def _anneal_lr()

in improved_diffusion/train_util.py [0:0]

• 7 lines of code
• 3 McCabe index (conditional complexity)

    def _anneal_lr(self):
        if not self.lr_anneal_steps:
            return
        frac_done = (self.step + self.resume_step) / self.lr_anneal_steps
        lr = self.lr * (1 - frac_done)
        for param_group in self.opt.param_groups:
            param_group["lr"] = lr