whisper-tiny-mongolian-ver_0.1
This model is a fine-tuned version of openai/whisper-tiny on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 2.3273
- Wer: 1.4787
- Cer: 1.1410
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: 3e-05
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 1500
- num_epochs: 40
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 4.7120 | 0.7828 | 200 | 2.2863 | 4.5120 | 3.2971 |
| 3.3904 | 1.5636 | 400 | 1.7828 | 6.8384 | 5.2278 |
| 2.7891 | 2.3444 | 600 | 1.5849 | 7.0761 | 5.7974 |
| 2.4542 | 3.1252 | 800 | 1.4924 | 6.5835 | 4.8679 |
| 2.2607 | 3.9080 | 1000 | 1.4447 | 7.2391 | 5.5257 |
| 1.9499 | 4.6888 | 1200 | 1.4320 | 7.4224 | 5.6994 |
| 1.6370 | 5.4697 | 1400 | 1.4568 | 3.7166 | 2.8622 |
| 1.2351 | 6.2505 | 1600 | 1.4645 | 4.5953 | 3.4377 |
| 1.2110 | 7.0313 | 1800 | 1.4919 | 3.1556 | 2.4271 |
| 0.9618 | 7.8141 | 2000 | 1.5336 | 1.8855 | 1.4706 |
| 0.6975 | 8.5949 | 2200 | 1.6017 | 2.1892 | 1.7822 |
| 0.4658 | 9.3757 | 2400 | 1.6394 | 2.2635 | 1.7316 |
| 0.3487 | 10.1566 | 2600 | 1.7045 | 3.0183 | 2.3890 |
| 0.3661 | 10.9393 | 2800 | 1.7281 | 2.0836 | 1.6340 |
| 0.2356 | 11.7202 | 3000 | 1.7828 | 2.3525 | 1.8445 |
| 0.1666 | 12.5010 | 3200 | 1.8371 | 2.9169 | 2.3832 |
| 0.1181 | 13.2818 | 3400 | 1.8699 | 2.3120 | 1.8057 |
| 0.1219 | 14.0626 | 3600 | 1.8986 | 2.4652 | 1.8801 |
| 0.1060 | 14.8454 | 3800 | 1.9228 | 2.6044 | 2.1090 |
| 0.0893 | 15.6262 | 4000 | 1.9650 | 2.4527 | 1.9012 |
| 0.0792 | 16.4070 | 4200 | 1.9731 | 2.1270 | 1.6754 |
| 0.0692 | 17.1879 | 4400 | 2.0064 | 2.1618 | 1.6681 |
| 0.0759 | 17.9706 | 4600 | 2.0330 | 2.2145 | 1.7385 |
| 0.0710 | 18.7515 | 4800 | 2.0608 | 1.6744 | 1.2720 |
| 0.0648 | 19.5323 | 5000 | 2.0659 | 1.9503 | 1.5219 |
| 0.0597 | 20.3131 | 5200 | 2.0910 | 1.8122 | 1.4153 |
| 0.0594 | 21.0939 | 5400 | 2.1002 | 1.5923 | 1.2476 |
| 0.0609 | 21.8767 | 5600 | 2.1201 | 1.7642 | 1.3755 |
| 0.0500 | 22.6575 | 5800 | 2.1471 | 1.8116 | 1.4039 |
| 0.0502 | 23.4384 | 6000 | 2.1587 | 1.9254 | 1.5046 |
| 0.0479 | 24.2192 | 6200 | 2.1605 | 1.6193 | 1.2676 |
| 0.0481 | 25.0 | 6400 | 2.1822 | 1.8192 | 1.4223 |
| 0.0486 | 25.7828 | 6600 | 2.1944 | 1.7241 | 1.3362 |
| 0.0470 | 26.5636 | 6800 | 2.2046 | 1.8708 | 1.5019 |
| 0.0466 | 27.3444 | 7000 | 2.2174 | 2.4272 | 1.8908 |
| 0.0456 | 28.1252 | 7200 | 2.2303 | 1.8501 | 1.4650 |
| 0.0462 | 28.9080 | 7400 | 2.2442 | 1.5173 | 1.1813 |
| 0.0473 | 29.6888 | 7600 | 2.2529 | 1.7588 | 1.3726 |
| 0.0453 | 30.4697 | 7800 | 2.2645 | 1.5270 | 1.1752 |
| 0.0441 | 31.2505 | 8000 | 2.2738 | 1.8864 | 1.4603 |
| 0.0446 | 32.0313 | 8200 | 2.2797 | 1.6456 | 1.2651 |
| 0.0456 | 32.8141 | 8400 | 2.2902 | 1.7098 | 1.3178 |
| 0.0441 | 33.5949 | 8600 | 2.2910 | 1.3894 | 1.0824 |
| 0.0443 | 34.3757 | 8800 | 2.3041 | 1.7588 | 1.3565 |
| 0.0442 | 35.1566 | 9000 | 2.3079 | 1.5122 | 1.1638 |
| 0.0434 | 35.9393 | 9200 | 2.3097 | 1.4354 | 1.1082 |
| 0.0440 | 36.7202 | 9400 | 2.3179 | 1.4892 | 1.1516 |
| 0.0432 | 37.5010 | 9600 | 2.3211 | 1.4881 | 1.1498 |
| 0.0440 | 38.2818 | 9800 | 2.3246 | 1.4495 | 1.1159 |
| 0.0430 | 39.0626 | 10000 | 2.3263 | 1.4698 | 1.1395 |
| 0.0430 | 39.8454 | 10200 | 2.3273 | 1.4787 | 1.1410 |
Framework versions
- Transformers 5.3.0.dev0
- Pytorch 2.7.1+cu128
- Datasets 3.6.0
- Tokenizers 0.22.2