w2v2-base-pretrained_lr1e-4_at0.8_da0.6

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: 2.5777
• Wer: 0.1820

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
15.4716	8.93	250	3.5805	1.0
3.156	17.86	500	3.1574	1.0
2.6095	26.79	750	1.2490	1.0
0.2716	35.71	1000	1.6097	0.2422
0.0901	44.64	1250	1.8537	0.2085
0.0526	53.57	1500	1.7249	0.2080
0.0328	62.5	1750	2.2934	0.1922
0.0237	71.43	2000	2.1689	0.1948
0.018	80.36	2250	2.4606	0.1918
0.0144	89.29	2500	2.6943	0.1837
0.0108	98.21	2750	2.5514	0.1807
0.0078	107.14	3000	2.5035	0.1858
0.0064	116.07	3250	2.6022	0.1845
0.0069	125.0	3500	2.5777	0.1820

Framework versions

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