Whisper Small UK - Bohdan's Fine-Tune
This model is a fine-tuned version of openai/whisper-small on the Google FLEURS (Ukrainian) dataset. It achieves the following results on the evaluation set:
- Loss: 0.3156
- Wer: 17.2136
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: 4
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 16
- 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: 500
- training_steps: 4000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.007 | 5.1020 | 1000 | 0.2776 | 17.6874 |
| 0.0009 | 10.2041 | 2000 | 0.2952 | 17.1358 |
| 0.0005 | 15.3061 | 3000 | 0.3098 | 17.1216 |
| 0.0004 | 20.4082 | 4000 | 0.3156 | 17.2136 |
Framework versions
- Transformers 4.57.1
- Pytorch 2.8.0+cu128
- Datasets 2.21.0
- Tokenizers 0.22.1