I thought this had something to do with the DGX Spark at first from the name haha. (Incidentally, a lot of recent work has gone into making the DGX Spark better at inference, like MTP yielded a 50-100% speedup, so DSpark will likely be very helpful to that end as well)