Repository navigation

[Fix] MiniMax-M3: default to the breakable prefill CUDA graph - #41845

Merged

hnyls2002 merged 14 commits into

sgl-project:mainfrom

thanhhao98:htphan/m3-tc-piecewise-traceable

Oct 9, 2026

Commits on Sep 30, 2026
- [Fix] Skip the deprecated getters' caller check while compiling
  Hao Phan
  committed
  05b91b9
- [Fix] Build the per-forward layer boundary values as plain classes
  Hao Phan
  committed
  9d34d3c
- [Fix] Build the deferred MoE finalize as a plain class
  Hao Phan
  committed
  50f7dd1
Commits on Oct 5, 2026
- Drop the explanatory comments; the code says it
  Hao Phan
  committed
  e1a748b
- Drop the no-dataclasses rule exception
  Hao Phan
  committed
  c8e203e
- Run sparse attention as an eager break under breakable
  Hao Phan
  committed
  6f999c3
- Default MiniMax-M3 prefill to the breakable CUDA graph
  Hao Phan
  committed
  c50fd69
- Merge origin/main into htphan/m3-tc-piecewise-traceable
  Hao Phan
  committed
  6287a23
- Port the fullgraph FFN-exit test to main's exit flags
  Hao Phan
  committed
  5a1b3ba
- Drop the tc_piecewise fixes
  Hao Phan
  committed
  0579e20
Commits on Oct 6, 2026
- Merge branch 'main' into htphan/m3-tc-piecewise-traceable
  nvpohanh
  authored
  84ead2b
Commits on Oct 8, 2026
- Merge branch 'main' into htphan/m3-tc-piecewise-traceable
  nvpohanh
  authored
  08c6b12
- Merge branch 'main' into htphan/m3-tc-piecewise-traceable
  kevin-mii
  authored
  981cf86
Commits on Oct 9, 2026
- Merge branch 'main' into htphan/m3-tc-piecewise-traceable
  
  nvpohanh
  and
  claude
  committed
  111f607