loss[loss=0.1661, simple_loss=0.2435, pruned_loss=0.04429, over 4859.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2452, pruned_loss=0.05081, over 954673.50 frames. ], batch size: 47, lr: 3.08e-03, grad_scale: 32.0 2023-03-27 04:23:19,986 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.906e+01 1.515e+02 1.768e+02 2.232e+02 6.495e+02, threshold=3.536e+02, percent-clipped=3.0 2023-03-27 04:23:36,001 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6398, 1.5687, 2.0684, 1.9252, 1.7116, 4.1342, 1.6332, 1.6272], device='cuda:3'), covar=tensor([0.1056, 0.2031, 0.1333, 0.1090, 0.1774, 0.0204, 0.1563, 0.2093], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0081, 0.0073, 0.0076, 0.0091, 0.0081, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 04:23:54,689 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=130704.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:23:55,883 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=130706.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:23:58,157 INFO [finetune.py:976] (3/7) Epoch 23, batch 4700, loss[loss=0.1675, simple_loss=0.2314, pruned_loss=0.05186, over 4708.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2423, pruned_loss=0.0501, over 955669.30 frames. ], batch size: 23, lr: 3.08e-03, grad_scale: 32.0 2023-03-27 04:24:18,049 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130740.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:24:27,207 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-27 04:24:31,370 INFO [finetune.py:976] (3/7) Epoch 23, batch 4750, loss[loss=0.1356, simple_loss=0.2031, pruned_loss=0.03403, over 3980.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2408, pruned_loss=0.04973, over 955348.79 frames. ], batch size: 17, lr: 3.08e-03, grad_scale: 32.0 2023-03-27 04:24:34,232 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.253e+02 1.528e+02 1.803e+02 2.150e+02 3.686e+02, threshold=3.606e+02, percent-clipped=2.0 2023-03-27 04:24:49,883 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130788.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:24:54,004 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130794.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:24:56,217 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-27 04:25:04,659 INFO [finetune.py:976] (3/7) Epoch 23, batch 4800, loss[loss=0.2119, simple_loss=0.2789, pruned_loss=0.07248, over 4912.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2439, pruned_loss=0.05092, over 954392.90 frames. ], batch size: 36, lr: 3.08e-03, grad_scale: 32.0 2023-03-27 04:25:09,491 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2143, 2.1284, 1.6087, 2.2232, 2.1048, 1.8711, 2.5074, 2.2233], device='cuda:3'), covar=tensor([0.1482, 0.2207, 0.3132, 0.2783, 0.2604, 0.1695, 0.3246, 0.1820], device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0189, 0.0236, 0.0255, 0.0249, 0.0206, 0.0215, 0.0202], 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') 2023-03-27 04:25:37,285 INFO [finetune.py:976] (3/7) Epoch 23, batch 4850, loss[loss=0.1532, simple_loss=0.2275, pruned_loss=0.03944, over 4296.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2463, pruned_loss=0.05157, over 952214.97 frames. ], batch size: 65, lr: 3.08e-03, grad_scale: 32.0 2023-03-27 04:25:37,427 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8286, 1.3261, 1.7084, 1.8248, 1.6004, 1.5859, 1.7405, 1.6990], device='cuda:3'), covar=tensor([0.4921, 0.4438, 0.4029, 0.4454, 0.5614, 0.4574, 0.5149, 0.3981], device='cuda:3'), in_proj_covar=tensor([0.0257, 0.0241, 0.0262, 0.0285, 0.0284, 0.0261, 0.0292, 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') 2023-03-27 04:25:40,095 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.117e+02 1.613e+02 1.947e+02 2.336e+02 6.046e+02, threshold=3.894e+02, percent-clipped=4.0 2023-03-27 04:26:15,092 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4106, 1.4611, 1.8935, 1.7061, 1.5991, 3.4979, 1.4068, 1.6009], device='cuda:3'), covar=tensor([0.1022, 0.1732, 0.1062, 0.0948, 0.1612, 0.0227, 0.1435, 0.1804], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0081, 0.0073, 0.0076, 0.0091, 0.0081, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 04:26:15,658 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130904.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 04:26:19,118 INFO [finetune.py:976] (3/7) Epoch 23, batch 4900, loss[loss=0.17, simple_loss=0.2503, pruned_loss=0.04487, over 4899.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2468, pruned_loss=0.05128, over 954072.46 frames. ], batch size: 37, lr: 3.08e-03, grad_scale: 32.0 2023-03-27 04:26:28,433 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130921.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:26:52,302 INFO [finetune.py:976] (3/7) Epoch 23, batch 4950, loss[loss=0.1997, simple_loss=0.2714, pruned_loss=0.06401, over 4909.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2477, pruned_loss=0.05155, over 952171.07 frames. ], batch size: 38, lr: 3.08e-03, grad_scale: 32.0 2023-03-27 04:26:57,593 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.290e+01 1.586e+02 1.789e+02 2.374e+02 3.586e+02, threshold=3.578e+02, percent-clipped=0.0 2023-03-27 04:27:24,435 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.15 vs. limit=5.0 2023-03-27 04:27:25,618 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9477, 1.7350, 1.5838, 2.0172, 2.4355, 2.0818, 1.6896, 1.5497], device='cuda:3'), covar=tensor([0.1957, 0.1850, 0.1801, 0.1517, 0.1469, 0.1082, 0.2120, 0.1817], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0209, 0.0212, 0.0196, 0.0243, 0.0189, 0.0215, 0.0204], 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') 2023-03-27 04:27:36,349 INFO [finetune.py:976] (3/7) Epoch 23, batch 5000, loss[loss=0.1647, simple_loss=0.2401, pruned_loss=0.0446, over 4932.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2459, pruned_loss=0.05074, over 951958.62 frames. ], batch size: 38, lr: 3.08e-03, grad_scale: 32.0 2023-03-27 04:27:57,587 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131040.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:28:09,926 INFO [finetune.py:976] (3/7) Epoch 23, batch 5050, loss[loss=0.1332, simple_loss=0.2099, pruned_loss=0.0283, over 4784.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.243, pruned_loss=0.05013, over 952819.19 frames. ], batch size: 26, lr: 3.08e-03, grad_scale: 32.0 2023-03-27 04:28:12,372 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.026e+02 1.381e+02 1.770e+02 2.059e+02 4.416e+02, threshold=3.539e+02, percent-clipped=4.0 2023-03-27 04:28:41,480 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=131088.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:28:41,508 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131088.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:28:45,137 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131094.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:28:57,923 INFO [finetune.py:976] (3/7) Epoch 23, batch 5100, loss[loss=0.1274, simple_loss=0.2045, pruned_loss=0.02514, over 3559.00 frames. ], tot_loss[loss=0.169, simple_loss=0.24, pruned_loss=0.04903, over 952860.99 frames. ], batch size: 15, lr: 3.08e-03, grad_scale: 32.0 2023-03-27 04:29:17,262 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=131136.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:29:20,883 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=131142.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:29:21,812 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-27 04:29:31,085 INFO [finetune.py:976] (3/7) Epoch 23, batch 5150, loss[loss=0.149, simple_loss=0.2144, pruned_loss=0.0418, over 4565.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2406, pruned_loss=0.04964, over 954233.74 frames. ], batch size: 20, lr: 3.08e-03, grad_scale: 32.0 2023-03-27 04:29:34,465 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.214e+01 1.572e+02 1.903e+02 2.241e+02 4.010e+02, threshold=3.805e+02, percent-clipped=1.0 2023-03-27 04:29:35,161 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3474, 1.5292, 0.8887, 2.2015, 2.6241, 1.9508, 2.1469, 2.0091], device='cuda:3'), covar=tensor([0.1260, 0.1931, 0.1925, 0.1057, 0.1589, 0.1670, 0.1263, 0.1877], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0094, 0.0110, 0.0093, 0.0119, 0.0094, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 04:29:41,821 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131174.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:30:01,340 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131204.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 04:30:04,269 INFO [finetune.py:976] (3/7) Epoch 23, batch 5200, loss[loss=0.1606, simple_loss=0.2349, pruned_loss=0.04313, over 4857.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2441, pruned_loss=0.05082, over 952403.60 frames. ], batch size: 44, lr: 3.08e-03, grad_scale: 32.0 2023-03-27 04:30:12,621 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131221.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:30:22,997 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131235.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 04:30:33,243 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=131252.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:30:33,902 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6109, 1.4812, 2.1998, 3.5426, 2.2574, 2.4093, 1.3964, 2.8523], device='cuda:3'), covar=tensor([0.1871, 0.1511, 0.1328, 0.0520, 0.0858, 0.1653, 0.1652, 0.0484], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0116, 0.0133, 0.0163, 0.0100, 0.0136, 0.0124, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 04:30:37,393 INFO [finetune.py:976] (3/7) Epoch 23, batch 5250, loss[loss=0.1772, simple_loss=0.2516, pruned_loss=0.05145, over 4907.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2462, pruned_loss=0.05175, over 950997.60 frames. ], batch size: 42, lr: 3.08e-03, grad_scale: 16.0 2023-03-27 04:30:40,886 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.130e+02 1.531e+02 1.792e+02 2.239e+02 3.281e+02, threshold=3.585e+02, percent-clipped=0.0 2023-03-27 04:30:41,185 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-03-27 04:30:44,385 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=131269.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:31:21,563 INFO [finetune.py:976] (3/7) Epoch 23, batch 5300, loss[loss=0.1288, simple_loss=0.1944, pruned_loss=0.03158, over 4388.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2461, pruned_loss=0.05169, over 949991.48 frames. ], batch size: 19, lr: 3.08e-03, grad_scale: 16.0 2023-03-27 04:31:37,675 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2132, 1.4807, 0.8072, 2.0518, 2.4991, 1.8743, 1.9517, 1.9334], device='cuda:3'), covar=tensor([0.1337, 0.1993, 0.2018, 0.1137, 0.1726, 0.1749, 0.1257, 0.1881], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0094, 0.0110, 0.0092, 0.0119, 0.0093, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 04:31:40,152 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.5169, 3.1464, 2.9093, 1.5487, 2.9301, 2.4686, 2.4288, 2.8055], device='cuda:3'), covar=tensor([0.0761, 0.0746, 0.1601, 0.2125, 0.1552, 0.1961, 0.1816, 0.1118], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0192, 0.0201, 0.0182, 0.0210, 0.0210, 0.0224, 0.0197], 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') 2023-03-27 04:31:42,002 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-03-27 04:31:47,370 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-03-27 04:31:54,360 INFO [finetune.py:976] (3/7) Epoch 23, batch 5350, loss[loss=0.134, simple_loss=0.2042, pruned_loss=0.03191, over 4783.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2458, pruned_loss=0.05141, over 952008.29 frames. ], batch size: 25, lr: 3.08e-03, grad_scale: 16.0 2023-03-27 04:31:57,384 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.087e+02 1.529e+02 1.830e+02 2.196e+02 3.219e+02, threshold=3.659e+02, percent-clipped=0.0 2023-03-27 04:32:15,434 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6710, 1.6714, 1.4529, 1.7552, 2.0035, 1.9957, 1.6984, 1.4662], device='cuda:3'), covar=tensor([0.0371, 0.0326, 0.0673, 0.0274, 0.0243, 0.0484, 0.0318, 0.0464], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0106, 0.0144, 0.0111, 0.0100, 0.0111, 0.0101, 0.0112], device='cuda:3'), out_proj_covar=tensor([7.7094e-05, 8.1319e-05, 1.1253e-04, 8.4962e-05, 7.7387e-05, 8.2108e-05, 7.5103e-05, 8.5255e-05], device='cuda:3') 2023-03-27 04:32:32,768 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2469, 1.9685, 1.4929, 0.5418, 1.7191, 1.9164, 1.7664, 1.8662], device='cuda:3'), covar=tensor([0.1058, 0.0868, 0.1762, 0.2212, 0.1427, 0.2535, 0.2422, 0.0955], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0192, 0.0200, 0.0181, 0.0209, 0.0209, 0.0223, 0.0196], 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') 2023-03-27 04:32:38,097 INFO [finetune.py:976] (3/7) Epoch 23, batch 5400, loss[loss=0.1846, simple_loss=0.2499, pruned_loss=0.0596, over 4736.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2452, pruned_loss=0.05183, over 952704.19 frames. ], batch size: 59, lr: 3.08e-03, grad_scale: 16.0 2023-03-27 04:32:38,213 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131409.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:32:42,430 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131416.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:32:46,608 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8008, 2.4249, 2.2096, 1.0773, 2.3368, 2.1215, 1.9812, 2.3695], device='cuda:3'), covar=tensor([0.0741, 0.0855, 0.1628, 0.2138, 0.1473, 0.1960, 0.1989, 0.0924], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0193, 0.0201, 0.0183, 0.0210, 0.0210, 0.0225, 0.0197], 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') 2023-03-27 04:32:47,433 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-27 04:33:11,755 INFO [finetune.py:976] (3/7) Epoch 23, batch 5450, loss[loss=0.1884, simple_loss=0.2409, pruned_loss=0.06794, over 4898.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2432, pruned_loss=0.05114, over 955208.71 frames. ], batch size: 35, lr: 3.08e-03, grad_scale: 16.0 2023-03-27 04:33:14,785 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.094e+02 1.514e+02 1.875e+02 2.409e+02 5.439e+02, threshold=3.749e+02, percent-clipped=4.0 2023-03-27 04:33:18,556 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131470.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:33:23,305 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131477.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:33:33,348 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131492.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:33:51,830 INFO [finetune.py:976] (3/7) Epoch 23, batch 5500, loss[loss=0.1347, simple_loss=0.2093, pruned_loss=0.03008, over 4750.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2396, pruned_loss=0.04982, over 954462.20 frames. ], batch size: 54, lr: 3.08e-03, grad_scale: 16.0 2023-03-27 04:34:12,687 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131530.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 04:34:29,168 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131553.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:34:33,174 INFO [finetune.py:976] (3/7) Epoch 23, batch 5550, loss[loss=0.2001, simple_loss=0.2797, pruned_loss=0.06025, over 4922.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2401, pruned_loss=0.04985, over 956602.50 frames. ], batch size: 42, lr: 3.08e-03, grad_scale: 16.0 2023-03-27 04:34:36,709 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.823e+01 1.422e+02 1.728e+02 2.186e+02 5.215e+02, threshold=3.457e+02, percent-clipped=2.0 2023-03-27 04:34:39,231 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0366, 1.9537, 2.2828, 1.5432, 2.1388, 2.3310, 1.6810, 2.4579], device='cuda:3'), covar=tensor([0.1360, 0.1885, 0.1581, 0.2018, 0.0903, 0.1489, 0.2722, 0.0846], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0205, 0.0191, 0.0189, 0.0172, 0.0213, 0.0214, 0.0197], 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') 2023-03-27 04:35:04,716 INFO [finetune.py:976] (3/7) Epoch 23, batch 5600, loss[loss=0.1748, simple_loss=0.2592, pruned_loss=0.04525, over 4871.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2419, pruned_loss=0.04965, over 954623.81 frames. ], batch size: 34, lr: 3.07e-03, grad_scale: 16.0 2023-03-27 04:35:13,322 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.4589, 3.9193, 4.1244, 4.3434, 4.2135, 3.9757, 4.5930, 1.4693], device='cuda:3'), covar=tensor([0.0807, 0.0895, 0.0893, 0.1022, 0.1147, 0.1557, 0.0636, 0.5643], device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0247, 0.0280, 0.0293, 0.0338, 0.0287, 0.0304, 0.0300], 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') 2023-03-27 04:35:34,625 INFO [finetune.py:976] (3/7) Epoch 23, batch 5650, loss[loss=0.1944, simple_loss=0.2763, pruned_loss=0.05624, over 4927.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2467, pruned_loss=0.05126, over 953231.30 frames. ], batch size: 42, lr: 3.07e-03, grad_scale: 16.0 2023-03-27 04:35:37,860 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.955e+01 1.494e+02 1.801e+02 2.339e+02 4.576e+02, threshold=3.601e+02, percent-clipped=4.0 2023-03-27 04:35:55,938 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-03-27 04:36:00,480 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0008, 1.8742, 2.2560, 1.6023, 1.9696, 2.2969, 1.8274, 2.3215], device='cuda:3'), covar=tensor([0.0998, 0.1748, 0.1117, 0.1485, 0.0826, 0.1014, 0.2407, 0.0749], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0206, 0.0191, 0.0190, 0.0172, 0.0213, 0.0215, 0.0198], 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') 2023-03-27 04:36:04,505 INFO [finetune.py:976] (3/7) Epoch 23, batch 5700, loss[loss=0.1531, simple_loss=0.2122, pruned_loss=0.04701, over 4244.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2444, pruned_loss=0.05053, over 938464.92 frames. ], batch size: 18, lr: 3.07e-03, grad_scale: 16.0 2023-03-27 04:36:40,049 INFO [finetune.py:976] (3/7) Epoch 24, batch 0, loss[loss=0.1394, simple_loss=0.2144, pruned_loss=0.03214, over 4744.00 frames. ], tot_loss[loss=0.1394, simple_loss=0.2144, pruned_loss=0.03214, over 4744.00 frames. ], batch size: 23, lr: 3.07e-03, grad_scale: 16.0 2023-03-27 04:36:40,049 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-27 04:36:49,687 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6685, 1.5913, 1.5429, 1.5897, 1.0602, 2.9249, 1.1945, 1.5631], device='cuda:3'), covar=tensor([0.3195, 0.2344, 0.2061, 0.2198, 0.1733, 0.0249, 0.2472, 0.1222], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0116, 0.0121, 0.0123, 0.0113, 0.0096, 0.0095, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 04:36:50,763 INFO [finetune.py:1010] (3/7) Epoch 24, validation: loss=0.1594, simple_loss=0.227, pruned_loss=0.04592, over 2265189.00 frames. 2023-03-27 04:36:50,763 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6469MB 2023-03-27 04:36:52,443 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7315, 1.1568, 0.8656, 1.5439, 2.0544, 1.3040, 1.4617, 1.6047], device='cuda:3'), covar=tensor([0.1360, 0.1971, 0.1805, 0.1147, 0.1859, 0.1980, 0.1349, 0.1736], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0094, 0.0110, 0.0092, 0.0119, 0.0094, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 04:36:55,490 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4898, 1.4053, 1.3444, 1.4225, 0.8294, 2.8944, 1.0501, 1.4123], device='cuda:3'), covar=tensor([0.3512, 0.2665, 0.2369, 0.2617, 0.2113, 0.0261, 0.2749, 0.1401], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0116, 0.0121, 0.0124, 0.0114, 0.0096, 0.0095, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 04:37:07,463 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 7.206e+01 1.398e+02 1.674e+02 2.004e+02 3.219e+02, threshold=3.348e+02, percent-clipped=0.0 2023-03-27 04:37:08,163 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131765.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:37:12,920 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131772.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:37:25,289 INFO [finetune.py:976] (3/7) Epoch 24, batch 50, loss[loss=0.1893, simple_loss=0.263, pruned_loss=0.05776, over 4878.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2472, pruned_loss=0.05086, over 217604.23 frames. ], batch size: 32, lr: 3.07e-03, grad_scale: 16.0 2023-03-27 04:38:02,599 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131830.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:38:07,276 INFO [finetune.py:976] (3/7) Epoch 24, batch 100, loss[loss=0.177, simple_loss=0.2421, pruned_loss=0.05595, over 4915.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2423, pruned_loss=0.05105, over 381517.53 frames. ], batch size: 46, lr: 3.07e-03, grad_scale: 16.0 2023-03-27 04:38:15,492 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131848.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:38:20,989 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131857.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:38:25,098 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.662e+01 1.465e+02 1.761e+02 2.142e+02 3.724e+02, threshold=3.523e+02, percent-clipped=1.0 2023-03-27 04:38:34,587 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=131878.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:38:40,501 INFO [finetune.py:976] (3/7) Epoch 24, batch 150, loss[loss=0.1272, simple_loss=0.1971, pruned_loss=0.02866, over 4782.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2356, pruned_loss=0.04765, over 508967.79 frames. ], batch size: 28, lr: 3.07e-03, grad_scale: 16.0 2023-03-27 04:39:10,819 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131918.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:39:29,546 INFO [finetune.py:976] (3/7) Epoch 24, batch 200, loss[loss=0.1854, simple_loss=0.2463, pruned_loss=0.0622, over 4825.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2354, pruned_loss=0.04813, over 608882.46 frames. ], batch size: 33, lr: 3.07e-03, grad_scale: 16.0 2023-03-27 04:39:51,182 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.027e+02 1.517e+02 1.799e+02 2.123e+02 6.232e+02, threshold=3.598e+02, percent-clipped=3.0 2023-03-27 04:40:00,906 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0069, 1.9458, 1.6235, 2.0447, 2.5787, 2.1161, 1.9982, 1.5406], device='cuda:3'), covar=tensor([0.2203, 0.1911, 0.1875, 0.1605, 0.1743, 0.1150, 0.1991, 0.1848], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0208, 0.0212, 0.0194, 0.0242, 0.0188, 0.0214, 0.0204], 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') 2023-03-27 04:40:06,649 INFO [finetune.py:976] (3/7) Epoch 24, batch 250, loss[loss=0.2094, simple_loss=0.2674, pruned_loss=0.07568, over 4832.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2412, pruned_loss=0.05068, over 683365.71 frames. ], batch size: 30, lr: 3.07e-03, grad_scale: 16.0 2023-03-27 04:40:13,593 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.61 vs. limit=5.0 2023-03-27 04:40:36,947 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132031.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:40:41,437 INFO [finetune.py:976] (3/7) Epoch 24, batch 300, loss[loss=0.1822, simple_loss=0.2638, pruned_loss=0.05025, over 4773.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2457, pruned_loss=0.05156, over 743265.33 frames. ], batch size: 28, lr: 3.07e-03, grad_scale: 16.0 2023-03-27 04:40:53,231 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132054.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:40:59,162 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.071e+02 1.635e+02 1.887e+02 2.261e+02 6.512e+02, threshold=3.774e+02, percent-clipped=2.0 2023-03-27 04:40:59,873 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132065.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:41:04,193 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132072.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:41:14,119 INFO [finetune.py:976] (3/7) Epoch 24, batch 350, loss[loss=0.1352, simple_loss=0.2086, pruned_loss=0.03093, over 4711.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2475, pruned_loss=0.05194, over 789187.88 frames. ], batch size: 23, lr: 3.07e-03, grad_scale: 16.0 2023-03-27 04:41:17,747 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132092.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:41:33,450 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=132113.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:41:34,739 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132115.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 04:41:42,165 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=132120.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:41:56,062 INFO [finetune.py:976] (3/7) Epoch 24, batch 400, loss[loss=0.1891, simple_loss=0.2538, pruned_loss=0.06222, over 4879.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2492, pruned_loss=0.05267, over 823586.16 frames. ], batch size: 32, lr: 3.07e-03, grad_scale: 16.0 2023-03-27 04:42:03,884 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132148.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:42:15,409 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.028e+02 1.594e+02 1.883e+02 2.349e+02 4.739e+02, threshold=3.766e+02, percent-clipped=1.0 2023-03-27 04:42:29,842 INFO [finetune.py:976] (3/7) Epoch 24, batch 450, loss[loss=0.1441, simple_loss=0.2149, pruned_loss=0.03661, over 4854.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2476, pruned_loss=0.05201, over 853633.68 frames. ], batch size: 44, lr: 3.07e-03, grad_scale: 16.0 2023-03-27 04:42:36,329 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=132196.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:42:55,015 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132213.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:43:13,241 INFO [finetune.py:976] (3/7) Epoch 24, batch 500, loss[loss=0.1782, simple_loss=0.2438, pruned_loss=0.05633, over 4221.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2455, pruned_loss=0.05138, over 875376.41 frames. ], batch size: 18, lr: 3.07e-03, grad_scale: 16.0 2023-03-27 04:43:32,460 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 7.902e+01 1.519e+02 1.809e+02 2.135e+02 3.897e+02, threshold=3.617e+02, percent-clipped=1.0 2023-03-27 04:43:46,923 INFO [finetune.py:976] (3/7) Epoch 24, batch 550, loss[loss=0.1399, simple_loss=0.2064, pruned_loss=0.03673, over 4741.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2425, pruned_loss=0.05043, over 891030.39 frames. ], batch size: 23, lr: 3.07e-03, grad_scale: 16.0 2023-03-27 04:44:28,679 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.38 vs. limit=5.0 2023-03-27 04:44:30,204 INFO [finetune.py:976] (3/7) Epoch 24, batch 600, loss[loss=0.2109, simple_loss=0.2665, pruned_loss=0.0777, over 4761.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2439, pruned_loss=0.05113, over 903321.70 frames. ], batch size: 27, lr: 3.07e-03, grad_scale: 16.0 2023-03-27 04:44:58,206 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.737e+01 1.576e+02 1.859e+02 2.330e+02 3.343e+02, threshold=3.718e+02, percent-clipped=0.0 2023-03-27 04:45:12,592 INFO [finetune.py:976] (3/7) Epoch 24, batch 650, loss[loss=0.2037, simple_loss=0.2842, pruned_loss=0.06162, over 4897.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2464, pruned_loss=0.05157, over 916068.78 frames. ], batch size: 43, lr: 3.07e-03, grad_scale: 16.0 2023-03-27 04:45:12,659 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132387.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:45:28,118 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132410.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 04:45:46,139 INFO [finetune.py:976] (3/7) Epoch 24, batch 700, loss[loss=0.1804, simple_loss=0.2472, pruned_loss=0.05684, over 4849.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2479, pruned_loss=0.05214, over 925156.63 frames. ], batch size: 44, lr: 3.07e-03, grad_scale: 16.0 2023-03-27 04:46:03,856 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.085e+02 1.592e+02 1.911e+02 2.262e+02 4.191e+02, threshold=3.822e+02, percent-clipped=2.0 2023-03-27 04:46:19,328 INFO [finetune.py:976] (3/7) Epoch 24, batch 750, loss[loss=0.1578, simple_loss=0.2356, pruned_loss=0.04004, over 4747.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2491, pruned_loss=0.0525, over 931977.18 frames. ], batch size: 26, lr: 3.06e-03, grad_scale: 16.0 2023-03-27 04:46:36,506 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132513.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:46:58,682 INFO [finetune.py:976] (3/7) Epoch 24, batch 800, loss[loss=0.2041, simple_loss=0.265, pruned_loss=0.07161, over 4874.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2468, pruned_loss=0.05132, over 936357.24 frames. ], batch size: 35, lr: 3.06e-03, grad_scale: 16.0 2023-03-27 04:47:17,153 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=132561.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:47:19,958 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.775e+01 1.480e+02 1.729e+02 2.075e+02 4.531e+02, threshold=3.459e+02, percent-clipped=1.0 2023-03-27 04:47:35,955 INFO [finetune.py:976] (3/7) Epoch 24, batch 850, loss[loss=0.1688, simple_loss=0.2432, pruned_loss=0.04721, over 4890.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2452, pruned_loss=0.05067, over 939652.69 frames. ], batch size: 32, lr: 3.06e-03, grad_scale: 16.0 2023-03-27 04:48:18,613 INFO [finetune.py:976] (3/7) Epoch 24, batch 900, loss[loss=0.1843, simple_loss=0.2603, pruned_loss=0.05417, over 4750.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2422, pruned_loss=0.04985, over 941693.73 frames. ], batch size: 23, lr: 3.06e-03, grad_scale: 16.0 2023-03-27 04:48:33,131 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132660.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:48:35,407 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.013e+02 1.417e+02 1.718e+02 2.002e+02 3.598e+02, threshold=3.436e+02, percent-clipped=1.0 2023-03-27 04:48:52,530 INFO [finetune.py:976] (3/7) Epoch 24, batch 950, loss[loss=0.2056, simple_loss=0.2782, pruned_loss=0.06647, over 4825.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2412, pruned_loss=0.04983, over 942104.22 frames. ], batch size: 39, lr: 3.06e-03, grad_scale: 16.0 2023-03-27 04:48:52,615 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132687.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:49:07,060 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132710.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 04:49:08,297 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9330, 1.7178, 2.3264, 1.6322, 2.1624, 2.3164, 1.6103, 2.4108], device='cuda:3'), covar=tensor([0.1272, 0.2154, 0.1276, 0.1780, 0.0861, 0.1224, 0.2816, 0.0768], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0207, 0.0191, 0.0191, 0.0172, 0.0214, 0.0216, 0.0199], 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') 2023-03-27 04:49:14,297 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132721.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:49:26,439 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=132735.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:49:27,540 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8087, 4.6235, 4.3777, 2.3654, 4.7283, 3.7147, 0.7740, 3.0579], device='cuda:3'), covar=tensor([0.2605, 0.2219, 0.1428, 0.3464, 0.0756, 0.0852, 0.5212, 0.1617], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0179, 0.0162, 0.0129, 0.0161, 0.0124, 0.0148, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 04:49:28,059 INFO [finetune.py:976] (3/7) Epoch 24, batch 1000, loss[loss=0.1764, simple_loss=0.245, pruned_loss=0.05386, over 4930.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2429, pruned_loss=0.05063, over 946100.80 frames. ], batch size: 37, lr: 3.06e-03, grad_scale: 16.0 2023-03-27 04:49:38,695 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132746.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:49:50,822 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=132758.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:49:54,880 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.132e+02 1.613e+02 1.803e+02 2.355e+02 4.590e+02, threshold=3.605e+02, percent-clipped=3.0 2023-03-27 04:50:17,565 INFO [finetune.py:976] (3/7) Epoch 24, batch 1050, loss[loss=0.1833, simple_loss=0.2556, pruned_loss=0.05546, over 4741.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2447, pruned_loss=0.05067, over 948674.89 frames. ], batch size: 59, lr: 3.06e-03, grad_scale: 16.0 2023-03-27 04:50:31,463 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132807.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:50:51,430 INFO [finetune.py:976] (3/7) Epoch 24, batch 1100, loss[loss=0.1642, simple_loss=0.2475, pruned_loss=0.04046, over 4885.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2463, pruned_loss=0.05089, over 951463.43 frames. ], batch size: 43, lr: 3.06e-03, grad_scale: 16.0 2023-03-27 04:51:08,747 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.005e+02 1.623e+02 1.902e+02 2.304e+02 4.937e+02, threshold=3.804e+02, percent-clipped=2.0 2023-03-27 04:51:24,186 INFO [finetune.py:976] (3/7) Epoch 24, batch 1150, loss[loss=0.1785, simple_loss=0.2567, pruned_loss=0.05017, over 4857.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2467, pruned_loss=0.05021, over 953227.06 frames. ], batch size: 44, lr: 3.06e-03, grad_scale: 16.0 2023-03-27 04:51:57,326 INFO [finetune.py:976] (3/7) Epoch 24, batch 1200, loss[loss=0.1504, simple_loss=0.2101, pruned_loss=0.04532, over 4303.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2451, pruned_loss=0.04939, over 953380.86 frames. ], batch size: 19, lr: 3.06e-03, grad_scale: 16.0 2023-03-27 04:52:24,199 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.5448, 3.9858, 4.1298, 4.3985, 4.3206, 3.9149, 4.5778, 1.4247], device='cuda:3'), covar=tensor([0.0696, 0.0801, 0.0926, 0.0915, 0.1019, 0.1743, 0.0621, 0.5940], device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0248, 0.0282, 0.0294, 0.0340, 0.0288, 0.0307, 0.0302], 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') 2023-03-27 04:52:24,717 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.055e+01 1.470e+02 1.716e+02 2.148e+02 3.548e+02, threshold=3.432e+02, percent-clipped=0.0 2023-03-27 04:52:28,525 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1252, 2.0338, 1.6177, 2.0239, 2.1046, 1.8038, 2.4274, 2.1030], device='cuda:3'), covar=tensor([0.1396, 0.2040, 0.3072, 0.2457, 0.2386, 0.1621, 0.2826, 0.1786], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0188, 0.0233, 0.0251, 0.0246, 0.0204, 0.0212, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 04:52:40,281 INFO [finetune.py:976] (3/7) Epoch 24, batch 1250, loss[loss=0.1511, simple_loss=0.2199, pruned_loss=0.04114, over 4819.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2425, pruned_loss=0.04881, over 952876.27 frames. ], batch size: 30, lr: 3.06e-03, grad_scale: 16.0 2023-03-27 04:52:41,302 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.08 vs. limit=5.0 2023-03-27 04:52:59,659 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133016.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:53:04,045 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5907, 2.3961, 1.9063, 2.6052, 2.4320, 2.1155, 3.0001, 2.5616], device='cuda:3'), covar=tensor([0.1250, 0.2315, 0.3097, 0.2701, 0.2662, 0.1736, 0.3295, 0.1823], device='cuda:3'), in_proj_covar=tensor([0.0186, 0.0187, 0.0233, 0.0250, 0.0246, 0.0203, 0.0211, 0.0200], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 04:53:15,460 INFO [finetune.py:976] (3/7) Epoch 24, batch 1300, loss[loss=0.1523, simple_loss=0.2263, pruned_loss=0.03915, over 4774.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2401, pruned_loss=0.04834, over 951874.50 frames. ], batch size: 27, lr: 3.06e-03, grad_scale: 16.0 2023-03-27 04:53:42,214 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.345e+01 1.496e+02 1.852e+02 2.141e+02 4.041e+02, threshold=3.705e+02, percent-clipped=2.0 2023-03-27 04:53:57,208 INFO [finetune.py:976] (3/7) Epoch 24, batch 1350, loss[loss=0.1793, simple_loss=0.2489, pruned_loss=0.05483, over 4814.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2413, pruned_loss=0.04952, over 953346.97 frames. ], batch size: 51, lr: 3.06e-03, grad_scale: 16.0 2023-03-27 04:54:07,946 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133102.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:54:18,374 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.53 vs. limit=5.0 2023-03-27 04:54:31,059 INFO [finetune.py:976] (3/7) Epoch 24, batch 1400, loss[loss=0.1618, simple_loss=0.2482, pruned_loss=0.03765, over 4828.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2424, pruned_loss=0.04931, over 951875.21 frames. ], batch size: 33, lr: 3.06e-03, grad_scale: 16.0 2023-03-27 04:54:35,888 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5694, 1.4527, 1.4646, 1.4308, 1.2662, 3.0424, 1.2349, 1.6167], device='cuda:3'), covar=tensor([0.4097, 0.3301, 0.2488, 0.3150, 0.1714, 0.0343, 0.2519, 0.1322], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0115, 0.0121, 0.0123, 0.0113, 0.0096, 0.0094, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0005, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 04:54:59,478 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.355e+01 1.624e+02 1.914e+02 2.315e+02 3.947e+02, threshold=3.828e+02, percent-clipped=1.0 2023-03-27 04:55:19,623 INFO [finetune.py:976] (3/7) Epoch 24, batch 1450, loss[loss=0.1838, simple_loss=0.2659, pruned_loss=0.05079, over 4855.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2445, pruned_loss=0.05016, over 952066.68 frames. ], batch size: 44, lr: 3.06e-03, grad_scale: 16.0 2023-03-27 04:55:35,367 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5546, 2.2586, 3.0095, 1.8793, 2.4897, 2.9893, 2.1809, 2.9544], device='cuda:3'), covar=tensor([0.1215, 0.1936, 0.1369, 0.2130, 0.1012, 0.1269, 0.2474, 0.0860], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0208, 0.0192, 0.0190, 0.0173, 0.0214, 0.0216, 0.0200], 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') 2023-03-27 04:55:56,686 INFO [finetune.py:976] (3/7) Epoch 24, batch 1500, loss[loss=0.1975, simple_loss=0.2392, pruned_loss=0.07794, over 3978.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2458, pruned_loss=0.05063, over 951970.47 frames. ], batch size: 17, lr: 3.06e-03, grad_scale: 16.0 2023-03-27 04:55:56,776 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1278, 3.6352, 3.7786, 4.0354, 3.8763, 3.6571, 4.2300, 1.3984], device='cuda:3'), covar=tensor([0.0802, 0.0863, 0.0955, 0.0863, 0.1220, 0.1618, 0.0772, 0.5435], device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0247, 0.0281, 0.0293, 0.0338, 0.0286, 0.0306, 0.0300], 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') 2023-03-27 04:55:57,034 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.72 vs. limit=5.0 2023-03-27 04:56:15,023 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.176e+02 1.579e+02 1.864e+02 2.355e+02 5.095e+02, threshold=3.727e+02, percent-clipped=2.0 2023-03-27 04:56:24,034 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6952, 2.4959, 2.0382, 2.8994, 2.6354, 2.2243, 3.1108, 2.6392], device='cuda:3'), covar=tensor([0.1345, 0.2302, 0.3117, 0.2350, 0.2494, 0.1775, 0.2954, 0.1865], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0188, 0.0235, 0.0251, 0.0247, 0.0204, 0.0213, 0.0201], 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') 2023-03-27 04:56:30,467 INFO [finetune.py:976] (3/7) Epoch 24, batch 1550, loss[loss=0.1833, simple_loss=0.254, pruned_loss=0.05627, over 4781.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2463, pruned_loss=0.05051, over 953077.40 frames. ], batch size: 51, lr: 3.06e-03, grad_scale: 32.0 2023-03-27 04:56:50,654 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133316.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:57:04,264 INFO [finetune.py:976] (3/7) Epoch 24, batch 1600, loss[loss=0.2038, simple_loss=0.2757, pruned_loss=0.06599, over 4825.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2458, pruned_loss=0.05088, over 952421.60 frames. ], batch size: 33, lr: 3.06e-03, grad_scale: 32.0 2023-03-27 04:57:28,473 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.062e+02 1.451e+02 1.796e+02 2.043e+02 3.402e+02, threshold=3.593e+02, percent-clipped=0.0 2023-03-27 04:57:28,563 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=133364.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:57:36,402 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3266, 2.0807, 2.2393, 0.9835, 2.5252, 2.7978, 2.4016, 2.0711], device='cuda:3'), covar=tensor([0.0915, 0.0699, 0.0508, 0.0728, 0.0656, 0.0560, 0.0402, 0.0762], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0148, 0.0127, 0.0122, 0.0131, 0.0130, 0.0141, 0.0148], device='cuda:3'), out_proj_covar=tensor([8.8879e-05, 1.0682e-04, 9.0526e-05, 8.5786e-05, 9.2008e-05, 9.2299e-05, 1.0044e-04, 1.0582e-04], device='cuda:3') 2023-03-27 04:57:38,828 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133374.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 04:57:39,506 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.77 vs. limit=5.0 2023-03-27 04:57:40,056 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133376.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:57:46,649 INFO [finetune.py:976] (3/7) Epoch 24, batch 1650, loss[loss=0.1297, simple_loss=0.2098, pruned_loss=0.02478, over 4865.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2426, pruned_loss=0.04977, over 952396.75 frames. ], batch size: 31, lr: 3.06e-03, grad_scale: 32.0 2023-03-27 04:57:52,072 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2338, 2.2382, 2.3569, 1.6103, 2.2277, 2.4301, 2.4545, 2.0248], device='cuda:3'), covar=tensor([0.0637, 0.0678, 0.0727, 0.0931, 0.0741, 0.0697, 0.0618, 0.1041], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0136, 0.0139, 0.0119, 0.0127, 0.0138, 0.0138, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 04:57:56,872 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133402.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:58:19,426 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133435.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 04:58:20,504 INFO [finetune.py:976] (3/7) Epoch 24, batch 1700, loss[loss=0.2302, simple_loss=0.292, pruned_loss=0.08419, over 4817.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2395, pruned_loss=0.04893, over 952508.67 frames. ], batch size: 41, lr: 3.06e-03, grad_scale: 32.0 2023-03-27 04:58:20,621 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133437.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:58:21,795 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0526, 1.8122, 2.3609, 4.2895, 2.8884, 2.8751, 1.2165, 3.5566], device='cuda:3'), covar=tensor([0.1815, 0.1551, 0.1578, 0.0413, 0.0753, 0.1470, 0.1998, 0.0366], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0132, 0.0162, 0.0100, 0.0136, 0.0124, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 04:58:31,232 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=133450.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:58:48,614 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.499e+01 1.457e+02 1.770e+02 2.219e+02 3.253e+02, threshold=3.541e+02, percent-clipped=0.0 2023-03-27 04:58:54,593 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.2407, 4.5626, 4.7770, 5.0315, 4.9853, 4.7484, 5.3594, 1.7587], device='cuda:3'), covar=tensor([0.0733, 0.0893, 0.0776, 0.0888, 0.1167, 0.1550, 0.0518, 0.5944], device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0248, 0.0282, 0.0293, 0.0338, 0.0287, 0.0307, 0.0302], 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') 2023-03-27 04:58:55,845 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133474.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:59:04,128 INFO [finetune.py:976] (3/7) Epoch 24, batch 1750, loss[loss=0.2082, simple_loss=0.2869, pruned_loss=0.06475, over 4734.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2438, pruned_loss=0.051, over 952483.48 frames. ], batch size: 59, lr: 3.06e-03, grad_scale: 32.0 2023-03-27 04:59:24,969 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9734, 1.3868, 0.7986, 1.8567, 2.3476, 1.7648, 1.6360, 1.7693], device='cuda:3'), covar=tensor([0.1426, 0.1927, 0.2059, 0.1135, 0.1909, 0.1980, 0.1352, 0.1885], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0093, 0.0108, 0.0091, 0.0118, 0.0093, 0.0097, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-27 04:59:36,818 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133535.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 04:59:37,911 INFO [finetune.py:976] (3/7) Epoch 24, batch 1800, loss[loss=0.2457, simple_loss=0.3098, pruned_loss=0.0908, over 4757.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2469, pruned_loss=0.05189, over 953990.69 frames. ], batch size: 59, lr: 3.06e-03, grad_scale: 32.0 2023-03-27 04:59:57,744 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.053e+02 1.575e+02 1.839e+02 2.282e+02 3.463e+02, threshold=3.677e+02, percent-clipped=0.0 2023-03-27 05:00:09,637 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0375, 1.3558, 1.9091, 1.9590, 1.7806, 1.7734, 1.8810, 1.8996], device='cuda:3'), covar=tensor([0.4334, 0.4188, 0.4035, 0.3976, 0.5363, 0.4468, 0.4940, 0.3664], device='cuda:3'), in_proj_covar=tensor([0.0259, 0.0244, 0.0263, 0.0287, 0.0287, 0.0263, 0.0295, 0.0248], 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') 2023-03-27 05:00:23,456 INFO [finetune.py:976] (3/7) Epoch 24, batch 1850, loss[loss=0.1826, simple_loss=0.2577, pruned_loss=0.05376, over 4830.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2476, pruned_loss=0.05148, over 954000.69 frames. ], batch size: 47, lr: 3.06e-03, grad_scale: 32.0 2023-03-27 05:00:52,139 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-27 05:00:56,576 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.3634, 3.7647, 4.0155, 4.1689, 4.1289, 3.8489, 4.4478, 1.3638], device='cuda:3'), covar=tensor([0.0708, 0.0863, 0.0834, 0.0933, 0.1095, 0.1549, 0.0705, 0.5672], device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0247, 0.0280, 0.0292, 0.0335, 0.0286, 0.0305, 0.0300], 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') 2023-03-27 05:00:59,960 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.8834, 3.4522, 3.7310, 3.5707, 3.4481, 3.4213, 4.0425, 1.2430], device='cuda:3'), covar=tensor([0.1260, 0.1701, 0.1410, 0.1886, 0.2233, 0.2506, 0.1482, 0.7671], device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0246, 0.0280, 0.0292, 0.0335, 0.0285, 0.0305, 0.0300], 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') 2023-03-27 05:01:04,055 INFO [finetune.py:976] (3/7) Epoch 24, batch 1900, loss[loss=0.1516, simple_loss=0.2311, pruned_loss=0.03606, over 4788.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2491, pruned_loss=0.05249, over 952125.80 frames. ], batch size: 29, lr: 3.06e-03, grad_scale: 32.0 2023-03-27 05:01:14,206 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133652.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:01:21,803 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.050e+02 1.540e+02 1.881e+02 2.277e+02 3.366e+02, threshold=3.762e+02, percent-clipped=0.0 2023-03-27 05:01:28,954 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3421, 1.2560, 1.2124, 1.3825, 1.5754, 1.5167, 1.3041, 1.1992], device='cuda:3'), covar=tensor([0.0379, 0.0306, 0.0638, 0.0278, 0.0252, 0.0409, 0.0365, 0.0419], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0107, 0.0145, 0.0112, 0.0100, 0.0112, 0.0102, 0.0113], device='cuda:3'), out_proj_covar=tensor([7.7950e-05, 8.1812e-05, 1.1346e-04, 8.5645e-05, 7.8006e-05, 8.3128e-05, 7.6071e-05, 8.5900e-05], device='cuda:3') 2023-03-27 05:01:35,358 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4861, 1.5041, 1.3381, 1.5770, 1.7835, 1.7087, 1.5105, 1.2796], device='cuda:3'), covar=tensor([0.0382, 0.0302, 0.0595, 0.0270, 0.0246, 0.0443, 0.0296, 0.0425], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0107, 0.0145, 0.0112, 0.0100, 0.0112, 0.0102, 0.0113], device='cuda:3'), out_proj_covar=tensor([7.7816e-05, 8.1684e-05, 1.1325e-04, 8.5518e-05, 7.7891e-05, 8.2988e-05, 7.5920e-05, 8.5748e-05], device='cuda:3') 2023-03-27 05:01:37,659 INFO [finetune.py:976] (3/7) Epoch 24, batch 1950, loss[loss=0.1511, simple_loss=0.2227, pruned_loss=0.03976, over 4928.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2469, pruned_loss=0.05083, over 951141.04 frames. ], batch size: 38, lr: 3.06e-03, grad_scale: 32.0 2023-03-27 05:01:55,029 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133713.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:02:06,299 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133730.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 05:02:07,529 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133732.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:02:11,409 INFO [finetune.py:976] (3/7) Epoch 24, batch 2000, loss[loss=0.1569, simple_loss=0.2258, pruned_loss=0.04399, over 4796.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2435, pruned_loss=0.04965, over 953929.98 frames. ], batch size: 51, lr: 3.06e-03, grad_scale: 32.0 2023-03-27 05:02:28,713 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.381e+01 1.373e+02 1.735e+02 2.243e+02 3.912e+02, threshold=3.469e+02, percent-clipped=2.0 2023-03-27 05:02:54,158 INFO [finetune.py:976] (3/7) Epoch 24, batch 2050, loss[loss=0.1789, simple_loss=0.2438, pruned_loss=0.05694, over 4817.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2405, pruned_loss=0.04867, over 954587.37 frames. ], batch size: 39, lr: 3.06e-03, grad_scale: 32.0 2023-03-27 05:03:23,707 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133830.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:03:27,931 INFO [finetune.py:976] (3/7) Epoch 24, batch 2100, loss[loss=0.1672, simple_loss=0.2366, pruned_loss=0.04887, over 4719.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2404, pruned_loss=0.04907, over 955230.52 frames. ], batch size: 23, lr: 3.06e-03, grad_scale: 32.0 2023-03-27 05:03:34,466 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5827, 1.5154, 1.5149, 1.5346, 1.0014, 3.3756, 1.2415, 1.6948], device='cuda:3'), covar=tensor([0.3237, 0.2417, 0.2063, 0.2317, 0.1865, 0.0210, 0.2763, 0.1304], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0121, 0.0123, 0.0113, 0.0097, 0.0095, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 05:03:47,588 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.827e+01 1.541e+02 1.860e+02 2.293e+02 6.118e+02, threshold=3.720e+02, percent-clipped=3.0 2023-03-27 05:04:07,689 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2511, 1.8818, 2.1563, 1.4348, 1.9934, 2.2237, 2.1729, 1.6901], device='cuda:3'), covar=tensor([0.0532, 0.0744, 0.0728, 0.0931, 0.0770, 0.0670, 0.0616, 0.1176], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0138, 0.0141, 0.0121, 0.0127, 0.0139, 0.0139, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 05:04:10,129 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6689, 1.9776, 0.9432, 2.2815, 2.8746, 2.4243, 2.2308, 2.2911], device='cuda:3'), covar=tensor([0.1604, 0.2419, 0.2595, 0.1440, 0.1808, 0.1949, 0.1878, 0.2528], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0093, 0.0110, 0.0092, 0.0119, 0.0093, 0.0098, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 05:04:11,231 INFO [finetune.py:976] (3/7) Epoch 24, batch 2150, loss[loss=0.19, simple_loss=0.2719, pruned_loss=0.05401, over 4814.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2436, pruned_loss=0.05005, over 954513.03 frames. ], batch size: 38, lr: 3.05e-03, grad_scale: 32.0 2023-03-27 05:04:35,077 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.3711, 2.9742, 3.1300, 3.2623, 3.1740, 2.9282, 3.3779, 1.0233], device='cuda:3'), covar=tensor([0.1041, 0.1069, 0.1152, 0.1106, 0.1458, 0.1962, 0.1108, 0.5583], device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0247, 0.0281, 0.0293, 0.0337, 0.0286, 0.0306, 0.0301], 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') 2023-03-27 05:04:35,131 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6488, 1.5618, 1.3968, 1.7254, 1.7192, 1.7077, 1.0984, 1.4137], device='cuda:3'), covar=tensor([0.2496, 0.2147, 0.2080, 0.1803, 0.1664, 0.1333, 0.2670, 0.2123], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0209, 0.0213, 0.0196, 0.0243, 0.0190, 0.0216, 0.0204], 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') 2023-03-27 05:04:35,739 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7568, 1.6425, 1.4592, 1.8575, 2.1101, 1.8416, 1.3423, 1.4346], device='cuda:3'), covar=tensor([0.2162, 0.2007, 0.1915, 0.1619, 0.1599, 0.1152, 0.2576, 0.2000], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0209, 0.0213, 0.0196, 0.0243, 0.0190, 0.0216, 0.0204], 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') 2023-03-27 05:04:38,021 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8634, 1.6138, 2.0866, 1.3732, 1.8708, 2.0600, 1.4932, 2.2747], device='cuda:3'), covar=tensor([0.1209, 0.2171, 0.1409, 0.1934, 0.0977, 0.1446, 0.3250, 0.0732], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0209, 0.0192, 0.0191, 0.0174, 0.0215, 0.0218, 0.0201], 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') 2023-03-27 05:04:44,944 INFO [finetune.py:976] (3/7) Epoch 24, batch 2200, loss[loss=0.1847, simple_loss=0.2491, pruned_loss=0.06016, over 4814.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2467, pruned_loss=0.05121, over 955042.22 frames. ], batch size: 25, lr: 3.05e-03, grad_scale: 32.0 2023-03-27 05:05:02,723 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.027e+02 1.464e+02 1.824e+02 2.239e+02 3.694e+02, threshold=3.648e+02, percent-clipped=0.0 2023-03-27 05:05:24,729 INFO [finetune.py:976] (3/7) Epoch 24, batch 2250, loss[loss=0.1471, simple_loss=0.2058, pruned_loss=0.04423, over 3872.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2491, pruned_loss=0.05254, over 955742.66 frames. ], batch size: 16, lr: 3.05e-03, grad_scale: 32.0 2023-03-27 05:05:35,450 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133997.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:05:49,820 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134008.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:06:02,134 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6459, 1.5794, 1.3386, 1.6207, 1.8880, 1.8744, 1.6124, 1.3636], device='cuda:3'), covar=tensor([0.0335, 0.0364, 0.0659, 0.0324, 0.0237, 0.0457, 0.0335, 0.0485], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0106, 0.0144, 0.0112, 0.0100, 0.0113, 0.0102, 0.0112], device='cuda:3'), out_proj_covar=tensor([7.7516e-05, 8.1606e-05, 1.1290e-04, 8.5527e-05, 7.7763e-05, 8.3349e-05, 7.5762e-05, 8.5543e-05], device='cuda:3') 2023-03-27 05:06:05,738 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7199, 1.5336, 2.1481, 3.3111, 2.2468, 2.4190, 1.1843, 2.7540], device='cuda:3'), covar=tensor([0.1524, 0.1336, 0.1168, 0.0509, 0.0773, 0.1315, 0.1572, 0.0417], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0114, 0.0132, 0.0161, 0.0100, 0.0135, 0.0123, 0.0099], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 05:06:07,562 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134030.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 05:06:09,734 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134032.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:06:09,777 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6944, 2.4896, 1.9968, 1.0139, 2.2414, 2.2203, 1.9972, 2.3171], device='cuda:3'), covar=tensor([0.0748, 0.0812, 0.1593, 0.2118, 0.1326, 0.1992, 0.2231, 0.0896], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0191, 0.0199, 0.0180, 0.0209, 0.0207, 0.0223, 0.0194], 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') 2023-03-27 05:06:12,681 INFO [finetune.py:976] (3/7) Epoch 24, batch 2300, loss[loss=0.1829, simple_loss=0.248, pruned_loss=0.05889, over 4922.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2494, pruned_loss=0.0523, over 955568.80 frames. ], batch size: 38, lr: 3.05e-03, grad_scale: 32.0 2023-03-27 05:06:27,050 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134058.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:06:31,055 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.051e+02 1.533e+02 1.723e+02 2.089e+02 4.293e+02, threshold=3.445e+02, percent-clipped=2.0 2023-03-27 05:06:40,132 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=134078.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 05:06:41,358 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=134080.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:06:46,519 INFO [finetune.py:976] (3/7) Epoch 24, batch 2350, loss[loss=0.1863, simple_loss=0.2594, pruned_loss=0.05659, over 4818.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2463, pruned_loss=0.05117, over 956410.43 frames. ], batch size: 30, lr: 3.05e-03, grad_scale: 32.0 2023-03-27 05:07:09,013 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8509, 3.9729, 3.8116, 2.0335, 4.0929, 2.9488, 0.9033, 2.8603], device='cuda:3'), covar=tensor([0.2309, 0.1911, 0.1446, 0.3452, 0.0868, 0.1085, 0.4838, 0.1524], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0179, 0.0160, 0.0129, 0.0160, 0.0124, 0.0148, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 05:07:14,968 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134130.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:07:19,176 INFO [finetune.py:976] (3/7) Epoch 24, batch 2400, loss[loss=0.1579, simple_loss=0.2292, pruned_loss=0.04332, over 4836.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2435, pruned_loss=0.05053, over 954139.81 frames. ], batch size: 44, lr: 3.05e-03, grad_scale: 32.0 2023-03-27 05:07:35,391 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-27 05:07:38,330 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.553e+01 1.434e+02 1.789e+02 2.166e+02 3.942e+02, threshold=3.577e+02, percent-clipped=1.0 2023-03-27 05:07:49,932 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=134178.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:07:55,475 INFO [finetune.py:976] (3/7) Epoch 24, batch 2450, loss[loss=0.178, simple_loss=0.2509, pruned_loss=0.05257, over 4818.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2417, pruned_loss=0.0502, over 955726.47 frames. ], batch size: 39, lr: 3.05e-03, grad_scale: 32.0 2023-03-27 05:08:12,572 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3426, 2.3615, 2.3754, 1.6403, 2.1678, 2.5268, 2.5141, 1.9521], device='cuda:3'), covar=tensor([0.0630, 0.0628, 0.0691, 0.0943, 0.1254, 0.0725, 0.0598, 0.1204], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0137, 0.0140, 0.0120, 0.0127, 0.0139, 0.0139, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 05:08:36,910 INFO [finetune.py:976] (3/7) Epoch 24, batch 2500, loss[loss=0.2002, simple_loss=0.2884, pruned_loss=0.05602, over 4777.00 frames. ], tot_loss[loss=0.174, simple_loss=0.245, pruned_loss=0.05147, over 954964.99 frames. ], batch size: 59, lr: 3.05e-03, grad_scale: 32.0 2023-03-27 05:08:55,716 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.396e+01 1.499e+02 1.865e+02 2.171e+02 5.575e+02, threshold=3.730e+02, percent-clipped=1.0 2023-03-27 05:09:01,382 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9307, 1.7926, 1.5295, 1.6759, 1.7067, 1.6750, 1.7255, 2.4062], device='cuda:3'), covar=tensor([0.3720, 0.3786, 0.3082, 0.3348, 0.3648, 0.2346, 0.3417, 0.1647], device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0262, 0.0234, 0.0275, 0.0256, 0.0226, 0.0253, 0.0235], 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') 2023-03-27 05:09:20,384 INFO [finetune.py:976] (3/7) Epoch 24, batch 2550, loss[loss=0.19, simple_loss=0.2488, pruned_loss=0.06562, over 4049.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.249, pruned_loss=0.0527, over 955138.10 frames. ], batch size: 66, lr: 3.05e-03, grad_scale: 32.0 2023-03-27 05:09:32,213 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134304.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:09:34,671 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134308.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:09:35,293 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134309.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:09:54,095 INFO [finetune.py:976] (3/7) Epoch 24, batch 2600, loss[loss=0.1965, simple_loss=0.2689, pruned_loss=0.06207, over 4927.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2487, pruned_loss=0.05236, over 955465.57 frames. ], batch size: 33, lr: 3.05e-03, grad_scale: 32.0 2023-03-27 05:10:04,838 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134353.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:10:07,200 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=134356.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:10:09,697 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7718, 4.1429, 3.8255, 2.1277, 4.2141, 3.3644, 1.4733, 2.9314], device='cuda:3'), covar=tensor([0.2288, 0.1705, 0.1413, 0.3179, 0.0836, 0.0918, 0.3819, 0.1420], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0180, 0.0162, 0.0130, 0.0161, 0.0124, 0.0149, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 05:10:11,577 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3459, 2.3816, 2.4110, 1.6415, 2.1483, 2.5665, 2.5712, 2.0747], device='cuda:3'), covar=tensor([0.0639, 0.0740, 0.0808, 0.0994, 0.1283, 0.0788, 0.0639, 0.1276], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0138, 0.0141, 0.0121, 0.0128, 0.0140, 0.0141, 0.0164], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 05:10:12,062 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.001e+02 1.471e+02 1.806e+02 2.184e+02 4.519e+02, threshold=3.612e+02, percent-clipped=2.0 2023-03-27 05:10:13,312 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134365.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:10:16,849 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134370.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:10:29,856 INFO [finetune.py:976] (3/7) Epoch 24, batch 2650, loss[loss=0.1764, simple_loss=0.2571, pruned_loss=0.04778, over 4847.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2495, pruned_loss=0.05247, over 954407.99 frames. ], batch size: 44, lr: 3.05e-03, grad_scale: 32.0 2023-03-27 05:10:55,602 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.3314, 2.9897, 3.0995, 3.2636, 3.1227, 2.8912, 3.3714, 0.9646], device='cuda:3'), covar=tensor([0.0978, 0.0878, 0.1012, 0.1010, 0.1490, 0.1788, 0.1049, 0.5187], device='cuda:3'), in_proj_covar=tensor([0.0346, 0.0246, 0.0281, 0.0292, 0.0336, 0.0286, 0.0307, 0.0300], 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') 2023-03-27 05:11:07,363 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8290, 1.7792, 1.5335, 1.9333, 2.1358, 1.9239, 1.4406, 1.4999], device='cuda:3'), covar=tensor([0.2168, 0.1899, 0.1907, 0.1601, 0.1584, 0.1204, 0.2471, 0.1984], device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0208, 0.0212, 0.0195, 0.0241, 0.0189, 0.0214, 0.0202], 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') 2023-03-27 05:11:14,488 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5953, 1.5328, 1.4437, 1.5216, 1.0157, 3.1667, 1.2379, 1.5714], device='cuda:3'), covar=tensor([0.3100, 0.2355, 0.2100, 0.2384, 0.1831, 0.0243, 0.2656, 0.1221], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0117, 0.0121, 0.0124, 0.0114, 0.0097, 0.0095, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 05:11:21,024 INFO [finetune.py:976] (3/7) Epoch 24, batch 2700, loss[loss=0.1831, simple_loss=0.2621, pruned_loss=0.05204, over 4901.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2482, pruned_loss=0.05168, over 951980.07 frames. ], batch size: 43, lr: 3.05e-03, grad_scale: 32.0 2023-03-27 05:11:39,167 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.916e+01 1.416e+02 1.758e+02 2.159e+02 3.599e+02, threshold=3.516e+02, percent-clipped=0.0 2023-03-27 05:11:44,531 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.1405, 2.8388, 2.6346, 1.3832, 2.7542, 2.2176, 2.1370, 2.6218], device='cuda:3'), covar=tensor([0.0927, 0.0797, 0.1715, 0.2051, 0.1726, 0.2066, 0.2078, 0.1004], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0190, 0.0199, 0.0180, 0.0209, 0.0208, 0.0223, 0.0195], 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') 2023-03-27 05:11:54,598 INFO [finetune.py:976] (3/7) Epoch 24, batch 2750, loss[loss=0.2143, simple_loss=0.2639, pruned_loss=0.08236, over 4914.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2451, pruned_loss=0.0507, over 952457.92 frames. ], batch size: 37, lr: 3.05e-03, grad_scale: 32.0 2023-03-27 05:12:27,873 INFO [finetune.py:976] (3/7) Epoch 24, batch 2800, loss[loss=0.1563, simple_loss=0.2261, pruned_loss=0.04319, over 4835.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2406, pruned_loss=0.04905, over 953414.52 frames. ], batch size: 30, lr: 3.05e-03, grad_scale: 32.0 2023-03-27 05:12:46,112 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.301e+01 1.456e+02 1.823e+02 2.196e+02 4.309e+02, threshold=3.645e+02, percent-clipped=3.0 2023-03-27 05:12:59,885 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2492, 2.1729, 2.3466, 1.4659, 2.2430, 2.3544, 2.4120, 1.8833], device='cuda:3'), covar=tensor([0.0627, 0.0699, 0.0608, 0.0837, 0.0749, 0.0663, 0.0576, 0.1169], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0136, 0.0139, 0.0119, 0.0126, 0.0138, 0.0139, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 05:13:01,599 INFO [finetune.py:976] (3/7) Epoch 24, batch 2850, loss[loss=0.132, simple_loss=0.2052, pruned_loss=0.02941, over 4762.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.239, pruned_loss=0.04868, over 955445.20 frames. ], batch size: 28, lr: 3.05e-03, grad_scale: 16.0 2023-03-27 05:13:45,376 INFO [finetune.py:976] (3/7) Epoch 24, batch 2900, loss[loss=0.1936, simple_loss=0.2698, pruned_loss=0.05867, over 4793.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2419, pruned_loss=0.04941, over 954146.21 frames. ], batch size: 51, lr: 3.05e-03, grad_scale: 16.0 2023-03-27 05:13:56,303 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134653.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:14:00,530 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134660.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:14:03,944 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.005e+02 1.559e+02 1.783e+02 2.063e+02 3.902e+02, threshold=3.566e+02, percent-clipped=1.0 2023-03-27 05:14:04,026 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134665.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:14:04,096 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6013, 1.5250, 1.4064, 1.7276, 1.7239, 1.6716, 1.0699, 1.3986], device='cuda:3'), covar=tensor([0.2305, 0.2101, 0.1951, 0.1636, 0.1644, 0.1276, 0.2436, 0.1929], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0210, 0.0213, 0.0196, 0.0243, 0.0190, 0.0215, 0.0203], 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') 2023-03-27 05:14:20,900 INFO [finetune.py:976] (3/7) Epoch 24, batch 2950, loss[loss=0.1632, simple_loss=0.2489, pruned_loss=0.03869, over 4755.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2433, pruned_loss=0.04934, over 953300.71 frames. ], batch size: 54, lr: 3.05e-03, grad_scale: 16.0 2023-03-27 05:14:37,442 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=134701.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:14:40,372 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6168, 3.8874, 3.6773, 1.8620, 4.0324, 2.9821, 0.8412, 2.6928], device='cuda:3'), covar=tensor([0.2531, 0.1819, 0.1529, 0.3463, 0.0891, 0.1020, 0.4609, 0.1503], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0181, 0.0162, 0.0130, 0.0162, 0.0125, 0.0149, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 05:14:51,611 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9094, 5.0608, 4.8883, 2.8316, 5.1171, 3.8360, 1.2757, 3.6048], device='cuda:3'), covar=tensor([0.2281, 0.1533, 0.1173, 0.3002, 0.0649, 0.0829, 0.4520, 0.1317], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0181, 0.0162, 0.0131, 0.0162, 0.0125, 0.0150, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 05:15:02,624 INFO [finetune.py:976] (3/7) Epoch 24, batch 3000, loss[loss=0.1967, simple_loss=0.2703, pruned_loss=0.0616, over 4818.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2455, pruned_loss=0.05081, over 952765.61 frames. ], batch size: 40, lr: 3.05e-03, grad_scale: 16.0 2023-03-27 05:15:02,624 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-27 05:15:06,777 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5489, 1.3829, 1.3379, 1.4918, 1.7168, 1.6877, 1.4603, 1.3043], device='cuda:3'), covar=tensor([0.0406, 0.0350, 0.0645, 0.0343, 0.0273, 0.0387, 0.0351, 0.0458], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0107, 0.0145, 0.0112, 0.0101, 0.0114, 0.0103, 0.0113], device='cuda:3'), out_proj_covar=tensor([7.7582e-05, 8.1824e-05, 1.1349e-04, 8.5894e-05, 7.8169e-05, 8.4084e-05, 7.6307e-05, 8.6140e-05], device='cuda:3') 2023-03-27 05:15:13,334 INFO [finetune.py:1010] (3/7) Epoch 24, validation: loss=0.1561, simple_loss=0.2251, pruned_loss=0.0436, over 2265189.00 frames. 2023-03-27 05:15:13,334 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6469MB 2023-03-27 05:15:31,259 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.188e+02 1.527e+02 1.858e+02 2.241e+02 5.364e+02, threshold=3.716e+02, percent-clipped=3.0 2023-03-27 05:15:48,109 INFO [finetune.py:976] (3/7) Epoch 24, batch 3050, loss[loss=0.19, simple_loss=0.2607, pruned_loss=0.05962, over 4774.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2475, pruned_loss=0.05152, over 953372.76 frames. ], batch size: 26, lr: 3.05e-03, grad_scale: 16.0 2023-03-27 05:16:35,973 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7994, 2.0857, 1.5633, 1.7516, 2.3135, 2.3601, 2.0747, 1.9079], device='cuda:3'), covar=tensor([0.0429, 0.0354, 0.0651, 0.0388, 0.0333, 0.0586, 0.0376, 0.0477], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0107, 0.0145, 0.0112, 0.0101, 0.0114, 0.0103, 0.0113], device='cuda:3'), out_proj_covar=tensor([7.7644e-05, 8.1786e-05, 1.1333e-04, 8.5926e-05, 7.8207e-05, 8.4119e-05, 7.6215e-05, 8.6107e-05], device='cuda:3') 2023-03-27 05:16:39,367 INFO [finetune.py:976] (3/7) Epoch 24, batch 3100, loss[loss=0.1964, simple_loss=0.2699, pruned_loss=0.06143, over 4900.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2459, pruned_loss=0.05106, over 954008.94 frames. ], batch size: 43, lr: 3.05e-03, grad_scale: 16.0 2023-03-27 05:16:56,670 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134862.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:16:58,376 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.835e+01 1.385e+02 1.668e+02 2.115e+02 5.080e+02, threshold=3.336e+02, percent-clipped=1.0 2023-03-27 05:17:00,347 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8050, 1.6939, 1.5516, 1.9273, 2.1669, 1.8894, 1.4415, 1.5441], device='cuda:3'), covar=tensor([0.2103, 0.1937, 0.1902, 0.1491, 0.1480, 0.1164, 0.2387, 0.1888], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0211, 0.0214, 0.0198, 0.0244, 0.0191, 0.0217, 0.0205], 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') 2023-03-27 05:17:08,762 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.95 vs. limit=5.0 2023-03-27 05:17:12,765 INFO [finetune.py:976] (3/7) Epoch 24, batch 3150, loss[loss=0.1654, simple_loss=0.2258, pruned_loss=0.0525, over 4258.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2431, pruned_loss=0.05034, over 952187.82 frames. ], batch size: 18, lr: 3.05e-03, grad_scale: 16.0 2023-03-27 05:17:36,023 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3917, 1.2698, 1.2171, 1.3067, 1.6056, 1.5319, 1.3768, 1.2176], device='cuda:3'), covar=tensor([0.0367, 0.0304, 0.0630, 0.0295, 0.0247, 0.0430, 0.0343, 0.0464], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0106, 0.0144, 0.0112, 0.0100, 0.0113, 0.0102, 0.0113], device='cuda:3'), out_proj_covar=tensor([7.7493e-05, 8.1375e-05, 1.1304e-04, 8.5672e-05, 7.7925e-05, 8.3842e-05, 7.6072e-05, 8.5642e-05], device='cuda:3') 2023-03-27 05:17:37,237 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134923.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:17:46,593 INFO [finetune.py:976] (3/7) Epoch 24, batch 3200, loss[loss=0.1619, simple_loss=0.2269, pruned_loss=0.04847, over 4812.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2393, pruned_loss=0.04903, over 952645.37 frames. ], batch size: 25, lr: 3.05e-03, grad_scale: 16.0 2023-03-27 05:17:51,453 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8418, 1.4885, 0.9000, 1.6364, 2.1387, 1.4077, 1.5934, 1.6619], device='cuda:3'), covar=tensor([0.1341, 0.1830, 0.1861, 0.1168, 0.1975, 0.1823, 0.1437, 0.1895], device='cuda:3'), in_proj_covar=tensor([0.0088, 0.0092, 0.0108, 0.0091, 0.0117, 0.0091, 0.0096, 0.0087], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-27 05:18:03,010 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134960.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:18:05,933 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.665e+01 1.513e+02 1.793e+02 2.252e+02 3.579e+02, threshold=3.586e+02, percent-clipped=1.0 2023-03-27 05:18:06,034 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134965.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:18:22,481 INFO [finetune.py:976] (3/7) Epoch 24, batch 3250, loss[loss=0.188, simple_loss=0.2573, pruned_loss=0.05936, over 4820.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.241, pruned_loss=0.05022, over 952623.90 frames. ], batch size: 38, lr: 3.05e-03, grad_scale: 16.0 2023-03-27 05:18:45,538 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=135008.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:18:48,600 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=135013.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:18:59,443 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6519, 1.6866, 1.5226, 1.7231, 1.4437, 4.2488, 1.6007, 1.9210], device='cuda:3'), covar=tensor([0.3303, 0.2428, 0.2111, 0.2293, 0.1594, 0.0132, 0.2530, 0.1258], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0116, 0.0121, 0.0123, 0.0113, 0.0096, 0.0095, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 05:19:04,066 INFO [finetune.py:976] (3/7) Epoch 24, batch 3300, loss[loss=0.1545, simple_loss=0.236, pruned_loss=0.03652, over 4799.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.245, pruned_loss=0.05176, over 951596.27 frames. ], batch size: 51, lr: 3.05e-03, grad_scale: 16.0 2023-03-27 05:19:13,248 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.06 vs. limit=5.0 2023-03-27 05:19:23,522 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.113e+02 1.563e+02 1.899e+02 2.275e+02 5.700e+02, threshold=3.799e+02, percent-clipped=2.0 2023-03-27 05:19:44,198 INFO [finetune.py:976] (3/7) Epoch 24, batch 3350, loss[loss=0.1667, simple_loss=0.2446, pruned_loss=0.04437, over 4897.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2447, pruned_loss=0.05088, over 951991.07 frames. ], batch size: 35, lr: 3.05e-03, grad_scale: 16.0 2023-03-27 05:20:21,440 INFO [finetune.py:976] (3/7) Epoch 24, batch 3400, loss[loss=0.1607, simple_loss=0.2344, pruned_loss=0.04352, over 4830.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2464, pruned_loss=0.05119, over 952990.21 frames. ], batch size: 49, lr: 3.05e-03, grad_scale: 16.0 2023-03-27 05:20:40,369 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.077e+02 1.584e+02 1.828e+02 2.150e+02 3.792e+02, threshold=3.656e+02, percent-clipped=0.0 2023-03-27 05:20:44,635 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135171.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:20:46,443 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5688, 3.4540, 3.2477, 1.5127, 3.6335, 2.7613, 1.0850, 2.3425], device='cuda:3'), covar=tensor([0.2396, 0.2135, 0.1857, 0.3651, 0.1076, 0.1052, 0.4215, 0.1600], device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0181, 0.0162, 0.0130, 0.0161, 0.0124, 0.0149, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 05:20:54,273 INFO [finetune.py:976] (3/7) Epoch 24, batch 3450, loss[loss=0.1569, simple_loss=0.2323, pruned_loss=0.04078, over 4909.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2447, pruned_loss=0.05011, over 954014.30 frames. ], batch size: 32, lr: 3.05e-03, grad_scale: 16.0 2023-03-27 05:21:27,777 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=135218.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:21:40,728 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135232.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:21:47,100 INFO [finetune.py:976] (3/7) Epoch 24, batch 3500, loss[loss=0.1706, simple_loss=0.2428, pruned_loss=0.04917, over 4815.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2433, pruned_loss=0.04991, over 953485.83 frames. ], batch size: 38, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:22:06,081 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.678e+01 1.506e+02 1.714e+02 2.011e+02 3.544e+02, threshold=3.428e+02, percent-clipped=0.0 2023-03-27 05:22:20,456 INFO [finetune.py:976] (3/7) Epoch 24, batch 3550, loss[loss=0.1397, simple_loss=0.2113, pruned_loss=0.03409, over 4826.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2412, pruned_loss=0.04941, over 953930.68 frames. ], batch size: 30, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:22:24,264 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1963, 2.0353, 1.7234, 1.8930, 2.1218, 1.8450, 2.2783, 2.1667], device='cuda:3'), covar=tensor([0.1360, 0.1942, 0.2987, 0.2594, 0.2719, 0.1780, 0.2991, 0.1725], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0191, 0.0237, 0.0255, 0.0251, 0.0207, 0.0215, 0.0204], 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') 2023-03-27 05:22:54,386 INFO [finetune.py:976] (3/7) Epoch 24, batch 3600, loss[loss=0.1931, simple_loss=0.2699, pruned_loss=0.05819, over 4865.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2391, pruned_loss=0.04864, over 953876.58 frames. ], batch size: 34, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:23:09,579 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 2023-03-27 05:23:12,795 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.699e+01 1.474e+02 1.759e+02 2.084e+02 3.295e+02, threshold=3.517e+02, percent-clipped=0.0 2023-03-27 05:23:28,231 INFO [finetune.py:976] (3/7) Epoch 24, batch 3650, loss[loss=0.1597, simple_loss=0.2321, pruned_loss=0.04368, over 4819.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2402, pruned_loss=0.04933, over 951661.15 frames. ], batch size: 33, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:23:35,901 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-27 05:24:11,249 INFO [finetune.py:976] (3/7) Epoch 24, batch 3700, loss[loss=0.1659, simple_loss=0.244, pruned_loss=0.04388, over 4827.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2439, pruned_loss=0.05051, over 952301.70 frames. ], batch size: 33, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:24:14,357 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7081, 1.5069, 1.0367, 0.2871, 1.3427, 1.5062, 1.4457, 1.4313], device='cuda:3'), covar=tensor([0.1196, 0.0925, 0.1726, 0.2086, 0.1467, 0.2613, 0.2437, 0.1028], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0189, 0.0197, 0.0179, 0.0208, 0.0206, 0.0220, 0.0193], 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') 2023-03-27 05:24:28,521 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.118e+01 1.614e+02 1.999e+02 2.429e+02 5.138e+02, threshold=3.998e+02, percent-clipped=6.0 2023-03-27 05:24:43,339 INFO [finetune.py:976] (3/7) Epoch 24, batch 3750, loss[loss=0.1731, simple_loss=0.2378, pruned_loss=0.05415, over 4861.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2461, pruned_loss=0.051, over 954012.91 frames. ], batch size: 31, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:25:12,912 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=135518.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:25:18,590 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-03-27 05:25:18,853 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=135527.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:25:26,760 INFO [finetune.py:976] (3/7) Epoch 24, batch 3800, loss[loss=0.1575, simple_loss=0.2378, pruned_loss=0.0386, over 4909.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2463, pruned_loss=0.05098, over 952184.74 frames. ], batch size: 38, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:25:44,703 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.088e+02 1.524e+02 1.815e+02 2.221e+02 4.659e+02, threshold=3.630e+02, percent-clipped=3.0 2023-03-27 05:25:45,389 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=135566.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:26:00,454 INFO [finetune.py:976] (3/7) Epoch 24, batch 3850, loss[loss=0.1853, simple_loss=0.2606, pruned_loss=0.05494, over 4903.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2451, pruned_loss=0.05039, over 951800.39 frames. ], batch size: 35, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:26:38,636 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135630.