pytorch - 💡(How to fix) Fix 2026.04.24 Pytorch nightly wheel for Windows is non functional (LE: cuDNN DLLs are missing, manually adding them solves the issue) [1 comments, 1 participants]

D:\Program Files\Microsoft Visual Studio\18\Insiders>python Python 3.14.4 (main, Apr 17 2026, 20:27:01) [MSC v.1944 64 bit (AMD64)] on win32 Type "help", "copyright", "credits" or "license" for more information.

import torch Traceback (most recent call last): File "<python-input-0>", line 1, in <module> import torch File "F:\Python314\install\Lib\site-packages\torch_init_.py", line 287, in <module> load_dll_libraries() ~~~~~~~~~~~~~~~~~~~^^ File "F:\Python314\install\Lib\site-packages\torch_init.py", line 283, in _load_dll_libraries raise err OSError: [WinError 126] The specified module could not be found. Error loading "F:\Python314\install\Lib\site-packages\torch\lib\shm.dll" or one of its dependencies. print(torch.version) Traceback (most recent call last): File "<python-input-1>", line 1, in <module> print(torch.version) ^^^^^ NameError: name 'torch' is not defined

TL;DR

The issue is likely due to missing DLLs in the torch\lib folder, which can be resolved by reinstalling the previous version of the wheel that worked well.

Guidance

• The error message indicates that the shm.dll or one of its dependencies is missing, which is consistent with the observation that some DLLs, including cuDNN DLLs, are missing from the torch\lib folder.
• The size difference between the two wheel files suggests that the newer version may be incomplete or corrupted.
• Try reinstalling the previous version of the wheel (torch-2.13.0.dev20260423+cu132-cp314-cp314-win_amd64.whl) that worked well to see if the issue is resolved.
• Verify that all required DLLs are present in the torch\lib folder after installation.

Notes

The issue may be specific to the particular version of the wheel, and reinstalling the previous version may not be a long-term solution. Further investigation may be needed to determine the root cause of the issue.

Recommendation

Apply workaround: Reinstall the previous version of the wheel (torch-2.13.0.dev20260423+cu132-cp314-cp314-win_amd64.whl) that worked well, as it is likely to resolve the issue temporarily.