Daily prebuilt llama.cpp shared libraries for Koharu.
The release workflow resolves the newest listed llama.cpp release, including prereleases, and builds CUDA, HIP, Vulkan, and Metal variants for Windows, Linux, and macOS. It runs daily at 03:17 UTC, on pushes to main, and when started manually.
Build jobs use the standard ubuntu-latest, windows-latest, and macos-latest GitHub-hosted runners.
Each archive holds llama, mtmd, and llama-common. llama-common is llama.cpp's common library with the C wrapper in cmake/wrapper, from llama-cpp-rs, so callers that load it at runtime can use it through llama_rs_* functions.
Generated binaries are published through this repository's releases.
This repository is dual-licensed under the MIT License or the Apache License, Version 2.0, at your option.
Third-party code and dependencies retain their respective licenses.