whisper-small-finetuned-iwslt-irish-welsh-50-50-synthetic
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: 3.3705
- Wer: 95.9162
- Bleu: 8.7162
- Chrf++: 29.5222
- Comet: 0.5384
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: 15000
- mixed_precision_training: Native AMP
- label_smoothing_factor: 0.1
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Bleu | Chrf++ | Comet |
|---|---|---|---|---|---|---|---|
| 2.9847 | 0.2132 | 2500 | 3.6076 | 111.2215 | 3.9335 | 21.7113 | 0.4847 |
| 2.3154 | 0.4264 | 5000 | 3.4639 | 99.7199 | 6.3091 | 25.9006 | 0.5034 |
| 2.1807 | 0.6396 | 7500 | 3.3756 | 103.3701 | 7.1207 | 26.9444 | 0.5208 |
| 2.0977 | 0.8528 | 10000 | 3.3650 | 91.2089 | 9.0337 | 28.8094 | 0.5329 |
| 1.9992 | 1.0660 | 12500 | 3.3632 | 95.9342 | 8.6023 | 29.5014 | 0.5384 |
| 1.9005 | 1.2792 | 15000 | 3.3705 | 95.9162 | 8.7162 | 29.5222 | 0.5384 |
Framework versions
- Transformers 4.52.0
- Pytorch 2.12.1+cu130
- Datasets 2.21.0
- Tokenizers 0.21.4