wav2vec2-large-xls-r-300m-turkish-colab
This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice dataset. It achieves the following results on the evaluation set:
- Loss: 0.7532
- Wer: 0.4020
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: Adam with betas=(0.9,0.999) and epsilon=1e-08
- 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 |
|---|---|---|---|---|
| 5.9542 | 1.96 | 400 | 1.5737 | 0.8827 |
| 0.8596 | 3.92 | 800 | 0.7296 | 0.5696 |
| 0.4729 | 5.88 | 1200 | 0.6004 | 0.4934 |
| 0.3364 | 7.84 | 1600 | 0.5776 | 0.4656 |
| 0.2684 | 9.8 | 2000 | 0.6178 | 0.4563 |
| 0.2143 | 11.76 | 2400 | 0.6408 | 0.4690 |
| 0.1744 | 13.72 | 2800 | 0.6704 | 0.4573 |
| 0.1458 | 15.68 | 3200 | 0.7015 | 0.4484 |
| 0.1201 | 17.65 | 3600 | 0.7151 | 0.4228 |
| 0.104 | 19.61 | 4000 | 0.7123 | 0.4195 |
| 0.0887 | 21.57 | 4400 | 0.7102 | 0.4234 |
| 0.0807 | 23.53 | 4800 | 0.7561 | 0.4132 |
| 0.0697 | 25.49 | 5200 | 0.7435 | 0.4075 |
| 0.0611 | 27.45 | 5600 | 0.7465 | 0.4034 |
| 0.0556 | 29.41 | 6000 | 0.7532 | 0.4020 |
Framework versions
- Transformers 4.18.0
- Pytorch 1.11.0
- Datasets 2.1.0
- Tokenizers 0.12.1