wav2vec2-xls-r-300m-Korean-children-pronunciation-jamo-based-semi-supervised_V2
This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.0802
- Cer: 0.1160
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: 1
- seed: 42
- gradient_accumulation_steps: 8
- total_train_batch_size: 32
- 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: cosine
- lr_scheduler_warmup_steps: 2000
- num_epochs: 4
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Cer |
|---|---|---|---|---|
| 1.6482 | 0.256 | 2000 | 0.9305 | 0.2902 |
| 0.5812 | 0.512 | 4000 | 0.3067 | 0.1785 |
| 0.4531 | 0.768 | 6000 | 0.2110 | 0.1556 |
| 0.3864 | 1.0239 | 8000 | 0.1677 | 0.1434 |
| 0.3448 | 1.2799 | 10000 | 0.1385 | 0.1346 |
| 0.311 | 1.5359 | 12000 | 0.1276 | 0.1310 |
| 0.2957 | 1.7919 | 14000 | 0.1057 | 0.1239 |
| 0.2712 | 2.0479 | 16000 | 0.1035 | 0.1235 |
| 0.2652 | 2.3039 | 18000 | 0.0946 | 0.1208 |
| 0.2544 | 2.5599 | 20000 | 0.0904 | 0.1188 |
| 0.2514 | 2.8159 | 22000 | 0.0870 | 0.1178 |
| 0.2332 | 3.0718 | 24000 | 0.0818 | 0.1165 |
| 0.2401 | 3.3278 | 26000 | 0.0821 | 0.1165 |
| 0.2323 | 3.5838 | 28000 | 0.0801 | 0.1162 |
| 0.2323 | 3.8398 | 30000 | 0.0802 | 0.1160 |
Framework versions
- Transformers 4.57.0
- Pytorch 2.9.1+cu128
- Datasets 4.2.0
- Tokenizers 0.22.1