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:26:45,797 INFO [finetune.py:976] (3/7) Epoch 24, batch 3900, loss[loss=0.1727, simple_loss=0.2415, pruned_loss=0.05197, over 4910.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2432, pruned_loss=0.05004, over 954106.77 frames. ], batch size: 35, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:27:08,432 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6679, 0.7480, 1.7716, 1.7015, 1.6070, 1.5319, 1.6031, 1.7194], device='cuda:3'), covar=tensor([0.3635, 0.3760, 0.3224, 0.3155, 0.4232, 0.3549, 0.4008, 0.2795], device='cuda:3'), in_proj_covar=tensor([0.0262, 0.0245, 0.0266, 0.0291, 0.0290, 0.0267, 0.0297, 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') 2023-03-27 05:27:10,709 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.672e+01 1.400e+02 1.667e+02 1.961e+02 4.314e+02, threshold=3.334e+02, percent-clipped=1.0 2023-03-27 05:27:26,033 INFO [finetune.py:976] (3/7) Epoch 24, batch 3950, loss[loss=0.1467, simple_loss=0.2133, pruned_loss=0.04001, over 4825.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.24, pruned_loss=0.0492, over 953348.82 frames. ], batch size: 30, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:27:29,115 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135691.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:27:39,367 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2995, 2.9326, 2.8159, 1.2894, 3.0275, 2.2243, 0.7281, 1.8897], device='cuda:3'), covar=tensor([0.2448, 0.2690, 0.1746, 0.3627, 0.1410, 0.1168, 0.4260, 0.1774], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0180, 0.0162, 0.0130, 0.0162, 0.0124, 0.0149, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 05:27:57,309 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4859, 1.4421, 1.6782, 2.5241, 1.6866, 2.2260, 0.9692, 2.1402], device='cuda:3'), covar=tensor([0.1694, 0.1302, 0.1121, 0.0689, 0.0854, 0.1040, 0.1429, 0.0629], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0114, 0.0132, 0.0162, 0.0100, 0.0135, 0.0123, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 05:27:58,422 INFO [finetune.py:976] (3/7) Epoch 24, batch 4000, loss[loss=0.1527, simple_loss=0.2329, pruned_loss=0.03621, over 4742.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2397, pruned_loss=0.04914, over 956017.91 frames. ], batch size: 27, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:28:16,421 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.536e+01 1.548e+02 1.897e+02 2.315e+02 3.943e+02, threshold=3.793e+02, percent-clipped=5.0 2023-03-27 05:28:31,229 INFO [finetune.py:976] (3/7) Epoch 24, batch 4050, loss[loss=0.2157, simple_loss=0.2813, pruned_loss=0.07508, over 4824.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2436, pruned_loss=0.05069, over 955993.29 frames. ], batch size: 39, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:28:59,137 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=135827.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:29:09,952 INFO [finetune.py:976] (3/7) Epoch 24, batch 4100, loss[loss=0.1974, simple_loss=0.275, pruned_loss=0.05988, over 4804.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2454, pruned_loss=0.05058, over 957398.70 frames. ], batch size: 40, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:29:29,730 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3182, 2.2085, 1.9162, 2.3798, 2.9167, 2.2981, 2.2699, 1.8518], device='cuda:3'), covar=tensor([0.2160, 0.1889, 0.1879, 0.1596, 0.1532, 0.1093, 0.1910, 0.1835], device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0208, 0.0212, 0.0195, 0.0241, 0.0188, 0.0214, 0.0203], 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') 2023-03-27 05:29:32,641 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.978e+01 1.562e+02 1.866e+02 2.353e+02 4.250e+02, threshold=3.731e+02, percent-clipped=2.0 2023-03-27 05:29:34,672 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-27 05:29:39,221 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=135875.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:29:46,949 INFO [finetune.py:976] (3/7) Epoch 24, batch 4150, loss[loss=0.2241, simple_loss=0.2873, pruned_loss=0.08048, over 4819.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2464, pruned_loss=0.05068, over 956536.51 frames. ], batch size: 39, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:30:30,442 INFO [finetune.py:976] (3/7) Epoch 24, batch 4200, loss[loss=0.1443, simple_loss=0.2193, pruned_loss=0.03467, over 4810.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2463, pruned_loss=0.05061, over 955186.19 frames. ], batch size: 39, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:30:40,105 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 2023-03-27 05:30:47,978 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-03-27 05:30:49,318 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.053e+01 1.587e+02 1.796e+02 2.438e+02 3.967e+02, threshold=3.591e+02, percent-clipped=1.0 2023-03-27 05:30:57,405 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-03-27 05:31:00,639 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135982.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:31:03,040 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=135986.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:31:03,581 INFO [finetune.py:976] (3/7) Epoch 24, batch 4250, loss[loss=0.1782, simple_loss=0.2437, pruned_loss=0.05635, over 4823.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2457, pruned_loss=0.05078, over 954870.40 frames. ], batch size: 30, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:31:08,457 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135994.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:31:45,377 INFO [finetune.py:976] (3/7) Epoch 24, batch 4300, loss[loss=0.1631, simple_loss=0.2282, pruned_loss=0.04896, over 4799.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2433, pruned_loss=0.05033, over 955460.09 frames. ], batch size: 25, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:31:49,210 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136043.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:31:49,268 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0 2023-03-27 05:32:02,584 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136055.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:32:04,966 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-03-27 05:32:14,138 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.579e+01 1.491e+02 1.827e+02 2.181e+02 5.621e+02, threshold=3.653e+02, percent-clipped=1.0 2023-03-27 05:32:20,781 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-27 05:32:31,275 INFO [finetune.py:976] (3/7) Epoch 24, batch 4350, loss[loss=0.1434, simple_loss=0.2261, pruned_loss=0.03037, over 4725.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2405, pruned_loss=0.04955, over 956416.06 frames. ], batch size: 23, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:32:40,939 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9158, 2.6406, 2.4607, 1.2327, 2.6243, 2.0858, 1.9636, 2.4578], device='cuda:3'), covar=tensor([0.1478, 0.0841, 0.1895, 0.2233, 0.1748, 0.2386, 0.2466, 0.1258], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0192, 0.0200, 0.0181, 0.0211, 0.0209, 0.0225, 0.0196], 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') 2023-03-27 05:33:04,547 INFO [finetune.py:976] (3/7) Epoch 24, batch 4400, loss[loss=0.168, simple_loss=0.2412, pruned_loss=0.0474, over 4830.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2415, pruned_loss=0.05035, over 953434.32 frames. ], batch size: 30, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:33:06,476 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5556, 2.6813, 2.6447, 1.8076, 2.5787, 2.9085, 2.9567, 2.3034], device='cuda:3'), covar=tensor([0.0610, 0.0583, 0.0659, 0.0854, 0.0711, 0.0698, 0.0546, 0.1033], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0135, 0.0138, 0.0118, 0.0125, 0.0137, 0.0137, 0.0160], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 05:33:08,163 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136142.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:33:13,272 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-03-27 05:33:23,890 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.065e+02 1.540e+02 1.819e+02 2.170e+02 3.954e+02, threshold=3.638e+02, percent-clipped=3.0 2023-03-27 05:33:37,769 INFO [finetune.py:976] (3/7) Epoch 24, batch 4450, loss[loss=0.1565, simple_loss=0.2424, pruned_loss=0.03533, over 4881.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2433, pruned_loss=0.04988, over 953077.82 frames. ], batch size: 32, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:33:48,792 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136203.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:33:53,528 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0072, 1.4606, 0.6570, 1.8039, 2.4033, 1.6653, 1.6368, 1.8908], device='cuda:3'), covar=tensor([0.1475, 0.2061, 0.2251, 0.1212, 0.1869, 0.2052, 0.1422, 0.1923], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0094, 0.0111, 0.0092, 0.0119, 0.0093, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 05:34:13,599 INFO [finetune.py:976] (3/7) Epoch 24, batch 4500, loss[loss=0.2148, simple_loss=0.2831, pruned_loss=0.07325, over 4831.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2456, pruned_loss=0.05048, over 953146.00 frames. ], batch size: 47, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:34:23,920 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-27 05:34:24,433 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5052, 2.7210, 2.4352, 1.6948, 2.4489, 2.7638, 2.7717, 2.2027], device='cuda:3'), covar=tensor([0.0562, 0.0513, 0.0687, 0.0899, 0.0815, 0.0591, 0.0535, 0.1008], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0135, 0.0138, 0.0118, 0.0125, 0.0137, 0.0137, 0.0160], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 05:34:26,817 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6901, 1.4625, 1.8810, 1.1951, 1.7640, 1.8798, 1.4465, 2.0831], device='cuda:3'), covar=tensor([0.1175, 0.2220, 0.1326, 0.1879, 0.0820, 0.1244, 0.2924, 0.0783], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0207, 0.0192, 0.0190, 0.0174, 0.0214, 0.0216, 0.0200], 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') 2023-03-27 05:34:32,750 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.33 vs. limit=5.0 2023-03-27 05:34:39,506 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.886e+01 1.509e+02 1.852e+02 2.239e+02 3.856e+02, threshold=3.704e+02, percent-clipped=1.0 2023-03-27 05:34:48,297 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.5144, 1.4711, 1.4368, 0.7145, 1.5000, 1.7878, 1.7027, 1.3703], device='cuda:3'), covar=tensor([0.1053, 0.0680, 0.0615, 0.0618, 0.0495, 0.0567, 0.0319, 0.0705], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0148, 0.0125, 0.0121, 0.0131, 0.0128, 0.0141, 0.0147], device='cuda:3'), out_proj_covar=tensor([8.8927e-05, 1.0655e-04, 8.9548e-05, 8.5115e-05, 9.1774e-05, 9.1243e-05, 1.0036e-04, 1.0548e-04], device='cuda:3') 2023-03-27 05:34:54,273 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136286.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:34:54,759 INFO [finetune.py:976] (3/7) Epoch 24, batch 4550, loss[loss=0.1633, simple_loss=0.2322, pruned_loss=0.04722, over 4827.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2464, pruned_loss=0.05031, over 952027.34 frames. ], batch size: 49, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:35:27,964 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-03-27 05:35:28,210 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=136334.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:35:28,890 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136335.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:35:30,007 INFO [finetune.py:976] (3/7) Epoch 24, batch 4600, loss[loss=0.1652, simple_loss=0.2256, pruned_loss=0.05242, over 4216.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2459, pruned_loss=0.05019, over 953005.53 frames. ], batch size: 18, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:35:35,227 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136338.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:35:45,768 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136350.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:35:56,270 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.033e+02 1.624e+02 1.856e+02 2.259e+02 4.732e+02, threshold=3.713e+02, percent-clipped=2.0 2023-03-27 05:36:11,519 INFO [finetune.py:976] (3/7) Epoch 24, batch 4650, loss[loss=0.1469, simple_loss=0.2261, pruned_loss=0.03382, over 4890.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2428, pruned_loss=0.04938, over 953675.49 frames. ], batch size: 35, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:36:17,113 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136396.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:36:22,550 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1657, 2.1951, 2.2241, 1.4868, 2.1415, 2.4243, 2.3779, 1.8799], device='cuda:3'), covar=tensor([0.0606, 0.0602, 0.0665, 0.0854, 0.0657, 0.0600, 0.0560, 0.1008], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0135, 0.0138, 0.0118, 0.0125, 0.0137, 0.0137, 0.0159], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 05:36:38,623 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.48 vs. limit=5.0 2023-03-27 05:36:45,426 INFO [finetune.py:976] (3/7) Epoch 24, batch 4700, loss[loss=0.1643, simple_loss=0.2354, pruned_loss=0.04666, over 4894.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2402, pruned_loss=0.04861, over 954336.37 frames. ], batch size: 35, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:36:47,947 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.8403, 3.3816, 3.5443, 3.6681, 3.6536, 3.4477, 3.8862, 1.1752], device='cuda:3'), covar=tensor([0.0831, 0.0826, 0.0964, 0.1085, 0.1179, 0.1476, 0.0826, 0.5506], device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0246, 0.0283, 0.0294, 0.0338, 0.0285, 0.0306, 0.0301], 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') 2023-03-27 05:37:13,732 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.054e+02 1.479e+02 1.754e+02 2.064e+02 3.231e+02, threshold=3.507e+02, percent-clipped=0.0 2023-03-27 05:37:38,205 INFO [finetune.py:976] (3/7) Epoch 24, batch 4750, loss[loss=0.1794, simple_loss=0.2599, pruned_loss=0.04942, over 4843.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2388, pruned_loss=0.04806, over 954819.93 frames. ], batch size: 49, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:37:44,920 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136498.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:37:56,942 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.0854, 1.8858, 1.9150, 0.8373, 2.1971, 2.4059, 2.1097, 1.7421], device='cuda:3'), covar=tensor([0.0942, 0.0767, 0.0549, 0.0726, 0.0461, 0.0714, 0.0447, 0.0757], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0149, 0.0126, 0.0121, 0.0130, 0.0129, 0.0141, 0.0148], device='cuda:3'), out_proj_covar=tensor([8.9213e-05, 1.0702e-04, 8.9999e-05, 8.5394e-05, 9.1472e-05, 9.1420e-05, 1.0074e-04, 1.0570e-04], device='cuda:3') 2023-03-27 05:38:10,346 INFO [finetune.py:976] (3/7) Epoch 24, batch 4800, loss[loss=0.1685, simple_loss=0.2419, pruned_loss=0.04752, over 4831.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2424, pruned_loss=0.04942, over 957159.69 frames. ], batch size: 33, lr: 3.04e-03, grad_scale: 16.0 2023-03-27 05:38:28,983 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.123e+02 1.582e+02 2.021e+02 2.347e+02 5.093e+02, threshold=4.042e+02, percent-clipped=3.0 2023-03-27 05:38:44,073 INFO [finetune.py:976] (3/7) Epoch 24, batch 4850, loss[loss=0.2135, simple_loss=0.2803, pruned_loss=0.07334, over 4884.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2461, pruned_loss=0.05035, over 957297.57 frames. ], batch size: 35, lr: 3.04e-03, grad_scale: 32.0 2023-03-27 05:38:51,264 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2810, 2.2096, 1.7636, 2.2153, 2.2114, 1.9278, 2.5417, 2.3847], device='cuda:3'), covar=tensor([0.1349, 0.1959, 0.2914, 0.2450, 0.2391, 0.1627, 0.3022, 0.1501], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0190, 0.0237, 0.0255, 0.0251, 0.0207, 0.0215, 0.0203], 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') 2023-03-27 05:39:00,597 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5568, 1.4696, 1.3076, 1.6410, 1.7203, 1.6450, 1.0768, 1.3243], device='cuda:3'), covar=tensor([0.2162, 0.2114, 0.2026, 0.1611, 0.1656, 0.1166, 0.2514, 0.1833], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0212, 0.0215, 0.0197, 0.0244, 0.0191, 0.0218, 0.0205], 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') 2023-03-27 05:39:05,907 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3165, 1.4866, 1.9357, 1.7099, 1.5459, 3.2772, 1.4480, 1.5895], device='cuda:3'), covar=tensor([0.1014, 0.1776, 0.1120, 0.0938, 0.1595, 0.0237, 0.1347, 0.1683], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0073, 0.0076, 0.0091, 0.0080, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 05:39:17,546 INFO [finetune.py:976] (3/7) Epoch 24, batch 4900, loss[loss=0.1758, simple_loss=0.2471, pruned_loss=0.05222, over 4811.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2472, pruned_loss=0.05094, over 955902.82 frames. ], batch size: 39, lr: 3.03e-03, grad_scale: 32.0 2023-03-27 05:39:18,262 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136638.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:39:22,263 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136643.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:39:26,545 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136650.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:39:42,312 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.069e+02 1.600e+02 1.925e+02 2.438e+02 3.559e+02, threshold=3.849e+02, percent-clipped=0.0 2023-03-27 05:39:59,909 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=136686.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:40:00,476 INFO [finetune.py:976] (3/7) Epoch 24, batch 4950, loss[loss=0.2277, simple_loss=0.2958, pruned_loss=0.07975, over 4095.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.249, pruned_loss=0.05135, over 956730.62 frames. ], batch size: 65, lr: 3.03e-03, grad_scale: 32.0 2023-03-27 05:40:03,951 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136691.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:40:08,765 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=136698.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:40:12,457 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136704.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:40:19,117 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6644, 1.5023, 1.1280, 0.2929, 1.4054, 1.4689, 1.4358, 1.5200], device='cuda:3'), covar=tensor([0.0921, 0.0834, 0.1329, 0.1869, 0.1210, 0.2385, 0.2157, 0.0821], device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0189, 0.0197, 0.0179, 0.0208, 0.0207, 0.0221, 0.0194], 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') 2023-03-27 05:40:33,840 INFO [finetune.py:976] (3/7) Epoch 24, batch 5000, loss[loss=0.1712, simple_loss=0.2468, pruned_loss=0.04783, over 4926.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2479, pruned_loss=0.05107, over 956426.57 frames. ], batch size: 33, lr: 3.03e-03, grad_scale: 32.0 2023-03-27 05:41:02,607 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.076e+02 1.520e+02 1.782e+02 2.173e+02 3.913e+02, threshold=3.563e+02, percent-clipped=1.0 2023-03-27 05:41:17,087 INFO [finetune.py:976] (3/7) Epoch 24, batch 5050, loss[loss=0.1564, simple_loss=0.2303, pruned_loss=0.04127, over 4922.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2438, pruned_loss=0.05, over 956333.93 frames. ], batch size: 36, lr: 3.03e-03, grad_scale: 32.0 2023-03-27 05:41:25,184 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136798.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:41:30,674 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136806.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:41:49,839 INFO [finetune.py:976] (3/7) Epoch 24, batch 5100, loss[loss=0.1526, simple_loss=0.2266, pruned_loss=0.03928, over 4823.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2406, pruned_loss=0.0494, over 956375.02 frames. ], batch size: 39, lr: 3.03e-03, grad_scale: 32.0 2023-03-27 05:41:56,303 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=136846.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:42:11,831 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.036e+02 1.576e+02 1.921e+02 2.257e+02 4.191e+02, threshold=3.841e+02, percent-clipped=1.0 2023-03-27 05:42:13,184 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136867.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:42:26,255 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9689, 1.7548, 1.6240, 1.8158, 2.1937, 2.1286, 1.8114, 1.6873], device='cuda:3'), covar=tensor([0.0345, 0.0366, 0.0565, 0.0325, 0.0193, 0.0371, 0.0364, 0.0394], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0107, 0.0146, 0.0112, 0.0101, 0.0114, 0.0103, 0.0113], device='cuda:3'), out_proj_covar=tensor([7.8086e-05, 8.1930e-05, 1.1405e-04, 8.5760e-05, 7.8507e-05, 8.4393e-05, 7.6473e-05, 8.6172e-05], device='cuda:3') 2023-03-27 05:42:35,177 INFO [finetune.py:976] (3/7) Epoch 24, batch 5150, loss[loss=0.1778, simple_loss=0.2673, pruned_loss=0.04412, over 4809.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.241, pruned_loss=0.04978, over 957475.23 frames. ], batch size: 40, lr: 3.03e-03, grad_scale: 32.0 2023-03-27 05:42:48,278 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-27 05:43:16,538 INFO [finetune.py:976] (3/7) Epoch 24, batch 5200, loss[loss=0.1896, simple_loss=0.2717, pruned_loss=0.05375, over 4896.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2436, pruned_loss=0.05043, over 956010.92 frames. ], batch size: 35, lr: 3.03e-03, grad_scale: 32.0 2023-03-27 05:43:35,513 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.133e+02 1.647e+02 1.940e+02 2.397e+02 3.428e+02, threshold=3.879e+02, percent-clipped=0.0 2023-03-27 05:43:41,130 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0552, 1.9205, 1.6766, 1.5701, 2.0245, 1.8199, 1.8949, 2.0430], device='cuda:3'), covar=tensor([0.1219, 0.1767, 0.2708, 0.2162, 0.2296, 0.1615, 0.2320, 0.1581], device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0188, 0.0235, 0.0252, 0.0248, 0.0204, 0.0213, 0.0201], 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') 2023-03-27 05:43:48,851 INFO [finetune.py:976] (3/7) Epoch 24, batch 5250, loss[loss=0.1829, simple_loss=0.2519, pruned_loss=0.05697, over 4861.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2461, pruned_loss=0.05102, over 956625.32 frames. ], batch size: 31, lr: 3.03e-03, grad_scale: 32.0 2023-03-27 05:43:49,662 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-03-27 05:43:51,862 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136991.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:43:57,215 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136999.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:44:22,624 INFO [finetune.py:976] (3/7) Epoch 24, batch 5300, loss[loss=0.187, simple_loss=0.2554, pruned_loss=0.05933, over 4917.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2478, pruned_loss=0.05185, over 957732.42 frames. ], batch size: 38, lr: 3.03e-03, grad_scale: 32.0 2023-03-27 05:44:23,378 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8646, 1.8162, 1.7624, 1.8032, 1.3376, 3.3697, 1.5111, 1.9615], device='cuda:3'), covar=tensor([0.2909, 0.2342, 0.1913, 0.2198, 0.1648, 0.0253, 0.2365, 0.1086], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0116, 0.0121, 0.0123, 0.0113, 0.0096, 0.0095, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 05:44:23,938 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=137039.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:44:24,095 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-03-27 05:44:42,412 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.035e+02 1.600e+02 1.832e+02 2.198e+02 3.821e+02, threshold=3.665e+02, percent-clipped=0.0 2023-03-27 05:45:05,791 INFO [finetune.py:976] (3/7) Epoch 24, batch 5350, loss[loss=0.1679, simple_loss=0.2447, pruned_loss=0.04556, over 4886.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.247, pruned_loss=0.05103, over 956363.48 frames. ], batch size: 32, lr: 3.03e-03, grad_scale: 32.0 2023-03-27 05:45:38,835 INFO [finetune.py:976] (3/7) Epoch 24, batch 5400, loss[loss=0.2015, simple_loss=0.2695, pruned_loss=0.06669, over 4820.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2443, pruned_loss=0.05033, over 953629.75 frames. ], batch size: 39, lr: 3.03e-03, grad_scale: 32.0 2023-03-27 05:45:57,170 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=137162.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:45:58,905 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.720e+01 1.449e+02 1.770e+02 2.209e+02 4.288e+02, threshold=3.541e+02, percent-clipped=1.0 2023-03-27 05:46:17,226 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2023-03-27 05:46:17,439 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4562, 2.3508, 1.8982, 2.4428, 2.3254, 2.0522, 2.6869, 2.5521], device='cuda:3'), covar=tensor([0.1309, 0.2018, 0.3041, 0.2353, 0.2516, 0.1698, 0.2933, 0.1552], device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0189, 0.0235, 0.0253, 0.0249, 0.0205, 0.0214, 0.0201], 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') 2023-03-27 05:46:22,348 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2807, 2.1989, 2.2915, 1.5493, 2.1501, 2.4235, 2.2956, 1.9008], device='cuda:3'), covar=tensor([0.0583, 0.0601, 0.0664, 0.0860, 0.0642, 0.0718, 0.0649, 0.1002], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0134, 0.0136, 0.0116, 0.0123, 0.0135, 0.0135, 0.0157], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 05:46:22,825 INFO [finetune.py:976] (3/7) Epoch 24, batch 5450, loss[loss=0.1815, simple_loss=0.2451, pruned_loss=0.05891, over 4921.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2408, pruned_loss=0.0491, over 954991.55 frames. ], batch size: 37, lr: 3.03e-03, grad_scale: 32.0 2023-03-27 05:46:52,457 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.5999, 1.6252, 1.6092, 0.8594, 1.7540, 1.9120, 1.9204, 1.4645], device='cuda:3'), covar=tensor([0.1005, 0.0684, 0.0506, 0.0566, 0.0450, 0.0654, 0.0356, 0.0761], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0150, 0.0128, 0.0123, 0.0132, 0.0131, 0.0143, 0.0150], device='cuda:3'), out_proj_covar=tensor([9.0207e-05, 1.0810e-04, 9.1438e-05, 8.6397e-05, 9.2903e-05, 9.3140e-05, 1.0202e-04, 1.0724e-04], device='cuda:3') 2023-03-27 05:46:53,079 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4793, 1.3375, 1.4701, 0.7837, 1.4603, 1.4947, 1.4456, 1.2903], device='cuda:3'), covar=tensor([0.0492, 0.0759, 0.0648, 0.0827, 0.1046, 0.0665, 0.0596, 0.1161], device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0134, 0.0136, 0.0117, 0.0124, 0.0136, 0.0136, 0.0158], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 05:46:55,994 INFO [finetune.py:976] (3/7) Epoch 24, batch 5500, loss[loss=0.1417, simple_loss=0.2036, pruned_loss=0.0399, over 4766.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2384, pruned_loss=0.04845, over 954536.05 frames. ], batch size: 27, lr: 3.03e-03, grad_scale: 32.0 2023-03-27 05:46:57,016 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-27 05:47:13,426 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.031e+02 1.493e+02 1.897e+02 2.213e+02 3.719e+02, threshold=3.794e+02, percent-clipped=2.0 2023-03-27 05:47:25,696 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.43 vs. limit=5.0 2023-03-27 05:47:36,857 INFO [finetune.py:976] (3/7) Epoch 24, batch 5550, loss[loss=0.1731, simple_loss=0.2448, pruned_loss=0.05069, over 4759.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2405, pruned_loss=0.04931, over 952869.99 frames. ], batch size: 27, lr: 3.03e-03, grad_scale: 32.0 2023-03-27 05:47:47,664 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=137299.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:47:50,453 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-27 05:48:20,577 INFO [finetune.py:976] (3/7) Epoch 24, batch 5600, loss[loss=0.2211, simple_loss=0.281, pruned_loss=0.08059, over 4818.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2445, pruned_loss=0.05043, over 952668.44 frames. ], batch size: 51, lr: 3.03e-03, grad_scale: 32.0 2023-03-27 05:48:26,395 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=137347.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:48:37,704 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.121e+02 1.584e+02 1.843e+02 2.259e+02 3.753e+02, threshold=3.686e+02, percent-clipped=0.0 2023-03-27 05:48:51,115 INFO [finetune.py:976] (3/7) Epoch 24, batch 5650, loss[loss=0.1219, simple_loss=0.1881, pruned_loss=0.02788, over 4198.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2465, pruned_loss=0.05084, over 949760.21 frames. ], batch size: 18, lr: 3.03e-03, grad_scale: 32.0 2023-03-27 05:48:52,083 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-03-27 05:49:19,334 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.0845, 2.8172, 2.4464, 1.4827, 2.6632, 2.4532, 2.2230, 2.6490], device='cuda:3'), covar=tensor([0.0621, 0.0657, 0.1211, 0.1660, 0.1055, 0.1558, 0.1721, 0.0720], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0191, 0.0200, 0.0181, 0.0210, 0.0210, 0.0224, 0.0196], 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') 2023-03-27 05:49:20,949 INFO [finetune.py:976] (3/7) Epoch 24, batch 5700, loss[loss=0.1448, simple_loss=0.2094, pruned_loss=0.04011, over 4256.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2438, pruned_loss=0.05054, over 934749.55 frames. ], batch size: 18, lr: 3.03e-03, grad_scale: 32.0 2023-03-27 05:49:35,729 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=137462.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:49:52,033 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.393e+01 1.414e+02 1.686e+02 2.138e+02 3.465e+02, threshold=3.373e+02, percent-clipped=0.0 2023-03-27 05:49:52,049 INFO [finetune.py:976] (3/7) Epoch 25, batch 0, loss[loss=0.2073, simple_loss=0.2713, pruned_loss=0.07163, over 4814.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2713, pruned_loss=0.07163, over 4814.00 frames. ], batch size: 33, lr: 3.03e-03, grad_scale: 32.0 2023-03-27 05:49:52,049 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-27 05:49:58,513 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6844, 3.6447, 3.4055, 1.4973, 3.6134, 2.7831, 0.6593, 2.4846], device='cuda:3'), covar=tensor([0.1871, 0.1725, 0.1507, 0.3453, 0.1005, 0.1012, 0.3789, 0.1461], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0177, 0.0160, 0.0129, 0.0161, 0.0123, 0.0147, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 05:50:06,683 INFO [finetune.py:1010] (3/7) Epoch 25, validation: loss=0.1587, simple_loss=0.2267, pruned_loss=0.04536, over 2265189.00 frames. 2023-03-27 05:50:06,683 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6469MB 2023-03-27 05:50:46,518 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=137510.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:50:50,024 INFO [finetune.py:976] (3/7) Epoch 25, batch 50, loss[loss=0.1312, simple_loss=0.2036, pruned_loss=0.0294, over 4752.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2477, pruned_loss=0.05066, over 217001.74 frames. ], batch size: 54, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 05:50:52,333 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4811, 1.3800, 1.3824, 1.3648, 0.9158, 2.2825, 0.8165, 1.2341], device='cuda:3'), covar=tensor([0.3235, 0.2581, 0.2246, 0.2525, 0.1791, 0.0347, 0.2791, 0.1343], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0116, 0.0121, 0.0123, 0.0113, 0.0095, 0.0094, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 05:50:54,350 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.18 vs. limit=5.0 2023-03-27 05:51:06,082 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.0732, 1.2989, 1.3747, 1.2873, 1.3543, 2.5275, 1.1199, 1.3400], device='cuda:3'), covar=tensor([0.1199, 0.2216, 0.1107, 0.1037, 0.1933, 0.0407, 0.1949, 0.2285], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0081, 0.0073, 0.0075, 0.0090, 0.0080, 0.0085, 0.0079], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 05:51:09,921 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6263, 1.5287, 2.1001, 3.3584, 2.1993, 2.3541, 1.0758, 2.7967], device='cuda:3'), covar=tensor([0.1933, 0.1799, 0.1547, 0.0830, 0.0929, 0.1387, 0.2105, 0.0543], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0116, 0.0133, 0.0163, 0.0101, 0.0136, 0.0124, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 05:51:12,267 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.0422, 0.9934, 0.9426, 0.3395, 0.8779, 1.1847, 1.1931, 0.9874], device='cuda:3'), covar=tensor([0.0969, 0.0607, 0.0594, 0.0592, 0.0691, 0.0634, 0.0430, 0.0733], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0150, 0.0127, 0.0123, 0.0131, 0.0130, 0.0142, 0.0149], device='cuda:3'), out_proj_covar=tensor([8.9609e-05, 1.0756e-04, 9.0843e-05, 8.6264e-05, 9.2124e-05, 9.2654e-05, 1.0144e-04, 1.0627e-04], device='cuda:3') 2023-03-27 05:51:25,239 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.123e+02 1.580e+02 1.851e+02 2.170e+02 4.183e+02, threshold=3.702e+02, percent-clipped=2.0 2023-03-27 05:51:25,255 INFO [finetune.py:976] (3/7) Epoch 25, batch 100, loss[loss=0.171, simple_loss=0.2379, pruned_loss=0.05208, over 4827.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2428, pruned_loss=0.05051, over 381254.55 frames. ], batch size: 30, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 05:51:59,266 INFO [finetune.py:976] (3/7) Epoch 25, batch 150, loss[loss=0.2086, simple_loss=0.2695, pruned_loss=0.07388, over 4863.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2389, pruned_loss=0.05008, over 508605.09 frames. ], batch size: 31, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 05:52:09,254 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.11 vs. limit=2.0 2023-03-27 05:52:33,562 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.113e+02 1.544e+02 1.791e+02 2.141e+02 4.771e+02, threshold=3.582e+02, percent-clipped=2.0 2023-03-27 05:52:33,578 INFO [finetune.py:976] (3/7) Epoch 25, batch 200, loss[loss=0.1831, simple_loss=0.2546, pruned_loss=0.0558, over 4750.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2386, pruned_loss=0.04989, over 607657.01 frames. ], batch size: 54, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 05:53:26,995 INFO [finetune.py:976] (3/7) Epoch 25, batch 250, loss[loss=0.1982, simple_loss=0.2767, pruned_loss=0.05986, over 4738.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2388, pruned_loss=0.04906, over 685806.30 frames. ], batch size: 59, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 05:53:31,855 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1373, 2.1577, 1.8089, 2.0331, 1.9577, 1.9613, 2.0375, 2.7278], device='cuda:3'), covar=tensor([0.3425, 0.3481, 0.3022, 0.3292, 0.3555, 0.2329, 0.3381, 0.1473], device='cuda:3'), in_proj_covar=tensor([0.0291, 0.0265, 0.0236, 0.0277, 0.0260, 0.0229, 0.0257, 0.0239], 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') 2023-03-27 05:54:00,392 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.518e+01 1.603e+02 1.998e+02 2.287e+02 4.515e+02, threshold=3.995e+02, percent-clipped=2.0 2023-03-27 05:54:00,408 INFO [finetune.py:976] (3/7) Epoch 25, batch 300, loss[loss=0.1601, simple_loss=0.2308, pruned_loss=0.0447, over 4774.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2428, pruned_loss=0.04986, over 746701.20 frames. ], batch size: 26, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 05:54:01,153 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7478, 1.1419, 1.8537, 1.8076, 1.6028, 1.5636, 1.7235, 1.6974], device='cuda:3'), covar=tensor([0.3764, 0.3888, 0.3102, 0.3335, 0.4736, 0.3769, 0.4271, 0.3181], device='cuda:3'), in_proj_covar=tensor([0.0262, 0.0246, 0.0266, 0.0291, 0.0291, 0.0267, 0.0297, 0.0249], 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') 2023-03-27 05:54:33,830 INFO [finetune.py:976] (3/7) Epoch 25, batch 350, loss[loss=0.1843, simple_loss=0.2551, pruned_loss=0.05676, over 4732.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2451, pruned_loss=0.05074, over 793053.67 frames. ], batch size: 54, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 05:55:00,346 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-03-27 05:55:07,123 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.080e+02 1.541e+02 1.822e+02 2.129e+02 2.910e+02, threshold=3.644e+02, percent-clipped=0.0 2023-03-27 05:55:07,139 INFO [finetune.py:976] (3/7) Epoch 25, batch 400, loss[loss=0.1323, simple_loss=0.2113, pruned_loss=0.02664, over 4761.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2449, pruned_loss=0.05046, over 829376.82 frames. ], batch size: 28, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 05:55:31,422 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137892.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:55:33,187 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137895.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:55:53,569 INFO [finetune.py:976] (3/7) Epoch 25, batch 450, loss[loss=0.1944, simple_loss=0.2572, pruned_loss=0.0658, over 4910.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2445, pruned_loss=0.0509, over 857305.67 frames. ], batch size: 35, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 05:55:54,829 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3097, 1.5262, 0.9102, 1.9325, 2.6185, 1.8803, 1.8462, 1.9641], device='cuda:3'), covar=tensor([0.1370, 0.1990, 0.1844, 0.1183, 0.1646, 0.1700, 0.1320, 0.1949], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0093, 0.0109, 0.0092, 0.0118, 0.0092, 0.0097, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-27 05:56:19,218 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137953.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:56:20,974 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137956.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:56:26,866 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.118e+02 1.489e+02 1.801e+02 2.267e+02 5.324e+02, threshold=3.602e+02, percent-clipped=3.0 2023-03-27 05:56:26,882 INFO [finetune.py:976] (3/7) Epoch 25, batch 500, loss[loss=0.1795, simple_loss=0.2518, pruned_loss=0.05365, over 4921.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2421, pruned_loss=0.04979, over 879003.67 frames. ], batch size: 37, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 05:56:57,229 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-03-27 05:56:58,230 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4867, 1.5957, 1.3461, 1.6375, 1.8897, 1.7555, 1.5573, 1.4203], device='cuda:3'), covar=tensor([0.0428, 0.0345, 0.0632, 0.0292, 0.0212, 0.0499, 0.0298, 0.0397], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0106, 0.0145, 0.0112, 0.0101, 0.0114, 0.0102, 0.0113], device='cuda:3'), out_proj_covar=tensor([7.7783e-05, 8.1419e-05, 1.1367e-04, 8.5842e-05, 7.8596e-05, 8.3952e-05, 7.5641e-05, 8.5857e-05], device='cuda:3') 2023-03-27 05:57:01,547 INFO [finetune.py:976] (3/7) Epoch 25, batch 550, loss[loss=0.1956, simple_loss=0.2745, pruned_loss=0.05832, over 4915.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2386, pruned_loss=0.04844, over 896033.10 frames. ], batch size: 37, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 05:57:34,657 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.805e+01 1.427e+02 1.740e+02 1.994e+02 3.808e+02, threshold=3.480e+02, percent-clipped=1.0 2023-03-27 05:57:34,673 INFO [finetune.py:976] (3/7) Epoch 25, batch 600, loss[loss=0.1337, simple_loss=0.1994, pruned_loss=0.03401, over 3502.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2389, pruned_loss=0.0486, over 907968.08 frames. ], batch size: 15, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 05:58:07,364 INFO [finetune.py:976] (3/7) Epoch 25, batch 650, loss[loss=0.241, simple_loss=0.311, pruned_loss=0.08546, over 4826.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2456, pruned_loss=0.05158, over 918156.29 frames. ], batch size: 40, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 05:58:34,861 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.4589, 1.5570, 1.6088, 0.8603, 1.7300, 1.9148, 1.8776, 1.4096], device='cuda:3'), covar=tensor([0.0855, 0.0640, 0.0577, 0.0567, 0.0440, 0.0535, 0.0394, 0.0681], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0149, 0.0127, 0.0122, 0.0131, 0.0129, 0.0141, 0.0148], device='cuda:3'), out_proj_covar=tensor([8.9281e-05, 1.0735e-04, 9.0633e-05, 8.5889e-05, 9.1829e-05, 9.2007e-05, 1.0089e-04, 1.0580e-04], device='cuda:3') 2023-03-27 05:58:59,108 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.045e+02 1.613e+02 1.882e+02 2.325e+02 5.163e+02, threshold=3.765e+02, percent-clipped=4.0 2023-03-27 05:58:59,124 INFO [finetune.py:976] (3/7) Epoch 25, batch 700, loss[loss=0.2464, simple_loss=0.3006, pruned_loss=0.09607, over 4132.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2473, pruned_loss=0.05164, over 924468.19 frames. ], batch size: 65, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 05:59:26,285 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-03-27 05:59:32,528 INFO [finetune.py:976] (3/7) Epoch 25, batch 750, loss[loss=0.1423, simple_loss=0.2283, pruned_loss=0.02811, over 4814.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2474, pruned_loss=0.05112, over 931903.14 frames. ], batch size: 39, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 05:59:53,522 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138248.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 05:59:55,845 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138251.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:00:05,193 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.386e+01 1.513e+02 1.803e+02 2.270e+02 6.862e+02, threshold=3.605e+02, percent-clipped=3.0 2023-03-27 06:00:05,209 INFO [finetune.py:976] (3/7) Epoch 25, batch 800, loss[loss=0.1773, simple_loss=0.2538, pruned_loss=0.05043, over 4848.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2459, pruned_loss=0.04994, over 938194.33 frames. ], batch size: 44, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 06:00:08,351 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138270.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:00:19,836 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.52 vs. limit=2.0 2023-03-27 06:00:38,347 INFO [finetune.py:976] (3/7) Epoch 25, batch 850, loss[loss=0.1665, simple_loss=0.242, pruned_loss=0.0455, over 4731.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2441, pruned_loss=0.04952, over 940929.27 frames. ], batch size: 59, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 06:00:44,521 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138322.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:00:45,807 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9338, 1.7155, 1.5807, 1.3379, 1.7131, 1.7250, 1.6934, 2.2636], device='cuda:3'), covar=tensor([0.3609, 0.3613, 0.3050, 0.3403, 0.3626, 0.2334, 0.3369, 0.1775], device='cuda:3'), in_proj_covar=tensor([0.0291, 0.0264, 0.0235, 0.0276, 0.0259, 0.0229, 0.0257, 0.0238], 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') 2023-03-27 06:00:54,537 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138331.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:01:24,851 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.023e+02 1.388e+02 1.706e+02 2.091e+02 3.563e+02, threshold=3.412e+02, percent-clipped=0.0 2023-03-27 06:01:24,867 INFO [finetune.py:976] (3/7) Epoch 25, batch 900, loss[loss=0.1867, simple_loss=0.2557, pruned_loss=0.05886, over 4766.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2411, pruned_loss=0.04829, over 945238.12 frames. ], batch size: 26, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 06:01:34,654 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4964, 2.5307, 2.2623, 2.5615, 2.4859, 4.9946, 2.4152, 2.8505], device='cuda:3'), covar=tensor([0.2670, 0.2070, 0.1777, 0.1838, 0.1060, 0.0151, 0.1799, 0.0950], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0121, 0.0124, 0.0113, 0.0096, 0.0094, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 06:01:36,364 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138383.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:01:55,436 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8879, 1.7921, 1.5789, 2.0471, 2.2393, 2.0416, 1.5154, 1.4894], device='cuda:3'), covar=tensor([0.2367, 0.2098, 0.2010, 0.1596, 0.1731, 0.1220, 0.2564, 0.2081], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0213, 0.0216, 0.0198, 0.0245, 0.0192, 0.0218, 0.0206], 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') 2023-03-27 06:01:57,678 INFO [finetune.py:976] (3/7) Epoch 25, batch 950, loss[loss=0.1479, simple_loss=0.2174, pruned_loss=0.03916, over 4823.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2398, pruned_loss=0.04858, over 948924.68 frames. ], batch size: 41, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 06:02:03,932 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4577, 1.5505, 1.2536, 1.4255, 1.7952, 1.7424, 1.5601, 1.3962], device='cuda:3'), covar=tensor([0.0373, 0.0294, 0.0593, 0.0309, 0.0210, 0.0494, 0.0323, 0.0397], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0107, 0.0146, 0.0112, 0.0101, 0.0113, 0.0102, 0.0113], device='cuda:3'), out_proj_covar=tensor([7.7886e-05, 8.1737e-05, 1.1419e-04, 8.5897e-05, 7.8502e-05, 8.3867e-05, 7.5927e-05, 8.6010e-05], device='cuda:3') 2023-03-27 06:02:30,846 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.004e+02 1.447e+02 1.781e+02 2.298e+02 4.571e+02, threshold=3.563e+02, percent-clipped=3.0 2023-03-27 06:02:30,862 INFO [finetune.py:976] (3/7) Epoch 25, batch 1000, loss[loss=0.2075, simple_loss=0.2739, pruned_loss=0.07057, over 4759.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2412, pruned_loss=0.04958, over 950730.53 frames. ], batch size: 59, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 06:02:35,740 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8397, 1.7099, 1.5026, 1.2895, 1.6554, 1.6429, 1.6154, 2.1743], device='cuda:3'), covar=tensor([0.3801, 0.3654, 0.3177, 0.3525, 0.3694, 0.2222, 0.3464, 0.1763], device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0263, 0.0234, 0.0275, 0.0258, 0.0228, 0.0256, 0.0237], 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') 2023-03-27 06:03:03,759 INFO [finetune.py:976] (3/7) Epoch 25, batch 1050, loss[loss=0.175, simple_loss=0.2551, pruned_loss=0.04745, over 4808.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2432, pruned_loss=0.04952, over 951262.36 frames. ], batch size: 41, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 06:03:10,893 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5294, 2.7095, 2.5347, 1.7742, 2.6834, 2.9063, 2.7397, 2.2572], device='cuda:3'), covar=tensor([0.0618, 0.0645, 0.0768, 0.0901, 0.0658, 0.0692, 0.0652, 0.1224], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0137, 0.0140, 0.0119, 0.0126, 0.0139, 0.0138, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 06:03:25,373 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138548.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:03:25,991 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9534, 1.5947, 2.2463, 1.5013, 2.1550, 2.2548, 1.5605, 2.2895], device='cuda:3'), covar=tensor([0.1062, 0.1955, 0.1054, 0.1681, 0.0627, 0.1080, 0.2592, 0.0741], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0207, 0.0192, 0.0190, 0.0174, 0.0214, 0.0217, 0.0199], 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') 2023-03-27 06:03:27,130 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138551.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:03:39,079 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.051e+02 1.570e+02 1.874e+02 2.177e+02 7.699e+02, threshold=3.747e+02, percent-clipped=3.0 2023-03-27 06:03:39,095 INFO [finetune.py:976] (3/7) Epoch 25, batch 1100, loss[loss=0.1889, simple_loss=0.2563, pruned_loss=0.0608, over 4881.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2456, pruned_loss=0.05053, over 951317.41 frames. ], batch size: 32, lr: 3.02e-03, grad_scale: 32.0 2023-03-27 06:04:17,870 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=138596.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:04:19,635 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=138599.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:04:29,871 INFO [finetune.py:976] (3/7) Epoch 25, batch 1150, loss[loss=0.1369, simple_loss=0.2117, pruned_loss=0.03107, over 4816.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2474, pruned_loss=0.05134, over 954494.46 frames. ], batch size: 25, lr: 3.02e-03, grad_scale: 64.0 2023-03-27 06:04:39,029 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138626.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:05:03,419 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.832e+01 1.520e+02 1.733e+02 2.222e+02 3.582e+02, threshold=3.466e+02, percent-clipped=0.0 2023-03-27 06:05:03,435 INFO [finetune.py:976] (3/7) Epoch 25, batch 1200, loss[loss=0.2133, simple_loss=0.2697, pruned_loss=0.07849, over 4873.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2458, pruned_loss=0.05086, over 955562.34 frames. ], batch size: 34, lr: 3.02e-03, grad_scale: 64.0 2023-03-27 06:05:13,801 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138678.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:05:15,605 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138681.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:05:18,695 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138686.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:05:25,320 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.4461, 1.5251, 1.5417, 0.8177, 1.6254, 1.8538, 1.8541, 1.4157], device='cuda:3'), covar=tensor([0.0889, 0.0581, 0.0505, 0.0528, 0.0480, 0.0512, 0.0303, 0.0692], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0149, 0.0127, 0.0122, 0.0132, 0.0130, 0.0141, 0.0148], device='cuda:3'), out_proj_covar=tensor([8.9339e-05, 1.0706e-04, 9.0676e-05, 8.6174e-05, 9.2435e-05, 9.2189e-05, 1.0094e-04, 1.0574e-04], device='cuda:3') 2023-03-27 06:05:25,909 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138697.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:05:37,221 INFO [finetune.py:976] (3/7) Epoch 25, batch 1250, loss[loss=0.183, simple_loss=0.2475, pruned_loss=0.05922, over 4897.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2443, pruned_loss=0.05107, over 952740.68 frames. ], batch size: 46, lr: 3.02e-03, grad_scale: 64.0 2023-03-27 06:05:55,608 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138742.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:05:59,136 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138747.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:06:05,711 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138758.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:06:12,322 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.092e+02 1.434e+02 1.700e+02 2.124e+02 3.591e+02, threshold=3.400e+02, percent-clipped=1.0 2023-03-27 06:06:12,338 INFO [finetune.py:976] (3/7) Epoch 25, batch 1300, loss[loss=0.1893, simple_loss=0.2607, pruned_loss=0.05897, over 4825.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.241, pruned_loss=0.04992, over 951526.85 frames. ], batch size: 39, lr: 3.02e-03, grad_scale: 64.0 2023-03-27 06:06:57,378 INFO [finetune.py:976] (3/7) Epoch 25, batch 1350, loss[loss=0.1574, simple_loss=0.2461, pruned_loss=0.03433, over 4795.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2423, pruned_loss=0.05055, over 952843.24 frames. ], batch size: 29, lr: 3.02e-03, grad_scale: 64.0 2023-03-27 06:07:31,275 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.011e+02 1.573e+02 1.999e+02 2.321e+02 4.595e+02, threshold=3.999e+02, percent-clipped=3.0 2023-03-27 06:07:31,291 INFO [finetune.py:976] (3/7) Epoch 25, batch 1400, loss[loss=0.1806, simple_loss=0.2612, pruned_loss=0.04996, over 4800.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.246, pruned_loss=0.05167, over 953484.39 frames. ], batch size: 51, lr: 3.02e-03, grad_scale: 64.0 2023-03-27 06:07:37,339 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.7163, 1.7036, 1.6439, 1.0057, 1.9230, 2.1187, 2.0275, 1.5007], device='cuda:3'), covar=tensor([0.0991, 0.0690, 0.0561, 0.0631, 0.0471, 0.0520, 0.0375, 0.0711], device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0150, 0.0128, 0.0123, 0.0132, 0.0131, 0.0143, 0.0149], device='cuda:3'), out_proj_covar=tensor([9.0075e-05, 1.0785e-04, 9.1346e-05, 8.6950e-05, 9.2839e-05, 9.2997e-05, 1.0213e-04, 1.0668e-04], device='cuda:3') 2023-03-27 06:07:46,616 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5399, 1.3686, 2.1072, 1.9060, 1.8043, 3.9745, 1.3881, 1.6243], device='cuda:3'), covar=tensor([0.1043, 0.1846, 0.1353, 0.0912, 0.1546, 0.0219, 0.1526, 0.1811], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0082, 0.0073, 0.0076, 0.0091, 0.0080, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 06:08:01,090 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138909.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:08:04,578 INFO [finetune.py:976] (3/7) Epoch 25, batch 1450, loss[loss=0.1702, simple_loss=0.2497, pruned_loss=0.04531, over 4818.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2473, pruned_loss=0.05183, over 951362.52 frames. ], batch size: 39, lr: 3.01e-03, grad_scale: 64.0 2023-03-27 06:08:11,164 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1696, 1.9847, 1.4745, 0.6189, 1.6938, 1.8111, 1.7417, 1.7820], device='cuda:3'), covar=tensor([0.0839, 0.0738, 0.1522, 0.1912, 0.1307, 0.2150, 0.2257, 0.0922], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0191, 0.0200, 0.0180, 0.0208, 0.0209, 0.0222, 0.0194], 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') 2023-03-27 06:08:11,764 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138926.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:08:15,131 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.7431, 1.5201, 1.5109, 0.8194, 1.6575, 1.8110, 1.7716, 1.4702], device='cuda:3'), covar=tensor([0.0878, 0.0657, 0.0434, 0.0526, 0.0454, 0.0504, 0.0318, 0.0551], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0149, 0.0127, 0.0123, 0.0131, 0.0130, 0.0142, 0.0148], device='cuda:3'), out_proj_covar=tensor([8.9607e-05, 1.0709e-04, 9.0805e-05, 8.6405e-05, 9.2241e-05, 9.2393e-05, 1.0150e-04, 1.0585e-04], device='cuda:3') 2023-03-27 06:08:23,886 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6761, 1.7251, 1.4730, 1.9174, 2.1869, 1.9886, 1.6324, 1.3660], device='cuda:3'), covar=tensor([0.2217, 0.1950, 0.1993, 0.1506, 0.1901, 0.1197, 0.2511, 0.1944], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0213, 0.0216, 0.0199, 0.0245, 0.0192, 0.0218, 0.0206], 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') 2023-03-27 06:08:38,070 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.768e+01 1.472e+02 1.793e+02 2.201e+02 3.947e+02, threshold=3.587e+02, percent-clipped=0.0 2023-03-27 06:08:38,086 INFO [finetune.py:976] (3/7) Epoch 25, batch 1500, loss[loss=0.2005, simple_loss=0.2695, pruned_loss=0.06572, over 4832.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2484, pruned_loss=0.05201, over 951813.58 frames. ], batch size: 30, lr: 3.01e-03, grad_scale: 64.0 2023-03-27 06:08:41,666 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138970.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 06:08:44,020 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=138974.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:08:46,501 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138978.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:09:23,189 INFO [finetune.py:976] (3/7) Epoch 25, batch 1550, loss[loss=0.1955, simple_loss=0.2524, pruned_loss=0.06935, over 4891.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2489, pruned_loss=0.05235, over 953012.66 frames. ], batch size: 35, lr: 3.01e-03, grad_scale: 64.0 2023-03-27 06:09:34,900 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=139026.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:09:47,267 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139037.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:09:50,291 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139042.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:09:57,453 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139053.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:10:02,886 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1537, 1.8975, 2.1747, 2.1910, 1.8773, 1.8825, 2.1537, 1.9969], device='cuda:3'), covar=tensor([0.3950, 0.3849, 0.3055, 0.3842, 0.4869, 0.4075, 0.4656, 0.2973], device='cuda:3'), in_proj_covar=tensor([0.0261, 0.0245, 0.0265, 0.0290, 0.0291, 0.0267, 0.0297, 0.0248], 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') 2023-03-27 06:10:05,109 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.522e+01 1.375e+02 1.683e+02 2.077e+02 3.862e+02, threshold=3.366e+02, percent-clipped=3.0 2023-03-27 06:10:05,125 INFO [finetune.py:976] (3/7) Epoch 25, batch 1600, loss[loss=0.1651, simple_loss=0.2267, pruned_loss=0.05175, over 4889.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2459, pruned_loss=0.0511, over 954312.59 frames. ], batch size: 43, lr: 3.01e-03, grad_scale: 64.0 2023-03-27 06:10:38,952 INFO [finetune.py:976] (3/7) Epoch 25, batch 1650, loss[loss=0.1854, simple_loss=0.2379, pruned_loss=0.06644, over 4867.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2431, pruned_loss=0.0504, over 956150.70 frames. ], batch size: 31, lr: 3.01e-03, grad_scale: 64.0 2023-03-27 06:11:12,572 INFO [finetune.py:976] (3/7) Epoch 25, batch 1700, loss[loss=0.1986, simple_loss=0.257, pruned_loss=0.07009, over 4901.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2408, pruned_loss=0.04952, over 957199.88 frames. ], batch size: 35, lr: 3.01e-03, grad_scale: 32.0 2023-03-27 06:11:13,176 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.831e+01 1.435e+02 1.759e+02 2.193e+02 3.727e+02, threshold=3.518e+02, percent-clipped=3.0 2023-03-27 06:11:52,851 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5796, 1.6410, 1.3086, 1.6588, 1.8766, 1.8225, 1.6100, 1.4050], device='cuda:3'), covar=tensor([0.0358, 0.0316, 0.0658, 0.0314, 0.0215, 0.0543, 0.0363, 0.0445], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0107, 0.0146, 0.0112, 0.0101, 0.0114, 0.0103, 0.0113], device='cuda:3'), out_proj_covar=tensor([7.7999e-05, 8.1901e-05, 1.1406e-04, 8.5684e-05, 7.8564e-05, 8.4341e-05, 7.6164e-05, 8.5778e-05], device='cuda:3') 2023-03-27 06:11:56,394 INFO [finetune.py:976] (3/7) Epoch 25, batch 1750, loss[loss=0.2352, simple_loss=0.2989, pruned_loss=0.08568, over 4808.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2416, pruned_loss=0.04968, over 957678.73 frames. ], batch size: 51, lr: 3.01e-03, grad_scale: 32.0 2023-03-27 06:12:20,694 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139238.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:12:21,982 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0 2023-03-27 06:12:31,172 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6217, 2.4792, 2.0362, 2.6772, 2.4724, 2.2354, 2.9893, 2.6839], device='cuda:3'), covar=tensor([0.1297, 0.2236, 0.2928, 0.2724, 0.2475, 0.1541, 0.2995, 0.1658], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0189, 0.0234, 0.0253, 0.0249, 0.0204, 0.0214, 0.0201], 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') 2023-03-27 06:12:39,405 INFO [finetune.py:976] (3/7) Epoch 25, batch 1800, loss[loss=0.1715, simple_loss=0.244, pruned_loss=0.04951, over 4911.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2441, pruned_loss=0.05001, over 957976.79 frames. ], batch size: 37, lr: 3.01e-03, grad_scale: 32.0 2023-03-27 06:12:39,473 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139265.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 06:12:39,956 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.974e+01 1.553e+02 1.833e+02 2.131e+02 4.022e+02, threshold=3.667e+02, percent-clipped=3.0 2023-03-27 06:12:46,664 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139276.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:12:55,156 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139289.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:13:02,281 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139299.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:13:04,600 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.82 vs. limit=5.0 2023-03-27 06:13:13,308 INFO [finetune.py:976] (3/7) Epoch 25, batch 1850, loss[loss=0.1549, simple_loss=0.2436, pruned_loss=0.03307, over 4889.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2456, pruned_loss=0.05105, over 956132.40 frames. ], batch size: 36, lr: 3.01e-03, grad_scale: 32.0 2023-03-27 06:13:27,874 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139337.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:13:27,898 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139337.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 06:13:30,858 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139342.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:13:36,178 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139350.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:13:38,379 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139353.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:13:46,548 INFO [finetune.py:976] (3/7) Epoch 25, batch 1900, loss[loss=0.201, simple_loss=0.2824, pruned_loss=0.05981, over 4814.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.246, pruned_loss=0.05039, over 955415.42 frames. ], batch size: 40, lr: 3.01e-03, grad_scale: 32.0 2023-03-27 06:13:46,656 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5488, 1.4556, 1.4092, 1.5460, 0.9549, 3.1859, 1.1836, 1.6447], device='cuda:3'), covar=tensor([0.3202, 0.2480, 0.2207, 0.2391, 0.1900, 0.0230, 0.2675, 0.1294], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0121, 0.0124, 0.0113, 0.0096, 0.0094, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 06:13:47,140 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.007e+02 1.553e+02 1.835e+02 2.150e+02 3.557e+02, threshold=3.671e+02, percent-clipped=0.0 2023-03-27 06:13:59,661 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=139385.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:14:03,129 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=139390.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:14:10,503 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=139401.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:14:18,138 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5383, 1.2686, 2.0352, 3.2769, 2.0975, 2.4545, 1.2045, 2.8620], device='cuda:3'), covar=tensor([0.2043, 0.1970, 0.1497, 0.0861, 0.1062, 0.1348, 0.1930, 0.0569], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0117, 0.0134, 0.0164, 0.0102, 0.0137, 0.0125, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 06:14:19,883 INFO [finetune.py:976] (3/7) Epoch 25, batch 1950, loss[loss=0.1816, simple_loss=0.25, pruned_loss=0.05653, over 4795.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2453, pruned_loss=0.05021, over 956401.37 frames. ], batch size: 45, lr: 3.01e-03, grad_scale: 32.0 2023-03-27 06:14:23,689 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 2023-03-27 06:15:11,617 INFO [finetune.py:976] (3/7) Epoch 25, batch 2000, loss[loss=0.1738, simple_loss=0.2377, pruned_loss=0.05494, over 4907.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2438, pruned_loss=0.05007, over 956410.31 frames. ], batch size: 36, lr: 3.01e-03, grad_scale: 32.0 2023-03-27 06:15:12,705 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.253e+01 1.362e+02 1.721e+02 2.187e+02 3.038e+02, threshold=3.442e+02, percent-clipped=0.0 2023-03-27 06:15:15,218 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6017, 1.5375, 1.4876, 1.5926, 1.1909, 2.8079, 1.1393, 1.6488], device='cuda:3'), covar=tensor([0.3080, 0.2292, 0.2018, 0.2219, 0.1525, 0.0283, 0.2720, 0.1120], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0121, 0.0124, 0.0113, 0.0096, 0.0094, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 06:15:38,519 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-03-27 06:15:45,240 INFO [finetune.py:976] (3/7) Epoch 25, batch 2050, loss[loss=0.1707, simple_loss=0.2434, pruned_loss=0.04898, over 4820.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2399, pruned_loss=0.04852, over 957965.36 frames. ], batch size: 38, lr: 3.01e-03, grad_scale: 32.0 2023-03-27 06:16:18,436 INFO [finetune.py:976] (3/7) Epoch 25, batch 2100, loss[loss=0.1229, simple_loss=0.1972, pruned_loss=0.02434, over 4786.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2385, pruned_loss=0.04776, over 958528.88 frames. ], batch size: 26, lr: 3.01e-03, grad_scale: 16.0 2023-03-27 06:16:18,552 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139565.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 06:16:20,121 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.352e+01 1.449e+02 1.714e+02 2.109e+02 3.824e+02, threshold=3.428e+02, percent-clipped=2.0 2023-03-27 06:16:22,149 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7279, 1.0647, 1.7646, 1.7067, 1.5355, 1.4712, 1.6789, 1.6858], device='cuda:3'), covar=tensor([0.3601, 0.3459, 0.2732, 0.3070, 0.4116, 0.3524, 0.3619, 0.2790], device='cuda:3'), in_proj_covar=tensor([0.0263, 0.0246, 0.0266, 0.0291, 0.0291, 0.0268, 0.0297, 0.0249], 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') 2023-03-27 06:16:23,224 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139571.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:16:30,381 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139582.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:16:38,034 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139594.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:16:50,723 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=139613.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:16:52,380 INFO [finetune.py:976] (3/7) Epoch 25, batch 2150, loss[loss=0.2263, simple_loss=0.2917, pruned_loss=0.08046, over 4916.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2423, pruned_loss=0.04967, over 955655.39 frames. ], batch size: 36, lr: 3.01e-03, grad_scale: 16.0 2023-03-27 06:17:09,469 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.6833, 4.1052, 4.3162, 4.5234, 4.4821, 4.1899, 4.8074, 1.6308], device='cuda:3'), covar=tensor([0.0702, 0.0891, 0.0788, 0.0900, 0.1001, 0.1662, 0.0548, 0.6025], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0249, 0.0284, 0.0298, 0.0340, 0.0288, 0.0308, 0.0305], 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') 2023-03-27 06:17:09,471 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139632.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 06:17:09,503 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139632.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:17:19,888 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-27 06:17:21,661 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139643.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:17:27,568 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139645.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:17:35,298 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.04 vs. limit=5.0 2023-03-27 06:17:43,825 INFO [finetune.py:976] (3/7) Epoch 25, batch 2200, loss[loss=0.1757, simple_loss=0.2406, pruned_loss=0.05543, over 4761.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2459, pruned_loss=0.05046, over 957144.66 frames. ], batch size: 26, lr: 3.01e-03, grad_scale: 16.0 2023-03-27 06:17:45,448 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.088e+02 1.512e+02 1.789e+02 2.111e+02 3.462e+02, threshold=3.578e+02, percent-clipped=1.0 2023-03-27 06:17:57,530 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8719, 1.7537, 1.5031, 1.9670, 2.4250, 2.0254, 1.8142, 1.4510], device='cuda:3'), covar=tensor([0.2054, 0.1949, 0.1838, 0.1531, 0.1478, 0.1061, 0.2040, 0.1818], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0211, 0.0215, 0.0198, 0.0245, 0.0191, 0.0217, 0.0205], 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') 2023-03-27 06:18:17,069 INFO [finetune.py:976] (3/7) Epoch 25, batch 2250, loss[loss=0.1635, simple_loss=0.2391, pruned_loss=0.04393, over 4818.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2476, pruned_loss=0.05136, over 957124.51 frames. ], batch size: 33, lr: 3.01e-03, grad_scale: 16.0 2023-03-27 06:18:27,556 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8013, 1.7478, 1.5783, 1.8017, 1.4866, 4.4890, 1.5513, 2.0051], device='cuda:3'), covar=tensor([0.3294, 0.2424, 0.2233, 0.2387, 0.1562, 0.0109, 0.2468, 0.1211], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0121, 0.0123, 0.0113, 0.0096, 0.0094, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 06:18:50,822 INFO [finetune.py:976] (3/7) Epoch 25, batch 2300, loss[loss=0.141, simple_loss=0.2144, pruned_loss=0.03386, over 4842.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2485, pruned_loss=0.0516, over 955611.99 frames. ], batch size: 49, lr: 3.01e-03, grad_scale: 16.0 2023-03-27 06:18:52,006 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.130e+02 1.533e+02 1.822e+02 2.118e+02 3.916e+02, threshold=3.645e+02, percent-clipped=1.0 2023-03-27 06:18:53,812 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139769.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:18:57,318 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9660, 1.7589, 1.6627, 2.0842, 2.2099, 2.0149, 1.4303, 1.6406], device='cuda:3'), covar=tensor([0.1827, 0.1629, 0.1606, 0.1274, 0.1323, 0.0948, 0.2147, 0.1623], device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0209, 0.0213, 0.0196, 0.0242, 0.0189, 0.0215, 0.0203], 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') 2023-03-27 06:19:06,795 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139789.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:19:16,873 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139804.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:19:18,130 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139806.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:19:22,815 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5613, 1.5473, 1.8580, 1.2105, 1.6674, 1.8352, 1.4445, 1.9880], device='cuda:3'), covar=tensor([0.1336, 0.2102, 0.1147, 0.1792, 0.0880, 0.1269, 0.3158, 0.0844], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0208, 0.0193, 0.0191, 0.0176, 0.0215, 0.0218, 0.0200], 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') 2023-03-27 06:19:23,929 INFO [finetune.py:976] (3/7) Epoch 25, batch 2350, loss[loss=0.2364, simple_loss=0.282, pruned_loss=0.09543, over 4891.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2458, pruned_loss=0.05068, over 956947.85 frames. ], batch size: 35, lr: 3.01e-03, grad_scale: 16.0 2023-03-27 06:19:29,490 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-27 06:19:35,043 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139830.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:19:54,872 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139850.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:19:54,882 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139850.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:19:57,087 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-03-27 06:20:07,031 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0414, 1.7767, 2.4132, 1.4785, 2.1168, 2.3108, 1.6702, 2.4362], device='cuda:3'), covar=tensor([0.1403, 0.2252, 0.1501, 0.2126, 0.0927, 0.1479, 0.2916, 0.0832], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0208, 0.0193, 0.0191, 0.0176, 0.0214, 0.0217, 0.0200], 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') 2023-03-27 06:20:07,502 INFO [finetune.py:976] (3/7) Epoch 25, batch 2400, loss[loss=0.1605, simple_loss=0.224, pruned_loss=0.04846, over 4929.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.243, pruned_loss=0.05008, over 958643.42 frames. ], batch size: 38, lr: 3.01e-03, grad_scale: 16.0 2023-03-27 06:20:07,612 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139865.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:20:10,610 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.173e+01 1.419e+02 1.768e+02 2.081e+02 3.267e+02, threshold=3.536e+02, percent-clipped=0.0 2023-03-27 06:20:10,763 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139867.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:20:14,387 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7361, 1.7003, 1.8157, 1.0692, 1.9259, 1.8825, 1.9084, 1.5266], device='cuda:3'), covar=tensor([0.0597, 0.0791, 0.0686, 0.0937, 0.0650, 0.0670, 0.0566, 0.1174], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0136, 0.0139, 0.0119, 0.0125, 0.0137, 0.0138, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 06:20:35,972 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139894.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:20:47,264 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139911.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:20:50,037 INFO [finetune.py:976] (3/7) Epoch 25, batch 2450, loss[loss=0.2025, simple_loss=0.2676, pruned_loss=0.0687, over 4809.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2397, pruned_loss=0.04885, over 955143.03 frames. ], batch size: 39, lr: 3.01e-03, grad_scale: 16.0 2023-03-27 06:20:53,430 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-03-27 06:20:57,353 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139927.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:21:01,395 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139932.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 06:21:05,455 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139938.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:21:08,389 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=139942.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:21:10,175 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139945.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:21:22,671 INFO [finetune.py:976] (3/7) Epoch 25, batch 2500, loss[loss=0.1571, simple_loss=0.2308, pruned_loss=0.04172, over 4783.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2404, pruned_loss=0.04922, over 953782.15 frames. ], batch size: 29, lr: 3.01e-03, grad_scale: 16.0 2023-03-27 06:21:24,361 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.029e+02 1.523e+02 1.884e+02 2.422e+02 3.755e+02, threshold=3.768e+02, percent-clipped=3.0 2023-03-27 06:21:24,835 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-03-27 06:21:32,355 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=139980.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:21:42,135 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=139993.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:21:57,548 INFO [finetune.py:976] (3/7) Epoch 25, batch 2550, loss[loss=0.1932, simple_loss=0.2529, pruned_loss=0.06676, over 4909.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2453, pruned_loss=0.05103, over 954078.39 frames. ], batch size: 28, lr: 3.01e-03, grad_scale: 16.0 2023-03-27 06:22:10,454 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8347, 1.7941, 1.6196, 1.9101, 1.3873, 4.4781, 1.6522, 2.0221], device='cuda:3'), covar=tensor([0.3011, 0.2398, 0.2089, 0.2097, 0.1562, 0.0126, 0.2296, 0.1186], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0121, 0.0123, 0.0113, 0.0096, 0.0094, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 06:22:36,078 INFO [finetune.py:976] (3/7) Epoch 25, batch 2600, loss[loss=0.1827, simple_loss=0.2543, pruned_loss=0.05552, over 4827.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2484, pruned_loss=0.05242, over 953572.20 frames. ], batch size: 33, lr: 3.01e-03, grad_scale: 16.0 2023-03-27 06:22:36,209 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.4578, 1.5803, 1.5687, 0.9069, 1.7068, 1.8697, 1.8925, 1.3563], device='cuda:3'), covar=tensor([0.0941, 0.0637, 0.0535, 0.0506, 0.0395, 0.0593, 0.0325, 0.0777], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0148, 0.0128, 0.0122, 0.0131, 0.0129, 0.0142, 0.0147], device='cuda:3'), out_proj_covar=tensor([8.9459e-05, 1.0662e-04, 9.0991e-05, 8.6054e-05, 9.1587e-05, 9.1986e-05, 1.0094e-04, 1.0534e-04], device='cuda:3') 2023-03-27 06:22:42,050 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.019e+02 1.547e+02 1.969e+02 2.320e+02 4.703e+02, threshold=3.938e+02, percent-clipped=1.0 2023-03-27 06:22:44,790 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-03-27 06:23:11,713 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0392, 1.8224, 2.0785, 2.1367, 1.8307, 1.8254, 2.0758, 2.0096], device='cuda:3'), covar=tensor([0.4414, 0.4152, 0.3366, 0.3987, 0.5016, 0.4321, 0.4980, 0.2951], device='cuda:3'), in_proj_covar=tensor([0.0262, 0.0245, 0.0264, 0.0289, 0.0290, 0.0267, 0.0296, 0.0247], 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') 2023-03-27 06:23:14,131 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140103.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:23:19,347 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140110.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:23:22,260 INFO [finetune.py:976] (3/7) Epoch 25, batch 2650, loss[loss=0.1511, simple_loss=0.2385, pruned_loss=0.03181, over 4807.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2491, pruned_loss=0.05238, over 953498.16 frames. ], batch size: 40, lr: 3.01e-03, grad_scale: 16.0 2023-03-27 06:23:28,792 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140125.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:23:41,813 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140145.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:23:51,796 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140160.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:23:53,483 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140162.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:23:55,138 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140164.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:23:55,633 INFO [finetune.py:976] (3/7) Epoch 25, batch 2700, loss[loss=0.1755, simple_loss=0.2393, pruned_loss=0.05579, over 4799.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2474, pruned_loss=0.0513, over 954038.59 frames. ], batch size: 51, lr: 3.01e-03, grad_scale: 16.0 2023-03-27 06:23:56,848 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.003e+02 1.476e+02 1.708e+02 2.136e+02 4.297e+02, threshold=3.417e+02, percent-clipped=1.0 2023-03-27 06:23:59,423 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140171.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:24:22,839 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140206.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:24:28,675 INFO [finetune.py:976] (3/7) Epoch 25, batch 2750, loss[loss=0.1543, simple_loss=0.2287, pruned_loss=0.03996, over 4890.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.245, pruned_loss=0.05078, over 954207.19 frames. ], batch size: 35, lr: 3.01e-03, grad_scale: 16.0 2023-03-27 06:24:36,541 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140227.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:24:43,717 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140238.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:24:57,688 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5172, 2.1068, 2.9465, 1.7463, 2.5787, 2.7246, 2.0122, 2.8819], device='cuda:3'), covar=tensor([0.1182, 0.2113, 0.1318, 0.1996, 0.0906, 0.1414, 0.2726, 0.0730], device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0208, 0.0193, 0.0192, 0.0176, 0.0214, 0.0217, 0.0200], 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') 2023-03-27 06:24:58,274 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140259.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:25:01,787 INFO [finetune.py:976] (3/7) Epoch 25, batch 2800, loss[loss=0.1618, simple_loss=0.2442, pruned_loss=0.03971, over 4826.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2403, pruned_loss=0.04912, over 952604.97 frames. ], batch size: 33, lr: 3.01e-03, grad_scale: 16.0 2023-03-27 06:25:02,939 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.021e+02 1.479e+02 1.751e+02 2.221e+02 3.486e+02, threshold=3.502e+02, percent-clipped=1.0 2023-03-27 06:25:10,749 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=140275.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:25:12,061 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5077, 1.4385, 1.2971, 1.4843, 1.7476, 1.6678, 1.4155, 1.3084], device='cuda:3'), covar=tensor([0.0338, 0.0336, 0.0666, 0.0294, 0.0207, 0.0490, 0.0376, 0.0430], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0107, 0.0146, 0.0111, 0.0100, 0.0114, 0.0103, 0.0112], device='cuda:3'), out_proj_covar=tensor([7.8067e-05, 8.1685e-05, 1.1384e-04, 8.5073e-05, 7.7789e-05, 8.4533e-05, 7.6411e-05, 8.5404e-05], device='cuda:3') 2023-03-27 06:25:22,268 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=140286.