<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->
bert-finetuned-ner
This model is a fine-tuned version of bert-base-cased on the bionlp2004 dataset. It achieves the following results on the evaluation set:
- Loss: 0.2098
- Precision: 0.7522
- Recall: 0.8140
- F1: 0.7819
- Accuracy: 0.9379
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: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3
Training results
Training Loss | Epoch | Step | Validation Loss | Precision | Recall | F1 | Accuracy |
---|---|---|---|---|---|---|---|
0.2255 | 1.0 | 2078 | 0.2073 | 0.7080 | 0.7877 | 0.7457 | 0.9305 |
0.1709 | 2.0 | 4156 | 0.1995 | 0.7479 | 0.8106 | 0.7780 | 0.9364 |
0.1324 | 3.0 | 6234 | 0.2098 | 0.7522 | 0.8140 | 0.7819 | 0.9379 |
Framework versions
- Transformers 4.25.1
- Pytorch 1.13.1+cu116
- Datasets 2.8.0
- Tokenizers 0.13.2