Package Details: llama.cpp-hip b6189-1

Git Clone URL: https://aur.archlinux.org/llama.cpp-hip.git (read-only, click to copy)
Package Base: llama.cpp-hip
Description: Port of Facebook's LLaMA model in C/C++ (with AMD ROCm optimizations)
Upstream URL: https://github.com/ggml-org/llama.cpp
Licenses: MIT
Conflicts: ggml, libggml, llama.cpp
Provides: llama.cpp
Submitter: txtsd
Maintainer: Orion-zhen
Last Packager: Orion-zhen
Votes: 8
Popularity: 1.32
First Submitted: 2024-10-26 19:54 (UTC)
Last Updated: 2025-08-18 00:40 (UTC)

Pinned Comments

Orion-zhen commented on 2025-08-16 10:24 (UTC)

Make sure ROCm is correctly set in your system before installing this package.

  1. sudo pacman -S rocm-hip-sdk rocm-hip-libraries rocm-hip-runtime
  2. Reboot (MUST, or the system will not be able to recognize ROCm)
  3. Setup environment variables, including ROCM_HOME=/opt/rocm

txtsd commented on 2024-10-26 20:15 (UTC) (edited on 2024-12-06 14:15 (UTC) by txtsd)

Alternate versions

llama.cpp
llama.cpp-vulkan
llama.cpp-sycl-fp16
llama.cpp-sycl-fp32
llama.cpp-cuda
llama.cpp-cuda-f16
llama.cpp-hip

Latest Comments

1 2 3 Next › Last »

Orion-zhen commented on 2025-08-16 10:24 (UTC)

Make sure ROCm is correctly set in your system before installing this package.

  1. sudo pacman -S rocm-hip-sdk rocm-hip-libraries rocm-hip-runtime
  2. Reboot (MUST, or the system will not be able to recognize ROCm)
  3. Setup environment variables, including ROCM_HOME=/opt/rocm

Orion-zhen commented on 2025-08-04 08:06 (UTC) (edited on 2025-08-04 12:25 (UTC) by Orion-zhen)

Hi @Valantur.

Actually, I removed service file and configuration file from the PKGBUILD, because my automatic update script have difficulty uploading assets files. And TBH, I have never used llama.cpp service before, because I usually run multiple models, such as chat model, embedding model and reranking model. Switching between these models via llama.cpp service is difficult. Instead of llama.cpp service, I recommend llama-swap, an application that switches model based on requests. So if you need llama.cpp service, please write it on your own, thanks.

Valantur commented on 2025-08-02 17:12 (UTC)

Upgraded today and the service wont' start because the service file is a symlink to your build system instead of a real text file.

panikal commented on 2025-07-20 05:44 (UTC) (edited on 2025-07-20 05:45 (UTC) by panikal)

This wouldn't build for me due to errors about not finding the ROCm

-- The HIP compiler identification is unknown
CMake Error at /usr/share/cmake/Modules/CMakeDetermineHIPCompiler.cmake:174
(message):
Failed to find ROCm root directory.
Call Stack (most recent call first):
ggml/src/ggml-hip/CMakeLists.txt:36 (enable_language)

I added this to my install to make it work, this should really be part of the build

ROCM_PATH=/opt/rocm
PATH=$ROCM_PATH/bin:$PATH
LD_LIBRARY_PATH=$ROCM_PATH/lib:$ROCM_PATH/lib64:$LD_LIBRARY_PATH

Jawzper commented on 2025-07-05 13:08 (UTC)

Thank you @edtoml, those PKGBUILD changes seem to have done the trick.

I removed libggml-hip-git during the installation to avoid conflicts, and now I'm not able to install it again (because of said conflicts). I only had it in the first place because it was a dependency for llama.cpp-hip. Is it fine?

edtoml commented on 2025-07-05 11:49 (UTC) (edited on 2025-07-05 11:50 (UTC) by edtoml)

edtoml commented on 2025-07-05 02:02 (UTC)

This has broken AGAIN due to libggml-hip-git being out of sync.

It can be made by changing option in the build() section (ON has been changed to OFF)

-DLLAMA_USE_SYSTEM_GGML=OFF

If you are on RDNA3 or RDNA4 adding the line below speeds up flash attention

-DGGML_HIP_ROCWMMA_FATTN=ON

You also need to remove the depend for libggml-hip in the depends=( list

There are dependencies you will need to resolve as files added by libggml-hip-git are going to conflict.

Jawzper commented on 2025-07-04 08:17 (UTC)

I am not able to update... can anyone help me?

https://pastebin.com/rqdQNXDq

edtoml commented on 2025-07-01 17:01 (UTC)

Looks like libggml-hip has been updated and after it rebuilds this package builds correctly.