PreciseCoder: Faithful and Generalizable Debugging Beyond Unit-Test Rewards

Key points

• Reinforcement learning (RL) from unit tests is the common recipe for post-training code models, but a unit test only asks whether the final program passes.1
• We analyze what this unit-test-only reward teaches a model during debugging and find that, while it raises pass rates, it also teaches the model to edit more and to hallucinate bugs in code that is already correct.1
• Trained only on single-line bugs, PreciseCoder can generalize to repository-level program repair, where it resolves more issues with smaller patches.1
• Linguistic analysis of the thinking traces shows that our reward encourages the model to reason about individual lines, name the buggy ones, and form consistent bug hypotheses.1

Sources (2)

• [1]PreciseCoder: Faithful and Generalizable Debugging Beyond Unit-Test Rewards
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 13, 05:19 PM

  We introduce PreciseCoder, a training algorithm that augments the unit-test reward with a signal for semantically verified, localized edits, which correlates with smaller edits and higher pass rates.

  Reinforcement learning (RL) from unit tests is the common recipe for post-training code models, but a unit test only asks whether the final program passes.

• [2]PreciseCoder: Faithful and Generalizable Debugging Beyond Unit-Test Rewards
  Hugging Face Daily Papers · Oct 13, 12:00 AM · same content

Extractive summary: sentences quoted from the sources.