arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2605.26720cs.AI
收藏

面向CUDA内核生成中自进化LLM代理的反馈到计划决策

Towards Feedback-to-Plan Decisions for Self-Evolving LLM Agents in CUDA Kernel Generation

Department of Computer Science and Technology, Tsinghua University, Beijing, China(清华大学计算机科学与技术系);Beijing National Research Center for Information Science and Technology, Beijing, China(北京信息科学国家研究中心)

Yee Hin Chong, Jiaming Wu, Youhui Zhang, Peng Qu

AI总结:

通过轨迹冻结和选择性反馈注入,提出CUDAnalyst框架以归因规划决策对反馈组件的贡献,揭示显式规划仅在反馈对齐时有效,且有效规划源于结构化多反馈交互。

摘要:

大型语言模型(LLMs)作为自进化代理在CUDA内核生成中展现出强大的实证收益,这得益于跨代际的反馈条件规划。然而,规划决策如何归因并组合异构反馈信号仍不透明。标准的端到端消融无法解决这一问题,因为迭代规划放大了早期扰动,并将反馈效应与轨迹依赖漂移混为一谈。我们引入 exttt{CUDAnalyst},一个统一的分析层,通过轨迹冻结和选择性反馈注入,实现对规划决策到反馈组件的受控、代际级归因。 exttt{CUDAnalyst}支持稳定的代际级评估和原则性的联盟式反馈效应及交互归因。我们的结果表明,显式规划仅在反馈对齐时有益,有效规划源于结构化的多反馈交互,且来自更强推理模型的高级规划可部分迁移至较弱模型。这些趋势在参考骨干网络、代表性工作负载和参考归纳机制中保持一致,表明在所研究的受控轴内,识别出的反馈到规划结构是稳健的。

英文摘要:

Large language models (LLMs) have shown strong empirical gains as self-evolving agents for CUDA kernel generation, driven by feedback-conditioned planning across generations. However, how planning decisions attribute and combine heterogeneous feedback signals remains opaque. Standard end-to-end ablations fail to resolve this question, as iterative planning amplifies early perturbations and conflates feedback effects with trajectory-dependent drift. We introduce \texttt{CUDAnalyst}, a unified analysis layer for controlled, generation-level attribution of planning decisions to feedback components via trajectory freezing and selective feedback injection. \texttt{CUDAnalyst} enables stable generation-level evaluation and principled coalitional-style attribution of feedback effects and interactions. Our results show that explicit planning is beneficial only when feedback is aligned, that effective planning emerges from structured multi-feedback interactions, and that high-level plans from stronger reasoning models can partially transfer to weaker ones. These trends hold across reference backbones, representative workloads, and reference induction regimes, indicating that the identified feedback-to-plan structure is robust within the controlled axes studied.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