Looking for PyTorch compatible with cuda and jetpack 6.2

Robotics & Edge Computing Jetson & Embedded Systems Jetson Orin Nano
1

I installed pytorch from jp6/cu126 index

The issue is that the torch (torch-2.8.0-cp310-cp310-linux_aarch64.whl) is not available for cuda.

4

I had to build it.

8

Hi,

Please see the updated command below

$ pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cu126 --force-reinstall

9

Hi,

Sorry for the incorrect message above.
Please use the packages from the jetson-ai-lab instead:

$ pip3 install torch torchvision --index-url https://pypi.jetson-ai-lab.io/jp6/cu126 --force-reinstall

Thanks

10

thanks, in the end it worked, but I had to use the cu129 folder instead.

11

Hi,

Could you share more about the fix?
Suppose you should have CUDA 12.6 installed when setting up with JetPack 6.2.x.

So it’s expected to use the packages in jp6/cu126.
Thanks.

12

The pytorch file in the cu126 folder was only enabling cpu and not CUDA support. I was able to install without issue but after running the test it was saying that cuda was not supported. When I tried the same for the file in the cu129 folder, it worked immediatly, saying cuda ‘supported’

13

Hi,

Do you have CUDA 12.6 installed in your environment?
Since the package depends on the library, it might use CPU instead if the required library doesn’t exist.

Thanks.

14

Hi, I am trying to launch a ros2 package which needs CUDA compatible PyTorch. I think there is a requirements file which installed torch with pip, which obviously is CPU only and does not work on GPU.

I am using Jetson Orin Nano 8GB with JetPack 6.2
nvidia@localhost:~$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2024 NVIDIA Corporation
Built on Wed_Aug_14_10:14:07_PDT_2024
Cuda compilation tools, release 12.6, V12.6.68
Build cuda_12.6.r12.6/compiler.34714021_0

Can you please let me know how to install compatible version of PyTorch? It would be really helpful.

Thanks.

15

Hi,

The package from jetson-ai-lab should work.
But do you have any dependencies for CUDA 12.9 with the ros setting?
That’s maybe the reason you need the packages from CUDA 12.9?

Thanks.