llama.cpp
CUDA: use mma FA kernel for gqa > 4 on RTX 4000
#15035

Merged

CUDA: use mma FA kernel for gqa > 4 on RTX 4000 #15035

JohannesGaessler merged 1 commit into ggml-org:master from JohannesGaessler:cuda-fa-kernel-choice

CUDA: use mma FA kernel for gqa > 4 on RTX 4000

069d410b

github-actions added Nvidia GPU

github-actions added ggml

ggerganov approved these changes on 2025-08-02

JohannesGaessler merged 03d46982 into master 217 days ago

Reviewers

ggerganov

Assignees

No one assigned

Labels

Nvidia GPU ggml

Milestone

No milestone