Explanation on why off-the-self MAE models do not need proportional attention in `ToMeAttention.forward()`?

[Lopa07]

For MAE off-the-self models, self._tome_info['prop_attn'] should be False as mentioned here. Why for MAE off-the-self models, in ToMeAttention.forward() proportional attention is not needed?