- it should automatically detect the best device to run on - We should require 0 manual configuration from the user, by default llama.cpp for example requires specifying the device