UniNER-7B-all

Description: This model is the best UniNER model. It is trained on the combinations of three data splits: (1) ChatGPT-generated Pile-NER-type data, (2) ChatGPT-generated Pile-NER-definition data, and (3) 40 supervised datasets in the Universal NER benchmark (see Fig. 4 in paper), where we randomly sample up to 10K instances from the train split of each dataset. Note that CrossNER and MIT datasets are excluded from training for OOD evaluation.

Check our paper for more information. Check our repo about how to use the model.

Inference

The template for inference instances is as follows:

Note: Inferences are based on one entity type at a time. For multiple entity types, create separate instances for each type.

License

This model and its associated data are released under the CC BY-NC 4.0 license. They are primarily used for research purposes.

Citation

@article{zhou2023universalner,
      title={UniversalNER: Targeted Distillation from Large Language Models for Open Named Entity Recognition}, 
      author={Wenxuan Zhou and Sheng Zhang and Yu Gu and Muhao Chen and Hoifung Poon},
      year={2023},
      eprint={2308.03279},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}