XLSR_Fine_Tuned_Urdu_V2
This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the common_voice_8_0 dataset. It achieves the following results on the evaluation set:
- Loss: 0.8023
- Wer: 0.4382
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: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 1000
- num_epochs: 30
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 6.424 | 3.25 | 1000 | 2.9777 | 1.0 |
| 1.4315 | 6.49 | 2000 | 0.8493 | 0.5896 |
| 0.6938 | 9.74 | 3000 | 0.7438 | 0.4978 |
| 0.5129 | 12.99 | 4000 | 0.7480 | 0.4785 |
| 0.4133 | 16.23 | 5000 | 0.7568 | 0.4600 |
| 0.3496 | 19.48 | 6000 | 0.7387 | 0.4471 |
| 0.3133 | 22.73 | 7000 | 0.7655 | 0.4426 |
| 0.2767 | 25.97 | 8000 | 0.8081 | 0.4530 |
| 0.2581 | 29.22 | 9000 | 0.8023 | 0.4382 |
Framework versions
- Transformers 4.21.0
- Pytorch 1.11.0+cu113
- Datasets 2.4.0
- Tokenizers 0.12.1