Name | Modified | Size | Downloads / Week |
---|---|---|---|
Parent folder | |||
koboldcpp-linux-x64-rocm | 2025-07-22 | 1.1 GB | |
koboldcpp-mac-arm64 | 2025-06-08 | 27.1 MB | |
koboldcpp-rocm-rolling source code.tar.gz | 2024-01-02 | 12.9 MB | |
koboldcpp-rocm-rolling source code.zip | 2024-01-02 | 13.1 MB | |
README.md | 2024-01-02 | 1.2 kB | |
Totals: 5 Items | 1.2 GB | 2 |
koboldcpp-rocm-rolling
Welcome to the KoboldCpp-rocm rolling release!
This release is a very experimental build that is build with rocm support for AMD devices. Builds are generated and uploaded automatically, performance and device compatibility are not guaranteed!
The binary is a 1GB self-extracting pyinstaller, that should work on x64 Linux environments with glibc equal to or newer than Ubuntu 22.04 and include ROCm 6.4.1 support (it works inside empty docker containers, system rocm may or may not be neccesary). Please consider upgrading or downgrading your AMD drivers for compatibility if you encounter issues. Limited support is provided for this build, as our maintainers do not personally have any AMD devices.
If you are looking for something more stable, please try the Vulkan (nocuda) release instead here: https://github.com/LostRuins/koboldcpp/releases/latest
Note: The source code zips are not accurate as the CI can't update these, refer to the concedo branch.
Because of incomplete support by AMD 9000 series is included without wmma support, we expect your GPU to be faster in ROCm 7.0. For other GPU's WMMA is included.