llama-cpp-python icon indicating copy to clipboard operation
llama-cpp-python copied to clipboard

HIP enabled binary, or allow me to replace with my own llama binary

Open madprops opened this issue 9 months ago • 0 comments

Hello. The rocm-hip-sdk on arch linux wants me to pull in 30gb of dependencies which is insane.

I just want to compile llama-cpp-python (llama.cpp) with hip support, to see if it helps.

CMAKE_ARGS="-DGGML_HIP=ON -DGGML_STATIC=OFF"

But I don't want to install 30gb of packages, my root partition can't even handle that.

My question is. Does this project provide a binary that is already hip enabled?

If not, is it possible to replace the llama binary this creates with my own?

For instance I'm seeing various hip enabled llama.cpp packages in the AUR.

I'm thinking I can just grab a made binary and use that.

madprops avatar Mar 25 '25 21:03 madprops