Enhancing Issue Localization Agent with Tool-Interactive Training
Issue localization, the process of identifying code locations that need modification to resolve software issues, is a critical yet challenging task in software development. The semantic gap between natural language issue descriptions and faulty code requires complex multi-hop reasoning through code dependencies. Existing LLM-based agents attempt to address this by integrating repository retrieval tools, which poses a higher demand for LLMs to effectively utilize various tools during multi-step reasoning for issue localization. To tackle this challenge, we present ToolTrain, a two-stage tool-interactive training framework combining rejection-sampled supervised fine-tuning and tool-interactive reinforcement learning to enhance LLMs’ ability to use retrieval tools for issue localization. Experimental results show that ToolTrain-trained models achieve state-of-the-art performance, with our 32B model even surpassing Claude-3.7 on function-level localization. The results also show that improved localization performance translates to better end-to-end issue resolution performance. This further demonstrates that training for issue localization is a viable and effective strategy for improving automated software development.
Fri 17 AprDisplayed time zone: Brasilia, Distrito Federal, Brazil change
16:00 - 17:30 | AI for Software Engineering 27Research Track / New Ideas and Emerging Results (NIER) at Asia IV Chair(s): Giuseppe Scanniello University of Salerno | ||
16:00 15mTalk | Setup AGent (SAG): A Dual-Model LLM Agent for Autonomous End-to-End Java Project Configuration New Ideas and Emerging Results (NIER) Chenhao Wei Stevens Institute of technology, Gengwu Zhao Stevens Institute of Technology, Xinyi Li Stevens Institute of Technology, Billy Ye Stevens Institute of Technology, Lu Xiao Stevens Institute of Technology | ||
16:15 15mTalk | MAJIT: Just-in-Time Detection of Compatibility Issues in Android and iOS Apps through Large Language Model-based Multi-Agent Collaboration New Ideas and Emerging Results (NIER) Jiaqi Wang Xidian University, Di Cui Xidian University, Shenghan Liu Douyin, Qiankang Mao Douyin, xiangxingqian Douyin, Qiaoyin Gan Douyin, Rui Li Media Attached | ||
16:30 15mTalk | Code Wars: Adversarial Self-Play for Evolving Software Validation Tools New Ideas and Emerging Results (NIER) | ||
16:45 15mTalk | SEAlign: Alignment Training for Software Engineering AgentDistinguished Paper Award Research Track Kechi Zhang Peking University, China, Huangzhao Zhang Verdent AI, Ge Li Peking University, Jinliang You Peking University, Jia Li , Yunfei Zhao Peking University, Zhi Jin Peking University, Wuhan University | ||
17:00 15mTalk | Atomizer: An LLM-based Collaborative Multi-Agent Framework for Intent-Driven Commit Untangling Research Track Kangchen Zhu National university of Defense Technology, Zhiliang Tian National University of Defense Technology, Shangwen Wang National University of Defense Technology, mingyue leng National University of Defense Technology, Xiaoguang Mao National University of Defense Technology Media Attached | ||
17:15 15mTalk | Enhancing Issue Localization Agent with Tool-Interactive Training Research Track Zexiong Ma Peking University, Chao Peng ByteDance, Qunhong Zeng Beijing Institute of Technology, Pengfei Gao ByteDance, Yanzhen Zou Peking University, Bing Xie Peking University Pre-print | ||