ImportError: cannot import name 'zero_gradients' from 'torch.autograd.gradcheck'

[ashok-arjun]

This issue exists because this line:
from advertorch.attacks.utils import zero_gradients

which refers to the zero_gradients function in PyTorch is obsolete.

This is because zero_gradients is removed from PyTorch 1.9. It was present until PyTorch 1.7.

[masoudhashemi]

This issue is fixed in the code. It may have not been updated in the pip repo. Please re-install advertorch from the code (pip install -e .). sorry for the inconvenience.

[LostOxygen]

Ah, that worked. Thanks!

[ashok-arjun]

Got it, thanks! I will still keep this issue open as it exists when install using normal pip.

Feel free to close it,

[BetterZhuang]

This issue appeared again.
from torch.autograd.gradcheck import zero_gradients ImportError: cannot import name 'zero_gradients' from 'torch.autograd.gradcheck' (C:\Users\pc\AppData\Roaming\Python\Python37\site-packages\torch\autograd\gradcheck.py)

[CHENBIN99]

This issue appeared again at pytorch 1.10
ImportError: cannot import name 'zero_gradients' from 'torch.autograd.gradcheck'

[fam4r]

This solved for me in Colab:

!pip install --upgrade git+https://github.com/BorealisAI/advertorch.git

@masoudhashemi @gwding can someone please update the pip package?

[anukriti15]

The issue still persists in version 1.12

[larksudo]

still has this problem, can anyone give any help?

[waite116]

The error seems to have changed a bit from the start, but still seems to be a persistent issue with zero_gradients:

cannot import name 'zero_gradients' from 'advertorch.attacks.utils' (/usr/local/lib/python3.9/dist-packages/advertorch/attacks/utils.py)

[lzb1616]

可以通过下面这个方法解决：
from typing import Collection
import torch
import torchvision
from torchvision import datasets, transforms
from torch.autograd import Variable

from torch.autograd.gradcheck import zero_gradients 这句话运行不了，改为下面这段函数

import torch.utils.data.dataloader as Data
import torch.nn as nn
from torchvision import models
import numpy as np
import cv2

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

[lzb1616]

`from typing import Collection
import torch
import torchvision
from torchvision import datasets, transforms
from torch.autograd import Variable

from torch.autograd.gradcheck import zero_gradients这句话运行不了，改为下面这段函数

import torch.utils.data.dataloader as Data
import torch.nn as nn
from torchvision import models
import numpy as np
import cv2

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

[linhanbin0453]

`from typing import Collection import torch import torchvision from torchvision import datasets, transforms from torch.autograd import Variable

from torch.autograd.gradcheck import zero_gradients这句话运行不了，改为下面这段函数

import torch.utils.data.dataloader as Data import torch.nn as nn from torchvision import models import numpy as np import cv2

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

don't know what the cv2 refers