CeLLaTe-tapt-pubmedbert-tokenizer-original-wwmask
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: 1.0329
- Accuracy: 0.7709
- Perplexity: 2.8091
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 |
|---|---|---|---|---|---|
| 1.2066 | 1.0 | 14 | 1.0552 | 0.7680 | 2.8725 |
| 1.2022 | 2.0 | 28 | 1.0389 | 0.7703 | 2.8262 |
| 1.1634 | 3.0 | 42 | 1.0517 | 0.7694 | 2.8624 |
| 1.1435 | 4.0 | 56 | 1.0598 | 0.7664 | 2.8858 |
| 1.1674 | 5.0 | 70 | 0.9973 | 0.7784 | 2.7109 |
| 1.1354 | 6.0 | 84 | 0.9737 | 0.7797 | 2.6478 |
| 1.1377 | 7.0 | 98 | 1.0435 | 0.7709 | 2.8391 |
| 1.1163 | 8.0 | 112 | 1.0534 | 0.7680 | 2.8675 |
| 1.1126 | 9.0 | 126 | 1.0454 | 0.7648 | 2.8445 |
| 1.0904 | 10.0 | 140 | 1.0425 | 0.7715 | 2.8363 |
| 1.0807 | 11.0 | 154 | 1.0148 | 0.7702 | 2.7587 |
Framework versions
- Transformers 4.48.2
- Pytorch 2.4.1+cu121
- Datasets 3.0.2
- Tokenizers 0.21.0