w2v2-base-pretrained_lr1e-4_at0.8_da0.4

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: 2.1115
• Wer: 0.1952

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
• training_steps: 3500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
21.2157	13.16	250	4.1634	1.0
3.2337	26.32	500	3.1231	1.0
3.0575	39.47	750	3.0466	1.0
2.0739	52.63	1000	1.0677	0.6284
0.1758	65.79	1250	1.3711	0.3170
0.0675	78.95	1500	1.6521	0.2268
0.0355	92.11	1750	1.7313	0.2332
0.0209	105.26	2000	1.9720	0.2114
0.0162	118.42	2250	1.7569	0.2085
0.0099	131.58	2500	2.1623	0.1944
0.0071	144.74	2750	2.2067	0.1922
0.0066	157.89	3000	2.1246	0.1944
0.0059	171.05	3250	2.1484	0.1922
0.0045	184.21	3500	2.1115	0.1952

Framework versions

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