why you use layer_norm ?

[taki0112]

Hi
code
In Decoder, you used the layer_norm

have you tried another normalization ? like instance, batch, group

[xunhuang1995]

Instance norm does not work well, since it removes the global feature mean and variance that capture the style information.

Batch norm is the same since we use batch size = 1.

The paper had been submitted before group normalization came out. So we haven't try it.

[taki0112]

I have some question...

If you look at configs, there is no information about animal translation.
When will you upload it?
Is it ok to just run as hyper-parameter like edge2shoes?