arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2405.05955cs.CL
收藏

Smurfs:采用上下文高效型DFSDT进行工具规划的多智能体系统

Smurfs: Multi-Agent System using Context-Efficient DFSDT for Tool Planning

The Chinese University of Hong Kong, Shenzhen(香港中文大学(深圳))

Junzhi Chen, Juhao Liang, Benyou Wang

AI总结:

针对现有工具使用框架的缺陷,提出模块化、上下文高效且免训练的多智能体系统Smurfs,在两类基准任务上表现更优,大幅降低token消耗,让小模型性能追平大模型的DFSDT方案。

摘要:

让大语言模型(LLM)学会使用工具解决复杂问题,可使其具备类人推理能力。ReAct及其变体是单智能体和多智能体系统中主流的工具使用框架。为解决ReAct存在的误差传播、探索受限等问题,深度优先搜索决策树(DFSDT)被提出,但它在单智能体设置下仍面临回滚不稳定、上下文冗余、过早终止等挑战。我们提出了名为“Smurfs”的新型多智能体系统(MAS),该系统采用模块化、上下文高效且无需训练的设计,对DFSDT进行增强。在开放式任务基准StableToolBench和封闭式任务基准HotpotQA上,Smurfs的表现均优于基线方法,与DFSDT相比token使用量减少了60.9%,且能让Mistral-7b的性能达到与GPT-4-DFSDT相当的水平。大量消融实验验证了Smurfs核心组件的有效性,为多智能体系统的构建与解读提供了宝贵思路,也为未来的探索铺平了道路。

英文摘要:

Teaching large language models (LLMs) to use tools for solving complex problems can grant them human-like reasoning abilities. ReAct and its variants are popular frameworks for tool use in both single-agent and multi-agent systems. To address issues like error propagation and limited exploration in ReAct, the Deep First Search Decision Tree (DFSDT) was proposed, but it faces challenges such as rollback instability, redundant context, and premature termination in single-agent settings. We introduce "Smurfs," a novel multi-agent system (MAS) that enhances DFSDT with a modular, context-efficient, and training-free design. Smurfs surpasses baseline methods in both the open-ended StableToolBench and the closed-ended HotpotQA tasks, reducing token usage by 60.9\% compared to DFSDT and enabling Mistral-7b to perform on par with GPT-4-DFSDT. Extensive ablation studies confirm the effectiveness of Smurfs' core components, offering valuable insights for the construction and interpretation of MAS, and paving the way for future exploration.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