Wenjie Gan, Zhaoxi Zhang, Jiahui Liang, Weikang Li, Wanyuan Wang, Li Zhang, Jian Dong · 2026 · arXiv (Cornell University)
DOI: 10.48550/arxiv.2610.27694 · cited by 0 works (OpenAlex) · open-access PDF
Große Sprachmodelle haben starke Codegenerationsfähigkeiten gezeigt, aber ihre Ausgänge enthalten oft subtile Fehler. Die automatische Reparatur solcher Codes erfordert informative Tests: Inputs, die fehlerhaftes Verhalten aufdecken, gepaart mit korrekten erwarteten Ausgängen. Neuere RL-basierte Testergenerationsmethoden wenden die gleiche Ergebnisbelohnung auf alle Begründungsschritte an, die nützliches Denken entmutigen und fehlerhafte Schritte verstärken können. Sie lernen auch nicht effektiv von Trajectories, die bug-triggerende Eingänge finden, aber falsche Ausgänge vorhersagen, so dass ihre verifizierten bug-triggerenden Fortschritte nicht ausreichend ausgenutzt werden. Um diese Einschränkungen anzugehen, sch
Machine translation (Meta M2M100); the English original follows.
Large Language Models have shown strong code generation capabilities, yet their outputs often contain subtle bugs. Automatically repairing such code requires informative tests: inputs that expose faulty behavior, paired with correct expected outputs. Recent RL-based test generation methods apply the same outcome reward to all reasoning steps, which can discourage useful reasoning and reinforce flawed steps. They also fail to effectively learn from trajectories that find bug-triggering inputs but predict incorrect outputs, leaving their verified bug-triggering progress insufficiently exploited. To address these limitations, we propose SCALF, a process-supervised RL approach that combines Step-level Credit Assignment with targeted Learning from valuable Failed trajectories. SCALF derives step-specific advantages from process rewards, allowing each reasoning stage to receive its own learning signal. It further routes trajectories that expose bugs but predict incorrect outputs to on-policy self-distillation, using execution feedback to learn from their valuable bug-triggering progress. Across three benchmarks, SCALF outperforms strong prompting and post-training baselines by up to 2.97, 5.26, and 3.53 percentage points in joint success rate, output correctness rate, and bug-triggering rate, respectively. Our code and datasets are available at https://anonymous.4open.science/r/SCALF-F4D1.
Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese
Topics: Correctness, Computer science, Process (computing), Code (set theory), Machine learning, Source code
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.27694, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange