wav2vec2-ehf-test

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: 0.0648
• Wer: 0.1661

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
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.3431	0.7812	250	3.0217	1.0
2.9086	1.5625	500	2.9322	1.0
2.0836	2.3438	750	0.6583	0.5862
0.5632	3.125	1000	0.2877	0.3624
0.3427	3.9062	1250	0.1959	0.2823
0.2548	4.6875	1500	0.1463	0.2464
0.217	5.4688	1750	0.1467	0.2340
0.1769	6.25	2000	0.1217	0.2162
0.1564	7.0312	2250	0.1100	0.2090
0.1351	7.8125	2500	0.1062	0.2074
0.12	8.5938	2750	0.1055	0.2022
0.1161	9.375	3000	0.1039	0.2011
0.1085	10.1562	3250	0.0988	0.1912
0.097	10.9375	3500	0.0931	0.1879
0.0895	11.7188	3750	0.0873	0.1869
0.0846	12.5	4000	0.0807	0.1846
0.0815	13.2812	4250	0.0826	0.1836
0.0787	14.0625	4500	0.0780	0.1798
0.0714	14.8438	4750	0.0732	0.1774
0.0702	15.625	5000	0.0745	0.1778
0.0637	16.4062	5250	0.0741	0.1764
0.0608	17.1875	5500	0.0788	0.1758
0.0575	17.9688	5750	0.0726	0.1727
0.0529	18.75	6000	0.0727	0.1726
0.0539	19.5312	6250	0.0704	0.1709
0.0533	20.3125	6500	0.0683	0.1702
0.0483	21.0938	6750	0.0643	0.1667
0.0461	21.875	7000	0.0650	0.1696
0.0442	22.6562	7250	0.0697	0.1687
0.042	23.4375	7500	0.0696	0.1687
0.0389	24.2188	7750	0.0689	0.1682
0.0402	25.0	8000	0.0702	0.1683
0.0365	25.7812	8250	0.0709	0.1669
0.0315	26.5625	8500	0.0695	0.1672
0.0349	27.3438	8750	0.0667	0.1662
0.0292	28.125	9000	0.0666	0.1669
0.0311	28.9062	9250	0.0652	0.1666
0.0322	29.6875	9500	0.0648	0.1661

Framework versions

• Transformers 4.45.1
• Pytorch 2.1.2
• Datasets 3.0.1
• Tokenizers 0.20.0