Whisper Small Persian
This model is a fine-tuned version of openai/whisper-small on the Fleurs Farsi dataset. It achieves the following results on the evaluation set:
- Loss: 0.3818
- Wer: 25.8171
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: 32
- eval_batch_size: 1
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 64
- 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: 100
- training_steps: 500
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.3277 | 2.0 | 50 | 0.3894 | 34.8873 |
| 0.1146 | 4.0 | 100 | 0.3268 | 28.6293 |
| 0.0373 | 6.0 | 150 | 0.3289 | 27.3372 |
| 0.0142 | 8.0 | 200 | 0.3390 | 27.5399 |
| 0.0036 | 10.0 | 250 | 0.3523 | 26.5265 |
| 0.0024 | 12.0 | 300 | 0.3677 | 26.5265 |
| 0.001 | 14.0 | 350 | 0.3734 | 26.5011 |
| 0.0007 | 16.0 | 400 | 0.3777 | 26.1971 |
| 0.0007 | 18.0 | 450 | 0.3807 | 25.9944 |
| 0.0006 | 20.0 | 500 | 0.3818 | 25.8171 |
Framework versions
- Transformers 4.48.0
- Pytorch 2.8.0+cu126
- Datasets 4.1.1
- Tokenizers 0.21.4