SWE-Fuse: Empowering Software Agents via Issue-free Trajectory Learning and Entropy-aware RLVR Training

2026-03-09

AI Summary

Existing software engineering agents often suffer from performance degradation due to noisy problem descriptions. To address this limitation, this work proposes SWE-Fuse, a framework that trains agents by fusing both description-guided and description-free samples. It introduces description-free trajectory learning to circumvent the adverse effects of description noise and incorporates entropy-aware reinforcement learning (RLVR) alongside a dynamic clipping mechanism to adaptively balance exploration and training stability. Combined with test-time scaling (TTS), SWE-Fuse achieves substantial improvements in solve rates on SWE-bench Verified: reaching 49.8% and 65.2% for 8B and 32B models, respectively—corresponding to relative gains of 43.0% and 60.2% over the strongest baseline.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Planning, Routing, and Scheduling: Replanning and Plan Repair
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Large language models (LLMs) have transformed the software engineering landscape. Recently, numerous LLM-based agents have been developed to address real-world software issue fixing tasks. Despite their state-of-the-art performance, Despite achieving state-of-the-art performance, these agents face a significant challenge: \textbf{Insufficient high-quality issue descriptions.} Real-world datasets often exhibit misalignments between issue descriptions and their corresponding solutions, introducing noise and ambiguity that mislead automated agents and limit their problem-solving effectiveness. We propose \textbf{\textit{SWE-Fuse}}, an issue-description-aware training framework that fuses issue-description-guided and issue-free samples for training SWE agents. It consists of two key modules: (1) An issue-free-driven trajectory learning module for mitigating potentially misleading issue descriptions while enabling the model to learn step-by-step debugging processes; and (2) An entropy-aware RLVR training module, which adaptively adjusts training dynamics through entropy-driven clipping. It applies relaxed clipping under high entropy to encourage exploration, and stricter clipping under low entropy to ensure training stability. We evaluate SWE-Fuse on the widely studied SWE-bench Verified benchmark shows to demonstrate its effectiveness in solving real-world software problems. Specifically, SWE-Fuse outperforms the best 8B and 32B baselines by 43.0\% and 60.2\% in solve rate, respectively. Furthermore, integrating SWE-Fuse with test-time scaling (TTS) enables further performance improvements, achieving solve rates of 49.8\% and 65.2\% under TTS@8 for the 8B and 32B models, respectively.

Problem

• issue description
• software engineering agents
• trajectory learning
• RLVR training
• SWE-bench

Innovation

• issue-free trajectory learning
• entropy-aware RLVR
• SWE agent training
• trajectory fusion
• adaptive clipping