This is my approach to self-updating llama.cpp.
- Create all the files listed in the
folder_structure.txtfile from this gist. - Open the latest
llama.cpprelease on GitHub, for example: https://github.com/ggml-org/llama.cpp/releases/tag/b9587 - Download the required archives.
In my case, these were:
llama-b9587-bin-win-cuda-13.3-x64.zip
cudart-llama-bin-win-cuda-13.3-x64.zip