CeLLaTe-ner-3class-pubmedbert-baseline
This model is a fine-tuned version of microsoft/BiomedNLP-BiomedBERT-base-uncased-abstract-fulltext on the OTAR3088/CeLLaTe-ner-3class-iob_final dataset. It achieves the following results on the evaluation set:
- Loss: 0.0923
- Precision: 0.7695
- Recall: 0.7647
- Micro F1: 0.7671
- Weighted F1: 0.7667
- Macro F1: 0.7626
- Accuracy: 0.9829
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: 2e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 3407
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.01
- num_epochs: 20
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Precision | Recall | Micro F1 | Weighted F1 | Macro F1 | Accuracy |
|---|---|---|---|---|---|---|---|---|---|
| 0.3707 | 1.0 | 263 | 0.0799 | 0.6219 | 0.5815 | 0.6010 | 0.5923 | 0.5770 | 0.9770 |
| 0.0585 | 2.0 | 526 | 0.0618 | 0.7373 | 0.7246 | 0.7309 | 0.7313 | 0.7277 | 0.9815 |
| 0.0354 | 3.0 | 789 | 0.0642 | 0.7269 | 0.7647 | 0.7453 | 0.7451 | 0.7433 | 0.9818 |
| 0.0257 | 4.0 | 1052 | 0.0811 | 0.7938 | 0.7054 | 0.7470 | 0.7457 | 0.7406 | 0.9823 |
| 0.0192 | 5.0 | 1315 | 0.0726 | 0.7556 | 0.7324 | 0.7439 | 0.7433 | 0.7413 | 0.9821 |
| 0.0145 | 6.0 | 1578 | 0.0841 | 0.7036 | 0.7574 | 0.7295 | 0.7315 | 0.7282 | 0.9808 |
| 0.0108 | 7.0 | 1841 | 0.0896 | 0.7809 | 0.7330 | 0.7562 | 0.7536 | 0.7491 | 0.9824 |
| 0.0086 | 8.0 | 2104 | 0.0914 | 0.7442 | 0.7574 | 0.7508 | 0.7510 | 0.7500 | 0.9823 |
| 0.0071 | 9.0 | 2367 | 0.0928 | 0.7695 | 0.7647 | 0.7671 | 0.7667 | 0.7626 | 0.9829 |
| 0.0061 | 10.0 | 2630 | 0.1001 | 0.7536 | 0.7402 | 0.7468 | 0.7469 | 0.7463 | 0.9822 |
| 0.0053 | 11.0 | 2893 | 0.0949 | 0.7722 | 0.7517 | 0.7618 | 0.7611 | 0.7580 | 0.9828 |
| 0.0042 | 12.0 | 3156 | 0.1045 | 0.7545 | 0.7423 | 0.7484 | 0.7490 | 0.7484 | 0.9823 |
| 0.0034 | 13.0 | 3419 | 0.1130 | 0.7660 | 0.7548 | 0.7604 | 0.7597 | 0.7571 | 0.9827 |
| 0.0032 | 14.0 | 3682 | 0.1128 | 0.7572 | 0.7371 | 0.7470 | 0.7472 | 0.7465 | 0.9819 |
Framework versions
- Transformers 4.48.2
- Pytorch 2.4.1+cu121
- Datasets 3.0.2
- Tokenizers 0.21.0