loss出现nan值,加上梯度的取值范围后,出现Tensor had NaN values 1、调低学习率为0.0001和0.00001后,依然出现Tensor had NaN values 2、查看模型中是否有除数为0的情况。范数为0