RuntimeError: Couldn't load custom C++ ops. This can happen if your PyTorch and torchvision versions are incompatible, or if you had errors while com

Please help me to solve this If anyone have solution for this. My model is Yolov8 segmentation model
jetpack version: 5.1.2
torch : 2.1.0
torchvision : 0.16.1
python: 3.8.10

Hello @pardeshi.grishma07, which Jetson model are you using? Because then I can move your post into the dedicated Jetson forums.


1 Like

Thanks for your response my Jetson Model is NVIDIA Orin NX Developer Kit

I went ahead and moved this over for you.

1 Like

Hi @pardeshi.grishma07, did you compile torchvision like as shown in the Installation section of this topic?

It’s possible that if you pip installed other package (like YOLO ultralytics for example), it may have re-installed the torchvision from pip (which wasn’t built with CUDA enabled). So you may need to re-install torchvision again. You can also use the l4t-pytorch container which already has these packages installed and working.

1 Like

yes earlier I have installed PyTorch using this link only but still i face issue. .

Thank you

It’s possible that if you installed YOLOv8 Ultralytics from pip, it uninstalled the GPU-accelerated version of PyTorch that you installed previously - so I would double-check that. And if you continue facing issues getting PyTorch/torchvision properly installed and working, I would recommend trying the l4t-pytorch container:

Yes thanks for this solution It works now code is running. . Thank you