wav2vec2-1b-Yspeed
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: 0.9193
- Cer: 22.7561
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 |
|---|---|---|---|---|
| 14.1833 | 0.2580 | 200 | 4.5491 | 97.4389 |
| 2.5916 | 0.5160 | 400 | 2.0400 | 48.7077 |
| 1.351 | 0.7741 | 600 | 1.9023 | 46.5695 |
| 1.0665 | 1.0321 | 800 | 1.2533 | 34.7039 |
| 0.8141 | 1.2901 | 1000 | 1.3127 | 32.2016 |
| 0.7672 | 1.5481 | 1200 | 1.3379 | 33.9814 |
| 0.6847 | 1.8062 | 1400 | 1.2278 | 32.0489 |
| 0.5985 | 2.0642 | 1600 | 1.1310 | 28.4657 |
| 0.5208 | 2.3222 | 1800 | 1.2212 | 29.4995 |
| 0.4664 | 2.5802 | 2000 | 1.0708 | 26.2512 |
| 0.4379 | 2.8383 | 2200 | 1.0705 | 26.3804 |
| 0.4332 | 3.0963 | 2400 | 1.0949 | 27.4260 |
| 0.3778 | 3.3543 | 2600 | 0.9809 | 25.3936 |
| 0.3273 | 3.6123 | 2800 | 1.0045 | 24.5418 |
| 0.3117 | 3.8703 | 3000 | 0.9171 | 23.0146 |
| 0.2586 | 4.1284 | 3200 | 0.9051 | 22.5329 |
| 0.2273 | 4.3864 | 3400 | 0.8966 | 22.3508 |
| 0.2169 | 4.6444 | 3600 | 0.9430 | 23.2202 |
| 0.2159 | 4.9024 | 3800 | 0.9193 | 22.7561 |
Framework versions
- Transformers 4.45.2
- Pytorch 2.3.1.post100
- Datasets 2.19.1
- Tokenizers 0.20.1