Skip to content

Latest commit

 

History

History
29 lines (19 loc) · 722 Bytes

File metadata and controls

29 lines (19 loc) · 722 Bytes

Introduction

Code for the paper: Towards Dynamic Dense Retrieval with Routing Strategy.

Preprocessing

please refer to the data_clearning to obtain the training datasets.

Training

sh examples/condenser_beir/train_phase1.sh sh examples/condenser_beir/train_phase2.sh

Evaluation

sh examples/condenser_beir/eval.sh sh examples/condenser_beir/eval_remop.sh

Search

sh examples/condenser_beir/search.sh sh examples/condenser_beir/search_remop.sh

Hard Negative Generation

sh data_cleaning/hn_generation/passage_retrieve.sh

Installation

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