LugandaASRwav2Vec300M
This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_13_0 dataset. It achieves the following results on the evaluation set:
- Loss: 0.1741
- Wer: 0.2231
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: 4
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 24
- total_train_batch_size: 96
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 200
- num_epochs: 3
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 6.4394 | 0.14 | 100 | 2.9784 | 1.0 |
| 2.8739 | 0.27 | 200 | 2.7056 | 1.0000 |
| 1.2203 | 0.41 | 300 | 0.5656 | 0.7264 |
| 0.4507 | 0.54 | 400 | 0.3978 | 0.5258 |
| 0.3657 | 0.68 | 500 | 0.3314 | 0.4416 |
| 0.3131 | 0.81 | 600 | 0.2996 | 0.4049 |
| 0.2886 | 0.95 | 700 | 0.2823 | 0.3766 |
| 0.2535 | 1.08 | 800 | 0.2517 | 0.3317 |
| 0.2279 | 1.22 | 900 | 0.2407 | 0.3190 |
| 0.2209 | 1.36 | 1000 | 0.2296 | 0.3077 |
| 0.2075 | 1.49 | 1100 | 0.2228 | 0.2931 |
| 0.1983 | 1.63 | 1200 | 0.2139 | 0.2809 |
| 0.1902 | 1.76 | 1300 | 0.2093 | 0.2688 |
| 0.1931 | 1.9 | 1400 | 0.2019 | 0.2666 |
| 0.1741 | 2.03 | 1500 | 0.1951 | 0.2521 |
| 0.1481 | 2.17 | 1600 | 0.1934 | 0.2435 |
| 0.1423 | 2.3 | 1700 | 0.1912 | 0.2409 |
| 0.1413 | 2.44 | 1800 | 0.1841 | 0.2368 |
| 0.1361 | 2.58 | 1900 | 0.1813 | 0.2310 |
| 0.1337 | 2.71 | 2000 | 0.1775 | 0.2279 |
| 0.1358 | 2.85 | 2100 | 0.1756 | 0.2247 |
| 0.133 | 2.98 | 2200 | 0.1741 | 0.2231 |
Framework versions
- Transformers 4.32.0
- Pytorch 2.0.1+cu118
- Datasets 2.13.0
- Tokenizers 0.13.3