wav2vec2-large-xlsr-53-vietnamese-finetuned-demo-v2.0-colab
This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:
- Loss: 0.0097
- Wer: 0.1424
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: 0.0003
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- optimizer: Use OptimizerNames.ADAMW_TORCH 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: 500
- num_epochs: 30
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 8.7632 | 3.2 | 400 | 3.4087 | 1.0 |
| 2.015 | 6.4 | 800 | 0.5443 | 0.5681 |
| 0.4694 | 9.6 | 1200 | 0.2067 | 0.3400 |
| 0.2993 | 12.8 | 1600 | 0.0970 | 0.2381 |
| 0.221 | 16.0 | 2000 | 0.0514 | 0.1885 |
| 0.176 | 19.2 | 2400 | 0.0311 | 0.1650 |
| 0.149 | 22.4 | 2800 | 0.0189 | 0.1510 |
| 0.1283 | 25.6 | 3200 | 0.0125 | 0.1456 |
| 0.1163 | 28.8 | 3600 | 0.0097 | 0.1424 |
Framework versions
- Transformers 4.48.3
- Pytorch 2.5.1+cu124
- Datasets 3.3.2
- Tokenizers 0.21.0