ImportError: cannot import name 'zero_gradients' from 'advertorch.attacks.utils'

[abdalimran]

Environment:

1. Python 3.10.13
2. advertorch-0.2.4
3. torch 2.2.1

Installed advertorch using pip install --upgrade git+https://github.com/BorealisAI/advertorch.git
While doing from advertorch.attacks import LinfPGDAttack, I got the following error:

---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
Cell In[2], [line 1](vscode-notebook-cell:?execution_count=2&line=1)
----> [1](vscode-notebook-cell:?execution_count=2&line=1) from advertorch.attacks import LinfPGDAttack

File [/opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/__init__.py:14](https://file+.vscode-resource.vscode-cdn.net/opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/__init__.py:14)
     [11](https://file+.vscode-resource.vscode-cdn.net/opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/__init__.py:11) with open(os.path.join(os.path.dirname(__file__), 'VERSION')) as f:
     [12](https://file+.vscode-resource.vscode-cdn.net/opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/__init__.py:12)     __version__ = f.read().strip()
---> [14](https://file+.vscode-resource.vscode-cdn.net/opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/__init__.py:14) from . import attacks  # noqa: F401
     [15](https://file+.vscode-resource.vscode-cdn.net/opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/__init__.py:15) from . import defenses

File [/opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/attacks/__init__.py:48](https://file+.vscode-resource.vscode-cdn.net/opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/attacks/__init__.py:48)
     [46](https://file+.vscode-resource.vscode-cdn.net/opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/attacks/__init__.py:46) from .spsa import LinfSPSAAttack
     [47](https://file+.vscode-resource.vscode-cdn.net/opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/attacks/__init__.py:47) from .fast_adaptive_boundary import FABAttack
---> [48](https://file+.vscode-resource.vscode-cdn.net/opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/attacks/__init__.py:48) from .fast_adaptive_boundary import LinfFABAttack
     [49](https://file+.vscode-resource.vscode-cdn.net/opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/attacks/__init__.py:49) from .fast_adaptive_boundary import L2FABAttack
     [50](https://file+.vscode-resource.vscode-cdn.net/opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/attacks/__init__.py:50) from .fast_adaptive_boundary import L1FABAttack

File [/opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/attacks/fast_adaptive_boundary.py:22](https://file+.vscode-resource.vscode-cdn.net/opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/attacks/fast_adaptive_boundary.py:22)
     [19](https://file+.vscode-resource.vscode-cdn.net/opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/attacks/fast_adaptive_boundary.py:19) except ImportError:
     [20](https://file+.vscode-resource.vscode-cdn.net/opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/attacks/fast_adaptive_boundary.py:20)     from advertorch.utils import torch_flip as flip
---> [22](https://file+.vscode-resource.vscode-cdn.net/opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/attacks/fast_adaptive_boundary.py:22) from advertorch.utils import replicate_input
     [24](https://file+.vscode-resource.vscode-cdn.net/opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/attacks/fast_adaptive_boundary.py:24) from .base import Attack
     [25](https://file+.vscode-resource.vscode-cdn.net/opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/attacks/fast_adaptive_boundary.py:25) from .base import LabelMixin

ImportError: cannot import name 'zero_gradients' from 'advertorch.attacks.utils' (/opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/attacks/utils.py)

[ocean1229-max]

Solution:
Change ~/anaconda3/envs/d2l/lib/python3.7/site-packages/advertorch/attacks/fast_adaptive_boundary.py(or /opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/attacks/utils.py, )
Delete "from torch.autograd.gradcheck import zero_gradients" and add this:
def zero_gradients(x):
if isinstance(x, torch.Tensor):
if x.grad is not None:
x.grad.detach_()
x.grad.zero_()
elif isinstance(x, collections.abc.Iterable):
for elem in x:
zero_gradients(elem)
this may help you

[jonggyujang0123]

Solution: Change ~/anaconda3/envs/d2l/lib/python3.7/site-packages/advertorch/attacks/fast_adaptive_boundary.py(or /opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/attacks/utils.py, ) Delete "from torch.autograd.gradcheck import zero_gradients" and add this: def zero_gradients(x): if isinstance(x, torch.Tensor): if x.grad is not None: x.grad.detach_() x.grad.zero_() elif isinstance(x, collections.abc.Iterable): for elem in x: zero_gradients(elem) this may help you

Thank you for this solution.

For more help, I converted this into a block code version.

1. vi ~/anaconda3/envs/[YOUR-ENV]/lib/python3.9/site-packages/advertorch/attacks/fast_adaptive_boundary.py
2. Remove Line 14 from torch.autograd.gradcheck import zero_gradients
3. Add the following function

def zero_gradients(x):
    if isinstance(x, torch.Tensor):
        if x.grad is not None:
            x.grad.detach_()
            x.grad.data.zero_()
    elif isinstance(x, container_abcs.Iterable):
        for elem in x:
            zero_gradients(elem)

[Oceanshy12-YANG]

Solution: Change ~/anaconda3/envs/d2l/lib/python3.7/site-packages/advertorch/attacks/fast_adaptive_boundary.py(or /opt/miniconda3/envs/deeplearning/lib/python3.10/site-packages/advertorch/attacks/utils.py, ) Delete "from torch.autograd.gradcheck import zero_gradients" and add this: def zero_gradients(x): if isinstance(x, torch.Tensor): if x.grad is not None: x.grad.detach_() x.grad.zero_() elif isinstance(x, collections.abc.Iterable): for elem in x: zero_gradients(elem) this may help you

Thank you for this solution.

For more help, I converted this into a block code version.

1. vi ~/anaconda3/envs/[YOUR-ENV]/lib/python3.9/site-packages/advertorch/attacks/fast_adaptive_boundary.py
2. Remove Line 14 from torch.autograd.gradcheck import zero_gradients
3. Add the following function

def zero_gradients(x):
    if isinstance(x, torch.Tensor):
        if x.grad is not None:
            x.grad.detach_()
            x.grad.data.zero_()
    elif isinstance(x, container_abcs.Iterable):
        for elem in x:
            zero_gradients(elem)

I'm glad to be able to help you. I also encountered this problem yesterday, and then I found this solution, and then solved it through this solution. You can read the original text of the Website: https://blog.csdn.net/TaoLinna/article/details/137058509