Learning Adaptive Parallel Execution for Efficient Code Localization

TL;DR

FuseSearch introduces an adaptive, efficiency-aware approach to parallel code localization, significantly improving speed and accuracy by dynamically balancing exploration and refinement stages.

Contribution

The paper presents FuseSearch, a novel method that optimizes parallel code localization by jointly considering quality and efficiency, using a two-phase training approach.

Findings

Achieves state-of-the-art performance with 84.7% file-level F1 score

Realizes 93.6% speedup over baseline methods

Uses 67.7% fewer search turns and 68.9% fewer tokens

Abstract

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 \textbf{FuseSearch}, reformulating parallel code localization as a \textbf{joint quality-efficiency optimization} task. Through defining \textbf{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 $F_1$ 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.