Wav2Vec2-XLS-TR
This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the Common Voice 17 dataset. It achieves the following results on the evaluation set:
- Loss: 2.7124
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: 1e-05
- train_batch_size: 32
- eval_batch_size: 32
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 30
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 23.6246 | 0.3446 | 500 | 16.2983 |
| 8.2656 | 0.6892 | 1000 | 5.9544 |
| 5.1083 | 1.0338 | 1500 | 4.6067 |
| 4.1663 | 1.3784 | 2000 | 3.8176 |
| 3.559 | 1.7229 | 2500 | 3.4058 |
| 3.2582 | 2.0675 | 3000 | 3.2021 |
| 2.9475 | 2.4121 | 3500 | 3.0052 |
| 1.9117 | 2.7567 | 4000 | 2.7124 |
| 1.2089 | 3.1013 | 4500 | 2.6678 |
| 0.9505 | 3.4459 | 5000 | 3.1079 |
| 0.7688 | 3.7905 | 5500 | 2.6448 |
| 0.6474 | 4.1351 | 6000 | 2.1442 |
| 0.5717 | 4.4797 | 6500 | 2.6866 |
| 0.5262 | 4.8243 | 7000 | 2.7227 |
| 0.4764 | 5.1688 | 7500 | 2.5182 |
| 0.4452 | 5.5134 | 8000 | 2.7089 |
| 0.4237 | 5.8580 | 8500 | 2.4970 |
| 0.4004 | 6.2026 | 9000 | 2.0918 |
| 0.3845 | 6.5472 | 9500 | 2.1695 |
| 0.3685 | 6.8918 | 10000 | 2.3281 |
| 0.3582 | 7.2364 | 10500 | 2.0681 |
| 0.337 | 7.5810 | 11000 | 2.2537 |
| 0.3301 | 7.9256 | 11500 | 2.0853 |
| 0.3142 | 8.2702 | 12000 | 2.2379 |
| 0.3087 | 8.6147 | 12500 | 2.0322 |
| 0.3026 | 8.9593 | 13000 | 2.3212 |
| 0.2894 | 9.3039 | 13500 | 2.3748 |
| 0.2828 | 9.6485 | 14000 | 2.0785 |
| 0.2798 | 9.9931 | 14500 | 1.9353 |
| 0.262 | 10.3377 | 15000 | 2.2790 |
| 0.2592 | 10.6823 | 15500 | 2.4833 |
| 0.2613 | 11.0269 | 16000 | 2.4773 |
| 0.2484 | 11.3715 | 16500 | 2.0992 |
| 0.246 | 11.7161 | 17000 | 2.3585 |
| 0.2423 | 12.0606 | 17500 | 2.3101 |
| 0.2341 | 12.4052 | 18000 | 2.2959 |
| 0.231 | 12.7498 | 18500 | 1.9466 |
| 0.2283 | 13.0944 | 19000 | 2.0529 |
| 0.224 | 13.4390 | 19500 | 2.2572 |
| 0.226 | 13.7836 | 20000 | 2.2684 |
| 0.2157 | 14.1282 | 20500 | 2.0767 |
| 0.2128 | 14.4728 | 21000 | 2.4035 |
| 0.2117 | 14.8174 | 21500 | 2.1512 |
| 0.2062 | 15.1620 | 22000 | 2.2322 |
| 0.2002 | 15.5065 | 22500 | 2.1550 |
| 0.2033 | 15.8511 | 23000 | 2.5139 |
| 0.1997 | 16.1957 | 23500 | 2.1999 |
| 0.1942 | 16.5403 | 24000 | 2.5342 |
| 0.1942 | 16.8849 | 24500 | 2.5350 |
| 0.1923 | 17.2295 | 25000 | 2.6199 |
| 0.1869 | 17.5741 | 25500 | 2.5196 |
| 0.1875 | 17.9187 | 26000 | 2.3316 |
| 0.1825 | 18.2633 | 26500 | 2.4622 |
| 0.1828 | 18.6079 | 27000 | 2.3920 |
| 0.1808 | 18.9524 | 27500 | 2.4571 |
| 0.1797 | 19.2970 | 28000 | 2.5495 |
| 0.1777 | 19.6416 | 28500 | 2.1355 |
| 0.1752 | 19.9862 | 29000 | 2.4399 |
| 0.1723 | 20.3308 | 29500 | 2.4222 |
| 0.1719 | 20.6754 | 30000 | 2.7283 |
| 0.1702 | 21.0200 | 30500 | 2.4584 |
| 0.1655 | 21.3646 | 31000 | 2.4646 |
| 0.1658 | 21.7092 | 31500 | 2.3659 |
| 0.1676 | 22.0538 | 32000 | 2.5923 |
| 0.1606 | 22.3983 | 32500 | 2.6588 |
| 0.1572 | 22.7429 | 33000 | 2.3206 |
| 0.1611 | 23.0875 | 33500 | 2.4420 |
| 0.156 | 23.4321 | 34000 | 2.6875 |
| 0.1587 | 23.7767 | 34500 | 2.6309 |
| 0.1596 | 24.1213 | 35000 | 2.4329 |
| 0.1572 | 24.4659 | 35500 | 2.6451 |
| 0.1552 | 24.8105 | 36000 | 2.6626 |
| 0.1529 | 25.1551 | 36500 | 2.6210 |
| 0.155 | 25.4997 | 37000 | 2.6027 |
| 0.1531 | 25.8442 | 37500 | 2.7159 |
| 0.1564 | 26.1888 | 38000 | 2.6926 |
| 0.1497 | 26.5334 | 38500 | 2.8263 |
| 0.1443 | 26.8780 | 39000 | 2.6340 |
| 0.15 | 27.2226 | 39500 | 2.6501 |
| 0.1456 | 27.5672 | 40000 | 2.7202 |
| 0.1489 | 27.9118 | 40500 | 2.6929 |
| 0.15 | 28.2564 | 41000 | 2.7965 |
| 0.147 | 28.6010 | 41500 | 2.6591 |
| 0.1471 | 28.9456 | 42000 | 2.7030 |
| 0.1473 | 29.2901 | 42500 | 2.6906 |
| 0.1458 | 29.6347 | 43000 | 2.7225 |
| 0.1438 | 29.9793 | 43500 | 2.7124 |
Framework versions
- Transformers 4.41.2
- Pytorch 2.3.1+cu121
- Datasets 2.20.0
- Tokenizers 0.19.1