Optimize convolution and groupnorm - #98

Open

ndryden wants to merge 8 commits into

mainfrom

Commits on Aug 5, 2026

Add a channels-last-native Triton GroupNorm
ndryden
committed
792cfff
Route the UNet GroupNorm through the Triton kernel
ndryden
committed
c6b6725
Add a Triton 3-D convolution package
ndryden
committed
0f9a38c
Route every convolution through the Triton rung
ndryden
committed
fcb6964
Cite findings rather than the scratch tree that produced them
ndryden
committed
b5fe853
Add helper script to run tests.
ndryden
committed
77cf50d
Satisfy the style checks CI actually runs
ndryden
committed
981e7af

Commits on Aug 6, 2026

Say which kernel each accelerated module ended up using
ndryden
committed
84e5624