Spectral Normalization for Recurrent Layers

Hi,

I am trying to implement a GAN model with a discriminator network containing a GRU layer. I use the utils function spectral_norm which works well with convolutional layers but unfortunately, it is not functional for recurrent layers. I tried to follow the implementation for a GRUCell but that’s not possible for the multilayer GRU as I cannot access its weight tensor. Any advice on how to fix that?

Best