让代码定位工具自适应并行，减少冗余调用，提速93.6%。
Learning Adaptive Parallel Execution for Efficient Code Localization
- 通过动态调节搜索广度，实现质量与效率协同优化
- 在SWE-bench上达84.7%文件级F1，速度提升93.6%
- 适合需要高效代码定位的自动化开发场景
代码定位是自动化软件开发中的关键瓶颈。尽管并行执行可加快发现速度，但现有智能体存在34.9%的冗余调用率，抵消了并行优势。本文提出FuseSearch，将并行代码定位重构为质量-效率联合优化问题。通过定义工具效率（单位调用带来的唯一信息增益），采用两阶段监督微调与强化学习训练，学习自适应并行策略。不同于固定宽度方法，FuseSearch根据任务上下文动态调整搜索广度，从探索阶段演进至精炼阶段。在SWE-bench Verified上，FuseSearch-4B实现84.7%文件级和56.4%函数级F1分数，达到93.6%速度提升，调用次数减少67.7%，令牌消耗降低68.9%。结果表明，注重效率的训练能自然提升性能，通过消除噪声冗余信号，实现高性能且低成本的定位代理。
原文摘要 · Abstract (English)
Code localization constitutes a key bottleneck in automated software development pipelines. While concurrent tool execution can enhance discovery speed, current agents demonstrate a 34.9% redundant invocation rate, which negates parallelism benefits. We propose FuseSearch, reformulating parallel code localization as a joint quality-efficiency optimization} task. Through defining tool efficiency -- the ratio of unique information gain to invocation count -- we utilize a two-phase SFT and RL training approach for learning adaptive parallel strategies. Different from fixed-breadth approaches, FuseSearch dynamically modulates search breadth according to task context, evolving from exploration phases to refinement stages. Evaluated on SWE-bench Verified, FuseSearch-4B achieves SOTA-level performance (84.7% file-level and 56.4% function-level F1 scores) with 93.6% speedup, utilizing 67.7% fewer turns and 68.9% fewer tokens. Results indicate that efficiency-aware training naturally improves quality through eliminating noisy redundant signals, enabling high-performance cost-effective localization agents.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。