CeLLaTe-tapt-pubmedbert-tokenizer-adapted-baseline
This model is a fine-tuned version of microsoft/BiomedNLP-BiomedBERT-base-uncased-abstract-fulltext on the Mardiyyah/TAPT_CeLLaTe2.0 dataset. It achieves the following results on the evaluation set:
- Loss: 1.1076
- Accuracy: 0.7643
- Perplexity: 3.0272
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: 3e-05
- train_batch_size: 32
- eval_batch_size: 32
- seed: 3407
- gradient_accumulation_steps: 2
- total_train_batch_size: 64
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-06 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.06
- num_epochs: 100
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy | Perplexity |
|---|---|---|---|---|---|
| 1.3684 | 1.0 | 14 | 1.2141 | 0.7641 | 3.3674 |
| 1.3406 | 2.0 | 28 | 1.1774 | 0.7648 | 3.2459 |
| 1.3246 | 3.0 | 42 | 1.1310 | 0.7689 | 3.0988 |
| 1.2929 | 4.0 | 56 | 1.1271 | 0.7675 | 3.0866 |
| 1.2489 | 5.0 | 70 | 1.1542 | 0.7641 | 3.1716 |
| 1.2216 | 6.0 | 84 | 1.0947 | 0.7733 | 2.9882 |
| 1.1987 | 7.0 | 98 | 1.0540 | 0.7750 | 2.8691 |
| 1.2006 | 8.0 | 112 | 1.0942 | 0.7706 | 2.9868 |
| 1.1659 | 9.0 | 126 | 1.0899 | 0.7701 | 2.9741 |
| 1.1546 | 10.0 | 140 | 1.0596 | 0.7757 | 2.8851 |
| 1.1216 | 11.0 | 154 | 1.0536 | 0.7735 | 2.8680 |
| 1.1113 | 12.0 | 168 | 1.0336 | 0.7779 | 2.8111 |
| 1.0872 | 13.0 | 182 | 1.0709 | 0.7716 | 2.9179 |
| 1.1014 | 14.0 | 196 | 1.0786 | 0.7746 | 2.9406 |
| 1.0882 | 15.0 | 210 | 1.0788 | 0.7727 | 2.9413 |
| 1.0683 | 16.0 | 224 | 1.0398 | 0.7796 | 2.8287 |
| 1.0612 | 17.0 | 238 | 1.0620 | 0.7712 | 2.8920 |
Framework versions
- Transformers 4.48.2
- Pytorch 2.4.1+cu121
- Datasets 3.0.2
- Tokenizers 0.21.0