Aren’t they already separated? Max is an inference runtime, written in Mojolang, which uses the MLIR compiler chain to target diverse accelerator hardware.