llama.cpp
CUDA: MoE helper in device code, better tile sizes
#15525

Merged

CUDA: MoE helper in device code, better tile sizes #15525

JohannesGaessler merged 6 commits into ggml-org:master from JohannesGaessler:cuda-mmq-moe-device-4

CUDA: MoE helper in device code, better tile sizes

8bb55de6

github-actions added Nvidia GPU

github-actions added ggml

reduce superfluous CUDA blocks

07c814b5

IMbackK requested changes on 2025-08-23

try AMD fix

e7b884da

IMbackK approved these changes on 2025-08-23

raise shared memory limit

57249900

reduce shared memory use

a2f702a9

add assert

1d609235

JohannesGaessler merged 5eff6ec9 into master 140 days ago

Reviewers

IMbackK

Assignees

No one assigned

Labels

Nvidia GPU ggml

Milestone

No milestone