english-filipino-wav2vec2-l-xls-r-test-08
This model is a fine-tuned version of jonatasgrosman/wav2vec2-large-xlsr-53-english on the filipino_voice dataset. It achieves the following results on the evaluation set:
- Loss: 0.5968
- Wer: 0.4255
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: 8
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 20
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 3.3434 | 2.09 | 400 | 2.2857 | 0.9625 |
| 1.6304 | 4.19 | 800 | 1.1547 | 0.7268 |
| 0.9231 | 6.28 | 1200 | 1.0252 | 0.6186 |
| 0.6098 | 8.38 | 1600 | 0.9371 | 0.5494 |
| 0.4922 | 10.47 | 2000 | 0.7092 | 0.5478 |
| 0.3652 | 12.57 | 2400 | 0.7358 | 0.5149 |
| 0.2735 | 14.66 | 2800 | 0.6270 | 0.4646 |
| 0.2038 | 16.75 | 3200 | 0.5717 | 0.4506 |
| 0.1552 | 18.85 | 3600 | 0.5968 | 0.4255 |
Framework versions
- Transformers 4.11.3
- Pytorch 1.10.0+cu113
- Datasets 1.18.3
- Tokenizers 0.10.3