Whisper Small Egyptian Arabic 10% - Mohanad Khaled
This model is a fine-tuned version of openai/whisper-small on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 1.7599
- Wer: 85.6220
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 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: 500
- num_epochs: 5
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 1.1753 | 1.0 | 62 | 1.7765 | 86.8052 |
| 0.8825 | 2.0 | 124 | 1.6395 | 81.6148 |
| 0.602 | 3.0 | 186 | 1.6509 | 85.8805 |
| 0.4529 | 4.0 | 248 | 1.6849 | 82.5694 |
| 0.3231 | 4.9306 | 305 | 1.7599 | 85.6220 |
Framework versions
- Transformers 4.49.0
- Pytorch 2.5.1+cu121
- Datasets 3.3.2
- Tokenizers 0.21.0
- Downloads last month
- 31
Inference Providers
NEW
This model is not currently available via any of the supported Inference Providers.
Model tree for Mohanad03/whisper-small-egyptian-arabic
Base model
openai/whisper-small