Whisper - Serbian Model
This model is a fine-tuned version of openai/whisper-large-v2 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.1714
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: 0.0009
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- 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: 400
- training_steps: 12000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 0.1206 | 0.0705 | 500 | 0.1705 |
| 0.1162 | 0.1409 | 1000 | 0.1626 |
| 0.1189 | 0.2114 | 1500 | 0.1548 |
| 0.117 | 0.2819 | 2000 | 0.1467 |
| 0.1117 | 0.3524 | 2500 | 0.1386 |
| 0.1148 | 0.4228 | 3000 | 0.1313 |
| 0.1128 | 0.4933 | 3500 | 0.1264 |
| 0.1277 | 0.5638 | 4000 | 0.1228 |
| 0.1851 | 0.6342 | 4500 | 0.1940 |
| 0.1678 | 0.7047 | 5000 | 0.1714 |
Framework versions
- PEFT 0.13.2
- Transformers 4.46.3
- Pytorch 2.5.1
- Datasets 3.0.0
- Tokenizers 0.20.3