llama.cpp

494503a0 - use no_alloc to get memory requirements for model load