wav2vec2-base-finetune-vi-v2

This model is a fine-tuned version of nguyenvulebinh/wav2vec2-base-vietnamese-250h on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2188
• Wer: 0.1391

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: 8
• 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: 24

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.3873	0.67	500	2.4321	0.9719
1.4812	1.34	1000	0.5449	0.3062
0.7731	2.0	1500	0.3793	0.2263
0.542	2.67	2000	0.3021	0.2002
0.4461	3.34	2500	0.2905	0.1862
0.4175	4.01	3000	0.2687	0.1771
0.3878	4.67	3500	0.2958	0.1751
0.3373	5.34	4000	0.2713	0.1721
0.3046	6.01	4500	0.2505	0.1616
0.2933	6.68	5000	0.2561	0.1611
0.285	7.34	5500	0.2405	0.1617
0.2998	8.01	6000	0.2363	0.1578
0.2486	8.68	6500	0.2254	0.1570
0.2682	9.35	7000	0.2306	0.1547
0.2327	10.01	7500	0.2289	0.1537
0.2141	10.68	8000	0.2383	0.1499
0.2124	11.35	8500	0.2261	0.15
0.2156	12.02	9000	0.2142	0.1511
0.2082	12.68	9500	0.2386	0.1467
0.1814	13.35	10000	0.2301	0.1448
0.1836	14.02	10500	0.2302	0.1446
0.18	14.69	11000	0.2244	0.1445
0.1756	15.35	11500	0.2280	0.1439
0.1693	16.02	12000	0.2307	0.1426
0.1588	16.69	12500	0.2164	0.1422
0.1587	17.36	13000	0.2198	0.1417
0.1738	18.02	13500	0.2282	0.1411
0.1524	18.69	14000	0.2274	0.1394
0.1569	19.36	14500	0.2178	0.1396
0.1433	20.03	15000	0.2200	0.1413
0.1512	20.69	15500	0.2193	0.1382
0.1375	21.36	16000	0.2174	0.1393
0.1302	22.03	16500	0.2246	0.1391
0.146	22.7	17000	0.2222	0.1392
0.1265	23.36	17500	0.2188	0.1391

Framework versions

• Transformers 4.30.2
• Pytorch 2.0.0
• Datasets 2.1.0
• Tokenizers 0.13.3