w2v2-base-pretrained_lr1e-4_at0.2_da1

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

• Loss: 5.8341
• Wer: 1.0477

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: 0.0001
• train_batch_size: 32
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 60
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
19.6307	3.91	250	4.5235	1.0
3.3476	7.81	500	3.6749	1.0
2.8074	11.72	750	2.6368	1.0
0.5923	15.62	1000	2.6950	1.0098
0.2651	19.53	1250	3.7328	1.0413
0.1514	23.44	1500	4.3897	1.0520
0.0994	27.34	1750	4.4070	1.0294
0.0682	31.25	2000	4.9965	1.0362
0.047	35.16	2250	5.0946	1.0448
0.0375	39.06	2500	5.6089	1.0482
0.0307	42.97	2750	5.5917	1.0409
0.0254	46.88	3000	5.5547	1.0465
0.0195	50.78	3250	5.5936	1.0418
0.0161	54.69	3500	5.7545	1.0431
0.0142	58.59	3750	5.8341	1.0477

Framework versions

• Transformers 4.35.0
• Pytorch 2.0.0
• Datasets 2.14.6
• Tokenizers 0.14.1