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:25:53,836 INFO [finetune.py:976] (3/7) Epoch 25, batch 2850, loss[loss=0.1934, simple_loss=0.2716, pruned_loss=0.05762, over 4809.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2386, pruned_loss=0.04907, over 950172.92 frames. ], batch size: 41, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:25:56,984 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140320.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:26:27,554 INFO [finetune.py:976] (3/7) Epoch 25, batch 2900, loss[loss=0.1719, simple_loss=0.2376, pruned_loss=0.05311, over 4831.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2413, pruned_loss=0.04952, over 950731.84 frames. ], batch size: 30, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:26:28,760 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.105e+02 1.583e+02 1.866e+02 2.190e+02 4.311e+02, threshold=3.732e+02, percent-clipped=1.0 2023-03-27 06:26:40,205 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1013, 2.0346, 1.7665, 2.0212, 1.8972, 1.9384, 1.9393, 2.6416], device='cuda:3'), covar=tensor([0.3661, 0.4183, 0.3152, 0.3793, 0.4126, 0.2456, 0.3642, 0.1655], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0263, 0.0236, 0.0277, 0.0260, 0.0230, 0.0257, 0.0238], 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') 2023-03-27 06:26:42,000 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140387.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:26:59,280 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6522, 1.5676, 1.4183, 1.8042, 2.1234, 1.7680, 1.4377, 1.3744], device='cuda:3'), covar=tensor([0.2225, 0.1999, 0.1921, 0.1511, 0.1453, 0.1246, 0.2282, 0.1930], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0212, 0.0215, 0.0199, 0.0246, 0.0192, 0.0217, 0.0206], 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') 2023-03-27 06:27:01,453 INFO [finetune.py:976] (3/7) Epoch 25, batch 2950, loss[loss=0.2031, simple_loss=0.2707, pruned_loss=0.06774, over 4906.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2449, pruned_loss=0.05057, over 949751.31 frames. ], batch size: 43, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:27:07,571 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140425.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:27:21,216 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140445.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:27:23,044 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140448.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:27:30,137 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140459.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:27:30,779 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140460.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:27:32,481 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140462.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:27:34,688 INFO [finetune.py:976] (3/7) Epoch 25, batch 3000, loss[loss=0.1402, simple_loss=0.2103, pruned_loss=0.0351, over 4765.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.246, pruned_loss=0.05075, over 950399.87 frames. ], batch size: 26, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:27:34,689 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-27 06:27:37,409 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.9085, 3.5219, 3.6601, 3.8160, 3.6815, 3.5461, 3.9651, 1.3951], device='cuda:3'), covar=tensor([0.0833, 0.0827, 0.0868, 0.0965, 0.1290, 0.1486, 0.0758, 0.5318], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0248, 0.0283, 0.0297, 0.0339, 0.0287, 0.0308, 0.0304], 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') 2023-03-27 06:27:39,243 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3520, 1.5574, 1.7508, 1.6131, 1.8388, 3.1017, 1.5338, 1.7157], device='cuda:3'), covar=tensor([0.0941, 0.1654, 0.0918, 0.0851, 0.1309, 0.0307, 0.1371, 0.1560], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0082, 0.0073, 0.0076, 0.0091, 0.0081, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 06:27:48,789 INFO [finetune.py:1010] (3/7) Epoch 25, validation: loss=0.1571, simple_loss=0.2254, pruned_loss=0.04443, over 2265189.00 frames. 2023-03-27 06:27:48,790 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6469MB 2023-03-27 06:27:49,493 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140466.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:27:50,499 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.023e+02 1.568e+02 1.888e+02 2.214e+02 4.503e+02, threshold=3.776e+02, percent-clipped=3.0 2023-03-27 06:27:59,536 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=140473.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:28:17,971 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140489.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:28:20,342 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=140493.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:28:24,121 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-03-27 06:28:29,179 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140506.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:28:30,343 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=140508.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:28:31,021 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140509.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:28:31,589 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=140510.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:28:34,543 INFO [finetune.py:976] (3/7) Epoch 25, batch 3050, loss[loss=0.1477, simple_loss=0.2172, pruned_loss=0.03908, over 4732.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2473, pruned_loss=0.05119, over 951286.97 frames. ], batch size: 27, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:28:58,635 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140550.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:29:00,944 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=140554.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:29:08,053 INFO [finetune.py:976] (3/7) Epoch 25, batch 3100, loss[loss=0.1963, simple_loss=0.256, pruned_loss=0.0683, over 4923.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2466, pruned_loss=0.05105, over 954640.76 frames. ], batch size: 33, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:29:09,242 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.313e+01 1.495e+02 1.767e+02 2.180e+02 4.499e+02, threshold=3.535e+02, percent-clipped=1.0 2023-03-27 06:29:12,210 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140570.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:29:17,392 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1045, 3.5758, 3.7716, 3.9099, 3.9000, 3.6735, 4.1866, 1.5413], device='cuda:3'), covar=tensor([0.0932, 0.0898, 0.1000, 0.1143, 0.1418, 0.1402, 0.0833, 0.5747], device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0249, 0.0284, 0.0298, 0.0339, 0.0288, 0.0309, 0.0304], 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') 2023-03-27 06:29:42,052 INFO [finetune.py:976] (3/7) Epoch 25, batch 3150, loss[loss=0.1423, simple_loss=0.2119, pruned_loss=0.03638, over 4825.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2436, pruned_loss=0.05008, over 954068.23 frames. ], batch size: 38, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:29:42,118 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140615.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:29:47,538 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0205, 1.6497, 2.2861, 1.5900, 2.1147, 2.2656, 1.6161, 2.3498], device='cuda:3'), covar=tensor([0.1259, 0.2170, 0.1540, 0.1979, 0.0864, 0.1438, 0.2937, 0.0807], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0206, 0.0190, 0.0189, 0.0174, 0.0212, 0.0215, 0.0198], 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') 2023-03-27 06:30:15,050 INFO [finetune.py:976] (3/7) Epoch 25, batch 3200, loss[loss=0.1609, simple_loss=0.2275, pruned_loss=0.04713, over 4771.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2411, pruned_loss=0.0495, over 955990.87 frames. ], batch size: 26, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:30:16,220 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.929e+01 1.486e+02 1.750e+02 2.144e+02 4.466e+02, threshold=3.500e+02, percent-clipped=2.0 2023-03-27 06:30:27,089 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8219, 1.2878, 1.9156, 1.8655, 1.6585, 1.6492, 1.7880, 1.8337], device='cuda:3'), covar=tensor([0.3937, 0.3975, 0.3247, 0.3638, 0.4756, 0.3806, 0.4574, 0.2942], device='cuda:3'), in_proj_covar=tensor([0.0264, 0.0247, 0.0267, 0.0292, 0.0293, 0.0268, 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') 2023-03-27 06:31:06,561 INFO [finetune.py:976] (3/7) Epoch 25, batch 3250, loss[loss=0.1631, simple_loss=0.2369, pruned_loss=0.04468, over 4833.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2409, pruned_loss=0.04951, over 951420.12 frames. ], batch size: 30, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:31:11,471 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140723.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:31:24,372 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.58 vs. limit=5.0 2023-03-27 06:31:25,169 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140743.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:31:35,360 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140759.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:31:39,361 INFO [finetune.py:976] (3/7) Epoch 25, batch 3300, loss[loss=0.1872, simple_loss=0.2613, pruned_loss=0.05654, over 4871.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2458, pruned_loss=0.05106, over 953544.28 frames. ], batch size: 34, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:31:40,555 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140766.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:31:41,064 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.042e+02 1.629e+02 1.945e+02 2.397e+02 4.021e+02, threshold=3.889e+02, percent-clipped=5.0 2023-03-27 06:31:44,869 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2668, 2.1340, 1.6663, 2.2724, 2.0643, 1.8345, 2.4756, 2.2454], device='cuda:3'), covar=tensor([0.1388, 0.2079, 0.2975, 0.2470, 0.2761, 0.1775, 0.3600, 0.1708], device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0189, 0.0234, 0.0252, 0.0247, 0.0205, 0.0212, 0.0200], 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') 2023-03-27 06:31:53,063 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140784.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:32:08,086 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=140807.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:32:09,340 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140809.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:32:09,368 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8711, 2.1151, 1.6495, 1.9018, 2.3255, 2.5430, 1.9807, 1.9812], device='cuda:3'), covar=tensor([0.0495, 0.0342, 0.0712, 0.0354, 0.0302, 0.0619, 0.0360, 0.0377], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0106, 0.0144, 0.0110, 0.0099, 0.0112, 0.0102, 0.0111], device='cuda:3'), out_proj_covar=tensor([7.7086e-05, 8.1084e-05, 1.1275e-04, 8.4197e-05, 7.7106e-05, 8.3086e-05, 7.5586e-05, 8.4566e-05], device='cuda:3') 2023-03-27 06:32:12,392 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=140814.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:32:12,930 INFO [finetune.py:976] (3/7) Epoch 25, batch 3350, loss[loss=0.2345, simple_loss=0.2988, pruned_loss=0.08511, over 4821.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2481, pruned_loss=0.05203, over 953861.28 frames. ], batch size: 49, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:32:33,951 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140845.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:32:43,720 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0558, 1.9659, 1.7086, 1.8381, 1.8941, 1.8847, 1.8436, 2.5504], device='cuda:3'), covar=tensor([0.3696, 0.4266, 0.3243, 0.4035, 0.3996, 0.2473, 0.3857, 0.1683], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0263, 0.0236, 0.0276, 0.0259, 0.0229, 0.0256, 0.0237], 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') 2023-03-27 06:32:46,606 INFO [finetune.py:976] (3/7) Epoch 25, batch 3400, loss[loss=0.2259, simple_loss=0.2891, pruned_loss=0.08136, over 4796.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2491, pruned_loss=0.05216, over 954546.80 frames. ], batch size: 45, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:32:46,672 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140865.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:32:47,794 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.030e+02 1.564e+02 1.878e+02 2.236e+02 3.278e+02, threshold=3.756e+02, percent-clipped=0.0 2023-03-27 06:32:49,707 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140870.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:32:59,519 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8232, 1.7806, 2.2945, 2.0540, 1.9681, 4.4806, 1.9499, 1.9769], device='cuda:3'), covar=tensor([0.0942, 0.1777, 0.1084, 0.0946, 0.1596, 0.0188, 0.1366, 0.1689], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0082, 0.0073, 0.0076, 0.0091, 0.0081, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 06:33:39,160 INFO [finetune.py:976] (3/7) Epoch 25, batch 3450, loss[loss=0.1574, simple_loss=0.2388, pruned_loss=0.03795, over 4883.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2479, pruned_loss=0.0514, over 955388.30 frames. ], batch size: 43, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:33:39,261 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140915.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:33:40,451 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140917.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:34:01,667 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140947.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:34:05,901 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2928, 1.4058, 1.5690, 1.5963, 1.6329, 3.0161, 1.4111, 1.5307], device='cuda:3'), covar=tensor([0.1065, 0.2035, 0.1172, 0.1003, 0.1722, 0.0352, 0.1682, 0.2030], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0082, 0.0073, 0.0076, 0.0091, 0.0081, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 06:34:09,536 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.4370, 3.8389, 4.0739, 4.1832, 4.2443, 3.9845, 4.5235, 1.8404], device='cuda:3'), covar=tensor([0.0709, 0.0787, 0.0773, 0.0848, 0.1051, 0.1297, 0.0604, 0.5004], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0247, 0.0282, 0.0296, 0.0337, 0.0287, 0.0308, 0.0301], 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') 2023-03-27 06:34:11,814 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=140963.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:34:12,458 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1004, 1.8815, 2.2408, 1.5503, 2.1201, 2.3342, 2.2654, 1.4232], device='cuda:3'), covar=tensor([0.0682, 0.0943, 0.0720, 0.0899, 0.0728, 0.0685, 0.0678, 0.1695], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0136, 0.0140, 0.0119, 0.0126, 0.0137, 0.0139, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 06:34:12,946 INFO [finetune.py:976] (3/7) Epoch 25, batch 3500, loss[loss=0.1662, simple_loss=0.2364, pruned_loss=0.04802, over 4822.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2444, pruned_loss=0.04986, over 956218.03 frames. ], batch size: 38, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:34:14,175 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.063e+02 1.468e+02 1.748e+02 2.204e+02 3.629e+02, threshold=3.496e+02, percent-clipped=0.0 2023-03-27 06:34:20,901 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140978.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:34:24,172 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.1964, 2.9471, 2.6426, 1.4759, 2.7810, 2.3023, 2.2631, 2.6287], device='cuda:3'), covar=tensor([0.1112, 0.0740, 0.1773, 0.2191, 0.1690, 0.2226, 0.1922, 0.1051], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0191, 0.0199, 0.0181, 0.0209, 0.0210, 0.0223, 0.0195], 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') 2023-03-27 06:34:41,922 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141008.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:34:46,049 INFO [finetune.py:976] (3/7) Epoch 25, batch 3550, loss[loss=0.1603, simple_loss=0.2383, pruned_loss=0.04116, over 4912.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2413, pruned_loss=0.04871, over 956731.57 frames. ], batch size: 32, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:35:04,136 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141043.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:35:19,349 INFO [finetune.py:976] (3/7) Epoch 25, batch 3600, loss[loss=0.1562, simple_loss=0.2107, pruned_loss=0.05089, over 3983.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.24, pruned_loss=0.04878, over 956364.58 frames. ], batch size: 17, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:35:19,518 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.64 vs. limit=2.0 2023-03-27 06:35:20,525 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.001e+02 1.456e+02 1.796e+02 2.356e+02 3.995e+02, threshold=3.592e+02, percent-clipped=1.0 2023-03-27 06:35:22,984 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3221, 1.3755, 1.6584, 1.6159, 1.5031, 3.0611, 1.3274, 1.5216], device='cuda:3'), covar=tensor([0.0996, 0.1656, 0.1230, 0.0932, 0.1528, 0.0260, 0.1436, 0.1677], device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0082, 0.0073, 0.0076, 0.0091, 0.0081, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 06:35:28,400 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141079.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:35:36,623 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=141091.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:36:00,471 INFO [finetune.py:976] (3/7) Epoch 25, batch 3650, loss[loss=0.2069, simple_loss=0.3015, pruned_loss=0.0561, over 4819.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2425, pruned_loss=0.0501, over 953623.00 frames. ], batch size: 38, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:36:30,905 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141143.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:36:32,104 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141145.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:36:34,517 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1558, 2.8512, 2.7011, 1.2013, 3.0085, 2.1517, 0.6490, 1.8895], device='cuda:3'), covar=tensor([0.2675, 0.2571, 0.1900, 0.3559, 0.1446, 0.1240, 0.4201, 0.1741], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0179, 0.0162, 0.0131, 0.0161, 0.0124, 0.0148, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 06:36:46,082 INFO [finetune.py:976] (3/7) Epoch 25, batch 3700, loss[loss=0.1795, simple_loss=0.2457, pruned_loss=0.05667, over 4818.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2459, pruned_loss=0.05108, over 953150.09 frames. ], batch size: 25, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:36:46,155 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141165.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:36:46,172 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141165.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:36:47,287 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.158e+02 1.708e+02 2.029e+02 2.382e+02 3.628e+02, threshold=4.058e+02, percent-clipped=1.0 2023-03-27 06:36:54,975 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-27 06:37:02,144 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2748, 1.2854, 1.5810, 1.0254, 1.4076, 1.4714, 1.2630, 1.6628], device='cuda:3'), covar=tensor([0.1106, 0.2217, 0.1079, 0.1427, 0.0758, 0.1149, 0.2994, 0.0734], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0208, 0.0192, 0.0191, 0.0174, 0.0213, 0.0217, 0.0199], 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') 2023-03-27 06:37:03,880 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=141193.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:37:11,216 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141204.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:37:14,561 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3593, 2.4199, 1.8570, 2.6078, 2.2768, 1.9402, 2.8539, 2.3895], device='cuda:3'), covar=tensor([0.1417, 0.2176, 0.2969, 0.2659, 0.2676, 0.1711, 0.2836, 0.1798], device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0190, 0.0234, 0.0253, 0.0248, 0.0205, 0.0213, 0.0202], 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') 2023-03-27 06:37:18,590 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=141213.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:37:19,744 INFO [finetune.py:976] (3/7) Epoch 25, batch 3750, loss[loss=0.1855, simple_loss=0.2546, pruned_loss=0.05823, over 4775.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2472, pruned_loss=0.05118, over 953422.23 frames. ], batch size: 28, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:37:20,438 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.2952, 3.6794, 3.9041, 4.1602, 4.1007, 3.8120, 4.3731, 1.3283], device='cuda:3'), covar=tensor([0.0759, 0.0870, 0.0862, 0.0952, 0.1159, 0.1497, 0.0706, 0.5909], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0248, 0.0283, 0.0296, 0.0338, 0.0288, 0.0308, 0.0302], 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') 2023-03-27 06:37:44,745 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9407, 1.9768, 1.6341, 1.9471, 1.9427, 1.9129, 1.9048, 2.4812], device='cuda:3'), covar=tensor([0.3724, 0.3484, 0.3323, 0.3758, 0.3894, 0.2393, 0.3402, 0.1779], device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0263, 0.0235, 0.0276, 0.0258, 0.0229, 0.0255, 0.0237], 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') 2023-03-27 06:37:52,662 INFO [finetune.py:976] (3/7) Epoch 25, batch 3800, loss[loss=0.1986, simple_loss=0.2714, pruned_loss=0.06292, over 4842.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.247, pruned_loss=0.05122, over 952703.44 frames. ], batch size: 49, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:37:54,343 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.018e+02 1.508e+02 1.827e+02 2.217e+02 6.513e+02, threshold=3.654e+02, percent-clipped=2.0 2023-03-27 06:37:58,031 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141273.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:38:19,459 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141303.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:38:21,132 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-03-27 06:38:32,076 INFO [finetune.py:976] (3/7) Epoch 25, batch 3850, loss[loss=0.2066, simple_loss=0.2678, pruned_loss=0.07269, over 4261.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2454, pruned_loss=0.05048, over 952335.39 frames. ], batch size: 65, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:38:49,946 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141329.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:39:16,918 INFO [finetune.py:976] (3/7) Epoch 25, batch 3900, loss[loss=0.1344, simple_loss=0.2113, pruned_loss=0.02874, over 4835.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2422, pruned_loss=0.04909, over 952398.91 frames. ], batch size: 47, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:39:18,105 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.088e+02 1.504e+02 1.773e+02 2.110e+02 6.012e+02, threshold=3.546e+02, percent-clipped=1.0 2023-03-27 06:39:26,413 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141379.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:39:33,575 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141390.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:39:45,073 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.0066, 3.4874, 3.6798, 3.8680, 3.7834, 3.5174, 4.0735, 1.3502], device='cuda:3'), covar=tensor([0.0873, 0.0881, 0.0953, 0.0999, 0.1233, 0.1672, 0.0826, 0.5634], device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0248, 0.0284, 0.0297, 0.0339, 0.0288, 0.0309, 0.0302], 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') 2023-03-27 06:39:49,668 INFO [finetune.py:976] (3/7) Epoch 25, batch 3950, loss[loss=0.1608, simple_loss=0.2136, pruned_loss=0.05396, over 4135.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2397, pruned_loss=0.04856, over 953395.29 frames. ], batch size: 17, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:39:51,005 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141417.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 06:39:58,916 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=141427.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:40:01,692 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-27 06:40:07,858 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141441.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:40:23,342 INFO [finetune.py:976] (3/7) Epoch 25, batch 4000, loss[loss=0.1746, simple_loss=0.2389, pruned_loss=0.05521, over 4738.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2395, pruned_loss=0.0487, over 953145.91 frames. ], batch size: 27, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:40:23,425 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141465.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:40:24,522 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.071e+02 1.514e+02 1.750e+02 2.113e+02 3.817e+02, threshold=3.500e+02, percent-clipped=1.0 2023-03-27 06:40:33,179 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141478.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 06:40:46,340 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141499.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:40:48,238 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141502.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:40:55,309 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=141513.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:40:56,983 INFO [finetune.py:976] (3/7) Epoch 25, batch 4050, loss[loss=0.1778, simple_loss=0.2521, pruned_loss=0.05175, over 4776.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2426, pruned_loss=0.04972, over 952119.73 frames. ], batch size: 28, lr: 3.00e-03, grad_scale: 16.0 2023-03-27 06:41:27,042 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2986, 2.1323, 1.7988, 2.1893, 2.2086, 1.9026, 2.4454, 2.2568], device='cuda:3'), covar=tensor([0.1326, 0.2084, 0.2876, 0.2362, 0.2509, 0.1728, 0.2865, 0.1605], device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0190, 0.0235, 0.0254, 0.0249, 0.0206, 0.0214, 0.0202], 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') 2023-03-27 06:41:49,013 INFO [finetune.py:976] (3/7) Epoch 25, batch 4100, loss[loss=0.1515, simple_loss=0.2304, pruned_loss=0.03631, over 4755.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2446, pruned_loss=0.04991, over 954303.18 frames. ], batch size: 54, lr: 3.00e-03, grad_scale: 32.0 2023-03-27 06:41:50,184 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.144e+02 1.605e+02 1.887e+02 2.173e+02 5.231e+02, threshold=3.774e+02, percent-clipped=3.0 2023-03-27 06:41:50,320 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4929, 1.3466, 1.2394, 1.5274, 1.7059, 1.5130, 1.1252, 1.2803], device='cuda:3'), covar=tensor([0.2107, 0.1931, 0.1793, 0.1556, 0.1418, 0.1216, 0.2310, 0.1766], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0209, 0.0213, 0.0196, 0.0243, 0.0190, 0.0215, 0.0204], 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') 2023-03-27 06:41:54,378 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141573.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:41:56,310 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8964, 1.4518, 2.0416, 1.9668, 1.7889, 1.7276, 1.9144, 1.8858], device='cuda:3'), covar=tensor([0.4021, 0.4221, 0.2960, 0.3511, 0.4422, 0.3800, 0.4205, 0.3012], device='cuda:3'), in_proj_covar=tensor([0.0265, 0.0248, 0.0268, 0.0293, 0.0294, 0.0270, 0.0299, 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') 2023-03-27 06:42:14,757 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141603.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:42:22,403 INFO [finetune.py:976] (3/7) Epoch 25, batch 4150, loss[loss=0.1536, simple_loss=0.2305, pruned_loss=0.03833, over 4920.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2463, pruned_loss=0.05054, over 953517.77 frames. ], batch size: 28, lr: 3.00e-03, grad_scale: 32.0 2023-03-27 06:42:26,083 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=141621.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:42:42,041 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-27 06:42:46,644 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=141651.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:42:52,136 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-27 06:42:55,970 INFO [finetune.py:976] (3/7) Epoch 25, batch 4200, loss[loss=0.1895, simple_loss=0.2588, pruned_loss=0.06012, over 4787.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2473, pruned_loss=0.05116, over 954510.42 frames. ], batch size: 51, lr: 3.00e-03, grad_scale: 32.0 2023-03-27 06:42:57,194 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.937e+01 1.553e+02 1.832e+02 2.223e+02 5.119e+02, threshold=3.664e+02, percent-clipped=2.0 2023-03-27 06:43:09,619 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141685.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:43:13,040 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4084, 2.2622, 1.7234, 0.7720, 1.9459, 1.9133, 1.8408, 1.9662], device='cuda:3'), covar=tensor([0.1001, 0.0818, 0.1675, 0.2235, 0.1450, 0.2581, 0.2234, 0.1001], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0192, 0.0200, 0.0182, 0.0210, 0.0211, 0.0224, 0.0196], 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') 2023-03-27 06:43:29,321 INFO [finetune.py:976] (3/7) Epoch 25, batch 4250, loss[loss=0.1398, simple_loss=0.2105, pruned_loss=0.03458, over 4768.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2458, pruned_loss=0.05081, over 955417.28 frames. ], batch size: 26, lr: 3.00e-03, grad_scale: 32.0 2023-03-27 06:44:21,246 INFO [finetune.py:976] (3/7) Epoch 25, batch 4300, loss[loss=0.1521, simple_loss=0.2253, pruned_loss=0.03949, over 4919.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2425, pruned_loss=0.04985, over 953722.28 frames. ], batch size: 37, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:44:22,425 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.814e+01 1.357e+02 1.630e+02 2.025e+02 3.929e+02, threshold=3.260e+02, percent-clipped=1.0 2023-03-27 06:44:26,617 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141773.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 06:44:35,647 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-03-27 06:44:43,604 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141797.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:44:44,841 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141799.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:44:55,201 INFO [finetune.py:976] (3/7) Epoch 25, batch 4350, loss[loss=0.1977, simple_loss=0.2694, pruned_loss=0.06303, over 4861.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2404, pruned_loss=0.04939, over 954947.19 frames. ], batch size: 44, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:45:16,952 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0699, 1.9530, 1.6905, 1.6713, 1.7952, 1.7784, 1.9452, 2.5365], device='cuda:3'), covar=tensor([0.3844, 0.3638, 0.3065, 0.3256, 0.3799, 0.2407, 0.3414, 0.1614], device='cuda:3'), in_proj_covar=tensor([0.0291, 0.0266, 0.0236, 0.0277, 0.0260, 0.0230, 0.0258, 0.0239], 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') 2023-03-27 06:45:17,467 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=141847.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:45:28,845 INFO [finetune.py:976] (3/7) Epoch 25, batch 4400, loss[loss=0.1398, simple_loss=0.2197, pruned_loss=0.02991, over 4904.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2415, pruned_loss=0.05036, over 954324.53 frames. ], batch size: 35, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:45:30,033 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.038e+02 1.595e+02 1.814e+02 2.202e+02 4.275e+02, threshold=3.628e+02, percent-clipped=6.0 2023-03-27 06:46:01,881 INFO [finetune.py:976] (3/7) Epoch 25, batch 4450, loss[loss=0.2012, simple_loss=0.2689, pruned_loss=0.06673, over 4872.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2455, pruned_loss=0.05137, over 956971.17 frames. ], batch size: 34, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:46:02,620 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141916.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:46:10,451 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141928.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:46:46,890 INFO [finetune.py:976] (3/7) Epoch 25, batch 4500, loss[loss=0.1587, simple_loss=0.2322, pruned_loss=0.04253, over 4885.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2457, pruned_loss=0.0512, over 953820.82 frames. ], batch size: 35, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:46:52,641 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.869e+01 1.594e+02 1.826e+02 2.236e+02 4.959e+02, threshold=3.653e+02, percent-clipped=2.0 2023-03-27 06:47:02,919 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141977.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:47:07,025 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2277, 1.7946, 2.2346, 2.2310, 1.9266, 1.9381, 2.1458, 2.1220], device='cuda:3'), covar=tensor([0.4278, 0.4144, 0.3280, 0.3990, 0.5419, 0.4147, 0.4876, 0.2963], device='cuda:3'), in_proj_covar=tensor([0.0264, 0.0247, 0.0267, 0.0293, 0.0293, 0.0269, 0.0300, 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') 2023-03-27 06:47:08,164 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141985.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:47:10,674 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141989.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:47:30,042 INFO [finetune.py:976] (3/7) Epoch 25, batch 4550, loss[loss=0.1782, simple_loss=0.2458, pruned_loss=0.05532, over 4924.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2472, pruned_loss=0.05166, over 953259.07 frames. ], batch size: 42, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:47:35,802 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-27 06:47:41,389 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=142033.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:48:03,344 INFO [finetune.py:976] (3/7) Epoch 25, batch 4600, loss[loss=0.1371, simple_loss=0.2171, pruned_loss=0.02857, over 4780.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2461, pruned_loss=0.05072, over 954868.80 frames. ], batch size: 26, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:48:04,587 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 7.716e+01 1.569e+02 1.799e+02 2.271e+02 3.318e+02, threshold=3.598e+02, percent-clipped=0.0 2023-03-27 06:48:08,722 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142073.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 06:48:23,764 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142097.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:48:36,588 INFO [finetune.py:976] (3/7) Epoch 25, batch 4650, loss[loss=0.2248, simple_loss=0.2809, pruned_loss=0.08432, over 4824.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2441, pruned_loss=0.05093, over 952869.87 frames. ], batch size: 33, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:48:40,333 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=142121.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 06:48:48,400 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-03-27 06:48:57,212 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=142145.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:49:19,997 INFO [finetune.py:976] (3/7) Epoch 25, batch 4700, loss[loss=0.1736, simple_loss=0.2372, pruned_loss=0.05497, over 4872.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2421, pruned_loss=0.05047, over 955203.29 frames. ], batch size: 34, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:49:21,182 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.163e+01 1.384e+02 1.765e+02 2.088e+02 3.764e+02, threshold=3.531e+02, percent-clipped=1.0 2023-03-27 06:49:52,963 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-27 06:50:00,899 INFO [finetune.py:976] (3/7) Epoch 25, batch 4750, loss[loss=0.2075, simple_loss=0.2648, pruned_loss=0.0751, over 4925.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2404, pruned_loss=0.04978, over 956075.40 frames. ], batch size: 38, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:50:04,580 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7759, 2.5794, 2.2223, 1.1120, 2.4242, 2.1294, 2.0182, 2.4048], device='cuda:3'), covar=tensor([0.0963, 0.0852, 0.1819, 0.2272, 0.1542, 0.2214, 0.2356, 0.1027], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0192, 0.0200, 0.0182, 0.0210, 0.0212, 0.0224, 0.0195], 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') 2023-03-27 06:50:16,735 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 2023-03-27 06:50:34,336 INFO [finetune.py:976] (3/7) Epoch 25, batch 4800, loss[loss=0.2078, simple_loss=0.2713, pruned_loss=0.07216, over 4743.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2419, pruned_loss=0.05007, over 956191.50 frames. ], batch size: 54, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:50:35,545 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.947e+01 1.535e+02 1.762e+02 2.238e+02 3.446e+02, threshold=3.524e+02, percent-clipped=1.0 2023-03-27 06:50:39,648 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142272.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:50:40,941 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.5314, 1.5962, 1.5771, 0.9071, 1.6688, 1.9241, 1.9319, 1.4595], device='cuda:3'), covar=tensor([0.0921, 0.0555, 0.0587, 0.0572, 0.0457, 0.0611, 0.0336, 0.0675], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0148, 0.0127, 0.0122, 0.0130, 0.0129, 0.0141, 0.0148], device='cuda:3'), out_proj_covar=tensor([8.8706e-05, 1.0604e-04, 9.0905e-05, 8.5437e-05, 9.0725e-05, 9.1760e-05, 1.0030e-04, 1.0579e-04], device='cuda:3') 2023-03-27 06:50:47,480 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142284.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:51:07,517 INFO [finetune.py:976] (3/7) Epoch 25, batch 4850, loss[loss=0.1596, simple_loss=0.2235, pruned_loss=0.0478, over 4247.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2454, pruned_loss=0.05038, over 954996.59 frames. ], batch size: 18, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:51:39,149 INFO [finetune.py:976] (3/7) Epoch 25, batch 4900, loss[loss=0.2029, simple_loss=0.2767, pruned_loss=0.06454, over 4718.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2467, pruned_loss=0.05107, over 954857.12 frames. ], batch size: 59, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:51:40,866 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.096e+02 1.551e+02 1.812e+02 2.135e+02 6.918e+02, threshold=3.624e+02, percent-clipped=2.0 2023-03-27 06:52:31,145 INFO [finetune.py:976] (3/7) Epoch 25, batch 4950, loss[loss=0.1892, simple_loss=0.2491, pruned_loss=0.06464, over 4281.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2475, pruned_loss=0.0513, over 954967.18 frames. ], batch size: 18, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:53:03,772 INFO [finetune.py:976] (3/7) Epoch 25, batch 5000, loss[loss=0.198, simple_loss=0.2697, pruned_loss=0.06314, over 4897.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2455, pruned_loss=0.05045, over 953666.54 frames. ], batch size: 43, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:53:04,976 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.190e+01 1.432e+02 1.813e+02 2.155e+02 3.992e+02, threshold=3.625e+02, percent-clipped=1.0 2023-03-27 06:53:19,161 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.0423, 4.4046, 4.5627, 4.8875, 4.7764, 4.5137, 5.1327, 1.5743], device='cuda:3'), covar=tensor([0.0743, 0.0846, 0.0844, 0.0844, 0.1187, 0.1581, 0.0577, 0.5982], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0248, 0.0282, 0.0295, 0.0335, 0.0287, 0.0306, 0.0301], 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') 2023-03-27 06:53:30,149 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.80 vs. limit=5.0 2023-03-27 06:53:30,535 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7749, 1.3279, 0.7991, 1.7245, 2.0949, 1.5272, 1.6624, 1.7688], device='cuda:3'), covar=tensor([0.1264, 0.1763, 0.1991, 0.1001, 0.1833, 0.1997, 0.1191, 0.1651], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0094, 0.0110, 0.0092, 0.0120, 0.0094, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 06:53:36,413 INFO [finetune.py:976] (3/7) Epoch 25, batch 5050, loss[loss=0.1523, simple_loss=0.223, pruned_loss=0.04082, over 4825.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.242, pruned_loss=0.04917, over 955082.74 frames. ], batch size: 41, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:54:09,843 INFO [finetune.py:976] (3/7) Epoch 25, batch 5100, loss[loss=0.174, simple_loss=0.2323, pruned_loss=0.05778, over 4833.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2395, pruned_loss=0.04845, over 954446.81 frames. ], batch size: 33, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:54:11,045 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.212e+01 1.519e+02 1.807e+02 2.247e+02 4.075e+02, threshold=3.613e+02, percent-clipped=2.0 2023-03-27 06:54:14,214 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142572.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:54:14,848 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142573.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:54:27,745 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142584.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:54:59,671 INFO [finetune.py:976] (3/7) Epoch 25, batch 5150, loss[loss=0.248, simple_loss=0.2992, pruned_loss=0.09837, over 4822.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2386, pruned_loss=0.04841, over 952980.15 frames. ], batch size: 38, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:55:03,298 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=142620.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:55:10,595 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=142632.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:55:12,840 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142634.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:55:15,860 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.35 vs. limit=5.0 2023-03-27 06:55:33,009 INFO [finetune.py:976] (3/7) Epoch 25, batch 5200, loss[loss=0.1832, simple_loss=0.2487, pruned_loss=0.05887, over 4830.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2422, pruned_loss=0.04916, over 951958.96 frames. ], batch size: 33, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:55:34,192 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.084e+02 1.563e+02 1.762e+02 2.093e+02 3.679e+02, threshold=3.523e+02, percent-clipped=1.0 2023-03-27 06:55:42,106 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8032, 1.7525, 1.5309, 1.9712, 2.3875, 1.9806, 1.7353, 1.4913], device='cuda:3'), covar=tensor([0.2413, 0.1975, 0.2056, 0.1616, 0.1665, 0.1234, 0.2280, 0.1981], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0211, 0.0215, 0.0198, 0.0245, 0.0192, 0.0218, 0.0206], 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') 2023-03-27 06:55:54,225 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.4412, 3.9311, 4.0827, 4.2542, 4.2248, 3.9721, 4.5151, 1.6374], device='cuda:3'), covar=tensor([0.0799, 0.0825, 0.0855, 0.1080, 0.1295, 0.1523, 0.0757, 0.5357], device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0248, 0.0282, 0.0297, 0.0337, 0.0289, 0.0307, 0.0302], 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') 2023-03-27 06:56:06,165 INFO [finetune.py:976] (3/7) Epoch 25, batch 5250, loss[loss=0.1751, simple_loss=0.2586, pruned_loss=0.04584, over 4813.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2445, pruned_loss=0.04986, over 953602.61 frames. ], batch size: 45, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:56:12,122 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142724.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:56:28,601 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4367, 2.2907, 1.8351, 2.2845, 2.2520, 2.0081, 2.5891, 2.3980], device='cuda:3'), covar=tensor([0.1296, 0.1892, 0.2942, 0.2337, 0.2634, 0.1647, 0.3140, 0.1571], device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0190, 0.0235, 0.0254, 0.0249, 0.0206, 0.0215, 0.0203], 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') 2023-03-27 06:56:32,204 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8666, 1.6922, 1.4676, 1.2422, 1.6181, 1.6378, 1.6208, 2.1938], device='cuda:3'), covar=tensor([0.3651, 0.3654, 0.3118, 0.3360, 0.3281, 0.2200, 0.3131, 0.1639], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0263, 0.0236, 0.0276, 0.0258, 0.0228, 0.0255, 0.0236], 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') 2023-03-27 06:56:39,094 INFO [finetune.py:976] (3/7) Epoch 25, batch 5300, loss[loss=0.175, simple_loss=0.248, pruned_loss=0.05099, over 4837.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2464, pruned_loss=0.05074, over 952623.62 frames. ], batch size: 49, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:56:40,274 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.031e+02 1.558e+02 1.826e+02 2.127e+02 3.045e+02, threshold=3.651e+02, percent-clipped=0.0 2023-03-27 06:56:51,725 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142785.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:57:19,998 INFO [finetune.py:976] (3/7) Epoch 25, batch 5350, loss[loss=0.1511, simple_loss=0.2169, pruned_loss=0.04263, over 4887.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2461, pruned_loss=0.05035, over 952174.59 frames. ], batch size: 32, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:58:06,041 INFO [finetune.py:976] (3/7) Epoch 25, batch 5400, loss[loss=0.1733, simple_loss=0.2403, pruned_loss=0.0531, over 4872.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2441, pruned_loss=0.04971, over 954254.95 frames. ], batch size: 34, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:58:07,257 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.131e+02 1.487e+02 1.682e+02 2.190e+02 4.832e+02, threshold=3.364e+02, percent-clipped=1.0 2023-03-27 06:58:38,660 INFO [finetune.py:976] (3/7) Epoch 25, batch 5450, loss[loss=0.1454, simple_loss=0.2222, pruned_loss=0.0343, over 4744.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2416, pruned_loss=0.04918, over 955846.28 frames. ], batch size: 26, lr: 2.99e-03, grad_scale: 32.0 2023-03-27 06:58:47,628 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142929.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 06:59:11,889 INFO [finetune.py:976] (3/7) Epoch 25, batch 5500, loss[loss=0.18, simple_loss=0.2483, pruned_loss=0.05582, over 4901.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2391, pruned_loss=0.04838, over 954370.81 frames. ], batch size: 43, lr: 2.99e-03, grad_scale: 16.0 2023-03-27 06:59:13,717 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.198e+01 1.372e+02 1.708e+02 2.223e+02 4.314e+02, threshold=3.415e+02, percent-clipped=3.0 2023-03-27 06:59:46,280 INFO [finetune.py:976] (3/7) Epoch 25, batch 5550, loss[loss=0.1911, simple_loss=0.251, pruned_loss=0.06554, over 4020.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2411, pruned_loss=0.04942, over 953432.49 frames. ], batch size: 17, lr: 2.99e-03, grad_scale: 16.0 2023-03-27 07:00:06,547 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2988, 2.9002, 2.7165, 1.1910, 2.9769, 2.2268, 0.7115, 1.9165], device='cuda:3'), covar=tensor([0.2589, 0.2725, 0.2160, 0.4045, 0.1481, 0.1172, 0.4653, 0.1893], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0180, 0.0163, 0.0131, 0.0162, 0.0124, 0.0149, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 07:00:29,924 INFO [finetune.py:976] (3/7) Epoch 25, batch 5600, loss[loss=0.1983, simple_loss=0.2576, pruned_loss=0.06953, over 4819.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2455, pruned_loss=0.05088, over 953263.51 frames. ], batch size: 33, lr: 2.99e-03, grad_scale: 16.0 2023-03-27 07:00:30,575 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.0263, 3.5310, 3.7533, 3.8997, 3.8141, 3.5666, 4.0993, 1.3089], device='cuda:3'), covar=tensor([0.0818, 0.0802, 0.0770, 0.0957, 0.1296, 0.1596, 0.0821, 0.5880], device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0247, 0.0280, 0.0295, 0.0335, 0.0287, 0.0306, 0.0299], 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') 2023-03-27 07:00:31,669 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.665e+01 1.700e+02 1.937e+02 2.306e+02 4.675e+02, threshold=3.875e+02, percent-clipped=1.0 2023-03-27 07:00:38,730 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=143080.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:01:00,341 INFO [finetune.py:976] (3/7) Epoch 25, batch 5650, loss[loss=0.1718, simple_loss=0.258, pruned_loss=0.04278, over 4862.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2485, pruned_loss=0.05137, over 953247.15 frames. ], batch size: 44, lr: 2.99e-03, grad_scale: 16.0 2023-03-27 07:01:29,840 INFO [finetune.py:976] (3/7) Epoch 25, batch 5700, loss[loss=0.1399, simple_loss=0.2084, pruned_loss=0.03571, over 4273.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2437, pruned_loss=0.05065, over 934855.97 frames. ], batch size: 18, lr: 2.98e-03, grad_scale: 16.0 2023-03-27 07:01:31,569 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.213e+01 1.339e+02 1.671e+02 2.043e+02 4.216e+02, threshold=3.342e+02, percent-clipped=1.0 2023-03-27 07:01:39,283 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8777, 2.4534, 3.1279, 4.5654, 3.3231, 3.3024, 1.5842, 3.9852], device='cuda:3'), covar=tensor([0.1376, 0.1275, 0.1214, 0.0481, 0.0644, 0.1092, 0.1732, 0.0324], device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0115, 0.0132, 0.0163, 0.0101, 0.0135, 0.0123, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 07:01:58,299 INFO [finetune.py:976] (3/7) Epoch 26, batch 0, loss[loss=0.1523, simple_loss=0.2314, pruned_loss=0.03662, over 4758.00 frames. ], tot_loss[loss=0.1523, simple_loss=0.2314, pruned_loss=0.03662, over 4758.00 frames. ], batch size: 26, lr: 2.98e-03, grad_scale: 16.0 2023-03-27 07:01:58,299 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-27 07:02:01,191 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7434, 1.6297, 1.9640, 1.3506, 1.7157, 1.9725, 1.5591, 2.0834], device='cuda:3'), covar=tensor([0.1141, 0.1930, 0.1217, 0.1633, 0.0922, 0.1212, 0.2751, 0.0743], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0205, 0.0190, 0.0190, 0.0173, 0.0212, 0.0215, 0.0197], 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') 2023-03-27 07:02:03,810 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4755, 1.3287, 1.3121, 1.4156, 1.6591, 1.6251, 1.3817, 1.2877], device='cuda:3'), covar=tensor([0.0388, 0.0367, 0.0635, 0.0363, 0.0278, 0.0490, 0.0458, 0.0476], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0107, 0.0146, 0.0112, 0.0102, 0.0116, 0.0104, 0.0113], device='cuda:3'), out_proj_covar=tensor([7.8500e-05, 8.1959e-05, 1.1422e-04, 8.5531e-05, 7.8734e-05, 8.5958e-05, 7.7183e-05, 8.6065e-05], device='cuda:3') 2023-03-27 07:02:06,836 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1456, 1.9271, 1.7573, 1.7286, 1.8878, 1.9002, 1.8922, 2.5679], device='cuda:3'), covar=tensor([0.3809, 0.4516, 0.3324, 0.3711, 0.3989, 0.2339, 0.3830, 0.1841], device='cuda:3'), in_proj_covar=tensor([0.0291, 0.0265, 0.0237, 0.0278, 0.0260, 0.0230, 0.0257, 0.0239], 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') 2023-03-27 07:02:16,993 INFO [finetune.py:1010] (3/7) Epoch 26, validation: loss=0.1591, simple_loss=0.2269, pruned_loss=0.04565, over 2265189.00 frames. 2023-03-27 07:02:16,993 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6469MB 2023-03-27 07:02:26,497 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-03-27 07:02:27,540 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0321, 1.6628, 2.3125, 1.6036, 2.0435, 2.2542, 1.6129, 2.3365], device='cuda:3'), covar=tensor([0.1126, 0.1960, 0.1294, 0.1877, 0.0876, 0.1272, 0.2571, 0.0739], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0205, 0.0191, 0.0190, 0.0174, 0.0213, 0.0216, 0.0198], 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') 2023-03-27 07:02:43,913 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143229.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:03:00,611 INFO [finetune.py:976] (3/7) Epoch 26, batch 50, loss[loss=0.1534, simple_loss=0.2314, pruned_loss=0.03767, over 4823.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2469, pruned_loss=0.05075, over 217183.99 frames. ], batch size: 33, lr: 2.98e-03, grad_scale: 16.0 2023-03-27 07:03:18,452 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.036e+02 1.460e+02 1.766e+02 2.058e+02 4.416e+02, threshold=3.532e+02, percent-clipped=3.0 2023-03-27 07:03:24,476 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=143277.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:03:34,112 INFO [finetune.py:976] (3/7) Epoch 26, batch 100, loss[loss=0.1798, simple_loss=0.2441, pruned_loss=0.05772, over 4938.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2396, pruned_loss=0.04811, over 382310.56 frames. ], batch size: 33, lr: 2.98e-03, grad_scale: 16.0 2023-03-27 07:03:43,418 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5931, 1.1957, 0.8868, 1.4896, 2.0379, 1.1260, 1.3667, 1.5896], device='cuda:3'), covar=tensor([0.1419, 0.1960, 0.1747, 0.1185, 0.1918, 0.1878, 0.1334, 0.1831], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0093, 0.0109, 0.0091, 0.0119, 0.0093, 0.0098, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 07:04:07,504 INFO [finetune.py:976] (3/7) Epoch 26, batch 150, loss[loss=0.1549, simple_loss=0.2241, pruned_loss=0.04289, over 4878.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2351, pruned_loss=0.04671, over 510983.00 frames. ], batch size: 34, lr: 2.98e-03, grad_scale: 16.0 2023-03-27 07:04:15,746 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.60 vs. limit=5.0 2023-03-27 07:04:25,693 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.744e+01 1.335e+02 1.679e+02 2.114e+02 2.886e+02, threshold=3.358e+02, percent-clipped=0.0 2023-03-27 07:04:33,624 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143380.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:04:41,259 INFO [finetune.py:976] (3/7) Epoch 26, batch 200, loss[loss=0.1709, simple_loss=0.2403, pruned_loss=0.05074, over 4776.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2356, pruned_loss=0.04772, over 610183.54 frames. ], batch size: 28, lr: 2.98e-03, grad_scale: 16.0 2023-03-27 07:04:46,838 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=143400.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 07:05:05,300 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=143428.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:05:05,381 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0980, 1.9645, 1.6481, 1.7358, 2.0061, 1.7335, 2.2015, 2.0422], device='cuda:3'), covar=tensor([0.1495, 0.1990, 0.3255, 0.2721, 0.3018, 0.2009, 0.2749, 0.1938], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0191, 0.0237, 0.0256, 0.0252, 0.0208, 0.0216, 0.0204], 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') 2023-03-27 07:05:22,139 INFO [finetune.py:976] (3/7) Epoch 26, batch 250, loss[loss=0.1828, simple_loss=0.2564, pruned_loss=0.05461, over 4859.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2392, pruned_loss=0.04849, over 687497.41 frames. ], batch size: 44, lr: 2.98e-03, grad_scale: 16.0 2023-03-27 07:05:48,870 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=143461.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 07:05:51,342 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4744, 2.4687, 1.9750, 0.9775, 2.1551, 1.8826, 1.8488, 2.1700], device='cuda:3'), covar=tensor([0.0949, 0.0692, 0.1542, 0.2180, 0.1440, 0.2262, 0.2086, 0.1036], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0193, 0.0202, 0.0183, 0.0212, 0.0213, 0.0226, 0.0197], 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') 2023-03-27 07:05:53,068 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.741e+01 1.618e+02 1.961e+02 2.394e+02 5.476e+02, threshold=3.922e+02, percent-clipped=2.0 2023-03-27 07:05:59,883 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=143473.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 07:06:12,189 INFO [finetune.py:976] (3/7) Epoch 26, batch 300, loss[loss=0.192, simple_loss=0.2805, pruned_loss=0.05177, over 4909.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2435, pruned_loss=0.04988, over 746750.91 frames. ], batch size: 42, lr: 2.98e-03, grad_scale: 16.0 2023-03-27 07:06:27,535 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5134, 2.0536, 2.7966, 1.8134, 2.4414, 2.7364, 1.8725, 2.8137], device='cuda:3'), covar=tensor([0.1322, 0.2197, 0.1496, 0.2280, 0.1050, 0.1520, 0.3076, 0.0800], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0204, 0.0190, 0.0189, 0.0173, 0.0212, 0.0215, 0.0197], 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') 2023-03-27 07:06:40,202 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=143534.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 07:06:44,913 INFO [finetune.py:976] (3/7) Epoch 26, batch 350, loss[loss=0.157, simple_loss=0.2382, pruned_loss=0.03785, over 4896.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2456, pruned_loss=0.05034, over 793765.97 frames. ], batch size: 36, lr: 2.98e-03, grad_scale: 16.0 2023-03-27 07:07:03,060 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.042e+02 1.441e+02 1.724e+02 2.077e+02 3.544e+02, threshold=3.448e+02, percent-clipped=0.0 2023-03-27 07:07:11,598 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4575, 3.3898, 3.1995, 1.4984, 3.4725, 2.5267, 0.8192, 2.3507], device='cuda:3'), covar=tensor([0.2145, 0.1794, 0.1588, 0.3164, 0.1064, 0.0990, 0.3919, 0.1372], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0178, 0.0161, 0.0130, 0.0160, 0.0123, 0.0148, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 07:07:18,101 INFO [finetune.py:976] (3/7) Epoch 26, batch 400, loss[loss=0.1713, simple_loss=0.2401, pruned_loss=0.05119, over 4863.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2467, pruned_loss=0.05058, over 829226.24 frames. ], batch size: 31, lr: 2.98e-03, grad_scale: 16.0 2023-03-27 07:07:54,070 INFO [finetune.py:976] (3/7) Epoch 26, batch 450, loss[loss=0.162, simple_loss=0.2377, pruned_loss=0.04321, over 4892.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.245, pruned_loss=0.04982, over 858315.97 frames. ], batch size: 35, lr: 2.98e-03, grad_scale: 16.0 2023-03-27 07:08:22,189 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.245e+01 1.544e+02 1.809e+02 2.165e+02 3.752e+02, threshold=3.619e+02, percent-clipped=3.0 2023-03-27 07:08:37,483 INFO [finetune.py:976] (3/7) Epoch 26, batch 500, loss[loss=0.1568, simple_loss=0.2328, pruned_loss=0.04038, over 4818.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2424, pruned_loss=0.04908, over 879163.46 frames. ], batch size: 25, lr: 2.98e-03, grad_scale: 16.0 2023-03-27 07:09:00,528 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2023, 2.0674, 1.9138, 2.2917, 2.7076, 2.3236, 2.1438, 1.7218], device='cuda:3'), covar=tensor([0.2063, 0.1802, 0.1852, 0.1602, 0.1580, 0.1057, 0.2012, 0.1759], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0210, 0.0216, 0.0197, 0.0244, 0.0191, 0.0216, 0.0205], 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') 2023-03-27 07:09:11,112 INFO [finetune.py:976] (3/7) Epoch 26, batch 550, loss[loss=0.216, simple_loss=0.2755, pruned_loss=0.07827, over 4872.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2384, pruned_loss=0.04777, over 895266.56 frames. ], batch size: 34, lr: 2.98e-03, grad_scale: 16.0 2023-03-27 07:09:20,261 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=143756.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 07:09:28,910 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.175e+01 1.443e+02 1.723e+02 1.984e+02 5.074e+02, threshold=3.446e+02, percent-clipped=2.0 2023-03-27 07:09:44,564 INFO [finetune.py:976] (3/7) Epoch 26, batch 600, loss[loss=0.1482, simple_loss=0.2275, pruned_loss=0.0344, over 4915.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2406, pruned_loss=0.04898, over 908361.46 frames. ], batch size: 37, lr: 2.98e-03, grad_scale: 16.0 2023-03-27 07:09:47,674 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6852, 3.4163, 3.2210, 1.4897, 3.5335, 2.5474, 0.7344, 2.3956], device='cuda:3'), covar=tensor([0.2321, 0.2674, 0.1880, 0.3977, 0.1394, 0.1203, 0.4828, 0.1716], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0178, 0.0160, 0.0129, 0.0160, 0.0123, 0.0148, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 07:10:09,699 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=143829.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 07:10:17,583 INFO [finetune.py:976] (3/7) Epoch 26, batch 650, loss[loss=0.145, simple_loss=0.2265, pruned_loss=0.03176, over 4779.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2426, pruned_loss=0.04889, over 916430.05 frames. ], batch size: 29, lr: 2.98e-03, grad_scale: 16.0 2023-03-27 07:10:41,255 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.012e+02 1.600e+02 1.821e+02 2.293e+02 5.159e+02, threshold=3.642e+02, percent-clipped=4.0 2023-03-27 07:11:07,601 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3498, 1.2586, 1.2491, 1.2818, 0.8959, 2.2565, 0.7631, 1.1182], device='cuda:3'), covar=tensor([0.3437, 0.2612, 0.2352, 0.2649, 0.1950, 0.0368, 0.3045, 0.1530], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0117, 0.0121, 0.0124, 0.0113, 0.0096, 0.0095, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 07:11:12,406 INFO [finetune.py:976] (3/7) Epoch 26, batch 700, loss[loss=0.2093, simple_loss=0.2894, pruned_loss=0.06458, over 4800.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2446, pruned_loss=0.04978, over 923522.17 frames. ], batch size: 45, lr: 2.98e-03, grad_scale: 16.0 2023-03-27 07:11:52,958 INFO [finetune.py:976] (3/7) Epoch 26, batch 750, loss[loss=0.1563, simple_loss=0.2369, pruned_loss=0.03789, over 4865.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.246, pruned_loss=0.05008, over 931451.69 frames. ], batch size: 34, lr: 2.98e-03, grad_scale: 16.0 2023-03-27 07:12:00,374 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0785, 1.6684, 2.1247, 2.0936, 1.8322, 1.8176, 2.0243, 1.9593], device='cuda:3'), covar=tensor([0.4137, 0.4293, 0.3175, 0.4052, 0.5175, 0.4026, 0.4981, 0.3149], device='cuda:3'), in_proj_covar=tensor([0.0264, 0.0246, 0.0266, 0.0293, 0.0293, 0.0270, 0.0299, 0.0249], 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') 2023-03-27 07:12:09,866 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.082e+02 1.567e+02 1.788e+02 2.169e+02 3.888e+02, threshold=3.576e+02, percent-clipped=1.0 2023-03-27 07:12:26,435 INFO [finetune.py:976] (3/7) Epoch 26, batch 800, loss[loss=0.1601, simple_loss=0.2291, pruned_loss=0.04557, over 4746.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2449, pruned_loss=0.04944, over 937875.34 frames. ], batch size: 23, lr: 2.98e-03, grad_scale: 16.0 2023-03-27 07:12:27,778 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2809, 2.1583, 1.9891, 2.3715, 2.6708, 2.3477, 2.2239, 1.7398], device='cuda:3'), covar=tensor([0.1946, 0.1797, 0.1735, 0.1456, 0.1676, 0.1000, 0.1895, 0.1801], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0209, 0.0215, 0.0197, 0.0244, 0.0191, 0.0215, 0.0204], 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') 2023-03-27 07:12:55,303 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6880, 1.6316, 1.5437, 1.7075, 1.2633, 3.7761, 1.4031, 1.9092], device='cuda:3'), covar=tensor([0.3271, 0.2535, 0.2164, 0.2372, 0.1641, 0.0166, 0.2497, 0.1224], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0117, 0.0120, 0.0124, 0.0113, 0.0096, 0.0095, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 07:13:00,661 INFO [finetune.py:976] (3/7) Epoch 26, batch 850, loss[loss=0.1472, simple_loss=0.2156, pruned_loss=0.03944, over 4775.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.242, pruned_loss=0.04887, over 940707.28 frames. ], batch size: 29, lr: 2.98e-03, grad_scale: 16.0 2023-03-27 07:13:09,693 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=144056.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 07:13:16,916 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.067e+02 1.460e+02 1.746e+02 2.115e+02 7.519e+02, threshold=3.492e+02, percent-clipped=2.0 2023-03-27 07:13:43,955 INFO [finetune.py:976] (3/7) Epoch 26, batch 900, loss[loss=0.154, simple_loss=0.2254, pruned_loss=0.04129, over 4817.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2381, pruned_loss=0.04725, over 943432.77 frames. ], batch size: 38, lr: 2.97e-03, grad_scale: 16.0 2023-03-27 07:13:45,339 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0160, 1.8689, 1.6870, 2.0763, 2.3908, 2.0850, 1.8807, 1.6444], device='cuda:3'), covar=tensor([0.1797, 0.1721, 0.1671, 0.1418, 0.1466, 0.0979, 0.1948, 0.1596], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0209, 0.0214, 0.0197, 0.0243, 0.0190, 0.0215, 0.0204], 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') 2023-03-27 07:13:51,327 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=144104.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 07:13:55,836 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-03-27 07:13:56,106 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8959, 1.8784, 1.7228, 1.9209, 1.5975, 4.4358, 1.5941, 2.0768], device='cuda:3'), covar=tensor([0.3117, 0.2332, 0.2014, 0.2175, 0.1456, 0.0117, 0.2387, 0.1162], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0120, 0.0124, 0.0113, 0.0095, 0.0095, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 07:14:07,459 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=144129.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 07:14:16,859 INFO [finetune.py:976] (3/7) Epoch 26, batch 950, loss[loss=0.2241, simple_loss=0.2893, pruned_loss=0.07944, over 4902.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2364, pruned_loss=0.04698, over 944891.73 frames. ], batch size: 36, lr: 2.97e-03, grad_scale: 16.0 2023-03-27 07:14:33,134 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.128e+02 1.468e+02 1.742e+02 2.065e+02 3.876e+02, threshold=3.485e+02, percent-clipped=2.0 2023-03-27 07:14:39,231 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=144177.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 07:14:50,350 INFO [finetune.py:976] (3/7) Epoch 26, batch 1000, loss[loss=0.1571, simple_loss=0.2371, pruned_loss=0.03853, over 4788.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2392, pruned_loss=0.04838, over 947435.38 frames. ], batch size: 59, lr: 2.97e-03, grad_scale: 16.0 2023-03-27 07:15:19,147 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-03-27 07:15:22,310 INFO [finetune.py:976] (3/7) Epoch 26, batch 1050, loss[loss=0.1926, simple_loss=0.2793, pruned_loss=0.05296, over 4852.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.242, pruned_loss=0.0488, over 949819.04 frames. ], batch size: 49, lr: 2.97e-03, grad_scale: 16.0 2023-03-27 07:15:40,005 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.069e+02 1.483e+02 1.785e+02 2.219e+02 5.161e+02, threshold=3.570e+02, percent-clipped=2.0 2023-03-27 07:16:01,562 INFO [finetune.py:976] (3/7) Epoch 26, batch 1100, loss[loss=0.201, simple_loss=0.2662, pruned_loss=0.06784, over 4793.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.243, pruned_loss=0.04883, over 951132.09 frames. ], batch size: 26, lr: 2.97e-03, grad_scale: 16.0 2023-03-27 07:16:45,339 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-03-27 07:16:45,531 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5926, 2.1920, 2.7015, 1.7257, 2.4594, 2.9092, 1.9492, 2.9535], device='cuda:3'), covar=tensor([0.1234, 0.2101, 0.1675, 0.2283, 0.1088, 0.1451, 0.2742, 0.0814], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0205, 0.0191, 0.0190, 0.0174, 0.0212, 0.0216, 0.0198], 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') 2023-03-27 07:16:56,472 INFO [finetune.py:976] (3/7) Epoch 26, batch 1150, loss[loss=0.1468, simple_loss=0.2163, pruned_loss=0.0386, over 4774.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2432, pruned_loss=0.0483, over 952057.56 frames. ], batch size: 26, lr: 2.97e-03, grad_scale: 16.0 2023-03-27 07:17:13,862 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.072e+02 1.499e+02 1.760e+02 2.197e+02 4.327e+02, threshold=3.521e+02, percent-clipped=2.0 2023-03-27 07:17:28,619 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4159, 2.2839, 2.3298, 1.5678, 2.2527, 2.5297, 2.4437, 1.8782], device='cuda:3'), covar=tensor([0.0572, 0.0609, 0.0692, 0.0897, 0.0707, 0.0624, 0.0579, 0.1126], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0138, 0.0141, 0.0120, 0.0128, 0.0139, 0.0141, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 07:17:30,185 INFO [finetune.py:976] (3/7) Epoch 26, batch 1200, loss[loss=0.1766, simple_loss=0.2404, pruned_loss=0.05643, over 4744.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2417, pruned_loss=0.04761, over 952647.91 frames. ], batch size: 59, lr: 2.97e-03, grad_scale: 16.0 2023-03-27 07:17:39,553 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-03-27 07:17:43,720 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=144411.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:18:03,375 INFO [finetune.py:976] (3/7) Epoch 26, batch 1250, loss[loss=0.1363, simple_loss=0.1953, pruned_loss=0.03869, over 4286.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2402, pruned_loss=0.04811, over 950918.19 frames. ], batch size: 18, lr: 2.97e-03, grad_scale: 16.0 2023-03-27 07:18:11,746 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.57 vs. limit=2.0 2023-03-27 07:18:21,714 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.894e+01 1.606e+02 1.805e+02 2.274e+02 3.881e+02, threshold=3.611e+02, percent-clipped=1.0 2023-03-27 07:18:24,295 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=144472.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:18:37,272 INFO [finetune.py:976] (3/7) Epoch 26, batch 1300, loss[loss=0.1671, simple_loss=0.2372, pruned_loss=0.04846, over 4815.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2378, pruned_loss=0.0474, over 951893.88 frames. ], batch size: 51, lr: 2.97e-03, grad_scale: 16.0 2023-03-27 07:19:12,959 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6252, 1.5241, 1.5106, 1.6680, 1.2282, 3.5827, 1.2958, 1.8562], device='cuda:3'), covar=tensor([0.3356, 0.2566, 0.2202, 0.2446, 0.1767, 0.0200, 0.2782, 0.1261], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0121, 0.0124, 0.0113, 0.0096, 0.0094, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 07:19:21,353 INFO [finetune.py:976] (3/7) Epoch 26, batch 1350, loss[loss=0.1958, simple_loss=0.2646, pruned_loss=0.0635, over 4823.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2381, pruned_loss=0.04775, over 951608.43 frames. ], batch size: 33, lr: 2.97e-03, grad_scale: 16.0 2023-03-27 07:19:39,468 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.034e+02 1.499e+02 1.803e+02 2.073e+02 4.281e+02, threshold=3.607e+02, percent-clipped=1.0 2023-03-27 07:19:54,503 INFO [finetune.py:976] (3/7) Epoch 26, batch 1400, loss[loss=0.1779, simple_loss=0.2712, pruned_loss=0.04231, over 4793.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2413, pruned_loss=0.04863, over 951913.95 frames. ], batch size: 29, lr: 2.97e-03, grad_scale: 16.0 2023-03-27 07:20:16,432 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.0406, 4.9265, 4.6420, 2.7918, 5.0514, 3.8744, 1.4721, 3.6043], device='cuda:3'), covar=tensor([0.2159, 0.2012, 0.1332, 0.2971, 0.0646, 0.0784, 0.3973, 0.1402], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0179, 0.0161, 0.0131, 0.0162, 0.0125, 0.0150, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 07:20:27,740 INFO [finetune.py:976] (3/7) Epoch 26, batch 1450, loss[loss=0.1741, simple_loss=0.2493, pruned_loss=0.04945, over 4874.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2438, pruned_loss=0.04877, over 953689.03 frames. ], batch size: 32, lr: 2.97e-03, grad_scale: 16.0 2023-03-27 07:20:45,824 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.092e+01 1.571e+02 1.855e+02 2.334e+02 4.645e+02, threshold=3.710e+02, percent-clipped=2.0 2023-03-27 07:20:51,889 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5774, 1.4891, 1.5089, 1.6143, 1.1030, 3.5708, 1.2072, 1.8168], device='cuda:3'), covar=tensor([0.3393, 0.2651, 0.2226, 0.2461, 0.1877, 0.0225, 0.2630, 0.1262], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0121, 0.0124, 0.0113, 0.0095, 0.0094, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 07:21:01,364 INFO [finetune.py:976] (3/7) Epoch 26, batch 1500, loss[loss=0.206, simple_loss=0.2721, pruned_loss=0.07001, over 4902.00 frames. ], tot_loss[loss=0.173, simple_loss=0.246, pruned_loss=0.05001, over 951735.06 frames. ], batch size: 36, lr: 2.97e-03, grad_scale: 16.0 2023-03-27 07:21:04,772 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-27 07:21:50,337 INFO [finetune.py:976] (3/7) Epoch 26, batch 1550, loss[loss=0.1843, simple_loss=0.2495, pruned_loss=0.05954, over 4901.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2459, pruned_loss=0.04999, over 951945.47 frames. ], batch size: 37, lr: 2.97e-03, grad_scale: 16.0 2023-03-27 07:22:14,455 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.85 vs. limit=5.0 2023-03-27 07:22:18,423 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=144767.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:22:18,964 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.202e+02 1.547e+02 1.850e+02 2.044e+02 4.068e+02, threshold=3.700e+02, percent-clipped=1.0 2023-03-27 07:22:35,465 INFO [finetune.py:976] (3/7) Epoch 26, batch 1600, loss[loss=0.1497, simple_loss=0.221, pruned_loss=0.03925, over 4700.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.244, pruned_loss=0.04945, over 953808.97 frames. ], batch size: 23, lr: 2.97e-03, grad_scale: 16.0 2023-03-27 07:23:09,214 INFO [finetune.py:976] (3/7) Epoch 26, batch 1650, loss[loss=0.1943, simple_loss=0.2556, pruned_loss=0.0665, over 4830.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2426, pruned_loss=0.04992, over 953688.01 frames. ], batch size: 33, lr: 2.97e-03, grad_scale: 16.0 2023-03-27 07:23:26,320 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.503e+01 1.460e+02 1.738e+02 2.010e+02 3.428e+02, threshold=3.475e+02, percent-clipped=0.0 2023-03-27 07:23:42,420 INFO [finetune.py:976] (3/7) Epoch 26, batch 1700, loss[loss=0.1537, simple_loss=0.2237, pruned_loss=0.04184, over 4820.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2405, pruned_loss=0.04898, over 953463.24 frames. ], batch size: 41, lr: 2.97e-03, grad_scale: 16.0 2023-03-27 07:23:57,309 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.5370, 5.1122, 4.8948, 3.4659, 5.1907, 4.0964, 1.7837, 4.0297], device='cuda:3'), covar=tensor([0.1770, 0.1578, 0.1262, 0.2411, 0.0933, 0.0678, 0.3871, 0.1038], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0178, 0.0160, 0.0130, 0.0160, 0.0124, 0.0148, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 07:24:19,071 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1311, 2.0715, 1.6217, 2.0947, 2.0614, 1.8029, 2.3962, 2.1251], device='cuda:3'), covar=tensor([0.1349, 0.1864, 0.2774, 0.2370, 0.2385, 0.1598, 0.2648, 0.1697], device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0190, 0.0236, 0.0254, 0.0251, 0.0207, 0.0215, 0.0203], 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') 2023-03-27 07:24:25,822 INFO [finetune.py:976] (3/7) Epoch 26, batch 1750, loss[loss=0.1665, simple_loss=0.2473, pruned_loss=0.04283, over 4904.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2422, pruned_loss=0.04968, over 953612.69 frames. ], batch size: 36, lr: 2.97e-03, grad_scale: 16.0 2023-03-27 07:24:42,913 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.967e+01 1.592e+02 1.823e+02 2.389e+02 4.337e+02, threshold=3.645e+02, percent-clipped=3.0 2023-03-27 07:24:59,586 INFO [finetune.py:976] (3/7) Epoch 26, batch 1800, loss[loss=0.2245, simple_loss=0.3115, pruned_loss=0.06872, over 4904.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.244, pruned_loss=0.04991, over 953740.39 frames. ], batch size: 43, lr: 2.97e-03, grad_scale: 32.0 2023-03-27 07:25:02,601 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=144996.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:25:23,064 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145027.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:25:33,497 INFO [finetune.py:976] (3/7) Epoch 26, batch 1850, loss[loss=0.1646, simple_loss=0.2394, pruned_loss=0.04491, over 4895.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2455, pruned_loss=0.05071, over 954569.50 frames. ], batch size: 37, lr: 2.97e-03, grad_scale: 32.0 2023-03-27 07:25:43,144 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145057.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 07:25:48,500 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145066.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:25:49,085 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145067.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:25:49,577 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.082e+02 1.533e+02 1.829e+02 2.183e+02 4.392e+02, threshold=3.659e+02, percent-clipped=3.0 2023-03-27 07:25:50,304 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1002, 2.0214, 2.1550, 1.4288, 2.0976, 2.2478, 2.1804, 1.7060], device='cuda:3'), covar=tensor([0.0577, 0.0598, 0.0682, 0.0875, 0.0671, 0.0604, 0.0577, 0.1101], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0136, 0.0140, 0.0118, 0.0127, 0.0137, 0.0139, 0.0160], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 07:26:01,862 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1442, 1.2795, 1.2091, 1.2947, 1.3586, 2.4720, 1.2309, 1.4238], device='cuda:3'), covar=tensor([0.1070, 0.1859, 0.1142, 0.0927, 0.1706, 0.0400, 0.1503, 0.1796], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0073, 0.0076, 0.0091, 0.0080, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 07:26:03,667 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145088.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:26:06,519 INFO [finetune.py:976] (3/7) Epoch 26, batch 1900, loss[loss=0.1163, simple_loss=0.2002, pruned_loss=0.0162, over 4791.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2471, pruned_loss=0.05151, over 954540.78 frames. ], batch size: 25, lr: 2.97e-03, grad_scale: 32.0 2023-03-27 07:26:21,297 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=145115.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:26:22,929 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.65 vs. limit=5.0 2023-03-27 07:26:29,576 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145127.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:26:35,661 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7404, 2.5064, 2.1628, 1.1268, 2.2788, 2.1217, 1.9591, 2.3830], device='cuda:3'), covar=tensor([0.0994, 0.0923, 0.1626, 0.2088, 0.1624, 0.2265, 0.2191, 0.0958], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0191, 0.0199, 0.0181, 0.0209, 0.0209, 0.0223, 0.0194], 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') 2023-03-27 07:26:46,688 INFO [finetune.py:976] (3/7) Epoch 26, batch 1950, loss[loss=0.1684, simple_loss=0.234, pruned_loss=0.05143, over 4720.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2462, pruned_loss=0.05113, over 955057.61 frames. ], batch size: 59, lr: 2.97e-03, grad_scale: 32.0 2023-03-27 07:27:15,913 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.125e+02 1.577e+02 1.831e+02 2.188e+02 4.363e+02, threshold=3.662e+02, percent-clipped=3.0 2023-03-27 07:27:40,103 INFO [finetune.py:976] (3/7) Epoch 26, batch 2000, loss[loss=0.1742, simple_loss=0.2495, pruned_loss=0.04947, over 4799.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.243, pruned_loss=0.05009, over 952636.81 frames. ], batch size: 51, lr: 2.97e-03, grad_scale: 32.0 2023-03-27 07:27:59,370 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8352, 1.5124, 1.8117, 1.1618, 1.8842, 1.8853, 1.8018, 1.2343], device='cuda:3'), covar=tensor([0.0599, 0.0995, 0.0767, 0.0961, 0.0775, 0.0680, 0.0734, 0.1897], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0135, 0.0139, 0.0117, 0.0125, 0.0136, 0.0138, 0.0158], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 07:28:13,277 INFO [finetune.py:976] (3/7) Epoch 26, batch 2050, loss[loss=0.1749, simple_loss=0.2468, pruned_loss=0.0515, over 4800.