what is the correct way to apply weight normalization (nn.utils.weight_norm) to a multi-layer lstm? It looks like the only way I can apply weight normalization is to use nn.utils.weight_norm as many times as the number of weight layers (not a very desirable way to do it.). I have the following weights:
weight_ih_l0
weight_hh_l0
weight_ih_l0_reverse
weight_hh_l0_reverse
weight_ih_l1
weight_hh_l1
weight_ih_l1_reverse
weight_hh_l1_reverse