fix: guard PRE_MLP NVFP4 fusion for dense layers - #17191

Open

zcxGGmu wants to merge 8 commits into

NVIDIA:mainfrom

zcxGGmu:fix/pre-mlp-nvfp4-unquantized-dense

Commits on Aug 3, 2026

fix: guard PRE_MLP NVFP4 fusion for dense layers
MaskerFather
committed
1d01e7a

Commits on Aug 5, 2026

test: assert PRE_MLP NVFP4 fallback branch
MaskerFather
committed
27d8384

Commits on Aug 6, 2026

test: add PRE_MLP NVFP4 guard to B200 test list
MaskerFather
committed
6dbe661

Commits on Aug 7, 2026

test: address PRE_MLP NVFP4 review feedback
MaskerFather
committed
2bdc269
test: collect nested PRE_MLP NVFP4 aliases
MaskerFather
committed
15dce66

Commits on Aug 8, 2026

test: tighten PRE_MLP NVFP4 alias detection
MaskerFather
committed
80c96cd

Commits on Aug 9, 2026

test: strengthen PRE_MLP NVFP4 guard test
MaskerFather
committed
cd6d0ff
test: drop obsolete split CPU stage entries
MaskerFather
committed
8167870