ParentFull threadAmbix·Is it possible to write llama.cpp in Futhark? Like do effective math manipulations on 4-bit vectors within GPU?View on HN