llama.cpp

2603b4c5 - use no_alloc to get memory requirements for model load