whisper-tiny-tunisian-arabic
This model is a fine-tuned version of openai/whisper-tiny on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.3653
- Wer: 0.4497
- Normalized Wer: 0.4478
- Normalized Cer: 0.2430
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: 1e-05
- train_batch_size: 16
- eval_batch_size: 16
- 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_ratio: 0.05
- num_epochs: 3.0
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Normalized Cer | Normalized Wer | Wer |
|---|---|---|---|---|---|---|
| 2.2906 | 0.2705 | 500 | 1.0503 | 0.5057 | 0.8108 | 0.8134 |
| 1.7705 | 0.5410 | 1000 | 0.8070 | 0.3889 | 0.6618 | 0.6656 |
| 1.5058 | 0.8115 | 1500 | 0.6639 | 0.3827 | 0.7675 | 0.7726 |
| 1.2915 | 1.0822 | 2000 | 0.5853 | 0.3094 | 0.6178 | 0.6217 |
| 1.1489 | 1.3527 | 2500 | 0.5212 | 0.5873 | 0.5847 | 0.2939 |
| 1.0187 | 1.6232 | 3000 | 0.4684 | 0.4694 | 0.4669 | 0.2528 |
| 0.9752 | 1.8937 | 3500 | 0.4411 | 0.4541 | 0.4522 | 0.2460 |
| 0.8932 | 2.1645 | 4000 | 0.4037 | 0.4510 | 0.4484 | 0.2434 |
| 0.839 | 2.4349 | 4500 | 0.3848 | 0.4459 | 0.4433 | 0.2422 |
| 0.8238 | 2.7054 | 5000 | 0.3719 | 0.4242 | 0.4223 | 0.2207 |
| 0.7789 | 2.9759 | 5500 | 0.3653 | 0.4497 | 0.4478 | 0.2430 |
Framework versions
- Transformers 4.47.1
- Pytorch 2.5.1+cu124
- Datasets 3.2.0
- Tokenizers 0.21.4