AEN_BERT question about s1_mean calcuation

[Pager07]

Greetings author,

You calcuated s1_mean as follow

s1_mean = torch.div(torch.sum(s1, dim=1), context_len.view(context_len.size(0), 1))

I fail to understand why s1 of shape (bs,target_seq_len,d_model) is being divided by the length of the context. Why is it not divided by the length of the target?

Thank you.
Sandeep