w2v2-base-pretrained_lr1e-4_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: 2.2354
• Wer: 0.1781

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
18.7165	25.0	250	3.6383	1.0
3.2049	50.0	500	3.1037	1.0
2.5797	75.0	750	1.3392	0.9987
0.2099	100.0	1000	1.5991	0.2947
0.0635	125.0	1250	1.7346	0.2119
0.0354	150.0	1500	2.0431	0.1939
0.0184	175.0	1750	1.9597	0.1935
0.0129	200.0	2000	2.2028	0.1991
0.0101	225.0	2250	1.9856	0.1820
0.0061	250.0	2500	1.9899	0.1897
0.0064	275.0	2750	2.1246	0.1811
0.0037	300.0	3000	2.2400	0.1803
0.0033	325.0	3250	2.1766	0.1786
0.0029	350.0	3500	2.2354	0.1781

Framework versions

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