Whisper - Serbian Model
This model is a fine-tuned version of openai/whisper-large-v3 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.1956
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: 0.002
- 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: 16000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 0.2482 | 0.0705 | 500 | 0.2655 |
| 0.227 | 0.1409 | 1000 | 0.2220 |
| 0.1997 | 0.2114 | 1500 | 0.2047 |
| 0.208 | 0.2819 | 2000 | 0.1990 |
| 0.1959 | 0.3524 | 2500 | 0.1945 |
| 0.2023 | 0.4228 | 3000 | 0.1962 |
| 0.1835 | 0.4933 | 3500 | 0.1882 |
| 0.194 | 0.5638 | 4000 | 0.1893 |
| 0.1876 | 0.6342 | 4500 | 0.1956 |
Framework versions
- PEFT 0.13.2
- Transformers 4.46.3
- Pytorch 2.5.1
- Datasets 3.0.0
- Tokenizers 0.20.3