8000 CUDA: skip fully masked-out KV in FA vec kernel by JohannesGaessler · Pull Request #13584 · ggml-org/llama.cpp · GitHub
[go: up one dir, main page]
Skip to content

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

Merged
JohannesGaessler merged 2 commits intoggml-org:masterfrom
JohannesGaessler:cuda-fa-opt-8
May 20, 2025

Commits

Commits on May 15, 2025

Commits on May 16, 2025
0