Whisper Small Kurdish - Sorani
This model is a fine-tuned version of openai/whisper-medium on the Common Voice 17.0 dataset. It achieves the following results on the evaluation set:
- Loss: 0.3857
- Wer: 34.2123
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: 16
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 8000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.4593 | 1.2330 | 1000 | 0.2504 | 42.2983 |
| 0.1318 | 2.4661 | 2000 | 0.2344 | 37.8429 |
| 0.0625 | 3.6991 | 3000 | 0.2582 | 35.9282 |
| 0.0274 | 4.9322 | 4000 | 0.2927 | 36.6139 |
| 0.009 | 6.1652 | 5000 | 0.3429 | 35.1365 |
| 0.0029 | 7.3983 | 6000 | 0.3625 | 34.7588 |
| 0.0008 | 8.6313 | 7000 | 0.3815 | 34.4740 |
| 0.0003 | 9.8644 | 8000 | 0.3857 | 34.2123 |
Framework versions
- Transformers 4.42.0.dev0
- Pytorch 2.3.1+cu118
- Datasets 2.20.0
- Tokenizers 0.19.1