GraphAlignCoder：对齐程序与证明图以提升代码生成

AI 摘要

GraphAlignCoder 是一种新的代码生成训练框架，通过将程序图与 Lean 证明图对齐，将显式的正确性结构注入代码生成。实验表明，相比 CodeRL，它在 LiveCodeBench v6 上解决数从 38 提升到 50，在 BigCodeBench Hard 上从 16 提升到 23，相对提升 31.6% 和 43.8%。该框架通过结构注入和巩固两个阶段，显著提升了代码生成的正确性。

正文节选

GraphAlignCoder: Aligning Program and Proof Graphs for Code Generation Abstract Code large language models (LLMs) can generate syntactically plausible programs that nevertheless violate hidden semantic constraints. Existing execution-feedback training methods identify whether a completed program fails, but provide limited supervision about how a correct solution should be organized. We introduce GraphAlignCoder, a training framework that transfers explicit correctness structure into code generat

发布时间：2026-08-13 12:00

抓取时间：2026-08-13 14:33

来源机构：arXiv