Question regarding residual connection module : x + self.dropout(sublayer(self.norm(x)))

[ssvision]

from the Vaswani et al. paper as highlighted ,

Q : the paper says LayerNorm(x + sublayer(x)) but why are we doing x + LayerNorm(sublayer(x)). i understood why we are doing sublayer(LayerNorm(x)) instead of LayerNorm(sublayer(x)) as it helps in better training as per certain papers. But my main question is why is "x + ..." part not included in LayerNorm?