]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: 96a181a)
metal: use shared buffers on eGPU (#17866)
authorJeremy Demeule <redacted>
Mon, 15 Dec 2025 14:14:49 +0000 (15:14 +0100)
committerGitHub <redacted>
Mon, 15 Dec 2025 14:14:49 +0000 (16:14 +0200)
commit165caaf5fbd359822b176b076f3ef4248309e56d
tree34f7a9ca4956f379c4b9c285b975e22082ac69fatree
parent96a181a933a34e92c316b0603f0f30cb7f30eefacommit | diff
metal: use shared buffers on eGPU (#17866)

* metal: use shared buffers on eGPU

With #15906, I noticed on important regression when using metal backend on eGPU.
This commit restore the previous behavior and add an option to force its activation.

* metal: use shared buffers on eGPU

* metal: use shared buffers on eGPU
ggml/src/ggml-metal/ggml-metal-device.m diff | blob | history
Packaging of ggml-org/llama.cpp