wav2vec2LugandaASR
This model is a fine-tuned version of Gemmar/wav2vec2LugandaASR on the common_voice_13_0 dataset. It achieves the following results on the evaluation set:
- Loss: 0.2014
- Wer: 0.2396
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: 32
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 128
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 200
- num_epochs: 5
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 5.8963 | 0.18 | 100 | 2.8825 | 1.0000 |
| 1.1814 | 0.36 | 200 | 0.3787 | 0.4585 |
| 0.3331 | 0.54 | 300 | 0.3166 | 0.3918 |
| 0.2939 | 0.72 | 400 | 0.2811 | 0.3483 |
| 0.2682 | 0.9 | 500 | 0.2652 | 0.3348 |
| 0.2389 | 1.08 | 600 | 0.2565 | 0.3207 |
| 0.2137 | 1.27 | 700 | 0.2452 | 0.3066 |
| 0.2062 | 1.45 | 800 | 0.2356 | 0.3092 |
| 0.2058 | 1.63 | 900 | 0.2346 | 0.2928 |
| 0.2055 | 1.81 | 1000 | 0.2252 | 0.2901 |
| 0.1979 | 1.99 | 1100 | 0.2215 | 0.2836 |
| 0.166 | 2.17 | 1200 | 0.2217 | 0.2811 |
| 0.1623 | 2.35 | 1300 | 0.2200 | 0.2685 |
| 0.1628 | 2.53 | 1400 | 0.2166 | 0.2707 |
| 0.1593 | 2.71 | 1500 | 0.2131 | 0.2634 |
| 0.1561 | 2.89 | 1600 | 0.2121 | 0.2661 |
| 0.146 | 3.07 | 1700 | 0.2128 | 0.2552 |
| 0.1339 | 3.25 | 1800 | 0.2119 | 0.2591 |
| 0.1314 | 3.43 | 1900 | 0.2090 | 0.2492 |
| 0.1296 | 3.62 | 2000 | 0.2058 | 0.2504 |
| 0.1304 | 3.8 | 2100 | 0.2057 | 0.2500 |
| 0.1276 | 3.98 | 2200 | 0.2028 | 0.2463 |
| 0.116 | 4.16 | 2300 | 0.2058 | 0.2461 |
| 0.1122 | 4.34 | 2400 | 0.2074 | 0.2443 |
| 0.1087 | 4.52 | 2500 | 0.2065 | 0.2411 |
| 0.1087 | 4.7 | 2600 | 0.2042 | 0.2412 |
| 0.11 | 4.88 | 2700 | 0.2014 | 0.2396 |
Framework versions
- Transformers 4.30.2
- Pytorch 2.0.1+cu118
- Datasets 2.13.0
- Tokenizers 0.13.3