wav2vec2-bert-mas-ex
This model is a fine-tuned version of facebook/w2v-bert-2.0 on the common_voice_16_0 dataset. It achieves the following results on the evaluation set:
- Loss: 0.7763
- Wer: 0.6301
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: 5e-05
- train_batch_size: 2
- 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: 300
- num_epochs: 5
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 2.424 | 0.12 | 300 | 1.3270 | 0.8863 |
| 1.2288 | 0.23 | 600 | 1.1525 | 0.8299 |
| 1.0443 | 0.35 | 900 | 0.9812 | 0.7729 |
| 1.0082 | 0.46 | 1200 | 0.9045 | 0.6852 |
| 0.8698 | 0.58 | 1500 | 0.9797 | 0.7063 |
| 0.8649 | 0.69 | 1800 | 0.9071 | 0.6724 |
| 0.8268 | 0.81 | 2100 | 0.8387 | 0.6716 |
| 0.8428 | 0.93 | 2400 | 0.8392 | 0.6623 |
| 0.6933 | 1.04 | 2700 | 0.7124 | 0.5966 |
| 0.6618 | 1.16 | 3000 | 0.7056 | 0.5688 |
| 0.6578 | 1.27 | 3300 | 0.7003 | 0.5708 |
| 0.6331 | 1.39 | 3600 | 0.6798 | 0.5578 |
| 0.5873 | 1.5 | 3900 | 0.6993 | 0.5453 |
| 0.6076 | 1.62 | 4200 | 0.6562 | 0.5268 |
| 0.5359 | 1.74 | 4500 | 0.6837 | 0.5735 |
| 0.6807 | 1.85 | 4800 | 0.6495 | 0.5272 |
| 0.5945 | 1.97 | 5100 | 0.6434 | 0.5058 |
| 0.5059 | 2.08 | 5400 | 0.6237 | 0.4855 |
| 0.5244 | 2.2 | 5700 | 0.6334 | 0.4749 |
| 0.5052 | 2.31 | 6000 | 0.6831 | 0.4976 |
| 0.5249 | 2.43 | 6300 | 0.6339 | 0.4919 |
| 0.5537 | 2.55 | 6600 | 0.6541 | 0.4990 |
| 0.6387 | 2.66 | 6900 | 0.8375 | 0.5829 |
| 0.669 | 2.78 | 7200 | 0.9152 | 0.6289 |
| 0.8881 | 2.89 | 7500 | 0.7704 | 0.6191 |
| 1.184 | 3.01 | 7800 | 0.8139 | 0.6866 |
| 1.0933 | 3.12 | 8100 | 0.7721 | 0.6518 |
| 1.3588 | 3.24 | 8400 | 0.7368 | 0.6152 |
| 1.4604 | 3.36 | 8700 | 0.7376 | 0.6158 |
| 1.2902 | 3.47 | 9000 | 0.7451 | 0.6188 |
| 1.3137 | 3.59 | 9300 | 0.7493 | 0.6194 |
| 1.3009 | 3.7 | 9600 | 0.7454 | 0.6164 |
| 1.3757 | 3.82 | 9900 | 0.7515 | 0.6289 |
| 1.2412 | 3.93 | 10200 | 0.7629 | 0.6237 |
| 1.2835 | 4.05 | 10500 | 0.7760 | 0.6351 |
| 1.3803 | 4.17 | 10800 | 0.7718 | 0.6273 |
| 1.325 | 4.28 | 11100 | 0.7763 | 0.6301 |
| 1.3798 | 4.4 | 11400 | 0.7763 | 0.6301 |
| 1.3421 | 4.51 | 11700 | 0.7763 | 0.6301 |
| 1.2834 | 4.63 | 12000 | 0.7763 | 0.6301 |
| 1.4757 | 4.74 | 12300 | 0.7763 | 0.6301 |
| 1.4171 | 4.86 | 12600 | 0.7763 | 0.6301 |
| 1.2838 | 4.97 | 12900 | 0.7763 | 0.6301 |
Framework versions
- Transformers 4.38.2
- Pytorch 2.2.1+cu121
- Datasets 2.15.0
- Tokenizers 0.15.2