Weird behavior for residue connection for Wave_head...

[chemgeeklian]

In https://github.com/PraljakReps/ProtWaveVAE_model/blob/main/ProtWave_VAE/wavenet_decoder.py

In forward of class Wave_head

           if ii == 0:
                x = orig_x + res # residual enter the next layer (most likely dilated)
            else:
                x = res + x
]

Why is the behavior different for ii==0 or not? When ii !=0, you use the gated product as x, but why not do that when ii==0?