llama.cpp
CUDA: size routed MoE MMQ N-tiles from typical expert width on RDNA3 (recreated)
#28552
Open
Open
CUDA: size routed MoE MMQ N-tiles from typical expert width on RDNA3 (recreated) #28552
pwilkin
wants to merge
2 commits into
master
from
pr-24546-recreated
CUDA: size routed MoE MMQ N-tiles from typical expert width on RDNA3 …
79800853
CUDA: pick MMQ tile size against ncols_opt set on the host side
f01a7549
JohannesGaessler
commented
on 2026-09-07
JohannesGaessler
approved these changes
on 2026-09-08
Login to write a write a comment.
Login via GitHub