llama.cpp

7d587e55 - ggml-metal: do not copy headers for embedded, use current binary dir for embedded (#18705)