luganda_wav2vec2_ctc
This model is a fine-tuned version of facebook/wav2vec2-base on the common_voice_7_0 dataset. It achieves the following results on the evaluation set:
- Loss: 0.7622
- Wer: 0.5422
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: 48
- 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: 60
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 4.2675 | 3.6 | 500 | 1.9999 | 0.9999 |
| 0.5754 | 7.19 | 1000 | 0.6976 | 0.7050 |
| 0.231 | 10.79 | 1500 | 0.6153 | 0.6440 |
| 0.1557 | 14.39 | 2000 | 0.6581 | 0.6130 |
| 0.1221 | 17.99 | 2500 | 0.6718 | 0.6063 |
| 0.1013 | 21.58 | 3000 | 0.6711 | 0.5934 |
| 0.0871 | 25.18 | 3500 | 0.6728 | 0.5731 |
| 0.0751 | 28.78 | 4000 | 0.6729 | 0.5726 |
| 0.0666 | 32.37 | 4500 | 0.6884 | 0.5689 |
| 0.0604 | 35.97 | 5000 | 0.7452 | 0.5609 |
| 0.0543 | 39.57 | 5500 | 0.7302 | 0.5616 |
| 0.0488 | 43.17 | 6000 | 0.7414 | 0.5480 |
| 0.0448 | 46.76 | 6500 | 0.7662 | 0.5560 |
| 0.042 | 50.36 | 7000 | 0.7629 | 0.5433 |
| 0.038 | 53.96 | 7500 | 0.7582 | 0.5479 |
| 0.0353 | 57.55 | 8000 | 0.7622 | 0.5422 |
Framework versions
- Transformers 4.37.2
- Pytorch 2.1.0+cu121
- Datasets 2.17.1
- Tokenizers 0.15.2