wav2vec2-1b-E50_speed
This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.5794
- Cer: 21.1466
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: 2
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 8
- total_train_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 50
- num_epochs: 5
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Cer |
|---|---|---|---|---|
| 22.602 | 0.2580 | 200 | 6.2336 | 100.0 |
| 4.8547 | 0.5160 | 400 | 4.5853 | 98.3318 |
| 4.6154 | 0.7741 | 600 | 4.6172 | 93.2331 |
| 4.4727 | 1.0321 | 800 | 4.3835 | 80.8506 |
| 4.2854 | 1.2901 | 1000 | 4.2819 | 77.9194 |
| 3.947 | 1.5481 | 1200 | 3.1503 | 61.7070 |
| 2.091 | 1.8062 | 1400 | 1.7032 | 40.4547 |
| 1.223 | 2.0642 | 1600 | 1.4472 | 39.8320 |
| 0.8867 | 2.3222 | 1800 | 1.2054 | 30.8506 |
| 0.7444 | 2.5802 | 2000 | 1.0326 | 31.4732 |
| 0.6529 | 2.8383 | 2200 | 0.8416 | 27.4671 |
| 0.5313 | 3.0963 | 2400 | 0.8239 | 25.0176 |
| 0.4148 | 3.3543 | 2600 | 0.7047 | 24.6417 |
| 0.3665 | 3.6123 | 2800 | 0.6761 | 22.5681 |
| 0.3435 | 3.8703 | 3000 | 0.6632 | 22.3567 |
| 0.2865 | 4.1284 | 3200 | 0.5879 | 21.1936 |
| 0.2307 | 4.3864 | 3400 | 0.6009 | 21.3875 |
| 0.2198 | 4.6444 | 3600 | 0.5810 | 21.2700 |
| 0.195 | 4.9024 | 3800 | 0.5794 | 21.1466 |
Framework versions
- Transformers 4.45.2
- Pytorch 2.3.1.post100
- Datasets 2.19.1
- Tokenizers 0.20.1