Look at Discriminator loss, it is constant. This means that it maybe overfitted or predicts just one class (for ex. all fake). It leads to problems with generator loss etc.

There are plenty of problems training GANs. Try to look at Discriminator preds at first.. Also this result is very data-dependent. Let me know about discr. preds :)

And sometimes you need just restart training process and everything will be fine. So, also, try to restart training with different optimizer learning rates. LR is very important here!