PyTorch 2.8 wheel for Jetpack 6.2

[jetson.lover]

TLDR: How can I install the latest version of PyTorch without compiling for the Jetson Orin Nano

Details:

On the Jetson Orin Nano (upgraded to the latest firmware) I want to install PyTorch 2.8. According to the PyTorch for Jetson Platform documentation, 2.8.0a0+5228986c39 should be compatible with Jetpack 6.2.

The latest wheel is PyTorch 2.3.0 for Jetpack 6 on this post. @dusty_nv mentions that future wheels are available at Installing PyTorch for Jetson Platform - NVIDIA Docs. There, it is stated that you can enter the version based on the compatibility matrix and obtain the wheels, but alas there is no offerings for Jetpack 6.2 as seen on this page.

Alternatively, on this post, it is mentioned that we can get the wheels from https://pypi.jetson-ai-lab.dev/jp6/cu126. This link does not work.

From the GitHub - dusty-nv/jetson-containers: Machine Learning Containers for NVIDIA Jetson and JetPack-L4T repo, I understand that the domain .dev has an issue.

Tried using jp6/cu126 index instead, which leads to a 404.

[carolyuu]

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

Based on the provided context, it appears that you are trying to install PyTorch on a Jetson Orin Nano device that has been upgraded to the latest firmware and is running Jetpack 6.2. You have consulted the PyTorch for Jetson Platform documentation and found that the compatible version of PyTorch is 1.12.0a0+280a052, but you are unable to find a wheel for Jetpack 6.2.

After further research, you found a post that mentions that future wheels are available at the NVIDIA Docs page, but unfortunately, there is no offering for Jetpack 6.2. You also tried to access the wheels from the link provided in the post, but it does not work.

You then tried to use the jp6cu126 index instead, but it leads to a 404 error.

Here are a few potential solutions you could try:

1. Check the NVIDIA Docs page again for any updates on the availability of PyTorch wheels for Jetpack 6.2.
2. Try to contact the author of the post, dustynv, to see if they have any information on when the wheels for Jetpack 6.2 will be available.
3. Look for alternative sources for PyTorch wheels, such as the PyTorch website or other third-party repositories.
4. Consider using a different version of PyTorch that is compatible with Jetpack 6.2, if available.
5. If you are comfortable with building PyTorch from source, you could try to build it yourself using the instructions provided on the PyTorch website.

It’s worth noting that the Jetson Orin Nano is a relatively new device, and it’s possible that the PyTorch wheels for Jetpack 6.2 are not yet available. You may want to keep an eye on the NVIDIA Docs page and the PyTorch website for updates on the availability of PyTorch wheels for your device.

In comparison to other devices, the Jetson Orin Nano is a powerful device that is capable of running complex AI models, but it may require more effort to install and configure PyTorch due to its relatively new release. Other devices, such as the Jetson Xavier NX, may have more established support for PyTorch and other AI frameworks.

In terms of the compatibility matrix, it’s clear that PyTorch 1.12.0a0+280a052 is compatible with Jetpack 6.2, but the availability of wheels for this version is limited. You may want to consider using a different version of PyTorch that is compatible with Jetpack 6.2, if available.

Overall, while there may be some challenges to installing PyTorch on the Jetson Orin Nano, it’s clear that the device has a lot of potential for running complex AI models, and with some persistence and creativity, you should be able to get PyTorch up and running on your device.

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

[jetson.lover]

You have consulted the PyTorch for Jetson Platform documentation and found that the compatible version of PyTorch is 1.12.0a0+280a052, but you are unable to find a wheel for Jetpack 6.2.

No, not PyTorch 1.x. It is PyTorch 2.8. One of the reasons I am moving from the Jetson Nano to the Jetson Orin Nano is for the latest PyTorch support. 🙂

After further research, you found a post that mentions that future wheels are available at the NVIDIA Docs page, but unfortunately, there is no offering for Jetpack 6.2.

Yes, exactly.

Try to contact the author of the post, dustynv, to see if they have any information on when the wheels for Jetpack 6.2 will be available.

@dusty_nv could you or your colleagues kindly advise?

If you are comfortable with building PyTorch from source, you could try to build it yourself using the instructions provided on the PyTorch website.

I am new to the Jetson Orin Nano platform. Already, I have resolved other issues (such as the Snap app breaking in 2.7), pip refusing to install and jtop thinking that Jetpack is missing. Took me over 3 full days to get to where I am now. At this point, I just want things to work with minimal friction if possible. 🙏

Other devices, such as the Jetson Xavier NX, may have more established support for PyTorch and other AI frameworks.

I assume that PyTorch support is an essential feature of the Jetson Orin Nano platform and would not have to move to another device for this purpose. Given its support for Jetpack 6.2, which the Jetson Nano lacks.

[elsaco]

PyTorch-2.8 is available for JON from jp6/cu126 index

After downloading the wheel it installed okay.

>>> import torch
>>> print(torch.__version__)
2.8.0
>>> print(torch.cuda.is_available())
True

[AastaLLL]

Hi,

Would you mind testing the jp6/cu126 index again?
The server contains a PyTorch 2.8 prebuilt package.

Thanks.

[jetson.lover]

The website looks back up. Will test it out tomorrow and get back to you. Thanks!

[jetson.lover]

I followed the installation guide on PyTorch for Jetson, but substituted for torch-2.8.0 on the jp6/cu126 index page.

wget https://pypi.jetson-ai-lab.io/jp6/cu126/+f/62a/1beee9f2f1470/torch-2.8.0-cp310-cp310-linux_aarch64.whl#sha256=62a1beee9f2f147076a974d2942c90060c12771c94740830327cae705b2595fc
sudo apt-get install python3-pip libopenblas-base libopenmpi-dev libomp-dev
pip3 install 'Cython<3'
pip3 install numpy torch-2.8.0-cp310-cp310-linux_aarch64.whl

• ✅ Tested it first with the code sample provided by @elsaco by calling python3:
  
• ✅ Also tried the code snippet on the PyTorch for Jetson page for verifying installation
  
  

Success and thank you @AastaLLL and @elsaco !