Torch模型导入

冻结param的3种方式

        for param in net.lstm.parameters():
            param.requires_grad = True

        net.lstm.requires_grad = True        # 无效
        net.lstm.state_dict()['weight_ih_l0'].requires_grad=False
		net.lstm.weight_ih_l0.requires_grad = False
        # dir(net.lstm) to validate attributes of lstm module

optimizer