Whisper Small Chinese MOE Response
This model is a fine-tuned version of Kathernie/whisper-small-all on the MOE Response Chinese dataset. It achieves the following results on the evaluation set:
- Loss: 0.2640
- Cer: 11.0180
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-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 250
- training_steps: 2000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Cer |
|---|---|---|---|---|
| 0.2252 | 1.0811 | 200 | 0.2339 | 12.4230 |
| 0.1268 | 2.1622 | 400 | 0.2223 | 10.9194 |
| 0.056 | 3.2432 | 600 | 0.2242 | 10.8701 |
| 0.023 | 4.3243 | 800 | 0.2387 | 11.3384 |
| 0.01 | 5.4054 | 1000 | 0.2546 | 11.2645 |
| 0.0044 | 6.4865 | 1200 | 0.2515 | 11.2891 |
| 0.0028 | 7.5676 | 1400 | 0.2552 | 10.9440 |
| 0.0017 | 8.6486 | 1600 | 0.2623 | 11.2645 |
| 0.0017 | 9.7297 | 1800 | 0.2624 | 10.9933 |
| 0.001 | 10.8108 | 2000 | 0.2640 | 11.0180 |
Framework versions
- Transformers 4.42.3
- Pytorch 2.2.0
- Datasets 2.20.0
- Tokenizers 0.19.1