xlsr-a-clp

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5263
• Wer: 0.4067

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0004
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 132
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.858	5.0	200	2.6999	1.0
2.2652	10.0	400	1.2382	1.0042
0.8587	15.0	600	0.5870	0.6960
0.3893	20.0	800	0.5675	0.6122
0.257	25.0	1000	0.4928	0.4822
0.1862	30.0	1200	0.4408	0.4885
0.148	35.0	1400	0.5587	0.4298
0.1203	40.0	1600	0.5393	0.4549
0.1094	45.0	1800	0.4889	0.4403
0.0908	50.0	2000	0.6539	0.4298
0.0766	55.0	2200	0.4469	0.4109
0.0675	60.0	2400	0.4109	0.4004
0.0622	65.0	2600	0.4543	0.3962
0.0617	70.0	2800	0.5189	0.3920
0.0529	75.0	3000	0.4261	0.3878
0.0514	80.0	3200	0.5161	0.3920
0.0473	85.0	3400	0.4789	0.3983
0.0419	90.0	3600	0.5161	0.4088
0.0405	95.0	3800	0.5263	0.4067

Framework versions

• Transformers 4.47.0.dev0
• Pytorch 2.4.0
• Datasets 3.0.1
• Tokenizers 0.20.0