Issues

• Low-latency Llama: the down-projection can read stale silu_out (the up/gate arrive has no release)

  HazyResearch/Megakernels#15 · Filter by author TanishGudise opened on Oct 27, 2026

• Running latency demo with cuda 13 generates invalid tokens, change to 12.8 fix the issue

  HazyResearch/Megakernels#9 · Filter by author AnDongLi opened on Mar 27, 2026

• How to generate the bueatiful timeline?

  HazyResearch/Megakernels#8 · Filter by author ziyuhuang123 opened on Dec 7, 2025

• How to run the code using 2H100?

  HazyResearch/Megakernels#7 · Filter by author ziyuhuang123 opened on Dec 7, 2025

• Error in vector TMA descriptor creation: invalid argument

  HazyResearch/Megakernels#5 · Filter by author jafioti opened on Oct 15, 2025
  1

• wait_for_kv() waits for K & V but not Q — what guarantees Q is ready?

  HazyResearch/Megakernels#4 · Filter by author Beibei-Zhou opened on Jul 1, 2025

• Failed to run in mk mode when batch_size is greater than 1

  HazyResearch/Megakernels#2 · Filter by author zhendonghua opened on Jun 3, 2025

• RTX 4090 Support - Compilation Issues with ThunderKittens

  HazyResearch/Megakernels#1 · Filter by author MIracleyin opened on May 28, 2025
  5