whisper-synthesized-turkish-2-hour-llr
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: 0.3901
- Wer: 20.1074
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-06
- 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 |
|---|---|---|---|---|
| 1.8697 | 2.08 | 100 | 1.5724 | 43.7232 |
| 1.2267 | 4.17 | 200 | 0.8637 | 90.9905 |
| 0.6451 | 6.25 | 300 | 0.6205 | 85.8473 |
| 0.4958 | 8.33 | 400 | 0.5447 | 93.4368 |
| 0.4053 | 10.42 | 500 | 0.4883 | 52.0644 |
| 0.3149 | 12.5 | 600 | 0.4480 | 32.4821 |
| 0.2608 | 14.58 | 700 | 0.4218 | 24.0334 |
| 0.2245 | 16.67 | 800 | 0.4045 | 22.0764 |
| 0.2038 | 18.75 | 900 | 0.3937 | 19.5227 |
| 0.183 | 20.83 | 1000 | 0.3901 | 20.1074 |
Framework versions
- Transformers 4.28.0
- Pytorch 2.0.0+cu118
- Datasets 2.11.0
- Tokenizers 0.13.3