hubert-finetuned-CASIA
This model is a fine-tuned version of facebook/hubert-base-ls960 on the audiofolder dataset. It achieves the following results on the evaluation set:
- Loss: 0.6783
- Accuracy: 0.8
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: 3e-05
- train_batch_size: 4
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 16
- optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 0.1
- num_epochs: 7
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|---|---|---|---|---|
| 6.6906 | 1.0 | 60 | 1.4090 | 0.325 |
| 5.6728 | 2.0 | 120 | 1.1344 | 0.5542 |
| 4.9187 | 3.0 | 180 | 0.9036 | 0.7042 |
| 4.3592 | 4.0 | 240 | 0.7911 | 0.75 |
| 3.4655 | 5.0 | 300 | 0.6783 | 0.8 |
| 3.1525 | 6.0 | 360 | 0.7754 | 0.7417 |
| 2.8022 | 7.0 | 420 | 0.7240 | 0.7333 |
Framework versions
- Transformers 5.0.0
- Pytorch 2.10.0+cu128
- Datasets 4.0.0
- Tokenizers 0.22.2