arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2611.06254cs.AI
收藏

HarnessFlow:基于蒙特卡洛树搜索(MCTS)的LLM智能体 harness 优化方法

HarnessFlow: MCTS-Guided Harness Optimization for LLM Agents

South China University of Technology(华南理工大学);Fudan University(复旦大学);Wuhan University(武汉大学);Cornell University(康奈尔大学);University of California, San Diego(加利福尼亚大学圣地亚哥分校);The Chinese University of Hong Kong, Shenzhen(香港中文大学(深圳));Microsoft Research Asia(微软亚洲研究院)

Zhi Hong, Qianlong Wang, Hongchen Wei, Aidan Wu, Daohong Ying, Mingze Kong, Bei Liu, Chong Luo

AI总结:

HarnessFlow 是基于 MCTS 的 LLM 智能体 harness 优化框架,在六个基准测试中优化后 harness 平均得分 69.48%,优于初始 harness 和 Meta-Harness,且执行成本更低。

摘要:

大型语言模型(LLM)智能体的性能既取决于底层模型,也取决于控制其与工具及环境交互的 harness。然而 harness 设计目前仍以手动为主,实现 harness 自动化需要在开放的搜索空间中进行成本高昂的端到端评估。本文提出 HarnessFlow,这是一种在候选评估预算有限、模型和任务环境固定的情况下优化完整可执行 harness 的框架。HarnessFlow 将候选 harness 及其评估历史组织在一个优化树中,使用基于蒙特卡洛树搜索(MCTS)的策略平衡利用与探索。编码智能体利用存档代码、执行反馈以及可重用 harness 机制库来优化选定的 harness。在每个基准测试中进行 15 次优化尝试后,HarnessFlow 在涵盖智能体任务、编码和网页交互的六个基准测试中,取得了对比方法中最高的平均测试分数。优化后的 harness 平均得分为 69.48%,分别比初始 harness 和 Meta-Harness 高出 10.89 和 3.95 个百分点,同时比手动设计的基线方法产生更低的平均单任务执行成本。这些结果表明,结构化 harness 搜索为自动化设计高效且能力强大的 LLM 智能体提供了一条有前景的路径。

英文摘要:

The performance of large language model (LLM) agents depends on both the underlying model and the harness that governs interaction with tools and the environment. Yet harness design remains largely manual, and automating it requires navigating an open-ended search space through costly end-to-end evaluations. We introduce HarnessFlow, a framework for optimizing complete executable harnesses under a limited candidate-evaluation budget, with the model and task environment held fixed. HarnessFlow organizes candidates and their evaluation histories in a refinement tree, using a Monte Carlo Tree Search (MCTS)-based policy to balance exploitation and exploration. A coding agent refines selected harnesses using archived code, execution feedback, and a library of reusable harness mechanisms. With 15 refinement attempts per benchmark, HarnessFlow achieves the highest mean test scores among the compared methods on six benchmarks spanning agentic tasks, coding, and web interaction. The optimized harnesses achieve an average score of 69.48%, outperforming the initial harnesses and Meta-Harness by 10.89 and 3.95 percentage points, respectively, while incurring lower average per-task execution costs than hand-engineered baselines. These results suggest that structured harness search offers a promising route toward automating the design of capable and efficient LLM agents.

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。

相关深度报道

↑