Hacker GitHub

← All repos

mirage-project/mirage

View on GitHub ↗ ★ 2,608 Cuda

Mirage Persistent Kernel: Compiling LLMs into a MegaKernel

4 mentions on Hacker News by 4 people: 2 stories, 2 comments · first 2024-05-14 · latest 2025-06-20

Mentions per year 4 mentions · busiest year 2024 (2)
20082026
AllStoriesComments NewestTopOldest
  1. …the code actually runs well. I can reproduce the results using Modal GPUs, leaving the code here: https://github.com/mirage-project/mirage/pull/327/files Triton + FlashInfer: Prompt length 39, generate length 264, per-token latency…
    comment by gongy · 2025-06-20 · in Compiling LLMs into a MegaKernel: A path to low-latency inference
  2. The github repo includes a tutorial for using MPK: https://github.com/mirage-project/mirage/tree/mpk
    comment by zhihaojia · 2025-06-19 · in Compiling LLMs into a MegaKernel: A path to low-latency inference