RankGPT
Basic Information
RankGPT is the research codebase accompanying the paper that investigates using large generative language models as re-ranking agents for information retrieval. It provides implementations to generate instructional permutations with LLMs, apply those permutations to re-rank candidate passages, and evaluate re-ranking performance on standard IR benchmarks. The repository includes example code showing how to call the permutation pipeline with models such as ChatGPT, a sliding window strategy to handle long candidate lists beyond token limits, and scripts to run end-to-end retrieval and evaluation using pyserini and trec_eval. It also contains data releases and precomputed ChatGPT permutations for MS MARCO, procedures to distill LLM outputs into compact supervised rankers, and utilities for reproducing the experiments reported in the paper.
Links
Categorization
App Details
-
$$$$ -
$$$$