Whisper medium — Libyan Arabic Dialect2
This model is a fine-tuned version of openai/whisper-medium on the Custom Libyan Arabic podcast — 258 manually transcribed clips dataset. It achieves the following results on the evaluation set:
- Loss: 0.7715
- Wer: 36.1968
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: 1
- eval_batch_size: 1
- seed: 42
- gradient_accumulation_steps: 16
- total_train_batch_size: 16
- optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 25
- training_steps: 500
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.1496 | 6.6897 | 100 | 0.5612 | 34.7790 |
| 0.0344 | 13.3448 | 200 | 0.6559 | 34.9458 |
| 0.0097 | 20.0 | 300 | 0.7456 | 35.2794 |
| 0.0021 | 26.6897 | 400 | 0.7563 | 35.7798 |
| 0.0015 | 33.3448 | 500 | 0.7715 | 36.1968 |
Framework versions
- Transformers 5.12.1
- Pytorch 2.11.0+cu128
- Datasets 4.0.0
- Tokenizers 0.22.2