In that case, I only decline in very exceptional situations - or, if I recommended a clear reject before and made clear that I think the errors are not fixable. Because then, I would not have anything new to add in the new round (and I would also find it unfair towards the author)