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2405, pruned_loss=0.04919, over 952154.65 frames. ], batch size: 51, lr: 2.97e-03, grad_scale: 32.0 2023-03-27 07:28:15,895 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.96 vs. limit=5.0 2023-03-27 07:28:17,509 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.5508, 1.4503, 1.3995, 0.7400, 1.5085, 1.6453, 1.7041, 1.3883], device='cuda:3'), covar=tensor([0.0797, 0.0583, 0.0488, 0.0504, 0.0396, 0.0555, 0.0256, 0.0538], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0147, 0.0128, 0.0123, 0.0130, 0.0129, 0.0142, 0.0149], device='cuda:3'), out_proj_covar=tensor([8.8926e-05, 1.0613e-04, 9.1001e-05, 8.6173e-05, 9.0811e-05, 9.1611e-05, 1.0081e-04, 1.0621e-04], device='cuda:3') 2023-03-27 07:28:30,375 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.187e+01 1.436e+02 1.792e+02 2.264e+02 4.038e+02, threshold=3.583e+02, percent-clipped=1.0 2023-03-27 07:28:32,298 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.0172, 4.2061, 3.9035, 2.4408, 4.2522, 3.3249, 1.4161, 3.2430], device='cuda:3'), covar=tensor([0.2158, 0.2030, 0.1916, 0.3092, 0.1134, 0.0975, 0.4266, 0.1289], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0178, 0.0160, 0.0130, 0.0160, 0.0123, 0.0148, 0.0123], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 07:28:40,034 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1881, 1.8093, 2.6970, 1.6360, 2.3174, 2.4920, 1.7603, 2.5214], device='cuda:3'), covar=tensor([0.1381, 0.2302, 0.1613, 0.2176, 0.0972, 0.1398, 0.2928, 0.0969], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0206, 0.0192, 0.0190, 0.0175, 0.0213, 0.0216, 0.0199], 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') 2023-03-27 07:28:45,861 INFO [finetune.py:976] (3/7) Epoch 26, batch 2100, loss[loss=0.1419, simple_loss=0.2067, pruned_loss=0.03856, over 4802.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2407, pruned_loss=0.04951, over 952861.37 frames. ], batch size: 25, lr: 2.97e-03, grad_scale: 32.0 2023-03-27 07:29:19,663 INFO [finetune.py:976] (3/7) Epoch 26, batch 2150, loss[loss=0.157, simple_loss=0.2373, pruned_loss=0.03836, over 4757.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2433, pruned_loss=0.04985, over 954638.14 frames. ], batch size: 27, lr: 2.97e-03, grad_scale: 32.0 2023-03-27 07:29:28,920 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145352.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 07:29:47,543 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.022e+02 1.494e+02 1.709e+02 2.298e+02 6.165e+02, threshold=3.419e+02, percent-clipped=3.0 2023-03-27 07:29:47,844 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-03-27 07:29:49,498 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145371.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:29:49,568 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-03-27 07:29:50,775 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6586, 0.7664, 1.7378, 1.7006, 1.5689, 1.5521, 1.6331, 1.7254], device='cuda:3'), covar=tensor([0.3918, 0.3939, 0.3428, 0.3387, 0.4901, 0.3739, 0.4110, 0.2999], device='cuda:3'), in_proj_covar=tensor([0.0264, 0.0247, 0.0265, 0.0293, 0.0293, 0.0270, 0.0299, 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') 2023-03-27 07:29:56,781 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145383.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:30:02,629 INFO [finetune.py:976] (3/7) Epoch 26, batch 2200, loss[loss=0.1558, simple_loss=0.2288, pruned_loss=0.04133, over 4814.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2443, pruned_loss=0.04981, over 954463.66 frames. ], batch size: 25, lr: 2.97e-03, grad_scale: 32.0 2023-03-27 07:30:22,393 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-03-27 07:30:23,254 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145422.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:30:29,366 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145432.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:30:36,242 INFO [finetune.py:976] (3/7) Epoch 26, batch 2250, loss[loss=0.1388, simple_loss=0.2024, pruned_loss=0.03765, over 4328.00 frames. ], tot_loss[loss=0.173, simple_loss=0.246, pruned_loss=0.05002, over 956080.23 frames. ], batch size: 18, lr: 2.97e-03, grad_scale: 32.0 2023-03-27 07:30:37,635 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0235, 1.8699, 1.7474, 1.7538, 2.2443, 2.2214, 1.9266, 1.8067], device='cuda:3'), covar=tensor([0.0358, 0.0318, 0.0574, 0.0337, 0.0218, 0.0414, 0.0299, 0.0370], device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0109, 0.0149, 0.0113, 0.0103, 0.0118, 0.0104, 0.0115], device='cuda:3'), out_proj_covar=tensor([8.0326e-05, 8.3024e-05, 1.1643e-04, 8.6571e-05, 7.9891e-05, 8.6772e-05, 7.7549e-05, 8.7307e-05], device='cuda:3') 2023-03-27 07:30:44,588 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2080, 1.3175, 1.3586, 1.3198, 1.5005, 2.4876, 1.2964, 1.4556], device='cuda:3'), covar=tensor([0.1033, 0.1900, 0.1123, 0.0938, 0.1633, 0.0371, 0.1558, 0.1830], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0072, 0.0076, 0.0090, 0.0080, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 07:30:53,946 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.189e+01 1.488e+02 1.760e+02 2.143e+02 3.776e+02, threshold=3.521e+02, percent-clipped=2.0 2023-03-27 07:31:08,986 INFO [finetune.py:976] (3/7) Epoch 26, batch 2300, loss[loss=0.1483, simple_loss=0.217, pruned_loss=0.03983, over 4854.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2463, pruned_loss=0.05009, over 954708.45 frames. ], batch size: 31, lr: 2.97e-03, grad_scale: 32.0 2023-03-27 07:31:24,704 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1675, 1.8090, 1.8294, 0.8818, 2.1059, 2.2549, 2.0794, 1.8262], device='cuda:3'), covar=tensor([0.0882, 0.0737, 0.0558, 0.0687, 0.0506, 0.0717, 0.0469, 0.0746], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0148, 0.0129, 0.0123, 0.0130, 0.0130, 0.0142, 0.0149], device='cuda:3'), out_proj_covar=tensor([8.9352e-05, 1.0681e-04, 9.1681e-05, 8.6634e-05, 9.1296e-05, 9.2064e-05, 1.0134e-04, 1.0685e-04], device='cuda:3') 2023-03-27 07:31:42,487 INFO [finetune.py:976] (3/7) Epoch 26, batch 2350, loss[loss=0.1707, simple_loss=0.2396, pruned_loss=0.05088, over 4813.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2444, pruned_loss=0.04982, over 955578.56 frames. ], batch size: 41, lr: 2.96e-03, grad_scale: 32.0 2023-03-27 07:32:06,624 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.008e+02 1.544e+02 1.840e+02 2.226e+02 4.643e+02, threshold=3.680e+02, percent-clipped=1.0 2023-03-27 07:32:34,351 INFO [finetune.py:976] (3/7) Epoch 26, batch 2400, loss[loss=0.1546, simple_loss=0.2235, pruned_loss=0.04283, over 4776.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2415, pruned_loss=0.04865, over 953840.53 frames. ], batch size: 26, lr: 2.96e-03, grad_scale: 32.0 2023-03-27 07:32:35,675 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145594.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:32:42,976 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-03-27 07:33:17,102 INFO [finetune.py:976] (3/7) Epoch 26, batch 2450, loss[loss=0.1423, simple_loss=0.2233, pruned_loss=0.03065, over 4892.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2391, pruned_loss=0.04787, over 954197.85 frames. ], batch size: 32, lr: 2.96e-03, grad_scale: 32.0 2023-03-27 07:33:17,836 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145643.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:33:23,878 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145652.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 07:33:26,201 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145655.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 07:33:34,916 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.521e+01 1.449e+02 1.824e+02 2.163e+02 4.630e+02, threshold=3.648e+02, percent-clipped=2.0 2023-03-27 07:33:42,864 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 2023-03-27 07:33:45,667 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145683.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:33:51,041 INFO [finetune.py:976] (3/7) Epoch 26, batch 2500, loss[loss=0.1821, simple_loss=0.2489, pruned_loss=0.05765, over 4806.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.239, pruned_loss=0.04765, over 951934.05 frames. ], batch size: 25, lr: 2.96e-03, grad_scale: 32.0 2023-03-27 07:33:55,945 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=145700.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:33:58,927 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145704.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:34:00,632 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8815, 1.4391, 1.9911, 1.9169, 1.6976, 1.6723, 1.8628, 1.8576], device='cuda:3'), covar=tensor([0.4292, 0.4199, 0.3216, 0.3875, 0.5026, 0.4014, 0.4829, 0.3151], device='cuda:3'), in_proj_covar=tensor([0.0267, 0.0248, 0.0267, 0.0296, 0.0295, 0.0272, 0.0302, 0.0252], 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') 2023-03-27 07:34:11,658 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145722.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:34:15,144 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145727.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:34:16,503 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-27 07:34:17,566 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=145731.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:34:24,601 INFO [finetune.py:976] (3/7) Epoch 26, batch 2550, loss[loss=0.2132, simple_loss=0.2785, pruned_loss=0.07394, over 4822.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2419, pruned_loss=0.04835, over 952391.29 frames. ], batch size: 40, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:34:42,325 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.531e+01 1.551e+02 1.807e+02 2.106e+02 4.459e+02, threshold=3.615e+02, percent-clipped=2.0 2023-03-27 07:34:43,553 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=145770.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:35:08,875 INFO [finetune.py:976] (3/7) Epoch 26, batch 2600, loss[loss=0.1526, simple_loss=0.2246, pruned_loss=0.04035, over 4752.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2439, pruned_loss=0.04931, over 951600.71 frames. ], batch size: 26, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:35:22,262 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.1869, 1.2409, 1.2921, 0.6960, 1.2720, 1.5001, 1.5428, 1.2018], device='cuda:3'), covar=tensor([0.0952, 0.0639, 0.0573, 0.0496, 0.0505, 0.0648, 0.0321, 0.0696], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0148, 0.0128, 0.0123, 0.0130, 0.0129, 0.0142, 0.0149], device='cuda:3'), out_proj_covar=tensor([8.9081e-05, 1.0634e-04, 9.1413e-05, 8.6157e-05, 9.1110e-05, 9.1515e-05, 1.0108e-04, 1.0665e-04], device='cuda:3') 2023-03-27 07:35:25,064 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-03-27 07:35:28,223 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145821.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 07:35:37,530 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0826, 1.9339, 1.7596, 1.9176, 1.8542, 1.8348, 1.8922, 2.4843], device='cuda:3'), covar=tensor([0.3205, 0.3563, 0.2834, 0.3066, 0.3635, 0.2195, 0.3343, 0.1386], device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0264, 0.0236, 0.0276, 0.0258, 0.0229, 0.0257, 0.0237], 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') 2023-03-27 07:35:42,709 INFO [finetune.py:976] (3/7) Epoch 26, batch 2650, loss[loss=0.1868, simple_loss=0.2666, pruned_loss=0.05357, over 4924.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2462, pruned_loss=0.0502, over 951296.97 frames. ], batch size: 42, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:35:44,118 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.77 vs. limit=5.0 2023-03-27 07:35:56,409 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0102, 1.8154, 2.3798, 4.0533, 2.7586, 2.7288, 0.7621, 3.4691], device='cuda:3'), covar=tensor([0.1730, 0.1379, 0.1433, 0.0520, 0.0749, 0.1716, 0.2121, 0.0377], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0133, 0.0163, 0.0101, 0.0134, 0.0125, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 07:36:00,020 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.986e+01 1.512e+02 1.783e+02 2.110e+02 4.476e+02, threshold=3.566e+02, percent-clipped=1.0 2023-03-27 07:36:09,563 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145882.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 07:36:13,003 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.4968, 3.9160, 4.1120, 4.3096, 4.2448, 3.9329, 4.5594, 1.3894], device='cuda:3'), covar=tensor([0.0780, 0.0808, 0.0826, 0.1132, 0.1217, 0.1471, 0.0689, 0.5604], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0248, 0.0280, 0.0296, 0.0336, 0.0287, 0.0304, 0.0301], 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') 2023-03-27 07:36:16,424 INFO [finetune.py:976] (3/7) Epoch 26, batch 2700, loss[loss=0.1504, simple_loss=0.2313, pruned_loss=0.03473, over 4701.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2435, pruned_loss=0.04869, over 951435.31 frames. ], batch size: 23, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:36:49,654 INFO [finetune.py:976] (3/7) Epoch 26, batch 2750, loss[loss=0.1738, simple_loss=0.2484, pruned_loss=0.04957, over 4768.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.242, pruned_loss=0.04847, over 952670.15 frames. ], batch size: 27, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:36:53,985 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.2165, 2.8380, 2.6390, 1.3800, 2.8882, 2.2734, 2.2153, 2.6319], device='cuda:3'), covar=tensor([0.0968, 0.0792, 0.1604, 0.2122, 0.1539, 0.2191, 0.2039, 0.1078], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0192, 0.0200, 0.0181, 0.0210, 0.0211, 0.0223, 0.0196], 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') 2023-03-27 07:36:55,118 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145950.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 07:37:01,835 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6908, 1.7409, 2.2852, 1.8895, 2.0679, 4.5806, 1.8888, 1.8974], device='cuda:3'), covar=tensor([0.0923, 0.1722, 0.1043, 0.0922, 0.1366, 0.0145, 0.1339, 0.1639], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0073, 0.0076, 0.0091, 0.0081, 0.0086, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-27 07:37:07,588 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.011e+02 1.569e+02 1.804e+02 2.200e+02 3.850e+02, threshold=3.609e+02, percent-clipped=2.0 2023-03-27 07:37:07,871 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-03-27 07:37:29,450 INFO [finetune.py:976] (3/7) Epoch 26, batch 2800, loss[loss=0.1624, simple_loss=0.2277, pruned_loss=0.0485, over 4919.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2389, pruned_loss=0.04775, over 954911.75 frames. ], batch size: 37, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:37:39,468 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145999.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:37:57,221 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2550, 2.2581, 1.7718, 0.9141, 1.9967, 1.8851, 1.7166, 2.0336], device='cuda:3'), covar=tensor([0.0941, 0.0594, 0.1224, 0.1724, 0.1128, 0.1930, 0.1951, 0.0737], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0192, 0.0199, 0.0181, 0.0210, 0.0210, 0.0223, 0.0195], 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') 2023-03-27 07:37:59,037 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.1737, 2.7236, 3.2456, 2.2193, 2.9869, 3.5305, 2.4471, 3.2237], device='cuda:3'), covar=tensor([0.1077, 0.1620, 0.1468, 0.1805, 0.0887, 0.1180, 0.2319, 0.0829], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0206, 0.0192, 0.0189, 0.0174, 0.0213, 0.0216, 0.0199], 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') 2023-03-27 07:38:14,434 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146027.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:38:14,489 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0934, 1.9712, 1.8550, 2.0721, 1.9104, 1.9706, 1.9616, 2.5546], device='cuda:3'), covar=tensor([0.3224, 0.4124, 0.3121, 0.3557, 0.4026, 0.2352, 0.3762, 0.1513], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0265, 0.0237, 0.0277, 0.0259, 0.0229, 0.0257, 0.0238], 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') 2023-03-27 07:38:24,536 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9709, 1.8520, 1.5114, 1.3959, 1.8810, 1.6547, 1.8235, 1.8981], device='cuda:3'), covar=tensor([0.1272, 0.1747, 0.2804, 0.2555, 0.2578, 0.1697, 0.3034, 0.1676], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0189, 0.0235, 0.0251, 0.0249, 0.0206, 0.0214, 0.0202], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 07:38:24,990 INFO [finetune.py:976] (3/7) Epoch 26, batch 2850, loss[loss=0.2441, simple_loss=0.3024, pruned_loss=0.09288, over 4892.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2381, pruned_loss=0.04796, over 954613.22 frames. ], batch size: 32, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:38:41,788 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0542, 1.9897, 1.7091, 1.8918, 1.8240, 1.8397, 1.9072, 2.5510], device='cuda:3'), covar=tensor([0.3508, 0.3791, 0.3008, 0.3555, 0.4025, 0.2291, 0.3459, 0.1656], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0265, 0.0236, 0.0277, 0.0258, 0.0229, 0.0257, 0.0238], 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') 2023-03-27 07:38:42,242 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.185e+01 1.523e+02 1.819e+02 2.110e+02 4.930e+02, threshold=3.638e+02, percent-clipped=2.0 2023-03-27 07:38:46,431 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=146075.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:38:58,297 INFO [finetune.py:976] (3/7) Epoch 26, batch 2900, loss[loss=0.1642, simple_loss=0.2441, pruned_loss=0.0421, over 4884.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2401, pruned_loss=0.04848, over 954284.09 frames. ], batch size: 32, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:39:31,499 INFO [finetune.py:976] (3/7) Epoch 26, batch 2950, loss[loss=0.2131, simple_loss=0.2775, pruned_loss=0.07439, over 4901.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2427, pruned_loss=0.04888, over 954146.00 frames. ], batch size: 35, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:39:36,365 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146149.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:39:45,151 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5919, 1.4037, 1.9984, 3.2629, 2.1139, 2.3517, 0.9174, 2.8515], device='cuda:3'), covar=tensor([0.1884, 0.1718, 0.1531, 0.0775, 0.0963, 0.1660, 0.2229, 0.0516], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0132, 0.0163, 0.0101, 0.0134, 0.0125, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 07:39:49,299 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.099e+02 1.550e+02 1.859e+02 2.106e+02 3.478e+02, threshold=3.719e+02, percent-clipped=0.0 2023-03-27 07:39:54,693 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146177.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 07:40:01,984 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1405, 1.9479, 1.7359, 2.0979, 2.6384, 2.1628, 2.2264, 1.6181], device='cuda:3'), covar=tensor([0.2051, 0.1928, 0.1847, 0.1493, 0.1732, 0.1137, 0.1900, 0.1890], device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0212, 0.0217, 0.0200, 0.0248, 0.0193, 0.0220, 0.0207], 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') 2023-03-27 07:40:04,822 INFO [finetune.py:976] (3/7) Epoch 26, batch 3000, loss[loss=0.1989, simple_loss=0.2777, pruned_loss=0.06001, over 4877.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2447, pruned_loss=0.04982, over 952781.75 frames. ], batch size: 43, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:40:04,822 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-27 07:40:19,937 INFO [finetune.py:1010] (3/7) Epoch 26, validation: loss=0.1577, simple_loss=0.2252, pruned_loss=0.04507, over 2265189.00 frames. 2023-03-27 07:40:19,938 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6469MB 2023-03-27 07:40:35,403 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146210.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:40:50,598 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-03-27 07:40:56,712 INFO [finetune.py:976] (3/7) Epoch 26, batch 3050, loss[loss=0.1272, simple_loss=0.2088, pruned_loss=0.02276, over 4847.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2452, pruned_loss=0.04907, over 953974.24 frames. ], batch size: 47, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:41:02,124 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146250.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:41:14,892 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.646e+01 1.408e+02 1.795e+02 2.163e+02 4.679e+02, threshold=3.589e+02, percent-clipped=3.0 2023-03-27 07:41:29,844 INFO [finetune.py:976] (3/7) Epoch 26, batch 3100, loss[loss=0.2016, simple_loss=0.2606, pruned_loss=0.07129, over 4842.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2436, pruned_loss=0.04903, over 952723.53 frames. ], batch size: 44, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:41:34,032 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=146298.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:41:34,682 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146299.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:41:34,708 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0537, 2.0314, 2.0430, 1.4156, 2.0815, 2.2091, 2.1032, 1.7654], device='cuda:3'), covar=tensor([0.0543, 0.0573, 0.0717, 0.0848, 0.0770, 0.0555, 0.0534, 0.0988], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0137, 0.0141, 0.0119, 0.0128, 0.0138, 0.0139, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 07:42:02,608 INFO [finetune.py:976] (3/7) Epoch 26, batch 3150, loss[loss=0.1429, simple_loss=0.2168, pruned_loss=0.03449, over 4780.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2416, pruned_loss=0.04874, over 952859.27 frames. ], batch size: 28, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:42:06,560 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=146347.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:42:21,140 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.044e+02 1.464e+02 1.851e+02 2.163e+02 3.423e+02, threshold=3.701e+02, percent-clipped=0.0 2023-03-27 07:42:38,056 INFO [finetune.py:976] (3/7) Epoch 26, batch 3200, loss[loss=0.1515, simple_loss=0.2167, pruned_loss=0.04309, over 4437.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2385, pruned_loss=0.0479, over 952455.65 frames. ], batch size: 19, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:42:57,670 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.3995, 1.3433, 1.3633, 0.7929, 1.4284, 1.5693, 1.6529, 1.2930], device='cuda:3'), covar=tensor([0.0817, 0.0584, 0.0472, 0.0460, 0.0467, 0.0555, 0.0280, 0.0575], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0149, 0.0128, 0.0123, 0.0131, 0.0129, 0.0142, 0.0149], device='cuda:3'), out_proj_covar=tensor([8.9237e-05, 1.0686e-04, 9.1113e-05, 8.6529e-05, 9.1407e-05, 9.1685e-05, 1.0143e-04, 1.0661e-04], device='cuda:3') 2023-03-27 07:43:28,048 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0 2023-03-27 07:43:35,343 INFO [finetune.py:976] (3/7) Epoch 26, batch 3250, loss[loss=0.1793, simple_loss=0.2605, pruned_loss=0.04909, over 4896.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2385, pruned_loss=0.04788, over 952811.87 frames. ], batch size: 35, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:43:53,729 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.041e+02 1.451e+02 1.808e+02 2.175e+02 4.535e+02, threshold=3.616e+02, percent-clipped=3.0 2023-03-27 07:43:58,660 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146477.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 07:44:08,659 INFO [finetune.py:976] (3/7) Epoch 26, batch 3300, loss[loss=0.1655, simple_loss=0.2499, pruned_loss=0.04055, over 4790.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2423, pruned_loss=0.04909, over 952707.66 frames. ], batch size: 29, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:44:17,123 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146505.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:44:30,705 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=146525.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 07:44:41,547 INFO [finetune.py:976] (3/7) Epoch 26, batch 3350, loss[loss=0.2006, simple_loss=0.2665, pruned_loss=0.06733, over 4796.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2445, pruned_loss=0.04926, over 953221.52 frames. ], batch size: 29, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:45:00,357 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.852e+01 1.562e+02 1.841e+02 2.282e+02 4.006e+02, threshold=3.682e+02, percent-clipped=1.0 2023-03-27 07:45:14,914 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146591.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:45:15,434 INFO [finetune.py:976] (3/7) Epoch 26, batch 3400, loss[loss=0.122, simple_loss=0.198, pruned_loss=0.02302, over 4748.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2451, pruned_loss=0.04983, over 950437.56 frames. ], batch size: 27, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:45:17,975 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6244, 1.5257, 1.8926, 1.2283, 1.7451, 1.8849, 1.4715, 2.0584], device='cuda:3'), covar=tensor([0.1194, 0.2274, 0.1337, 0.1815, 0.0927, 0.1286, 0.3090, 0.0889], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0207, 0.0193, 0.0190, 0.0175, 0.0213, 0.0217, 0.0200], 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') 2023-03-27 07:45:27,443 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2569, 2.2078, 1.7981, 2.1989, 2.2119, 1.8985, 2.4206, 2.2472], device='cuda:3'), covar=tensor([0.1374, 0.1818, 0.2975, 0.2344, 0.2466, 0.1752, 0.2479, 0.1720], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0188, 0.0234, 0.0251, 0.0247, 0.0205, 0.0213, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 07:45:48,138 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146626.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:45:58,658 INFO [finetune.py:976] (3/7) Epoch 26, batch 3450, loss[loss=0.167, simple_loss=0.2393, pruned_loss=0.04738, over 4786.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2447, pruned_loss=0.04957, over 950088.27 frames. ], batch size: 51, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:46:04,883 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146652.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:46:17,061 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.040e+02 1.450e+02 1.708e+02 2.017e+02 4.995e+02, threshold=3.417e+02, percent-clipped=1.0 2023-03-27 07:46:28,550 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146687.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 07:46:32,400 INFO [finetune.py:976] (3/7) Epoch 26, batch 3500, loss[loss=0.1932, simple_loss=0.2545, pruned_loss=0.06596, over 4839.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.244, pruned_loss=0.05006, over 951105.70 frames. ], batch size: 33, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:46:32,508 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4474, 1.4603, 1.9405, 1.7142, 1.6454, 3.4907, 1.4145, 1.6137], device='cuda:3'), covar=tensor([0.0997, 0.1771, 0.1091, 0.0979, 0.1548, 0.0234, 0.1442, 0.1754], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0081, 0.0073, 0.0076, 0.0090, 0.0080, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 07:46:45,220 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.1067, 4.4785, 4.6664, 4.9122, 4.8685, 4.5406, 5.2146, 1.7508], device='cuda:3'), covar=tensor([0.0760, 0.0879, 0.0908, 0.0972, 0.1172, 0.1594, 0.0561, 0.5856], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0249, 0.0281, 0.0296, 0.0339, 0.0287, 0.0304, 0.0301], 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') 2023-03-27 07:47:05,274 INFO [finetune.py:976] (3/7) Epoch 26, batch 3550, loss[loss=0.1467, simple_loss=0.2239, pruned_loss=0.03479, over 4861.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.242, pruned_loss=0.04969, over 953354.78 frames. ], batch size: 31, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:47:22,723 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.051e+02 1.430e+02 1.748e+02 2.315e+02 5.079e+02, threshold=3.497e+02, percent-clipped=6.0 2023-03-27 07:47:38,113 INFO [finetune.py:976] (3/7) Epoch 26, batch 3600, loss[loss=0.1376, simple_loss=0.2082, pruned_loss=0.03345, over 4769.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2394, pruned_loss=0.04917, over 949939.35 frames. ], batch size: 28, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:47:47,323 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146805.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:48:04,665 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2985, 2.0994, 1.8020, 1.8547, 2.1697, 1.9057, 2.2600, 2.2268], device='cuda:3'), covar=tensor([0.1249, 0.1767, 0.2916, 0.2467, 0.2611, 0.1752, 0.2901, 0.1610], device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0189, 0.0235, 0.0251, 0.0248, 0.0206, 0.0214, 0.0201], device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 07:48:06,909 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3932, 1.2698, 1.6021, 2.4404, 1.6229, 2.0533, 0.8032, 2.1745], device='cuda:3'), covar=tensor([0.1747, 0.1481, 0.1189, 0.0696, 0.0990, 0.1386, 0.1731, 0.0588], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0132, 0.0163, 0.0101, 0.0134, 0.0124, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 07:48:26,738 INFO [finetune.py:976] (3/7) Epoch 26, batch 3650, loss[loss=0.1569, simple_loss=0.2383, pruned_loss=0.03778, over 4782.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2398, pruned_loss=0.04891, over 950175.33 frames. ], batch size: 26, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:48:28,094 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146844.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:48:39,244 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=146853.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:48:49,566 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8082, 1.6619, 1.4319, 1.2974, 1.6283, 1.6270, 1.5916, 2.1579], device='cuda:3'), covar=tensor([0.3801, 0.3293, 0.3330, 0.3533, 0.3704, 0.2493, 0.3256, 0.1938], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0265, 0.0237, 0.0277, 0.0259, 0.0230, 0.0258, 0.0239], 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') 2023-03-27 07:48:53,562 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.488e+01 1.511e+02 1.813e+02 2.229e+02 3.524e+02, threshold=3.627e+02, percent-clipped=1.0 2023-03-27 07:49:12,460 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0277, 1.6548, 2.1860, 1.5718, 2.0621, 2.1535, 1.5929, 2.1842], device='cuda:3'), covar=tensor([0.0929, 0.1665, 0.1123, 0.1498, 0.0678, 0.1009, 0.2425, 0.0729], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0208, 0.0193, 0.0190, 0.0176, 0.0214, 0.0218, 0.0200], 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') 2023-03-27 07:49:12,953 INFO [finetune.py:976] (3/7) Epoch 26, batch 3700, loss[loss=0.2501, simple_loss=0.3117, pruned_loss=0.09425, over 4201.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2439, pruned_loss=0.05013, over 951982.21 frames. ], batch size: 65, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:49:21,431 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146905.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:49:37,171 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.37 vs. limit=5.0 2023-03-27 07:49:46,522 INFO [finetune.py:976] (3/7) Epoch 26, batch 3750, loss[loss=0.1816, simple_loss=0.2564, pruned_loss=0.05345, over 4764.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.246, pruned_loss=0.05045, over 953644.84 frames. ], batch size: 26, lr: 2.96e-03, grad_scale: 16.0 2023-03-27 07:49:49,619 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146947.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:49:52,548 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4515, 1.3364, 1.5344, 2.4767, 1.6399, 2.2115, 0.8435, 2.1472], device='cuda:3'), covar=tensor([0.1801, 0.1422, 0.1225, 0.0675, 0.0982, 0.1053, 0.1746, 0.0615], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0116, 0.0133, 0.0165, 0.0102, 0.0134, 0.0125, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 07:50:03,847 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.820e+01 1.503e+02 1.791e+02 2.461e+02 5.017e+02, threshold=3.581e+02, percent-clipped=5.0 2023-03-27 07:50:12,660 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146982.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 07:50:19,582 INFO [finetune.py:976] (3/7) Epoch 26, batch 3800, loss[loss=0.167, simple_loss=0.2377, pruned_loss=0.04819, over 4834.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2478, pruned_loss=0.05116, over 956106.20 frames. ], batch size: 30, lr: 2.95e-03, grad_scale: 16.0 2023-03-27 07:50:37,799 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-03-27 07:50:52,325 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6821, 3.7273, 3.5537, 1.7601, 3.8997, 2.8818, 0.7212, 2.6448], device='cuda:3'), covar=tensor([0.2234, 0.1787, 0.1398, 0.3133, 0.0863, 0.0961, 0.4635, 0.1461], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0180, 0.0161, 0.0130, 0.0162, 0.0124, 0.0150, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 07:50:55,309 INFO [finetune.py:976] (3/7) Epoch 26, batch 3850, loss[loss=0.1606, simple_loss=0.2408, pruned_loss=0.04021, over 4770.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2456, pruned_loss=0.0499, over 956375.29 frames. ], batch size: 51, lr: 2.95e-03, grad_scale: 16.0 2023-03-27 07:51:11,223 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6193, 1.4433, 2.1321, 3.3202, 2.1932, 2.4032, 0.9515, 2.7560], device='cuda:3'), covar=tensor([0.1682, 0.1475, 0.1256, 0.0577, 0.0851, 0.1399, 0.1898, 0.0477], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0116, 0.0133, 0.0165, 0.0102, 0.0135, 0.0125, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 07:51:13,932 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-03-27 07:51:17,259 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8604, 1.3179, 1.9419, 1.8764, 1.7033, 1.6399, 1.8423, 1.8550], device='cuda:3'), covar=tensor([0.3941, 0.3840, 0.3021, 0.3445, 0.4657, 0.3664, 0.4288, 0.2784], device='cuda:3'), in_proj_covar=tensor([0.0266, 0.0247, 0.0267, 0.0295, 0.0295, 0.0271, 0.0302, 0.0252], 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') 2023-03-27 07:51:21,146 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.070e+02 1.517e+02 1.854e+02 2.266e+02 5.483e+02, threshold=3.707e+02, percent-clipped=2.0 2023-03-27 07:51:25,821 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-27 07:51:37,020 INFO [finetune.py:976] (3/7) Epoch 26, batch 3900, loss[loss=0.1669, simple_loss=0.2365, pruned_loss=0.04869, over 4894.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2437, pruned_loss=0.04989, over 956855.91 frames. ], batch size: 32, lr: 2.95e-03, grad_scale: 16.0 2023-03-27 07:51:58,816 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6020, 2.3887, 1.9083, 0.9733, 2.0992, 2.0164, 1.8565, 2.1892], device='cuda:3'), covar=tensor([0.0838, 0.0756, 0.1480, 0.1998, 0.1279, 0.2167, 0.2219, 0.0906], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0191, 0.0199, 0.0181, 0.0209, 0.0210, 0.0223, 0.0195], 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') 2023-03-27 07:52:09,602 INFO [finetune.py:976] (3/7) Epoch 26, batch 3950, loss[loss=0.1494, simple_loss=0.2181, pruned_loss=0.04032, over 4876.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2402, pruned_loss=0.04882, over 956007.93 frames. ], batch size: 31, lr: 2.95e-03, grad_scale: 16.0 2023-03-27 07:52:27,910 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.014e+02 1.492e+02 1.682e+02 1.976e+02 2.814e+02, threshold=3.365e+02, percent-clipped=0.0 2023-03-27 07:52:42,788 INFO [finetune.py:976] (3/7) Epoch 26, batch 4000, loss[loss=0.1843, simple_loss=0.2567, pruned_loss=0.05595, over 4822.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2412, pruned_loss=0.04982, over 954730.16 frames. ], batch size: 33, lr: 2.95e-03, grad_scale: 16.0 2023-03-27 07:52:48,864 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147200.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:53:15,821 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147241.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:53:16,324 INFO [finetune.py:976] (3/7) Epoch 26, batch 4050, loss[loss=0.1568, simple_loss=0.2416, pruned_loss=0.03598, over 4743.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2441, pruned_loss=0.0506, over 952878.58 frames. ], batch size: 27, lr: 2.95e-03, grad_scale: 16.0 2023-03-27 07:53:21,622 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147247.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:53:48,586 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.158e+02 1.660e+02 1.920e+02 2.375e+02 4.575e+02, threshold=3.840e+02, percent-clipped=6.0 2023-03-27 07:53:58,884 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147282.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:54:08,318 INFO [finetune.py:976] (3/7) Epoch 26, batch 4100, loss[loss=0.2013, simple_loss=0.2793, pruned_loss=0.06167, over 4803.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2463, pruned_loss=0.05093, over 955064.79 frames. ], batch size: 39, lr: 2.95e-03, grad_scale: 16.0 2023-03-27 07:54:13,960 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=147295.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:54:18,785 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147302.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:54:24,648 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147310.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:54:27,683 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5132, 2.0906, 2.8090, 1.7292, 2.4266, 2.5638, 1.9834, 2.6537], device='cuda:3'), covar=tensor([0.1093, 0.2103, 0.1398, 0.2049, 0.0888, 0.1435, 0.2709, 0.0859], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0206, 0.0193, 0.0189, 0.0175, 0.0213, 0.0216, 0.0199], 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') 2023-03-27 07:54:37,225 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=147330.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:54:44,910 INFO [finetune.py:976] (3/7) Epoch 26, batch 4150, loss[loss=0.1708, simple_loss=0.2469, pruned_loss=0.04736, over 4221.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2468, pruned_loss=0.05087, over 953082.99 frames. ], batch size: 66, lr: 2.95e-03, grad_scale: 16.0 2023-03-27 07:55:03,461 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.008e+02 1.520e+02 1.873e+02 2.208e+02 5.004e+02, threshold=3.746e+02, percent-clipped=1.0 2023-03-27 07:55:05,306 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147371.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:55:18,277 INFO [finetune.py:976] (3/7) Epoch 26, batch 4200, loss[loss=0.1802, simple_loss=0.2554, pruned_loss=0.05255, over 4170.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2465, pruned_loss=0.05097, over 950418.72 frames. ], batch size: 65, lr: 2.95e-03, grad_scale: 16.0 2023-03-27 07:55:28,558 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147407.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:55:46,456 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147434.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:55:51,222 INFO [finetune.py:976] (3/7) Epoch 26, batch 4250, loss[loss=0.1949, simple_loss=0.2607, pruned_loss=0.06461, over 4865.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2441, pruned_loss=0.04984, over 950746.64 frames. ], batch size: 31, lr: 2.95e-03, grad_scale: 16.0 2023-03-27 07:56:01,801 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8460, 1.6023, 2.3144, 3.4986, 2.3236, 2.5340, 1.1153, 2.8299], device='cuda:3'), covar=tensor([0.1700, 0.1349, 0.1271, 0.0712, 0.0777, 0.1286, 0.1881, 0.0561], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0132, 0.0163, 0.0100, 0.0134, 0.0124, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 07:56:16,331 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147468.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:56:16,792 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.028e+02 1.473e+02 1.815e+02 2.255e+02 8.587e+02, threshold=3.630e+02, percent-clipped=2.0 2023-03-27 07:56:34,765 INFO [finetune.py:976] (3/7) Epoch 26, batch 4300, loss[loss=0.2019, simple_loss=0.2703, pruned_loss=0.06671, over 4868.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2411, pruned_loss=0.04934, over 950965.15 frames. ], batch size: 31, lr: 2.95e-03, grad_scale: 16.0 2023-03-27 07:56:36,732 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147495.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:56:40,201 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147500.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:56:45,436 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5272, 2.3342, 2.1655, 2.3866, 2.2181, 2.2645, 2.2655, 2.9321], device='cuda:3'), covar=tensor([0.3393, 0.4003, 0.3019, 0.3568, 0.3748, 0.2473, 0.3760, 0.1707], device='cuda:3'), in_proj_covar=tensor([0.0291, 0.0267, 0.0239, 0.0279, 0.0262, 0.0232, 0.0260, 0.0241], 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') 2023-03-27 07:57:08,538 INFO [finetune.py:976] (3/7) Epoch 26, batch 4350, loss[loss=0.1676, simple_loss=0.2286, pruned_loss=0.05326, over 4925.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2376, pruned_loss=0.04806, over 951547.52 frames. ], batch size: 33, lr: 2.95e-03, grad_scale: 16.0 2023-03-27 07:57:12,265 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=147548.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:57:26,972 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.870e+01 1.462e+02 1.667e+02 1.917e+02 5.708e+02, threshold=3.333e+02, percent-clipped=2.0 2023-03-27 07:57:27,067 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.6591, 3.1875, 3.3683, 3.4744, 3.4367, 3.2511, 3.7017, 1.5857], device='cuda:3'), covar=tensor([0.0851, 0.0933, 0.0843, 0.1010, 0.1306, 0.1550, 0.0939, 0.5284], device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0251, 0.0285, 0.0299, 0.0341, 0.0290, 0.0308, 0.0304], 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') 2023-03-27 07:57:42,396 INFO [finetune.py:976] (3/7) Epoch 26, batch 4400, loss[loss=0.175, simple_loss=0.2586, pruned_loss=0.04574, over 4834.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2394, pruned_loss=0.04873, over 950730.73 frames. ], batch size: 47, lr: 2.95e-03, grad_scale: 16.0 2023-03-27 07:57:45,507 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147597.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:57:56,493 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0 2023-03-27 07:58:08,078 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-27 07:58:16,297 INFO [finetune.py:976] (3/7) Epoch 26, batch 4450, loss[loss=0.143, simple_loss=0.2124, pruned_loss=0.0368, over 4719.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2436, pruned_loss=0.04991, over 951111.52 frames. ], batch size: 23, lr: 2.95e-03, grad_scale: 16.0 2023-03-27 07:58:27,312 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147659.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:58:34,814 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147666.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:58:36,564 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.042e+02 1.542e+02 1.864e+02 2.192e+02 3.736e+02, threshold=3.727e+02, percent-clipped=4.0 2023-03-27 07:59:06,170 INFO [finetune.py:976] (3/7) Epoch 26, batch 4500, loss[loss=0.1867, simple_loss=0.2556, pruned_loss=0.05894, over 4804.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2451, pruned_loss=0.05023, over 953198.25 frames. ], batch size: 45, lr: 2.95e-03, grad_scale: 16.0 2023-03-27 07:59:37,016 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147720.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 07:59:52,138 INFO [finetune.py:976] (3/7) Epoch 26, batch 4550, loss[loss=0.1382, simple_loss=0.2315, pruned_loss=0.02243, over 4769.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2471, pruned_loss=0.05107, over 953183.46 frames. ], batch size: 28, lr: 2.95e-03, grad_scale: 32.0 2023-03-27 08:00:04,825 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147763.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:00:09,336 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.722e+01 1.488e+02 1.756e+02 2.293e+02 4.562e+02, threshold=3.512e+02, percent-clipped=2.0 2023-03-27 08:00:24,540 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147790.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:00:25,691 INFO [finetune.py:976] (3/7) Epoch 26, batch 4600, loss[loss=0.197, simple_loss=0.2684, pruned_loss=0.06281, over 4917.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2462, pruned_loss=0.05096, over 951428.76 frames. ], batch size: 38, lr: 2.95e-03, grad_scale: 32.0 2023-03-27 08:00:29,984 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2547, 2.8621, 2.7367, 1.1644, 3.0001, 2.2132, 0.8265, 1.8872], device='cuda:3'), covar=tensor([0.2524, 0.2399, 0.1940, 0.3483, 0.1574, 0.1210, 0.3849, 0.1658], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0180, 0.0161, 0.0130, 0.0162, 0.0125, 0.0149, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 08:00:39,898 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147815.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:00:59,127 INFO [finetune.py:976] (3/7) Epoch 26, batch 4650, loss[loss=0.1544, simple_loss=0.2271, pruned_loss=0.04084, over 4790.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2435, pruned_loss=0.05052, over 949687.81 frames. ], batch size: 25, lr: 2.95e-03, grad_scale: 32.0 2023-03-27 08:01:08,834 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147857.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:01:15,987 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 7.955e+01 1.458e+02 1.712e+02 2.175e+02 4.467e+02, threshold=3.424e+02, percent-clipped=3.0 2023-03-27 08:01:21,850 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147876.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:01:39,629 INFO [finetune.py:976] (3/7) Epoch 26, batch 4700, loss[loss=0.1408, simple_loss=0.2186, pruned_loss=0.03153, over 4765.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2412, pruned_loss=0.04953, over 952282.36 frames. ], batch size: 28, lr: 2.95e-03, grad_scale: 32.0 2023-03-27 08:01:46,504 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147897.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:01:55,416 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147911.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:01:59,699 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147918.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:02:16,230 INFO [finetune.py:976] (3/7) Epoch 26, batch 4750, loss[loss=0.1724, simple_loss=0.2359, pruned_loss=0.05444, over 4752.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2394, pruned_loss=0.04896, over 952231.51 frames. ], batch size: 26, lr: 2.95e-03, grad_scale: 32.0 2023-03-27 08:02:18,626 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=147945.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:02:32,375 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147966.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:02:34,090 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.372e+01 1.470e+02 1.654e+02 2.049e+02 2.990e+02, threshold=3.309e+02, percent-clipped=0.0 2023-03-27 08:02:35,996 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147972.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:02:50,080 INFO [finetune.py:976] (3/7) Epoch 26, batch 4800, loss[loss=0.1423, simple_loss=0.2282, pruned_loss=0.02826, over 4877.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2439, pruned_loss=0.05062, over 952573.38 frames. ], batch size: 34, lr: 2.95e-03, grad_scale: 32.0 2023-03-27 08:03:06,226 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=148014.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:03:06,844 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148015.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:03:24,538 INFO [finetune.py:976] (3/7) Epoch 26, batch 4850, loss[loss=0.1816, simple_loss=0.265, pruned_loss=0.0491, over 4825.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2458, pruned_loss=0.05088, over 953571.35 frames. ], batch size: 47, lr: 2.95e-03, grad_scale: 32.0 2023-03-27 08:03:39,224 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148063.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:03:42,783 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.092e+02 1.530e+02 1.908e+02 2.333e+02 3.886e+02, threshold=3.817e+02, percent-clipped=4.0 2023-03-27 08:04:04,118 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148090.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:04:05,251 INFO [finetune.py:976] (3/7) Epoch 26, batch 4900, loss[loss=0.2129, simple_loss=0.2795, pruned_loss=0.07311, over 4879.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.247, pruned_loss=0.051, over 953816.72 frames. ], batch size: 43, lr: 2.95e-03, grad_scale: 32.0 2023-03-27 08:04:30,815 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=148111.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:04:35,812 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7962, 1.6543, 1.8199, 1.2087, 1.7952, 1.8565, 1.7702, 1.5161], device='cuda:3'), covar=tensor([0.0556, 0.0735, 0.0721, 0.0892, 0.0834, 0.0716, 0.0692, 0.1182], device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0137, 0.0142, 0.0120, 0.0128, 0.0139, 0.0141, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 08:04:57,743 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=148138.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:05:00,560 INFO [finetune.py:976] (3/7) Epoch 26, batch 4950, loss[loss=0.1321, simple_loss=0.22, pruned_loss=0.02211, over 4752.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2477, pruned_loss=0.05116, over 953909.00 frames. ], batch size: 28, lr: 2.95e-03, grad_scale: 32.0 2023-03-27 08:05:18,897 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.030e+01 1.572e+02 1.871e+02 2.257e+02 5.603e+02, threshold=3.742e+02, percent-clipped=1.0 2023-03-27 08:05:20,235 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148171.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:05:27,036 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5309, 2.3881, 1.9113, 0.9504, 2.1007, 1.9719, 1.8495, 2.1641], device='cuda:3'), covar=tensor([0.0879, 0.0785, 0.1655, 0.2060, 0.1364, 0.2173, 0.2031, 0.0956], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0192, 0.0200, 0.0182, 0.0209, 0.0211, 0.0224, 0.0196], 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') 2023-03-27 08:05:33,993 INFO [finetune.py:976] (3/7) Epoch 26, batch 5000, loss[loss=0.1603, simple_loss=0.2285, pruned_loss=0.046, over 4815.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2455, pruned_loss=0.05031, over 954238.04 frames. ], batch size: 33, lr: 2.95e-03, grad_scale: 32.0 2023-03-27 08:05:40,386 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-27 08:05:47,620 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6817, 3.6986, 3.4674, 1.9225, 3.7503, 2.6883, 0.8407, 2.5627], device='cuda:3'), covar=tensor([0.2537, 0.1879, 0.1779, 0.3045, 0.1123, 0.1176, 0.4401, 0.1657], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0180, 0.0162, 0.0130, 0.0162, 0.0125, 0.0149, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 08:05:48,838 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148213.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:06:07,383 INFO [finetune.py:976] (3/7) Epoch 26, batch 5050, loss[loss=0.1743, simple_loss=0.2436, pruned_loss=0.05253, over 4815.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2419, pruned_loss=0.04881, over 953011.38 frames. ], batch size: 41, lr: 2.95e-03, grad_scale: 32.0 2023-03-27 08:06:19,280 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6989, 1.5483, 1.5196, 1.5768, 0.9712, 3.2884, 1.2386, 1.6243], device='cuda:3'), covar=tensor([0.3175, 0.2437, 0.2003, 0.2272, 0.1877, 0.0206, 0.2522, 0.1233], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0121, 0.0124, 0.0113, 0.0095, 0.0094, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 08:06:25,059 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148267.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:06:26,160 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.004e+02 1.490e+02 1.796e+02 2.082e+02 4.496e+02, threshold=3.592e+02, percent-clipped=3.0 2023-03-27 08:06:37,566 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148287.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:06:40,479 INFO [finetune.py:976] (3/7) Epoch 26, batch 5100, loss[loss=0.1749, simple_loss=0.2319, pruned_loss=0.05889, over 4818.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2394, pruned_loss=0.04809, over 954474.91 frames. ], batch size: 38, lr: 2.95e-03, grad_scale: 32.0 2023-03-27 08:06:54,244 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.4498, 1.4619, 1.5264, 0.7855, 1.6299, 1.8330, 1.7735, 1.4056], device='cuda:3'), covar=tensor([0.1042, 0.0835, 0.0637, 0.0601, 0.0581, 0.0584, 0.0401, 0.0875], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0148, 0.0128, 0.0122, 0.0131, 0.0129, 0.0142, 0.0149], device='cuda:3'), out_proj_covar=tensor([8.8618e-05, 1.0669e-04, 9.1132e-05, 8.5811e-05, 9.1448e-05, 9.1601e-05, 1.0067e-04, 1.0681e-04], device='cuda:3') 2023-03-27 08:07:02,510 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148315.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:07:22,780 INFO [finetune.py:976] (3/7) Epoch 26, batch 5150, loss[loss=0.1768, simple_loss=0.2555, pruned_loss=0.04903, over 4745.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2396, pruned_loss=0.04838, over 955900.69 frames. ], batch size: 54, lr: 2.95e-03, grad_scale: 32.0 2023-03-27 08:07:25,436 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.28 vs. limit=5.0 2023-03-27 08:07:27,032 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148348.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:07:36,971 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=148363.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:07:41,442 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.157e+02 1.549e+02 1.828e+02 2.233e+02 5.689e+02, threshold=3.657e+02, percent-clipped=4.0 2023-03-27 08:07:55,902 INFO [finetune.py:976] (3/7) Epoch 26, batch 5200, loss[loss=0.1953, simple_loss=0.2646, pruned_loss=0.06296, over 4931.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2445, pruned_loss=0.04986, over 956789.21 frames. ], batch size: 36, lr: 2.95e-03, grad_scale: 16.0 2023-03-27 08:08:11,027 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.44 vs. limit=5.0 2023-03-27 08:08:21,971 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148429.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:08:29,695 INFO [finetune.py:976] (3/7) Epoch 26, batch 5250, loss[loss=0.208, simple_loss=0.2822, pruned_loss=0.06688, over 4823.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.245, pruned_loss=0.04985, over 956979.77 frames. ], batch size: 39, lr: 2.94e-03, grad_scale: 16.0 2023-03-27 08:08:32,275 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1974, 2.1716, 1.8350, 2.1942, 2.0980, 2.0710, 2.0790, 2.9570], device='cuda:3'), covar=tensor([0.3692, 0.4737, 0.3496, 0.4323, 0.4315, 0.2564, 0.4251, 0.1637], device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0265, 0.0237, 0.0277, 0.0260, 0.0230, 0.0258, 0.0240], 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') 2023-03-27 08:08:44,689 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7990, 1.7737, 1.5085, 2.0126, 2.3377, 1.9632, 1.8674, 1.4617], device='cuda:3'), covar=tensor([0.2061, 0.1856, 0.1777, 0.1498, 0.1487, 0.1088, 0.1966, 0.1793], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0210, 0.0214, 0.0198, 0.0245, 0.0191, 0.0217, 0.0204], 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') 2023-03-27 08:08:48,649 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.102e+02 1.514e+02 1.742e+02 2.193e+02 4.299e+02, threshold=3.484e+02, percent-clipped=1.0 2023-03-27 08:08:49,343 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148471.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:09:02,320 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148490.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:09:02,657 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-27 08:09:03,426 INFO [finetune.py:976] (3/7) Epoch 26, batch 5300, loss[loss=0.1839, simple_loss=0.2577, pruned_loss=0.05507, over 4808.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2459, pruned_loss=0.04999, over 954383.06 frames. ], batch size: 40, lr: 2.94e-03, grad_scale: 16.0 2023-03-27 08:09:07,101 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148498.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:09:17,175 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-27 08:09:20,021 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148513.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:09:28,368 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=148519.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:09:46,973 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-27 08:09:54,004 INFO [finetune.py:976] (3/7) Epoch 26, batch 5350, loss[loss=0.1955, simple_loss=0.2467, pruned_loss=0.0721, over 4389.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2468, pruned_loss=0.05016, over 954791.33 frames. ], batch size: 19, lr: 2.94e-03, grad_scale: 16.0 2023-03-27 08:10:12,425 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148559.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:10:13,594 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=148561.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:10:17,754 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148567.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:10:19,465 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.029e+02 1.412e+02 1.653e+02 1.941e+02 3.220e+02, threshold=3.306e+02, percent-clipped=0.0 2023-03-27 08:10:34,702 INFO [finetune.py:976] (3/7) Epoch 26, batch 5400, loss[loss=0.1239, simple_loss=0.1969, pruned_loss=0.02545, over 4819.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2438, pruned_loss=0.0498, over 953696.44 frames. ], batch size: 25, lr: 2.94e-03, grad_scale: 16.0 2023-03-27 08:10:49,746 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=148615.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:11:07,907 INFO [finetune.py:976] (3/7) Epoch 26, batch 5450, loss[loss=0.149, simple_loss=0.2196, pruned_loss=0.0392, over 4829.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2409, pruned_loss=0.049, over 953837.30 frames. ], batch size: 25, lr: 2.94e-03, grad_scale: 16.0 2023-03-27 08:11:08,561 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148643.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:11:13,475 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148651.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:11:25,817 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.053e+02 1.571e+02 1.855e+02 2.174e+02 4.125e+02, threshold=3.711e+02, percent-clipped=2.0 2023-03-27 08:11:41,100 INFO [finetune.py:976] (3/7) Epoch 26, batch 5500, loss[loss=0.148, simple_loss=0.2174, pruned_loss=0.03931, over 4158.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2374, pruned_loss=0.04789, over 953315.40 frames. ], batch size: 18, lr: 2.94e-03, grad_scale: 16.0 2023-03-27 08:11:44,453 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.27 vs. limit=5.0 2023-03-27 08:11:53,975 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148712.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:11:56,407 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148716.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:12:09,356 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1392, 2.2482, 1.7747, 1.8905, 2.7171, 2.7480, 2.1681, 2.1116], device='cuda:3'), covar=tensor([0.0420, 0.0354, 0.0624, 0.0393, 0.0203, 0.0421, 0.0356, 0.0404], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0106, 0.0147, 0.0111, 0.0101, 0.0115, 0.0103, 0.0113], device='cuda:3'), out_proj_covar=tensor([7.8329e-05, 8.0843e-05, 1.1429e-04, 8.4759e-05, 7.8133e-05, 8.4839e-05, 7.6635e-05, 8.5330e-05], device='cuda:3') 2023-03-27 08:12:24,636 INFO [finetune.py:976] (3/7) Epoch 26, batch 5550, loss[loss=0.2109, simple_loss=0.2853, pruned_loss=0.0683, over 4850.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.239, pruned_loss=0.04891, over 952410.12 frames. ], batch size: 47, lr: 2.94e-03, grad_scale: 16.0 2023-03-27 08:12:31,218 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-27 08:12:42,333 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.664e+01 1.578e+02 1.917e+02 2.288e+02 4.413e+02, threshold=3.834e+02, percent-clipped=2.0 2023-03-27 08:12:47,504 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148777.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:12:52,458 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148785.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:12:55,444 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4082, 1.1646, 1.0744, 1.0978, 1.5731, 1.5649, 1.3136, 1.1551], device='cuda:3'), covar=tensor([0.0331, 0.0425, 0.0894, 0.0412, 0.0292, 0.0455, 0.0369, 0.0495], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0106, 0.0147, 0.0111, 0.0101, 0.0115, 0.0103, 0.0112], device='cuda:3'), out_proj_covar=tensor([7.8259e-05, 8.0854e-05, 1.1413e-04, 8.4750e-05, 7.8125e-05, 8.4802e-05, 7.6536e-05, 8.5294e-05], device='cuda:3') 2023-03-27 08:12:56,501 INFO [finetune.py:976] (3/7) Epoch 26, batch 5600, loss[loss=0.1769, simple_loss=0.2482, pruned_loss=0.05278, over 4816.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2427, pruned_loss=0.04944, over 950538.15 frames. ], batch size: 33, lr: 2.94e-03, grad_scale: 16.0 2023-03-27 08:13:10,059 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4894, 1.3994, 1.3473, 1.4267, 1.2067, 3.5024, 1.4055, 1.7687], device='cuda:3'), covar=tensor([0.4281, 0.3380, 0.2592, 0.3028, 0.1876, 0.0244, 0.2791, 0.1296], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0116, 0.0121, 0.0124, 0.0113, 0.0095, 0.0094, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 08:13:25,694 INFO [finetune.py:976] (3/7) Epoch 26, batch 5650, loss[loss=0.1495, simple_loss=0.2268, pruned_loss=0.03609, over 4815.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2447, pruned_loss=0.04934, over 949598.86 frames. ], batch size: 25, lr: 2.94e-03, grad_scale: 16.0 2023-03-27 08:13:32,822 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148854.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:13:42,328 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.032e+02 1.500e+02 1.770e+02 2.150e+02 4.859e+02, threshold=3.539e+02, percent-clipped=1.0 2023-03-27 08:13:51,286 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3855, 2.2612, 2.4128, 1.7122, 2.2833, 2.4654, 2.4183, 1.9520], device='cuda:3'), covar=tensor([0.0488, 0.0557, 0.0590, 0.0814, 0.0894, 0.0591, 0.0614, 0.1039], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0136, 0.0141, 0.0119, 0.0128, 0.0138, 0.0139, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 08:13:52,320 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-03-27 08:13:55,308 INFO [finetune.py:976] (3/7) Epoch 26, batch 5700, loss[loss=0.1913, simple_loss=0.257, pruned_loss=0.06273, over 4035.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2397, pruned_loss=0.04876, over 931043.37 frames. ], batch size: 17, lr: 2.94e-03, grad_scale: 8.0 2023-03-27 08:14:05,589 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-27 08:14:24,188 INFO [finetune.py:976] (3/7) Epoch 27, batch 0, loss[loss=0.1768, simple_loss=0.2503, pruned_loss=0.05161, over 4855.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2503, pruned_loss=0.05161, over 4855.00 frames. ], batch size: 31, lr: 2.94e-03, grad_scale: 8.0 2023-03-27 08:14:24,188 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-27 08:14:29,674 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6515, 1.2176, 0.9293, 1.6543, 2.1316, 1.1703, 1.5463, 1.5327], device='cuda:3'), covar=tensor([0.1453, 0.1960, 0.1772, 0.1134, 0.1829, 0.2003, 0.1301, 0.1975], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0094, 0.0110, 0.0092, 0.0120, 0.0094, 0.0099, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 08:14:30,166 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5006, 1.4349, 1.3472, 1.4648, 1.7081, 1.7414, 1.4703, 1.3132], device='cuda:3'), covar=tensor([0.0412, 0.0344, 0.0662, 0.0343, 0.0266, 0.0357, 0.0386, 0.0462], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0105, 0.0146, 0.0110, 0.0100, 0.0114, 0.0103, 0.0112], device='cuda:3'), out_proj_covar=tensor([7.7685e-05, 8.0337e-05, 1.1349e-04, 8.4398e-05, 7.7821e-05, 8.4339e-05, 7.6199e-05, 8.4954e-05], device='cuda:3') 2023-03-27 08:14:40,695 INFO [finetune.py:1010] (3/7) Epoch 27, validation: loss=0.1593, simple_loss=0.2269, pruned_loss=0.04586, over 2265189.00 frames. 2023-03-27 08:14:40,696 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6469MB 2023-03-27 08:14:57,038 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148943.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:15:20,327 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-27 08:15:27,448 INFO [finetune.py:976] (3/7) Epoch 27, batch 50, loss[loss=0.171, simple_loss=0.2396, pruned_loss=0.05115, over 4783.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.246, pruned_loss=0.05031, over 216999.82 frames. ], batch size: 29, lr: 2.94e-03, grad_scale: 8.0 2023-03-27 08:15:28,074 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.253e+01 1.427e+02 1.731e+02 2.058e+02 3.661e+02, threshold=3.462e+02, percent-clipped=4.0 2023-03-27 08:15:35,636 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1523, 3.6293, 3.8348, 4.0137, 3.8979, 3.6488, 4.2279, 1.2677], device='cuda:3'), covar=tensor([0.0831, 0.0998, 0.0910, 0.0969, 0.1275, 0.1750, 0.0791, 0.5913], device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0250, 0.0285, 0.0298, 0.0338, 0.0289, 0.0307, 0.0302], 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') 2023-03-27 08:15:44,091 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=148991.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:15:54,006 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149007.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:16:03,651 INFO [finetune.py:976] (3/7) Epoch 27, batch 100, loss[loss=0.1703, simple_loss=0.2308, pruned_loss=0.05489, over 4831.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2437, pruned_loss=0.051, over 382795.89 frames. ], batch size: 38, lr: 2.94e-03, grad_scale: 8.0 2023-03-27 08:16:13,173 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6284, 2.7159, 2.5895, 1.9157, 2.4818, 2.8778, 2.7965, 2.2203], device='cuda:3'), covar=tensor([0.0583, 0.0540, 0.0666, 0.0853, 0.0876, 0.0622, 0.0621, 0.1059], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0136, 0.0141, 0.0119, 0.0128, 0.0139, 0.0140, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 08:16:28,911 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.5755, 3.9502, 4.1961, 4.3541, 4.3405, 4.1677, 4.6793, 1.3792], device='cuda:3'), covar=tensor([0.0787, 0.0888, 0.0866, 0.0957, 0.1182, 0.1483, 0.0658, 0.6092], device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0250, 0.0284, 0.0297, 0.0337, 0.0289, 0.0306, 0.0301], 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') 2023-03-27 08:16:36,428 INFO [finetune.py:976] (3/7) Epoch 27, batch 150, loss[loss=0.1406, simple_loss=0.2154, pruned_loss=0.03292, over 4821.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2386, pruned_loss=0.04925, over 509845.14 frames. ], batch size: 39, lr: 2.94e-03, grad_scale: 8.0 2023-03-27 08:16:37,489 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.570e+01 1.451e+02 1.770e+02 2.054e+02 3.397e+02, threshold=3.539e+02, percent-clipped=0.0 2023-03-27 08:16:39,138 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149072.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:16:47,475 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149085.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:16:52,676 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-27 08:17:09,497 INFO [finetune.py:976] (3/7) Epoch 27, batch 200, loss[loss=0.1807, simple_loss=0.2525, pruned_loss=0.05443, over 4916.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2357, pruned_loss=0.04764, over 610264.92 frames. ], batch size: 37, lr: 2.94e-03, grad_scale: 8.0 2023-03-27 08:17:09,737 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-27 08:17:19,407 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=149133.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:17:39,183 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149154.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:17:52,877 INFO [finetune.py:976] (3/7) Epoch 27, batch 250, loss[loss=0.1452, simple_loss=0.2086, pruned_loss=0.04091, over 4523.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2396, pruned_loss=0.04936, over 686674.23 frames. ], batch size: 19, lr: 2.94e-03, grad_scale: 8.0 2023-03-27 08:17:53,480 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.507e+01 1.547e+02 1.763e+02 2.073e+02 3.560e+02, threshold=3.526e+02, percent-clipped=1.0 2023-03-27 08:18:14,779 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=149202.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:18:25,587 INFO [finetune.py:976] (3/7) Epoch 27, batch 300, loss[loss=0.151, simple_loss=0.2372, pruned_loss=0.03245, over 4902.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2457, pruned_loss=0.05113, over 748449.83 frames. ], batch size: 32, lr: 2.94e-03, grad_scale: 8.0 2023-03-27 08:18:55,222 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1541, 3.6042, 3.7996, 4.0147, 3.8957, 3.5884, 4.2180, 1.1849], device='cuda:3'), covar=tensor([0.0799, 0.0905, 0.0941, 0.0913, 0.1321, 0.1888, 0.0807, 0.5924], device='cuda:3'), in_proj_covar=tensor([0.0350, 0.0248, 0.0284, 0.0296, 0.0336, 0.0288, 0.0305, 0.0299], 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') 2023-03-27 08:18:58,794 INFO [finetune.py:976] (3/7) Epoch 27, batch 350, loss[loss=0.1865, simple_loss=0.2608, pruned_loss=0.05614, over 4749.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2476, pruned_loss=0.05177, over 794236.25 frames. ], batch size: 27, lr: 2.94e-03, grad_scale: 8.0 2023-03-27 08:18:59,397 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.129e+02 1.601e+02 1.876e+02 2.140e+02 5.128e+02, threshold=3.753e+02, percent-clipped=1.0 2023-03-27 08:19:14,644 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2097, 2.4265, 1.8644, 2.4620, 2.3473, 2.1568, 2.2991, 3.0522], device='cuda:3'), covar=tensor([0.3504, 0.4130, 0.3318, 0.3834, 0.3904, 0.2489, 0.3716, 0.1636], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0263, 0.0235, 0.0275, 0.0259, 0.0229, 0.0257, 0.0238], 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') 2023-03-27 08:19:24,861 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149307.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:19:32,720 INFO [finetune.py:976] (3/7) Epoch 27, batch 400, loss[loss=0.1588, simple_loss=0.2374, pruned_loss=0.0401, over 4762.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2481, pruned_loss=0.05193, over 829594.38 frames. ], batch size: 26, lr: 2.94e-03, grad_scale: 8.0 2023-03-27 08:19:34,676 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6228, 1.6133, 1.4077, 1.5059, 1.9557, 1.8858, 1.5757, 1.4643], device='cuda:3'), covar=tensor([0.0345, 0.0368, 0.0709, 0.0383, 0.0236, 0.0462, 0.0412, 0.0445], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0106, 0.0147, 0.0111, 0.0101, 0.0115, 0.0104, 0.0112], device='cuda:3'), out_proj_covar=tensor([7.7942e-05, 8.1004e-05, 1.1428e-04, 8.5041e-05, 7.8375e-05, 8.4871e-05, 7.7135e-05, 8.5210e-05], device='cuda:3') 2023-03-27 08:19:35,595 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.81 vs. limit=5.0 2023-03-27 08:20:07,342 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=149355.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:20:10,500 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2905, 2.3046, 1.9334, 2.3523, 2.2185, 2.1815, 2.1623, 3.1109], device='cuda:3'), covar=tensor([0.3720, 0.4460, 0.3374, 0.4136, 0.4234, 0.2508, 0.4271, 0.1580], device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0263, 0.0235, 0.0275, 0.0259, 0.0229, 0.0258, 0.0238], 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') 2023-03-27 08:20:16,473 INFO [finetune.py:976] (3/7) Epoch 27, batch 450, loss[loss=0.1396, simple_loss=0.2048, pruned_loss=0.03723, over 4822.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2455, pruned_loss=0.05063, over 857944.85 frames. ], batch size: 25, lr: 2.94e-03, grad_scale: 8.0 2023-03-27 08:20:17,064 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.956e+01 1.496e+02 1.736e+02 2.126e+02 4.914e+02, threshold=3.471e+02, percent-clipped=1.0 2023-03-27 08:20:17,776 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149372.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:20:57,050 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-03-27 08:21:04,694 INFO [finetune.py:976] (3/7) Epoch 27, batch 500, loss[loss=0.1728, simple_loss=0.2389, pruned_loss=0.05339, over 4822.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2416, pruned_loss=0.04895, over 878594.15 frames. ], batch size: 41, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:21:04,765 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=149420.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:21:38,464 INFO [finetune.py:976] (3/7) Epoch 27, batch 550, loss[loss=0.176, simple_loss=0.2427, pruned_loss=0.05462, over 4774.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2392, pruned_loss=0.04817, over 895689.23 frames. ], batch size: 26, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:21:39,060 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.918e+01 1.466e+02 1.717e+02 2.125e+02 3.295e+02, threshold=3.435e+02, percent-clipped=0.0 2023-03-27 08:22:08,680 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149514.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:22:12,134 INFO [finetune.py:976] (3/7) Epoch 27, batch 600, loss[loss=0.1342, simple_loss=0.2052, pruned_loss=0.03159, over 4768.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2401, pruned_loss=0.04856, over 908331.93 frames. ], batch size: 27, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:22:45,156 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149565.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:22:48,083 INFO [finetune.py:976] (3/7) Epoch 27, batch 650, loss[loss=0.2148, simple_loss=0.2762, pruned_loss=0.07668, over 4904.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2433, pruned_loss=0.04952, over 918562.92 frames. ], batch size: 37, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:22:53,183 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.097e+02 1.592e+02 1.975e+02 2.434e+02 4.045e+02, threshold=3.949e+02, percent-clipped=4.0 2023-03-27 08:22:55,791 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149575.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:23:20,430 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.40 vs. limit=5.0 2023-03-27 08:23:29,821 INFO [finetune.py:976] (3/7) Epoch 27, batch 700, loss[loss=0.1365, simple_loss=0.2184, pruned_loss=0.0273, over 4853.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2435, pruned_loss=0.04823, over 927748.19 frames. ], batch size: 25, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:23:33,654 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149626.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 08:23:33,827 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-27 08:24:03,080 INFO [finetune.py:976] (3/7) Epoch 27, batch 750, loss[loss=0.1451, simple_loss=0.2172, pruned_loss=0.03657, over 4737.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2432, pruned_loss=0.04802, over 932511.52 frames. ], batch size: 23, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:24:03,698 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.028e+02 1.520e+02 1.783e+02 2.094e+02 3.998e+02, threshold=3.567e+02, percent-clipped=1.0 2023-03-27 08:24:15,992 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0 2023-03-27 08:24:23,403 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0878, 2.0578, 1.7094, 2.1607, 2.1822, 1.9220, 2.5154, 2.1707], device='cuda:3'), covar=tensor([0.1378, 0.2045, 0.2828, 0.2324, 0.2261, 0.1599, 0.2556, 0.1619], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0190, 0.0238, 0.0255, 0.0250, 0.0208, 0.0215, 0.0203], 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') 2023-03-27 08:24:35,238 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1300, 2.0320, 1.7393, 1.9782, 1.9705, 1.9483, 2.0107, 2.6726], device='cuda:3'), covar=tensor([0.3975, 0.4379, 0.3435, 0.4021, 0.3912, 0.2598, 0.3829, 0.1841], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0264, 0.0236, 0.0277, 0.0260, 0.0230, 0.0259, 0.0240], 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') 2023-03-27 08:24:36,878 INFO [finetune.py:976] (3/7) Epoch 27, batch 800, loss[loss=0.1962, simple_loss=0.2602, pruned_loss=0.06616, over 4810.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2443, pruned_loss=0.04907, over 937244.67 frames. ], batch size: 39, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:24:53,195 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149746.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 08:24:57,215 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7865, 1.5818, 1.4653, 1.7605, 2.2760, 1.8411, 1.7911, 1.4801], device='cuda:3'), covar=tensor([0.2060, 0.2061, 0.1981, 0.1664, 0.1675, 0.1400, 0.2273, 0.1905], device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0210, 0.0213, 0.0197, 0.0244, 0.0191, 0.0216, 0.0204], 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') 2023-03-27 08:25:09,835 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-27 08:25:20,609 INFO [finetune.py:976] (3/7) Epoch 27, batch 850, loss[loss=0.1487, simple_loss=0.2277, pruned_loss=0.0348, over 4819.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2431, pruned_loss=0.04931, over 940992.59 frames. ], batch size: 38, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:25:21,209 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.021e+02 1.421e+02 1.714e+02 1.950e+02 4.580e+02, threshold=3.429e+02, percent-clipped=2.0 2023-03-27 08:25:26,437 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 2023-03-27 08:25:56,751 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149807.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 08:26:09,034 INFO [finetune.py:976] (3/7) Epoch 27, batch 900, loss[loss=0.1495, simple_loss=0.2293, pruned_loss=0.03483, over 4789.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2402, pruned_loss=0.04821, over 944503.51 frames. ], batch size: 29, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:26:17,054 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4368, 1.2916, 1.3023, 1.3230, 0.8066, 2.2862, 0.7261, 1.2432], device='cuda:3'), covar=tensor([0.3188, 0.2572, 0.2216, 0.2435, 0.2012, 0.0340, 0.2752, 0.1291], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0116, 0.0120, 0.0124, 0.0113, 0.0095, 0.0094, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 08:26:42,227 INFO [finetune.py:976] (3/7) Epoch 27, batch 950, loss[loss=0.2573, simple_loss=0.308, pruned_loss=0.1033, over 4913.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2387, pruned_loss=0.04816, over 947098.59 frames. ], batch size: 37, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:26:42,297 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149870.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:26:42,815 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.619e+01 1.503e+02 1.866e+02 2.296e+02 3.689e+02, threshold=3.732e+02, percent-clipped=3.0 2023-03-27 08:27:15,533 INFO [finetune.py:976] (3/7) Epoch 27, batch 1000, loss[loss=0.1566, simple_loss=0.213, pruned_loss=0.05016, over 4259.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2415, pruned_loss=0.04917, over 950115.17 frames. ], batch size: 18, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:27:16,176 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149921.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 08:27:48,829 INFO [finetune.py:976] (3/7) Epoch 27, batch 1050, loss[loss=0.2324, simple_loss=0.2904, pruned_loss=0.08717, over 4866.