llama.cpp

a98b1633 - nix : add cuda, use a symlinked toolkit for cmake (#3202)