ASR-Somali
This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.3660
- Wer: 0.3060
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 |
|---|---|---|---|---|
| 3.1435 | 2.09 | 400 | 0.7624 | 0.7706 |
| 0.5829 | 4.18 | 800 | 0.3646 | 0.3935 |
| 0.3634 | 6.27 | 1200 | 0.3318 | 0.3944 |
| 0.2942 | 8.36 | 1600 | 0.3148 | 0.3403 |
| 0.2419 | 10.44 | 2000 | 0.3000 | 0.3255 |
| 0.2104 | 12.53 | 2400 | 0.2951 | 0.3312 |
| 0.1864 | 14.62 | 2800 | 0.3296 | 0.3083 |
| 0.1666 | 16.71 | 3200 | 0.3264 | 0.3153 |
| 0.148 | 18.8 | 3600 | 0.3188 | 0.3028 |
| 0.1305 | 20.89 | 4000 | 0.3448 | 0.3002 |
| 0.1206 | 22.98 | 4400 | 0.3660 | 0.3060 |
Framework versions
- Transformers 4.28.0
- Pytorch 2.0.0
- Datasets 1.18.3
- Tokenizers 0.13.3