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2428, pruned_loss=0.04884, over 952263.24 frames. ], batch size: 44, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:27:49,413 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.130e+02 1.561e+02 1.767e+02 2.240e+02 3.870e+02, threshold=3.534e+02, percent-clipped=1.0 2023-03-27 08:27:55,516 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5261, 1.3976, 1.9044, 2.9495, 2.0367, 2.1322, 0.8314, 2.5421], device='cuda:3'), covar=tensor([0.1723, 0.1336, 0.1281, 0.0674, 0.0800, 0.1471, 0.1849, 0.0500], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0132, 0.0163, 0.0100, 0.0136, 0.0124, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 08:28:33,249 INFO [finetune.py:976] (3/7) Epoch 27, batch 1100, loss[loss=0.2143, simple_loss=0.2779, pruned_loss=0.07535, over 4803.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2437, pruned_loss=0.04857, over 955408.95 frames. ], batch size: 40, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:28:49,558 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6944, 2.5460, 2.2635, 1.1314, 2.3617, 2.0074, 1.8808, 2.2587], device='cuda:3'), covar=tensor([0.0918, 0.0848, 0.1574, 0.2091, 0.1518, 0.2754, 0.2299, 0.1096], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0192, 0.0201, 0.0182, 0.0209, 0.0211, 0.0223, 0.0196], 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') 2023-03-27 08:29:06,477 INFO [finetune.py:976] (3/7) Epoch 27, batch 1150, loss[loss=0.1884, simple_loss=0.2595, pruned_loss=0.05869, over 4736.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2449, pruned_loss=0.04907, over 955891.24 frames. ], batch size: 54, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:29:07,079 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.913e+01 1.470e+02 1.766e+02 2.217e+02 3.439e+02, threshold=3.531e+02, percent-clipped=0.0 2023-03-27 08:29:13,022 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150079.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:29:13,636 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6853, 1.5315, 1.0871, 0.2991, 1.2915, 1.4964, 1.5252, 1.4519], device='cuda:3'), covar=tensor([0.0961, 0.0861, 0.1395, 0.2008, 0.1359, 0.2516, 0.2298, 0.0907], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0192, 0.0201, 0.0182, 0.0208, 0.0211, 0.0224, 0.0196], 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') 2023-03-27 08:29:26,983 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150102.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 08:29:39,271 INFO [finetune.py:976] (3/7) Epoch 27, batch 1200, loss[loss=0.166, simple_loss=0.2295, pruned_loss=0.05125, over 4709.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2437, pruned_loss=0.04901, over 955833.27 frames. ], batch size: 23, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:29:52,950 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150140.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:30:14,457 INFO [finetune.py:976] (3/7) Epoch 27, batch 1250, loss[loss=0.1846, simple_loss=0.2535, pruned_loss=0.05782, over 4829.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2415, pruned_loss=0.04846, over 956016.48 frames. ], batch size: 39, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:30:15,036 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150170.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:30:15,079 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.7838, 1.5426, 1.4388, 0.8170, 1.6516, 1.7660, 1.7270, 1.5010], device='cuda:3'), covar=tensor([0.0852, 0.0620, 0.0519, 0.0627, 0.0495, 0.0484, 0.0344, 0.0594], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0148, 0.0129, 0.0124, 0.0131, 0.0130, 0.0142, 0.0150], device='cuda:3'), out_proj_covar=tensor([8.8849e-05, 1.0613e-04, 9.1666e-05, 8.6773e-05, 9.1616e-05, 9.2020e-05, 1.0123e-04, 1.0766e-04], device='cuda:3') 2023-03-27 08:30:15,534 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.012e+02 1.554e+02 1.886e+02 2.235e+02 6.588e+02, threshold=3.772e+02, percent-clipped=2.0 2023-03-27 08:30:48,283 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-27 08:30:52,761 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6636, 1.5991, 1.5616, 1.6502, 1.2187, 3.5224, 1.3932, 1.7599], device='cuda:3'), covar=tensor([0.3257, 0.2488, 0.2114, 0.2368, 0.1727, 0.0202, 0.2577, 0.1249], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0116, 0.0120, 0.0124, 0.0113, 0.0095, 0.0094, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 08:30:56,392 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=150218.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:30:57,558 INFO [finetune.py:976] (3/7) Epoch 27, batch 1300, loss[loss=0.1397, simple_loss=0.2168, pruned_loss=0.0313, over 4858.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2388, pruned_loss=0.0472, over 956169.08 frames. ], batch size: 31, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:31:02,270 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6721, 2.4730, 2.1752, 1.0905, 2.2879, 2.0276, 1.9264, 2.3492], device='cuda:3'), covar=tensor([0.1015, 0.0831, 0.1887, 0.2179, 0.1357, 0.2199, 0.2149, 0.1000], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0193, 0.0201, 0.0183, 0.0210, 0.0212, 0.0225, 0.0196], 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') 2023-03-27 08:31:02,850 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150221.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:31:42,213 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=150269.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:31:42,765 INFO [finetune.py:976] (3/7) Epoch 27, batch 1350, loss[loss=0.1349, simple_loss=0.2026, pruned_loss=0.03364, over 4763.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2382, pruned_loss=0.04685, over 957306.52 frames. ], batch size: 26, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:31:43,344 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.009e+02 1.453e+02 1.768e+02 2.125e+02 3.830e+02, threshold=3.537e+02, percent-clipped=1.0 2023-03-27 08:31:45,802 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150274.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:32:01,016 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.97 vs. limit=5.0 2023-03-27 08:32:15,654 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-03-27 08:32:16,593 INFO [finetune.py:976] (3/7) Epoch 27, batch 1400, loss[loss=0.1745, simple_loss=0.2607, pruned_loss=0.0441, over 4792.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2411, pruned_loss=0.04772, over 955896.67 frames. ], batch size: 54, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:32:28,271 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150335.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:32:49,842 INFO [finetune.py:976] (3/7) Epoch 27, batch 1450, loss[loss=0.1614, simple_loss=0.2346, pruned_loss=0.04414, over 4796.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2431, pruned_loss=0.04795, over 956471.53 frames. ], batch size: 25, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:32:50,437 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.045e+02 1.587e+02 1.925e+02 2.309e+02 4.827e+02, threshold=3.851e+02, percent-clipped=3.0 2023-03-27 08:33:09,285 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2575, 1.9058, 2.4976, 1.7238, 2.2687, 2.4857, 1.7271, 2.6096], device='cuda:3'), covar=tensor([0.1283, 0.1954, 0.1388, 0.1918, 0.0916, 0.1286, 0.2643, 0.0729], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0207, 0.0194, 0.0191, 0.0175, 0.0214, 0.0219, 0.0199], 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') 2023-03-27 08:33:11,824 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150402.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 08:33:29,004 INFO [finetune.py:976] (3/7) Epoch 27, batch 1500, loss[loss=0.158, simple_loss=0.2295, pruned_loss=0.04326, over 4671.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.243, pruned_loss=0.04775, over 955288.24 frames. ], batch size: 23, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:33:42,988 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150435.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:33:52,710 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.58 vs. limit=5.0 2023-03-27 08:33:53,600 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=150450.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 08:34:05,547 INFO [finetune.py:976] (3/7) Epoch 27, batch 1550, loss[loss=0.1564, simple_loss=0.2401, pruned_loss=0.03633, over 4785.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2444, pruned_loss=0.04866, over 955456.36 frames. ], batch size: 29, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:34:06,131 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.290e+01 1.580e+02 1.863e+02 2.206e+02 4.598e+02, threshold=3.727e+02, percent-clipped=2.0 2023-03-27 08:34:14,660 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4327, 3.8828, 3.6110, 1.8029, 3.9714, 2.8128, 0.9182, 2.6447], device='cuda:3'), covar=tensor([0.2367, 0.1622, 0.1441, 0.3093, 0.0945, 0.1045, 0.4219, 0.1381], device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0179, 0.0161, 0.0129, 0.0161, 0.0124, 0.0148, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 08:34:27,944 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5189, 2.5336, 2.0239, 2.8549, 2.5409, 2.1463, 3.1885, 2.5478], device='cuda:3'), covar=tensor([0.1481, 0.2220, 0.2968, 0.2433, 0.2467, 0.1755, 0.2852, 0.1745], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0190, 0.0236, 0.0253, 0.0249, 0.0207, 0.0215, 0.0202], 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') 2023-03-27 08:34:38,728 INFO [finetune.py:976] (3/7) Epoch 27, batch 1600, loss[loss=0.1621, simple_loss=0.2215, pruned_loss=0.05132, over 4826.00 frames. ], tot_loss[loss=0.169, simple_loss=0.242, pruned_loss=0.04803, over 956113.15 frames. ], batch size: 25, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:34:50,075 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150537.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:35:11,518 INFO [finetune.py:976] (3/7) Epoch 27, batch 1650, loss[loss=0.1674, simple_loss=0.2373, pruned_loss=0.04877, over 4820.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2406, pruned_loss=0.0481, over 953912.35 frames. ], batch size: 33, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:35:11,648 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9592, 2.1642, 1.7588, 1.8423, 2.4541, 2.4951, 2.0936, 1.9822], device='cuda:3'), covar=tensor([0.0446, 0.0365, 0.0562, 0.0315, 0.0262, 0.0617, 0.0384, 0.0403], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0106, 0.0146, 0.0111, 0.0101, 0.0115, 0.0104, 0.0113], device='cuda:3'), out_proj_covar=tensor([7.7689e-05, 8.0848e-05, 1.1382e-04, 8.4640e-05, 7.7959e-05, 8.4976e-05, 7.6980e-05, 8.5676e-05], device='cuda:3') 2023-03-27 08:35:12,139 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.130e+02 1.535e+02 1.741e+02 2.182e+02 5.670e+02, threshold=3.482e+02, percent-clipped=1.0 2023-03-27 08:35:37,663 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150598.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:35:45,401 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3856, 2.5705, 2.3862, 1.8014, 2.3432, 2.6295, 2.6937, 2.2083], device='cuda:3'), covar=tensor([0.0631, 0.0587, 0.0741, 0.0867, 0.0924, 0.0669, 0.0603, 0.0986], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0137, 0.0141, 0.0119, 0.0128, 0.0138, 0.0140, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 08:35:54,937 INFO [finetune.py:976] (3/7) Epoch 27, batch 1700, loss[loss=0.1713, simple_loss=0.2465, pruned_loss=0.04809, over 4808.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2399, pruned_loss=0.04824, over 955278.17 frames. ], batch size: 45, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:35:55,690 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2902, 2.3559, 1.9119, 2.3895, 2.2152, 2.2160, 2.2094, 3.1268], device='cuda:3'), covar=tensor([0.3641, 0.4207, 0.3322, 0.3971, 0.4564, 0.2504, 0.4054, 0.1559], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0265, 0.0237, 0.0277, 0.0260, 0.0230, 0.0260, 0.0239], 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') 2023-03-27 08:36:01,036 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150630.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:36:41,945 INFO [finetune.py:976] (3/7) Epoch 27, batch 1750, loss[loss=0.2201, simple_loss=0.2922, pruned_loss=0.07402, over 4923.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2408, pruned_loss=0.04848, over 956170.67 frames. ], batch size: 38, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:36:42,539 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.512e+01 1.530e+02 1.821e+02 2.198e+02 3.521e+02, threshold=3.642e+02, percent-clipped=1.0 2023-03-27 08:36:50,604 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.8250, 1.6658, 1.8792, 1.1541, 1.9568, 2.1564, 1.9418, 1.7231], device='cuda:3'), covar=tensor([0.0947, 0.0880, 0.0522, 0.0568, 0.0512, 0.0557, 0.0479, 0.0656], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0147, 0.0128, 0.0122, 0.0130, 0.0129, 0.0141, 0.0149], device='cuda:3'), out_proj_covar=tensor([8.8337e-05, 1.0537e-04, 9.1037e-05, 8.5955e-05, 9.1176e-05, 9.1219e-05, 1.0039e-04, 1.0701e-04], device='cuda:3') 2023-03-27 08:36:54,028 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5205, 1.2482, 1.7134, 1.7949, 1.4238, 3.1538, 1.1734, 1.3481], device='cuda:3'), covar=tensor([0.1046, 0.1943, 0.1219, 0.0946, 0.1711, 0.0238, 0.1722, 0.1973], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0073, 0.0076, 0.0091, 0.0080, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 08:37:05,050 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-27 08:37:13,852 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0 2023-03-27 08:37:15,428 INFO [finetune.py:976] (3/7) Epoch 27, batch 1800, loss[loss=0.215, simple_loss=0.2769, pruned_loss=0.07657, over 4918.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.244, pruned_loss=0.04957, over 955457.79 frames. ], batch size: 36, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:37:23,560 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3307, 1.2648, 1.5695, 1.0484, 1.3315, 1.3655, 1.2664, 1.5820], device='cuda:3'), covar=tensor([0.1137, 0.2298, 0.1244, 0.1590, 0.0955, 0.1277, 0.2907, 0.0856], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0207, 0.0194, 0.0192, 0.0175, 0.0214, 0.0219, 0.0199], 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') 2023-03-27 08:37:27,759 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150732.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:37:33,296 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150735.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:37:38,137 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3051, 1.2716, 1.6273, 1.0208, 1.3196, 1.3878, 1.2977, 1.5857], device='cuda:3'), covar=tensor([0.1307, 0.2267, 0.1222, 0.1568, 0.1011, 0.1373, 0.3090, 0.0972], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0208, 0.0195, 0.0192, 0.0176, 0.0215, 0.0219, 0.0199], 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') 2023-03-27 08:37:51,385 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.3251, 1.3386, 1.3176, 0.8166, 1.3598, 1.6206, 1.6105, 1.2709], device='cuda:3'), covar=tensor([0.0965, 0.0692, 0.0664, 0.0566, 0.0567, 0.0539, 0.0347, 0.0734], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0147, 0.0128, 0.0123, 0.0131, 0.0129, 0.0142, 0.0150], device='cuda:3'), out_proj_covar=tensor([8.8628e-05, 1.0569e-04, 9.1365e-05, 8.6301e-05, 9.1764e-05, 9.1537e-05, 1.0113e-04, 1.0732e-04], device='cuda:3') 2023-03-27 08:37:57,187 INFO [finetune.py:976] (3/7) Epoch 27, batch 1850, loss[loss=0.1697, simple_loss=0.2406, pruned_loss=0.04936, over 4747.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2462, pruned_loss=0.05027, over 955633.72 frames. ], batch size: 28, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:37:57,786 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.063e+02 1.537e+02 1.800e+02 2.248e+02 4.542e+02, threshold=3.600e+02, percent-clipped=6.0 2023-03-27 08:38:05,038 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=150783.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:38:11,140 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150793.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 08:38:30,231 INFO [finetune.py:976] (3/7) Epoch 27, batch 1900, loss[loss=0.1985, simple_loss=0.285, pruned_loss=0.05598, over 4880.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2483, pruned_loss=0.05089, over 954980.81 frames. ], batch size: 43, lr: 2.93e-03, grad_scale: 8.0 2023-03-27 08:39:05,881 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.70 vs. limit=5.0 2023-03-27 08:39:14,072 INFO [finetune.py:976] (3/7) Epoch 27, batch 1950, loss[loss=0.1455, simple_loss=0.2273, pruned_loss=0.03179, over 4848.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2459, pruned_loss=0.05002, over 954168.23 frames. ], batch size: 44, lr: 2.92e-03, grad_scale: 8.0 2023-03-27 08:39:14,654 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.033e+02 1.460e+02 1.651e+02 1.933e+02 3.642e+02, threshold=3.302e+02, percent-clipped=1.0 2023-03-27 08:39:19,596 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9854, 1.9334, 1.7426, 1.9536, 2.0294, 1.7941, 2.1814, 2.0925], device='cuda:3'), covar=tensor([0.1164, 0.1812, 0.2548, 0.2122, 0.2205, 0.1483, 0.2453, 0.1395], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0190, 0.0236, 0.0253, 0.0249, 0.0207, 0.0214, 0.0203], 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') 2023-03-27 08:39:20,793 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2005, 1.7512, 2.3419, 4.0291, 2.7258, 2.8706, 0.9877, 3.3760], device='cuda:3'), covar=tensor([0.1627, 0.1389, 0.1498, 0.0507, 0.0787, 0.1461, 0.2014, 0.0375], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0132, 0.0163, 0.0101, 0.0136, 0.0125, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 08:39:28,775 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150893.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:39:33,110 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150900.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:39:39,816 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4507, 1.3553, 1.3422, 1.3781, 0.8424, 2.2352, 0.6806, 1.1382], device='cuda:3'), covar=tensor([0.3528, 0.2600, 0.2295, 0.2577, 0.1986, 0.0383, 0.2821, 0.1418], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0116, 0.0120, 0.0124, 0.0113, 0.0095, 0.0093, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0005, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 08:39:42,121 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1723, 3.6228, 3.8176, 3.9915, 3.9305, 3.7491, 4.2606, 1.3057], device='cuda:3'), covar=tensor([0.0877, 0.0905, 0.0907, 0.1047, 0.1362, 0.1711, 0.0859, 0.5921], device='cuda:3'), in_proj_covar=tensor([0.0356, 0.0251, 0.0286, 0.0300, 0.0340, 0.0291, 0.0310, 0.0306], 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') 2023-03-27 08:39:47,871 INFO [finetune.py:976] (3/7) Epoch 27, batch 2000, loss[loss=0.1984, simple_loss=0.261, pruned_loss=0.06794, over 4918.00 frames. ], tot_loss[loss=0.172, simple_loss=0.244, pruned_loss=0.04996, over 955897.99 frames. ], batch size: 43, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:39:54,535 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150930.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:40:15,344 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150961.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:40:21,603 INFO [finetune.py:976] (3/7) Epoch 27, batch 2050, loss[loss=0.129, simple_loss=0.2063, pruned_loss=0.02588, over 4805.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2398, pruned_loss=0.04825, over 954961.79 frames. ], batch size: 25, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:40:22,190 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.304e+01 1.432e+02 1.658e+02 2.071e+02 3.830e+02, threshold=3.317e+02, percent-clipped=1.0 2023-03-27 08:40:27,052 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=150978.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:40:56,342 INFO [finetune.py:976] (3/7) Epoch 27, batch 2100, loss[loss=0.1708, simple_loss=0.2464, pruned_loss=0.04756, over 4820.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2397, pruned_loss=0.04797, over 957257.62 frames. ], batch size: 45, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:41:11,411 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.0057, 4.3279, 4.5918, 4.7766, 4.7623, 4.4335, 5.0956, 1.7110], device='cuda:3'), covar=tensor([0.0773, 0.0902, 0.0796, 0.0951, 0.1246, 0.1860, 0.0626, 0.5867], device='cuda:3'), in_proj_covar=tensor([0.0356, 0.0251, 0.0285, 0.0299, 0.0339, 0.0291, 0.0310, 0.0305], 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') 2023-03-27 08:41:12,044 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0704, 1.9498, 1.5751, 0.6619, 1.6738, 1.7176, 1.5572, 1.8610], device='cuda:3'), covar=tensor([0.0985, 0.0797, 0.1507, 0.2050, 0.1342, 0.2160, 0.2469, 0.0923], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0192, 0.0201, 0.0182, 0.0209, 0.0211, 0.0225, 0.0197], 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') 2023-03-27 08:41:47,143 INFO [finetune.py:976] (3/7) Epoch 27, batch 2150, loss[loss=0.1743, simple_loss=0.2494, pruned_loss=0.04957, over 4826.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.241, pruned_loss=0.04801, over 956590.46 frames. ], batch size: 30, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:41:48,291 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.039e+02 1.525e+02 1.813e+02 2.166e+02 3.448e+02, threshold=3.626e+02, percent-clipped=1.0 2023-03-27 08:42:02,497 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151088.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 08:42:08,517 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2472, 1.1744, 1.5830, 1.0654, 1.2917, 1.4284, 1.1431, 1.5240], device='cuda:3'), covar=tensor([0.1426, 0.2442, 0.1367, 0.1574, 0.1168, 0.1347, 0.3348, 0.1173], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0207, 0.0193, 0.0190, 0.0175, 0.0213, 0.0218, 0.0198], 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') 2023-03-27 08:42:23,614 INFO [finetune.py:976] (3/7) Epoch 27, batch 2200, loss[loss=0.1562, simple_loss=0.2398, pruned_loss=0.03632, over 4759.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2428, pruned_loss=0.04858, over 955777.22 frames. ], batch size: 28, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:42:24,875 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.54 vs. limit=5.0 2023-03-27 08:43:01,672 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-27 08:43:04,289 INFO [finetune.py:976] (3/7) Epoch 27, batch 2250, loss[loss=0.1796, simple_loss=0.2619, pruned_loss=0.04862, over 4891.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2455, pruned_loss=0.05005, over 956568.45 frames. ], batch size: 37, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:43:04,889 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.120e+01 1.457e+02 1.754e+02 2.221e+02 3.820e+02, threshold=3.509e+02, percent-clipped=1.0 2023-03-27 08:43:08,324 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4054, 2.2080, 1.8220, 0.8520, 1.9448, 1.8770, 1.7739, 2.1089], device='cuda:3'), covar=tensor([0.0816, 0.0889, 0.1673, 0.2237, 0.1300, 0.2254, 0.2241, 0.0881], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0192, 0.0201, 0.0182, 0.0209, 0.0211, 0.0224, 0.0196], 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') 2023-03-27 08:43:14,285 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.78 vs. limit=5.0 2023-03-27 08:43:14,767 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151184.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:43:20,788 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151193.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:43:37,566 INFO [finetune.py:976] (3/7) Epoch 27, batch 2300, loss[loss=0.1734, simple_loss=0.2497, pruned_loss=0.04861, over 4837.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2457, pruned_loss=0.04991, over 956203.66 frames. ], batch size: 49, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:43:51,735 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=151241.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:43:54,685 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151245.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:44:06,983 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151256.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:44:08,912 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.34 vs. limit=5.0 2023-03-27 08:44:18,928 INFO [finetune.py:976] (3/7) Epoch 27, batch 2350, loss[loss=0.1544, simple_loss=0.2335, pruned_loss=0.03771, over 4789.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2432, pruned_loss=0.04915, over 954810.73 frames. ], batch size: 51, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:44:19,965 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 7.208e+01 1.503e+02 1.827e+02 2.189e+02 3.264e+02, threshold=3.653e+02, percent-clipped=0.0 2023-03-27 08:44:28,775 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151282.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:44:39,438 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151298.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:44:52,608 INFO [finetune.py:976] (3/7) Epoch 27, batch 2400, loss[loss=0.1603, simple_loss=0.2274, pruned_loss=0.04659, over 4825.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2408, pruned_loss=0.04841, over 956932.15 frames. ], batch size: 39, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:45:09,231 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151343.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:45:19,405 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151359.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:45:26,013 INFO [finetune.py:976] (3/7) Epoch 27, batch 2450, loss[loss=0.1408, simple_loss=0.2158, pruned_loss=0.03293, over 4800.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.238, pruned_loss=0.04789, over 956063.94 frames. ], batch size: 29, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:45:26,602 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.046e+02 1.415e+02 1.689e+02 1.968e+02 4.441e+02, threshold=3.378e+02, percent-clipped=1.0 2023-03-27 08:45:35,683 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-03-27 08:45:38,488 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151388.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:45:58,929 INFO [finetune.py:976] (3/7) Epoch 27, batch 2500, loss[loss=0.2111, simple_loss=0.2928, pruned_loss=0.06465, over 4735.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2404, pruned_loss=0.0494, over 953580.38 frames. ], batch size: 59, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:46:12,824 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=151436.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:46:48,696 INFO [finetune.py:976] (3/7) Epoch 27, batch 2550, loss[loss=0.1737, simple_loss=0.2469, pruned_loss=0.05022, over 4833.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2433, pruned_loss=0.04971, over 953986.74 frames. ], batch size: 39, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:46:49,282 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.209e+01 1.472e+02 1.881e+02 2.470e+02 3.912e+02, threshold=3.762e+02, percent-clipped=2.0 2023-03-27 08:46:56,412 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-27 08:47:24,844 INFO [finetune.py:976] (3/7) Epoch 27, batch 2600, loss[loss=0.1389, simple_loss=0.189, pruned_loss=0.04445, over 3605.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2449, pruned_loss=0.05055, over 953256.30 frames. ], batch size: 15, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:47:42,187 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151540.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:48:03,486 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151556.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:48:10,383 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-03-27 08:48:16,178 INFO [finetune.py:976] (3/7) Epoch 27, batch 2650, loss[loss=0.1712, simple_loss=0.2436, pruned_loss=0.04939, over 4799.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2449, pruned_loss=0.04995, over 951693.33 frames. ], batch size: 51, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:48:16,787 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.199e+02 1.649e+02 1.887e+02 2.270e+02 4.456e+02, threshold=3.774e+02, percent-clipped=3.0 2023-03-27 08:48:39,917 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=151604.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:48:49,946 INFO [finetune.py:976] (3/7) Epoch 27, batch 2700, loss[loss=0.1802, simple_loss=0.2504, pruned_loss=0.05504, over 4893.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2443, pruned_loss=0.04954, over 952648.25 frames. ], batch size: 43, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:49:01,310 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151638.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:49:03,198 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2648, 1.2390, 1.2022, 1.3043, 1.5265, 1.4799, 1.3561, 1.1966], device='cuda:3'), covar=tensor([0.0391, 0.0286, 0.0598, 0.0294, 0.0221, 0.0435, 0.0319, 0.0421], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0106, 0.0146, 0.0111, 0.0101, 0.0115, 0.0103, 0.0113], device='cuda:3'), out_proj_covar=tensor([7.7605e-05, 8.0999e-05, 1.1402e-04, 8.4815e-05, 7.8258e-05, 8.5017e-05, 7.6351e-05, 8.5796e-05], device='cuda:3') 2023-03-27 08:49:12,823 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151654.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:49:29,775 INFO [finetune.py:976] (3/7) Epoch 27, batch 2750, loss[loss=0.1865, simple_loss=0.25, pruned_loss=0.06147, over 4807.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.242, pruned_loss=0.04917, over 952889.49 frames. ], batch size: 41, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:49:30,372 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.940e+01 1.418e+02 1.693e+02 2.178e+02 3.976e+02, threshold=3.385e+02, percent-clipped=1.0 2023-03-27 08:49:44,697 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151688.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:49:45,292 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.0977, 1.2443, 1.4215, 1.3053, 1.3018, 2.4852, 1.0413, 1.2864], device='cuda:3'), covar=tensor([0.1276, 0.2566, 0.1206, 0.1159, 0.2177, 0.0437, 0.2309, 0.2759], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0083, 0.0074, 0.0076, 0.0092, 0.0080, 0.0086, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-27 08:50:06,331 INFO [finetune.py:976] (3/7) Epoch 27, batch 2800, loss[loss=0.1517, simple_loss=0.233, pruned_loss=0.03519, over 4755.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2406, pruned_loss=0.04897, over 952392.79 frames. ], batch size: 26, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:50:12,929 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.2673, 3.6895, 3.8954, 4.0942, 4.0065, 3.7820, 4.3430, 1.3468], device='cuda:3'), covar=tensor([0.0905, 0.0982, 0.0868, 0.1076, 0.1259, 0.1730, 0.0832, 0.6091], device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0249, 0.0282, 0.0297, 0.0335, 0.0288, 0.0307, 0.0302], 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') 2023-03-27 08:50:24,973 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151749.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:50:39,489 INFO [finetune.py:976] (3/7) Epoch 27, batch 2850, loss[loss=0.2066, simple_loss=0.285, pruned_loss=0.06412, over 4905.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2385, pruned_loss=0.04805, over 952882.41 frames. ], batch size: 37, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:50:40,100 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.092e+02 1.485e+02 1.795e+02 2.169e+02 3.375e+02, threshold=3.589e+02, percent-clipped=0.0 2023-03-27 08:50:51,022 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6570, 1.5332, 1.9494, 1.2280, 1.6853, 1.8807, 1.4478, 2.1006], device='cuda:3'), covar=tensor([0.1275, 0.2020, 0.1251, 0.1751, 0.0900, 0.1295, 0.2841, 0.0727], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0207, 0.0193, 0.0190, 0.0175, 0.0213, 0.0218, 0.0198], 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') 2023-03-27 08:50:59,493 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151801.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:51:12,912 INFO [finetune.py:976] (3/7) Epoch 27, batch 2900, loss[loss=0.1917, simple_loss=0.2643, pruned_loss=0.0595, over 4824.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2408, pruned_loss=0.0488, over 954051.72 frames. ], batch size: 39, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:51:25,537 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151840.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:51:34,455 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2713, 2.9117, 2.8156, 1.1816, 3.0331, 2.2046, 0.7494, 1.8877], device='cuda:3'), covar=tensor([0.2212, 0.2228, 0.1724, 0.3628, 0.1367, 0.1183, 0.4032, 0.1709], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0179, 0.0160, 0.0130, 0.0160, 0.0124, 0.0148, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 08:51:54,451 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151862.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:51:56,256 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4847, 1.0724, 0.7960, 1.3288, 1.9179, 0.7770, 1.2739, 1.2913], device='cuda:3'), covar=tensor([0.1584, 0.2117, 0.1595, 0.1195, 0.1831, 0.1901, 0.1471, 0.2037], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0092, 0.0107, 0.0090, 0.0117, 0.0091, 0.0097, 0.0087], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-27 08:52:04,134 INFO [finetune.py:976] (3/7) Epoch 27, batch 2950, loss[loss=0.1482, simple_loss=0.235, pruned_loss=0.03067, over 4810.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2438, pruned_loss=0.04923, over 954726.20 frames. ], batch size: 38, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:52:04,749 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.132e+02 1.531e+02 1.876e+02 2.281e+02 4.815e+02, threshold=3.752e+02, percent-clipped=2.0 2023-03-27 08:52:15,645 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=151888.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:52:26,517 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.0582, 1.2650, 1.4805, 1.2894, 1.3625, 2.3653, 1.1877, 1.3960], device='cuda:3'), covar=tensor([0.1042, 0.1731, 0.1110, 0.0909, 0.1586, 0.0380, 0.1477, 0.1726], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0073, 0.0076, 0.0091, 0.0080, 0.0086, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 08:52:37,407 INFO [finetune.py:976] (3/7) Epoch 27, batch 3000, loss[loss=0.1575, simple_loss=0.2434, pruned_loss=0.03579, over 4893.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2456, pruned_loss=0.04949, over 956382.91 frames. ], batch size: 43, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:52:37,407 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-27 08:52:50,759 INFO [finetune.py:1010] (3/7) Epoch 27, validation: loss=0.1572, simple_loss=0.2248, pruned_loss=0.04486, over 2265189.00 frames. 2023-03-27 08:52:50,760 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6469MB 2023-03-27 08:53:01,270 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1065, 2.1339, 1.8175, 2.0154, 2.0053, 1.9672, 2.0638, 2.6890], device='cuda:3'), covar=tensor([0.3399, 0.3851, 0.3007, 0.3549, 0.3859, 0.2154, 0.3497, 0.1630], device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0263, 0.0236, 0.0274, 0.0260, 0.0229, 0.0258, 0.0238], 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') 2023-03-27 08:53:01,838 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151938.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:53:14,180 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151954.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:53:32,162 INFO [finetune.py:976] (3/7) Epoch 27, batch 3050, loss[loss=0.1839, simple_loss=0.2372, pruned_loss=0.06532, over 4756.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2464, pruned_loss=0.04992, over 956211.85 frames. ], batch size: 28, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:53:32,747 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.022e+02 1.546e+02 1.837e+02 2.199e+02 4.500e+02, threshold=3.674e+02, percent-clipped=2.0 2023-03-27 08:53:44,360 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=151986.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:53:55,280 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.9753, 4.7385, 4.5327, 2.6667, 4.8552, 3.6913, 1.0289, 3.3979], device='cuda:3'), covar=tensor([0.2164, 0.1762, 0.1468, 0.2901, 0.0687, 0.0824, 0.4535, 0.1437], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0178, 0.0159, 0.0129, 0.0160, 0.0123, 0.0148, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 08:53:55,893 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=152002.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:54:06,689 INFO [finetune.py:976] (3/7) Epoch 27, batch 3100, loss[loss=0.1322, simple_loss=0.2026, pruned_loss=0.03095, over 4937.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2442, pruned_loss=0.04895, over 956533.43 frames. ], batch size: 38, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:54:23,681 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152044.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:54:29,204 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-27 08:54:32,356 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-03-27 08:54:41,700 INFO [finetune.py:976] (3/7) Epoch 27, batch 3150, loss[loss=0.2119, simple_loss=0.2748, pruned_loss=0.07452, over 4826.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2417, pruned_loss=0.04859, over 952055.76 frames. ], batch size: 41, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:54:42,282 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.945e+01 1.491e+02 1.827e+02 2.202e+02 3.039e+02, threshold=3.654e+02, percent-clipped=0.0 2023-03-27 08:55:21,807 INFO [finetune.py:976] (3/7) Epoch 27, batch 3200, loss[loss=0.1654, simple_loss=0.2396, pruned_loss=0.04566, over 4912.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2389, pruned_loss=0.0476, over 954008.10 frames. ], batch size: 36, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:55:24,377 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8190, 2.4060, 3.1118, 2.0209, 2.6687, 3.1014, 2.2033, 3.1852], device='cuda:3'), covar=tensor([0.1353, 0.1909, 0.1434, 0.2064, 0.1113, 0.1406, 0.2548, 0.0788], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0208, 0.0194, 0.0190, 0.0175, 0.0214, 0.0219, 0.0199], 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') 2023-03-27 08:55:46,791 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152157.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:55:53,516 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0298, 1.6543, 2.2298, 3.8079, 2.5132, 2.7472, 0.8068, 3.1453], device='cuda:3'), covar=tensor([0.1618, 0.1371, 0.1411, 0.0473, 0.0788, 0.1772, 0.2006, 0.0381], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0132, 0.0163, 0.0100, 0.0135, 0.0124, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 08:55:54,656 INFO [finetune.py:976] (3/7) Epoch 27, batch 3250, loss[loss=0.1935, simple_loss=0.2591, pruned_loss=0.06395, over 4766.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2403, pruned_loss=0.04856, over 954515.35 frames. ], batch size: 54, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:55:55,261 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.795e+01 1.453e+02 1.756e+02 2.073e+02 3.538e+02, threshold=3.512e+02, percent-clipped=0.0 2023-03-27 08:56:32,279 INFO [finetune.py:976] (3/7) Epoch 27, batch 3300, loss[loss=0.2004, simple_loss=0.2778, pruned_loss=0.0615, over 4768.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2449, pruned_loss=0.04977, over 956282.82 frames. ], batch size: 54, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:56:35,395 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152225.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:57:13,841 INFO [finetune.py:976] (3/7) Epoch 27, batch 3350, loss[loss=0.1879, simple_loss=0.2565, pruned_loss=0.05966, over 4893.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2462, pruned_loss=0.05027, over 954764.76 frames. ], batch size: 35, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:57:14,395 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.581e+01 1.606e+02 1.884e+02 2.337e+02 3.345e+02, threshold=3.768e+02, percent-clipped=0.0 2023-03-27 08:57:20,869 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152273.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:57:33,308 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-27 08:57:33,557 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152286.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:57:46,034 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2372, 2.0413, 1.7214, 2.1420, 2.6204, 2.2122, 2.1479, 1.6485], device='cuda:3'), covar=tensor([0.2040, 0.1759, 0.1786, 0.1582, 0.1672, 0.1081, 0.1970, 0.1849], device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0212, 0.0217, 0.0201, 0.0248, 0.0192, 0.0220, 0.0207], 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') 2023-03-27 08:57:46,655 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8872, 1.6899, 1.6184, 1.3436, 1.6890, 1.7177, 1.7351, 2.1541], device='cuda:3'), covar=tensor([0.3508, 0.3660, 0.3032, 0.3270, 0.3574, 0.2376, 0.3038, 0.1824], device='cuda:3'), in_proj_covar=tensor([0.0293, 0.0266, 0.0239, 0.0279, 0.0263, 0.0232, 0.0261, 0.0241], 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') 2023-03-27 08:58:01,025 INFO [finetune.py:976] (3/7) Epoch 27, batch 3400, loss[loss=0.1564, simple_loss=0.2378, pruned_loss=0.03755, over 4756.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.247, pruned_loss=0.0501, over 955417.81 frames. ], batch size: 27, lr: 2.92e-03, grad_scale: 16.0 2023-03-27 08:58:07,864 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.4829, 1.5616, 1.5731, 0.9003, 1.6188, 1.9327, 1.8671, 1.4717], device='cuda:3'), covar=tensor([0.0985, 0.0637, 0.0579, 0.0616, 0.0538, 0.0567, 0.0382, 0.0687], device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0148, 0.0130, 0.0123, 0.0132, 0.0131, 0.0143, 0.0151], device='cuda:3'), out_proj_covar=tensor([8.9338e-05, 1.0625e-04, 9.2668e-05, 8.6602e-05, 9.2713e-05, 9.2687e-05, 1.0130e-04, 1.0790e-04], device='cuda:3') 2023-03-27 08:58:09,647 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152334.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:58:16,672 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152344.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:58:36,175 INFO [finetune.py:976] (3/7) Epoch 27, batch 3450, loss[loss=0.1349, simple_loss=0.2157, pruned_loss=0.02702, over 4759.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2475, pruned_loss=0.0502, over 956512.51 frames. ], batch size: 28, lr: 2.91e-03, grad_scale: 16.0 2023-03-27 08:58:36,743 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.916e+01 1.467e+02 1.787e+02 2.252e+02 4.149e+02, threshold=3.573e+02, percent-clipped=3.0 2023-03-27 08:58:58,955 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=152392.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:59:17,727 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5469, 1.6530, 1.9764, 1.8224, 1.7624, 3.5902, 1.4996, 1.7319], device='cuda:3'), covar=tensor([0.0911, 0.1789, 0.0991, 0.0893, 0.1472, 0.0227, 0.1426, 0.1674], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0083, 0.0074, 0.0077, 0.0092, 0.0080, 0.0086, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-27 08:59:18,831 INFO [finetune.py:976] (3/7) Epoch 27, batch 3500, loss[loss=0.1895, simple_loss=0.2646, pruned_loss=0.05724, over 4832.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2453, pruned_loss=0.04994, over 957595.59 frames. ], batch size: 33, lr: 2.91e-03, grad_scale: 16.0 2023-03-27 08:59:34,557 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152445.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:59:43,765 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152457.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 08:59:52,116 INFO [finetune.py:976] (3/7) Epoch 27, batch 3550, loss[loss=0.1648, simple_loss=0.2242, pruned_loss=0.05275, over 4072.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2427, pruned_loss=0.0494, over 957393.97 frames. ], batch size: 17, lr: 2.91e-03, grad_scale: 16.0 2023-03-27 08:59:52,706 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.528e+01 1.381e+02 1.664e+02 2.040e+02 3.997e+02, threshold=3.328e+02, percent-clipped=1.0 2023-03-27 09:00:25,117 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=152505.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:00:26,320 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152506.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:00:36,271 INFO [finetune.py:976] (3/7) Epoch 27, batch 3600, loss[loss=0.1283, simple_loss=0.1956, pruned_loss=0.03052, over 4186.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2399, pruned_loss=0.04864, over 957210.30 frames. ], batch size: 18, lr: 2.91e-03, grad_scale: 16.0 2023-03-27 09:01:10,221 INFO [finetune.py:976] (3/7) Epoch 27, batch 3650, loss[loss=0.1774, simple_loss=0.2548, pruned_loss=0.04999, over 4819.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2425, pruned_loss=0.05046, over 954115.85 frames. ], batch size: 38, lr: 2.91e-03, grad_scale: 16.0 2023-03-27 09:01:10,828 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.982e+01 1.594e+02 1.907e+02 2.265e+02 4.160e+02, threshold=3.814e+02, percent-clipped=3.0 2023-03-27 09:01:17,559 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152581.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:01:18,186 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152582.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:01:22,485 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152589.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:01:46,436 INFO [finetune.py:976] (3/7) Epoch 27, batch 3700, loss[loss=0.1783, simple_loss=0.2685, pruned_loss=0.04404, over 4820.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2458, pruned_loss=0.0508, over 954079.88 frames. ], batch size: 47, lr: 2.91e-03, grad_scale: 16.0 2023-03-27 09:01:52,531 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152629.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:02:01,196 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152643.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:02:05,487 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152650.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:02:22,198 INFO [finetune.py:976] (3/7) Epoch 27, batch 3750, loss[loss=0.2152, simple_loss=0.2716, pruned_loss=0.07936, over 4832.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2458, pruned_loss=0.0502, over 953566.86 frames. ], batch size: 30, lr: 2.91e-03, grad_scale: 16.0 2023-03-27 09:02:22,799 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.041e+02 1.491e+02 1.751e+02 2.166e+02 4.226e+02, threshold=3.502e+02, percent-clipped=3.0 2023-03-27 09:03:12,604 INFO [finetune.py:976] (3/7) Epoch 27, batch 3800, loss[loss=0.1677, simple_loss=0.2322, pruned_loss=0.05156, over 4694.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2467, pruned_loss=0.0505, over 953217.18 frames. ], batch size: 23, lr: 2.91e-03, grad_scale: 16.0 2023-03-27 09:03:16,875 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152726.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:03:18,104 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8911, 1.6599, 1.6125, 1.9250, 1.9644, 2.0127, 1.3657, 1.6254], device='cuda:3'), covar=tensor([0.2049, 0.1942, 0.1821, 0.1595, 0.1581, 0.1096, 0.2522, 0.1882], device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0211, 0.0215, 0.0200, 0.0246, 0.0192, 0.0219, 0.0205], 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') 2023-03-27 09:03:21,151 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6329, 2.6136, 2.5406, 1.8592, 2.4778, 2.8778, 2.8730, 2.2151], device='cuda:3'), covar=tensor([0.0523, 0.0577, 0.0689, 0.0856, 0.0763, 0.0578, 0.0482, 0.1051], device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0138, 0.0142, 0.0121, 0.0130, 0.0140, 0.0142, 0.0164], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 09:03:24,104 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.0797, 1.3149, 1.3805, 1.2467, 1.4304, 2.3875, 1.2234, 1.4865], device='cuda:3'), covar=tensor([0.1087, 0.1920, 0.1122, 0.0968, 0.1705, 0.0383, 0.1578, 0.1815], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0083, 0.0074, 0.0077, 0.0092, 0.0080, 0.0086, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-27 09:03:25,033 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-27 09:03:37,654 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-03-27 09:03:45,581 INFO [finetune.py:976] (3/7) Epoch 27, batch 3850, loss[loss=0.1584, simple_loss=0.2306, pruned_loss=0.04316, over 4927.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2449, pruned_loss=0.04927, over 955670.14 frames. ], batch size: 33, lr: 2.91e-03, grad_scale: 16.0 2023-03-27 09:03:46,653 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.206e+01 1.335e+02 1.631e+02 2.144e+02 3.589e+02, threshold=3.262e+02, percent-clipped=1.0 2023-03-27 09:04:00,006 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152787.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:04:02,014 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-27 09:04:10,897 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5252, 2.3609, 2.9255, 1.9421, 2.4317, 2.8482, 1.9683, 2.9365], device='cuda:3'), covar=tensor([0.1392, 0.1854, 0.1514, 0.2049, 0.1102, 0.1473, 0.2785, 0.0788], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0207, 0.0194, 0.0190, 0.0175, 0.0214, 0.0218, 0.0198], 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') 2023-03-27 09:04:12,777 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152801.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:04:28,201 INFO [finetune.py:976] (3/7) Epoch 27, batch 3900, loss[loss=0.1369, simple_loss=0.2148, pruned_loss=0.02949, over 4857.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.242, pruned_loss=0.04839, over 957684.44 frames. ], batch size: 31, lr: 2.91e-03, grad_scale: 16.0 2023-03-27 09:04:41,786 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2072, 1.6870, 2.1099, 2.1363, 1.8445, 1.8818, 2.0709, 2.0295], device='cuda:3'), covar=tensor([0.4286, 0.4357, 0.3356, 0.4488, 0.5114, 0.4774, 0.4954, 0.3075], device='cuda:3'), in_proj_covar=tensor([0.0266, 0.0248, 0.0268, 0.0297, 0.0295, 0.0272, 0.0301, 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') 2023-03-27 09:04:47,900 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.89 vs. limit=5.0 2023-03-27 09:05:01,434 INFO [finetune.py:976] (3/7) Epoch 27, batch 3950, loss[loss=0.1498, simple_loss=0.2186, pruned_loss=0.04045, over 4846.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2398, pruned_loss=0.04757, over 954491.31 frames. ], batch size: 49, lr: 2.91e-03, grad_scale: 16.0 2023-03-27 09:05:02,042 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.044e+02 1.440e+02 1.686e+02 2.039e+02 3.105e+02, threshold=3.372e+02, percent-clipped=0.0 2023-03-27 09:05:09,753 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152881.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:05:43,211 INFO [finetune.py:976] (3/7) Epoch 27, batch 4000, loss[loss=0.1731, simple_loss=0.2394, pruned_loss=0.05344, over 4900.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2385, pruned_loss=0.04735, over 952867.02 frames. ], batch size: 35, lr: 2.91e-03, grad_scale: 32.0 2023-03-27 09:05:49,731 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=152929.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:05:49,767 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152929.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:05:52,627 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4784, 2.2768, 2.0436, 0.9807, 2.0978, 1.8188, 1.7438, 2.2048], device='cuda:3'), covar=tensor([0.0995, 0.0908, 0.1566, 0.2111, 0.1330, 0.2700, 0.2377, 0.1031], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0192, 0.0202, 0.0183, 0.0210, 0.0212, 0.0226, 0.0196], 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') 2023-03-27 09:05:54,059 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-27 09:05:56,133 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152938.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:06:00,351 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152945.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:06:16,491 INFO [finetune.py:976] (3/7) Epoch 27, batch 4050, loss[loss=0.2158, simple_loss=0.2989, pruned_loss=0.06641, over 4840.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2416, pruned_loss=0.04812, over 953960.79 frames. ], batch size: 47, lr: 2.91e-03, grad_scale: 32.0 2023-03-27 09:06:17,094 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.013e+02 1.469e+02 1.767e+02 2.180e+02 3.425e+02, threshold=3.534e+02, percent-clipped=1.0 2023-03-27 09:06:20,836 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=152977.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:06:49,257 INFO [finetune.py:976] (3/7) Epoch 27, batch 4100, loss[loss=0.1345, simple_loss=0.2119, pruned_loss=0.02858, over 4793.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2421, pruned_loss=0.04807, over 950888.38 frames. ], batch size: 25, lr: 2.91e-03, grad_scale: 32.0 2023-03-27 09:07:04,364 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-27 09:07:22,844 INFO [finetune.py:976] (3/7) Epoch 27, batch 4150, loss[loss=0.2077, simple_loss=0.2758, pruned_loss=0.0698, over 4737.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2443, pruned_loss=0.04888, over 952170.80 frames. ], batch size: 54, lr: 2.91e-03, grad_scale: 32.0 2023-03-27 09:07:23,441 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.058e+02 1.644e+02 1.926e+02 2.373e+02 3.999e+02, threshold=3.851e+02, percent-clipped=3.0 2023-03-27 09:07:31,207 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153082.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:07:51,108 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153101.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:08:07,060 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153117.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:08:13,346 INFO [finetune.py:976] (3/7) Epoch 27, batch 4200, loss[loss=0.1868, simple_loss=0.2614, pruned_loss=0.0561, over 4901.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2445, pruned_loss=0.04848, over 954168.47 frames. ], batch size: 46, lr: 2.91e-03, grad_scale: 32.0 2023-03-27 09:08:36,712 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=153149.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:08:38,015 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-27 09:08:49,826 INFO [finetune.py:976] (3/7) Epoch 27, batch 4250, loss[loss=0.1572, simple_loss=0.2301, pruned_loss=0.04212, over 4902.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2423, pruned_loss=0.04812, over 953381.47 frames. ], batch size: 46, lr: 2.91e-03, grad_scale: 32.0 2023-03-27 09:08:50,417 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.682e+01 1.570e+02 1.909e+02 2.227e+02 3.978e+02, threshold=3.818e+02, percent-clipped=1.0 2023-03-27 09:08:54,790 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153178.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:09:02,977 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5207, 1.5846, 1.9452, 1.7966, 1.7464, 3.5193, 1.5530, 1.7753], device='cuda:3'), covar=tensor([0.1027, 0.1832, 0.1058, 0.0966, 0.1588, 0.0199, 0.1484, 0.1719], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0083, 0.0074, 0.0077, 0.0091, 0.0080, 0.0086, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-27 09:09:33,254 INFO [finetune.py:976] (3/7) Epoch 27, batch 4300, loss[loss=0.1289, simple_loss=0.2089, pruned_loss=0.0244, over 4925.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.24, pruned_loss=0.04751, over 955823.86 frames. ], batch size: 33, lr: 2.91e-03, grad_scale: 32.0 2023-03-27 09:09:42,854 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7878, 1.6562, 1.4652, 1.6296, 2.1262, 2.0459, 1.7496, 1.5044], device='cuda:3'), covar=tensor([0.0351, 0.0335, 0.0649, 0.0353, 0.0205, 0.0474, 0.0303, 0.0446], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0107, 0.0149, 0.0113, 0.0102, 0.0116, 0.0104, 0.0114], device='cuda:3'), out_proj_covar=tensor([7.8541e-05, 8.1590e-05, 1.1568e-04, 8.6001e-05, 7.8876e-05, 8.5681e-05, 7.7433e-05, 8.6510e-05], device='cuda:3') 2023-03-27 09:09:44,704 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153238.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:09:49,331 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153245.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:09:49,361 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2442, 2.0554, 2.1281, 0.9573, 2.4825, 2.6985, 2.3399, 1.9871], device='cuda:3'), covar=tensor([0.0973, 0.0793, 0.0505, 0.0720, 0.0540, 0.0634, 0.0455, 0.0828], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0147, 0.0130, 0.0123, 0.0132, 0.0130, 0.0142, 0.0150], device='cuda:3'), out_proj_covar=tensor([8.9025e-05, 1.0571e-04, 9.2473e-05, 8.6261e-05, 9.2266e-05, 9.2429e-05, 1.0131e-04, 1.0739e-04], device='cuda:3') 2023-03-27 09:10:00,397 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0 2023-03-27 09:10:06,730 INFO [finetune.py:976] (3/7) Epoch 27, batch 4350, loss[loss=0.1502, simple_loss=0.2139, pruned_loss=0.04325, over 4822.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.237, pruned_loss=0.04671, over 956905.94 frames. ], batch size: 30, lr: 2.91e-03, grad_scale: 32.0 2023-03-27 09:10:07,329 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.429e+01 1.453e+02 1.753e+02 2.112e+02 4.699e+02, threshold=3.507e+02, percent-clipped=1.0 2023-03-27 09:10:09,514 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-03-27 09:10:11,663 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153278.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:10:16,465 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=153286.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:10:18,802 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4880, 1.5184, 1.8309, 1.7449, 1.6333, 3.2823, 1.4003, 1.6412], device='cuda:3'), covar=tensor([0.1001, 0.1842, 0.1075, 0.0950, 0.1590, 0.0226, 0.1505, 0.1824], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0073, 0.0077, 0.0091, 0.0080, 0.0086, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-27 09:10:21,197 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=153293.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:10:41,648 INFO [finetune.py:976] (3/7) Epoch 27, batch 4400, loss[loss=0.1396, simple_loss=0.2291, pruned_loss=0.02505, over 4818.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2395, pruned_loss=0.04786, over 957017.97 frames. ], batch size: 40, lr: 2.91e-03, grad_scale: 32.0 2023-03-27 09:11:01,181 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153339.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:11:23,046 INFO [finetune.py:976] (3/7) Epoch 27, batch 4450, loss[loss=0.2031, simple_loss=0.2749, pruned_loss=0.06571, over 4821.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2424, pruned_loss=0.04853, over 956695.21 frames. ], batch size: 40, lr: 2.91e-03, grad_scale: 32.0 2023-03-27 09:11:23,634 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.005e+02 1.494e+02 1.793e+02 2.132e+02 3.020e+02, threshold=3.586e+02, percent-clipped=0.0 2023-03-27 09:11:30,907 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153382.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:11:56,789 INFO [finetune.py:976] (3/7) Epoch 27, batch 4500, loss[loss=0.2218, simple_loss=0.2867, pruned_loss=0.07844, over 4905.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.245, pruned_loss=0.04963, over 956150.85 frames. ], batch size: 37, lr: 2.91e-03, grad_scale: 32.0 2023-03-27 09:11:57,475 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153421.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:12:03,375 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=153430.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:12:11,377 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0056, 1.5095, 2.3750, 1.5212, 1.9865, 2.1323, 1.3976, 2.2280], device='cuda:3'), covar=tensor([0.1389, 0.2313, 0.1338, 0.1904, 0.1035, 0.1586, 0.3080, 0.1126], device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0208, 0.0194, 0.0189, 0.0175, 0.0215, 0.0218, 0.0199], 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') 2023-03-27 09:12:29,935 INFO [finetune.py:976] (3/7) Epoch 27, batch 4550, loss[loss=0.1583, simple_loss=0.2387, pruned_loss=0.03895, over 4808.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2455, pruned_loss=0.04937, over 955334.56 frames. ], batch size: 41, lr: 2.91e-03, grad_scale: 32.0 2023-03-27 09:12:30,509 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.518e+01 1.590e+02 1.867e+02 2.229e+02 3.919e+02, threshold=3.734e+02, percent-clipped=1.0 2023-03-27 09:12:31,762 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153473.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:12:37,680 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153482.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:12:54,371 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2589, 2.2204, 1.8067, 2.2247, 2.1741, 1.9462, 2.4942, 2.2929], device='cuda:3'), covar=tensor([0.1219, 0.1911, 0.2760, 0.2409, 0.2342, 0.1563, 0.2951, 0.1502], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0191, 0.0237, 0.0254, 0.0250, 0.0207, 0.0215, 0.0203], 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') 2023-03-27 09:13:14,340 INFO [finetune.py:976] (3/7) Epoch 27, batch 4600, loss[loss=0.1524, simple_loss=0.2235, pruned_loss=0.04062, over 4790.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2451, pruned_loss=0.04897, over 956252.18 frames. ], batch size: 29, lr: 2.91e-03, grad_scale: 32.0 2023-03-27 09:13:33,450 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7841, 1.5344, 1.9122, 1.2718, 1.8239, 1.9413, 1.4722, 2.0874], device='cuda:3'), covar=tensor([0.1093, 0.2108, 0.1334, 0.1656, 0.0828, 0.1216, 0.2868, 0.0767], device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0209, 0.0196, 0.0191, 0.0175, 0.0216, 0.0218, 0.0200], 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') 2023-03-27 09:13:38,993 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2721, 2.2301, 1.7355, 2.2394, 2.1505, 1.8857, 2.5163, 2.2752], device='cuda:3'), covar=tensor([0.1324, 0.1939, 0.2838, 0.2530, 0.2397, 0.1683, 0.2884, 0.1614], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0191, 0.0237, 0.0254, 0.0250, 0.0207, 0.0216, 0.0203], 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') 2023-03-27 09:13:41,973 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5957, 1.0488, 0.8480, 1.4137, 2.0438, 1.0925, 1.2828, 1.4450], device='cuda:3'), covar=tensor([0.1514, 0.2333, 0.1897, 0.1243, 0.1891, 0.2052, 0.1669, 0.1955], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0093, 0.0109, 0.0091, 0.0119, 0.0093, 0.0098, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 09:13:56,975 INFO [finetune.py:976] (3/7) Epoch 27, batch 4650, loss[loss=0.1776, simple_loss=0.2462, pruned_loss=0.05448, over 4863.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2426, pruned_loss=0.04873, over 958759.12 frames. ], batch size: 34, lr: 2.91e-03, grad_scale: 32.0 2023-03-27 09:13:57,586 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.034e+02 1.461e+02 1.737e+02 2.165e+02 3.643e+02, threshold=3.475e+02, percent-clipped=0.0 2023-03-27 09:14:31,815 INFO [finetune.py:976] (3/7) Epoch 27, batch 4700, loss[loss=0.1802, simple_loss=0.2372, pruned_loss=0.0616, over 4865.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2398, pruned_loss=0.04817, over 956957.58 frames. ], batch size: 31, lr: 2.91e-03, grad_scale: 32.0 2023-03-27 09:14:47,931 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153634.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:14:53,331 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4207, 2.2644, 2.3780, 1.7684, 2.4875, 2.7682, 2.6571, 1.7242], device='cuda:3'), covar=tensor([0.0701, 0.0940, 0.0901, 0.1071, 0.0871, 0.0821, 0.0779, 0.1930], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0137, 0.0140, 0.0119, 0.0128, 0.0138, 0.0140, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 09:15:12,023 INFO [finetune.py:976] (3/7) Epoch 27, batch 4750, loss[loss=0.1994, simple_loss=0.2568, pruned_loss=0.07102, over 4826.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2382, pruned_loss=0.04802, over 958134.84 frames. ], batch size: 30, lr: 2.91e-03, grad_scale: 32.0 2023-03-27 09:15:13,078 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.023e+02 1.473e+02 1.795e+02 2.173e+02 4.465e+02, threshold=3.590e+02, percent-clipped=3.0 2023-03-27 09:15:28,192 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8578, 1.7634, 1.5398, 1.4576, 1.8819, 1.6377, 1.8605, 1.8481], device='cuda:3'), covar=tensor([0.1363, 0.1772, 0.2791, 0.2357, 0.2398, 0.1605, 0.2672, 0.1559], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0191, 0.0237, 0.0255, 0.0251, 0.0208, 0.0217, 0.0203], 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') 2023-03-27 09:15:45,873 INFO [finetune.py:976] (3/7) Epoch 27, batch 4800, loss[loss=0.1929, simple_loss=0.2793, pruned_loss=0.05324, over 4807.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.241, pruned_loss=0.0487, over 954358.23 frames. ], batch size: 45, lr: 2.91e-03, grad_scale: 32.0 2023-03-27 09:16:28,411 INFO [finetune.py:976] (3/7) Epoch 27, batch 4850, loss[loss=0.1954, simple_loss=0.2723, pruned_loss=0.05923, over 4899.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2448, pruned_loss=0.04942, over 955047.83 frames. ], batch size: 43, lr: 2.91e-03, grad_scale: 32.0 2023-03-27 09:16:28,977 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.100e+02 1.544e+02 1.777e+02 2.223e+02 4.381e+02, threshold=3.554e+02, percent-clipped=2.0 2023-03-27 09:16:30,767 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153773.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:16:33,638 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153777.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:16:37,258 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1338, 1.8354, 1.4244, 0.5627, 1.6052, 1.7002, 1.3493, 1.7261], device='cuda:3'), covar=tensor([0.0837, 0.0989, 0.1872, 0.2412, 0.1607, 0.2413, 0.2819, 0.1002], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0191, 0.0201, 0.0182, 0.0210, 0.0210, 0.0223, 0.0196], 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') 2023-03-27 09:17:00,915 INFO [finetune.py:976] (3/7) Epoch 27, batch 4900, loss[loss=0.1985, simple_loss=0.2674, pruned_loss=0.06477, over 4829.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2464, pruned_loss=0.04984, over 955994.26 frames. ], batch size: 30, lr: 2.91e-03, grad_scale: 32.0 2023-03-27 09:17:01,616 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=153821.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:17:20,895 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-03-27 09:17:34,600 INFO [finetune.py:976] (3/7) Epoch 27, batch 4950, loss[loss=0.2507, simple_loss=0.3133, pruned_loss=0.09408, over 4819.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2477, pruned_loss=0.0502, over 954287.53 frames. ], batch size: 38, lr: 2.90e-03, grad_scale: 32.0 2023-03-27 09:17:35,195 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.062e+02 1.496e+02 1.750e+02 2.158e+02 3.393e+02, threshold=3.501e+02, percent-clipped=0.0 2023-03-27 09:18:10,031 INFO [finetune.py:976] (3/7) Epoch 27, batch 5000, loss[loss=0.1462, simple_loss=0.2201, pruned_loss=0.03616, over 4770.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2454, pruned_loss=0.04957, over 954656.51 frames. ], batch size: 26, lr: 2.90e-03, grad_scale: 32.0 2023-03-27 09:18:12,427 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153923.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:18:27,322 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-27 09:18:28,727 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153934.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:19:01,714 INFO [finetune.py:976] (3/7) Epoch 27, batch 5050, loss[loss=0.1433, simple_loss=0.2162, pruned_loss=0.03523, over 4876.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2412, pruned_loss=0.04806, over 955804.88 frames. ], batch size: 34, lr: 2.90e-03, grad_scale: 32.0 2023-03-27 09:19:02,311 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.210e+01 1.435e+02 1.808e+02 2.168e+02 4.775e+02, threshold=3.617e+02, percent-clipped=1.0 2023-03-27 09:19:07,836 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0130, 2.0590, 1.6058, 1.8997, 1.9244, 1.8728, 1.9457, 2.5308], device='cuda:3'), covar=tensor([0.3802, 0.3570, 0.3303, 0.3887, 0.3909, 0.2526, 0.3433, 0.1849], device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0265, 0.0238, 0.0276, 0.0262, 0.0231, 0.0260, 0.0240], 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') 2023-03-27 09:19:10,569 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=153982.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:19:11,851 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153984.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:19:36,570 INFO [finetune.py:976] (3/7) Epoch 27, batch 5100, loss[loss=0.1861, simple_loss=0.2469, pruned_loss=0.06266, over 4871.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2391, pruned_loss=0.04741, over 957277.05 frames. ], batch size: 49, lr: 2.90e-03, grad_scale: 32.0 2023-03-27 09:20:06,446 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154049.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 09:20:19,664 INFO [finetune.py:976] (3/7) Epoch 27, batch 5150, loss[loss=0.1709, simple_loss=0.2524, pruned_loss=0.04474, over 4836.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2398, pruned_loss=0.04754, over 956584.63 frames. ], batch size: 30, lr: 2.90e-03, grad_scale: 32.0 2023-03-27 09:20:20,255 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.883e+01 1.500e+02 1.789e+02 2.109e+02 4.792e+02, threshold=3.578e+02, percent-clipped=3.0 2023-03-27 09:20:23,493 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154076.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:20:24,067 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154077.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:20:28,828 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154084.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:20:29,475 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-03-27 09:20:39,110 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7768, 1.7501, 1.5436, 1.9264, 2.1492, 1.8827, 1.6064, 1.5143], device='cuda:3'), covar=tensor([0.2291, 0.2013, 0.2012, 0.1788, 0.1556, 0.1243, 0.2338, 0.2039], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0211, 0.0215, 0.0201, 0.0246, 0.0191, 0.0218, 0.0205], 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') 2023-03-27 09:20:46,968 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154110.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 09:20:50,240 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-03-27 09:20:53,305 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=5.08 vs. limit=5.0 2023-03-27 09:20:53,422 INFO [finetune.py:976] (3/7) Epoch 27, batch 5200, loss[loss=0.1605, simple_loss=0.238, pruned_loss=0.0415, over 4872.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2426, pruned_loss=0.0482, over 956132.76 frames. ], batch size: 34, lr: 2.90e-03, grad_scale: 32.0 2023-03-27 09:20:55,327 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.7043, 1.6889, 1.7285, 1.0612, 1.8422, 2.1576, 2.0533, 1.5386], device='cuda:3'), covar=tensor([0.0850, 0.0686, 0.0499, 0.0536, 0.0430, 0.0599, 0.0317, 0.0617], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0147, 0.0129, 0.0122, 0.0131, 0.0130, 0.0142, 0.0150], device='cuda:3'), out_proj_covar=tensor([8.8232e-05, 1.0526e-04, 9.1810e-05, 8.5690e-05, 9.1681e-05, 9.2316e-05, 1.0065e-04, 1.0710e-04], device='cuda:3') 2023-03-27 09:20:55,349 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2729, 2.1043, 1.8648, 2.1192, 2.0327, 2.0448, 2.0644, 2.8123], device='cuda:3'), covar=tensor([0.3527, 0.4505, 0.3212, 0.4016, 0.4125, 0.2421, 0.3814, 0.1676], device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0266, 0.0238, 0.0277, 0.0263, 0.0231, 0.0260, 0.0240], 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') 2023-03-27 09:20:56,441 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=154125.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:21:04,333 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154137.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 09:21:12,347 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154145.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:21:34,726 INFO [finetune.py:976] (3/7) Epoch 27, batch 5250, loss[loss=0.183, simple_loss=0.255, pruned_loss=0.0555, over 4895.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2447, pruned_loss=0.04916, over 956431.49 frames. ], batch size: 36, lr: 2.90e-03, grad_scale: 32.0 2023-03-27 09:21:35,333 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.122e+02 1.556e+02 1.889e+02 2.346e+02 3.556e+02, threshold=3.778e+02, percent-clipped=0.0 2023-03-27 09:22:08,469 INFO [finetune.py:976] (3/7) Epoch 27, batch 5300, loss[loss=0.1756, simple_loss=0.2527, pruned_loss=0.0492, over 4192.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2458, pruned_loss=0.04925, over 956238.88 frames. ], batch size: 65, lr: 2.90e-03, grad_scale: 32.0 2023-03-27 09:22:09,173 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154221.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:22:41,917 INFO [finetune.py:976] (3/7) Epoch 27, batch 5350, loss[loss=0.1327, simple_loss=0.2047, pruned_loss=0.03036, over 4817.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2459, pruned_loss=0.04918, over 955813.62 frames. ], batch size: 30, lr: 2.90e-03, grad_scale: 32.0 2023-03-27 09:22:42,511 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.999e+01 1.513e+02 1.798e+02 2.139e+02 3.270e+02, threshold=3.596e+02, percent-clipped=0.0 2023-03-27 09:22:47,877 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154279.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:22:49,759 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154282.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 09:22:52,193 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154286.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:22:55,592 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154291.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:23:00,765 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8916, 1.8194, 1.7054, 1.8604, 1.4756, 4.0413, 1.6012, 2.0127], device='cuda:3'), covar=tensor([0.2869, 0.2263, 0.1958, 0.2064, 0.1446, 0.0165, 0.2409, 0.1127], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0120, 0.0124, 0.0113, 0.0095, 0.0094, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 09:23:15,341 INFO [finetune.py:976] (3/7) Epoch 27, batch 5400, loss[loss=0.1364, simple_loss=0.2079, pruned_loss=0.03247, over 4912.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2438, pruned_loss=0.04897, over 956732.26 frames. ], batch size: 37, lr: 2.90e-03, grad_scale: 32.0 2023-03-27 09:23:33,027 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154337.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 09:23:36,096 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9005, 1.3467, 1.9782, 1.8524, 1.6854, 1.6677, 1.8347, 1.8920], device='cuda:3'), covar=tensor([0.3899, 0.4261, 0.3352, 0.4019, 0.5045, 0.3925, 0.4344, 0.2987], device='cuda:3'), in_proj_covar=tensor([0.0266, 0.0248, 0.0270, 0.0298, 0.0297, 0.0274, 0.0302, 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') 2023-03-27 09:23:42,751 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154347.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:23:49,149 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154352.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:24:08,869 INFO [finetune.py:976] (3/7) Epoch 27, batch 5450, loss[loss=0.1368, simple_loss=0.2059, pruned_loss=0.03381, over 4757.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.24, pruned_loss=0.04763, over 957195.88 frames. ], batch size: 27, lr: 2.90e-03, grad_scale: 32.0 2023-03-27 09:24:09,470 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.752e+01 1.439e+02 1.730e+02 2.063e+02 4.741e+02, threshold=3.460e+02, percent-clipped=1.0 2023-03-27 09:24:14,998 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5748, 1.5181, 1.9173, 2.9955, 2.0760, 2.2402, 1.0284, 2.5602], device='cuda:3'), covar=tensor([0.1676, 0.1412, 0.1221, 0.0670, 0.0779, 0.1199, 0.1770, 0.0469], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0116, 0.0132, 0.0164, 0.0100, 0.0135, 0.0124, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 09:24:18,891 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-03-27 09:24:26,429 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154398.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 09:24:32,220 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154405.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 09:24:37,961 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3224, 2.2046, 1.9243, 2.1976, 2.0732, 2.1075, 2.1148, 2.8582], device='cuda:3'), covar=tensor([0.3544, 0.4339, 0.3246, 0.3904, 0.4165, 0.2434, 0.3828, 0.1584], device='cuda:3'), in_proj_covar=tensor([0.0291, 0.0265, 0.0238, 0.0276, 0.0262, 0.0231, 0.0260, 0.0239], 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') 2023-03-27 09:24:42,541 INFO [finetune.py:976] (3/7) Epoch 27, batch 5500, loss[loss=0.1471, simple_loss=0.2241, pruned_loss=0.03502, over 4907.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2372, pruned_loss=0.04687, over 954883.99 frames. ], batch size: 36, lr: 2.90e-03, grad_scale: 32.0 2023-03-27 09:24:49,892 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154432.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 09:24:54,771 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154440.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:25:26,806 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.38 vs. limit=5.0 2023-03-27 09:25:27,012 INFO [finetune.py:976] (3/7) Epoch 27, batch 5550, loss[loss=0.1786, simple_loss=0.2724, pruned_loss=0.04237, over 4924.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2406, pruned_loss=0.04837, over 953860.43 frames. ], batch size: 33, lr: 2.90e-03, grad_scale: 32.0 2023-03-27 09:25:27,597 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.282e+01 1.501e+02 1.802e+02 2.038e+02 5.335e+02, threshold=3.603e+02, percent-clipped=2.0 2023-03-27 09:25:29,541 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154474.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:25:54,655 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4083, 1.2960, 1.5589, 1.6035, 1.4472, 2.8652, 1.1480, 1.4111], device='cuda:3'), covar=tensor([0.1036, 0.1892, 0.1214, 0.0923, 0.1659, 0.0279, 0.1653, 0.1857], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0083, 0.0073, 0.0077, 0.0092, 0.0080, 0.0086, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-27 09:25:57,467 INFO [finetune.py:976] (3/7) Epoch 27, batch 5600, loss[loss=0.1508, simple_loss=0.2302, pruned_loss=0.03565, over 4819.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2431, pruned_loss=0.04906, over 950176.37 frames. ], batch size: 33, lr: 2.90e-03, grad_scale: 32.0 2023-03-27 09:26:07,318 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154535.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:26:23,197 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 2023-03-27 09:26:30,304 INFO [finetune.py:976] (3/7) Epoch 27, batch 5650, loss[loss=0.1929, simple_loss=0.2632, pruned_loss=0.06129, over 4899.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2445, pruned_loss=0.04869, over 951180.21 frames. ], batch size: 36, lr: 2.90e-03, grad_scale: 32.0 2023-03-27 09:26:30,863 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.301e+01 1.398e+02 1.740e+02 2.119e+02 4.723e+02, threshold=3.480e+02, percent-clipped=2.0 2023-03-27 09:26:39,109 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154577.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 09:26:40,303 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154579.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:27:03,428 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 2023-03-27 09:27:07,858 INFO [finetune.py:976] (3/7) Epoch 27, batch 5700, loss[loss=0.1197, simple_loss=0.1806, pruned_loss=0.02944, over 4044.