w2v2-base-pretrained_lr5e-5_at1_da1-p4

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: 0.6341
• Wer: 0.1039

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: 5e-05
• 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
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
18.6256	5.21	250	4.2622	1.0
3.3901	10.42	500	3.2209	1.0
3.0963	15.62	750	3.1175	1.0
2.0992	20.83	1000	0.5962	0.4402
0.2069	26.04	1250	0.4456	0.1310
0.0849	31.25	1500	0.4902	0.1200
0.0596	36.46	1750	0.5079	0.1176
0.0437	41.67	2000	0.5362	0.1136
0.0355	46.88	2250	0.5433	0.1156
0.0281	52.08	2500	0.5994	0.1136
0.0238	57.29	2750	0.6018	0.1112
0.02	62.5	3000	0.5970	0.1120
0.0181	67.71	3250	0.6282	0.1083
0.0167	72.92	3500	0.6120	0.1075
0.0145	78.12	3750	0.6404	0.1047
0.014	83.33	4000	0.6341	0.1039

Framework versions

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