bert-mini-sst2-distilled
This model is a fine-tuned version of google/bert_uncased_L-4_H-256_A-4 on the glue dataset. It achieves the following results on the evaluation set:
- Loss: 1.1792
- Accuracy: 0.8567
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: 0.00021185586235152412
- train_batch_size: 1024
- eval_batch_size: 1024
- seed: 33
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 8
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|---|---|---|---|---|
| 2.1552 | 1.0 | 66 | 1.4847 | 0.8349 |
| 0.8451 | 2.0 | 132 | 1.3495 | 0.8624 |
| 0.5864 | 3.0 | 198 | 1.2257 | 0.8532 |
| 0.4553 | 4.0 | 264 | 1.2571 | 0.8544 |
| 0.3708 | 5.0 | 330 | 1.2132 | 0.8658 |
| 0.3086 | 6.0 | 396 | 1.2370 | 0.8589 |
| 0.2701 | 7.0 | 462 | 1.1900 | 0.8635 |
| 0.246 | 8.0 | 528 | 1.1792 | 0.8567 |
Framework versions
- Transformers 4.12.3
- Pytorch 1.9.1
- Datasets 1.15.1
- Tokenizers 0.10.3