indeed, can't wait for it to be supported by llama.cpp (or other engines)?