bert-finetuned-ner3
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: 0.0603
- Precision: 0.9296
- Recall: 0.9490
- F1: 0.9392
- Accuracy: 0.9863
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 |
---|---|---|---|---|---|---|---|
0.0855 | 1.0 | 1756 | 0.0673 | 0.9130 | 0.9340 | 0.9234 | 0.9827 |
0.0345 | 2.0 | 3512 | 0.0590 | 0.9284 | 0.9445 | 0.9363 | 0.9855 |
0.0229 | 3.0 | 5268 | 0.0603 | 0.9296 | 0.9490 | 0.9392 | 0.9863 |
Framework versions
- Transformers 4.18.0
- Pytorch 1.10.0+cu111
- Datasets 2.1.0
- Tokenizers 0.12.1
- Downloads last month
- 2
Inference Providers
NEW
This model is not currently available via any of the supported Inference Providers.
Dataset used to train Ghost1/bert-finetuned-ner3
Evaluation results
- Precision on conll2003self-reported0.930
- Recall on conll2003self-reported0.949
- F1 on conll2003self-reported0.939
- Accuracy on conll2003self-reported0.986