dgx1_whisper_tiny_finetune_teacher_no_noise_mozilla_100_epochs_batch_32_resume_training
This model is a fine-tuned version of rohitp1/dgx1_whisper_tiny_finetune_teacher_no_noise_mozilla_40_epochs_batch_32 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 1.1796
- Wer: 37.3233
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: 100
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.0059 | 7.35 | 500 | 0.9865 | 36.5295 |
| 0.0368 | 14.7 | 1000 | 1.0064 | 37.8595 |
| 0.0545 | 22.06 | 1500 | 1.0131 | 37.9918 |
| 0.0308 | 29.41 | 2000 | 1.0712 | 37.3895 |
| 0.0189 | 36.76 | 2500 | 1.1299 | 37.4278 |
| 0.0146 | 44.12 | 3000 | 1.1796 | 37.3233 |
Framework versions
- Transformers 4.25.1
- Pytorch 1.12.1
- Datasets 2.8.0
- Tokenizers 0.13.2