ipython下运行代码出现<module '__main__' (built-in)> is a built-in class错误

这个问题是我在做模型剪枝的时候遇到的

1. 先载入VGG16模型，更改全连接层，使输出单元个数与自己待处理的数据类别个数一致。
2. 冻住卷积层，只训练全连接层。学习率设为0.0001，momentum=0.9。
3. 当设定的epoch数目满足的时候，把模型保存，使用命令：

torch.save(model, "./cifar-10/5epochs_cifar10_vgg_model")

4 . 然后在ipython使用以下命令加载模型：

model = torch.load("/home/smiles/tsq/PyTorch/pytorch-pruning/cifar-10/5epochs_cifar10_vgg_model").cuda()

这个时候就出现错误：

<module '__main__' (built-in)> is a built-in class

关于这个问题：网上有很多人遇到。应该是一个Bug。
python社区有关于这个问题的讨论：https://bugs.python.org/issue12920
在cifar10_prune.py中写代码，显示出各层是否需要训练：

if __name__ == '__main__':    args = get_args()    if args.train:        # model = ModifiedVGG16Model().cuda() # use GPU and finetune         model = torch.load("/home/smiles/tsq/PyTorch/pytorch-pruning/cifar-10/10epochs_cifar10_vgg_model").cuda()        for param in model.features.parameters():            print "features's requires_grad is: ", param.requires_grad        for param in model.classifier.parameters():            print "classifier's requires_grad is: ", param.requires_grad        elif args.prune:        model = torch.load("./cifar-10/cifar10_vgg_model").cuda

然后在终端下输入命令：

python cifar10_prune.py --train --train_path ./cifar-10/ --test_path ./cifar-10/features's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falsefeatures's requires_grad is:  Falseclassifier's requires_grad is:  Trueclassifier's requires_grad is:  Trueclassifier's requires_grad is:  Trueclassifier's requires_grad is:  Trueclassifier's requires_grad is:  Trueclassifier's requires_grad is:  True

这说明，使用torch.save命令保存模型，除了把各层的参数保存了之外，还保存了各层参数是否需要进行训练这一设置，即requires_grad，当然还保存了其他，比如说网络结构，总之模型所有东西都保存了。