This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA-FOUNDATION/COMMON_VOICE_9_0 - OR dataset. It achieves the following results on the evaluation set:
- Loss: 0.7886
- Wer: 0.5495
- Cer: 0.1311
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: 7.5e-05
- train_batch_size: 64
- eval_batch_size: 64
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 128
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- training_steps: 3071
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 3.5875 | 66.62 | 400 | 3.4289 | 1.0 | 1.0 |
| 1.4065 | 133.31 | 800 | 0.7243 | 0.6619 | 0.1734 |
| 1.007 | 199.92 | 1200 | 0.6611 | 0.5831 | 0.1457 |
| 0.7984 | 266.62 | 1600 | 0.6387 | 0.5520 | 0.1332 |
| 0.6117 | 333.31 | 2000 | 0.7424 | 0.5682 | 0.1376 |
| 0.4926 | 399.92 | 2400 | 0.7627 | 0.5514 | 0.1314 |
| 0.416 | 466.62 | 2800 | 0.7816 | 0.5604 | 0.1320 |
Framework versions
- Transformers 4.19.0.dev0
- Pytorch 1.11.0+cu102
- Datasets 2.1.1.dev0
- Tokenizers 0.12.1