示例#1
0
 def _clip_gradients(self, max_norm):
     nn.utils.clip_grad_norm_(master_params(self.optimizer), max_norm)
示例#2
0
 def _rescale_gradients(self, scale):
     for param in master_params(self.optimizer):
         if param.grad is not None:
             param.grad.data.mul_(scale)