ReMaKE

This repository contains the data and codes for our paper "Retrieval-augmented Multilingual Knowledge Editing".

1. Data & Model

MzsRE is located in ./data/MzsRE/

Models are located in ./model/ You can download from google drive

2. Edit

python run_bizsre.py --editing_method=IKE --hparams_dir=./hparams/IKE/llama2-7b-16.yaml --data_dir=./data --metrics_save_dir ./results/llama2-7b/16shot/ --backbone llama2_7b-16shot_classifier --search classifier

3. Evaluate

python evaluate.py

Acknowledgement

• Our codes are based on Bi-ZsRE, and we thank their outstanding open-source contributions.
• Our data is based on vanilla ZsRE dataset (Levy et al., 2017), Bi-ZsRE dataset Wang et al., 2023 and the portability QA pairs collect by Yao et al. (2023).
  • Zero-Shot Relation Extraction via Reading Comprehension (CoNLL 2017)
  • Editing Large Language Models: Problems, Methods, and Opportunities (arXiv preprint 2023)
  • Cross-Lingual Knowledge Editing in Large Language Models (arXiv preprint 2023)

Citation

If you find this work is useful or use the data in your work, please consider cite our paper:

@inproceedings{DBLP:conf/acl/WangHB24,
  author       = {Weixuan Wang and
                  Barry Haddow and
                  Alexandra Birch},
  editor       = {Lun{-}Wei Ku and
                  Andre Martins and
                  Vivek Srikumar},
  title        = {Retrieval-Augmented Multilingual Knowledge Editing},
  booktitle    = {Proceedings of the 62nd Annual Meeting of the Association for Computational
                  Linguistics (Volume 1: Long Papers), {ACL} 2024, Bangkok, Thailand,
                  August 11-16, 2024},
  pages        = {335--354},
  publisher    = {Association for Computational Linguistics},
  year         = {2024},
  url          = {https://doi.org/10.18653/v1/2024.acl-long.21},
  doi          = {10.18653/V1/2024.ACL-LONG.21},
  timestamp    = {Sun, 19 Jan 2025 13:21:44 +0100},
  biburl       = {https://dblp.org/rec/conf/acl/WangHB24.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}