Towards Feedback-to-Plan Decisions for Self-Evolving LLM Agents in CUDA Kernel Generation
The paper discusses advancements in self-evolving large language models (LLMs) for CUDA kernel generation. It introduces a new analysis tool called CUDAnalyst, which helps in understanding how feedback influences planning decisions. The findings suggest that effective planning relies on aligned feedback and that stronger reasoning models can enhance weaker ones.
- ▪Large language models have shown strong gains as self-evolving agents for CUDA kernel generation.
- ▪CUDAnalyst provides a framework for analyzing the impact of feedback on planning decisions.
- ▪The study reveals that effective planning emerges from structured interactions of multiple feedback sources.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.26720 (cs) [Submitted on 26 May 2026] Title:Towards Feedback-to-Plan Decisions for Self-Evolving LLM Agents in CUDA Kernel Generation Authors:Yee Hin Chong, Jiaming Wu, Youhui Zhang, Peng Qu View a PDF of the paper titled Towards Feedback-to-Plan Decisions for Self-Evolving LLM Agents in CUDA Kernel Generation, by Yee Hin Chong and 3 other authors View PDF HTML (experimental) Abstract: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.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
