question about computing contrastive loss

[Pixie8888]

Hi,
Why you use Variable(self.lip_embedding_norm.data) instead of direct use self.lip_embedding_norm here? And why denominator is 2*batch_size instead of batch_size^2?