# AttributeError: module ‘torch.cuda’ has no attribute ‘amp’

**URL:** <https://forums.developer.nvidia.com/t/attributeerror-module-torch-cuda-has-no-attribute-amp/198112>\
**Category:** CUDA Programming and Performance\
**Tags:** cuda\
**Created:** [December 15, 2021, 1:54pm UTC](https://forums.developer.nvidia.com/t/attributeerror-module-torch-cuda-has-no-attribute-amp/198112 "2021-12-15T13:54:50Z")\
**Posts on this page:** 4\
**Page:** 1

<div class="post-metadata">

**Author:** ![ssuresh.mca](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@ssuresh.mca](https://forums.developer.nvidia.com/u/ssuresh.mca)\
**Post date:** [December 15, 2021, 1:54pm UTC](https://forums.developer.nvidia.com/t/attributeerror-module-torch-cuda-has-no-attribute-amp/198112/1 "2021-12-15T13:54:50Z")

</div>

Issue : AttributeError: module ‘torch.cuda’ has no attribute ‘amp’

Traceback (most recent call last):  
File “tools/train\_net.py”, line 15, in  
from maskrcnn\_benchmark.data import make\_data\_loader  
File “/miniconda3/lib/python3.7/site-packages/maskrcnn\_benchmark/data/ **init**.py”, line 2, in  
from .build import make\_data\_loader  
File “/miniconda3/lib/python3.7/site-packages/maskrcnn\_benchmark/data/build.py”, line 11, in  
from . import datasets as D  
File “/miniconda3/lib/python3.7/site-packages/maskrcnn\_benchmark/data/datasets/ **init**.py”, line 3, in  
from .coco import COCODataset  
File “/miniconda3/lib/python3.7/site-packages/maskrcnn\_benchmark/data/datasets/coco.py”, line 6, in  
from maskrcnn\_benchmark.structures.segmentation\_mask import SegmentationMask  
File “/miniconda3/lib/python3.7/site-packages/maskrcnn\_benchmark/structures/segmentation\_mask.py”, line 5, in  
from maskrcnn\_benchmark.layers.misc import interpolate  
File “/miniconda3/lib/python3.7/site-packages/maskrcnn\_benchmark/layers/ **init**.py”, line 10, in  
from .nms import nms  
File “/miniconda3/lib/python3.7/site-packages/maskrcnn\_benchmark/layers/nms.py”, line 5, in  
from apex import amp  
File “/miniconda3/lib/python3.7/site-packages/apex/ **init**.py”, line 23, in  
from . import transformer  
File “/miniconda3/lib/python3.7/site-packages/apex/transformer/ **init**.py”, line 1, in  
from apex.transformer import amp  
File “/miniconda3/lib/python3.7/site-packages/apex/transformer/amp/ **init**.py”, line 1, in  
from apex.transformer.amp.grad\_scaler import GradScaler  
File “/miniconda3/lib/python3.7/site-packages/apex/transformer/amp/grad\_scaler.py”, line 8, in  
class GradScaler(torch.cuda.amp.GradScaler):  
AttributeError: module ‘torch.cuda’ has no attribute ‘amp’

Environment:  
GPU : RTX 8000  
CUDA: 10.0  
Pytorch 1.0.1  
torchvision 0.2.2  
apex 0.1

Question:

Same application is working fine in Tesla T4 CUDA10.0 directly on the same software environment at the GPU server (without using docker image)

If i use RTX 8000 CUDA 10.0 on the same software environment, I am getting error " **AttributeError: module ‘torch.cuda’ has no attribute ‘amp’**"

What is the issue? How can i solve it

---

<div class="post-metadata">

**Author:** ![Robert\_Crovella](https://sea2.discourse-cdn.com/nvidia/user_avatar/forums.developer.nvidia.com/robert_crovella/32/14043_2.png) [@Robert\_Crovella](https://forums.developer.nvidia.com/u/Robert_Crovella)\
**Post date:** [December 15, 2021, 6:20pm UTC](https://forums.developer.nvidia.com/t/attributeerror-module-torch-cuda-has-no-attribute-amp/198112/2 "2021-12-15T18:20:40Z")

</div>

This subforum is not for support of pytorch. You might get better help posting on the pytorch discussion forum. Searching there, there are similar reports like [this one](https://discuss.pytorch.org/t/module-torch-cuda-has-no-attribute-amp-with-torch-1-6-0/92360) that may give some hints.

---

<div class="post-metadata">

**Author:** ![njuffa](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@njuffa](https://forums.developer.nvidia.com/u/njuffa)\
**Post date:** [December 15, 2021, 6:34pm UTC](https://forums.developer.nvidia.com/t/attributeerror-module-torch-cuda-has-no-attribute-amp/198112/3 "2021-12-15T18:34:06Z")

</div>

Outside my area of expertise. Existing discussions on this error message at the PyTorch forum and GitHub _appear to suggest_ upgrading to a PyTorch version \>= 1.5. As Robert Crovella already indicated, this is not the proper venue for PyTorch questions.

---

<div class="post-metadata">

**Author:** ![ali.nehrani1](https://sea2.discourse-cdn.com/nvidia/user_avatar/forums.developer.nvidia.com/ali.nehrani1/32/82039_2.png) [@ali.nehrani1](https://forums.developer.nvidia.com/u/ali.nehrani1)\
**Post date:** [September 13, 2022, 7:32pm UTC](https://forums.developer.nvidia.com/t/attributeerror-module-torch-cuda-has-no-attribute-amp/198112/4 "2022-09-13T19:32:55Z")

</div>

new version of torch (1.12 LTS) with cuda 11.6 is working fine for example