subhadeep_whisper_small_finetune_teacher_no_noise_libri_360_hours_100_epochs_batch_8
This model is a fine-tuned version of openai/whisper-small.en on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.1135
- Wer: 9.0383
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: 0.0005
- train_batch_size: 4
- eval_batch_size: 1
- seed: 42
- gradient_accumulation_steps: 256
- total_train_batch_size: 1024
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine_with_restarts
- lr_scheduler_warmup_ratio: 0.2
- num_epochs: 100
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.5483 | 0.98 | 100 | 0.1273 | 9.9482 |
| 0.0795 | 1.97 | 200 | 0.0815 | 8.6467 |
| 0.0415 | 2.96 | 300 | 0.0788 | 8.3986 |
| 0.0257 | 3.95 | 400 | 0.0849 | 8.3857 |
| 0.0325 | 4.95 | 500 | 0.0993 | 8.8471 |
| 0.0219 | 5.94 | 600 | 0.0951 | 8.7350 |
| 0.018 | 6.93 | 700 | 0.0952 | 8.7000 |
| 0.0159 | 7.92 | 800 | 0.1098 | 8.7901 |
| 0.017 | 8.91 | 900 | 0.1135 | 9.0383 |
Framework versions
- Transformers 4.25.1
- Pytorch 1.12.1
- Datasets 2.8.0
- Tokenizers 0.13.2