w2v-bert-2.0-600m-turkish-colab
This model is a fine-tuned version of ylacombe/w2v-bert-2.0 on the common_voice_16_0 dataset. It achieves the following results on the evaluation set:
- Loss: 0.1441
- Wer: 0.1373
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: 16
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 1000
- num_epochs: 5
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.252 | 0.29 | 400 | 0.3121 | 0.3150 |
| 0.2541 | 0.58 | 800 | 0.3786 | 0.3441 |
| 0.2505 | 0.88 | 1200 | 0.4106 | 0.3766 |
| 0.1958 | 1.17 | 1600 | 0.2974 | 0.2877 |
| 0.1686 | 1.46 | 2000 | 0.2854 | 0.2736 |
| 0.1498 | 1.75 | 2400 | 0.2508 | 0.2486 |
| 0.1343 | 2.05 | 2800 | 0.2315 | 0.2263 |
| 0.1045 | 2.34 | 3200 | 0.2207 | 0.2243 |
| 0.0983 | 2.63 | 3600 | 0.2109 | 0.2046 |
| 0.089 | 2.92 | 4000 | 0.1970 | 0.1896 |
| 0.0726 | 3.21 | 4400 | 0.1963 | 0.1799 |
| 0.0552 | 3.51 | 4800 | 0.1879 | 0.1778 |
| 0.0573 | 3.8 | 5200 | 0.1821 | 0.1693 |
| 0.0421 | 4.09 | 5600 | 0.1602 | 0.1517 |
| 0.0363 | 4.38 | 6000 | 0.1564 | 0.1485 |
| 0.0345 | 4.67 | 6400 | 0.1466 | 0.1437 |
| 0.0294 | 4.97 | 6800 | 0.1441 | 0.1373 |
Framework versions
- Transformers 4.37.0.dev0
- Pytorch 2.1.0+cu121
- Datasets 2.16.1
- Tokenizers 0.15.0