whisper-small-finetuned-iwslt-irish-transfer-welsh-no-frozen-decoder
This model is a fine-tuned version of openai/whisper-small on the None dataset. It achieves the following results on the evaluation set:
- Loss: 6.7416
- Wer: 105.8005
- Bleu: 0.6736
- Chrf++: 12.5667
- Comet: 0.4173
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 |
|---|---|---|---|---|---|---|---|
| 1.887 | 2.6709 | 1250 | 5.9918 | 111.1673 | 0.9638 | 13.3780 | 0.4054 |
| 1.5006 | 5.3419 | 2500 | 6.3732 | 103.9031 | 0.7768 | 12.2773 | 0.4111 |
| 1.4611 | 8.0128 | 3750 | 6.4941 | 103.6050 | 0.7533 | 12.2820 | 0.4108 |
| 1.423 | 10.6838 | 5000 | 6.7416 | 105.8005 | 0.6736 | 12.5667 | 0.4173 |
Framework versions
- Transformers 4.52.0
- Pytorch 2.12.1+cu130
- Datasets 2.21.0
- Tokenizers 0.21.4