daVinci-kernel: Co-Evolving Skill Selection, Summarization, and Utilization via RL for GPU Kernel Optimization

daVinci-kernel is a reinforcement learning framework designed for GPU kernel optimization, integrating skill selection, summarization, and utilization through a shared LLM backbone. It features three agents: a Skill Selection Agent using BM25 and LLM reranking, a Policy Agent generating CUDA/Triton kernels, and a Skill Summary Agent distilling successful rollouts. Achieving benchmark results of 37.2%, 70.6%, and 32.2% on KernelBench at Level 1, Level 2, and Level 3 respectively, daVinci-kernel-14B surpasses the performance of the previous leading model, Dr.Kernel-14B, highlighting its potential for enhancing execution efficiency in GPU programming.