Anchored Self-Play for Code Repair

• URL: http://arxiv.org/abs/2607.03523v1
• Date: Fri, 03 Jul 2026 17:51:52 GMT
• Title: Anchored Self-Play for Code Repair
• Abstract summary: Code repair is an important capability for language models (LMs)<n>We propose __generator--fixer self-play__, in which a single model is trained with reinforcement learning to generate bugs and fix them.<n>On BugSourceBench, we find that self-play drifts toward difficult but unrealistic bugs.<n>We propose Anchored Self-Play (ASP), which anchors self-play with a small reference set by adding a code-embedding similarity reward for generation and mixing reference bugs into fixer training.
• Score: 68.0728003844994
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Code repair is an important capability for language models (LMs): given a buggy program and unit tests, an LM must produce a fixed program that passes the tests. Because code repair data is limited, we aim to scale supervision by using an LM to generate bug--fix tasks. We propose __generator--fixer self-play__, in which a single model is trained with reinforcement learning to generate bugs and fix them. As the fixer improves, the generator adapts to produce more difficult bugs, yielding an automatic curriculum. To test whether this curriculum generalizes, we introduce BugSourceBench, a repair benchmark spanning realistic bug sources: bugs in human-written code, LM-generated code, and human-edited LM-generated code. On BugSourceBench, we find that self-play drifts toward difficult but unrealistic bugs, improving on synthetic bugs but degrading on human-authored ones. We propose Anchored Self-Play (ASP), which anchors self-play with a small reference set by adding a code-embedding similarity reward for generation and mixing reference bugs into fixer training. Across bug sources, ASP achieves the best fix rates, improving average fix rate over standard self-play by $+24\%$ relative / $+7.0$ pp absolute, with gains on bugs from both LMs and humans.

Related papers

• BugPilot: Complex Bug Generation for Efficient Learning of SWE Skills [59.003563837981886]
  High quality bugs are key to training the next generation of language model based software engineering (SWE) agents.<n>We introduce a novel method for synthetic generation of difficult and diverse bugs.
  arXiv  Detail & Related papers  (2025-10-22T17:58:56Z)
• Do AI models help produce verified bug fixes? [62.985237003585674]
  Large Language Models are used to produce corrections to software bugs.<n>This paper investigates how programmers use Large Language Models to complement their own skills.<n>The results are a first step towards a proper role for AI and LLMs in providing guaranteed-correct fixes to program bugs.
  arXiv  Detail & Related papers  (2025-07-21T17:30:16Z)
• Learning to Solve and Verify: A Self-Play Framework for Code and Test Generation [69.62857948698436]
  Recent advances in large language models (LLMs) have improved their performance on coding benchmarks.<n>However, improvement is plateauing due to the exhaustion of readily available high-quality data.<n>We propose Sol-Ver, a self-play solver-verifier framework that jointly improves a single model's code and test generation capacity.
  arXiv  Detail & Related papers  (2025-02-20T18:32:19Z)
• PATCH: Empowering Large Language Model with Programmer-Intent Guidance and Collaborative-Behavior Simulation for Automatic Bug Fixing [34.768989900184636]
  Bug fixing holds significant importance in software development and maintenance.<n>Recent research has made substantial strides in exploring the potential of large language models (LLMs) for automatically resolving software bugs.
  arXiv  Detail & Related papers  (2025-01-27T15:43:04Z)
• Investigating the Transferability of Code Repair for Low-Resource Programming Languages [57.62712191540067]
  Large language models (LLMs) have shown remarkable performance on code generation tasks. Recent works augment the code repair process by integrating modern techniques such as chain-of-thought reasoning or distillation. We investigate the benefits of distilling code repair for both high and low resource languages.
  arXiv  Detail & Related papers  (2024-06-21T05:05:39Z)
• A Deep Dive into Large Language Models for Automated Bug Localization and Repair [12.756202755547024]
  Large language models (LLMs) have shown impressive effectiveness in various software engineering tasks, including automated program repair (APR) In this study, we take a deep dive into automated bug fixing utilizing LLMs. This methodological separation of bug localization and fixing using different LLMs enables effective integration of diverse contextual information. Toggle achieves the new state-of-the-art (SOTA) performance on the CodeXGLUE code refinement benchmark.
  arXiv  Detail & Related papers  (2024-04-17T17:48:18Z)
• Assessing the Latent Automated Program Repair Capabilities of Large Language Models using Round-Trip Translation [44.3761164214368]
  We investigate Round-Trip Translation (RTT): translating code from one programming language into another programming or natural language and back.<n>We perform a detailed quantitative and qualitative analysis of RTT-generated patches in Java.<n>We find that RTT through English generates plausible patches for 100 of 164 bugs with GPT-4 on the HumanEval-Java benchmark, and 97 are found to be correct in our manual assessment.
  arXiv  Detail & Related papers  (2024-01-15T22:36:31Z)
• Challenging Bug Prediction and Repair Models with Synthetic Bugs [7.285779914390472]
  BugFarm transforms arbitrary code into multiple complex bugs.<n>A comprehensive evaluation of 435k+ bugs from over 1.9M mutants generated by BUGFARM.
  arXiv  Detail & Related papers  (2023-10-03T20:01:51Z)
• Large Language Models of Code Fail at Completing Code with Potential Bugs [30.80172644795715]
  We study the buggy-code completion problem inspired by real-time code suggestion. We find that the presence of potential bugs significantly degrades the generation performance of the high-performing Code-LLMs.
  arXiv  Detail & Related papers  (2023-06-06T06:35:27Z)
• Prompting Is All You Need: Automated Android Bug Replay with Large Language Models [28.69675481931385]
  We propose AdbGPT, a new lightweight approach to automatically reproduce the bugs from bug reports through prompt engineering. AdbGPT leverages few-shot learning and chain-of-thought reasoning to elicit human knowledge and logical reasoning from LLMs. Our evaluations demonstrate the effectiveness and efficiency of our AdbGPT to reproduce 81.3% of bug reports in 253.6 seconds.
  arXiv  Detail & Related papers  (2023-06-03T03:03:52Z)
• Break-It-Fix-It: Unsupervised Learning for Program Repair [90.55497679266442]
  We propose a new training approach, Break-It-Fix-It (BIFI), which has two key ideas. We use the critic to check a fixer's output on real bad inputs and add good (fixed) outputs to the training data. Based on these ideas, we iteratively update the breaker and the fixer while using them in conjunction to generate more paired data. BIFI outperforms existing methods, obtaining 90.5% repair accuracy on GitHub-Python and 71.7% on DeepFix.
  arXiv  Detail & Related papers  (2021-06-11T20:31:04Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.