Request: torchaudio / torchvision Wheel for JetPack 6.2.1 + torch 2.5.0a0 (Jetson AGX Orin)

Robotics & Edge Computing Jetson Systems Jetson Orin NX
1

Request: torchaudio / torchvision Wheel for JetPack 6.2.1 + torch 2.5.0a0 (Jetson AGX Orin)

Platform:

  • Jetson AGX Orin
  • JetPack 6.2.1
  • Python 3.10
  • CUDA 11.x (from JetPack 6.2.1)

PyTorch Version:
torch-2.5.0a0+872d972e41.nv24.08 (NVIDIA official wheel for JetPack 6.2.1, installed from local .whl)

torchaudio Version(s) Attempted:

  • torchaudio-2.3.0+952ea74-cp310-cp310-linux_aarch64.whl (built from source and from community wheels)
  • torchaudio built from source (matching torch version, with and without CUDA, with all recommended flags)

Errors Encountered

  1. ImportError / OSError (ABI mismatch):
    When running:
    import torch
    import torchaudio

OSError: /home/neo/.local/lib/python3.10/site-packages/torchaudio/lib/libtorchaudio.so: undefined symbol: _ZN3c1010Dispatcher17runRecordFunctionERN2at14RecordFunctionESt17reference_wrapperIKNS_14FunctionSchemaEENS_11DispatchKeyE

This occurs with both prebuilt wheels and source builds.
All builds are done in a clean environment, with only the NVIDIA torch wheel installed.

  1. Build Issues:
  • Building torchaudio from source (as per official Jetson instructions) completes, but the resulting wheel always fails to import with the above ABI error.
  • All dependencies (ffmpeg, CUDA, etc.) are present and working.
  1. No System or PyPI torch/torchaudio present:
  • Only user site-packages contains torch and torchaudio.
  • No stray or conflicting installs.

What I Need
A torchaudio wheel (or build recipe) that is ABI-compatible with NVIDIA’s torch-2.5.0a0+872d972e41.nv24.08 for JetPack 6.2.1 (Python 3.10, aarch64).

Or, official guidance from NVIDIA on how to build torchaudio for this environment.

What I’ve Tried

  • All community and official wheels (PyPI, Qengineering, Jetson Zoo, GitHub) — none are compatible.
  • Building from source with and without CUDA, with all recommended flags and environment variables.
  • Full environment audit and clean install.

If you have a working wheel or build instructions for this exact setup, please share.

Edit
So I seem to have torch audio installed now, but have a torchvision issue:

Is there a torchvision wheel that’s compatible with:

  • PyTorch 2.5.0a0+872d972e41.nv24.08
  • JetPack 6.2.1
  • CUDA 12.6
  • cuDNN 9.3

Solution needed:

  • A torchvision build that’s exactly compatible with your PyTorch 2.5.0a0 version
  • Either an NVIDIA wheel or a source build from the correct branch/tag that matches your PyTorch build

Or complete package / resource is available somewhere?

I’ve tried https://pypi.jetson-ai-lab.dev but seems down?

Thank you.

3

Hi,

We have changed the link to https://pypi.jetson-ai-lab.io.
Please check the package in the above link.

Thanks.

4

Hi, Thank you for the reply.
I’m almost there, but https://pypi.jetson-ai-lab.io/jp6/cu126/torch/2.8.0 link keeps timing out.
I’ve tried manually downloading using various different browsers and wget, it seems there is an issue there.

seems gone down now also 502 Bad Gateway

5

Hi,

Could you try it again?
We can download it without issue today.

Thanks.

1 Like
7

Hello, I am having the same predicament, trying to install the same software on the same product today (2025-08-01). I confirmed on two separate devices on two separate networks that access to https://pypi.jetson-ai-lab.io ends in 502 Bad Gateway. Also, I must say that as a user, having three pypi indexes is pretty confusing:

I would greatly appreciate any working solution, as for now I can’t deploy on the device.

8

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.