CeLLaTe-tapt-pubmedbert-tokenizer-original_v2
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: 1.0433
- Accuracy: 0.7695
- Perplexity: 2.8385
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 |
|---|---|---|---|---|---|
| 1.209 | 1.0 | 14 | 1.0540 | 0.7678 | 2.8692 |
| 1.1983 | 2.0 | 28 | 1.0407 | 0.7700 | 2.8313 |
| 1.1651 | 3.0 | 42 | 1.0530 | 0.7689 | 2.8662 |
| 1.1403 | 4.0 | 56 | 1.0620 | 0.7659 | 2.8920 |
| 1.1665 | 5.0 | 70 | 1.0056 | 0.7774 | 2.7337 |
| 1.1431 | 6.0 | 84 | 0.9861 | 0.7798 | 2.6806 |
| 1.1417 | 7.0 | 98 | 1.0499 | 0.7707 | 2.8572 |
| 1.1194 | 8.0 | 112 | 1.0551 | 0.7675 | 2.8722 |
| 1.1158 | 9.0 | 126 | 1.0471 | 0.7644 | 2.8494 |
| 1.0882 | 10.0 | 140 | 1.0437 | 0.7717 | 2.8396 |
| 1.0809 | 11.0 | 154 | 1.0141 | 0.7710 | 2.7570 |
Framework versions
- Transformers 4.48.2
- Pytorch 2.4.1+cu121
- Datasets 3.0.2
- Tokenizers 0.21.0