bert-finetuned-ner
This model is a fine-tuned version of bert-base-uncased on the conll2003 dataset. It achieves the following results on the evaluation set:
- Loss: 0.0557
- Precision: 0.9390
- Recall: 0.9485
- F1: 0.9437
- Accuracy: 0.9867
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: 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 |
---|---|---|---|---|---|---|---|
0.2174 | 1.0 | 878 | 0.0619 | 0.9296 | 0.9349 | 0.9322 | 0.9841 |
0.0498 | 2.0 | 1756 | 0.0550 | 0.9337 | 0.9442 | 0.9389 | 0.9861 |
0.0257 | 3.0 | 2634 | 0.0557 | 0.9390 | 0.9485 | 0.9437 | 0.9867 |
Framework versions
- Transformers 4.24.0
- Pytorch 1.12.1+cu113
- Datasets 2.7.1
- Tokenizers 0.13.2
- Downloads last month
- 11
Dataset used to train Z3rOs2/bert-finetuned-ner
Evaluation results
- Precision on conll2003self-reported0.939
- Recall on conll2003self-reported0.949
- F1 on conll2003self-reported0.944
- Accuracy on conll2003self-reported0.987