whisper-small-finetuned-iwslt-welsh
This model is a fine-tuned version of openai/whisper-small on the covost2 dataset. It achieves the following results on the evaluation set:
- Loss: 3.7340
- Wer: 82.1085
- Bleu: 14.0988
- Chrf++: 33.5907
- Comet: 0.5541
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: 5e-05
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- 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: 5000
- mixed_precision_training: Native AMP
- label_smoothing_factor: 0.1
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Bleu | Chrf++ | Comet |
|---|---|---|---|---|---|---|---|
| 1.6851 | 16.0256 | 1250 | 3.2626 | 89.7478 | 11.2753 | 31.8587 | 0.5606 |
| 1.4611 | 32.0513 | 2500 | 3.4611 | 82.4896 | 14.0591 | 34.1147 | 0.5704 |
| 1.4471 | 48.0769 | 3750 | 3.6301 | 81.8000 | 14.0228 | 33.4928 | 0.5625 |
| 1.4415 | 64.1026 | 5000 | 3.7340 | 82.1085 | 14.0988 | 33.5907 | 0.5541 |
Framework versions
- Transformers 4.52.0
- Pytorch 2.11.0+cu128
- Datasets 2.21.0
- Tokenizers 0.21.4