wav2vec2-large-xls-r-300m-slovak
This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA-FOUNDATION/COMMON_VOICE_7_0 - SK dataset. It achieves the following results on the evaluation set:
- Loss: 0.2915
- Wer: 0.2481
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: 7e-05
- train_batch_size: 32
- eval_batch_size: 1
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 3000
- num_epochs: 100.0
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 1.0076 | 19.74 | 3000 | 0.3274 | 0.3806 |
| 0.6889 | 39.47 | 6000 | 0.2824 | 0.2942 |
| 0.5863 | 59.21 | 9000 | 0.2700 | 0.2735 |
| 0.4798 | 78.95 | 12000 | 0.2844 | 0.2602 |
| 0.4399 | 98.68 | 15000 | 0.2907 | 0.2489 |
Framework versions
- Transformers 4.16.0.dev0
- Pytorch 1.10.1+cu102
- Datasets 1.17.1.dev0
- Tokenizers 0.11.0