JakeATX@JakeATX · Sep 7A llama.cpp variant optimised for Ampere cards, reporting very high token rates on 3090 and 3090 Ti, relevant for Flash-Next hosts.
