bert-tiny-emotion-KD-BERT_and_distilBERT
This model is a fine-tuned version of google/bert_uncased_L-2_H-128_A-2 on the emotion dataset. It achieves the following results on the evaluation set:
- Loss: 0.8780
- Accuracy: 0.918
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: 16
- eval_batch_size: 16
- seed: 33
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 50
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|---|---|---|---|---|
| 7.1848 | 1.0 | 1000 | 4.7404 | 0.774 |
| 3.856 | 2.0 | 2000 | 2.7317 | 0.8685 |
| 2.3973 | 3.0 | 3000 | 1.8329 | 0.8895 |
| 1.5273 | 4.0 | 4000 | 1.2938 | 0.898 |
| 1.113 | 5.0 | 5000 | 1.1298 | 0.8985 |
| 0.9099 | 6.0 | 6000 | 1.0746 | 0.907 |
| 0.831 | 7.0 | 7000 | 1.0071 | 0.907 |
| 0.6813 | 8.0 | 8000 | 0.9556 | 0.9115 |
| 0.6432 | 9.0 | 9000 | 0.9746 | 0.913 |
| 0.5745 | 10.0 | 10000 | 0.8780 | 0.918 |
| 0.5319 | 11.0 | 11000 | 0.9410 | 0.909 |
| 0.4787 | 12.0 | 12000 | 0.9103 | 0.913 |
| 0.4529 | 13.0 | 13000 | 0.8829 | 0.915 |
Framework versions
- Transformers 4.22.1
- Pytorch 1.12.1+cu113
- Datasets 2.5.1
- Tokenizers 0.12.1