Trouble of Pytorch 2.9.1 on Jetson Orin Nano Super

Robotics & Edge Computing Jetson Systems Jetson Orin Nano
,
1

Hi,

I am using Jetson Orin Nano with Jetpack 6.2, Python 3.10.12, CUDA 12.6. The system is just flashed using SDK manager.

I installed torch 2.9.1 from this link jp6/cu126 index

However, when I import torch, it appears this problem “ImportError: libcudss.so.0: cannot open shared object file: No such file or directory“.

When I try install previous version of torch from other user’s blog through the link of https://pypi.jetson-ai-lab.dev/jp6/cu126 , however it cannot be access now.

Please provide me any solution. Thank you.

Screenshot from 2026-02-02 21-10-02
Screenshot from 2026-02-02 21-10-021366×768 121 KB

3

*** Please note that this reply is generated by LLM automatically ***

Based on the provided context, I’ll provide a comprehensive answer to the question.

The issue you’re experiencing with importing torch on your Jetson Orin Nano device with Jetpack 6.2, Python 3.10.12, and CUDA 12.6 is likely due to a compatibility issue between the torch version and the CUDA version.

The torch version you installed (1.9.1) might not be compatible with CUDA 12.6. You can try installing a different version of torch that is compatible with CUDA 12.6.

Here are a few possible solutions:

  1. Install a compatible torch version: You can try installing a different version of torch that is compatible with CUDA 12.6. You can check the compatibility matrix on the PyTorch website to find a compatible version.

  2. Install torch from the NVIDIA repository: You can try installing torch from the NVIDIA repository, which is specifically designed for Jetson devices. You can use the following command to install torch:

sudo apt-get install libtorch-cpu

   or
   ```bash
sudo apt-get install libtorch-cuda

depending on whether you want to use CPU or CUDA acceleration.

  1. Build torch from source: If you’re unable to find a compatible pre-built torch version, you can try building torch from source. This will allow you to build torch with the specific CUDA version you’re using. You can follow the instructions on the PyTorch website to build torch from source.

  2. Check the CUDA version: Make sure that the CUDA version you’re using is compatible with the torch version you’re trying to install. You can check the CUDA version using the following command:

nvcc --version

   If the CUDA version is not compatible, you may need to update or downgrade CUDA to a compatible version.

5. **Check the Python version**: Make sure that the Python version you're using is compatible with the torch version you're trying to install. You can check the Python version using the following command:
   ```bash
python --version

If the Python version is not compatible, you may need to update or downgrade Python to a compatible version.

In comparison to the previous version of torch you tried to install from the other user’s blog, the version you installed from the official PyTorch repository (1.9.1) might not be compatible with CUDA 12.6. You can try installing a different version of torch that is compatible with CUDA 12.6, or try building torch from source to ensure compatibility.

I hope this helps you resolve the issue with importing torch on your Jetson Orin Nano device.

*** Please give the thumbs up if you get answers you like. Or provide feedback to help us improve the answer. ***

4

I have checked all these. Please note that I am using the latest version of Torch provided by you all in the pypi jetson ai lab. Please provide me the solution. Thank you.

7

Hi,

Please download the requirement packages.

Thanks

8

Thank you for your reply. I managed to try on a different way. I think both method can be worked, just in case someone facing the same problem with me.

<>

I using the command provided in this link. https://forums.developer.nvidia.com/t/website-for-jetpack-download-not-working-https-pypi-jetson-ai-lab-dev-jp6-cu126/348887/6

pip uninstall torch torchvision torchaudio

pip install https://pypi.jetson-ai-lab.io/jp6/cu126/+f/62a/1beee9f2f1470/torch-2.8.0-cp310-cp310-linux_aarch64.whl

pip install https://pypi.jetson-ai-lab.io/jp6/cu126/+f/907/c4c1933789645/torchvision-0.23.0-cp310-cp310-linux_aarch64.whl

pip install https://pypi.jetson-ai-lab.io/jp6/cu126/+f/81a/775c8af36ac85/torchaudio-2.8.0-cp310-cp310-linux_aarch64.whl

By using torch version 2.8, torchvision 0.23.0 and torchaudio 2.8.0, I managed to use the CUDA. Also, the tensor can be found.

True
0
Orin
tensor([[0.9745, 0.6260, 0.0240],
[0.8729, 0.9838, 0.3861],
[0.6786, 0.9686, 0.8946]], device=‘cuda:0’)

Screenshot from 2026-02-03 14-39-03
Screenshot from 2026-02-03 14-39-031366×768 109 KB

9

Hi,

Thank you for your feedback.

Since the command and result look good, please mark the solution as ✅ so we can close the topic.

Thanks