Smurfs: Multi-Agent System using Context-Efficient DFSDT for Tool Planning
This addresses the challenge of efficient tool planning in multi-agent systems for AI researchers, though it appears incremental as it builds on existing DFSDT methods.
The paper tackles the problem of error propagation and limited exploration in tool-use frameworks for LLMs by introducing Smurfs, a multi-agent system that enhances DFSDT, achieving a 60.9% reduction in token usage and enabling Mistral-7b to match GPT-4-DFSDT performance.
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.