XLS-R-demo-google-colab-Ezra_William_Prod_3
This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the common_voice_13_0 dataset. It achieves the following results on the evaluation set:
- Loss: 0.7896
- Wer: 0.6979
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: 0.0001
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 12
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 5.4479 | 1.0 | 121 | 2.9741 | 1.0 |
| 2.9543 | 2.0 | 242 | 2.9297 | 1.0 |
| 2.9306 | 3.0 | 363 | 2.9112 | 1.0 |
| 2.9216 | 4.0 | 484 | 2.9071 | 1.0 |
| 2.8968 | 5.0 | 605 | 2.8713 | 1.0 |
| 2.8822 | 6.0 | 726 | 2.8446 | 1.0 |
| 2.8421 | 7.0 | 847 | 2.5157 | 1.0 |
| 2.5763 | 8.0 | 968 | 1.5780 | 0.9964 |
| 1.9449 | 9.0 | 1089 | 0.9864 | 0.8132 |
| 1.0398 | 10.0 | 1210 | 0.8565 | 0.7348 |
| 0.9162 | 11.0 | 1331 | 0.7941 | 0.7043 |
| 0.8909 | 12.0 | 1452 | 0.7896 | 0.6979 |
Framework versions
- Transformers 4.42.3
- Pytorch 2.3.1+cu121
- Datasets 2.20.0
- Tokenizers 0.19.1