llama.cpp
CUDA: batched+noncont MMQ, refactor bs>1 MoE code
#13199

Merged

CUDA: batched+noncont MMQ, refactor bs>1 MoE code #13199

JohannesGaessler merged 1 commit into ggml-org:master from JohannesGaessler:cuda-moe-mmq-5

CUDA: batched+noncont MMQ, refactor bs>1 MoE code

12bef0d3

github-actions added testing

github-actions added Nvidia GPU

github-actions added ggml

slaren approved these changes on 2025-04-30

JohannesGaessler merged e1e8e099 into master 1 year ago

Reviewers

slaren

Assignees

No one assigned

Labels

testing Nvidia GPU ggml

Milestone

No milestone