How do you combine the token-reconstruction loss and diffusion loss?

Thank you for your excellent job!
After reviewing your code I have a question that how you combine your two kinds of loss, I think it is needed for the training of the HART model.