]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: 08f3f4a)
CUDA: Replace init_offsets kernel with iterators in cub-based argsort (#18930)
authorOliver Simons <redacted>
Tue, 20 Jan 2026 12:11:01 +0000 (13:11 +0100)
committerGitHub <redacted>
Tue, 20 Jan 2026 12:11:01 +0000 (20:11 +0800)
commitd1e3556481c8b351f9b7b69ba3febf6cb77fffa6
tree42adf2eb68175fdc2850c9c3036e63fd2b6f32c7tree
parent08f3f4a8a30633491b031bf833441de2a1ab5029commit | diff
CUDA: Replace init_offsets kernel with iterators in cub-based argsort (#18930)

* CUDA: Replace `init_offsets` with iterators in argsort

This is a QOL improvement, saving us the cost of materializing the
iterator

* Remove unnecessary include from top-k.cu
ggml/src/ggml-cuda/argsort.cu diff | blob | history
ggml/src/ggml-cuda/top-k.cu diff | blob | history
Packaging of ggml-org/llama.cpp