wav2vec2-xlsr-dora-afrispeech-all
This model is a fine-tuned version of jonatasgrosman/wav2vec2-large-xlsr-53-english on the afrispeech-200 dataset. It achieves the following results on the evaluation set:
- Loss: 0.4410
- Wer: 0.305
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
- gradient_accumulation_steps: 2
- total_train_batch_size: 16
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 3589
- num_epochs: 10
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 1.4649 | 1.0 | 3590 | 0.5842 | 0.38 |
| 1.4877 | 2.0 | 7180 | 0.5156 | 0.348 |
| 1.1878 | 3.0 | 10770 | 0.4959 | 0.333 |
| 1.0824 | 4.0 | 14360 | 0.4787 | 0.322 |
| 1.2068 | 5.0 | 17950 | 0.4624 | 0.317 |
| 1.0674 | 6.0 | 21540 | 0.4546 | 0.312 |
| 0.9825 | 7.0 | 25130 | 0.4503 | 0.309 |
| 1.0736 | 8.0 | 28720 | 0.4467 | 0.308 |
| 0.9760 | 9.0 | 32310 | 0.4424 | 0.305 |
| 0.9802 | 10.0 | 35900 | 0.4410 | 0.305 |
Framework versions
- PEFT 0.18.1
- Transformers 5.3.0
- Pytorch 2.12.0.dev20260307+cu128
- Datasets 2.19.0
- Tokenizers 0.22.2