LongRAG

LongRAG

by Tsinghua University, CAS, ZhiPu
LongRAG is a dual-perspective robust retrieval-augmented generation (RAG) framework designed for long-context question answering (LCQA).

What is LongRAG?

LongRAG is a dual-perspective robust retrieval-augmented generation (RAG) framework developed by Tsinghua University, the Chinese Academy of Sciences (CAS), and ZhiPu. It is designed to address long-context question answering (LCQA) by integrating global context understanding and factual detail recognition.

Main Features of LongRAG

  • Dual-Perspective Information Processing: Combines global information and factual details for comprehensive answers.
  • Hybrid Retriever: Efficiently retrieves relevant information from large datasets.
  • LLM-Enhanced Information Extractor: Maps retrieved snippets back to the original text for context restoration.
  • CoT-Guided Filter: Uses Chain of Thought (CoT) to focus on relevant information.
  • LLM-Enhanced Generator: Generates final answers by combining global and detailed information.
  • Automated Fine-Tuning Data Construction: Enhances model performance through automated dataset creation.

Technical Principles

  • Retrieval-Augmented Generation (RAG): Retrieves external knowledge to assist in answer generation.
  • Global and Detailed Information Integration: Balances local factual details with global context.
  • Mapping Strategy: Restores contextual information by mapping snippets to the original text.
  • Chain of Thought (CoT): Guides the model to focus on relevant knowledge.
  • Filtering Strategy: Retains key factual details while filtering out irrelevant information.

Application Scenarios

  • Customer Service and Support: Answers lengthy customer queries accurately.
  • Medical Consultation: Processes patient records and medical literature for complex questions.
  • Legal Consultation: Analyzes legal documents and cases for in-depth advice.
  • Education and Research: Assists in understanding academic articles and research reports.
  • Corporate Decision Support: Provides insights from market research and corporate reports.

Getting Started

Framework Features

Supported Tasks
Long-Context Question Answering Information Retrieval Text Generation Fine-Tuning
Tags
Retrieval-Augmented Generation Long-Context Question Answering LLM Natural Language Processing AI Research Hybrid Retriever Chain of Thought Information Extraction Automated Fine-Tuning Global Context Understanding

Getting Started

Pricing
free

Screenshots & Images

Primary Screenshot
Additional Images
View Repository

Stats

0 Views
0 Favorites
101 GitHub Stars

Community & Support

GitHub Repository

Similar Frameworks

TPO
0
Phantom by ByteDance
0
AgentSociety by Tsinghua University
0

Recently Viewed

AgentQ Framework