Works with latest llama.cpp version. (05/06/23)