Comment by thelastparadise

But do we know that the same techniques won't scale if trained in the huge clusters?