LongAgent: Scaling Language Models to 128k Context through Multi-Agent Collaboration

TL;DR

LongAgent introduces a multi-agent collaboration framework that scales language models to handle 128k token contexts, improving long-text processing tasks like retrieval and multi-hop QA over existing models such as GPT-4.

Contribution

The paper presents LongAgent, a novel multi-agent approach that enables existing LLMs to process much longer contexts effectively, addressing hallucination issues through inter-agent communication.

Findings

LongAgent with LLaMA-7B outperforms GPT-4 in long-text retrieval.

The approach effectively reduces hallucinations via inter-member communication.

Significant improvements in multi-hop question answering tasks.

Abstract

Large language models (LLMs) have demonstrated impressive performance in understanding language and executing complex reasoning tasks. However, LLMs with long context windows have been notorious for their expensive training costs and high inference latency. Even the most advanced models such as GPT-4 and Claude2 often make mistakes when processing inputs of over $100k$ tokens, a phenomenon also known as \textit{lost in the middle}. In this paper, we propose \textsc{LongAgent}, a method based on multi-agent collaboration, which scales LLMs (e.g., LLaMA) to a context of 128K and demonstrates potential superiority in long-text processing compared to GPT-4. In \textsc{LongAgent}, a leader is responsible for understanding user intent and directing team members to acquire information from documents. Due to members' hallucinations, it is non-trivial for a leader to obtain accurate information from the responses of dozens to hundreds of members. To address this, we develop an \textit{inter-member communication} mechanism to resolve response conflicts caused by hallucinations through information sharing. Our experimental results indicate that \textsc{LongAgent} offers a promising alternative for long-text processing. The agent team instantiated with LLaMA-7B achieves significant improvements in tasks such as 128k-long text retrieval, multi-hop question answering, compared to GPT-4.