Repository contains tooling necessary to build llama.cpp with CUDA support for Linux systems.
Focuses on SUSE binary-compatibility targeting OpenSUSE Leap, and the similar commercial SLES.

The build assets are locally compiled and uploaded.
Past release version assets might be deleted at any time to preserve storage space.

Important

You need to have cuda-toolkit 13 (dynamically linked) and associated gpu installed for the precompiled binaries. You also need openblas (+openmp) installed.

Build signed with committer GitHub public key.

ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIAjHQTVR0rwFW0l2ocQC9+WJqEA+QbwNeAu9Ro2r0DTa signer.pub

Verifying file signature using

ssh-keygen -Y verify \
-f <(echo "security@webesque.agency ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIAjHQTVR0rwFW0l2ocQC9+WJqEA+QbwNeAu9Ro2r0DTa") \
-I security@webesque.agency \
-n file \
-s build.tar.xz.sig \
< build.tar.xz

Good "file" signature for security@webesque.agency with ED25519 key SHA256:dtIP/UBbUaZ6q2qA...

License

Files in this repository are Apache 2.0 licensed.

llama.cpp source code licensed under MIT.

cuda-toolkit, and associated NVIDIA development libraries installed in container image, and used during build, licensed under NVIDIA Software License Agreement and CUDA Supplement to Software License Agreement.