<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->
wav2vec2-xls-r-300m-ar-11
This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice dataset. It achieves the following results on the evaluation set:
- Loss: 60.5659
- Wer: 0.2144
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: 64
- eval_batch_size: 16
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 256
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 30
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 16849.2529 | 1.0 | 85 | 1458.4645 | 1.0 |
| 4474.9085 | 2.0 | 170 | 687.2793 | 1.0 |
| 2937.0309 | 3.0 | 255 | 632.0456 | 1.0 |
| 2853.7682 | 4.0 | 340 | 621.7872 | 1.0 |
| 2786.243 | 5.0 | 425 | 611.4717 | 1.0 |
| 2738.1844 | 6.0 | 510 | 578.0577 | 1.0 |
| 2118.4608 | 7.0 | 595 | 253.0534 | 0.9927 |
| 1026.4239 | 8.0 | 680 | 140.3523 | 0.6430 |
| 682.4369 | 9.0 | 765 | 106.5226 | 0.4990 |
| 516.4381 | 10.0 | 850 | 85.3184 | 0.4126 |
| 434.9369 | 11.0 | 935 | 79.4750 | 0.3683 |
| 369.3786 | 12.0 | 1020 | 73.2318 | 0.3290 |
| 324.2687 | 13.0 | 1105 | 69.6444 | 0.3160 |
| 292.8527 | 14.0 | 1190 | 66.7714 | 0.2922 |
| 266.229 | 15.0 | 1275 | 68.2237 | 0.2839 |
| 242.3606 | 16.0 | 1360 | 66.0233 | 0.2745 |
| 227.9846 | 17.0 | 1445 | 66.8503 | 0.2668 |
| 210.1087 | 18.0 | 1530 | 63.1035 | 0.2539 |
| 201.326 | 19.0 | 1615 | 63.9665 | 0.2481 |
| 189.019 | 20.0 | 1700 | 60.9628 | 0.2418 |
| 181.3091 | 21.0 | 1785 | 62.5716 | 0.2387 |
| 168.631 | 22.0 | 1870 | 62.4718 | 0.2342 |
| 165.8396 | 23.0 | 1955 | 61.0784 | 0.2287 |
| 161.4992 | 24.0 | 2040 | 62.2299 | 0.2257 |
| 153.6809 | 25.0 | 2125 | 60.4889 | 0.2235 |
| 145.4282 | 26.0 | 2210 | 60.8189 | 0.2208 |
| 144.6855 | 27.0 | 2295 | 61.8122 | 0.2203 |
| 138.6269 | 28.0 | 2380 | 60.4600 | 0.2172 |
| 137.6246 | 29.0 | 2465 | 61.4417 | 0.2167 |
| 134.6211 | 30.0 | 2550 | 60.5659 | 0.2144 |
Framework versions
- Transformers 4.17.0
- Pytorch 1.11.0
- Datasets 1.18.4
- Tokenizers 0.11.6
