Skip to content

LongRAG

A dual-perspective retrieval system for long-context QA

Visit official site

What LongRAG is

The system is described as a dual-perspective and robust retrieval-augmented generation paradigm built on large language models for understanding complex long texts.

What you can do with it

  • Read a long text from global and detail perspectives
  • Combine retrieval with generation for long-context QA
  • Work through multi-step reasoning questions
  • Adjust parameters for different QA tasks and datasets

Getting started

  • Install the dependencies from the requirements file
  • Download and standardise the training and evaluation datasets
  • Run the instruction and index generation scripts
  • Place the built data into the LLaMA-Factory data directory

Pros & cons

✓ What we like

  • The model code is open source and free to modify
  • Retrieval and generation are combined for long-context QA

! What to watch out for

  • Data preparation requires downloading and standardising training datasets
  • Training uses LLaMA-Factory with an edited dataset configuration file

FAQ

What kind of QA does it target?

The material describes long-context question answering with global information and factual details.

Which datasets are named?

The examples name HotpotQA, 2WikiMultiHopQA and MusiQue.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review