About MLN（Modified Layer Normalization）

[youngtboy]

This paper provides new perspectives about Transformer block, but I have some questions about one of the details. As far as I know, the LayerNorm officially provided by Pytorch implements the same function as the MLN, which computes the mean and variance along token and channel dimensions. So where is the improvement? The official example : #Image Example N, C, H, W = 20, 5, 10, 10 input = torch.randn(N, C, H, W) #Normalize over the last three dimensions (i.e. the channel and spatial dimensions) #as shown in the image below layer_norm = nn.LayerNorm([C, H, W]) output = layer_norm(input)

[yuweihao]

Hi @youngtboy ,

Thanks for your attention. Please refer to this issue #9 .

[youngtboy]

Hi @youngtboy ,

Thanks for your attention. Please refer to this issue #9 .

Thanks for your explanation!

[yuweihao]

Duplicate of #9