errors in class MultiHeadedAttention? about annotated-transformer

[AwsomeName]

[l(x).view(nbatches, -1, self.h, self.d_k).transpose(1, 2)
for l, x in zip(self.linears, (query, key, value))]
the same line, this func apply 3 linears , return result [ l1(query), l2(key), l3(value) ]. i think this may not the idea of the multi-head attention. do i wrong??, please tell me if im wrong, thank you !

[Menandalbee]

[l(x).view(nbatches, -1, self.h, self.d_k).transpose(1, 2)
for l, x in zip(self.linears, (query, key, value))]
the same line, this func apply 3 linears , return result [ l1(query), l2(key), l3(value) ]. i think this may not the idea of the multi-head attention. do i wrong??, please tell me if im wrong, thank you !

Note that the multi-head attention has been considered when self.linears is defined, since its dimentsion is (d_model, d_model) rather than (d_model, d_k).