Support dedicated ML node

Follow up to what @ylwu mentioned before, we closed the loop on ML nodes in 2.1.