w2v2-base-pretrained_lr5e-5_at0.4_da1

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: 1.4372
• Wer: 0.1666

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
18.4042	4.03	250	4.2497	1.0
3.3741	8.06	500	3.2004	1.0
3.1004	12.1	750	3.1159	1.0
2.3298	16.13	1000	1.0486	0.7809
0.5044	20.16	1250	0.6083	0.3464
0.27	24.19	1500	0.6948	0.2456
0.1833	28.23	1750	0.9908	0.1956
0.1324	32.26	2000	1.0134	0.1995
0.1027	36.29	2250	1.3176	0.1760
0.0852	40.32	2500	1.1929	0.1837
0.0703	44.35	2750	1.3824	0.1670
0.0601	48.39	3000	1.3337	0.1674
0.0546	52.42	3250	1.3566	0.1717
0.05	56.45	3500	1.4653	0.1670
0.046	60.48	3750	1.4321	0.1696
0.0452	64.52	4000	1.4372	0.1666

Framework versions

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