CeLLaTe-tapt-pubmedbert-tokenizer-adapted-spanmask
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: 2.3029
- Accuracy: 0.5761
- Perplexity: 10.0034
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 |
|---|---|---|---|---|---|
| 3.0089 | 1.0 | 14 | 2.6409 | 0.5507 | 14.0253 |
| 2.7347 | 2.0 | 28 | 2.4055 | 0.5698 | 11.0835 |
| 2.6256 | 3.0 | 42 | 2.4142 | 0.5723 | 11.1810 |
| 2.5855 | 4.0 | 56 | 2.3657 | 0.5720 | 10.6515 |
| 2.5716 | 5.0 | 70 | 2.3596 | 0.5713 | 10.5871 |
| 2.5212 | 6.0 | 84 | 2.3339 | 0.5698 | 10.3177 |
| 2.4673 | 7.0 | 98 | 2.3420 | 0.5751 | 10.4017 |
| 2.4827 | 8.0 | 112 | 2.2801 | 0.5824 | 9.7779 |
| 2.4762 | 9.0 | 126 | 2.2619 | 0.5839 | 9.6009 |
| 2.457 | 10.0 | 140 | 2.2566 | 0.5845 | 9.5509 |
| 2.4403 | 11.0 | 154 | 2.3328 | 0.5761 | 10.3072 |
| 2.3889 | 12.0 | 168 | 2.2309 | 0.5902 | 9.3085 |
| 2.3968 | 13.0 | 182 | 2.2488 | 0.5802 | 9.4764 |
| 2.3618 | 14.0 | 196 | 2.2365 | 0.5814 | 9.3607 |
| 2.3805 | 15.0 | 210 | 2.2719 | 0.5776 | 9.6975 |
| 2.348 | 16.0 | 224 | 2.3000 | 0.5791 | 9.9740 |
| 2.3436 | 17.0 | 238 | 2.2701 | 0.5824 | 9.6807 |
Framework versions
- Transformers 4.48.2
- Pytorch 2.4.1+cu121
- Datasets 3.0.2
- Tokenizers 0.21.0