distilbert-base-uncased-finetuned-ner
This model is a fine-tuned version of distilbert-base-uncased on the conll2003 dataset. It achieves the following results on the evaluation set:
- Loss: 0.0629
- Precision: 0.9265
- Recall: 0.9357
- F1: 0.9310
- Accuracy: 0.9835
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 |
---|---|---|---|---|---|---|---|
0.2396 | 1.0 | 878 | 0.0706 | 0.9172 | 0.9186 | 0.9179 | 0.9810 |
0.0539 | 2.0 | 1756 | 0.0627 | 0.9264 | 0.9334 | 0.9299 | 0.9831 |
0.03 | 3.0 | 2634 | 0.0629 | 0.9265 | 0.9357 | 0.9310 | 0.9835 |
Framework versions
- Transformers 4.25.1
- Pytorch 1.13.0+cu116
- Datasets 2.8.0
- Tokenizers 0.13.2
- Downloads last month
- 117
Inference Providers
NEW
This model is not currently available via any of the supported Inference Providers.
Dataset used to train AlonCohen/distilbert-base-uncased-finetuned-ner
Evaluation results
- Precision on conll2003self-reported0.926
- Recall on conll2003self-reported0.936
- F1 on conll2003self-reported0.931
- Accuracy on conll2003self-reported0.984