暹罗网络中可拉斯对称问题

I'm trying to implement a siamese network is Keras for textual similarity, but my network doesn't seem to be symmetrical. When I'm testing, the similarity score it is giving for A, B is different from B, A. How can I fix it? I have tried dot mode as well, as well as all Dense layers. My code is,I'm trying to implement a siamese network is Ke