dgx1_whisper_tiny_finetune_teacher_no_noise_mozilla_40_epochs_batch_32
This model is a fine-tuned version of openai/whisper-tiny.en on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.8362
- Wer: 33.6328
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: 32
- eval_batch_size: 1
- seed: 42
- gradient_accumulation_steps: 32
- 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: 40
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.5135 | 7.35 | 500 | 0.6979 | 36.6444 |
| 0.1026 | 14.7 | 1000 | 0.8056 | 35.3631 |
| 0.0074 | 22.06 | 1500 | 0.8321 | 34.1480 |
| 0.0004 | 29.41 | 2000 | 0.8341 | 33.6606 |
| 0.0003 | 36.76 | 2500 | 0.8362 | 33.6328 |
Framework versions
- Transformers 4.25.1
- Pytorch 1.12.1
- Datasets 2.8.0
- Tokenizers 0.13.2