[Auracle]

Unless I missed something just from skimming their tutorial it looks like they can do parallelism to speed things up with some models, not actually split the model (apart from the usual chunk offloading techniques).