GNITOAHC / YuLan-IR

YuLan-IR: Information Retrieval Boosted LMs

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

YuLan-IR

YuLan-IR is part of YuLan, an open source LLM initiative proposed by Gaoling School of Artificial Intelligence, Renmin University of China.

Overview

In this repository, we hope to explore the combination between Information Retrieval (IR) and generative language models, including but not limited to:

  • Augmenting generation with information retrieval to alleviate the hallucination of language models.
  • Improving information retrieval with language models. For example, using language models to help IR models determine whether a document / passage is useful.
  • etc.

Current Progress

RETA-LLM

RETA-LLM is a RETreival-Augmented LLM toolkit to support research in retrieval-augmented generation and to help users build their own in-down LLM-based systems. Find more about it in the repo.

WebBrain

WebBrain is a new benchmark for retrieval-augmented generation. We collect both the first passage of Wikipedia and the corresponding references. Given a user query, retrieval models are required to return relevant references, while generation methods should generate a response with the help of the references. The whole pipeline simulates the working process of New Bing, while all data are accessible. More details can be found in the repo.

About

YuLan-IR: Information Retrieval Boosted LMs

License:MIT License


Languages

Language:Python 99.5%Language:Shell 0.5%