Error loading "\lib\site-packages\torch\lib\shm.dll" or one of its dependencie

22

I had the same issue (Windows 11 Pro, Cuda 12.4, Python 3.11.9)
I was using torch==2.4.1+cu124, torchvision==0.19.1+cu124, torchaudio==2.4.1+cu124.

Going back to torch==2.4.0+cu124, torchvision==0.19.0+cu124, torchaudio==2.4.0+cu124 fixed this for me. Install with:
“pip install torch==2.4.0 torchvision==0.19.0 torchaudio==2.4.0 --index-url https://download.pytorch.org/whl/cu124
Looking in indexes: https://download.pytorch.org/whl/cu124”

1 Like
23

ok i have had encountering similar errors. after trying different versions of torch, the follow works:

windows 11 / i9-14900HX / RTX4070
numpy 1.24
torch 2.2.2
torchvision 0.17.2
torchaudio 2.2.2
cuda 12.1

24

Thank you， your method worked for me too. My envirements are : windows 11, cuda 12.8, pytorch 2.2.0. This env can work.

25

Same issue with windows 10 and pythorch 2.6.0. I had to go back to pytorch 2.4.0 to solve it. CPU version.

26

Still an issue on Windows 11. Using Python 3.11.0. This addressed the issue:

pip install torch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 --index-url https://download.pytorch.org/whl/cpu

EDIT: Never mind. It installs now, but it throws the same error in runtime

27

Having this same issue:
Windows 11 / RTX 5080 / python 3.13.9 / torch 2.9.1+cu128 / torchvision 0.24.1+cu128

Has no one found a recent solution? I’ve tried following this video from a Reddit post but the info is outdated and it looks like these changes have already been made to init.py for more recent version of torch: https://www.youtube.com/watch?v=ca34C8ZUI0A

Need help

28

Maybe it can help someone.
I was just having the same issue, by running python -c "import torch” inside the terminal I also got the error OMP: Error #15: Initializing libiomp5md.dll, but found libiomp5md.dll already initialized.

I had first installed numpy from conda-forge, from what I uderstood that uses MKL and resulted in a conflict with torch crashing, but inside a jupyter notebook the reported error was only Error loading “\lib\site-packages\torch\lib\shm.dll” or one of its dependencies.

My fix was first installing torch via pip in a new clean enviroment, which automatically installs numpy with OpenBLAS, or installing the “nomkl” package before installing numpy from conda-forge forcing to install package versions without MKL.