I think something wrong with my Keras multiple outputs coding, which causes a high loss comparing with the Sequential model. Please help me out which part of it is wrong.