whisper-synthesized-turkish-2-hour-hlr
This model is a fine-tuned version of openai/whisper-small on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.5741
- Wer: 20.5967
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: 0.0001
- 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: 1000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.7454 | 2.08 | 100 | 0.3140 | 18.8186 |
| 0.0822 | 4.17 | 200 | 0.3872 | 17.8640 |
| 0.0577 | 6.25 | 300 | 0.4162 | 22.8520 |
| 0.0552 | 8.33 | 400 | 0.5068 | 21.3126 |
| 0.0638 | 10.42 | 500 | 0.5803 | 24.4749 |
| 0.0571 | 12.5 | 600 | 0.5954 | 23.7112 |
| 0.0351 | 14.58 | 700 | 0.6020 | 22.5060 |
| 0.0159 | 16.67 | 800 | 0.6010 | 22.9594 |
| 0.0088 | 18.75 | 900 | 0.5819 | 21.6826 |
| 0.0012 | 20.83 | 1000 | 0.5741 | 20.5967 |
Framework versions
- Transformers 4.28.0
- Pytorch 2.0.0+cu118
- Datasets 2.11.0
- Tokenizers 0.13.3