研究人员推出了一种名为协调Pass@K策略优化（CPPO）的新方法，通过同时探索多种不同的算法策略来增强代码生成能力。与抽取独立样本的标准方法不同，CPPO训练一个联合策略，其中规划器提出$K=4$个备选方法，共享求解器尝试为每个方法找到解决方案。这种协调探索在APPS、CodeContests和LiveCodeBench-v6等多个基准测试中，显著提高了pass@K指标。 AI
影响 这种协调策略探索有望带来更强大、更多样化的代码生成能力，尤其是在竞争性编程场景中。
排序理由 该集群包含一篇详细介绍代码推理和生成新方法的论文。
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
arXiv:2605.27000v1 Announce Type: cross Abstract: Repeated sampling with a verifier is the standard way to allocate test-time compute for code generation, with pass@$K$ as the canonical metric. Yet the standard policy class draws $K$ independent samples from a single answer distr…
Repeated sampling with a verifier is the standard way to allocate test-time compute for code generation, with pass@$K$ as the canonical metric. Yet the standard policy class draws $K$ independent samples from a single answer distribution, so attempts often collapse onto near-dupl…