llama.cpp

428772f3 - use no_alloc to get memory requirements for model load