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2394, pruned_loss=0.04791, over 929838.81 frames. ], batch size: 17, lr: 2.90e-03, grad_scale: 32.0 2023-03-27 09:27:12,036 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=154627.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:27:12,683 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154628.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:27:20,864 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154642.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:27:34,185 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154647.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:27:34,737 INFO [finetune.py:976] (3/7) Epoch 28, batch 0, loss[loss=0.1486, simple_loss=0.2315, pruned_loss=0.0329, over 4878.00 frames. ], tot_loss[loss=0.1486, simple_loss=0.2315, pruned_loss=0.0329, over 4878.00 frames. ], batch size: 43, lr: 2.90e-03, grad_scale: 32.0 2023-03-27 09:27:34,737 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-27 09:27:44,985 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.1758, 1.4781, 1.5189, 0.7571, 1.5299, 1.6889, 1.7463, 1.4045], device='cuda:3'), covar=tensor([0.1152, 0.0716, 0.0702, 0.0601, 0.0628, 0.0830, 0.0379, 0.0865], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0147, 0.0129, 0.0121, 0.0131, 0.0130, 0.0142, 0.0150], device='cuda:3'), out_proj_covar=tensor([8.8337e-05, 1.0504e-04, 9.1824e-05, 8.5222e-05, 9.1862e-05, 9.1862e-05, 1.0083e-04, 1.0708e-04], device='cuda:3') 2023-03-27 09:27:54,286 INFO [finetune.py:1010] (3/7) Epoch 28, validation: loss=0.1583, simple_loss=0.2265, pruned_loss=0.04511, over 2265189.00 frames. 2023-03-27 09:27:54,287 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6469MB 2023-03-27 09:27:55,412 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0800, 1.3586, 0.8544, 2.0321, 2.4086, 1.9410, 1.7337, 1.8228], device='cuda:3'), covar=tensor([0.1347, 0.1981, 0.1888, 0.1084, 0.1759, 0.1822, 0.1258, 0.1871], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0093, 0.0109, 0.0091, 0.0119, 0.0093, 0.0098, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 09:27:59,642 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4321, 1.2677, 1.7827, 2.7468, 1.8010, 2.2951, 0.9206, 2.4551], device='cuda:3'), covar=tensor([0.1913, 0.2114, 0.1536, 0.1065, 0.1100, 0.1268, 0.2245, 0.0701], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0132, 0.0163, 0.0100, 0.0135, 0.0124, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 09:28:08,716 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.011e+02 1.488e+02 1.773e+02 2.221e+02 3.199e+02, threshold=3.546e+02, percent-clipped=0.0 2023-03-27 09:28:19,996 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3067, 2.2718, 2.4023, 1.5725, 2.2857, 2.4303, 2.4341, 1.9161], device='cuda:3'), covar=tensor([0.0590, 0.0586, 0.0647, 0.0958, 0.0628, 0.0656, 0.0560, 0.1028], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0138, 0.0141, 0.0120, 0.0129, 0.0140, 0.0140, 0.0164], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 09:28:21,213 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154689.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:28:24,051 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154693.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 09:28:27,013 INFO [finetune.py:976] (3/7) Epoch 28, batch 50, loss[loss=0.1905, simple_loss=0.2594, pruned_loss=0.0608, over 4847.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2482, pruned_loss=0.05144, over 216497.44 frames. ], batch size: 44, lr: 2.90e-03, grad_scale: 32.0 2023-03-27 09:28:32,659 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154705.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 09:28:35,795 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1897, 2.0782, 1.9346, 2.2035, 1.9693, 2.0443, 1.9787, 2.6538], device='cuda:3'), covar=tensor([0.3089, 0.3828, 0.2899, 0.3543, 0.4079, 0.2160, 0.3623, 0.1454], device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0263, 0.0236, 0.0274, 0.0260, 0.0228, 0.0258, 0.0237], 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') 2023-03-27 09:28:41,928 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154720.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:28:52,104 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154732.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:28:57,933 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154740.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:29:03,129 INFO [finetune.py:976] (3/7) Epoch 28, batch 100, loss[loss=0.1461, simple_loss=0.22, pruned_loss=0.03611, over 4902.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.242, pruned_loss=0.04935, over 380484.75 frames. ], batch size: 36, lr: 2.90e-03, grad_scale: 32.0 2023-03-27 09:29:10,681 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7282, 3.9613, 3.7491, 1.9173, 4.0480, 3.0302, 1.1779, 2.8790], device='cuda:3'), covar=tensor([0.2135, 0.1926, 0.1499, 0.3404, 0.1055, 0.1058, 0.4049, 0.1461], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0179, 0.0159, 0.0130, 0.0161, 0.0123, 0.0148, 0.0124], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 09:29:11,913 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=154753.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 09:29:13,665 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2939, 2.1626, 2.3567, 1.5321, 2.3585, 2.3428, 2.2999, 1.8732], device='cuda:3'), covar=tensor([0.0465, 0.0527, 0.0510, 0.0750, 0.0689, 0.0512, 0.0453, 0.1066], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0138, 0.0141, 0.0120, 0.0128, 0.0139, 0.0140, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 09:29:26,905 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.078e+02 1.413e+02 1.713e+02 2.093e+02 4.180e+02, threshold=3.426e+02, percent-clipped=2.0 2023-03-27 09:29:32,447 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=154780.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:29:33,111 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154781.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:29:37,834 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=154788.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:29:44,901 INFO [finetune.py:976] (3/7) Epoch 28, batch 150, loss[loss=0.1655, simple_loss=0.2346, pruned_loss=0.0482, over 4937.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2372, pruned_loss=0.04833, over 508769.63 frames. ], batch size: 38, lr: 2.90e-03, grad_scale: 32.0 2023-03-27 09:29:50,533 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-03-27 09:29:56,987 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8479, 1.8417, 1.5966, 2.0261, 2.3835, 2.0545, 1.7338, 1.5480], device='cuda:3'), covar=tensor([0.2144, 0.1874, 0.1886, 0.1460, 0.1595, 0.1143, 0.2194, 0.1872], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0212, 0.0216, 0.0200, 0.0246, 0.0192, 0.0218, 0.0205], 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') 2023-03-27 09:29:57,853 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-03-27 09:29:59,446 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0 2023-03-27 09:30:06,035 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154830.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:30:18,389 INFO [finetune.py:976] (3/7) Epoch 28, batch 200, loss[loss=0.157, simple_loss=0.2313, pruned_loss=0.04139, over 4905.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.2359, pruned_loss=0.04826, over 608685.95 frames. ], batch size: 36, lr: 2.89e-03, grad_scale: 32.0 2023-03-27 09:30:40,608 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.190e+01 1.565e+02 1.831e+02 2.234e+02 3.641e+02, threshold=3.662e+02, percent-clipped=1.0 2023-03-27 09:30:48,051 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154877.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 09:31:02,680 INFO [finetune.py:976] (3/7) Epoch 28, batch 250, loss[loss=0.1397, simple_loss=0.216, pruned_loss=0.0317, over 4780.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2395, pruned_loss=0.04953, over 685101.04 frames. ], batch size: 26, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:31:14,954 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.74 vs. limit=5.0 2023-03-27 09:31:20,578 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=154925.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:31:31,424 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154942.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:31:34,983 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154947.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:31:35,502 INFO [finetune.py:976] (3/7) Epoch 28, batch 300, loss[loss=0.1856, simple_loss=0.2627, pruned_loss=0.05422, over 4854.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2438, pruned_loss=0.05028, over 743538.17 frames. ], batch size: 44, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:31:42,230 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3150, 1.8597, 1.9179, 0.9271, 2.3018, 2.2695, 2.1425, 1.7444], device='cuda:3'), covar=tensor([0.0864, 0.0753, 0.0595, 0.0671, 0.0510, 0.0740, 0.0484, 0.0732], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0147, 0.0129, 0.0122, 0.0131, 0.0130, 0.0142, 0.0150], device='cuda:3'), out_proj_covar=tensor([8.8628e-05, 1.0545e-04, 9.1969e-05, 8.5533e-05, 9.1986e-05, 9.1993e-05, 1.0103e-04, 1.0706e-04], device='cuda:3') 2023-03-27 09:31:51,473 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.026e+02 1.523e+02 1.869e+02 2.212e+02 3.864e+02, threshold=3.739e+02, percent-clipped=1.0 2023-03-27 09:31:59,670 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154976.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:32:07,904 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154984.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:32:11,536 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=154990.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:32:13,403 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154993.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 09:32:15,017 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=154995.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:32:16,795 INFO [finetune.py:976] (3/7) Epoch 28, batch 350, loss[loss=0.2474, simple_loss=0.312, pruned_loss=0.0914, over 4202.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2473, pruned_loss=0.05156, over 787330.78 frames. ], batch size: 65, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:32:28,691 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0606, 2.0356, 1.6487, 1.8694, 1.9982, 1.7426, 2.2121, 2.0512], device='cuda:3'), covar=tensor([0.1340, 0.1888, 0.2794, 0.2455, 0.2592, 0.1679, 0.2771, 0.1616], device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0190, 0.0236, 0.0254, 0.0250, 0.0207, 0.0215, 0.0202], 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') 2023-03-27 09:32:31,991 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.2843, 3.6979, 3.9374, 4.1339, 4.0834, 3.7914, 4.3799, 1.4030], device='cuda:3'), covar=tensor([0.0797, 0.0874, 0.0764, 0.0876, 0.1123, 0.1518, 0.0672, 0.5482], device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0248, 0.0285, 0.0297, 0.0336, 0.0288, 0.0306, 0.0304], 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') 2023-03-27 09:32:43,051 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155037.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:32:45,383 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=155041.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 09:32:49,880 INFO [finetune.py:976] (3/7) Epoch 28, batch 400, loss[loss=0.1712, simple_loss=0.2452, pruned_loss=0.04855, over 4795.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2475, pruned_loss=0.0513, over 824439.10 frames. ], batch size: 45, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:33:13,472 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155069.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:33:15,062 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.906e+01 1.559e+02 1.879e+02 2.352e+02 4.263e+02, threshold=3.758e+02, percent-clipped=3.0 2023-03-27 09:33:18,283 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155076.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:33:19,631 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.91 vs. limit=5.0 2023-03-27 09:33:31,495 INFO [finetune.py:976] (3/7) Epoch 28, batch 450, loss[loss=0.1384, simple_loss=0.2171, pruned_loss=0.02985, over 4861.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2455, pruned_loss=0.05033, over 853253.84 frames. ], batch size: 34, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:33:54,284 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155130.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:33:54,328 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155130.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:34:03,764 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.79 vs. limit=5.0 2023-03-27 09:34:05,151 INFO [finetune.py:976] (3/7) Epoch 28, batch 500, loss[loss=0.18, simple_loss=0.2474, pruned_loss=0.05625, over 4818.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2423, pruned_loss=0.04901, over 876735.89 frames. ], batch size: 33, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:34:28,553 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.687e+01 1.475e+02 1.683e+02 2.204e+02 4.497e+02, threshold=3.366e+02, percent-clipped=1.0 2023-03-27 09:34:37,156 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=155178.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:34:49,246 INFO [finetune.py:976] (3/7) Epoch 28, batch 550, loss[loss=0.1609, simple_loss=0.2243, pruned_loss=0.04878, over 4910.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2382, pruned_loss=0.04772, over 895381.71 frames. ], batch size: 35, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:35:01,235 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7999, 2.5265, 2.1901, 1.0224, 2.3400, 2.1533, 1.9600, 2.2980], device='cuda:3'), covar=tensor([0.0860, 0.0888, 0.1711, 0.2194, 0.1393, 0.2194, 0.2220, 0.0982], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0192, 0.0202, 0.0182, 0.0210, 0.0210, 0.0224, 0.0197], 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') 2023-03-27 09:35:11,080 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2619, 2.1564, 1.8814, 1.9312, 2.2629, 2.0171, 2.3011, 2.2460], device='cuda:3'), covar=tensor([0.1333, 0.2009, 0.2878, 0.2533, 0.2444, 0.1744, 0.2291, 0.1766], device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0189, 0.0235, 0.0253, 0.0249, 0.0206, 0.0214, 0.0201], 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') 2023-03-27 09:35:14,009 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5128, 2.0490, 2.7366, 2.0017, 2.5388, 2.8425, 1.8020, 2.7883], device='cuda:3'), covar=tensor([0.1477, 0.2364, 0.1700, 0.1962, 0.1004, 0.1313, 0.3247, 0.1029], device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0208, 0.0194, 0.0189, 0.0174, 0.0213, 0.0219, 0.0198], 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') 2023-03-27 09:35:23,081 INFO [finetune.py:976] (3/7) Epoch 28, batch 600, loss[loss=0.178, simple_loss=0.2439, pruned_loss=0.05604, over 4910.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2404, pruned_loss=0.04911, over 907617.54 frames. ], batch size: 43, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:35:38,977 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.640e+01 1.454e+02 1.702e+02 1.998e+02 4.828e+02, threshold=3.403e+02, percent-clipped=2.0 2023-03-27 09:35:53,266 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155284.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:36:05,099 INFO [finetune.py:976] (3/7) Epoch 28, batch 650, loss[loss=0.1322, simple_loss=0.1975, pruned_loss=0.03345, over 4791.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2421, pruned_loss=0.04905, over 916895.83 frames. ], batch size: 25, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:36:28,001 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9059, 1.6682, 1.6109, 1.2711, 1.7317, 1.7053, 1.7206, 2.2354], device='cuda:3'), covar=tensor([0.3416, 0.3035, 0.2811, 0.3099, 0.3116, 0.2073, 0.2629, 0.1577], device='cuda:3'), in_proj_covar=tensor([0.0291, 0.0265, 0.0238, 0.0277, 0.0262, 0.0230, 0.0260, 0.0240], 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') 2023-03-27 09:36:29,093 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=155332.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:36:29,103 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155332.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:36:38,732 INFO [finetune.py:976] (3/7) Epoch 28, batch 700, loss[loss=0.1164, simple_loss=0.1871, pruned_loss=0.02284, over 4741.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2433, pruned_loss=0.0493, over 926473.03 frames. ], batch size: 23, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:36:54,658 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.124e+02 1.563e+02 1.812e+02 2.261e+02 4.160e+02, threshold=3.625e+02, percent-clipped=3.0 2023-03-27 09:36:57,795 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155376.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:37:05,078 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.9108, 4.2792, 4.5179, 4.7706, 4.6704, 4.3247, 5.0182, 1.5750], device='cuda:3'), covar=tensor([0.0719, 0.0811, 0.0750, 0.0860, 0.1101, 0.1645, 0.0549, 0.6171], device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0247, 0.0284, 0.0297, 0.0335, 0.0287, 0.0305, 0.0302], 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') 2023-03-27 09:37:19,493 INFO [finetune.py:976] (3/7) Epoch 28, batch 750, loss[loss=0.1579, simple_loss=0.233, pruned_loss=0.04141, over 4919.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2447, pruned_loss=0.05006, over 934492.47 frames. ], batch size: 33, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:37:40,158 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=155424.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:37:41,308 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155425.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:37:49,653 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6525, 1.5590, 1.5375, 1.6168, 1.1981, 3.4924, 1.2547, 1.6430], device='cuda:3'), covar=tensor([0.3246, 0.2514, 0.2161, 0.2322, 0.1695, 0.0212, 0.2630, 0.1280], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0121, 0.0124, 0.0113, 0.0095, 0.0094, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 09:37:56,713 INFO [finetune.py:976] (3/7) Epoch 28, batch 800, loss[loss=0.1623, simple_loss=0.2434, pruned_loss=0.04059, over 4793.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2452, pruned_loss=0.04982, over 940442.89 frames. ], batch size: 25, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:38:02,347 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155457.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:38:11,704 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-03-27 09:38:11,935 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.111e+02 1.495e+02 1.724e+02 1.967e+02 3.002e+02, threshold=3.447e+02, percent-clipped=0.0 2023-03-27 09:38:39,879 INFO [finetune.py:976] (3/7) Epoch 28, batch 850, loss[loss=0.1795, simple_loss=0.2476, pruned_loss=0.0557, over 4818.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2437, pruned_loss=0.04911, over 946502.73 frames. ], batch size: 25, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:38:52,665 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155518.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:39:03,294 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9258, 1.0646, 1.8432, 1.8571, 1.6787, 1.6498, 1.7468, 1.8341], device='cuda:3'), covar=tensor([0.3682, 0.3787, 0.3355, 0.3498, 0.4667, 0.3814, 0.3918, 0.2957], device='cuda:3'), in_proj_covar=tensor([0.0266, 0.0248, 0.0268, 0.0297, 0.0296, 0.0273, 0.0302, 0.0252], 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') 2023-03-27 09:39:07,980 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2179, 2.2047, 1.9350, 2.3411, 2.2885, 2.0882, 2.4214, 2.3213], device='cuda:3'), covar=tensor([0.1185, 0.1972, 0.2481, 0.2071, 0.1959, 0.1337, 0.2722, 0.1450], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0190, 0.0236, 0.0255, 0.0249, 0.0207, 0.0215, 0.0203], 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') 2023-03-27 09:39:09,205 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6673, 2.5891, 2.2040, 2.8935, 2.6821, 2.4031, 3.1211, 2.7067], device='cuda:3'), covar=tensor([0.1375, 0.2207, 0.3068, 0.2574, 0.2500, 0.1644, 0.2894, 0.1835], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0190, 0.0236, 0.0255, 0.0249, 0.0207, 0.0215, 0.0203], 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') 2023-03-27 09:39:13,738 INFO [finetune.py:976] (3/7) Epoch 28, batch 900, loss[loss=0.188, simple_loss=0.2554, pruned_loss=0.06026, over 4826.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2412, pruned_loss=0.04854, over 947355.61 frames. ], batch size: 51, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:39:28,252 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.783e+01 1.413e+02 1.787e+02 2.288e+02 4.282e+02, threshold=3.575e+02, percent-clipped=3.0 2023-03-27 09:39:28,427 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8239, 1.0778, 1.8949, 1.8699, 1.6917, 1.6620, 1.7029, 1.8746], device='cuda:3'), covar=tensor([0.3691, 0.3856, 0.3187, 0.3375, 0.4812, 0.3635, 0.4092, 0.2812], device='cuda:3'), in_proj_covar=tensor([0.0267, 0.0248, 0.0269, 0.0298, 0.0296, 0.0273, 0.0302, 0.0252], 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') 2023-03-27 09:39:35,832 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-03-27 09:39:54,476 INFO [finetune.py:976] (3/7) Epoch 28, batch 950, loss[loss=0.2241, simple_loss=0.2893, pruned_loss=0.07941, over 4912.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2405, pruned_loss=0.04872, over 948452.39 frames. ], batch size: 46, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:40:02,629 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155605.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:40:15,282 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1407, 1.9763, 1.6125, 0.5884, 1.7377, 1.7985, 1.6075, 1.7720], device='cuda:3'), covar=tensor([0.0892, 0.0749, 0.1232, 0.1807, 0.1081, 0.2029, 0.2295, 0.0817], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0192, 0.0201, 0.0182, 0.0210, 0.0211, 0.0224, 0.0197], 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') 2023-03-27 09:40:20,529 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155632.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:40:31,723 INFO [finetune.py:976] (3/7) Epoch 28, batch 1000, loss[loss=0.1155, simple_loss=0.1896, pruned_loss=0.02073, over 4739.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2444, pruned_loss=0.05016, over 949232.51 frames. ], batch size: 23, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:40:37,310 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155657.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:40:42,776 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155666.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:40:45,714 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.619e+01 1.519e+02 1.814e+02 2.190e+02 3.109e+02, threshold=3.628e+02, percent-clipped=0.0 2023-03-27 09:40:54,173 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=155680.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:41:13,974 INFO [finetune.py:976] (3/7) Epoch 28, batch 1050, loss[loss=0.2045, simple_loss=0.2782, pruned_loss=0.06547, over 4829.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2469, pruned_loss=0.05024, over 950240.24 frames. ], batch size: 47, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:41:26,782 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155718.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:41:29,279 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-03-27 09:41:31,009 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155725.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:41:35,924 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.64 vs. limit=2.0 2023-03-27 09:41:45,039 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1229, 2.0073, 1.6454, 2.1709, 2.6660, 2.1726, 2.3345, 1.5660], device='cuda:3'), covar=tensor([0.2058, 0.1845, 0.1894, 0.1467, 0.1636, 0.1093, 0.1755, 0.1853], device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0213, 0.0217, 0.0201, 0.0248, 0.0192, 0.0218, 0.0206], 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') 2023-03-27 09:41:46,713 INFO [finetune.py:976] (3/7) Epoch 28, batch 1100, loss[loss=0.1713, simple_loss=0.2508, pruned_loss=0.04592, over 4897.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.247, pruned_loss=0.05014, over 951241.28 frames. ], batch size: 37, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:41:49,475 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-03-27 09:41:56,301 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4689, 2.3349, 2.1247, 1.0840, 2.2629, 1.9151, 1.7995, 2.2087], device='cuda:3'), covar=tensor([0.1035, 0.0856, 0.1584, 0.2205, 0.1404, 0.2350, 0.2166, 0.1011], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0192, 0.0202, 0.0182, 0.0210, 0.0212, 0.0225, 0.0197], 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') 2023-03-27 09:42:01,622 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.104e+02 1.629e+02 1.915e+02 2.256e+02 9.973e+02, threshold=3.830e+02, percent-clipped=2.0 2023-03-27 09:42:02,899 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=155773.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:42:10,612 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155784.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:42:12,861 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-03-27 09:42:21,324 INFO [finetune.py:976] (3/7) Epoch 28, batch 1150, loss[loss=0.163, simple_loss=0.2432, pruned_loss=0.04144, over 4757.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2471, pruned_loss=0.04974, over 954221.32 frames. ], batch size: 54, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:42:23,726 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-03-27 09:42:39,765 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155813.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:42:55,939 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3605, 2.3686, 2.0319, 1.0154, 2.1336, 1.9431, 1.7904, 2.1653], device='cuda:3'), covar=tensor([0.0964, 0.0649, 0.1493, 0.1949, 0.1166, 0.2135, 0.2076, 0.0879], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0191, 0.0201, 0.0181, 0.0210, 0.0211, 0.0224, 0.0197], 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') 2023-03-27 09:43:00,221 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-27 09:43:01,194 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155845.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 09:43:02,855 INFO [finetune.py:976] (3/7) Epoch 28, batch 1200, loss[loss=0.1848, simple_loss=0.2528, pruned_loss=0.05844, over 4900.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2441, pruned_loss=0.04893, over 954102.01 frames. ], batch size: 37, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:43:18,194 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.094e+02 1.526e+02 1.746e+02 2.167e+02 3.236e+02, threshold=3.492e+02, percent-clipped=0.0 2023-03-27 09:43:26,155 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.6608, 1.3880, 1.4241, 0.8138, 1.5449, 1.8073, 1.6658, 1.3816], device='cuda:3'), covar=tensor([0.0931, 0.0921, 0.0614, 0.0631, 0.0540, 0.0696, 0.0479, 0.0847], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0147, 0.0130, 0.0123, 0.0132, 0.0131, 0.0143, 0.0150], device='cuda:3'), out_proj_covar=tensor([8.8845e-05, 1.0563e-04, 9.2553e-05, 8.6181e-05, 9.2435e-05, 9.2454e-05, 1.0168e-04, 1.0740e-04], device='cuda:3') 2023-03-27 09:43:45,613 INFO [finetune.py:976] (3/7) Epoch 28, batch 1250, loss[loss=0.1359, simple_loss=0.2063, pruned_loss=0.03269, over 4834.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2414, pruned_loss=0.04832, over 952862.79 frames. ], batch size: 30, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:43:58,869 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.9892, 4.3723, 4.5594, 4.8211, 4.7461, 4.4206, 5.1169, 1.6318], device='cuda:3'), covar=tensor([0.0800, 0.0871, 0.0817, 0.0984, 0.1229, 0.1721, 0.0572, 0.6214], device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0247, 0.0285, 0.0297, 0.0335, 0.0289, 0.0306, 0.0303], 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') 2023-03-27 09:44:22,108 INFO [finetune.py:976] (3/7) Epoch 28, batch 1300, loss[loss=0.1967, simple_loss=0.2659, pruned_loss=0.06379, over 4933.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2397, pruned_loss=0.04808, over 953825.36 frames. ], batch size: 38, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:44:32,019 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155961.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:44:38,007 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.818e+01 1.476e+02 1.729e+02 2.197e+02 4.050e+02, threshold=3.458e+02, percent-clipped=1.0 2023-03-27 09:44:53,644 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8755, 1.7390, 1.4785, 1.3954, 1.6890, 1.6459, 1.6531, 2.2314], device='cuda:3'), covar=tensor([0.3530, 0.3357, 0.3115, 0.3242, 0.3268, 0.2383, 0.3139, 0.1634], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0265, 0.0238, 0.0276, 0.0261, 0.0231, 0.0260, 0.0239], 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') 2023-03-27 09:44:55,320 INFO [finetune.py:976] (3/7) Epoch 28, batch 1350, loss[loss=0.1758, simple_loss=0.2517, pruned_loss=0.04991, over 4822.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2405, pruned_loss=0.04848, over 955478.25 frames. ], batch size: 40, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:44:56,001 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4916, 3.4510, 3.2024, 1.4323, 3.5464, 2.6373, 0.7452, 2.3213], device='cuda:3'), covar=tensor([0.2605, 0.2333, 0.1813, 0.3819, 0.1234, 0.1048, 0.4709, 0.1711], device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0179, 0.0160, 0.0130, 0.0163, 0.0123, 0.0148, 0.0125], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 09:45:10,153 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156013.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:45:30,427 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5036, 1.3867, 1.9883, 2.9276, 1.9495, 2.1689, 1.1509, 2.4674], device='cuda:3'), covar=tensor([0.1702, 0.1429, 0.1166, 0.0692, 0.0858, 0.1504, 0.1590, 0.0520], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0117, 0.0134, 0.0167, 0.0101, 0.0137, 0.0126, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 09:45:31,628 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.6705, 4.0873, 4.2683, 4.5112, 4.4293, 4.1902, 4.7669, 1.5543], device='cuda:3'), covar=tensor([0.0869, 0.0880, 0.0923, 0.1043, 0.1361, 0.1575, 0.0664, 0.6138], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0245, 0.0284, 0.0296, 0.0334, 0.0287, 0.0304, 0.0302], 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') 2023-03-27 09:45:32,753 INFO [finetune.py:976] (3/7) Epoch 28, batch 1400, loss[loss=0.1675, simple_loss=0.245, pruned_loss=0.04496, over 4868.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2447, pruned_loss=0.05003, over 954960.30 frames. ], batch size: 34, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:45:48,226 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156070.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:45:48,708 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.069e+02 1.532e+02 1.806e+02 2.221e+02 4.474e+02, threshold=3.612e+02, percent-clipped=3.0 2023-03-27 09:46:06,150 INFO [finetune.py:976] (3/7) Epoch 28, batch 1450, loss[loss=0.1668, simple_loss=0.2415, pruned_loss=0.04602, over 4732.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2447, pruned_loss=0.04965, over 954725.25 frames. ], batch size: 27, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:46:14,583 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9566, 1.9237, 1.6508, 2.0375, 2.5500, 2.1671, 1.7537, 1.5693], device='cuda:3'), covar=tensor([0.2031, 0.1877, 0.1808, 0.1610, 0.1504, 0.1059, 0.2232, 0.1830], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0212, 0.0216, 0.0200, 0.0246, 0.0190, 0.0217, 0.0205], 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') 2023-03-27 09:46:23,161 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156113.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:46:35,152 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156131.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 09:46:40,535 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156140.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 09:46:45,776 INFO [finetune.py:976] (3/7) Epoch 28, batch 1500, loss[loss=0.2099, simple_loss=0.2779, pruned_loss=0.07095, over 4738.00 frames. ], tot_loss[loss=0.174, simple_loss=0.247, pruned_loss=0.05045, over 954894.05 frames. ], batch size: 54, lr: 2.89e-03, grad_scale: 64.0 2023-03-27 09:46:47,657 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-27 09:46:54,178 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=156161.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:47:02,075 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.065e+02 1.585e+02 1.899e+02 2.331e+02 3.577e+02, threshold=3.799e+02, percent-clipped=0.0 2023-03-27 09:47:18,929 INFO [finetune.py:976] (3/7) Epoch 28, batch 1550, loss[loss=0.1783, simple_loss=0.2536, pruned_loss=0.05147, over 4886.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2452, pruned_loss=0.04973, over 952304.09 frames. ], batch size: 43, lr: 2.89e-03, grad_scale: 32.0 2023-03-27 09:47:21,365 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-03-27 09:47:30,566 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156216.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:47:59,289 INFO [finetune.py:976] (3/7) Epoch 28, batch 1600, loss[loss=0.1311, simple_loss=0.1931, pruned_loss=0.03455, over 4740.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2434, pruned_loss=0.04941, over 953444.06 frames. ], batch size: 23, lr: 2.89e-03, grad_scale: 32.0 2023-03-27 09:48:08,680 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156261.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:48:15,839 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.043e+02 1.435e+02 1.767e+02 2.111e+02 3.704e+02, threshold=3.535e+02, percent-clipped=0.0 2023-03-27 09:48:19,947 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156277.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:48:27,880 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156290.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:48:32,613 INFO [finetune.py:976] (3/7) Epoch 28, batch 1650, loss[loss=0.1711, simple_loss=0.2344, pruned_loss=0.05389, over 4909.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2396, pruned_loss=0.04784, over 955961.01 frames. ], batch size: 36, lr: 2.89e-03, grad_scale: 32.0 2023-03-27 09:48:38,736 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9868, 1.9302, 1.6258, 2.0833, 2.4953, 2.2166, 1.9394, 1.6326], device='cuda:3'), covar=tensor([0.2003, 0.1758, 0.1717, 0.1475, 0.1579, 0.1021, 0.2040, 0.1849], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0212, 0.0215, 0.0199, 0.0245, 0.0190, 0.0216, 0.0204], 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') 2023-03-27 09:48:40,816 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=156309.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:48:43,231 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156313.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:49:18,367 INFO [finetune.py:976] (3/7) Epoch 28, batch 1700, loss[loss=0.1894, simple_loss=0.2634, pruned_loss=0.05769, over 4735.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.238, pruned_loss=0.04696, over 956963.51 frames. ], batch size: 54, lr: 2.88e-03, grad_scale: 32.0 2023-03-27 09:49:20,357 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156351.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:49:27,454 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=156361.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:49:34,434 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.260e+01 1.525e+02 1.796e+02 2.204e+02 4.546e+02, threshold=3.593e+02, percent-clipped=3.0 2023-03-27 09:49:51,285 INFO [finetune.py:976] (3/7) Epoch 28, batch 1750, loss[loss=0.1915, simple_loss=0.2527, pruned_loss=0.06513, over 4829.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2398, pruned_loss=0.04804, over 956968.70 frames. ], batch size: 33, lr: 2.88e-03, grad_scale: 32.0 2023-03-27 09:49:59,094 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.65 vs. limit=5.0 2023-03-27 09:50:08,507 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2023-03-27 09:50:09,512 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156426.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 09:50:19,441 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156440.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 09:50:19,735 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-03-27 09:50:24,158 INFO [finetune.py:976] (3/7) Epoch 28, batch 1800, loss[loss=0.2393, simple_loss=0.3158, pruned_loss=0.08137, over 4860.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.244, pruned_loss=0.04948, over 959355.44 frames. ], batch size: 44, lr: 2.88e-03, grad_scale: 32.0 2023-03-27 09:50:39,987 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.070e+02 1.522e+02 1.851e+02 2.291e+02 4.651e+02, threshold=3.702e+02, percent-clipped=5.0 2023-03-27 09:50:51,504 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=156488.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:50:57,542 INFO [finetune.py:976] (3/7) Epoch 28, batch 1850, loss[loss=0.1742, simple_loss=0.2428, pruned_loss=0.05282, over 4883.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2457, pruned_loss=0.05006, over 959365.55 frames. ], batch size: 35, lr: 2.88e-03, grad_scale: 32.0 2023-03-27 09:51:40,565 INFO [finetune.py:976] (3/7) Epoch 28, batch 1900, loss[loss=0.1581, simple_loss=0.2347, pruned_loss=0.04075, over 4745.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.248, pruned_loss=0.05086, over 958115.82 frames. ], batch size: 54, lr: 2.88e-03, grad_scale: 32.0 2023-03-27 09:51:47,429 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-27 09:51:56,031 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.831e+01 1.540e+02 1.871e+02 2.242e+02 4.934e+02, threshold=3.741e+02, percent-clipped=1.0 2023-03-27 09:51:56,113 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156572.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:52:11,756 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.52 vs. limit=5.0 2023-03-27 09:52:13,684 INFO [finetune.py:976] (3/7) Epoch 28, batch 1950, loss[loss=0.2059, simple_loss=0.2672, pruned_loss=0.07231, over 4827.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2461, pruned_loss=0.05013, over 957025.36 frames. ], batch size: 33, lr: 2.88e-03, grad_scale: 32.0 2023-03-27 09:52:18,855 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6945, 1.6540, 1.4991, 1.5927, 1.9732, 1.9745, 1.7774, 1.5268], device='cuda:3'), covar=tensor([0.0393, 0.0353, 0.0661, 0.0348, 0.0266, 0.0486, 0.0376, 0.0444], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0106, 0.0149, 0.0112, 0.0102, 0.0116, 0.0104, 0.0114], device='cuda:3'), out_proj_covar=tensor([7.8499e-05, 8.1265e-05, 1.1577e-04, 8.5296e-05, 7.8650e-05, 8.5386e-05, 7.7034e-05, 8.6153e-05], device='cuda:3') 2023-03-27 09:52:20,679 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.2186, 1.2051, 1.1934, 0.6897, 1.2211, 1.3563, 1.4912, 1.1694], device='cuda:3'), covar=tensor([0.0889, 0.0638, 0.0614, 0.0448, 0.0545, 0.0707, 0.0358, 0.0706], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0147, 0.0130, 0.0122, 0.0131, 0.0130, 0.0142, 0.0150], device='cuda:3'), out_proj_covar=tensor([8.8340e-05, 1.0546e-04, 9.2176e-05, 8.5836e-05, 9.1817e-05, 9.2303e-05, 1.0089e-04, 1.0705e-04], device='cuda:3') 2023-03-27 09:52:46,376 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156646.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:52:47,505 INFO [finetune.py:976] (3/7) Epoch 28, batch 2000, loss[loss=0.1847, simple_loss=0.2479, pruned_loss=0.06073, over 4929.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2437, pruned_loss=0.04954, over 957183.17 frames. ], batch size: 33, lr: 2.88e-03, grad_scale: 32.0 2023-03-27 09:53:04,320 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.919e+01 1.529e+02 1.760e+02 2.169e+02 4.761e+02, threshold=3.520e+02, percent-clipped=1.0 2023-03-27 09:53:20,897 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 2023-03-27 09:53:29,982 INFO [finetune.py:976] (3/7) Epoch 28, batch 2050, loss[loss=0.157, simple_loss=0.2392, pruned_loss=0.0374, over 4753.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2406, pruned_loss=0.04881, over 956709.37 frames. ], batch size: 26, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 09:53:47,462 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156726.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 09:54:06,185 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9746, 1.5388, 1.9650, 2.0001, 1.7807, 1.7398, 1.9629, 1.8474], device='cuda:3'), covar=tensor([0.3944, 0.3913, 0.3461, 0.3808, 0.4772, 0.4056, 0.4453, 0.3196], device='cuda:3'), in_proj_covar=tensor([0.0269, 0.0249, 0.0270, 0.0298, 0.0297, 0.0274, 0.0304, 0.0254], 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') 2023-03-27 09:54:08,961 INFO [finetune.py:976] (3/7) Epoch 28, batch 2100, loss[loss=0.1411, simple_loss=0.222, pruned_loss=0.03006, over 4794.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2399, pruned_loss=0.04825, over 955165.12 frames. ], batch size: 29, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 09:54:09,707 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156749.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:54:37,771 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.070e+02 1.521e+02 1.844e+02 2.179e+02 3.224e+02, threshold=3.687e+02, percent-clipped=0.0 2023-03-27 09:54:38,477 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=156774.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:54:45,059 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3839, 2.0255, 2.7792, 1.6432, 2.2769, 2.6441, 1.8542, 2.6575], device='cuda:3'), covar=tensor([0.1478, 0.2206, 0.1521, 0.2281, 0.1178, 0.1567, 0.2986, 0.0945], device='cuda:3'), in_proj_covar=tensor([0.0194, 0.0209, 0.0195, 0.0191, 0.0176, 0.0214, 0.0220, 0.0201], 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') 2023-03-27 09:54:54,963 INFO [finetune.py:976] (3/7) Epoch 28, batch 2150, loss[loss=0.1844, simple_loss=0.2626, pruned_loss=0.05306, over 4764.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2424, pruned_loss=0.04846, over 956483.54 frames. ], batch size: 28, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 09:55:03,485 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156810.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:55:13,216 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0 2023-03-27 09:55:23,024 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.23 vs. limit=5.0 2023-03-27 09:55:27,788 INFO [finetune.py:976] (3/7) Epoch 28, batch 2200, loss[loss=0.1795, simple_loss=0.2465, pruned_loss=0.05621, over 4859.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2438, pruned_loss=0.04908, over 955022.66 frames. ], batch size: 34, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 09:55:44,140 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156872.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:55:44,654 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.040e+02 1.543e+02 1.740e+02 2.127e+02 4.555e+02, threshold=3.480e+02, percent-clipped=1.0 2023-03-27 09:55:52,580 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156885.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:56:01,289 INFO [finetune.py:976] (3/7) Epoch 28, batch 2250, loss[loss=0.143, simple_loss=0.2181, pruned_loss=0.03398, over 4925.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2452, pruned_loss=0.04967, over 955400.92 frames. ], batch size: 33, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 09:56:15,883 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-03-27 09:56:16,205 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=156920.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:56:17,826 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.82 vs. limit=5.0 2023-03-27 09:56:32,864 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156946.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:56:32,885 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156946.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:56:33,999 INFO [finetune.py:976] (3/7) Epoch 28, batch 2300, loss[loss=0.1694, simple_loss=0.2446, pruned_loss=0.04709, over 4803.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2449, pruned_loss=0.04857, over 955492.89 frames. ], batch size: 25, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 09:57:00,119 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 1.437e+02 1.655e+02 2.039e+02 3.893e+02, threshold=3.311e+02, percent-clipped=1.0 2023-03-27 09:57:17,551 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=156994.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:57:20,394 INFO [finetune.py:976] (3/7) Epoch 28, batch 2350, loss[loss=0.1552, simple_loss=0.2224, pruned_loss=0.04402, over 4778.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.243, pruned_loss=0.04818, over 957272.78 frames. ], batch size: 28, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 09:57:28,245 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157010.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:57:52,968 INFO [finetune.py:976] (3/7) Epoch 28, batch 2400, loss[loss=0.1346, simple_loss=0.2055, pruned_loss=0.03186, over 4729.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2404, pruned_loss=0.04774, over 956604.37 frames. ], batch size: 59, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 09:58:07,295 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2590, 2.0900, 1.8967, 2.0509, 2.0005, 2.0281, 2.0813, 2.8032], device='cuda:3'), covar=tensor([0.3234, 0.3938, 0.2832, 0.3458, 0.3734, 0.2301, 0.3282, 0.1395], device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0264, 0.0238, 0.0275, 0.0261, 0.0230, 0.0259, 0.0239], 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') 2023-03-27 09:58:08,928 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157071.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:58:12,821 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.019e+02 1.490e+02 1.799e+02 2.218e+02 3.254e+02, threshold=3.597e+02, percent-clipped=0.0 2023-03-27 09:58:28,594 INFO [finetune.py:976] (3/7) Epoch 28, batch 2450, loss[loss=0.1974, simple_loss=0.2658, pruned_loss=0.06447, over 4894.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2365, pruned_loss=0.04618, over 958004.80 frames. ], batch size: 32, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 09:58:33,536 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157105.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:58:34,803 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6571, 1.5719, 1.4854, 1.5751, 1.1844, 3.5509, 1.2912, 1.8384], device='cuda:3'), covar=tensor([0.3201, 0.2462, 0.2137, 0.2392, 0.1748, 0.0199, 0.2756, 0.1212], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0121, 0.0124, 0.0114, 0.0096, 0.0094, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 09:58:57,710 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157142.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 09:59:01,796 INFO [finetune.py:976] (3/7) Epoch 28, batch 2500, loss[loss=0.1767, simple_loss=0.2563, pruned_loss=0.04852, over 4897.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2387, pruned_loss=0.04748, over 955425.72 frames. ], batch size: 32, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 09:59:23,963 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.120e+02 1.586e+02 1.871e+02 2.257e+02 5.817e+02, threshold=3.742e+02, percent-clipped=3.0 2023-03-27 09:59:35,836 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.9353, 3.4330, 3.6504, 3.8464, 3.7293, 3.3924, 3.9417, 1.2634], device='cuda:3'), covar=tensor([0.0800, 0.0888, 0.0933, 0.0922, 0.1117, 0.1667, 0.0786, 0.5791], device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0246, 0.0284, 0.0294, 0.0335, 0.0286, 0.0305, 0.0303], 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') 2023-03-27 09:59:51,557 INFO [finetune.py:976] (3/7) Epoch 28, batch 2550, loss[loss=0.2171, simple_loss=0.3026, pruned_loss=0.06583, over 4833.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2427, pruned_loss=0.04906, over 954104.98 frames. ], batch size: 47, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 09:59:55,368 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157203.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:00:20,725 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157241.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:00:24,904 INFO [finetune.py:976] (3/7) Epoch 28, batch 2600, loss[loss=0.1676, simple_loss=0.2422, pruned_loss=0.04644, over 4766.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2443, pruned_loss=0.04961, over 954491.92 frames. ], batch size: 28, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 10:00:34,240 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2607, 1.3797, 1.6106, 1.5605, 1.4756, 2.8695, 1.2512, 1.4612], device='cuda:3'), covar=tensor([0.1086, 0.1895, 0.1143, 0.0985, 0.1672, 0.0312, 0.1570, 0.1741], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0074, 0.0076, 0.0091, 0.0080, 0.0086, 0.0081], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-27 10:00:41,899 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.034e+02 1.561e+02 1.846e+02 2.212e+02 4.271e+02, threshold=3.692e+02, percent-clipped=1.0 2023-03-27 10:00:57,758 INFO [finetune.py:976] (3/7) Epoch 28, batch 2650, loss[loss=0.1334, simple_loss=0.2061, pruned_loss=0.03039, over 4772.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2446, pruned_loss=0.04897, over 956936.03 frames. ], batch size: 26, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 10:01:14,925 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9393, 1.4399, 2.0172, 2.0130, 1.7816, 1.7356, 1.9139, 1.8855], device='cuda:3'), covar=tensor([0.4055, 0.4042, 0.3173, 0.3510, 0.4908, 0.3719, 0.4522, 0.2998], device='cuda:3'), in_proj_covar=tensor([0.0269, 0.0250, 0.0270, 0.0299, 0.0299, 0.0274, 0.0304, 0.0254], 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') 2023-03-27 10:01:30,663 INFO [finetune.py:976] (3/7) Epoch 28, batch 2700, loss[loss=0.1339, simple_loss=0.2095, pruned_loss=0.02916, over 4892.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2431, pruned_loss=0.04833, over 956540.87 frames. ], batch size: 35, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 10:01:35,644 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157356.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:01:42,598 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157366.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:01:47,173 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.080e+02 1.455e+02 1.749e+02 2.146e+02 4.370e+02, threshold=3.498e+02, percent-clipped=1.0 2023-03-27 10:01:53,267 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157382.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:02:12,808 INFO [finetune.py:976] (3/7) Epoch 28, batch 2750, loss[loss=0.1984, simple_loss=0.2572, pruned_loss=0.06974, over 4226.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2407, pruned_loss=0.04799, over 956186.67 frames. ], batch size: 65, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 10:02:20,320 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157404.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 10:02:20,954 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157405.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:02:21,573 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8219, 1.6927, 1.4956, 1.9270, 2.2770, 1.9379, 1.5175, 1.5183], device='cuda:3'), covar=tensor([0.2071, 0.1862, 0.1903, 0.1607, 0.1489, 0.1205, 0.2391, 0.1911], device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0212, 0.0216, 0.0200, 0.0246, 0.0191, 0.0217, 0.0205], 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') 2023-03-27 10:02:22,857 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.97 vs. limit=5.0 2023-03-27 10:02:29,172 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157417.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:02:33,285 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3400, 2.2276, 1.7605, 2.3531, 2.2068, 1.9532, 2.5956, 2.3396], device='cuda:3'), covar=tensor([0.1195, 0.2058, 0.2882, 0.2435, 0.2409, 0.1636, 0.2726, 0.1567], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0191, 0.0238, 0.0255, 0.0251, 0.0208, 0.0215, 0.0204], 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') 2023-03-27 10:02:37,512 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5863, 3.6595, 3.4591, 1.6698, 3.7029, 2.9291, 0.8765, 2.7123], device='cuda:3'), covar=tensor([0.2441, 0.2160, 0.1599, 0.3174, 0.1220, 0.1006, 0.4151, 0.1471], device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0180, 0.0160, 0.0130, 0.0164, 0.0124, 0.0149, 0.0126], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 10:02:46,866 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157443.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:02:50,279 INFO [finetune.py:976] (3/7) Epoch 28, batch 2800, loss[loss=0.1852, simple_loss=0.2407, pruned_loss=0.06483, over 4282.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2374, pruned_loss=0.0471, over 956213.64 frames. ], batch size: 65, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 10:02:53,300 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=157453.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:02:53,353 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7520, 1.6692, 1.6171, 1.6769, 1.1820, 3.6916, 1.4694, 2.0114], device='cuda:3'), covar=tensor([0.3239, 0.2585, 0.2055, 0.2241, 0.1728, 0.0178, 0.2417, 0.1121], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0121, 0.0124, 0.0114, 0.0095, 0.0094, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 10:02:57,156 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-03-27 10:03:01,073 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157465.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 10:03:06,316 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.288e+01 1.520e+02 1.688e+02 2.071e+02 7.416e+02, threshold=3.376e+02, percent-clipped=4.0 2023-03-27 10:03:18,788 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6888, 1.5511, 1.5025, 1.6569, 1.8823, 1.9010, 1.6936, 1.4822], device='cuda:3'), covar=tensor([0.0404, 0.0338, 0.0578, 0.0336, 0.0251, 0.0448, 0.0319, 0.0411], device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0106, 0.0148, 0.0111, 0.0102, 0.0116, 0.0104, 0.0113], device='cuda:3'), out_proj_covar=tensor([7.8687e-05, 8.0964e-05, 1.1521e-04, 8.5012e-05, 7.8758e-05, 8.5260e-05, 7.7045e-05, 8.5938e-05], device='cuda:3') 2023-03-27 10:03:23,414 INFO [finetune.py:976] (3/7) Epoch 28, batch 2850, loss[loss=0.1635, simple_loss=0.2399, pruned_loss=0.04357, over 4893.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2355, pruned_loss=0.04603, over 956997.12 frames. ], batch size: 35, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 10:03:23,481 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157498.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:03:28,866 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157506.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 10:03:52,496 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157541.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:03:57,080 INFO [finetune.py:976] (3/7) Epoch 28, batch 2900, loss[loss=0.1841, simple_loss=0.2548, pruned_loss=0.05668, over 4918.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2396, pruned_loss=0.0474, over 955003.38 frames. ], batch size: 36, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 10:04:09,788 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157567.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 10:04:10,367 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157568.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:04:13,244 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.581e+01 1.484e+02 1.768e+02 2.098e+02 4.175e+02, threshold=3.535e+02, percent-clipped=1.0 2023-03-27 10:04:18,532 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2219, 2.8879, 2.7761, 1.2953, 3.0380, 2.2936, 0.9530, 1.9555], device='cuda:3'), covar=tensor([0.2557, 0.2569, 0.1772, 0.3451, 0.1428, 0.1086, 0.3705, 0.1614], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0182, 0.0161, 0.0131, 0.0165, 0.0125, 0.0150, 0.0127], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 10:04:24,446 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=157589.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:04:32,418 INFO [finetune.py:976] (3/7) Epoch 28, batch 2950, loss[loss=0.1547, simple_loss=0.2397, pruned_loss=0.03482, over 4811.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2428, pruned_loss=0.04862, over 955097.87 frames. ], batch size: 38, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 10:04:41,096 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-03-27 10:05:06,828 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157629.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:05:23,714 INFO [finetune.py:976] (3/7) Epoch 28, batch 3000, loss[loss=0.182, simple_loss=0.2578, pruned_loss=0.05312, over 4819.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2441, pruned_loss=0.04926, over 953468.14 frames. ], batch size: 38, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 10:05:23,714 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-27 10:05:27,669 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1100, 2.0427, 1.8220, 2.0699, 2.1342, 1.9054, 2.2332, 2.1595], device='cuda:3'), covar=tensor([0.1258, 0.2200, 0.2767, 0.2058, 0.2224, 0.1527, 0.2837, 0.1600], device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0190, 0.0237, 0.0254, 0.0250, 0.0207, 0.0215, 0.0203], 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') 2023-03-27 10:05:30,427 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.1329, 1.3988, 1.4357, 0.6479, 1.4352, 1.6283, 1.7030, 1.3575], device='cuda:3'), covar=tensor([0.0927, 0.0616, 0.0513, 0.0513, 0.0529, 0.0544, 0.0342, 0.0678], device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0149, 0.0130, 0.0123, 0.0132, 0.0131, 0.0143, 0.0152], device='cuda:3'), out_proj_covar=tensor([8.8922e-05, 1.0667e-04, 9.2651e-05, 8.6451e-05, 9.2698e-05, 9.2844e-05, 1.0146e-04, 1.0827e-04], device='cuda:3') 2023-03-27 10:05:32,954 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.8399, 3.4833, 3.5598, 3.7634, 3.5831, 3.4362, 3.9146, 1.2726], device='cuda:3'), covar=tensor([0.0894, 0.0811, 0.0907, 0.0904, 0.1431, 0.1749, 0.0783, 0.5889], device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0245, 0.0285, 0.0294, 0.0334, 0.0285, 0.0305, 0.0303], 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') 2023-03-27 10:05:34,509 INFO [finetune.py:1010] (3/7) Epoch 28, validation: loss=0.1567, simple_loss=0.2243, pruned_loss=0.04455, over 2265189.00 frames. 2023-03-27 10:05:34,510 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6469MB 2023-03-27 10:05:46,488 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157666.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:05:50,618 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.041e+02 1.524e+02 1.841e+02 2.248e+02 4.082e+02, threshold=3.682e+02, percent-clipped=3.0 2023-03-27 10:06:07,190 INFO [finetune.py:976] (3/7) Epoch 28, batch 3050, loss[loss=0.154, simple_loss=0.2329, pruned_loss=0.03756, over 4926.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2458, pruned_loss=0.04975, over 953245.36 frames. ], batch size: 38, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 10:06:16,640 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157712.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:06:17,821 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=157714.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:06:33,278 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157738.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:06:38,535 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6822, 2.3465, 2.9305, 1.9565, 2.6272, 2.9333, 1.9919, 2.9073], device='cuda:3'), covar=tensor([0.1203, 0.1858, 0.1537, 0.1914, 0.0961, 0.1282, 0.2725, 0.0840], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0205, 0.0192, 0.0188, 0.0173, 0.0211, 0.0217, 0.0197], 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') 2023-03-27 10:06:40,234 INFO [finetune.py:976] (3/7) Epoch 28, batch 3100, loss[loss=0.141, simple_loss=0.2198, pruned_loss=0.0311, over 4762.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2439, pruned_loss=0.04905, over 955432.27 frames. ], batch size: 27, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 10:06:48,817 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157760.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 10:06:57,050 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.055e+02 1.389e+02 1.744e+02 2.105e+02 3.209e+02, threshold=3.488e+02, percent-clipped=0.0 2023-03-27 10:07:19,510 INFO [finetune.py:976] (3/7) Epoch 28, batch 3150, loss[loss=0.188, simple_loss=0.2556, pruned_loss=0.06015, over 4795.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2406, pruned_loss=0.04783, over 955985.05 frames. ], batch size: 51, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 10:07:20,085 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157798.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:08:03,993 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=157846.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:08:05,667 INFO [finetune.py:976] (3/7) Epoch 28, batch 3200, loss[loss=0.1326, simple_loss=0.2074, pruned_loss=0.02892, over 4788.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2384, pruned_loss=0.04769, over 956146.96 frames. ], batch size: 26, lr: 2.88e-03, grad_scale: 16.0 2023-03-27 10:08:09,350 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157853.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:08:15,275 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157862.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 10:08:22,849 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.817e+01 1.555e+02 1.831e+02 2.254e+02 7.078e+02, threshold=3.662e+02, percent-clipped=7.0 2023-03-27 10:08:38,480 INFO [finetune.py:976] (3/7) Epoch 28, batch 3250, loss[loss=0.1398, simple_loss=0.2117, pruned_loss=0.03399, over 4720.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2398, pruned_loss=0.04862, over 956167.60 frames. ], batch size: 23, lr: 2.87e-03, grad_scale: 16.0 2023-03-27 10:08:49,845 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157914.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:08:56,864 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157924.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:09:11,706 INFO [finetune.py:976] (3/7) Epoch 28, batch 3300, loss[loss=0.1576, simple_loss=0.2426, pruned_loss=0.03626, over 4818.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2409, pruned_loss=0.04831, over 953692.76 frames. ], batch size: 30, lr: 2.87e-03, grad_scale: 16.0 2023-03-27 10:09:29,133 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.081e+02 1.521e+02 1.758e+02 2.140e+02 4.599e+02, threshold=3.515e+02, percent-clipped=1.0 2023-03-27 10:09:40,583 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6207, 3.3463, 3.2648, 1.6574, 3.5860, 2.7786, 1.3302, 2.4579], device='cuda:3'), covar=tensor([0.2944, 0.2125, 0.1433, 0.3189, 0.1103, 0.0990, 0.3606, 0.1486], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0181, 0.0161, 0.0131, 0.0165, 0.0125, 0.0151, 0.0127], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 10:09:44,716 INFO [finetune.py:976] (3/7) Epoch 28, batch 3350, loss[loss=0.1882, simple_loss=0.2673, pruned_loss=0.05451, over 4814.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2435, pruned_loss=0.04943, over 952888.10 frames. ], batch size: 38, lr: 2.87e-03, grad_scale: 16.0 2023-03-27 10:09:58,173 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158012.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:10:16,562 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1012, 3.5483, 3.7017, 3.9285, 3.8077, 3.5644, 4.1606, 1.3243], device='cuda:3'), covar=tensor([0.0857, 0.0967, 0.0931, 0.1053, 0.1558, 0.1726, 0.0834, 0.5921], device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0247, 0.0287, 0.0296, 0.0338, 0.0287, 0.0307, 0.0305], 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') 2023-03-27 10:10:32,973 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158038.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:10:38,995 INFO [finetune.py:976] (3/7) Epoch 28, batch 3400, loss[loss=0.1723, simple_loss=0.2359, pruned_loss=0.05432, over 4855.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2435, pruned_loss=0.04898, over 954485.78 frames. ], batch size: 31, lr: 2.87e-03, grad_scale: 16.0 2023-03-27 10:10:46,886 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=158060.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:10:46,931 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158060.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 10:10:55,547 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.103e+02 1.550e+02 1.897e+02 2.350e+02 3.360e+02, threshold=3.793e+02, percent-clipped=0.0 2023-03-27 10:11:04,974 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=158086.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:11:12,585 INFO [finetune.py:976] (3/7) Epoch 28, batch 3450, loss[loss=0.1654, simple_loss=0.2344, pruned_loss=0.04821, over 4179.00 frames. ], tot_loss[loss=0.17, simple_loss=0.243, pruned_loss=0.04851, over 954623.41 frames. ], batch size: 66, lr: 2.87e-03, grad_scale: 16.0 2023-03-27 10:11:18,607 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=158108.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 10:11:28,436 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158122.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:11:46,008 INFO [finetune.py:976] (3/7) Epoch 28, batch 3500, loss[loss=0.1565, simple_loss=0.23, pruned_loss=0.04148, over 4935.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2404, pruned_loss=0.04781, over 955815.48 frames. ], batch size: 38, lr: 2.87e-03, grad_scale: 16.0 2023-03-27 10:11:51,037 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-03-27 10:11:54,574 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158162.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 10:12:02,509 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.003e+02 1.432e+02 1.800e+02 2.074e+02 3.411e+02, threshold=3.600e+02, percent-clipped=0.0 2023-03-27 10:12:09,601 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158183.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:12:19,503 INFO [finetune.py:976] (3/7) Epoch 28, batch 3550, loss[loss=0.1437, simple_loss=0.2089, pruned_loss=0.03929, over 4756.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2389, pruned_loss=0.04745, over 955365.79 frames. ], batch size: 54, lr: 2.87e-03, grad_scale: 16.0 2023-03-27 10:12:28,587 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158209.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:12:29,188 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=158210.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 10:12:31,674 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4754, 1.4685, 2.0691, 1.8480, 1.6849, 3.6554, 1.3527, 1.7218], device='cuda:3'), covar=tensor([0.0987, 0.1903, 0.1027, 0.0975, 0.1560, 0.0214, 0.1594, 0.1826], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0073, 0.0076, 0.0091, 0.0080, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-27 10:12:38,713 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158224.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:13:02,523 INFO [finetune.py:976] (3/7) Epoch 28, batch 3600, loss[loss=0.1835, simple_loss=0.2601, pruned_loss=0.05341, over 4833.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2372, pruned_loss=0.04701, over 956435.90 frames. ], batch size: 40, lr: 2.87e-03, grad_scale: 16.0 2023-03-27 10:13:18,161 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=158272.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:13:18,717 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.065e+02 1.425e+02 1.679e+02 2.016e+02 3.584e+02, threshold=3.358e+02, percent-clipped=0.0 2023-03-27 10:13:36,304 INFO [finetune.py:976] (3/7) Epoch 28, batch 3650, loss[loss=0.1833, simple_loss=0.262, pruned_loss=0.05232, over 4831.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2394, pruned_loss=0.04805, over 957110.96 frames. ], batch size: 47, lr: 2.87e-03, grad_scale: 16.0 2023-03-27 10:14:10,062 INFO [finetune.py:976] (3/7) Epoch 28, batch 3700, loss[loss=0.172, simple_loss=0.2436, pruned_loss=0.05014, over 4811.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2415, pruned_loss=0.04801, over 955686.89 frames. ], batch size: 45, lr: 2.87e-03, grad_scale: 16.0 2023-03-27 10:14:23,365 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0593, 1.9752, 1.7349, 1.9896, 1.8853, 1.8951, 1.8961, 2.6188], device='cuda:3'), covar=tensor([0.3790, 0.4367, 0.3419, 0.3830, 0.4125, 0.2322, 0.4145, 0.1712], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0266, 0.0239, 0.0277, 0.0262, 0.0232, 0.0261, 0.0240], 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') 2023-03-27 10:14:26,132 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 1.573e+02 1.954e+02 2.338e+02 5.991e+02, threshold=3.909e+02, percent-clipped=5.0 2023-03-27 10:14:33,449 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158384.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:14:43,249 INFO [finetune.py:976] (3/7) Epoch 28, batch 3750, loss[loss=0.1693, simple_loss=0.2362, pruned_loss=0.05125, over 4779.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2439, pruned_loss=0.04917, over 954585.96 frames. ], batch size: 26, lr: 2.87e-03, grad_scale: 16.0 2023-03-27 10:15:19,907 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158445.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:15:22,040 INFO [finetune.py:976] (3/7) Epoch 28, batch 3800, loss[loss=0.2083, simple_loss=0.2732, pruned_loss=0.07175, over 4819.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2454, pruned_loss=0.0495, over 955804.86 frames. ], batch size: 30, lr: 2.87e-03, grad_scale: 16.0 2023-03-27 10:15:51,764 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.090e+02 1.580e+02 1.909e+02 2.258e+02 3.504e+02, threshold=3.818e+02, percent-clipped=0.0 2023-03-27 10:15:55,436 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158478.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:16:08,885 INFO [finetune.py:976] (3/7) Epoch 28, batch 3850, loss[loss=0.1285, simple_loss=0.2009, pruned_loss=0.028, over 4874.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2433, pruned_loss=0.04837, over 954979.79 frames. ], batch size: 31, lr: 2.87e-03, grad_scale: 16.0 2023-03-27 10:16:16,604 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158509.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:16:40,985 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2918, 2.0730, 1.8710, 2.0986, 1.9944, 2.0233, 2.0432, 2.7669], device='cuda:3'), covar=tensor([0.3719, 0.4710, 0.3301, 0.3652, 0.4117, 0.2461, 0.3802, 0.1686], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0265, 0.0238, 0.0276, 0.0261, 0.0231, 0.0260, 0.0238], 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') 2023-03-27 10:16:42,035 INFO [finetune.py:976] (3/7) Epoch 28, batch 3900, loss[loss=0.1763, simple_loss=0.2404, pruned_loss=0.05612, over 4891.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2411, pruned_loss=0.04794, over 953141.97 frames. ], batch size: 32, lr: 2.87e-03, grad_scale: 16.0 2023-03-27 10:16:48,975 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=158557.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:16:58,922 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.091e+02 1.423e+02 1.682e+02 2.011e+02 3.434e+02, threshold=3.365e+02, percent-clipped=0.0 2023-03-27 10:17:15,468 INFO [finetune.py:976] (3/7) Epoch 28, batch 3950, loss[loss=0.1797, simple_loss=0.2391, pruned_loss=0.06014, over 4866.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2384, pruned_loss=0.04749, over 953091.27 frames. ], batch size: 34, lr: 2.87e-03, grad_scale: 16.0 2023-03-27 10:17:48,877 INFO [finetune.py:976] (3/7) Epoch 28, batch 4000, loss[loss=0.171, simple_loss=0.2325, pruned_loss=0.05474, over 4882.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2389, pruned_loss=0.04827, over 953647.57 frames. ], batch size: 32, lr: 2.87e-03, grad_scale: 16.0 2023-03-27 10:18:15,602 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.131e+01 1.444e+02 1.852e+02 2.094e+02 7.470e+02, threshold=3.703e+02, percent-clipped=2.0 2023-03-27 10:18:19,891 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.4944, 1.5070, 1.5410, 1.0621, 1.6820, 1.8894, 1.8460, 1.4660], device='cuda:3'), covar=tensor([0.0967, 0.0778, 0.0591, 0.0495, 0.0492, 0.0604, 0.0306, 0.0687], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0148, 0.0130, 0.0122, 0.0132, 0.0130, 0.0143, 0.0151], device='cuda:3'), out_proj_covar=tensor([8.8554e-05, 1.0598e-04, 9.2561e-05, 8.5687e-05, 9.2149e-05, 9.1947e-05, 1.0152e-04, 1.0794e-04], device='cuda:3') 2023-03-27 10:18:32,177 INFO [finetune.py:976] (3/7) Epoch 28, batch 4050, loss[loss=0.1647, simple_loss=0.2339, pruned_loss=0.04776, over 4789.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2409, pruned_loss=0.04912, over 952813.40 frames. ], batch size: 25, lr: 2.87e-03, grad_scale: 32.0 2023-03-27 10:18:59,960 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158740.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:19:05,205 INFO [finetune.py:976] (3/7) Epoch 28, batch 4100, loss[loss=0.1239, simple_loss=0.1931, pruned_loss=0.02737, over 4719.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2434, pruned_loss=0.04945, over 952610.55 frames. ], batch size: 23, lr: 2.87e-03, grad_scale: 32.0 2023-03-27 10:19:22,719 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.067e+02 1.534e+02 1.832e+02 2.269e+02 3.411e+02, threshold=3.665e+02, percent-clipped=0.0 2023-03-27 10:19:25,288 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158777.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:19:25,879 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158778.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:19:29,400 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158783.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:19:38,842 INFO [finetune.py:976] (3/7) Epoch 28, batch 4150, loss[loss=0.1819, simple_loss=0.2645, pruned_loss=0.04964, over 4734.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2454, pruned_loss=0.05013, over 953642.93 frames. ], batch size: 27, lr: 2.87e-03, grad_scale: 32.0 2023-03-27 10:19:58,147 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=158826.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:20:03,650 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158835.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 10:20:05,953 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158838.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:20:09,553 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158844.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:20:11,899 INFO [finetune.py:976] (3/7) Epoch 28, batch 4200, loss[loss=0.1485, simple_loss=0.2253, pruned_loss=0.03587, over 4166.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2453, pruned_loss=0.04916, over 954950.38 frames. ], batch size: 65, lr: 2.87e-03, grad_scale: 32.0 2023-03-27 10:20:35,369 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.763e+01 1.457e+02 1.627e+02 2.050e+02 3.601e+02, threshold=3.253e+02, percent-clipped=0.0 2023-03-27 10:20:58,728 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158896.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 10:21:04,568 INFO [finetune.py:976] (3/7) Epoch 28, batch 4250, loss[loss=0.1491, simple_loss=0.2231, pruned_loss=0.03757, over 4755.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2428, pruned_loss=0.04781, over 956224.27 frames. ], batch size: 26, lr: 2.87e-03, grad_scale: 32.0 2023-03-27 10:21:46,471 INFO [finetune.py:976] (3/7) Epoch 28, batch 4300, loss[loss=0.146, simple_loss=0.2158, pruned_loss=0.03804, over 4912.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2399, pruned_loss=0.04726, over 956109.56 frames. ], batch size: 43, lr: 2.87e-03, grad_scale: 32.0 2023-03-27 10:22:01,406 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.5538, 3.2028, 2.9226, 1.6132, 3.0764, 2.4478, 2.3869, 2.8406], device='cuda:3'), covar=tensor([0.0901, 0.0721, 0.1433, 0.2039, 0.1452, 0.1906, 0.1908, 0.1027], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0191, 0.0200, 0.0180, 0.0211, 0.0210, 0.0224, 0.0196], 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') 2023-03-27 10:22:03,614 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.096e+02 1.473e+02 1.789e+02 2.045e+02 3.501e+02, threshold=3.577e+02, percent-clipped=2.0 2023-03-27 10:22:08,869 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3239, 2.2371, 2.3761, 1.6302, 2.2594, 2.4346, 2.5380, 2.0090], device='cuda:3'), covar=tensor([0.0549, 0.0664, 0.0653, 0.0832, 0.0695, 0.0738, 0.0536, 0.1034], device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0137, 0.0139, 0.0118, 0.0128, 0.0139, 0.0139, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 10:22:20,212 INFO [finetune.py:976] (3/7) Epoch 28, batch 4350, loss[loss=0.1691, simple_loss=0.2402, pruned_loss=0.04905, over 4227.00 frames. ], tot_loss[loss=0.165, simple_loss=0.2372, pruned_loss=0.04643, over 956532.35 frames. ], batch size: 65, lr: 2.87e-03, grad_scale: 32.0 2023-03-27 10:22:25,126 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-03-27 10:22:48,312 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159040.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:22:50,141 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6054, 1.5063, 2.0509, 2.7616, 1.9193, 2.1544, 1.2638, 2.3156], device='cuda:3'), covar=tensor([0.1495, 0.1164, 0.0943, 0.0628, 0.0777, 0.1815, 0.1391, 0.0524], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0115, 0.0132, 0.0163, 0.0100, 0.0134, 0.0124, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 10:22:53,109 INFO [finetune.py:976] (3/7) Epoch 28, batch 4400, loss[loss=0.157, simple_loss=0.2373, pruned_loss=0.03837, over 4695.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2379, pruned_loss=0.04685, over 952934.43 frames. ], batch size: 23, lr: 2.87e-03, grad_scale: 32.0 2023-03-27 10:23:07,412 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159069.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:23:09,713 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.014e+02 1.529e+02 1.726e+02 2.218e+02 4.795e+02, threshold=3.452e+02, percent-clipped=2.0 2023-03-27 10:23:24,710 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=159088.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:23:35,374 INFO [finetune.py:976] (3/7) Epoch 28, batch 4450, loss[loss=0.1659, simple_loss=0.2176, pruned_loss=0.0571, over 4036.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2408, pruned_loss=0.04801, over 951044.08 frames. ], batch size: 17, lr: 2.87e-03, grad_scale: 32.0 2023-03-27 10:24:00,789 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159130.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:24:02,993 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159133.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:24:07,591 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159139.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:24:12,914 INFO [finetune.py:976] (3/7) Epoch 28, batch 4500, loss[loss=0.1736, simple_loss=0.2434, pruned_loss=0.05196, over 4869.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2434, pruned_loss=0.04887, over 952852.41 frames. ], batch size: 34, lr: 2.87e-03, grad_scale: 32.0 2023-03-27 10:24:20,178 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159159.