Whisper Base Korean
This model is a fine-tuned version of openai/whisper-base on the google/fleurs ko_kr dataset. It achieves the following results on the evaluation set:
- Loss: 0.4901
- Wer: 27.4344
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: 5e-07
- train_batch_size: 32
- eval_batch_size: 32
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 5000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.3225 | 66.0 | 500 | 0.5002 | 27.9275 |
| 0.1185 | 133.0 | 1000 | 0.4901 | 27.4344 |
| 0.0468 | 199.0 | 1500 | 0.5047 | 27.4696 |
| 0.0268 | 266.0 | 2000 | 0.5147 | 27.8746 |
| 0.0189 | 333.0 | 2500 | 0.5218 | 28.0507 |
| 0.0145 | 399.0 | 3000 | 0.5273 | 28.4733 |
| 0.0121 | 466.0 | 3500 | 0.5318 | 28.6318 |
| 0.0107 | 533.0 | 4000 | 0.5352 | 28.6846 |
| 0.0098 | 599.0 | 4500 | 0.5376 | 28.8079 |
| 0.0095 | 666.0 | 5000 | 0.5385 | 28.8079 |
Framework versions
- Transformers 4.38.0.dev0
- Pytorch 2.1.2+cu121
- Datasets 2.16.2.dev0
- Tokenizers 0.15.0