llama.cpp
CUDA: fix overflow in MMA kernel without stream-k
#17939

Merged

CUDA: fix overflow in MMA kernel without stream-k #17939

JohannesGaessler merged 1 commit into ggml-org:master from JohannesGaessler:cuda-fa-fix-ampere

CUDA: fix overflow in MMA kernel without stream-k

ed39524e

github-actions added Nvidia GPU

github-actions added ggml

ggerganov approved these changes on 2025-12-12

JohannesGaessler merged 48221143 into master 6 days ago

Reviewers

ggerganov

Assignees

No one assigned

Labels

Nvidia GPU ggml

Milestone

No milestone