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:24:28,959 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.131e+02 1.553e+02 1.848e+02 2.152e+02 3.966e+02, threshold=3.696e+02, percent-clipped=2.0 2023-03-27 10:24:29,059 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.9709, 3.4170, 3.6694, 3.8383, 3.7428, 3.5126, 4.0610, 1.3393], device='cuda:3'), covar=tensor([0.0836, 0.0898, 0.0951, 0.1018, 0.1284, 0.1573, 0.0772, 0.5863], device='cuda:3'), in_proj_covar=tensor([0.0356, 0.0249, 0.0289, 0.0299, 0.0342, 0.0289, 0.0309, 0.0307], 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') 2023-03-27 10:24:35,746 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-27 10:24:41,940 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159191.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 10:24:46,583 INFO [finetune.py:976] (3/7) Epoch 28, batch 4550, loss[loss=0.1658, simple_loss=0.2451, pruned_loss=0.04325, over 4733.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2434, pruned_loss=0.04823, over 954416.93 frames. ], batch size: 27, lr: 2.87e-03, grad_scale: 32.0 2023-03-27 10:24:46,652 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2719, 2.9495, 2.7972, 1.1339, 3.0384, 2.2187, 0.6038, 1.9020], device='cuda:3'), covar=tensor([0.2498, 0.2265, 0.2025, 0.3722, 0.1483, 0.1161, 0.4254, 0.1726], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0183, 0.0162, 0.0132, 0.0165, 0.0126, 0.0151, 0.0128], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 10:25:00,525 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159220.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:25:09,813 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0626, 1.9900, 2.1411, 1.5001, 2.0160, 2.1483, 2.2001, 1.7422], device='cuda:3'), covar=tensor([0.0589, 0.0696, 0.0687, 0.0863, 0.0779, 0.0704, 0.0641, 0.1160], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0139, 0.0141, 0.0120, 0.0129, 0.0140, 0.0141, 0.0164], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 10:25:20,093 INFO [finetune.py:976] (3/7) Epoch 28, batch 4600, loss[loss=0.1469, simple_loss=0.221, pruned_loss=0.03637, over 4915.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2431, pruned_loss=0.04801, over 953045.37 frames. ], batch size: 33, lr: 2.87e-03, grad_scale: 32.0 2023-03-27 10:25:35,683 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.005e+02 1.454e+02 1.668e+02 2.062e+02 3.965e+02, threshold=3.336e+02, percent-clipped=3.0 2023-03-27 10:25:38,702 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2893, 2.0512, 1.8813, 2.2222, 2.6858, 2.3062, 2.1705, 1.7613], device='cuda:3'), covar=tensor([0.1982, 0.1923, 0.1819, 0.1619, 0.1750, 0.1079, 0.1999, 0.1812], device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0215, 0.0217, 0.0202, 0.0248, 0.0193, 0.0219, 0.0207], 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') 2023-03-27 10:25:54,149 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159290.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:26:05,014 INFO [finetune.py:976] (3/7) Epoch 28, batch 4650, loss[loss=0.1426, simple_loss=0.2139, pruned_loss=0.0356, over 4757.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2411, pruned_loss=0.04759, over 952021.52 frames. ], batch size: 28, lr: 2.87e-03, grad_scale: 32.0 2023-03-27 10:26:54,498 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1548, 1.2959, 1.4842, 1.3308, 1.4356, 2.4105, 1.2503, 1.4560], device='cuda:3'), covar=tensor([0.1025, 0.1841, 0.1014, 0.0909, 0.1592, 0.0409, 0.1522, 0.1737], device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0083, 0.0074, 0.0077, 0.0092, 0.0081, 0.0086, 0.0081], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-27 10:26:55,598 INFO [finetune.py:976] (3/7) Epoch 28, batch 4700, loss[loss=0.1376, simple_loss=0.2127, pruned_loss=0.03126, over 4905.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2386, pruned_loss=0.04699, over 954043.19 frames. ], batch size: 37, lr: 2.87e-03, grad_scale: 32.0 2023-03-27 10:26:58,036 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159351.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:27:11,660 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.356e+01 1.407e+02 1.592e+02 1.978e+02 3.098e+02, threshold=3.183e+02, percent-clipped=0.0 2023-03-27 10:27:28,643 INFO [finetune.py:976] (3/7) Epoch 28, batch 4750, loss[loss=0.2097, simple_loss=0.2847, pruned_loss=0.06738, over 4845.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.2379, pruned_loss=0.04725, over 955181.09 frames. ], batch size: 47, lr: 2.87e-03, grad_scale: 32.0 2023-03-27 10:27:46,459 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159425.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:27:51,842 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159433.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:27:55,936 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159439.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:27:56,314 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-03-27 10:28:02,259 INFO [finetune.py:976] (3/7) Epoch 28, batch 4800, loss[loss=0.166, simple_loss=0.241, pruned_loss=0.04548, over 4899.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2408, pruned_loss=0.04791, over 953507.26 frames. ], batch size: 36, lr: 2.86e-03, grad_scale: 32.0 2023-03-27 10:28:18,791 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.120e+02 1.565e+02 1.779e+02 2.195e+02 3.956e+02, threshold=3.558e+02, percent-clipped=1.0 2023-03-27 10:28:23,706 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=159481.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:28:27,815 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=159487.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:28:30,784 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159491.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 10:28:36,823 INFO [finetune.py:976] (3/7) Epoch 28, batch 4850, loss[loss=0.1496, simple_loss=0.2401, pruned_loss=0.02953, over 4770.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2433, pruned_loss=0.04869, over 952708.98 frames. ], batch size: 26, lr: 2.86e-03, grad_scale: 32.0 2023-03-27 10:28:45,267 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.2835, 2.9158, 3.0567, 3.2197, 3.0764, 2.8901, 3.3375, 0.8681], device='cuda:3'), covar=tensor([0.1066, 0.0982, 0.1072, 0.1200, 0.1516, 0.1763, 0.1031, 0.5692], device='cuda:3'), in_proj_covar=tensor([0.0357, 0.0251, 0.0290, 0.0301, 0.0343, 0.0291, 0.0310, 0.0308], 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') 2023-03-27 10:28:57,882 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159515.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:29:13,993 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0 2023-03-27 10:29:17,409 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=159539.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 10:29:19,312 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0716, 1.9778, 1.7411, 1.8350, 1.9605, 1.8753, 1.9656, 2.5731], device='cuda:3'), covar=tensor([0.3851, 0.4092, 0.3176, 0.3675, 0.3991, 0.2469, 0.3597, 0.1703], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0264, 0.0237, 0.0275, 0.0261, 0.0231, 0.0259, 0.0238], 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') 2023-03-27 10:29:23,227 INFO [finetune.py:976] (3/7) Epoch 28, batch 4900, loss[loss=0.1519, simple_loss=0.2321, pruned_loss=0.03585, over 4816.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2452, pruned_loss=0.04923, over 952357.93 frames. ], batch size: 51, lr: 2.86e-03, grad_scale: 32.0 2023-03-27 10:29:40,325 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.030e+02 1.553e+02 1.825e+02 2.271e+02 5.584e+02, threshold=3.651e+02, percent-clipped=3.0 2023-03-27 10:29:56,962 INFO [finetune.py:976] (3/7) Epoch 28, batch 4950, loss[loss=0.1763, simple_loss=0.2474, pruned_loss=0.05262, over 4897.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2458, pruned_loss=0.04939, over 952479.41 frames. ], batch size: 37, lr: 2.86e-03, grad_scale: 32.0 2023-03-27 10:30:18,769 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159631.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:30:28,813 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159646.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:30:29,440 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1560, 1.9416, 1.5853, 0.6066, 1.7553, 1.7629, 1.6240, 1.8554], device='cuda:3'), covar=tensor([0.0845, 0.0787, 0.1329, 0.1930, 0.1155, 0.2390, 0.2216, 0.0787], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0191, 0.0201, 0.0181, 0.0211, 0.0210, 0.0224, 0.0196], 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') 2023-03-27 10:30:29,930 INFO [finetune.py:976] (3/7) Epoch 28, batch 5000, loss[loss=0.1776, simple_loss=0.2492, pruned_loss=0.05305, over 4899.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.244, pruned_loss=0.04882, over 953489.65 frames. ], batch size: 43, lr: 2.86e-03, grad_scale: 32.0 2023-03-27 10:30:47,438 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.023e+02 1.490e+02 1.734e+02 1.957e+02 3.436e+02, threshold=3.469e+02, percent-clipped=0.0 2023-03-27 10:30:59,472 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159692.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:31:05,616 INFO [finetune.py:976] (3/7) Epoch 28, batch 5050, loss[loss=0.1592, simple_loss=0.224, pruned_loss=0.04726, over 4804.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2418, pruned_loss=0.0489, over 954058.21 frames. ], batch size: 51, lr: 2.86e-03, grad_scale: 32.0 2023-03-27 10:31:32,284 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159725.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:31:55,161 INFO [finetune.py:976] (3/7) Epoch 28, batch 5100, loss[loss=0.1745, simple_loss=0.2454, pruned_loss=0.05178, over 4904.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2385, pruned_loss=0.04752, over 953326.94 frames. ], batch size: 46, lr: 2.86e-03, grad_scale: 32.0 2023-03-27 10:32:14,921 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.1100, 1.2990, 1.4581, 1.2700, 1.4497, 2.3054, 1.2686, 1.4596], device='cuda:3'), covar=tensor([0.0873, 0.1579, 0.0879, 0.0804, 0.1446, 0.0389, 0.1309, 0.1556], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0073, 0.0076, 0.0091, 0.0080, 0.0086, 0.0081], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-27 10:32:20,230 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0555, 1.8819, 1.6729, 1.8170, 1.8153, 1.7716, 1.8342, 2.4765], device='cuda:3'), covar=tensor([0.3314, 0.3814, 0.2931, 0.3226, 0.3700, 0.2264, 0.3374, 0.1557], device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0264, 0.0238, 0.0275, 0.0261, 0.0231, 0.0259, 0.0238], 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') 2023-03-27 10:32:21,761 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.008e+02 1.553e+02 1.866e+02 2.180e+02 3.771e+02, threshold=3.731e+02, percent-clipped=1.0 2023-03-27 10:32:21,834 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=159773.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:32:37,624 INFO [finetune.py:976] (3/7) Epoch 28, batch 5150, loss[loss=0.1771, simple_loss=0.2534, pruned_loss=0.05039, over 4903.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2388, pruned_loss=0.04789, over 953381.17 frames. ], batch size: 43, lr: 2.86e-03, grad_scale: 32.0 2023-03-27 10:32:46,269 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3014, 2.2420, 1.9466, 1.0370, 2.1028, 1.7942, 1.6620, 2.1082], device='cuda:3'), covar=tensor([0.0951, 0.0644, 0.1472, 0.1828, 0.1322, 0.1997, 0.2052, 0.0931], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0190, 0.0201, 0.0180, 0.0210, 0.0210, 0.0224, 0.0196], 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') 2023-03-27 10:32:49,257 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159815.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:33:11,651 INFO [finetune.py:976] (3/7) Epoch 28, batch 5200, loss[loss=0.1714, simple_loss=0.2465, pruned_loss=0.0481, over 4009.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2425, pruned_loss=0.04878, over 953821.60 frames. ], batch size: 65, lr: 2.86e-03, grad_scale: 32.0 2023-03-27 10:33:16,460 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159855.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:33:21,714 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=159863.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:33:28,750 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.072e+02 1.499e+02 1.871e+02 2.174e+02 4.313e+02, threshold=3.743e+02, percent-clipped=1.0 2023-03-27 10:33:44,908 INFO [finetune.py:976] (3/7) Epoch 28, batch 5250, loss[loss=0.1694, simple_loss=0.2377, pruned_loss=0.05058, over 4800.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2436, pruned_loss=0.04931, over 950766.51 frames. ], batch size: 51, lr: 2.86e-03, grad_scale: 16.0 2023-03-27 10:33:53,304 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159910.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:33:59,375 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159916.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:34:26,200 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159946.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:34:27,321 INFO [finetune.py:976] (3/7) Epoch 28, batch 5300, loss[loss=0.1727, simple_loss=0.2377, pruned_loss=0.05391, over 4760.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2456, pruned_loss=0.05027, over 949012.07 frames. ], batch size: 26, lr: 2.86e-03, grad_scale: 16.0 2023-03-27 10:34:50,634 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159971.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:34:52,306 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.066e+02 1.498e+02 1.765e+02 2.107e+02 3.764e+02, threshold=3.530e+02, percent-clipped=1.0 2023-03-27 10:35:01,758 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159987.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:35:06,016 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=159994.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:35:08,412 INFO [finetune.py:976] (3/7) Epoch 28, batch 5350, loss[loss=0.1628, simple_loss=0.2459, pruned_loss=0.03989, over 4903.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2464, pruned_loss=0.05057, over 950617.93 frames. ], batch size: 37, lr: 2.86e-03, grad_scale: 16.0 2023-03-27 10:35:12,860 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0630, 2.0172, 1.5694, 0.7562, 1.7522, 1.6598, 1.5693, 1.8569], device='cuda:3'), covar=tensor([0.1114, 0.0727, 0.1478, 0.1789, 0.1269, 0.2825, 0.2569, 0.0843], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0191, 0.0202, 0.0181, 0.0211, 0.0211, 0.0225, 0.0197], 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') 2023-03-27 10:35:26,985 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([5.1243, 4.4525, 4.7199, 5.0020, 4.9034, 4.5579, 5.2567, 1.6960], device='cuda:3'), covar=tensor([0.0658, 0.0860, 0.0755, 0.0819, 0.1140, 0.1466, 0.0525, 0.5652], device='cuda:3'), in_proj_covar=tensor([0.0359, 0.0252, 0.0290, 0.0302, 0.0345, 0.0293, 0.0310, 0.0309], 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') 2023-03-27 10:35:33,821 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5292, 1.5072, 1.3634, 1.5355, 1.8812, 1.8389, 1.6269, 1.4115], device='cuda:3'), covar=tensor([0.0355, 0.0313, 0.0638, 0.0312, 0.0216, 0.0435, 0.0341, 0.0449], device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0106, 0.0148, 0.0111, 0.0102, 0.0116, 0.0104, 0.0114], device='cuda:3'), out_proj_covar=tensor([7.8445e-05, 8.0949e-05, 1.1486e-04, 8.4836e-05, 7.8525e-05, 8.5497e-05, 7.7234e-05, 8.6710e-05], device='cuda:3') 2023-03-27 10:35:43,140 INFO [finetune.py:976] (3/7) Epoch 28, batch 5400, loss[loss=0.1736, simple_loss=0.2393, pruned_loss=0.05393, over 4892.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2435, pruned_loss=0.04962, over 951122.61 frames. ], batch size: 35, lr: 2.86e-03, grad_scale: 16.0 2023-03-27 10:36:00,196 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.280e+01 1.494e+02 1.877e+02 2.215e+02 3.734e+02, threshold=3.755e+02, percent-clipped=1.0 2023-03-27 10:36:16,671 INFO [finetune.py:976] (3/7) Epoch 28, batch 5450, loss[loss=0.1494, simple_loss=0.2259, pruned_loss=0.03643, over 4772.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2401, pruned_loss=0.04857, over 952585.48 frames. ], batch size: 27, lr: 2.86e-03, grad_scale: 16.0 2023-03-27 10:37:07,790 INFO [finetune.py:976] (3/7) Epoch 28, batch 5500, loss[loss=0.1801, simple_loss=0.256, pruned_loss=0.05209, over 4936.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2372, pruned_loss=0.04713, over 955331.49 frames. ], batch size: 33, lr: 2.86e-03, grad_scale: 16.0 2023-03-27 10:37:38,240 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.035e+02 1.497e+02 1.808e+02 2.088e+02 3.346e+02, threshold=3.616e+02, percent-clipped=0.0 2023-03-27 10:37:41,301 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8689, 1.6553, 1.4635, 1.3116, 1.6412, 1.5699, 1.6216, 2.1809], device='cuda:3'), covar=tensor([0.3235, 0.3384, 0.2878, 0.3201, 0.3470, 0.2225, 0.3053, 0.1600], device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0264, 0.0236, 0.0275, 0.0260, 0.0231, 0.0258, 0.0237], 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') 2023-03-27 10:37:55,153 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5729, 1.3573, 2.0519, 3.1256, 1.9968, 2.2689, 0.9323, 2.7332], device='cuda:3'), covar=tensor([0.1698, 0.1480, 0.1319, 0.0585, 0.0883, 0.1489, 0.1787, 0.0427], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0116, 0.0134, 0.0166, 0.0101, 0.0136, 0.0125, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 10:37:55,667 INFO [finetune.py:976] (3/7) Epoch 28, batch 5550, loss[loss=0.1861, simple_loss=0.2665, pruned_loss=0.05289, over 4824.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2393, pruned_loss=0.04778, over 952594.63 frames. ], batch size: 40, lr: 2.86e-03, grad_scale: 16.0 2023-03-27 10:38:03,855 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160211.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:38:09,882 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9409, 1.9273, 1.6103, 2.0172, 2.6887, 2.0268, 2.2140, 1.5379], device='cuda:3'), covar=tensor([0.2195, 0.1904, 0.1926, 0.1676, 0.1679, 0.1199, 0.1895, 0.1832], device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0212, 0.0215, 0.0199, 0.0246, 0.0191, 0.0217, 0.0205], 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') 2023-03-27 10:38:21,291 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-27 10:38:23,020 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-27 10:38:27,210 INFO [finetune.py:976] (3/7) Epoch 28, batch 5600, loss[loss=0.2193, simple_loss=0.2805, pruned_loss=0.079, over 4822.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2413, pruned_loss=0.04771, over 953024.54 frames. ], batch size: 39, lr: 2.86e-03, grad_scale: 16.0 2023-03-27 10:38:37,607 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160266.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:38:40,617 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-27 10:38:42,223 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.097e+02 1.448e+02 1.891e+02 2.366e+02 4.690e+02, threshold=3.782e+02, percent-clipped=2.0 2023-03-27 10:38:49,864 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160287.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:38:56,601 INFO [finetune.py:976] (3/7) Epoch 28, batch 5650, loss[loss=0.2215, simple_loss=0.2875, pruned_loss=0.07775, over 4783.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2439, pruned_loss=0.04801, over 953872.55 frames. ], batch size: 54, lr: 2.86e-03, grad_scale: 16.0 2023-03-27 10:39:01,987 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160307.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 10:39:24,564 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=160335.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:39:25,249 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6083, 1.5441, 1.3837, 1.7218, 1.5701, 1.6290, 1.1527, 1.4305], device='cuda:3'), covar=tensor([0.1898, 0.1683, 0.1631, 0.1369, 0.1502, 0.1114, 0.2116, 0.1676], device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0211, 0.0214, 0.0198, 0.0244, 0.0190, 0.0216, 0.0204], 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') 2023-03-27 10:39:36,257 INFO [finetune.py:976] (3/7) Epoch 28, batch 5700, loss[loss=0.1534, simple_loss=0.2085, pruned_loss=0.0492, over 4383.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2397, pruned_loss=0.04761, over 934480.02 frames. ], batch size: 19, lr: 2.86e-03, grad_scale: 16.0 2023-03-27 10:39:43,110 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-03-27 10:39:48,028 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160368.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 10:39:50,254 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-03-27 10:39:53,238 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.932e+01 1.353e+02 1.695e+02 2.046e+02 5.031e+02, threshold=3.390e+02, percent-clipped=3.0 2023-03-27 10:40:10,861 INFO [finetune.py:976] (3/7) Epoch 29, batch 0, loss[loss=0.1727, simple_loss=0.2473, pruned_loss=0.04901, over 4863.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2473, pruned_loss=0.04901, over 4863.00 frames. ], batch size: 34, lr: 2.86e-03, grad_scale: 16.0 2023-03-27 10:40:10,861 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-27 10:40:21,882 INFO [finetune.py:1010] (3/7) Epoch 29, validation: loss=0.1588, simple_loss=0.2262, pruned_loss=0.04569, over 2265189.00 frames. 2023-03-27 10:40:21,882 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6469MB 2023-03-27 10:40:24,831 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6560, 1.5291, 2.0747, 3.2962, 2.2740, 2.2878, 0.9957, 2.8529], device='cuda:3'), covar=tensor([0.1557, 0.1225, 0.1214, 0.0531, 0.0781, 0.1555, 0.1679, 0.0375], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0116, 0.0134, 0.0165, 0.0101, 0.0136, 0.0125, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 10:40:57,961 INFO [finetune.py:976] (3/7) Epoch 29, batch 50, loss[loss=0.1943, simple_loss=0.2733, pruned_loss=0.05761, over 4730.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2438, pruned_loss=0.04803, over 215393.11 frames. ], batch size: 59, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:41:07,359 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-27 10:41:17,644 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.41 vs. limit=5.0 2023-03-27 10:41:23,846 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0 2023-03-27 10:41:38,854 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.219e+02 1.505e+02 1.852e+02 2.145e+02 8.823e+02, threshold=3.704e+02, percent-clipped=1.0 2023-03-27 10:41:39,936 INFO [finetune.py:976] (3/7) Epoch 29, batch 100, loss[loss=0.1418, simple_loss=0.2161, pruned_loss=0.03373, over 4746.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2374, pruned_loss=0.0464, over 379057.99 frames. ], batch size: 27, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:42:14,759 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160511.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:42:34,272 INFO [finetune.py:976] (3/7) Epoch 29, batch 150, loss[loss=0.1676, simple_loss=0.2364, pruned_loss=0.04942, over 4759.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.236, pruned_loss=0.04671, over 506262.84 frames. ], batch size: 59, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:42:43,894 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8678, 1.7388, 1.5271, 1.5203, 1.9136, 1.6532, 1.8689, 1.8787], device='cuda:3'), covar=tensor([0.1402, 0.1916, 0.3025, 0.2456, 0.2492, 0.1744, 0.2907, 0.1775], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0191, 0.0238, 0.0254, 0.0251, 0.0209, 0.0217, 0.0204], 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') 2023-03-27 10:42:55,945 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=160559.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:43:00,754 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160566.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:43:06,535 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.027e+02 1.446e+02 1.774e+02 2.135e+02 3.412e+02, threshold=3.547e+02, percent-clipped=0.0 2023-03-27 10:43:07,131 INFO [finetune.py:976] (3/7) Epoch 29, batch 200, loss[loss=0.1868, simple_loss=0.25, pruned_loss=0.06179, over 4900.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2359, pruned_loss=0.04696, over 607025.51 frames. ], batch size: 43, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:43:12,402 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2015, 1.8481, 2.6268, 4.1642, 2.9209, 2.7604, 1.1061, 3.5844], device='cuda:3'), covar=tensor([0.1594, 0.1382, 0.1317, 0.0576, 0.0693, 0.1473, 0.1792, 0.0317], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0116, 0.0133, 0.0165, 0.0100, 0.0135, 0.0125, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 10:43:32,501 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=160614.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:43:40,981 INFO [finetune.py:976] (3/7) Epoch 29, batch 250, loss[loss=0.1594, simple_loss=0.2254, pruned_loss=0.04668, over 4762.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2384, pruned_loss=0.04789, over 684349.69 frames. ], batch size: 28, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:44:05,536 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160663.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 10:44:12,915 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.082e+02 1.455e+02 1.868e+02 2.134e+02 3.558e+02, threshold=3.736e+02, percent-clipped=1.0 2023-03-27 10:44:13,967 INFO [finetune.py:976] (3/7) Epoch 29, batch 300, loss[loss=0.1797, simple_loss=0.2508, pruned_loss=0.05433, over 4806.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.242, pruned_loss=0.04833, over 744142.67 frames. ], batch size: 41, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:44:26,385 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-03-27 10:44:36,654 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-27 10:44:57,147 INFO [finetune.py:976] (3/7) Epoch 29, batch 350, loss[loss=0.1931, simple_loss=0.2598, pruned_loss=0.06318, over 4922.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2448, pruned_loss=0.04959, over 788537.79 frames. ], batch size: 42, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:45:20,172 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1709, 2.3940, 2.3145, 1.7636, 2.2053, 2.6463, 2.6344, 2.1258], device='cuda:3'), covar=tensor([0.0652, 0.0636, 0.0751, 0.0831, 0.0986, 0.0706, 0.0522, 0.1008], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0138, 0.0140, 0.0119, 0.0128, 0.0140, 0.0139, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 10:45:32,638 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-27 10:45:37,469 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.928e+01 1.581e+02 1.864e+02 2.210e+02 3.251e+02, threshold=3.728e+02, percent-clipped=0.0 2023-03-27 10:45:38,107 INFO [finetune.py:976] (3/7) Epoch 29, batch 400, loss[loss=0.1629, simple_loss=0.2443, pruned_loss=0.0407, over 4908.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2448, pruned_loss=0.04911, over 824367.26 frames. ], batch size: 36, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:45:59,569 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2301, 1.8089, 1.8322, 0.9319, 2.0446, 2.2065, 2.0605, 1.7935], device='cuda:3'), covar=tensor([0.0810, 0.0674, 0.0560, 0.0647, 0.0599, 0.0570, 0.0486, 0.0645], device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0148, 0.0130, 0.0122, 0.0131, 0.0130, 0.0142, 0.0151], device='cuda:3'), out_proj_covar=tensor([8.7830e-05, 1.0579e-04, 9.2256e-05, 8.5358e-05, 9.1659e-05, 9.1751e-05, 1.0126e-04, 1.0756e-04], device='cuda:3') 2023-03-27 10:46:04,343 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 2023-03-27 10:46:11,857 INFO [finetune.py:976] (3/7) Epoch 29, batch 450, loss[loss=0.1531, simple_loss=0.2249, pruned_loss=0.0406, over 4927.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2439, pruned_loss=0.0486, over 852365.09 frames. ], batch size: 33, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:46:55,073 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.046e+02 1.473e+02 1.674e+02 2.082e+02 4.783e+02, threshold=3.348e+02, percent-clipped=2.0 2023-03-27 10:46:55,687 INFO [finetune.py:976] (3/7) Epoch 29, batch 500, loss[loss=0.1376, simple_loss=0.2111, pruned_loss=0.03203, over 4753.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2395, pruned_loss=0.04725, over 873158.13 frames. ], batch size: 28, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:47:03,665 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-27 10:47:15,116 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160901.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:47:36,926 INFO [finetune.py:976] (3/7) Epoch 29, batch 550, loss[loss=0.1481, simple_loss=0.2217, pruned_loss=0.03729, over 4894.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.237, pruned_loss=0.04644, over 891529.82 frames. ], batch size: 35, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:48:08,213 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160962.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:48:08,821 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160963.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 10:48:10,053 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160965.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:48:11,245 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4861, 1.3069, 1.3622, 1.3388, 1.6663, 1.6208, 1.4538, 1.2652], device='cuda:3'), covar=tensor([0.0348, 0.0341, 0.0646, 0.0341, 0.0255, 0.0406, 0.0381, 0.0479], device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0106, 0.0148, 0.0111, 0.0102, 0.0117, 0.0105, 0.0115], device='cuda:3'), out_proj_covar=tensor([7.8707e-05, 8.1178e-05, 1.1546e-04, 8.4904e-05, 7.8574e-05, 8.5927e-05, 7.7588e-05, 8.6940e-05], device='cuda:3') 2023-03-27 10:48:15,382 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.000e+01 1.387e+02 1.769e+02 2.082e+02 3.377e+02, threshold=3.538e+02, percent-clipped=1.0 2023-03-27 10:48:16,012 INFO [finetune.py:976] (3/7) Epoch 29, batch 600, loss[loss=0.18, simple_loss=0.265, pruned_loss=0.04747, over 4855.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2368, pruned_loss=0.04647, over 906303.27 frames. ], batch size: 44, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:48:41,020 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=161011.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 10:48:41,642 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9516, 1.7560, 2.1559, 1.4229, 1.8921, 2.1680, 1.6536, 2.2864], device='cuda:3'), covar=tensor([0.1266, 0.2114, 0.1442, 0.1925, 0.1145, 0.1238, 0.3021, 0.0856], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0206, 0.0193, 0.0190, 0.0174, 0.0214, 0.0219, 0.0199], 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') 2023-03-27 10:48:44,050 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4227, 2.5114, 2.4666, 1.8883, 2.4117, 2.9294, 2.9686, 2.1948], device='cuda:3'), covar=tensor([0.0602, 0.0572, 0.0727, 0.0887, 0.0899, 0.0629, 0.0517, 0.1058], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0138, 0.0141, 0.0119, 0.0128, 0.0141, 0.0140, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 10:48:49,430 INFO [finetune.py:976] (3/7) Epoch 29, batch 650, loss[loss=0.1686, simple_loss=0.2479, pruned_loss=0.04459, over 4830.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2384, pruned_loss=0.04676, over 916434.38 frames. ], batch size: 33, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:48:50,193 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161026.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:48:57,692 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-03-27 10:49:22,505 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.055e+02 1.587e+02 1.904e+02 2.267e+02 4.706e+02, threshold=3.807e+02, percent-clipped=2.0 2023-03-27 10:49:23,101 INFO [finetune.py:976] (3/7) Epoch 29, batch 700, loss[loss=0.2071, simple_loss=0.2777, pruned_loss=0.06831, over 4923.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2407, pruned_loss=0.04725, over 925700.26 frames. ], batch size: 42, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:50:03,478 INFO [finetune.py:976] (3/7) Epoch 29, batch 750, loss[loss=0.1505, simple_loss=0.2418, pruned_loss=0.02959, over 4736.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.242, pruned_loss=0.04744, over 931341.55 frames. ], batch size: 27, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:50:05,375 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4664, 1.3702, 1.3213, 1.3674, 1.7136, 1.6642, 1.4874, 1.3009], device='cuda:3'), covar=tensor([0.0389, 0.0302, 0.0620, 0.0309, 0.0221, 0.0396, 0.0318, 0.0392], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0107, 0.0150, 0.0113, 0.0103, 0.0118, 0.0105, 0.0116], device='cuda:3'), out_proj_covar=tensor([7.9837e-05, 8.2037e-05, 1.1639e-04, 8.5855e-05, 7.9271e-05, 8.6834e-05, 7.8120e-05, 8.7713e-05], device='cuda:3') 2023-03-27 10:50:46,352 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.613e+01 1.645e+02 1.965e+02 2.346e+02 5.118e+02, threshold=3.931e+02, percent-clipped=1.0 2023-03-27 10:50:46,986 INFO [finetune.py:976] (3/7) Epoch 29, batch 800, loss[loss=0.1777, simple_loss=0.2506, pruned_loss=0.05238, over 4890.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2419, pruned_loss=0.04713, over 936218.26 frames. ], batch size: 43, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:51:15,286 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161216.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:51:16,487 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.2713, 1.4614, 1.8401, 1.5925, 1.5760, 3.0998, 1.4106, 1.6120], device='cuda:3'), covar=tensor([0.1031, 0.1670, 0.1031, 0.0906, 0.1449, 0.0262, 0.1338, 0.1574], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0073, 0.0076, 0.0091, 0.0080, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-27 10:51:18,884 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8201, 2.5583, 2.1219, 1.2162, 2.2683, 2.3575, 2.0690, 2.3608], device='cuda:3'), covar=tensor([0.0761, 0.0716, 0.1519, 0.1725, 0.1118, 0.1651, 0.1872, 0.0810], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0188, 0.0200, 0.0179, 0.0207, 0.0208, 0.0221, 0.0194], 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') 2023-03-27 10:51:20,586 INFO [finetune.py:976] (3/7) Epoch 29, batch 850, loss[loss=0.1484, simple_loss=0.2268, pruned_loss=0.03497, over 4779.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.24, pruned_loss=0.04678, over 939903.37 frames. ], batch size: 29, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:51:24,348 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161231.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:51:48,481 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161257.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:52:04,267 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.030e+02 1.396e+02 1.676e+02 2.040e+02 3.102e+02, threshold=3.353e+02, percent-clipped=0.0 2023-03-27 10:52:04,923 INFO [finetune.py:976] (3/7) Epoch 29, batch 900, loss[loss=0.1644, simple_loss=0.2322, pruned_loss=0.04832, over 4935.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2374, pruned_loss=0.04599, over 941570.85 frames. ], batch size: 38, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:52:06,236 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161277.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:52:15,373 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161292.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:52:35,802 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161321.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:52:38,180 INFO [finetune.py:976] (3/7) Epoch 29, batch 950, loss[loss=0.1596, simple_loss=0.2293, pruned_loss=0.04498, over 4776.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2362, pruned_loss=0.04591, over 945248.15 frames. ], batch size: 29, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:52:56,965 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6643, 1.3201, 0.8469, 1.6428, 2.1956, 1.2934, 1.4669, 1.5616], device='cuda:3'), covar=tensor([0.1629, 0.2256, 0.2035, 0.1304, 0.1961, 0.1979, 0.1571, 0.2203], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0094, 0.0109, 0.0093, 0.0120, 0.0093, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 10:53:00,286 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-03-27 10:53:28,290 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.815e+01 1.511e+02 1.765e+02 2.170e+02 3.612e+02, threshold=3.529e+02, percent-clipped=1.0 2023-03-27 10:53:28,920 INFO [finetune.py:976] (3/7) Epoch 29, batch 1000, loss[loss=0.1992, simple_loss=0.2762, pruned_loss=0.06111, over 4811.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.238, pruned_loss=0.04631, over 946424.58 frames. ], batch size: 40, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:53:48,061 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.0340, 1.8853, 2.0177, 1.0034, 2.3021, 2.4081, 2.1671, 1.8049], device='cuda:3'), covar=tensor([0.0935, 0.0843, 0.0512, 0.0668, 0.0609, 0.0875, 0.0508, 0.0793], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0148, 0.0131, 0.0123, 0.0132, 0.0130, 0.0143, 0.0152], device='cuda:3'), out_proj_covar=tensor([8.8578e-05, 1.0634e-04, 9.3134e-05, 8.6019e-05, 9.2506e-05, 9.2271e-05, 1.0194e-04, 1.0874e-04], device='cuda:3') 2023-03-27 10:53:49,435 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161401.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:54:06,940 INFO [finetune.py:976] (3/7) Epoch 29, batch 1050, loss[loss=0.1402, simple_loss=0.2201, pruned_loss=0.03015, over 4782.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2412, pruned_loss=0.04737, over 950058.95 frames. ], batch size: 26, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:54:13,083 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6760, 1.7773, 1.5872, 1.5961, 2.3295, 2.4209, 2.0271, 1.8713], device='cuda:3'), covar=tensor([0.0498, 0.0401, 0.0685, 0.0399, 0.0288, 0.0574, 0.0394, 0.0479], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0107, 0.0149, 0.0113, 0.0102, 0.0118, 0.0105, 0.0115], device='cuda:3'), out_proj_covar=tensor([7.9509e-05, 8.1921e-05, 1.1587e-04, 8.5688e-05, 7.9166e-05, 8.6512e-05, 7.8162e-05, 8.7466e-05], device='cuda:3') 2023-03-27 10:54:28,549 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161460.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:54:30,822 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161462.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:54:39,302 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.696e+01 1.443e+02 1.825e+02 2.108e+02 3.139e+02, threshold=3.650e+02, percent-clipped=0.0 2023-03-27 10:54:39,917 INFO [finetune.py:976] (3/7) Epoch 29, batch 1100, loss[loss=0.1584, simple_loss=0.2395, pruned_loss=0.03862, over 4903.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2443, pruned_loss=0.04885, over 951687.60 frames. ], batch size: 37, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:55:11,600 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161521.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:55:14,356 INFO [finetune.py:976] (3/7) Epoch 29, batch 1150, loss[loss=0.1708, simple_loss=0.2469, pruned_loss=0.04732, over 4914.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2451, pruned_loss=0.0494, over 952458.24 frames. ], batch size: 42, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:55:15,166 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0 2023-03-27 10:55:33,551 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-27 10:55:39,778 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161557.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:55:50,721 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161572.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:55:51,850 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.769e+01 1.420e+02 1.779e+02 2.194e+02 3.095e+02, threshold=3.558e+02, percent-clipped=0.0 2023-03-27 10:55:52,952 INFO [finetune.py:976] (3/7) Epoch 29, batch 1200, loss[loss=0.1829, simple_loss=0.2627, pruned_loss=0.05152, over 4882.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2432, pruned_loss=0.0487, over 952493.61 frames. ], batch size: 35, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:56:03,031 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161587.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:56:21,541 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-03-27 10:56:21,963 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-27 10:56:22,443 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=161605.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:56:33,596 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161621.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:56:36,405 INFO [finetune.py:976] (3/7) Epoch 29, batch 1250, loss[loss=0.1512, simple_loss=0.2176, pruned_loss=0.04243, over 4907.00 frames. ], tot_loss[loss=0.169, simple_loss=0.2411, pruned_loss=0.04844, over 951995.45 frames. ], batch size: 46, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:56:40,705 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-03-27 10:56:51,966 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6665, 0.7210, 1.7761, 1.6750, 1.5874, 1.4992, 1.5928, 1.7220], device='cuda:3'), covar=tensor([0.3594, 0.3583, 0.3144, 0.3181, 0.4570, 0.3480, 0.4006, 0.2848], device='cuda:3'), in_proj_covar=tensor([0.0271, 0.0251, 0.0271, 0.0300, 0.0300, 0.0277, 0.0307, 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') 2023-03-27 10:57:07,718 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=161669.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:57:07,783 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3831, 2.3198, 1.8387, 0.8787, 1.9893, 1.8880, 1.7770, 2.1093], device='cuda:3'), covar=tensor([0.0945, 0.0651, 0.1486, 0.1950, 0.1220, 0.2196, 0.2058, 0.0799], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0188, 0.0200, 0.0179, 0.0208, 0.0209, 0.0222, 0.0194], 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') 2023-03-27 10:57:07,794 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.3055, 1.4606, 1.5234, 0.8628, 1.5323, 1.7686, 1.7784, 1.3525], device='cuda:3'), covar=tensor([0.0958, 0.0690, 0.0505, 0.0515, 0.0494, 0.0600, 0.0343, 0.0777], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0148, 0.0131, 0.0122, 0.0132, 0.0130, 0.0143, 0.0152], device='cuda:3'), out_proj_covar=tensor([8.8020e-05, 1.0589e-04, 9.3013e-05, 8.5697e-05, 9.2468e-05, 9.2013e-05, 1.0156e-04, 1.0853e-04], device='cuda:3') 2023-03-27 10:57:10,128 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3938, 1.4878, 1.3000, 1.4868, 1.7682, 1.6874, 1.5183, 1.3255], device='cuda:3'), covar=tensor([0.0396, 0.0336, 0.0626, 0.0318, 0.0240, 0.0588, 0.0336, 0.0498], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0107, 0.0149, 0.0112, 0.0102, 0.0118, 0.0105, 0.0115], device='cuda:3'), out_proj_covar=tensor([7.9511e-05, 8.1784e-05, 1.1567e-04, 8.5501e-05, 7.9208e-05, 8.6742e-05, 7.8247e-05, 8.7668e-05], device='cuda:3') 2023-03-27 10:57:11,726 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 1.453e+02 1.676e+02 2.058e+02 3.499e+02, threshold=3.351e+02, percent-clipped=0.0 2023-03-27 10:57:12,869 INFO [finetune.py:976] (3/7) Epoch 29, batch 1300, loss[loss=0.1609, simple_loss=0.2362, pruned_loss=0.04281, over 4873.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2385, pruned_loss=0.04732, over 950815.62 frames. ], batch size: 34, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:57:47,280 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-03-27 10:57:54,736 INFO [finetune.py:976] (3/7) Epoch 29, batch 1350, loss[loss=0.1758, simple_loss=0.2542, pruned_loss=0.04871, over 4753.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2398, pruned_loss=0.04798, over 953381.66 frames. ], batch size: 27, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:58:29,754 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161757.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:58:49,497 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.001e+02 1.561e+02 1.829e+02 2.338e+02 6.236e+02, threshold=3.658e+02, percent-clipped=4.0 2023-03-27 10:58:50,570 INFO [finetune.py:976] (3/7) Epoch 29, batch 1400, loss[loss=0.1601, simple_loss=0.2267, pruned_loss=0.04672, over 4782.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2421, pruned_loss=0.04871, over 952214.93 frames. ], batch size: 26, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:59:17,535 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161815.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:59:18,091 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161816.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:59:23,988 INFO [finetune.py:976] (3/7) Epoch 29, batch 1450, loss[loss=0.1167, simple_loss=0.1779, pruned_loss=0.02778, over 4456.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2431, pruned_loss=0.04878, over 950941.22 frames. ], batch size: 19, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:59:33,487 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.28 vs. limit=5.0 2023-03-27 10:59:55,405 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161872.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 10:59:56,505 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.032e+02 1.509e+02 1.810e+02 2.171e+02 4.074e+02, threshold=3.620e+02, percent-clipped=1.0 2023-03-27 10:59:57,114 INFO [finetune.py:976] (3/7) Epoch 29, batch 1500, loss[loss=0.1447, simple_loss=0.2215, pruned_loss=0.03393, over 4781.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2441, pruned_loss=0.04919, over 951517.32 frames. ], batch size: 26, lr: 2.85e-03, grad_scale: 16.0 2023-03-27 10:59:58,327 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161876.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:00:05,889 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161887.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:00:27,816 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=161920.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:00:32,889 INFO [finetune.py:976] (3/7) Epoch 29, batch 1550, loss[loss=0.1214, simple_loss=0.1953, pruned_loss=0.02375, over 4694.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2446, pruned_loss=0.04913, over 951256.44 frames. ], batch size: 23, lr: 2.85e-03, grad_scale: 32.0 2023-03-27 11:00:44,438 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=161935.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:01:16,695 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.053e+02 1.413e+02 1.667e+02 1.984e+02 3.261e+02, threshold=3.334e+02, percent-clipped=0.0 2023-03-27 11:01:17,326 INFO [finetune.py:976] (3/7) Epoch 29, batch 1600, loss[loss=0.1572, simple_loss=0.2348, pruned_loss=0.03975, over 4921.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2424, pruned_loss=0.0485, over 952078.84 frames. ], batch size: 38, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:01:53,799 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-03-27 11:01:57,790 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9452, 1.7798, 1.5416, 1.3953, 1.9491, 1.6891, 1.8073, 1.9173], device='cuda:3'), covar=tensor([0.1351, 0.1976, 0.2981, 0.2588, 0.2598, 0.1771, 0.2804, 0.1768], device='cuda:3'), in_proj_covar=tensor([0.0189, 0.0190, 0.0236, 0.0252, 0.0249, 0.0208, 0.0215, 0.0203], 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') 2023-03-27 11:01:59,455 INFO [finetune.py:976] (3/7) Epoch 29, batch 1650, loss[loss=0.1633, simple_loss=0.239, pruned_loss=0.04379, over 4907.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2406, pruned_loss=0.04802, over 952672.70 frames. ], batch size: 32, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:02:22,371 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162057.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:02:34,952 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.499e+01 1.468e+02 1.758e+02 2.171e+02 4.899e+02, threshold=3.516e+02, percent-clipped=3.0 2023-03-27 11:02:35,583 INFO [finetune.py:976] (3/7) Epoch 29, batch 1700, loss[loss=0.1795, simple_loss=0.2608, pruned_loss=0.04912, over 4891.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2378, pruned_loss=0.04702, over 954227.56 frames. ], batch size: 35, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:02:36,905 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162077.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:03:01,317 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=162105.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:03:08,003 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162116.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:03:09,241 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4233, 1.3037, 1.3047, 1.3244, 0.9657, 2.2143, 0.8544, 1.2359], device='cuda:3'), covar=tensor([0.3343, 0.2575, 0.2219, 0.2537, 0.1789, 0.0375, 0.2898, 0.1368], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0121, 0.0124, 0.0113, 0.0096, 0.0093, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 11:03:16,124 INFO [finetune.py:976] (3/7) Epoch 29, batch 1750, loss[loss=0.1859, simple_loss=0.2669, pruned_loss=0.05248, over 4822.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2406, pruned_loss=0.04784, over 954399.92 frames. ], batch size: 40, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:03:18,004 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8120, 1.3002, 0.7636, 1.6720, 2.1875, 1.2406, 1.5601, 1.6277], device='cuda:3'), covar=tensor([0.1402, 0.1990, 0.1902, 0.1142, 0.1799, 0.1847, 0.1308, 0.1913], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0093, 0.0108, 0.0093, 0.0119, 0.0092, 0.0097, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-27 11:03:24,594 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162138.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:03:59,951 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=162164.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:04:04,278 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162171.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:04:10,046 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.681e+01 1.626e+02 1.856e+02 2.284e+02 4.131e+02, threshold=3.712e+02, percent-clipped=2.0 2023-03-27 11:04:10,644 INFO [finetune.py:976] (3/7) Epoch 29, batch 1800, loss[loss=0.1876, simple_loss=0.2708, pruned_loss=0.05218, over 4246.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2433, pruned_loss=0.04824, over 954768.56 frames. ], batch size: 65, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:04:16,238 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.0725, 1.2174, 1.3539, 1.2937, 1.3968, 2.4417, 1.2345, 1.3762], device='cuda:3'), covar=tensor([0.1108, 0.1982, 0.1169, 0.0984, 0.1759, 0.0344, 0.1582, 0.1973], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0073, 0.0076, 0.0091, 0.0080, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-27 11:04:30,453 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6114, 1.1699, 0.8797, 1.4931, 2.0540, 1.1156, 1.4000, 1.5164], device='cuda:3'), covar=tensor([0.1372, 0.1973, 0.1702, 0.1142, 0.1808, 0.1809, 0.1351, 0.1837], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0093, 0.0108, 0.0093, 0.0119, 0.0092, 0.0097, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:3') 2023-03-27 11:04:40,279 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5519, 1.9969, 2.7657, 1.8072, 2.4065, 2.6971, 1.7577, 2.7463], device='cuda:3'), covar=tensor([0.1083, 0.2033, 0.1457, 0.2103, 0.0862, 0.1233, 0.2893, 0.0758], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0204, 0.0191, 0.0187, 0.0173, 0.0211, 0.0216, 0.0196], 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') 2023-03-27 11:04:44,452 INFO [finetune.py:976] (3/7) Epoch 29, batch 1850, loss[loss=0.166, simple_loss=0.245, pruned_loss=0.04354, over 4830.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2446, pruned_loss=0.04907, over 956625.30 frames. ], batch size: 47, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:04:48,179 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162231.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:05:17,381 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.145e+02 1.604e+02 1.802e+02 2.267e+02 3.875e+02, threshold=3.605e+02, percent-clipped=1.0 2023-03-27 11:05:17,839 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-03-27 11:05:17,994 INFO [finetune.py:976] (3/7) Epoch 29, batch 1900, loss[loss=0.1326, simple_loss=0.2183, pruned_loss=0.02348, over 4807.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2456, pruned_loss=0.04923, over 956585.78 frames. ], batch size: 38, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:05:28,436 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162292.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:05:39,671 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-27 11:05:51,672 INFO [finetune.py:976] (3/7) Epoch 29, batch 1950, loss[loss=0.1766, simple_loss=0.2388, pruned_loss=0.05717, over 4817.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2444, pruned_loss=0.04882, over 957544.86 frames. ], batch size: 30, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:06:28,896 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8488, 2.8378, 2.8578, 1.8922, 2.8427, 3.1601, 3.1560, 2.5063], device='cuda:3'), covar=tensor([0.0512, 0.0536, 0.0587, 0.0798, 0.0547, 0.0591, 0.0477, 0.0931], device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0136, 0.0138, 0.0117, 0.0126, 0.0138, 0.0138, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 11:06:36,623 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.724e+01 1.430e+02 1.855e+02 2.130e+02 3.474e+02, threshold=3.709e+02, percent-clipped=0.0 2023-03-27 11:06:37,251 INFO [finetune.py:976] (3/7) Epoch 29, batch 2000, loss[loss=0.1745, simple_loss=0.2516, pruned_loss=0.0487, over 4895.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2412, pruned_loss=0.04776, over 957593.89 frames. ], batch size: 35, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:07:03,745 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.55 vs. limit=5.0 2023-03-27 11:07:14,705 INFO [finetune.py:976] (3/7) Epoch 29, batch 2050, loss[loss=0.1403, simple_loss=0.2161, pruned_loss=0.03226, over 4829.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2377, pruned_loss=0.0468, over 956555.94 frames. ], batch size: 39, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:07:19,650 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162433.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:07:45,861 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162471.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:07:47,599 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.116e+01 1.601e+02 1.964e+02 2.330e+02 4.874e+02, threshold=3.928e+02, percent-clipped=3.0 2023-03-27 11:07:48,226 INFO [finetune.py:976] (3/7) Epoch 29, batch 2100, loss[loss=0.1447, simple_loss=0.2142, pruned_loss=0.03762, over 4707.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2369, pruned_loss=0.04628, over 958192.23 frames. ], batch size: 23, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:08:22,564 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162511.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:08:26,836 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.52 vs. limit=2.0 2023-03-27 11:08:28,431 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=162519.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:08:34,950 INFO [finetune.py:976] (3/7) Epoch 29, batch 2150, loss[loss=0.2121, simple_loss=0.2933, pruned_loss=0.06543, over 4815.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2404, pruned_loss=0.04727, over 956762.38 frames. ], batch size: 45, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:08:36,161 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5069, 1.3782, 1.8482, 1.7456, 1.5959, 3.3171, 1.3558, 1.5465], device='cuda:3'), covar=tensor([0.1015, 0.1973, 0.1026, 0.0946, 0.1680, 0.0225, 0.1618, 0.1897], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0073, 0.0076, 0.0091, 0.0080, 0.0085, 0.0080], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-27 11:09:14,963 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162572.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 11:09:18,164 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.053e+02 1.475e+02 1.779e+02 2.116e+02 3.329e+02, threshold=3.558e+02, percent-clipped=0.0 2023-03-27 11:09:18,785 INFO [finetune.py:976] (3/7) Epoch 29, batch 2200, loss[loss=0.1949, simple_loss=0.2624, pruned_loss=0.06369, over 4891.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2432, pruned_loss=0.04851, over 953715.58 frames. ], batch size: 35, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:09:27,173 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162587.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:10:00,795 INFO [finetune.py:976] (3/7) Epoch 29, batch 2250, loss[loss=0.1868, simple_loss=0.2631, pruned_loss=0.05521, over 4889.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.243, pruned_loss=0.04809, over 953224.88 frames. ], batch size: 43, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:10:11,439 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.7130, 1.3533, 1.2912, 0.8192, 1.5577, 1.5587, 1.6270, 1.3200], device='cuda:3'), covar=tensor([0.0749, 0.0611, 0.0613, 0.0503, 0.0508, 0.0620, 0.0384, 0.0603], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0148, 0.0131, 0.0122, 0.0132, 0.0130, 0.0142, 0.0152], device='cuda:3'), out_proj_covar=tensor([8.8233e-05, 1.0599e-04, 9.3324e-05, 8.5728e-05, 9.2231e-05, 9.2130e-05, 1.0064e-04, 1.0841e-04], device='cuda:3') 2023-03-27 11:10:19,418 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5356, 2.1042, 2.9500, 1.6518, 2.3354, 2.7817, 2.0035, 2.7563], device='cuda:3'), covar=tensor([0.1318, 0.2114, 0.1380, 0.2391, 0.1084, 0.1544, 0.2663, 0.0916], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0206, 0.0193, 0.0190, 0.0175, 0.0213, 0.0219, 0.0199], 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') 2023-03-27 11:10:33,519 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.222e+01 1.558e+02 1.861e+02 2.054e+02 5.864e+02, threshold=3.723e+02, percent-clipped=1.0 2023-03-27 11:10:34,118 INFO [finetune.py:976] (3/7) Epoch 29, batch 2300, loss[loss=0.2028, simple_loss=0.2646, pruned_loss=0.07052, over 4817.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.244, pruned_loss=0.04827, over 953416.88 frames. ], batch size: 33, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:10:46,493 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2844, 2.2547, 1.7893, 2.1959, 2.1302, 1.9825, 2.0769, 2.8449], device='cuda:3'), covar=tensor([0.3624, 0.3814, 0.3470, 0.3641, 0.3811, 0.2628, 0.3938, 0.1614], device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0264, 0.0239, 0.0275, 0.0261, 0.0232, 0.0260, 0.0239], 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') 2023-03-27 11:11:09,593 INFO [finetune.py:976] (3/7) Epoch 29, batch 2350, loss[loss=0.1543, simple_loss=0.2285, pruned_loss=0.04, over 4904.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2419, pruned_loss=0.04741, over 954076.76 frames. ], batch size: 36, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:11:20,045 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162733.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:11:35,784 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6932, 1.5449, 1.5219, 1.5631, 1.8929, 1.8797, 1.6624, 1.4569], device='cuda:3'), covar=tensor([0.0419, 0.0332, 0.0557, 0.0342, 0.0255, 0.0400, 0.0358, 0.0433], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0106, 0.0149, 0.0112, 0.0102, 0.0118, 0.0105, 0.0115], device='cuda:3'), out_proj_covar=tensor([7.9357e-05, 8.1320e-05, 1.1577e-04, 8.5163e-05, 7.9087e-05, 8.6584e-05, 7.7741e-05, 8.7363e-05], device='cuda:3') 2023-03-27 11:11:39,137 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.60 vs. limit=5.0 2023-03-27 11:11:50,493 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.467e+01 1.478e+02 1.866e+02 2.325e+02 4.091e+02, threshold=3.732e+02, percent-clipped=2.0 2023-03-27 11:11:51,109 INFO [finetune.py:976] (3/7) Epoch 29, batch 2400, loss[loss=0.14, simple_loss=0.212, pruned_loss=0.03399, over 4730.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2392, pruned_loss=0.04644, over 954107.59 frames. ], batch size: 23, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:11:56,643 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=162781.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:12:33,618 INFO [finetune.py:976] (3/7) Epoch 29, batch 2450, loss[loss=0.214, simple_loss=0.2892, pruned_loss=0.0694, over 4858.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2378, pruned_loss=0.04633, over 955037.83 frames. ], batch size: 47, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:12:35,952 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3901, 1.3419, 1.7241, 2.4850, 1.6103, 2.3236, 1.0480, 2.1951], device='cuda:3'), covar=tensor([0.1820, 0.1410, 0.1114, 0.0720, 0.0981, 0.1077, 0.1537, 0.0583], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0114, 0.0132, 0.0163, 0.0100, 0.0135, 0.0125, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 11:12:35,976 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162827.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:13:02,268 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162867.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 11:13:09,318 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.738e+01 1.481e+02 1.739e+02 2.016e+02 4.521e+02, threshold=3.479e+02, percent-clipped=1.0 2023-03-27 11:13:09,948 INFO [finetune.py:976] (3/7) Epoch 29, batch 2500, loss[loss=0.1878, simple_loss=0.2575, pruned_loss=0.059, over 4787.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2402, pruned_loss=0.0472, over 952289.31 frames. ], batch size: 51, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:13:27,484 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162887.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:13:28,111 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162888.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:13:29,249 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0841, 2.1088, 1.4859, 2.1811, 1.9918, 1.6745, 2.8745, 2.1615], device='cuda:3'), covar=tensor([0.1568, 0.2122, 0.3469, 0.3323, 0.3164, 0.1961, 0.2673, 0.2037], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0191, 0.0236, 0.0253, 0.0251, 0.0209, 0.0216, 0.0204], 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') 2023-03-27 11:13:40,477 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1258, 1.9426, 2.0744, 1.4443, 2.0273, 2.2065, 2.1321, 1.5736], device='cuda:3'), covar=tensor([0.0618, 0.0710, 0.0704, 0.0951, 0.0759, 0.0652, 0.0667, 0.1295], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0138, 0.0140, 0.0119, 0.0128, 0.0140, 0.0140, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 11:13:52,413 INFO [finetune.py:976] (3/7) Epoch 29, batch 2550, loss[loss=0.1673, simple_loss=0.2452, pruned_loss=0.04473, over 4803.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2439, pruned_loss=0.04834, over 952994.03 frames. ], batch size: 51, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:14:01,041 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4894, 1.4043, 1.2402, 1.3665, 1.8151, 1.7299, 1.4995, 1.2832], device='cuda:3'), covar=tensor([0.0349, 0.0383, 0.0666, 0.0384, 0.0206, 0.0505, 0.0331, 0.0435], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0107, 0.0149, 0.0113, 0.0103, 0.0118, 0.0105, 0.0116], device='cuda:3'), out_proj_covar=tensor([7.9631e-05, 8.1608e-05, 1.1610e-04, 8.5718e-05, 7.9392e-05, 8.6870e-05, 7.8210e-05, 8.7732e-05], device='cuda:3') 2023-03-27 11:14:01,593 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=162935.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:14:01,981 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-03-27 11:14:09,138 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.4201, 2.2146, 1.7976, 0.8281, 2.0033, 1.9004, 1.6938, 2.0890], device='cuda:3'), covar=tensor([0.0974, 0.0774, 0.1553, 0.2043, 0.1191, 0.2355, 0.2324, 0.0868], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0187, 0.0201, 0.0180, 0.0208, 0.0209, 0.0222, 0.0194], 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') 2023-03-27 11:14:31,202 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3662, 1.2870, 1.4944, 1.0482, 1.3785, 1.4604, 1.3011, 1.6317], device='cuda:3'), covar=tensor([0.1152, 0.2316, 0.1231, 0.1525, 0.0926, 0.1326, 0.2894, 0.0778], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0206, 0.0193, 0.0189, 0.0174, 0.0212, 0.0218, 0.0197], 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') 2023-03-27 11:14:36,962 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.986e+01 1.516e+02 1.791e+02 2.248e+02 3.208e+02, threshold=3.582e+02, percent-clipped=0.0 2023-03-27 11:14:37,597 INFO [finetune.py:976] (3/7) Epoch 29, batch 2600, loss[loss=0.1789, simple_loss=0.254, pruned_loss=0.0519, over 4821.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2456, pruned_loss=0.04908, over 952659.04 frames. ], batch size: 33, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:15:17,742 INFO [finetune.py:976] (3/7) Epoch 29, batch 2650, loss[loss=0.1708, simple_loss=0.2406, pruned_loss=0.05051, over 4892.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2455, pruned_loss=0.04876, over 951575.18 frames. ], batch size: 35, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:15:51,094 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.228e+01 1.484e+02 1.662e+02 2.061e+02 3.704e+02, threshold=3.324e+02, percent-clipped=1.0 2023-03-27 11:15:51,717 INFO [finetune.py:976] (3/7) Epoch 29, batch 2700, loss[loss=0.1697, simple_loss=0.2454, pruned_loss=0.04698, over 4912.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2431, pruned_loss=0.04781, over 951791.22 frames. ], batch size: 46, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:16:02,227 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163090.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:16:25,137 INFO [finetune.py:976] (3/7) Epoch 29, batch 2750, loss[loss=0.1797, simple_loss=0.2441, pruned_loss=0.05767, over 4822.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2414, pruned_loss=0.04755, over 953671.94 frames. ], batch size: 41, lr: 2.84e-03, grad_scale: 32.0 2023-03-27 11:16:52,297 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163151.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:16:53,567 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-27 11:17:03,863 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=163167.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 11:17:10,458 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.013e+02 1.483e+02 1.714e+02 1.995e+02 3.279e+02, threshold=3.429e+02, percent-clipped=0.0 2023-03-27 11:17:10,474 INFO [finetune.py:976] (3/7) Epoch 29, batch 2800, loss[loss=0.1624, simple_loss=0.2377, pruned_loss=0.04351, over 4879.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2389, pruned_loss=0.04715, over 956050.72 frames. ], batch size: 34, lr: 2.84e-03, grad_scale: 16.0 2023-03-27 11:17:11,203 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6932, 1.6100, 1.5985, 1.6073, 1.1996, 3.5313, 1.3769, 1.7236], device='cuda:3'), covar=tensor([0.3199, 0.2626, 0.2127, 0.2363, 0.1777, 0.0210, 0.2518, 0.1290], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0121, 0.0124, 0.0113, 0.0095, 0.0094, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 11:17:15,460 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=163183.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:17:22,843 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-03-27 11:17:37,946 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=163215.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:17:38,634 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0476, 2.0003, 1.6727, 1.8944, 2.0496, 1.7844, 2.2451, 2.0505], device='cuda:3'), covar=tensor([0.1205, 0.1723, 0.2646, 0.2139, 0.2251, 0.1513, 0.2974, 0.1552], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0191, 0.0237, 0.0254, 0.0251, 0.0209, 0.0217, 0.0204], 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') 2023-03-27 11:17:41,005 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-03-27 11:17:44,448 INFO [finetune.py:976] (3/7) Epoch 29, batch 2850, loss[loss=0.137, simple_loss=0.2171, pruned_loss=0.0284, over 4781.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2363, pruned_loss=0.04563, over 954713.25 frames. ], batch size: 29, lr: 2.84e-03, grad_scale: 16.0 2023-03-27 11:17:59,020 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163240.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:18:31,673 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.921e+01 1.528e+02 1.890e+02 2.266e+02 3.566e+02, threshold=3.779e+02, percent-clipped=1.0 2023-03-27 11:18:31,689 INFO [finetune.py:976] (3/7) Epoch 29, batch 2900, loss[loss=0.1445, simple_loss=0.2231, pruned_loss=0.03296, over 4894.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2401, pruned_loss=0.04715, over 954539.16 frames. ], batch size: 37, lr: 2.84e-03, grad_scale: 16.0 2023-03-27 11:18:52,072 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163301.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:19:05,421 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6981, 1.5910, 1.5103, 1.6539, 1.2436, 3.7419, 1.4971, 2.0951], device='cuda:3'), covar=tensor([0.3221, 0.2468, 0.2079, 0.2319, 0.1759, 0.0177, 0.2446, 0.1104], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0116, 0.0120, 0.0123, 0.0113, 0.0095, 0.0093, 0.0094], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 11:19:08,249 INFO [finetune.py:976] (3/7) Epoch 29, batch 2950, loss[loss=0.1968, simple_loss=0.2692, pruned_loss=0.06224, over 4787.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2434, pruned_loss=0.04814, over 955657.17 frames. ], batch size: 29, lr: 2.84e-03, grad_scale: 16.0 2023-03-27 11:19:33,153 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.1797, 2.8701, 2.9699, 3.0723, 2.9630, 2.7725, 3.2282, 0.9896], device='cuda:3'), covar=tensor([0.1090, 0.1070, 0.1097, 0.1110, 0.1649, 0.1837, 0.1011, 0.4980], device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0245, 0.0286, 0.0296, 0.0339, 0.0286, 0.0305, 0.0302], 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') 2023-03-27 11:19:39,653 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1794, 2.0949, 2.2516, 1.5673, 2.1700, 2.3660, 2.3354, 1.7244], device='cuda:3'), covar=tensor([0.0548, 0.0603, 0.0655, 0.0885, 0.0650, 0.0565, 0.0492, 0.1137], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0137, 0.0139, 0.0118, 0.0127, 0.0138, 0.0139, 0.0161], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 11:19:49,296 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.216e+02 1.530e+02 1.927e+02 2.309e+02 4.929e+02, threshold=3.855e+02, percent-clipped=3.0 2023-03-27 11:19:49,311 INFO [finetune.py:976] (3/7) Epoch 29, batch 3000, loss[loss=0.1644, simple_loss=0.2358, pruned_loss=0.04648, over 4737.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2443, pruned_loss=0.04894, over 954463.05 frames. ], batch size: 23, lr: 2.84e-03, grad_scale: 16.0 2023-03-27 11:19:49,312 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-27 11:19:55,949 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2682, 2.1143, 1.8324, 1.9852, 2.2135, 1.9997, 2.3747, 2.2354], device='cuda:3'), covar=tensor([0.1314, 0.2053, 0.2859, 0.2262, 0.2544, 0.1652, 0.2961, 0.1767], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0191, 0.0238, 0.0255, 0.0251, 0.0210, 0.0217, 0.0205], 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') 2023-03-27 11:20:05,061 INFO [finetune.py:1010] (3/7) Epoch 29, validation: loss=0.158, simple_loss=0.2251, pruned_loss=0.04545, over 2265189.00 frames. 2023-03-27 11:20:05,062 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6469MB 2023-03-27 11:20:17,421 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.0170, 0.9882, 0.9793, 1.0467, 1.1708, 1.1308, 0.9873, 0.9508], device='cuda:3'), covar=tensor([0.0421, 0.0322, 0.0748, 0.0360, 0.0313, 0.0514, 0.0404, 0.0453], device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0107, 0.0150, 0.0112, 0.0103, 0.0118, 0.0106, 0.0116], device='cuda:3'), out_proj_covar=tensor([7.9567e-05, 8.1697e-05, 1.1633e-04, 8.5586e-05, 7.9895e-05, 8.6936e-05, 7.8549e-05, 8.7878e-05], device='cuda:3') 2023-03-27 11:20:43,033 INFO [finetune.py:976] (3/7) Epoch 29, batch 3050, loss[loss=0.1593, simple_loss=0.2316, pruned_loss=0.04355, over 4841.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2449, pruned_loss=0.04877, over 956387.55 frames. ], batch size: 49, lr: 2.84e-03, grad_scale: 16.0 2023-03-27 11:20:57,926 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=163446.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:21:02,226 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6874, 1.5922, 1.5796, 1.5972, 1.0444, 3.3327, 1.2774, 1.6984], device='cuda:3'), covar=tensor([0.2933, 0.2251, 0.1915, 0.2216, 0.1687, 0.0222, 0.2543, 0.1166], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0121, 0.0124, 0.0113, 0.0095, 0.0094, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 11:21:16,336 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.820e+01 1.385e+02 1.622e+02 1.937e+02 3.745e+02, threshold=3.244e+02, percent-clipped=0.0 2023-03-27 11:21:16,351 INFO [finetune.py:976] (3/7) Epoch 29, batch 3100, loss[loss=0.2001, simple_loss=0.2592, pruned_loss=0.07048, over 4823.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2429, pruned_loss=0.04851, over 956159.84 frames. ], batch size: 30, lr: 2.84e-03, grad_scale: 16.0 2023-03-27 11:21:22,349 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=163483.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:21:41,997 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7351, 1.6630, 1.5640, 1.6994, 1.2306, 3.6222, 1.4436, 1.8736], device='cuda:3'), covar=tensor([0.3380, 0.2388, 0.2088, 0.2442, 0.1801, 0.0220, 0.2527, 0.1194], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0116, 0.0121, 0.0124, 0.0113, 0.0095, 0.0094, 0.0095], device='cuda:3'), out_proj_covar=tensor([0.0006, 0.0006, 0.0005, 0.0006, 0.0005, 0.0004, 0.0005, 0.0004], device='cuda:3') 2023-03-27 11:21:51,333 INFO [finetune.py:976] (3/7) Epoch 29, batch 3150, loss[loss=0.1502, simple_loss=0.2209, pruned_loss=0.03979, over 4875.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2392, pruned_loss=0.04695, over 957218.32 frames. ], batch size: 31, lr: 2.84e-03, grad_scale: 16.0 2023-03-27 11:21:55,026 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=163531.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:22:27,153 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-03-27 11:22:33,520 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.004e+02 1.497e+02 1.704e+02 2.227e+02 4.213e+02, threshold=3.407e+02, percent-clipped=5.0 2023-03-27 11:22:33,536 INFO [finetune.py:976] (3/7) Epoch 29, batch 3200, loss[loss=0.2014, simple_loss=0.2644, pruned_loss=0.06922, over 4850.00 frames. ], tot_loss[loss=0.163, simple_loss=0.2357, pruned_loss=0.04518, over 957525.82 frames. ], batch size: 47, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:22:49,214 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=163596.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:22:59,110 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6433, 1.5665, 1.9907, 3.0628, 2.1182, 2.2063, 1.0120, 2.5975], device='cuda:3'), covar=tensor([0.1641, 0.1297, 0.1200, 0.0555, 0.0781, 0.1776, 0.1771, 0.0469], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0115, 0.0133, 0.0164, 0.0100, 0.0135, 0.0125, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 11:23:01,326 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-03-27 11:23:07,473 INFO [finetune.py:976] (3/7) Epoch 29, batch 3250, loss[loss=0.1788, simple_loss=0.2483, pruned_loss=0.05466, over 4938.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2371, pruned_loss=0.04626, over 956866.64 frames. ], batch size: 33, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:23:52,619 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.061e+02 1.629e+02 1.981e+02 2.506e+02 4.570e+02, threshold=3.962e+02, percent-clipped=6.0 2023-03-27 11:23:52,636 INFO [finetune.py:976] (3/7) Epoch 29, batch 3300, loss[loss=0.1726, simple_loss=0.2645, pruned_loss=0.04035, over 4750.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2427, pruned_loss=0.0487, over 956027.38 frames. ], batch size: 54, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:24:29,878 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5280, 1.4889, 1.4293, 1.4399, 1.8472, 1.7583, 1.5762, 1.3603], device='cuda:3'), covar=tensor([0.0377, 0.0358, 0.0620, 0.0367, 0.0239, 0.0484, 0.0373, 0.0508], device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0106, 0.0148, 0.0112, 0.0103, 0.0117, 0.0105, 0.0115], device='cuda:3'), out_proj_covar=tensor([7.9039e-05, 8.1144e-05, 1.1525e-04, 8.4967e-05, 7.9363e-05, 8.6344e-05, 7.7773e-05, 8.7268e-05], device='cuda:3') 2023-03-27 11:24:34,747 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.64 vs. limit=2.0 2023-03-27 11:24:37,005 INFO [finetune.py:976] (3/7) Epoch 29, batch 3350, loss[loss=0.1199, simple_loss=0.1992, pruned_loss=0.02035, over 4702.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2433, pruned_loss=0.04842, over 954470.31 frames. ], batch size: 23, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:24:55,817 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=163746.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:25:21,357 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.117e+02 1.486e+02 1.801e+02 2.174e+02 4.171e+02, threshold=3.603e+02, percent-clipped=1.0 2023-03-27 11:25:21,373 INFO [finetune.py:976] (3/7) Epoch 29, batch 3400, loss[loss=0.1658, simple_loss=0.2579, pruned_loss=0.03685, over 4829.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2439, pruned_loss=0.04883, over 953157.43 frames. ], batch size: 33, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:25:37,708 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=163794.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:25:54,479 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-03-27 11:25:58,981 INFO [finetune.py:976] (3/7) Epoch 29, batch 3450, loss[loss=0.1431, simple_loss=0.2154, pruned_loss=0.0354, over 4808.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2427, pruned_loss=0.04827, over 950988.75 frames. ], batch size: 45, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:26:05,968 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-27 11:26:41,241 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.103e+01 1.539e+02 1.877e+02 2.200e+02 3.171e+02, threshold=3.754e+02, percent-clipped=0.0 2023-03-27 11:26:41,257 INFO [finetune.py:976] (3/7) Epoch 29, batch 3500, loss[loss=0.1667, simple_loss=0.2431, pruned_loss=0.04518, over 4827.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.239, pruned_loss=0.04658, over 952995.59 frames. ], batch size: 38, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:26:55,487 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=163896.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:27:17,139 INFO [finetune.py:976] (3/7) Epoch 29, batch 3550, loss[loss=0.1452, simple_loss=0.2122, pruned_loss=0.03917, over 3997.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2368, pruned_loss=0.04604, over 951246.16 frames. ], batch size: 17, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:27:29,339 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.2756, 2.8347, 2.6831, 1.3993, 2.9418, 2.2510, 2.1995, 2.6923], device='cuda:3'), covar=tensor([0.1041, 0.0840, 0.1720, 0.2326, 0.1355, 0.2567, 0.2211, 0.1095], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0190, 0.0203, 0.0182, 0.0211, 0.0212, 0.0225, 0.0197], 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') 2023-03-27 11:27:38,583 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=163944.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:27:51,230 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.16 vs. limit=5.0 2023-03-27 11:27:59,306 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.875e+01 1.446e+02 1.734e+02 2.162e+02 4.667e+02, threshold=3.468e+02, percent-clipped=1.0 2023-03-27 11:27:59,322 INFO [finetune.py:976] (3/7) Epoch 29, batch 3600, loss[loss=0.1585, simple_loss=0.234, pruned_loss=0.04148, over 4771.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.236, pruned_loss=0.04605, over 951678.43 frames. ], batch size: 27, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:28:36,274 INFO [finetune.py:976] (3/7) Epoch 29, batch 3650, loss[loss=0.2253, simple_loss=0.3003, pruned_loss=0.07512, over 4805.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2394, pruned_loss=0.04772, over 951429.84 frames. ], batch size: 51, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:29:19,111 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.932e+01 1.541e+02 1.828e+02 2.331e+02 7.133e+02, threshold=3.656e+02, percent-clipped=4.0 2023-03-27 11:29:19,127 INFO [finetune.py:976] (3/7) Epoch 29, batch 3700, loss[loss=0.1925, simple_loss=0.263, pruned_loss=0.061, over 4920.