Failed to install llama-cpp-python with Metal on M2 Ultra

reno_queen

I followed the instruction on https://llama-cpp-python.readthedocs.io/en/latest/install/macos/ . My macOS version is Sonoma 14.4, and xcode-select is already installed (version: 15.3.0.0.1.1708646388). I created a conda environment with "conda version : 24.1.2" "python version : 3.10.13.final.0" "platform : osx-arm64". And I run the following command to install llama-cpp-python CMAKE_ARGS="-DLLAMA_METAL=on" pip install -U llama-cpp-python...

Originally asked by ooyeon on Stack Exchange (stackoverflow) (CC BY-SA 4.0).

uphilltech

Great question — answering as the UphillTech team:

Firstly, I'd like to thank Spo1ler's finding. I only installed xcode-select not Xcode.app and its path was /Library/Developer/CommandLineTools Based on https://github.com/gfx-rs/gfx/issues/2309#issuecomment-506130902 , I installed Xcode.app from Apple App Store. Then, I can see that the path of xcode-select is changed automatically to /Applications/Xcode.app/Contents/Developer Then, I had to run Xcode.app first to agree to the Xcode license agreements. After that, the following commands work. export CMAKE_ARGS="-DLLAMA_METAL=on" export FORCE_CMAKE=1 pip install llama-cpp-python --no-cache-dir But, I got another error message when I call LlamaCpp module as follows. ggml_metal_init:...

Full answer by ooyeon: https://stackoverflow.stackexchange.com/a/78190308 (CC BY-SA 4.0).

Need it done for your business? UphillTech.