So then these models could be used by llama.cpp today with the -md switch?

Interesting, must try tomorrow.