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2414, pruned_loss=0.04766, over 952826.66 frames. ], batch size: 38, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:29:36,830 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164090.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:30:00,904 INFO [finetune.py:976] (3/7) Epoch 29, batch 3750, loss[loss=0.1554, simple_loss=0.2244, pruned_loss=0.0432, over 4099.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2419, pruned_loss=0.04745, over 952055.76 frames. ], batch size: 65, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:30:07,064 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3080, 1.5852, 0.7925, 2.0726, 2.4943, 1.8226, 1.8773, 1.8595], device='cuda:3'), covar=tensor([0.1330, 0.1982, 0.2028, 0.1118, 0.1808, 0.1881, 0.1455, 0.1971], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0094, 0.0109, 0.0093, 0.0120, 0.0092, 0.0097, 0.0088], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 11:30:17,242 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164151.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:30:20,044 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164155.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:30:37,150 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.292e+01 1.655e+02 1.832e+02 2.179e+02 3.362e+02, threshold=3.664e+02, percent-clipped=0.0 2023-03-27 11:30:37,165 INFO [finetune.py:976] (3/7) Epoch 29, batch 3800, loss[loss=0.197, simple_loss=0.2743, pruned_loss=0.0598, over 4816.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2439, pruned_loss=0.04849, over 950376.79 frames. ], batch size: 47, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:31:03,884 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164216.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:31:09,782 INFO [finetune.py:976] (3/7) Epoch 29, batch 3850, loss[loss=0.1669, simple_loss=0.2385, pruned_loss=0.04765, over 4899.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2435, pruned_loss=0.04882, over 951709.59 frames. ], batch size: 43, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:31:20,478 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2039, 2.9286, 2.7237, 1.3429, 2.9971, 2.2717, 0.9253, 1.8558], device='cuda:3'), covar=tensor([0.2602, 0.2217, 0.1904, 0.3548, 0.1442, 0.1116, 0.3990, 0.1800], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0180, 0.0161, 0.0131, 0.0163, 0.0124, 0.0150, 0.0126], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 11:31:23,608 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2641, 1.8918, 2.5012, 1.6784, 2.1806, 2.5317, 1.7883, 2.5892], device='cuda:3'), covar=tensor([0.1141, 0.1827, 0.1276, 0.1852, 0.0940, 0.1230, 0.2457, 0.0781], device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0203, 0.0190, 0.0187, 0.0172, 0.0209, 0.0214, 0.0194], 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') 2023-03-27 11:31:45,643 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.748e+01 1.394e+02 1.747e+02 2.221e+02 3.425e+02, threshold=3.494e+02, percent-clipped=0.0 2023-03-27 11:31:45,659 INFO [finetune.py:976] (3/7) Epoch 29, batch 3900, loss[loss=0.1769, simple_loss=0.2522, pruned_loss=0.05083, over 4915.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2414, pruned_loss=0.04784, over 952910.40 frames. ], batch size: 37, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:32:27,497 INFO [finetune.py:976] (3/7) Epoch 29, batch 3950, loss[loss=0.1606, simple_loss=0.2301, pruned_loss=0.0455, over 4759.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2383, pruned_loss=0.0465, over 952433.59 frames. ], batch size: 27, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:32:58,626 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164360.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:32:58,937 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.42 vs. limit=5.0 2023-03-27 11:33:03,097 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9183, 1.9641, 1.6564, 2.2279, 2.4189, 2.2275, 1.8675, 1.5331], device='cuda:3'), covar=tensor([0.2349, 0.1937, 0.1958, 0.1571, 0.1735, 0.1109, 0.2267, 0.2138], device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0212, 0.0217, 0.0199, 0.0246, 0.0191, 0.0218, 0.0206], 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') 2023-03-27 11:33:11,862 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.819e+01 1.491e+02 1.748e+02 1.970e+02 3.605e+02, threshold=3.496e+02, percent-clipped=1.0 2023-03-27 11:33:11,878 INFO [finetune.py:976] (3/7) Epoch 29, batch 4000, loss[loss=0.2036, simple_loss=0.2741, pruned_loss=0.06659, over 4909.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.238, pruned_loss=0.0467, over 953113.36 frames. ], batch size: 36, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:33:42,975 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164421.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:33:45,306 INFO [finetune.py:976] (3/7) Epoch 29, batch 4050, loss[loss=0.144, simple_loss=0.205, pruned_loss=0.04149, over 4716.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2409, pruned_loss=0.04756, over 953794.35 frames. ], batch size: 23, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:34:06,764 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164446.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:34:15,379 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5412, 1.3662, 1.9303, 1.7685, 1.5757, 3.4312, 1.4671, 1.5466], device='cuda:3'), covar=tensor([0.0991, 0.2027, 0.1117, 0.0961, 0.1628, 0.0240, 0.1458, 0.1882], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0083, 0.0073, 0.0077, 0.0091, 0.0081, 0.0086, 0.0081], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-27 11:34:29,030 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.788e+01 1.600e+02 1.814e+02 2.182e+02 3.811e+02, threshold=3.628e+02, percent-clipped=1.0 2023-03-27 11:34:29,045 INFO [finetune.py:976] (3/7) Epoch 29, batch 4100, loss[loss=0.1701, simple_loss=0.2395, pruned_loss=0.05034, over 4808.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2429, pruned_loss=0.04791, over 953602.21 frames. ], batch size: 38, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:35:04,642 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164511.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:35:13,895 INFO [finetune.py:976] (3/7) Epoch 29, batch 4150, loss[loss=0.1371, simple_loss=0.2124, pruned_loss=0.03088, over 4832.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2442, pruned_loss=0.04831, over 954212.82 frames. ], batch size: 30, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:35:42,372 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164568.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:35:46,475 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.025e+02 1.582e+02 1.835e+02 2.360e+02 4.097e+02, threshold=3.670e+02, percent-clipped=1.0 2023-03-27 11:35:46,491 INFO [finetune.py:976] (3/7) Epoch 29, batch 4200, loss[loss=0.2082, simple_loss=0.2676, pruned_loss=0.07443, over 4197.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2454, pruned_loss=0.04855, over 954006.01 frames. ], batch size: 65, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:35:48,885 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164578.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:35:53,158 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.52 vs. limit=5.0 2023-03-27 11:36:20,292 INFO [finetune.py:976] (3/7) Epoch 29, batch 4250, loss[loss=0.1791, simple_loss=0.2546, pruned_loss=0.05178, over 4790.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2416, pruned_loss=0.04679, over 952603.26 frames. ], batch size: 29, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:36:23,232 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164629.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:36:29,275 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164639.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 11:36:29,962 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.62 vs. limit=5.0 2023-03-27 11:36:43,168 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164658.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:36:53,780 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.101e+02 1.450e+02 1.663e+02 2.112e+02 3.483e+02, threshold=3.326e+02, percent-clipped=0.0 2023-03-27 11:36:53,796 INFO [finetune.py:976] (3/7) Epoch 29, batch 4300, loss[loss=0.1767, simple_loss=0.2423, pruned_loss=0.05554, over 4938.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.2394, pruned_loss=0.04655, over 953681.37 frames. ], batch size: 38, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:37:04,230 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.1803, 3.6208, 3.8370, 3.9979, 3.9235, 3.6228, 4.2301, 1.3982], device='cuda:3'), covar=tensor([0.0920, 0.0977, 0.1076, 0.1229, 0.1427, 0.1758, 0.0799, 0.6025], device='cuda:3'), in_proj_covar=tensor([0.0355, 0.0248, 0.0288, 0.0297, 0.0340, 0.0287, 0.0309, 0.0303], 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') 2023-03-27 11:37:39,035 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-03-27 11:37:39,418 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164716.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:37:41,329 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164719.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:37:44,824 INFO [finetune.py:976] (3/7) Epoch 29, batch 4350, loss[loss=0.1326, simple_loss=0.2076, pruned_loss=0.02881, over 4795.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2374, pruned_loss=0.04621, over 954490.03 frames. ], batch size: 29, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:37:56,291 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.6745, 3.1338, 2.9284, 1.6042, 3.0947, 2.6433, 2.5332, 2.8951], device='cuda:3'), covar=tensor([0.0579, 0.0808, 0.1455, 0.1989, 0.1284, 0.1678, 0.1775, 0.0972], device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0191, 0.0205, 0.0183, 0.0212, 0.0213, 0.0226, 0.0199], 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') 2023-03-27 11:37:58,681 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=164746.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:37:58,694 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164746.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:38:13,635 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7412, 1.5413, 2.2329, 3.4405, 2.2356, 2.4719, 0.9908, 3.0184], device='cuda:3'), covar=tensor([0.1665, 0.1405, 0.1208, 0.0636, 0.0816, 0.1322, 0.1864, 0.0402], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0115, 0.0132, 0.0164, 0.0100, 0.0135, 0.0124, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 11:38:20,795 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.196e+01 1.395e+02 1.786e+02 2.060e+02 3.738e+02, threshold=3.572e+02, percent-clipped=2.0 2023-03-27 11:38:20,811 INFO [finetune.py:976] (3/7) Epoch 29, batch 4400, loss[loss=0.1638, simple_loss=0.2435, pruned_loss=0.04201, over 4835.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2384, pruned_loss=0.04713, over 954657.49 frames. ], batch size: 47, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:38:21,829 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.43 vs. limit=5.0 2023-03-27 11:38:33,467 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=164794.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:38:43,353 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164807.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:38:46,286 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=164811.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:38:48,164 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4080, 1.4069, 2.1454, 1.8455, 1.6658, 3.9757, 1.4410, 1.6022], device='cuda:3'), covar=tensor([0.0997, 0.1951, 0.1230, 0.0959, 0.1644, 0.0178, 0.1464, 0.1822], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0083, 0.0073, 0.0077, 0.0092, 0.0081, 0.0086, 0.0081], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-27 11:38:54,769 INFO [finetune.py:976] (3/7) Epoch 29, batch 4450, loss[loss=0.1743, simple_loss=0.2513, pruned_loss=0.04868, over 4773.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2395, pruned_loss=0.0471, over 952760.10 frames. ], batch size: 26, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:39:23,088 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=164859.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:39:28,213 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0873, 1.6000, 2.1800, 1.5782, 1.9458, 2.1349, 1.5807, 2.3375], device='cuda:3'), covar=tensor([0.1068, 0.2104, 0.1516, 0.1861, 0.0896, 0.1333, 0.2737, 0.0770], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0207, 0.0195, 0.0190, 0.0175, 0.0214, 0.0220, 0.0198], 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') 2023-03-27 11:39:29,013 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-03-27 11:39:31,199 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.2501, 1.6790, 2.2089, 2.2373, 2.0295, 1.9944, 2.1336, 2.1556], device='cuda:3'), covar=tensor([0.3979, 0.3787, 0.3242, 0.3287, 0.4624, 0.3669, 0.4645, 0.2884], device='cuda:3'), in_proj_covar=tensor([0.0270, 0.0250, 0.0270, 0.0300, 0.0299, 0.0277, 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') 2023-03-27 11:39:37,138 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.009e+02 1.552e+02 1.889e+02 2.217e+02 4.905e+02, threshold=3.778e+02, percent-clipped=1.0 2023-03-27 11:39:37,153 INFO [finetune.py:976] (3/7) Epoch 29, batch 4500, loss[loss=0.1971, simple_loss=0.2746, pruned_loss=0.05985, over 4879.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2412, pruned_loss=0.0479, over 950725.62 frames. ], batch size: 43, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:40:22,139 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164924.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:40:22,678 INFO [finetune.py:976] (3/7) Epoch 29, batch 4550, loss[loss=0.2352, simple_loss=0.3061, pruned_loss=0.08217, over 4889.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2428, pruned_loss=0.04808, over 951701.42 frames. ], batch size: 35, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:40:28,155 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164934.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 11:40:45,465 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8738, 1.3044, 1.9153, 1.9037, 1.7306, 1.6480, 1.8944, 1.8293], device='cuda:3'), covar=tensor([0.3993, 0.3602, 0.2986, 0.3426, 0.4143, 0.3632, 0.3842, 0.2802], device='cuda:3'), in_proj_covar=tensor([0.0270, 0.0250, 0.0270, 0.0300, 0.0299, 0.0277, 0.0305, 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') 2023-03-27 11:40:47,138 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0524, 1.8610, 1.7175, 1.7766, 1.7198, 1.7037, 1.7880, 2.3876], device='cuda:3'), covar=tensor([0.2899, 0.3106, 0.2661, 0.2787, 0.3430, 0.2017, 0.2824, 0.1403], device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0265, 0.0240, 0.0275, 0.0262, 0.0233, 0.0260, 0.0239], 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') 2023-03-27 11:40:55,994 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.117e+02 1.494e+02 1.764e+02 2.048e+02 3.220e+02, threshold=3.528e+02, percent-clipped=0.0 2023-03-27 11:40:56,010 INFO [finetune.py:976] (3/7) Epoch 29, batch 4600, loss[loss=0.1334, simple_loss=0.2053, pruned_loss=0.03077, over 4828.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2411, pruned_loss=0.04735, over 948502.02 frames. ], batch size: 30, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:41:09,901 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.8999, 1.8148, 1.5638, 1.4717, 1.9260, 1.6762, 1.9008, 1.9298], device='cuda:3'), covar=tensor([0.1431, 0.1868, 0.2965, 0.2493, 0.2669, 0.1761, 0.2809, 0.1773], device='cuda:3'), in_proj_covar=tensor([0.0192, 0.0193, 0.0239, 0.0256, 0.0253, 0.0211, 0.0217, 0.0206], 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') 2023-03-27 11:41:22,215 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165014.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:41:23,409 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165016.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:41:29,265 INFO [finetune.py:976] (3/7) Epoch 29, batch 4650, loss[loss=0.1661, simple_loss=0.2307, pruned_loss=0.05068, over 4826.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2396, pruned_loss=0.04716, over 949693.08 frames. ], batch size: 39, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:41:53,749 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=165064.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:42:01,328 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.062e+02 1.478e+02 1.798e+02 2.114e+02 1.110e+03, threshold=3.597e+02, percent-clipped=3.0 2023-03-27 11:42:01,344 INFO [finetune.py:976] (3/7) Epoch 29, batch 4700, loss[loss=0.1326, simple_loss=0.2067, pruned_loss=0.02928, over 4912.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2375, pruned_loss=0.04607, over 952721.16 frames. ], batch size: 36, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:42:27,136 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165102.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:42:43,744 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165123.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:42:44,888 INFO [finetune.py:976] (3/7) Epoch 29, batch 4750, loss[loss=0.1835, simple_loss=0.256, pruned_loss=0.05552, over 4806.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2368, pruned_loss=0.04674, over 951767.12 frames. ], batch size: 41, lr: 2.83e-03, grad_scale: 16.0 2023-03-27 11:43:21,526 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.139e+02 1.579e+02 1.786e+02 2.031e+02 3.721e+02, threshold=3.572e+02, percent-clipped=1.0 2023-03-27 11:43:21,542 INFO [finetune.py:976] (3/7) Epoch 29, batch 4800, loss[loss=0.1951, simple_loss=0.265, pruned_loss=0.0626, over 4773.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2387, pruned_loss=0.0471, over 952531.50 frames. ], batch size: 54, lr: 2.82e-03, grad_scale: 32.0 2023-03-27 11:43:27,594 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165184.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:43:28,208 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0898, 2.0821, 1.7015, 2.0448, 1.9322, 1.8768, 1.9603, 2.6597], device='cuda:3'), covar=tensor([0.3577, 0.3910, 0.3128, 0.3645, 0.3952, 0.2468, 0.3732, 0.1558], device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0263, 0.0239, 0.0273, 0.0260, 0.0232, 0.0259, 0.0238], 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') 2023-03-27 11:43:53,911 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165224.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:43:54,433 INFO [finetune.py:976] (3/7) Epoch 29, batch 4850, loss[loss=0.1611, simple_loss=0.246, pruned_loss=0.03811, over 4921.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2413, pruned_loss=0.04769, over 952172.73 frames. ], batch size: 42, lr: 2.82e-03, grad_scale: 32.0 2023-03-27 11:44:00,487 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165234.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 11:44:24,478 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=165272.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:44:26,731 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.028e+02 1.742e+02 1.982e+02 2.317e+02 4.079e+02, threshold=3.965e+02, percent-clipped=3.0 2023-03-27 11:44:26,747 INFO [finetune.py:976] (3/7) Epoch 29, batch 4900, loss[loss=0.1753, simple_loss=0.2558, pruned_loss=0.04744, over 4903.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2424, pruned_loss=0.04759, over 953366.76 frames. ], batch size: 36, lr: 2.82e-03, grad_scale: 32.0 2023-03-27 11:44:41,265 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=165282.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:44:57,726 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-03-27 11:45:03,720 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165314.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:45:16,147 INFO [finetune.py:976] (3/7) Epoch 29, batch 4950, loss[loss=0.1571, simple_loss=0.2348, pruned_loss=0.03973, over 4746.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2435, pruned_loss=0.04735, over 953796.98 frames. ], batch size: 27, lr: 2.82e-03, grad_scale: 32.0 2023-03-27 11:45:48,520 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=165362.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:45:51,011 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.6506, 2.4383, 2.1049, 1.0231, 2.2581, 1.9236, 1.8405, 2.3193], device='cuda:3'), covar=tensor([0.0769, 0.0781, 0.1532, 0.2213, 0.1399, 0.2319, 0.2270, 0.0954], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0189, 0.0203, 0.0180, 0.0209, 0.0210, 0.0222, 0.0196], 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') 2023-03-27 11:45:56,865 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.088e+01 1.449e+02 1.743e+02 2.087e+02 3.629e+02, threshold=3.486e+02, percent-clipped=0.0 2023-03-27 11:45:56,881 INFO [finetune.py:976] (3/7) Epoch 29, batch 5000, loss[loss=0.1233, simple_loss=0.1989, pruned_loss=0.02385, over 4267.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2415, pruned_loss=0.04684, over 953973.98 frames. ], batch size: 18, lr: 2.82e-03, grad_scale: 32.0 2023-03-27 11:46:15,382 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165402.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:46:19,106 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-03-27 11:46:30,165 INFO [finetune.py:976] (3/7) Epoch 29, batch 5050, loss[loss=0.2027, simple_loss=0.2736, pruned_loss=0.06597, over 4815.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2394, pruned_loss=0.04614, over 955068.61 frames. ], batch size: 40, lr: 2.82e-03, grad_scale: 32.0 2023-03-27 11:46:47,817 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=165450.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:47:03,406 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.069e+02 1.397e+02 1.700e+02 2.095e+02 3.650e+02, threshold=3.400e+02, percent-clipped=1.0 2023-03-27 11:47:03,422 INFO [finetune.py:976] (3/7) Epoch 29, batch 5100, loss[loss=0.1572, simple_loss=0.2245, pruned_loss=0.04502, over 4700.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2372, pruned_loss=0.04576, over 954009.84 frames. ], batch size: 23, lr: 2.82e-03, grad_scale: 32.0 2023-03-27 11:47:06,335 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165479.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:47:23,354 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5696, 2.3882, 1.9191, 0.8802, 2.1045, 1.9475, 1.8946, 2.1803], device='cuda:3'), covar=tensor([0.0799, 0.0825, 0.1650, 0.2039, 0.1459, 0.2411, 0.2117, 0.0941], device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0190, 0.0203, 0.0181, 0.0209, 0.0211, 0.0223, 0.0197], 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') 2023-03-27 11:47:46,062 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-03-27 11:47:46,452 INFO [finetune.py:976] (3/7) Epoch 29, batch 5150, loss[loss=0.2081, simple_loss=0.2788, pruned_loss=0.06873, over 4804.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2379, pruned_loss=0.04623, over 953054.80 frames. ], batch size: 51, lr: 2.82e-03, grad_scale: 32.0 2023-03-27 11:48:04,762 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165551.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:48:05,556 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.03 vs. limit=5.0 2023-03-27 11:48:20,242 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.818e+01 1.606e+02 1.856e+02 2.249e+02 3.990e+02, threshold=3.713e+02, percent-clipped=3.0 2023-03-27 11:48:20,258 INFO [finetune.py:976] (3/7) Epoch 29, batch 5200, loss[loss=0.2509, simple_loss=0.3109, pruned_loss=0.09544, over 4817.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2408, pruned_loss=0.04716, over 953298.02 frames. ], batch size: 41, lr: 2.82e-03, grad_scale: 32.0 2023-03-27 11:48:45,764 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165612.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:48:53,938 INFO [finetune.py:976] (3/7) Epoch 29, batch 5250, loss[loss=0.1769, simple_loss=0.2624, pruned_loss=0.04568, over 4932.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2419, pruned_loss=0.04745, over 953281.65 frames. ], batch size: 42, lr: 2.82e-03, grad_scale: 32.0 2023-03-27 11:49:26,773 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.319e+01 1.511e+02 1.772e+02 2.250e+02 3.554e+02, threshold=3.544e+02, percent-clipped=0.0 2023-03-27 11:49:26,789 INFO [finetune.py:976] (3/7) Epoch 29, batch 5300, loss[loss=0.1559, simple_loss=0.2257, pruned_loss=0.04301, over 4227.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2436, pruned_loss=0.04824, over 953493.59 frames. ], batch size: 18, lr: 2.82e-03, grad_scale: 32.0 2023-03-27 11:50:10,070 INFO [finetune.py:976] (3/7) Epoch 29, batch 5350, loss[loss=0.1931, simple_loss=0.2607, pruned_loss=0.06275, over 4814.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2434, pruned_loss=0.04797, over 954369.47 frames. ], batch size: 40, lr: 2.82e-03, grad_scale: 32.0 2023-03-27 11:50:33,767 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165750.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:50:55,027 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5622, 1.3377, 1.7174, 2.4926, 1.6883, 2.2977, 0.9067, 2.1802], device='cuda:3'), covar=tensor([0.1605, 0.1288, 0.1052, 0.0694, 0.0909, 0.1104, 0.1519, 0.0555], device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0114, 0.0132, 0.0163, 0.0099, 0.0134, 0.0123, 0.0100], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 11:50:59,727 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.046e+02 1.485e+02 1.788e+02 2.126e+02 3.231e+02, threshold=3.576e+02, percent-clipped=0.0 2023-03-27 11:50:59,743 INFO [finetune.py:976] (3/7) Epoch 29, batch 5400, loss[loss=0.1336, simple_loss=0.201, pruned_loss=0.0331, over 4877.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2409, pruned_loss=0.04736, over 952489.32 frames. ], batch size: 31, lr: 2.82e-03, grad_scale: 32.0 2023-03-27 11:51:01,710 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5629, 1.5557, 1.4721, 1.5650, 1.9356, 1.8943, 1.6192, 1.4026], device='cuda:3'), covar=tensor([0.0415, 0.0367, 0.0643, 0.0321, 0.0223, 0.0478, 0.0393, 0.0463], device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0107, 0.0149, 0.0112, 0.0104, 0.0119, 0.0105, 0.0117], device='cuda:3'), out_proj_covar=tensor([8.0211e-05, 8.1948e-05, 1.1591e-04, 8.5127e-05, 8.0178e-05, 8.7529e-05, 7.8264e-05, 8.8568e-05], device='cuda:3') 2023-03-27 11:51:02,241 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165779.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:51:11,750 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5515, 2.4157, 1.9280, 2.6981, 2.5098, 2.1509, 2.9678, 2.6035], device='cuda:3'), covar=tensor([0.1239, 0.2162, 0.2895, 0.2346, 0.2316, 0.1621, 0.2786, 0.1610], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0192, 0.0239, 0.0254, 0.0252, 0.0210, 0.0217, 0.0205], 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') 2023-03-27 11:51:18,372 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165803.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:51:24,189 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165811.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:51:31,592 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-03-27 11:51:33,003 INFO [finetune.py:976] (3/7) Epoch 29, batch 5450, loss[loss=0.1493, simple_loss=0.2243, pruned_loss=0.03714, over 4770.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.2383, pruned_loss=0.04694, over 953946.92 frames. ], batch size: 28, lr: 2.82e-03, grad_scale: 32.0 2023-03-27 11:51:34,301 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=165827.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:51:42,873 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-03-27 11:51:59,775 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165864.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:52:06,354 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.128e+01 1.456e+02 1.721e+02 2.001e+02 5.868e+02, threshold=3.442e+02, percent-clipped=2.0 2023-03-27 11:52:06,370 INFO [finetune.py:976] (3/7) Epoch 29, batch 5500, loss[loss=0.1768, simple_loss=0.2371, pruned_loss=0.05825, over 4904.00 frames. ], tot_loss[loss=0.1625, simple_loss=0.2345, pruned_loss=0.04523, over 955630.74 frames. ], batch size: 36, lr: 2.82e-03, grad_scale: 32.0 2023-03-27 11:52:27,324 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165907.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:52:40,109 INFO [finetune.py:976] (3/7) Epoch 29, batch 5550, loss[loss=0.1546, simple_loss=0.2321, pruned_loss=0.03858, over 4810.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2366, pruned_loss=0.04597, over 956313.93 frames. ], batch size: 25, lr: 2.82e-03, grad_scale: 32.0 2023-03-27 11:52:43,809 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.5228, 3.9710, 4.1713, 4.3749, 4.2944, 4.0169, 4.6225, 1.4519], device='cuda:3'), covar=tensor([0.0769, 0.0881, 0.0910, 0.0947, 0.1226, 0.1727, 0.0611, 0.6077], device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0248, 0.0288, 0.0298, 0.0338, 0.0288, 0.0306, 0.0304], 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') 2023-03-27 11:53:13,795 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.55 vs. limit=5.0 2023-03-27 11:53:22,711 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.003e+02 1.570e+02 1.926e+02 2.203e+02 4.633e+02, threshold=3.853e+02, percent-clipped=1.0 2023-03-27 11:53:22,727 INFO [finetune.py:976] (3/7) Epoch 29, batch 5600, loss[loss=0.186, simple_loss=0.2652, pruned_loss=0.05335, over 4810.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.2397, pruned_loss=0.04639, over 955314.30 frames. ], batch size: 41, lr: 2.82e-03, grad_scale: 32.0 2023-03-27 11:53:28,795 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-03-27 11:53:54,008 INFO [finetune.py:976] (3/7) Epoch 29, batch 5650, loss[loss=0.2218, simple_loss=0.2891, pruned_loss=0.0773, over 4758.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.244, pruned_loss=0.04766, over 954816.81 frames. ], batch size: 54, lr: 2.82e-03, grad_scale: 32.0 2023-03-27 11:54:11,189 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166054.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 11:54:19,652 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-03-27 11:54:23,584 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.094e+01 1.435e+02 1.719e+02 2.053e+02 3.744e+02, threshold=3.437e+02, percent-clipped=0.0 2023-03-27 11:54:23,600 INFO [finetune.py:976] (3/7) Epoch 29, batch 5700, loss[loss=0.1517, simple_loss=0.2174, pruned_loss=0.04302, over 4173.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2389, pruned_loss=0.04578, over 941720.61 frames. ], batch size: 18, lr: 2.82e-03, grad_scale: 32.0 2023-03-27 11:54:27,892 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.0716, 2.7872, 2.3927, 1.3201, 2.5504, 2.3749, 2.1861, 2.6253], device='cuda:3'), covar=tensor([0.0708, 0.0800, 0.1384, 0.2071, 0.1201, 0.1866, 0.1995, 0.0814], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0188, 0.0201, 0.0180, 0.0207, 0.0209, 0.0221, 0.0195], 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') 2023-03-27 11:54:50,334 INFO [finetune.py:976] (3/7) Epoch 30, batch 0, loss[loss=0.2227, simple_loss=0.2826, pruned_loss=0.08142, over 4914.00 frames. ], tot_loss[loss=0.2227, simple_loss=0.2826, pruned_loss=0.08142, over 4914.00 frames. ], batch size: 33, lr: 2.82e-03, grad_scale: 32.0 2023-03-27 11:54:50,334 INFO [finetune.py:1001] (3/7) Computing validation loss 2023-03-27 11:54:52,276 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0152, 1.2534, 2.1440, 2.0437, 1.9327, 1.8427, 1.8943, 2.0730], device='cuda:3'), covar=tensor([0.3819, 0.4131, 0.3467, 0.3913, 0.4958, 0.3714, 0.4507, 0.2910], device='cuda:3'), in_proj_covar=tensor([0.0270, 0.0250, 0.0269, 0.0301, 0.0300, 0.0277, 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') 2023-03-27 11:55:03,949 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1639, 1.9950, 1.9470, 1.8600, 1.9197, 1.9644, 1.9686, 2.5928], device='cuda:3'), covar=tensor([0.3450, 0.4427, 0.3238, 0.3529, 0.3586, 0.2435, 0.3466, 0.1554], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0264, 0.0239, 0.0274, 0.0261, 0.0232, 0.0259, 0.0239], 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') 2023-03-27 11:55:06,736 INFO [finetune.py:1010] (3/7) Epoch 30, validation: loss=0.1598, simple_loss=0.2264, pruned_loss=0.04658, over 2265189.00 frames. 2023-03-27 11:55:06,737 INFO [finetune.py:1011] (3/7) Maximum memory allocated so far is 6469MB 2023-03-27 11:55:11,870 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166106.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:55:21,158 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166115.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 11:55:43,738 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166147.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:55:52,370 INFO [finetune.py:976] (3/7) Epoch 30, batch 50, loss[loss=0.1756, simple_loss=0.2429, pruned_loss=0.05417, over 4769.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2487, pruned_loss=0.05173, over 218001.77 frames. ], batch size: 26, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 11:56:02,575 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166159.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:56:11,892 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7324, 1.2572, 0.8120, 1.6418, 2.1582, 1.5577, 1.4149, 1.6834], device='cuda:3'), covar=tensor([0.1385, 0.1992, 0.1959, 0.1109, 0.1807, 0.1787, 0.1346, 0.1828], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0094, 0.0109, 0.0093, 0.0120, 0.0092, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 11:56:16,046 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.060e+01 1.403e+02 1.686e+02 1.983e+02 3.736e+02, threshold=3.372e+02, percent-clipped=1.0 2023-03-27 11:56:34,673 INFO [finetune.py:976] (3/7) Epoch 30, batch 100, loss[loss=0.2137, simple_loss=0.2816, pruned_loss=0.0729, over 4902.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2398, pruned_loss=0.04805, over 381407.76 frames. ], batch size: 43, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 11:56:38,187 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166207.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:56:39,342 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166208.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:57:03,065 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5975, 3.4909, 3.2475, 1.3428, 3.4967, 2.6749, 0.8866, 2.4632], device='cuda:3'), covar=tensor([0.2394, 0.1803, 0.1782, 0.3986, 0.1239, 0.1064, 0.4390, 0.1637], device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0179, 0.0160, 0.0131, 0.0163, 0.0124, 0.0149, 0.0126], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 11:57:07,204 INFO [finetune.py:976] (3/7) Epoch 30, batch 150, loss[loss=0.1594, simple_loss=0.2294, pruned_loss=0.04467, over 4883.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2353, pruned_loss=0.04674, over 510135.47 frames. ], batch size: 35, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 11:57:08,957 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=166255.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:57:21,829 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.042e+02 1.442e+02 1.804e+02 2.095e+02 4.016e+02, threshold=3.609e+02, percent-clipped=1.0 2023-03-27 11:57:39,804 INFO [finetune.py:976] (3/7) Epoch 30, batch 200, loss[loss=0.1215, simple_loss=0.1892, pruned_loss=0.02688, over 4766.00 frames. ], tot_loss[loss=0.1623, simple_loss=0.2333, pruned_loss=0.04568, over 609827.83 frames. ], batch size: 23, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 11:57:42,843 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166307.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:58:14,802 INFO [finetune.py:976] (3/7) Epoch 30, batch 250, loss[loss=0.1729, simple_loss=0.2565, pruned_loss=0.04462, over 4808.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2371, pruned_loss=0.04654, over 686684.18 frames. ], batch size: 51, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 11:58:26,039 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166368.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:58:30,118 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.080e+02 1.508e+02 1.830e+02 2.341e+02 4.348e+02, threshold=3.661e+02, percent-clipped=2.0 2023-03-27 11:58:48,139 INFO [finetune.py:976] (3/7) Epoch 30, batch 300, loss[loss=0.1761, simple_loss=0.2626, pruned_loss=0.0448, over 4902.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2429, pruned_loss=0.04866, over 747750.09 frames. ], batch size: 35, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 11:58:50,084 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166406.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:58:51,690 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-03-27 11:58:52,503 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166410.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 11:59:21,479 INFO [finetune.py:976] (3/7) Epoch 30, batch 350, loss[loss=0.1483, simple_loss=0.2289, pruned_loss=0.03383, over 4789.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.245, pruned_loss=0.049, over 794836.16 frames. ], batch size: 29, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 11:59:22,659 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=166454.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:59:25,723 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166459.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:59:37,684 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 1.476e+02 1.813e+02 2.161e+02 3.890e+02, threshold=3.626e+02, percent-clipped=2.0 2023-03-27 11:59:41,439 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166481.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:59:55,115 INFO [finetune.py:976] (3/7) Epoch 30, batch 400, loss[loss=0.1851, simple_loss=0.261, pruned_loss=0.05462, over 4835.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2444, pruned_loss=0.04803, over 831301.44 frames. ], batch size: 47, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 11:59:55,185 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166503.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:59:58,013 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=166507.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:59:59,263 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166509.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 11:59:59,338 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-27 12:00:20,215 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166531.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:00:35,348 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166542.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:00:45,991 INFO [finetune.py:976] (3/7) Epoch 30, batch 450, loss[loss=0.1398, simple_loss=0.2133, pruned_loss=0.03312, over 4753.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2421, pruned_loss=0.04738, over 856291.24 frames. ], batch size: 27, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 12:00:57,399 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166570.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:01:00,793 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.283e+01 1.415e+02 1.694e+02 2.083e+02 4.695e+02, threshold=3.388e+02, percent-clipped=2.0 2023-03-27 12:01:22,345 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166592.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:01:32,866 INFO [finetune.py:976] (3/7) Epoch 30, batch 500, loss[loss=0.1487, simple_loss=0.2217, pruned_loss=0.03788, over 4932.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2397, pruned_loss=0.0469, over 879028.66 frames. ], batch size: 33, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 12:01:57,904 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0519, 1.6884, 2.3053, 1.5218, 2.0769, 2.1959, 1.6315, 2.2656], device='cuda:3'), covar=tensor([0.1188, 0.2033, 0.1348, 0.1899, 0.0860, 0.1348, 0.2860, 0.0804], device='cuda:3'), in_proj_covar=tensor([0.0191, 0.0207, 0.0193, 0.0189, 0.0174, 0.0212, 0.0217, 0.0198], 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') 2023-03-27 12:02:06,081 INFO [finetune.py:976] (3/7) Epoch 30, batch 550, loss[loss=0.1317, simple_loss=0.2038, pruned_loss=0.02982, over 4334.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2363, pruned_loss=0.04601, over 896645.33 frames. ], batch size: 19, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 12:02:12,189 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166663.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:02:20,414 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 7.871e+01 1.519e+02 1.786e+02 2.255e+02 3.834e+02, threshold=3.573e+02, percent-clipped=3.0 2023-03-27 12:02:26,361 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6846, 1.4790, 2.2179, 3.2511, 2.1985, 2.4223, 1.7136, 2.8024], device='cuda:3'), covar=tensor([0.1733, 0.1364, 0.1158, 0.0624, 0.0801, 0.1437, 0.1242, 0.0476], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0115, 0.0133, 0.0164, 0.0100, 0.0135, 0.0125, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 12:02:39,364 INFO [finetune.py:976] (3/7) Epoch 30, batch 600, loss[loss=0.1561, simple_loss=0.2341, pruned_loss=0.03907, over 4459.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2379, pruned_loss=0.04671, over 911604.37 frames. ], batch size: 19, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 12:02:43,710 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166710.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 12:02:45,125 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-03-27 12:03:00,926 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166735.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:03:12,732 INFO [finetune.py:976] (3/7) Epoch 30, batch 650, loss[loss=0.1655, simple_loss=0.2323, pruned_loss=0.04929, over 4176.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2399, pruned_loss=0.0474, over 919410.35 frames. ], batch size: 65, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 12:03:15,790 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=166758.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 12:03:19,478 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166764.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:03:21,888 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0630, 1.7819, 2.0829, 2.0733, 1.8203, 1.8482, 2.0979, 2.0201], device='cuda:3'), covar=tensor([0.4336, 0.3989, 0.3259, 0.4128, 0.5211, 0.4494, 0.4805, 0.2906], device='cuda:3'), in_proj_covar=tensor([0.0271, 0.0251, 0.0270, 0.0301, 0.0301, 0.0279, 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') 2023-03-27 12:03:26,540 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.144e+02 1.552e+02 1.821e+02 2.218e+02 3.611e+02, threshold=3.642e+02, percent-clipped=1.0 2023-03-27 12:03:41,563 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166796.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:03:45,812 INFO [finetune.py:976] (3/7) Epoch 30, batch 700, loss[loss=0.1593, simple_loss=0.2387, pruned_loss=0.03993, over 4842.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2417, pruned_loss=0.04763, over 927538.11 frames. ], batch size: 47, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 12:03:45,917 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166803.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:04:00,343 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166825.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:04:08,129 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166837.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:04:18,070 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=166851.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:04:19,224 INFO [finetune.py:976] (3/7) Epoch 30, batch 750, loss[loss=0.1237, simple_loss=0.1883, pruned_loss=0.02959, over 4116.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2427, pruned_loss=0.04819, over 933818.60 frames. ], batch size: 18, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 12:04:27,146 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166865.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:04:33,237 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.605e+01 1.409e+02 1.752e+02 2.054e+02 3.389e+02, threshold=3.504e+02, percent-clipped=0.0 2023-03-27 12:04:41,695 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166887.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:04:52,671 INFO [finetune.py:976] (3/7) Epoch 30, batch 800, loss[loss=0.2141, simple_loss=0.2644, pruned_loss=0.08193, over 4855.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2415, pruned_loss=0.04752, over 936868.98 frames. ], batch size: 44, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 12:05:09,486 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166929.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:05:21,685 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.5950, 2.4286, 2.1338, 1.0320, 2.2304, 2.0117, 1.8797, 2.3502], device='cuda:3'), covar=tensor([0.0806, 0.0772, 0.1791, 0.2098, 0.1451, 0.2052, 0.2043, 0.0959], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0189, 0.0201, 0.0180, 0.0209, 0.0210, 0.0223, 0.0196], 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') 2023-03-27 12:05:32,104 INFO [finetune.py:976] (3/7) Epoch 30, batch 850, loss[loss=0.2082, simple_loss=0.2673, pruned_loss=0.07456, over 4809.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2404, pruned_loss=0.04752, over 942079.32 frames. ], batch size: 41, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 12:05:35,542 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.09 vs. limit=5.0 2023-03-27 12:05:42,368 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166963.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:05:52,745 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7673, 1.2686, 0.8962, 1.6161, 2.1920, 1.4474, 1.5778, 1.5722], device='cuda:3'), covar=tensor([0.1503, 0.2384, 0.2098, 0.1373, 0.1915, 0.1935, 0.1585, 0.2117], device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0094, 0.0109, 0.0093, 0.0119, 0.0092, 0.0097, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 12:05:53,841 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.050e+02 1.507e+02 1.769e+02 2.105e+02 4.355e+02, threshold=3.539e+02, percent-clipped=2.0 2023-03-27 12:06:07,547 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166990.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:06:11,718 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([0.3305, 1.4220, 1.5229, 0.7976, 1.5248, 1.7543, 1.7730, 1.3749], device='cuda:3'), covar=tensor([0.0880, 0.0588, 0.0500, 0.0474, 0.0412, 0.0540, 0.0295, 0.0732], device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0146, 0.0131, 0.0121, 0.0131, 0.0130, 0.0141, 0.0151], device='cuda:3'), out_proj_covar=tensor([8.8062e-05, 1.0470e-04, 9.2522e-05, 8.4448e-05, 9.1761e-05, 9.2035e-05, 9.9985e-05, 1.0798e-04], device='cuda:3') 2023-03-27 12:06:16,006 INFO [finetune.py:976] (3/7) Epoch 30, batch 900, loss[loss=0.162, simple_loss=0.2401, pruned_loss=0.04199, over 4858.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2383, pruned_loss=0.04728, over 943110.08 frames. ], batch size: 49, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 12:06:23,180 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=167011.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:06:59,307 INFO [finetune.py:976] (3/7) Epoch 30, batch 950, loss[loss=0.1747, simple_loss=0.2433, pruned_loss=0.05307, over 4822.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2368, pruned_loss=0.04677, over 947665.86 frames. ], batch size: 41, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 12:07:09,145 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.18 vs. limit=5.0 2023-03-27 12:07:13,660 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.122e+02 1.522e+02 1.848e+02 2.259e+02 3.601e+02, threshold=3.696e+02, percent-clipped=1.0 2023-03-27 12:07:24,435 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167091.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:07:30,894 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=4.25 vs. limit=5.0 2023-03-27 12:07:31,401 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167100.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:07:33,103 INFO [finetune.py:976] (3/7) Epoch 30, batch 1000, loss[loss=0.1817, simple_loss=0.2421, pruned_loss=0.06071, over 4833.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2388, pruned_loss=0.04714, over 947716.47 frames. ], batch size: 30, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 12:07:33,849 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167104.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:07:44,461 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167120.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:07:52,689 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.64 vs. limit=5.0 2023-03-27 12:07:55,375 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167137.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:08:01,849 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-03-27 12:08:06,884 INFO [finetune.py:976] (3/7) Epoch 30, batch 1050, loss[loss=0.1748, simple_loss=0.2533, pruned_loss=0.04812, over 4802.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2425, pruned_loss=0.04812, over 949647.89 frames. ], batch size: 51, lr: 2.81e-03, grad_scale: 64.0 2023-03-27 12:08:09,424 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5837, 1.5147, 2.0009, 3.0191, 2.0338, 2.2881, 1.0540, 2.6748], device='cuda:3'), covar=tensor([0.1725, 0.1312, 0.1200, 0.0554, 0.0795, 0.1199, 0.1745, 0.0509], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0115, 0.0133, 0.0164, 0.0100, 0.0135, 0.0125, 0.0101], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 12:08:11,867 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167161.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:08:14,777 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167165.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:08:14,801 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167165.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:08:21,246 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.094e+02 1.490e+02 1.818e+02 2.230e+02 3.401e+02, threshold=3.636e+02, percent-clipped=0.0 2023-03-27 12:08:27,352 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=167185.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:08:28,586 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167187.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:08:39,896 INFO [finetune.py:976] (3/7) Epoch 30, batch 1100, loss[loss=0.1909, simple_loss=0.2685, pruned_loss=0.05663, over 4912.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2444, pruned_loss=0.04885, over 952574.88 frames. ], batch size: 36, lr: 2.81e-03, grad_scale: 64.0 2023-03-27 12:08:45,818 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4915, 1.4036, 2.2322, 1.7727, 1.6765, 3.6053, 1.2857, 1.4689], device='cuda:3'), covar=tensor([0.1056, 0.2084, 0.1405, 0.1070, 0.1703, 0.0271, 0.1768, 0.2139], device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0083, 0.0073, 0.0076, 0.0092, 0.0081, 0.0086, 0.0081], device='cuda:3'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0005], device='cuda:3') 2023-03-27 12:08:46,971 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=167213.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:09:01,420 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=167235.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:09:13,766 INFO [finetune.py:976] (3/7) Epoch 30, batch 1150, loss[loss=0.1064, simple_loss=0.1725, pruned_loss=0.02011, over 4271.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2455, pruned_loss=0.04948, over 951481.55 frames. ], batch size: 18, lr: 2.81e-03, grad_scale: 64.0 2023-03-27 12:09:28,392 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.072e+02 1.511e+02 1.772e+02 2.131e+02 4.281e+02, threshold=3.544e+02, percent-clipped=3.0 2023-03-27 12:09:34,971 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167285.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:09:47,290 INFO [finetune.py:976] (3/7) Epoch 30, batch 1200, loss[loss=0.1725, simple_loss=0.229, pruned_loss=0.05801, over 4315.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2453, pruned_loss=0.04961, over 951903.26 frames. ], batch size: 65, lr: 2.81e-03, grad_scale: 64.0 2023-03-27 12:10:20,450 INFO [finetune.py:976] (3/7) Epoch 30, batch 1250, loss[loss=0.1489, simple_loss=0.2221, pruned_loss=0.03783, over 4816.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2422, pruned_loss=0.04817, over 953489.22 frames. ], batch size: 30, lr: 2.81e-03, grad_scale: 64.0 2023-03-27 12:10:36,983 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.068e+02 1.462e+02 1.711e+02 2.091e+02 4.240e+02, threshold=3.422e+02, percent-clipped=1.0 2023-03-27 12:10:47,818 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.95 vs. limit=5.0 2023-03-27 12:10:56,488 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167391.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:11:06,189 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7721, 1.2773, 0.8684, 1.6459, 2.1789, 1.5466, 1.5837, 1.7455], device='cuda:3'), covar=tensor([0.1355, 0.2047, 0.1913, 0.1172, 0.1789, 0.1796, 0.1386, 0.1674], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0094, 0.0109, 0.0093, 0.0120, 0.0092, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 12:11:09,173 INFO [finetune.py:976] (3/7) Epoch 30, batch 1300, loss[loss=0.1488, simple_loss=0.2213, pruned_loss=0.03817, over 4788.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.2387, pruned_loss=0.04686, over 955273.02 frames. ], batch size: 25, lr: 2.81e-03, grad_scale: 64.0 2023-03-27 12:11:24,565 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167420.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:11:32,407 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167432.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:11:38,996 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=167439.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:11:47,737 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.6874, 1.5304, 1.4310, 1.5574, 1.8905, 1.8235, 1.5475, 1.4211], device='cuda:3'), covar=tensor([0.0364, 0.0334, 0.0654, 0.0354, 0.0244, 0.0502, 0.0369, 0.0423], device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0107, 0.0149, 0.0112, 0.0103, 0.0118, 0.0104, 0.0116], device='cuda:3'), out_proj_covar=tensor([7.9990e-05, 8.1833e-05, 1.1603e-04, 8.4729e-05, 7.9501e-05, 8.7181e-05, 7.7443e-05, 8.7955e-05], device='cuda:3') 2023-03-27 12:11:55,876 INFO [finetune.py:976] (3/7) Epoch 30, batch 1350, loss[loss=0.1412, simple_loss=0.2124, pruned_loss=0.03504, over 4796.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2374, pruned_loss=0.04624, over 955357.02 frames. ], batch size: 25, lr: 2.81e-03, grad_scale: 64.0 2023-03-27 12:11:58,249 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167456.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:12:00,698 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167460.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:12:07,068 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=167468.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:12:11,228 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.213e+01 1.429e+02 1.665e+02 1.960e+02 3.889e+02, threshold=3.329e+02, percent-clipped=2.0 2023-03-27 12:12:23,215 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167493.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:12:29,688 INFO [finetune.py:976] (3/7) Epoch 30, batch 1400, loss[loss=0.1328, simple_loss=0.218, pruned_loss=0.02381, over 4757.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2411, pruned_loss=0.04741, over 955566.29 frames. ], batch size: 26, lr: 2.81e-03, grad_scale: 64.0 2023-03-27 12:13:02,946 INFO [finetune.py:976] (3/7) Epoch 30, batch 1450, loss[loss=0.2127, simple_loss=0.2824, pruned_loss=0.0715, over 4834.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2439, pruned_loss=0.04841, over 957595.81 frames. ], batch size: 49, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 12:13:19,228 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.032e+02 1.535e+02 1.844e+02 2.174e+02 4.375e+02, threshold=3.689e+02, percent-clipped=4.0 2023-03-27 12:13:25,327 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167585.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:13:36,788 INFO [finetune.py:976] (3/7) Epoch 30, batch 1500, loss[loss=0.1653, simple_loss=0.2433, pruned_loss=0.04369, over 4853.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2446, pruned_loss=0.04846, over 955452.15 frames. ], batch size: 31, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 12:13:52,817 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-03-27 12:13:57,376 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=167633.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:14:10,430 INFO [finetune.py:976] (3/7) Epoch 30, batch 1550, loss[loss=0.1905, simple_loss=0.259, pruned_loss=0.06099, over 4828.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2447, pruned_loss=0.04832, over 955253.56 frames. ], batch size: 49, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 12:14:26,674 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.931e+01 1.416e+02 1.751e+02 2.105e+02 4.024e+02, threshold=3.503e+02, percent-clipped=1.0 2023-03-27 12:14:44,011 INFO [finetune.py:976] (3/7) Epoch 30, batch 1600, loss[loss=0.1642, simple_loss=0.2335, pruned_loss=0.04749, over 4893.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2427, pruned_loss=0.04803, over 956640.29 frames. ], batch size: 32, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 12:14:54,734 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167719.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:15:17,593 INFO [finetune.py:976] (3/7) Epoch 30, batch 1650, loss[loss=0.1443, simple_loss=0.2155, pruned_loss=0.03655, over 4852.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2384, pruned_loss=0.04612, over 955422.71 frames. ], batch size: 44, lr: 2.81e-03, grad_scale: 32.0 2023-03-27 12:15:19,516 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167756.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:15:21,897 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167760.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:15:26,694 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167767.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:15:32,525 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.722e+01 1.428e+02 1.633e+02 1.926e+02 4.440e+02, threshold=3.266e+02, percent-clipped=1.0 2023-03-27 12:15:36,064 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167780.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:15:39,494 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.3651, 2.4428, 2.3204, 1.7218, 2.3315, 2.6408, 2.6041, 2.1624], device='cuda:3'), covar=tensor([0.0660, 0.0675, 0.0799, 0.0885, 0.0861, 0.0673, 0.0636, 0.1152], device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0138, 0.0141, 0.0119, 0.0129, 0.0139, 0.0139, 0.0163], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 12:15:41,290 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167788.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:15:55,581 INFO [finetune.py:976] (3/7) Epoch 30, batch 1700, loss[loss=0.1482, simple_loss=0.231, pruned_loss=0.03272, over 4773.00 frames. ], tot_loss[loss=0.163, simple_loss=0.2357, pruned_loss=0.04519, over 955017.26 frames. ], batch size: 28, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:15:56,726 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=167804.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:16:03,659 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=167808.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:16:25,159 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167828.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:16:42,118 INFO [finetune.py:976] (3/7) Epoch 30, batch 1750, loss[loss=0.1393, simple_loss=0.2063, pruned_loss=0.03612, over 4703.00 frames. ], tot_loss[loss=0.1628, simple_loss=0.2356, pruned_loss=0.04503, over 955219.74 frames. ], batch size: 23, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:16:47,667 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7619, 2.8218, 2.7456, 1.9618, 2.7209, 2.9963, 3.0162, 2.4406], device='cuda:3'), covar=tensor([0.0591, 0.0561, 0.0687, 0.0790, 0.0636, 0.0668, 0.0596, 0.1045], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0137, 0.0141, 0.0118, 0.0128, 0.0138, 0.0139, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 12:16:54,723 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7553, 1.3992, 2.1443, 3.2867, 2.1659, 2.4015, 1.2026, 2.7858], device='cuda:3'), covar=tensor([0.1651, 0.1525, 0.1284, 0.0591, 0.0821, 0.1961, 0.1639, 0.0476], device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0116, 0.0133, 0.0164, 0.0100, 0.0136, 0.0125, 0.0102], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:3') 2023-03-27 12:16:56,579 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.3862, 1.3853, 1.1941, 1.3661, 1.6832, 1.5848, 1.4098, 1.2353], device='cuda:3'), covar=tensor([0.0353, 0.0335, 0.0676, 0.0314, 0.0216, 0.0546, 0.0327, 0.0445], device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0107, 0.0150, 0.0112, 0.0103, 0.0119, 0.0105, 0.0116], device='cuda:3'), out_proj_covar=tensor([7.9997e-05, 8.2062e-05, 1.1646e-04, 8.5244e-05, 7.9962e-05, 8.7409e-05, 7.8003e-05, 8.8186e-05], device='cuda:3') 2023-03-27 12:16:57,034 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.117e+02 1.486e+02 1.829e+02 2.140e+02 4.770e+02, threshold=3.658e+02, percent-clipped=2.0 2023-03-27 12:17:20,240 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9649, 1.3950, 0.8001, 1.8176, 2.2390, 1.6703, 1.8954, 1.7845], device='cuda:3'), covar=tensor([0.1319, 0.2021, 0.2008, 0.1199, 0.1857, 0.1866, 0.1308, 0.1965], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0094, 0.0109, 0.0094, 0.0120, 0.0093, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 12:17:25,521 INFO [finetune.py:976] (3/7) Epoch 30, batch 1800, loss[loss=0.143, simple_loss=0.2227, pruned_loss=0.03162, over 4890.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2391, pruned_loss=0.04599, over 956274.27 frames. ], batch size: 32, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:17:58,717 INFO [finetune.py:976] (3/7) Epoch 30, batch 1850, loss[loss=0.2129, simple_loss=0.2846, pruned_loss=0.07062, over 4801.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2405, pruned_loss=0.04688, over 954725.60 frames. ], batch size: 41, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:18:13,649 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.002e+02 1.458e+02 1.787e+02 2.144e+02 3.700e+02, threshold=3.573e+02, percent-clipped=1.0 2023-03-27 12:18:33,563 INFO [finetune.py:976] (3/7) Epoch 30, batch 1900, loss[loss=0.1807, simple_loss=0.2449, pruned_loss=0.05826, over 4791.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.243, pruned_loss=0.04801, over 956531.24 frames. ], batch size: 25, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:18:45,500 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.7902, 4.4466, 4.2269, 2.2188, 4.5022, 3.4237, 0.8134, 3.1653], device='cuda:3'), covar=tensor([0.2577, 0.1713, 0.1400, 0.2946, 0.0781, 0.0814, 0.4431, 0.1218], device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0181, 0.0161, 0.0131, 0.0164, 0.0125, 0.0149, 0.0127], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:3') 2023-03-27 12:18:55,002 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168035.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:18:55,593 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.5880, 1.0973, 0.8967, 1.5055, 1.9651, 1.2335, 1.4569, 1.5302], device='cuda:3'), covar=tensor([0.1474, 0.2033, 0.1841, 0.1186, 0.2049, 0.1986, 0.1384, 0.1821], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0095, 0.0110, 0.0094, 0.0121, 0.0093, 0.0099, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 12:18:56,839 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.4582, 1.0204, 0.7745, 1.3566, 1.8902, 0.8677, 1.2952, 1.3652], device='cuda:3'), covar=tensor([0.1514, 0.2161, 0.1721, 0.1280, 0.2044, 0.1874, 0.1488, 0.1963], device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0095, 0.0110, 0.0094, 0.0121, 0.0093, 0.0099, 0.0090], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 12:19:07,330 INFO [finetune.py:976] (3/7) Epoch 30, batch 1950, loss[loss=0.1608, simple_loss=0.2349, pruned_loss=0.04338, over 4901.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2409, pruned_loss=0.04668, over 955421.85 frames. ], batch size: 43, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:19:21,549 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168075.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:19:22,071 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.065e+02 1.458e+02 1.758e+02 2.118e+02 3.555e+02, threshold=3.516e+02, percent-clipped=0.0 2023-03-27 12:19:30,461 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168088.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:19:35,809 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168096.0, num_to_drop=1, layers_to_drop={3} 2023-03-27 12:19:40,900 INFO [finetune.py:976] (3/7) Epoch 30, batch 2000, loss[loss=0.1081, simple_loss=0.1719, pruned_loss=0.02214, over 4775.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2385, pruned_loss=0.04613, over 956270.84 frames. ], batch size: 23, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:19:54,026 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168123.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:20:01,897 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=168136.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:20:14,572 INFO [finetune.py:976] (3/7) Epoch 30, batch 2050, loss[loss=0.1427, simple_loss=0.2207, pruned_loss=0.0323, over 4761.00 frames. ], tot_loss[loss=0.1635, simple_loss=0.2358, pruned_loss=0.04564, over 956811.65 frames. ], batch size: 27, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:20:24,870 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0187, 1.7009, 2.2724, 1.5586, 2.0268, 2.2230, 1.5514, 2.3811], device='cuda:3'), covar=tensor([0.1213, 0.2118, 0.1564, 0.1974, 0.0955, 0.1354, 0.3214, 0.0852], device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0207, 0.0193, 0.0189, 0.0174, 0.0212, 0.0219, 0.0198], 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') 2023-03-27 12:20:28,417 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9159, 1.7842, 1.5300, 1.9181, 2.2295, 1.9419, 1.4685, 1.5212], device='cuda:3'), covar=tensor([0.1832, 0.1735, 0.1769, 0.1514, 0.1506, 0.1103, 0.2378, 0.1784], device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0215, 0.0219, 0.0202, 0.0249, 0.0195, 0.0222, 0.0209], 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') 2023-03-27 12:20:29,509 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.016e+02 1.484e+02 1.729e+02 2.115e+02 4.273e+02, threshold=3.459e+02, percent-clipped=3.0 2023-03-27 12:20:29,641 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.9149, 1.7491, 1.8387, 1.2858, 1.8660, 1.9165, 1.9566, 1.5427], device='cuda:3'), covar=tensor([0.0539, 0.0674, 0.0638, 0.0788, 0.0835, 0.0634, 0.0506, 0.1176], device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0137, 0.0141, 0.0119, 0.0128, 0.0138, 0.0139, 0.0162], device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:3') 2023-03-27 12:20:47,508 INFO [finetune.py:976] (3/7) Epoch 30, batch 2100, loss[loss=0.1836, simple_loss=0.2486, pruned_loss=0.05928, over 4828.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2364, pruned_loss=0.04668, over 956519.33 frames. ], batch size: 30, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:21:19,162 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168233.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:21:20,798 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 2023-03-27 12:21:41,933 INFO [finetune.py:976] (3/7) Epoch 30, batch 2150, loss[loss=0.1656, simple_loss=0.2549, pruned_loss=0.03813, over 4790.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2392, pruned_loss=0.04722, over 956322.76 frames. ], batch size: 29, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:21:58,233 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.8279, 2.5744, 2.4514, 1.3679, 2.6031, 1.9919, 1.9471, 2.4398], device='cuda:3'), covar=tensor([0.1274, 0.0872, 0.1929, 0.2237, 0.1660, 0.2548, 0.2472, 0.1240], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0189, 0.0202, 0.0181, 0.0209, 0.0210, 0.0224, 0.0196], 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') 2023-03-27 12:22:00,992 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.149e+02 1.551e+02 1.892e+02 2.224e+02 4.404e+02, threshold=3.784e+02, percent-clipped=3.0 2023-03-27 12:22:12,543 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168294.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:22:18,785 INFO [finetune.py:976] (3/7) Epoch 30, batch 2200, loss[loss=0.1772, simple_loss=0.259, pruned_loss=0.04767, over 4906.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2415, pruned_loss=0.04784, over 954355.52 frames. ], batch size: 46, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:22:39,300 INFO [scaling.py:679] (3/7) Whitening: num_groups=1, num_channels=384, metric=3.85 vs. limit=5.0 2023-03-27 12:22:47,301 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.0927, 1.8549, 1.4785, 0.6766, 1.6582, 1.8384, 1.6769, 1.7726], device='cuda:3'), covar=tensor([0.0713, 0.0692, 0.1198, 0.1559, 0.1072, 0.1690, 0.1812, 0.0672], device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0187, 0.0201, 0.0180, 0.0208, 0.0209, 0.0223, 0.0195], 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') 2023-03-27 12:23:02,560 INFO [finetune.py:976] (3/7) Epoch 30, batch 2250, loss[loss=0.1469, simple_loss=0.2295, pruned_loss=0.03211, over 4889.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2431, pruned_loss=0.04847, over 953263.68 frames. ], batch size: 43, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:23:17,396 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168375.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:23:17,907 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.029e+01 1.513e+02 1.826e+02 2.132e+02 3.584e+02, threshold=3.652e+02, percent-clipped=0.0 2023-03-27 12:23:28,072 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168391.0, num_to_drop=1, layers_to_drop={1} 2023-03-27 12:23:34,124 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7696, 1.3195, 1.9239, 1.8205, 1.6343, 1.5789, 1.7920, 1.8466], device='cuda:3'), covar=tensor([0.4194, 0.3772, 0.2856, 0.3521, 0.4332, 0.3694, 0.3858, 0.2694], device='cuda:3'), in_proj_covar=tensor([0.0271, 0.0251, 0.0270, 0.0301, 0.0302, 0.0280, 0.0307, 0.0256], 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') 2023-03-27 12:23:36,282 INFO [finetune.py:976] (3/7) Epoch 30, batch 2300, loss[loss=0.1638, simple_loss=0.2416, pruned_loss=0.043, over 4821.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2432, pruned_loss=0.04842, over 951891.64 frames. ], batch size: 30, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:23:40,052 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-03-27 12:23:49,962 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=168423.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:23:50,009 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168423.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:24:09,556 INFO [finetune.py:976] (3/7) Epoch 30, batch 2350, loss[loss=0.1262, simple_loss=0.2103, pruned_loss=0.02108, over 4835.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2417, pruned_loss=0.04802, over 951307.84 frames. ], batch size: 30, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:24:21,868 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=168471.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:24:24,771 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.258e+01 1.458e+02 1.699e+02 2.116e+02 4.301e+02, threshold=3.398e+02, percent-clipped=1.0 2023-03-27 12:24:42,043 INFO [finetune.py:976] (3/7) Epoch 30, batch 2400, loss[loss=0.1755, simple_loss=0.248, pruned_loss=0.05147, over 4927.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2382, pruned_loss=0.0469, over 953200.64 frames. ], batch size: 33, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:25:15,072 INFO [finetune.py:976] (3/7) Epoch 30, batch 2450, loss[loss=0.179, simple_loss=0.2434, pruned_loss=0.05731, over 4832.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2356, pruned_loss=0.04616, over 956288.90 frames. ], batch size: 30, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:25:30,927 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.020e+02 1.487e+02 1.838e+02 2.245e+02 3.083e+02, threshold=3.676e+02, percent-clipped=0.0 2023-03-27 12:25:39,369 INFO [zipformer.py:1188] (3/7) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168589.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:25:48,869 INFO [finetune.py:976] (3/7) Epoch 30, batch 2500, loss[loss=0.1718, simple_loss=0.2441, pruned_loss=0.04977, over 4764.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2376, pruned_loss=0.04666, over 954652.62 frames. ], batch size: 27, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:26:27,843 INFO [finetune.py:976] (3/7) Epoch 30, batch 2550, loss[loss=0.1727, simple_loss=0.2459, pruned_loss=0.04982, over 4861.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2409, pruned_loss=0.04711, over 953330.30 frames. ], batch size: 31, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:26:55,336 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.157e+02 1.538e+02 1.821e+02 2.163e+02 4.307e+02, threshold=3.641e+02, percent-clipped=2.0 2023-03-27 12:27:09,671 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168691.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:27:17,864 INFO [finetune.py:976] (3/7) Epoch 30, batch 2600, loss[loss=0.1961, simple_loss=0.2833, pruned_loss=0.05444, over 4906.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2426, pruned_loss=0.04745, over 954175.94 frames. ], batch size: 36, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:27:26,310 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([4.7272, 4.1622, 4.2987, 4.4940, 4.4734, 4.2334, 4.8408, 1.5785], device='cuda:3'), covar=tensor([0.0906, 0.0857, 0.0771, 0.1244, 0.1387, 0.1655, 0.0662, 0.6325], device='cuda:3'), in_proj_covar=tensor([0.0353, 0.0247, 0.0287, 0.0298, 0.0337, 0.0287, 0.0307, 0.0303], 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') 2023-03-27 12:27:44,434 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=168739.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:28:01,792 INFO [finetune.py:976] (3/7) Epoch 30, batch 2650, loss[loss=0.1722, simple_loss=0.2402, pruned_loss=0.05211, over 4878.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2427, pruned_loss=0.04736, over 955915.83 frames. ], batch size: 32, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:28:21,696 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 1.069e+02 1.510e+02 1.724e+02 1.979e+02 3.263e+02, threshold=3.448e+02, percent-clipped=0.0 2023-03-27 12:28:43,727 INFO [finetune.py:976] (3/7) Epoch 30, batch 2700, loss[loss=0.1554, simple_loss=0.2322, pruned_loss=0.03934, over 4756.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2414, pruned_loss=0.04644, over 955463.83 frames. ], batch size: 27, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:29:01,583 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([1.7647, 1.1753, 0.9276, 1.6588, 2.1057, 1.5367, 1.5025, 1.4797], device='cuda:3'), covar=tensor([0.1545, 0.2107, 0.2008, 0.1179, 0.1942, 0.2100, 0.1461, 0.1966], device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0094, 0.0109, 0.0093, 0.0120, 0.0093, 0.0098, 0.0089], device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0004, 0.0003, 0.0004, 0.0003], device='cuda:3') 2023-03-27 12:29:10,130 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 2023-03-27 12:29:17,014 INFO [finetune.py:976] (3/7) Epoch 30, batch 2750, loss[loss=0.1443, simple_loss=0.2297, pruned_loss=0.02944, over 4927.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2385, pruned_loss=0.04606, over 956254.52 frames. ], batch size: 33, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:29:27,235 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-03-27 12:29:28,388 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([2.1197, 1.9964, 1.7823, 1.7979, 1.8977, 1.8751, 1.9007, 2.6118], device='cuda:3'), covar=tensor([0.3281, 0.3674, 0.3050, 0.3514, 0.3740, 0.2399, 0.3392, 0.1589], device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0264, 0.0239, 0.0276, 0.0263, 0.0232, 0.0260, 0.0240], 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') 2023-03-27 12:29:32,261 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.461e+01 1.436e+02 1.670e+02 1.989e+02 2.987e+02, threshold=3.340e+02, percent-clipped=0.0 2023-03-27 12:29:40,005 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-03-27 12:29:41,552 INFO [zipformer.py:1188] (3/7) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168889.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:29:50,505 INFO [finetune.py:976] (3/7) Epoch 30, batch 2800, loss[loss=0.1239, simple_loss=0.1967, pruned_loss=0.02552, over 4798.00 frames. ], tot_loss[loss=0.1635, simple_loss=0.2359, pruned_loss=0.04552, over 957728.01 frames. ], batch size: 26, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:30:13,012 INFO [zipformer.py:1188] (3/7) warmup_begin=666.7, warmup_end=1333.3, batch_count=168937.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:30:23,986 INFO [finetune.py:976] (3/7) Epoch 30, batch 2850, loss[loss=0.1764, simple_loss=0.2361, pruned_loss=0.05837, over 4825.00 frames. ], tot_loss[loss=0.1632, simple_loss=0.235, pruned_loss=0.04576, over 955295.26 frames. ], batch size: 30, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:30:33,493 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168967.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:30:34,686 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168969.0, num_to_drop=1, layers_to_drop={0} 2023-03-27 12:30:38,878 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 8.606e+01 1.456e+02 1.728e+02 2.104e+02 5.266e+02, threshold=3.457e+02, percent-clipped=2.0 2023-03-27 12:30:57,836 INFO [finetune.py:976] (3/7) Epoch 30, batch 2900, loss[loss=0.1858, simple_loss=0.2592, pruned_loss=0.05618, over 4921.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2372, pruned_loss=0.04608, over 956454.99 frames. ], batch size: 36, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:31:14,729 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169028.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:31:15,927 INFO [zipformer.py:1188] (3/7) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169030.0, num_to_drop=1, layers_to_drop={2} 2023-03-27 12:31:18,977 INFO [zipformer.py:2441] (3/7) attn_weights_entropy = tensor([3.4928, 2.9165, 2.7877, 1.4618, 2.9236, 2.3389, 2.3078, 2.6895], device='cuda:3'), covar=tensor([0.1014, 0.0893, 0.1758, 0.2196, 0.1383, 0.2118, 0.2071, 0.1217], device='cuda:3'), in_proj_covar=tensor([0.0171, 0.0188, 0.0201, 0.0180, 0.0209, 0.0210, 0.0224, 0.0196], 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') 2023-03-27 12:31:31,772 INFO [finetune.py:976] (3/7) Epoch 30, batch 2950, loss[loss=0.1791, simple_loss=0.2515, pruned_loss=0.05337, over 4893.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2398, pruned_loss=0.04638, over 955543.02 frames. ], batch size: 32, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:31:47,441 INFO [scaling.py:679] (3/7) Whitening: num_groups=8, num_channels=192, metric=1.59 vs. limit=2.0 2023-03-27 12:31:49,074 INFO [optim.py:369] (3/7) Clipping_scale=2.0, grad-norm quartiles 9.406e+01 1.615e+02 1.887e+02 2.255e+02 4.054e+02, threshold=3.773e+02, percent-clipped=1.0 2023-03-27 12:31:53,285 INFO [zipformer.py:1188] (3/7) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169082.0, num_to_drop=0, layers_to_drop=set() 2023-03-27 12:32:19,262 INFO [finetune.py:976] (3/7) Epoch 30, batch 3000, loss[loss=0.1626, simple_loss=0.2344, pruned_loss=0.04545, over 4768.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2415, pruned_loss=0.04724, over 954299.46 frames. ], batch size: 26, lr: 2.80e-03, grad_scale: 32.0 2023-03-27 12:32:19,262 INFO [finetune.py:1001] (3/7) Computing validation loss