Introduce Mega-C++ to reduce CPU overhead - #3099

Open

zhongbozhu wants to merge 8 commits into

NVIDIA:mainfrom

zhongbozhu:main_megacpp_grouped_mlp

Commits on Jun 13, 2026

support scaled swiglu, scaled srelu and scaled clamp swiglu
zhongbozhu
committed
18d4d2c

Commits on Jun 16, 2026

vectorized loading improvement
zhongbozhu
committed
953c469
fix bug for backward kernel
zhongbozhu
committed
e3ae293
optimize
zhongbozhu
committed
84cbdec
fix unit test failure
zhongbozhu
committed
c73c8ea
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot]
committed
3eb18a6
megacpp grouped mlp
zhongbozhu
committed
7044104
integrate fused scaled swiglu and srelu
zhongbozhu
committed
07b2836