wav2vec2-fleur-mms-batch6-epoch16-finetunning-2

This model is a fine-tuned version of Nachuwu/wav2vec2-fleur-mms-batch6-epoch16-finetunning on the audiofolder dataset. It achieves the following results on the evaluation set:

• Loss: 0.7540
• Wer: 0.5148

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: 0.001
• train_batch_size: 6
• 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: 100
• num_epochs: 14
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7213	0.42	100	0.8689	0.5824
0.8363	0.84	200	0.9158	0.6085
0.8347	1.27	300	0.9091	0.5362
0.874	1.69	400	0.8487	0.5307
0.7452	2.11	500	0.8507	0.5472
0.791	2.53	600	0.8075	0.5410
0.7681	2.95	700	0.8389	0.5575
0.7645	3.38	800	0.8122	0.5610
0.739	3.8	900	0.8625	0.5569
0.7989	4.22	1000	0.8166	0.5624
0.7618	4.64	1100	0.8646	0.5796
0.7689	5.06	1200	0.8462	0.5445
0.738	5.49	1300	0.8014	0.5431
0.6966	5.91	1400	0.7993	0.5217
0.7275	6.33	1500	0.8049	0.5252
0.7304	6.75	1600	0.7870	0.5231
0.7225	7.17	1700	0.8037	0.5472
0.7148	7.59	1800	0.7913	0.5389
0.7225	8.02	1900	0.7908	0.5183
0.7123	8.44	2000	0.7894	0.5245
0.692	8.86	2100	0.7693	0.5224
0.7855	9.28	2200	0.7614	0.5252
0.6421	9.7	2300	0.7870	0.5169
0.6831	10.13	2400	0.7820	0.5121
0.7293	10.55	2500	0.7640	0.5245
0.6625	10.97	2600	0.7912	0.5575
0.6612	11.39	2700	0.7578	0.5203
0.6873	11.81	2800	0.7579	0.5231
0.7113	12.24	2900	0.7595	0.5258
0.7393	12.66	3000	0.7532	0.5148
0.7297	13.08	3100	0.7532	0.5183
0.6303	13.5	3200	0.7514	0.5162
0.6444	13.92	3300	0.7540	0.5148

Framework versions

• Transformers 4.36.0.dev0
• Pytorch 2.1.0+cu118
• Datasets 2.15.0
• Tokenizers 0.15.0