w2v2-base-pretrained_lr5e-5_at0.8_da0.2

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: 2.3279
• Wer: 0.2674

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
25.9775	25.0	250	4.0805	1.0
3.4724	50.0	500	3.1777	1.0
3.0775	75.0	750	3.0307	1.0
2.7225	100.0	1000	2.2425	1.0
0.8725	125.0	1250	1.5857	0.4584
0.1745	150.0	1500	1.8015	0.3503
0.0937	175.0	1750	1.8837	0.3131
0.0645	200.0	2000	2.0461	0.2947
0.0528	225.0	2250	2.1465	0.2853
0.0366	250.0	2500	2.1419	0.2875
0.0336	275.0	2750	2.1437	0.2755
0.026	300.0	3000	2.1733	0.2717
0.0247	325.0	3250	2.2546	0.2683
0.0207	350.0	3500	2.2603	0.2700
0.0185	375.0	3750	2.3492	0.2670
0.0186	400.0	4000	2.3279	0.2674

Framework versions

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