GPT-RE-FT

Code for reproducing the GPT-RE with fine-tuned RE representations, EMNLP 2023 paper "GPT-RE: In-context Learning for Relation Extraction using Large Language Models". For better code readability, this repository includes only the GPT-RE-FT implementation; please refer to GPT-RE for the remaining methods.

GPT-RE-FT
 |-- GPT-RE (Retrieval-Based In-Context Learning)
 |-- REbaseline (Training Representation Models)

Requirements

• torch >= 1.8.1
• transformers >= 3.4.0
• wandb
• ujson
• tqdm
• faiss
• openai

Dataset

Please refer to GPT-RE or PURE for preparing the datasets. The expected structure of files is:

RE_improved_baseline
 |-- data
 |    |-- semeval
 |    |    |-- train.json        
 |    |    |-- dev.json
 |    |    |-- test.json
 |    |-- scierc
 |    |    |-- train.json        
 |    |    |-- dev.json
 |    |    |-- test.json

Training The RE Representation Model (REbaseline)

To train the representation models, we adopt an efficient implementation of entity-marker based method "An Improved Baseline for Sentence-level Relation Extractions". The commands and hyper-parameters for running experiments can be found in the scripts folder.

>> sh run_bert_ace.sh
>> sh run_bert_scierc.sh
>> sh run_bert_semeval.sh

Checkpoints

Dataset	Model	Download
Semeval	bert-base-uncased	link
SciERC	scibert_scivocab_uncased	link
ACE	bert-base-uncased	link

Retrieval-Based In-Context Learning (GPT-RE)

>> sh run_relation.sh

Please refer to GPT-RE for detailed description of augments. An example for knn-retrieved demonstrations could be found in results/knn-semeval.

Citation

@inproceedings{wan2023gpt,
  title={GPT-RE: In-context Learning for Relation Extraction using Large Language Models},
  author={Wan, Zhen and Cheng, Fei and Mao, Zhuoyuan and Liu, Qianying and Song, Haiyue and Li, Jiwei and Kurohashi, Sadao},
  booktitle={Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing},
  year={2023}
}

@inproceedings{han-etal-2025-amr,
    title = "{AMR}-{RE}: {A}bstract {M}eaning {R}epresentations for Retrieval-Based In-Context Learning in Relation Extraction",
    author = "Han, Peitao  and
      Pereira, Lis  and
      Cheng, Fei  and
      She, Wan Jou  and
      Aramaki, Eiji",
    booktitle = "Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 4: Student Research Workshop)",
    year = "2025",
}

Acknowledgements

Our code is based on GPT-RE and RE-improved-baseline