The loss function is abnormal when the edge network is trained

[TT-mouse]

When training the edge network, I've seen generator and discriminator losses stay high, with generator losses reaching 100. Does anyone know why that is? Thank you very much!