Lecture 79 Mirage (MPK): Compiling LLMs into Mega Kernels

Talk by Mengdi Wu and Xinhao Cheng on Mirage. Mirage Persistent Kernel (MPK) is a compiler and runtime system that automatically transforms LLM inference into a single megakernel—a fused GPU kernel that performs all necessary computation and communication within a single kernel launch. This end-to-end GPU fusion approach reduces LLM inference latency by 1.2× to 6.7×, all while requiring minimal developer effort. Repo: https://github.com/mirage-project/mirage