Whisper Small Tamil (Saaras-style)
This model is a fine-tuned version of openai/whisper-small on the abar-uwc/tamil-common-voice_v21 dataset. It achieves the following results on the evaluation set:
- Loss: 0.2188
- Wer: 35.4991
- Cer: 5.8835
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: 32
- eval_batch_size: 16
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 200
- training_steps: 2000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 0.2661 | 2.0 | 500 | 0.1550 | 35.9699 | 6.2758 |
| 0.1193 | 4.0 | 1000 | 0.1628 | 35.8084 | 6.2833 |
| 0.0282 | 6.0 | 1500 | 0.1957 | 36.3331 | 6.3862 |
| 0.0060 | 8.0 | 2000 | 0.2188 | 35.4991 | 5.8835 |
Framework versions
- Transformers 5.12.1
- Pytorch 2.10.0+cu128
- Datasets 5.0.0
- Tokenizers 0.22.2