<!-- 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-6
This model is a fine-tuned version of MeshalAlamr/wav2vec2-xls-r-300m-ar-6 on the common_voice dataset. It achieves the following results on the evaluation set:
- Loss: 78.2951
- Wer: 0.2040
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 |
|---|---|---|---|---|
| No log | 1.0 | 85 | 75.3576 | 0.2131 |
| No log | 2.0 | 170 | 75.3215 | 0.2150 |
| No log | 3.0 | 255 | 75.5332 | 0.2201 |
| No log | 4.0 | 340 | 81.2835 | 0.2315 |
| 94.75 | 5.0 | 425 | 78.3768 | 0.2422 |
| 94.75 | 6.0 | 510 | 82.9389 | 0.2520 |
| 94.75 | 7.0 | 595 | 76.7272 | 0.2496 |
| 94.75 | 8.0 | 680 | 79.9325 | 0.2506 |
| 94.75 | 9.0 | 765 | 82.2568 | 0.2507 |
| 124.0193 | 10.0 | 850 | 78.7011 | 0.2415 |
| 124.0193 | 11.0 | 935 | 81.2829 | 0.2396 |
| 124.0193 | 12.0 | 1020 | 77.2370 | 0.2357 |
| 124.0193 | 13.0 | 1105 | 77.4057 | 0.2347 |
| 124.0193 | 14.0 | 1190 | 74.4764 | 0.2271 |
| 112.7824 | 15.0 | 1275 | 78.7320 | 0.2355 |
| 112.7824 | 16.0 | 1360 | 79.0120 | 0.2294 |
| 112.7824 | 17.0 | 1445 | 82.3663 | 0.2240 |
| 112.7824 | 18.0 | 1530 | 79.2765 | 0.2236 |
| 98.8702 | 19.0 | 1615 | 78.1527 | 0.2242 |
| 98.8702 | 20.0 | 1700 | 75.7842 | 0.2198 |
| 98.8702 | 21.0 | 1785 | 78.2980 | 0.2217 |
| 98.8702 | 22.0 | 1870 | 79.3180 | 0.2168 |
| 98.8702 | 23.0 | 1955 | 77.7381 | 0.2155 |
| 84.537 | 24.0 | 2040 | 78.1512 | 0.2131 |
| 84.537 | 25.0 | 2125 | 80.4068 | 0.2116 |
| 84.537 | 26.0 | 2210 | 75.5718 | 0.2075 |
| 84.537 | 27.0 | 2295 | 78.4438 | 0.2078 |
| 84.537 | 28.0 | 2380 | 79.6891 | 0.2086 |
| 74.4149 | 29.0 | 2465 | 77.9115 | 0.2069 |
| 74.4149 | 30.0 | 2550 | 78.2951 | 0.2040 |
Framework versions
- Transformers 4.17.0
- Pytorch 1.11.0
- Datasets 1.18.4
- Tokenizers 0.11.6
