openai/whisper-small
This model is a fine-tuned version of openai/whisper-small on the pphuc25/FrenchMed dataset. It achieves the following results on the evaluation set:
- Loss: 1.3133
- Wer: 40.9824
- Cer: 29.0078
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: 0.0001
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 100
- num_epochs: 20
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 0.9467 | 1.0 | 215 | 0.9214 | 52.4194 | 46.7318 |
| 0.4998 | 2.0 | 430 | 0.9703 | 53.0792 | 36.7002 |
| 0.2067 | 3.0 | 645 | 1.0581 | 40.7625 | 26.8749 |
| 0.1092 | 4.0 | 860 | 1.1350 | 63.1965 | 43.3329 |
| 0.0699 | 5.0 | 1075 | 1.1698 | 40.6158 | 26.9988 |
| 0.0602 | 6.0 | 1290 | 1.1891 | 51.3930 | 32.2279 |
| 0.0409 | 7.0 | 1505 | 1.2566 | 53.5191 | 34.8424 |
| 0.036 | 8.0 | 1720 | 1.2424 | 39.8827 | 27.3841 |
| 0.0183 | 9.0 | 1935 | 1.2718 | 39.8094 | 26.1869 |
| 0.0125 | 10.0 | 2150 | 1.2755 | 37.6833 | 25.6915 |
| 0.0072 | 11.0 | 2365 | 1.3042 | 39.4428 | 27.0125 |
| 0.0118 | 12.0 | 2580 | 1.2765 | 44.3548 | 30.6591 |
| 0.0062 | 13.0 | 2795 | 1.2951 | 40.7625 | 28.5537 |
| 0.0045 | 14.0 | 3010 | 1.2897 | 40.4692 | 28.2923 |
| 0.0005 | 15.0 | 3225 | 1.3089 | 42.0088 | 29.4344 |
| 0.0018 | 16.0 | 3440 | 1.3013 | 42.2287 | 29.3794 |
| 0.0004 | 17.0 | 3655 | 1.3101 | 41.1290 | 29.4069 |
| 0.0003 | 18.0 | 3870 | 1.3113 | 40.9824 | 29.0216 |
| 0.0002 | 19.0 | 4085 | 1.3128 | 40.9824 | 29.0078 |
| 0.0002 | 20.0 | 4300 | 1.3133 | 40.9824 | 29.0078 |
Framework versions
- Transformers 4.41.1
- Pytorch 2.3.0
- Datasets 2.19.1
- Tokenizers 0.19.1