whisper_large_v2_arabic_aug
This model is a fine-tuned version of Seyfelislem/whisper_large_ar on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.2033
- Wer: 11.9749
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: 32
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- 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 |
|---|---|---|---|---|
| 0.0872 | 0.33 | 400 | 0.1768 | 13.3808 |
| 0.0686 | 0.67 | 800 | 0.1776 | 13.1368 |
| 0.073 | 1.0 | 1200 | 0.1714 | 12.7051 |
| 0.0265 | 1.33 | 1600 | 0.1789 | 12.5511 |
| 0.0179 | 1.66 | 2000 | 0.1787 | 12.1438 |
| 0.0239 | 2.0 | 2400 | 0.1919 | 13.1743 |
| 0.0089 | 2.33 | 2800 | 0.1945 | 12.2152 |
| 0.0093 | 2.66 | 3200 | 0.1953 | 11.8811 |
| 0.0088 | 2.99 | 3600 | 0.1947 | 12.0763 |
| 0.0017 | 3.33 | 4000 | 0.2033 | 11.9749 |
Framework versions
- Transformers 4.29.1
- Pytorch 1.13.1
- Datasets 2.12.0
- Tokenizers 0.13.3