All pull requests

Pull requests

• refactor: rename the Python extension and organize layout helpers

  #111·Fang-Yuhang opened Nov 4, 2026Collaborator·

  ·
  2

• refactor(common): organize device helpers and assertion macros

  #108·Fang-Yuhang opened Nov 4, 2026Collaborator·

  ·
  5

• feat: add MegaMoE UT format and HGTX profiling

  #106·yonghaogu666 opened Nov 4, 2026·

  ·
  1

• perf(jit): reduce warm CPU overhead for dense GEMM

  #105·fangshoujia opened Nov 4, 2026Collaborator·

  ·
  1

• Optimize BF16 M=32 dense GEMM on 810E

  #102·fangshoujia opened Nov 4, 2026Collaborator·

  ·
  11

• Fix invalid expert handling and uninitialized metadata in MoE alignment

  #99·mingyue300 opened Nov 3, 2026·

  ·
  2

• refactor(jit): consolidate native compilation and kernel loading

  #92·Fang-Yuhang opened Oct 30, 2026Collaborator·

  ·
  6

• perf: optimize W4FA8 grouped MoE GEMM on M890

  #91·lfwang56 opened Oct 30, 2026·

  ·
  37

• support BF16 x FP32 dense GEMM

  #85·tangjian-spkey opened Oct 29, 2026Collaborator·

  ·
  4

• feat: add blockwise and channelwise FP8 GEMM + ReduceScatter

  #82·lfwang56 opened Oct 29, 2026·

  ·
  1

  21

• feat: add paged mqa

  #76·Yefeng-Li opened Oct 27, 2026Collaborator·

  ·
  13

• feat(tuner): add experimental legacy JSON to LUT conversion

  #30·Fang-Yuhang opened Oct 14, 2026Collaborator·1 task done·

  ·
  

• feat: add C++ JIT GEMM autotuning and LUT export

  #20·tieshengyao opened Oct 13, 2026Collaborator·

  ·
  3