Show HN: Auto GPU Kernel – Autonomous GPU-kernel discovery and optimizer
Category: devtools
Tags: gpu-kernel, optimization, ai-agent, triton, mlsys
Score: 7.3/10 (Innovation: 8, Technical: 8, Documentation: 7, Utility: 6)
Auto GPU Kernel is an autonomous agent that discovers and optimizes GPU kernels, achieving a 34.93x speedup in the MLSys 2026 contest. It uses a loop of profiling, research, and benchmarking to iteratively improve kernel performance, integrating with FlashInfer and Modal for cloud execution. Its innovation lies in applying an AI agent to automate GPU kernel optimization, a traditionally manual and expert-driven task.
Target audience: backend devs, data engineers, ai-ml researchers
Repository: https://github.com/Dogacel/auto-gpu-kernel · Python · 38 stars
View on Hacker News