# passing embedding through dense networks
qs = self.W_Q(embedding)  # [batch_size x sequence_len x (query_key_dim * n_heads)]
ks = self.W_K(embedding)  # [batch_size x sequence_len x (query_key_dim * n_heads)]
vs = self.W_V(embedding)  # [batch_size x sequence_len x (value_dim * n_heads)]