bert-large-uncased_ner_conll2003
This model is a fine-tuned version of bert-large-uncased on the conll2003 dataset. It achieves the following results on the evaluation set:
- Loss: 0.0516
- Precision: 0.9424
- Recall: 0.9530
- F1: 0.9477
- Accuracy: 0.9898
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: cosine
- num_epochs: 5
Training results
| Training Loss | Epoch | Step | Validation Loss | Precision | Recall | F1 | Accuracy |
|---|---|---|---|---|---|---|---|
| 0.1605 | 1.0 | 878 | 0.0533 | 0.9252 | 0.9329 | 0.9290 | 0.9864 |
| 0.032 | 2.0 | 1756 | 0.0433 | 0.9320 | 0.9475 | 0.9397 | 0.9887 |
| 0.0125 | 3.0 | 2634 | 0.0454 | 0.9424 | 0.9524 | 0.9474 | 0.9897 |
| 0.006 | 4.0 | 3512 | 0.0507 | 0.9417 | 0.9519 | 0.9468 | 0.9896 |
| 0.0036 | 5.0 | 4390 | 0.0516 | 0.9424 | 0.9530 | 0.9477 | 0.9898 |
Framework versions
- Transformers 4.20.1
- Pytorch 1.11.0
- Datasets 2.1.0
- Tokenizers 0.12.1