whisper-large-arabic-dialects-v5
This model is a fine-tuned version of openai/whisper-large-v3 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.3185
- Wer: 18.5555
- Cer: 9.6321
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: 8e-06
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- 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: cosine
- lr_scheduler_warmup_steps: 8000
- num_epochs: 5
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 0.383 | 0.1390 | 5000 | 0.3887 | 29.3404 | 16.0198 |
| 0.3528 | 0.2780 | 10000 | 0.3820 | 26.1219 | 12.6343 |
| 0.3443 | 0.4170 | 15000 | 0.3583 | 25.1779 | 11.8628 |
| 0.3422 | 0.5559 | 20000 | 0.3410 | 23.6417 | 11.4832 |
| 0.3103 | 0.6949 | 25000 | 0.3280 | 22.2940 | 10.9650 |
| 0.2942 | 0.8339 | 30000 | 0.3199 | 22.1227 | 11.0281 |
| 0.3076 | 0.9729 | 35000 | 0.3111 | 21.6923 | 10.8906 |
| 0.2502 | 1.1119 | 40000 | 0.3059 | 21.1143 | 10.5784 |
| 0.243 | 1.2509 | 45000 | 0.2973 | 20.7546 | 10.4810 |
| 0.2236 | 1.3899 | 50000 | 0.2958 | 20.3886 | 10.3351 |
| 0.2341 | 1.5288 | 55000 | 0.2906 | 20.1640 | 10.0796 |
| 0.2161 | 1.6678 | 60000 | 0.2840 | 19.7242 | 9.9189 |
| 0.232 | 1.8068 | 65000 | 0.2781 | 19.6928 | 9.9768 |
| 0.2296 | 1.9458 | 70000 | 0.2747 | 19.0896 | 9.5917 |
| 0.161 | 2.0848 | 75000 | 0.2826 | 19.1571 | 9.6719 |
| 0.1626 | 2.2238 | 80000 | 0.2798 | 19.2042 | 9.8972 |
| 0.1429 | 2.3628 | 85000 | 0.2832 | 18.9545 | 9.5847 |
| 0.1539 | 2.5017 | 90000 | 0.2766 | 18.6718 | 9.4762 |
| 0.1544 | 2.6407 | 95000 | 0.2724 | 18.4471 | 9.3818 |
| 0.1553 | 2.7797 | 100000 | 0.2715 | 18.4471 | 9.4508 |
| 0.1546 | 2.9187 | 105000 | 0.2683 | 18.2147 | 9.3637 |
| 0.0886 | 3.0577 | 110000 | 0.2911 | 18.6372 | 9.5890 |
| 0.0847 | 3.1967 | 115000 | 0.2894 | 18.4943 | 9.4961 |
| 0.0792 | 3.3356 | 120000 | 0.2909 | 18.2524 | 9.4680 |
| 0.0809 | 3.4746 | 125000 | 0.2901 | 18.3168 | 9.4424 |
| 0.0719 | 3.6136 | 130000 | 0.2929 | 18.3827 | 9.4867 |
| 0.074 | 3.7526 | 135000 | 0.2914 | 18.1518 | 9.3884 |
| 0.0802 | 3.8916 | 140000 | 0.2890 | 18.1220 | 9.3788 |
| 0.0383 | 4.0306 | 145000 | 0.3115 | 18.5068 | 9.6366 |
| 0.0425 | 4.1696 | 150000 | 0.3164 | 18.4000 | 9.5148 |
| 0.0367 | 4.3085 | 155000 | 0.3168 | 18.6529 | 9.6903 |
| 0.0343 | 4.4475 | 160000 | 0.3184 | 18.5272 | 9.5974 |
| 0.0329 | 4.5865 | 165000 | 0.3182 | 18.4660 | 9.5148 |
| 0.0352 | 4.7255 | 170000 | 0.3189 | 18.6388 | 9.6954 |
| 0.0392 | 4.8645 | 175000 | 0.3185 | 18.5555 | 9.6321 |
Framework versions
- Transformers 4.55.4
- Pytorch 2.7.1+cu128
- Datasets 3.6.0
- Tokenizers 0.21.4