llama.cpp

4adfcb22 - move model memory estimation to subprocess