Can these newly added keys be removed after the function profile is executed?

[newcomertzc]

My Python version: 3.8.12 / 3.8.13
My Pytorch version: 1.8.2 / 1.10.1

Run the following code:

from torchvision import models
from thop.profile import profile
import torch

net = models.resnet50()
x = torch.randn(1, 3, 224, 224)
oringinal_keys = net.state_dict().keys()

profile(net, (x,))
current_keys = net.state_dict().keys()

print(current_keys - oringinal_keys)

the output is as follows:

{'layer1.0.downsample.total_ops',
 'layer1.0.downsample.total_params',
 'layer1.0.total_ops',
 'layer1.0.total_params',
 'layer1.1.total_ops',
 'layer1.1.total_params',
 'layer1.2.total_ops',
 'layer1.2.total_params',
 'layer1.total_ops',
 'layer1.total_params',
 'layer2.0.downsample.total_ops',
 'layer2.0.downsample.total_params',
 'layer2.0.total_ops',
 'layer2.0.total_params',
 'layer2.1.total_ops',
 'layer2.1.total_params',
 'layer2.2.total_ops',
 'layer2.2.total_params',
 'layer2.3.total_ops',
 'layer2.3.total_params',
 'layer2.total_ops',
 'layer2.total_params',
 'layer3.0.downsample.total_ops',
 'layer3.0.downsample.total_params',
 'layer3.0.total_ops',
 'layer3.0.total_params',
 'layer3.1.total_ops',
 'layer3.1.total_params',
 'layer3.2.total_ops',
 'layer3.2.total_params',
 'layer3.3.total_ops',
 'layer3.3.total_params',
 'layer3.4.total_ops',
 'layer3.4.total_params',
 'layer3.5.total_ops',
 'layer3.5.total_params',
 'layer3.total_ops',
 'layer3.total_params',
 'layer4.0.downsample.total_ops',
 'layer4.0.downsample.total_params',
 'layer4.0.total_ops',
 'layer4.0.total_params',
 'layer4.1.total_ops',
 'layer4.1.total_params',
 'layer4.2.total_ops',
 'layer4.2.total_params',
 'layer4.total_ops',
 'layer4.total_params',
 'total_ops',
 'total_params'}

[Omar Moured]

Did you manage to find a way to remove the additional keys? Anyone from the authors to solve this issue, please!

[Mingzhe LV]

Did you manage to find a way to remove the additional keys? Anyone from the authors to solve this issue, please!

I think this issue has been solved by the authors, please update your version.