llama.cpp

be46a509 - model: using single llm_build per arch