wav2vec2-7

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: 0.6017
• Wer: 0.5200

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: 400
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.1311	1.56	200	2.9839	1.0
2.5727	3.12	400	1.4962	1.0209
1.0187	4.69	600	0.7562	0.7859
0.637	6.25	800	0.6529	0.6960
0.4847	7.81	1000	0.6609	0.6745
0.3952	9.38	1200	0.5808	0.6220
0.3343	10.94	1400	0.5622	0.6004
0.2897	12.5	1600	0.8842	0.5980
0.2549	14.06	1800	0.6047	0.5765
0.2334	15.62	2000	0.6436	0.5699
0.2144	17.19	2200	0.5831	0.5593
0.1982	18.75	2400	0.6327	0.5620
0.1817	20.31	2600	0.8790	0.5456
0.1713	21.88	2800	0.9603	0.5362
0.163	23.44	3000	0.5940	0.5384
0.1539	25.0	3200	0.6058	0.5311
0.1392	26.56	3400	0.6131	0.5221
0.1386	28.12	3600	0.6066	0.5258
0.1351	29.69	3800	0.6017	0.5200

Framework versions

• Transformers 4.19.2
• Pytorch 1.11.0+cu113
• Datasets 2.2.2
• Tokenizers 0.12.1