wav2vec2-xlsr-53-CV-demo-google-colab-Ezra_William_Prod16
This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_13_0 dataset. It achieves the following results on the evaluation set:
- Loss: 0.3355
- Wer: 0.3031
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 |
|---|---|---|---|---|
| 2.9606 | 1.0 | 278 | 2.9210 | 1.0 |
| 2.8429 | 2.0 | 556 | 2.1290 | 1.0 |
| 0.9644 | 3.0 | 834 | 0.5957 | 0.5614 |
| 0.6414 | 4.0 | 1112 | 0.4595 | 0.4643 |
| 0.5396 | 5.0 | 1390 | 0.4189 | 0.4090 |
| 0.4334 | 6.0 | 1668 | 0.3778 | 0.3670 |
| 0.3939 | 7.0 | 1946 | 0.3777 | 0.3544 |
| 0.3738 | 8.0 | 2224 | 0.3511 | 0.3355 |
| 0.3387 | 9.0 | 2502 | 0.3569 | 0.3240 |
| 0.3071 | 10.0 | 2780 | 0.3405 | 0.3165 |
| 0.3129 | 11.0 | 3058 | 0.3313 | 0.3065 |
| 0.2971 | 12.0 | 3336 | 0.3355 | 0.3031 |
Framework versions
- Transformers 4.40.0
- Pytorch 2.2.2+cu121
- Datasets 2.19.0
- Tokenizers 0.19.1