That isn’t possible, training is non-deterministic.