Introduction

Code for the paper: Modular Retrieval for Generalization and Interpretation [link]. REMOP (Retrieval with modular prompt tuning) is a simple-yet-effective implementation of modular retrieval, and the code is mainly develped based on DPTDR, coCondenser and P-tuning v2.

Installation

For environment, please run sh install_env.sh in a clean conda environment of python>=3.7. Then just run pip install -e.

Reproduction for BEIR

Please refer to examples/condener_beir.

References

@misc{liang2023modular,
      title={Modular Retrieval for Generalization and Interpretation}, 
      author={Juhao Liang and Chen Zhang and Zhengyang Tang and Jie Fu and Dawei Song and Benyou Wang},
      year={2023},
      eprint={2303.13419},
      archivePrefix={arXiv},
      primaryClass={cs.IR}
}