why discriminator is designed like this?

[YunLord]

 loss_D_real = criterion_GAN(pred_real, torch.ones_like(pred_real));i think this should be      loss_D_real = criterion_GAN(pred_real, torch.zeros_like(pred_real))

[FloCF]

Not sure why you think that. At least that is the usual way to train GANs, check out https://pytorch.org/tutorials/beginner/dcgan_faces_tutorial.html