when to save model for inference

[ghost]

Hi, thank you for all these implementations,
sorry to bother you,
I am new to GAN,
I want to know when to save the models for inference since the losses of gan models are unstable,
Do I need to split training data for validation (different from testing, like cross-validation)?
Are there some indications like metrics I can use for stopping training?