Model_S_D_Wav2Vec2

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0464
• Wer: 0.2319
• Cer: 0.0598

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: 0.0003
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.5768	0.85	400	0.6152	0.5812	0.1905
0.3226	1.71	800	0.1026	0.3195	0.0722
0.1827	2.56	1200	0.0725	0.2048	0.0454
0.129	3.41	1600	0.0671	0.2393	0.0525
0.1075	4.26	2000	0.0556	0.2312	0.0497
0.0924	5.12	2400	0.0572	0.2040	0.0478
0.076	5.97	2800	0.0596	0.1472	0.0346
0.0695	6.82	3200	0.0608	0.2274	0.0510
0.0707	7.68	3600	0.0490	0.2665	0.0660
0.0597	8.53	4000	0.0509	0.2442	0.0593
0.0557	9.38	4400	0.0501	0.2533	0.0610
0.0503	10.23	4800	0.0519	0.2534	0.0622
0.0471	11.09	5200	0.0512	0.2585	0.0638
0.0417	11.94	5600	0.0497	0.2522	0.0610
0.0415	12.79	6000	0.0508	0.2547	0.0629
0.0372	13.65	6400	0.0497	0.2580	0.0643
0.0364	14.5	6800	0.0448	0.2498	0.0600
0.034	15.35	7200	0.0522	0.2419	0.0593
0.0306	16.2	7600	0.0510	0.2433	0.0560
0.0345	17.06	8000	0.0503	0.2610	0.0657
0.0266	17.91	8400	0.0462	0.2434	0.0620
0.0273	18.76	8800	0.0507	0.2456	0.0622
0.0216	19.62	9200	0.0466	0.2214	0.0531
0.0208	20.47	9600	0.0497	0.2396	0.0598
0.0201	21.32	10000	0.0470	0.2332	0.0559
0.0174	22.17	10400	0.0418	0.2346	0.0590
0.0198	23.03	10800	0.0472	0.2386	0.0602
0.0149	23.88	11200	0.0490	0.2446	0.0638
0.0133	24.73	11600	0.0497	0.2430	0.0632
0.0118	25.59	12000	0.0498	0.2368	0.0620
0.0106	26.44	12400	0.0453	0.2309	0.0590
0.0104	27.29	12800	0.0452	0.2296	0.0583
0.0085	28.14	13200	0.0467	0.2352	0.0604
0.0081	29.0	13600	0.0470	0.2310	0.0592
0.0079	29.85	14000	0.0464	0.2319	0.0598

Framework versions

• Transformers 4.31.0
• Pytorch 2.0.1+cu117
• Datasets 1.18.3
• Tokenizers 0.13.3