Efficient implementations of Dion and Muon optimizers for distributed ML