llama.cpp
CUDA: skip fully masked-out KV in FA vec kernel
#13584

Merged

CUDA: skip fully masked-out KV in FA vec kernel #13584

JohannesGaessler merged 2 commits into ggml-org:master from JohannesGaessler:cuda-fa-opt-8

CUDA: skip fully masked-out KV in FA vec kernel

98543709

github-actions added Nvidia GPU

github-actions added ggml

fix AMD compilation

69647be0

ggerganov approved these changes on 2025-05-20

JohannesGaessler merged b69f1647 into master 252 days ago

ggerganov commented on 2025-05-24

Reviewers

ggerganov

Assignees

No one assigned

Labels

Nvidia GPU ggml

Milestone

No milestone