bert-base-uncased-ner-conll2003
This model is a fine-tuned version of bert-base-cased on the conll2003 dataset. It achieves the following results on the evaluation set:
- Loss: 0.0602
- Precision: 0.9342
- Recall: 0.9536
- F1: 0.9438
- Accuracy: 0.9870
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 |
|---|---|---|---|---|---|---|---|
| 0.0871 | 1.0 | 1756 | 0.0728 | 0.9138 | 0.9275 | 0.9206 | 0.9811 |
| 0.0331 | 2.0 | 3512 | 0.0591 | 0.9311 | 0.9514 | 0.9411 | 0.9866 |
| 0.0173 | 3.0 | 5268 | 0.0602 | 0.9342 | 0.9536 | 0.9438 | 0.9870 |
Framework versions
- Transformers 4.21.3
- Pytorch 1.12.1+cu113
- Datasets 2.4.0
- Tokenizers 0.12.1