Closed terrainer closed 6 months ago
The generator and discriminator do not need to be using separate GradScalers. The gradients are cleared after the generator and discriminator steps, meaning there is no reason to separate them.
Good PR :+1:
The generator and discriminator do not need to be using separate GradScalers. The gradients are cleared after the generator and discriminator steps, meaning there is no reason to separate them.