Whisper Small Vi - finetune viVoice - 70000
This model is a fine-tuned version of openai/whisper-small on the viVoice dataset. It achieves the following results on the evaluation set:
- Loss: 5.7260
- Wer: 14.0767
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: 1.25e-05
- train_batch_size: 32
- eval_batch_size: 8
- seed: 42
- distributed_type: multi-GPU
- num_devices: 2
- total_train_batch_size: 64
- total_eval_batch_size: 16
- 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: 1000
- training_steps: 80000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.1892 | 0.05 | 4000 | 3.5308 | 18.7775 |
| 0.1551 | 0.1 | 8000 | 4.2465 | 18.1171 |
| 0.1444 | 0.15 | 12000 | 4.4830 | 16.9775 |
| 0.1097 | 1.0266 | 16000 | 4.4955 | 16.1357 |
| 0.0966 | 1.0766 | 20000 | 4.8873 | 15.6825 |
| 0.0915 | 1.1266 | 24000 | 4.8408 | 15.6177 |
| 0.0853 | 2.0032 | 28000 | 5.0293 | 15.1904 |
| 0.065 | 2.0532 | 32000 | 5.0290 | 15.8120 |
| 0.0644 | 2.1032 | 36000 | 5.1940 | 14.5299 |
| 0.0584 | 2.1532 | 40000 | 5.3418 | 15.1515 |
| 0.0466 | 3.0298 | 44000 | 5.2564 | 15.2422 |
| 0.0405 | 3.0798 | 48000 | 5.4065 | 14.7112 |
| 0.0412 | 3.1298 | 52000 | 5.5395 | 14.1414 |
| 0.0344 | 4.0064 | 56000 | 5.6079 | 14.5947 |
| 0.0288 | 4.0564 | 60000 | 5.5141 | 14.4911 |
| 0.0257 | 4.1064 | 64000 | 5.6983 | 14.7242 |
| 0.0249 | 4.1564 | 68000 | 5.7079 | 14.0378 |
| 0.0209 | 5.033 | 72000 | 5.5744 | 13.8177 |
| 0.0192 | 5.083 | 76000 | 5.7272 | 14.1803 |
| 0.0185 | 5.133 | 80000 | 5.7260 | 14.0767 |
Framework versions
- Transformers 4.47.1
- Pytorch 2.5.1+cu121
- Datasets 3.2.0
- Tokenizers 0.21.0