技术报告：NVIDIA Blackwell上的人工智能辅助门控DeltaNet优化
Technical Report: AI-Assisted Gated DeltaNet Optimization on NVIDIA Blackwell
浏览论文内容
中文总结 AI 辅助
研究以MSInfer团队提交给MLSys 2026 FlashInfer竞赛的作品为例，探讨人工智能辅助GPU编程。该作品优化门控DeltaNet解码和预填充，在NVIDIA B200/Blackwell上实现1.58倍加速，揭示竞赛级优化需考虑多方面，将其视为端到端系统问题。
中文摘要 AI 辅助
人工智能辅助的GPU编程通常被构建为一个内核生成循环：要求模型生成更快的CUDA代码，对结果进行基准测试，然后重复。本案例研究认为，竞赛级别的优化不仅仅是改进内核主体。我们研究了我们团队MSInfer提交给MLSys 2026 FlashInfer竞赛的Agent-Assisted作品。该作品在NVIDIA B200/Blackwell上对门控DeltaNet解码和预填充进行了优化，实现了官方1.58倍的加速，解码的平均延迟约为9.315微秒，预填充的平均延迟约为239.4微秒。我们的经验表明，当工作负载需要结构重新制定和与评估器对齐的测量时，即使是有效的局部内核改进也可能达到瓶颈。因此，我们将人工智能辅助的内核优化描述为一个端到端的系统问题，它包括算法设计、工作负载专业化、测量工具、构建和评估表面、评估器对齐以及人工解释。
英文摘要
AI-assisted GPU programming is often framed as a kernel-generation loop: ask a model to produce faster CUDA code, benchmark the result, and repeat. This case study argues that contest-grade optimization involves more than improving the kernel body. We examine the Agent-Assisted submission by our team, MSInfer, to the MLSys 2026 FlashInfer Contest. The submission optimized Gated DeltaNet decode and prefill on NVIDIA B200/Blackwell and achieved an official $1.58\times$ speedup, with approximate average latencies of $9.315\,μ\mathrm{s}$ for decode and $239.48\,μ\mathrm{s}$ for prefill. Our experience shows that even effective local kernel improvements can plateau when a workload requires structural reformulation and evaluator-aligned measurement. We therefore characterize AI-assisted kernel optimization as an end-to-end systems problem that encompasses algorithm design, workload specialization, measurement tooling, build and evaluation surfaces, evaluator alignment, and human interpretation.
