Nvidia Xavier Dev Board Jetpack 5.1.2 torchvision installation

[mfatihaydogdu]

Hello

I am trying to generate a python 3.8 environment containing torch, torchvision and torchaudio on Nvidia AGX Xavier with Jetpack 5.1.2.

The proper torch version is
https://developer.download.nvidia.cn/compute/redist/jp/v512/pytorch/torch-2.1.0a0+41361538.nv23.06-cp38-cp38-linux_aarch64.whl as stated in here.

But how can I install torchvision and torchaudio compatible with this .whl file?

We have containers for torch, torchvision and torchaudio but what about whl files?

[mfatihaydogdu]

Can I use the .whl’s below???

https://download.pytorch.org/whl/torchvision/
https://download.pytorch.org/whl/torchaudio/

[jan.devera]

yes noticed this too… its confusing… in JetPack 6 its all listed…

[AastaLLL]

Hi,

For JetPack 5, we only have a prebuilt PyTorch package.
Please build Torchaudio and Torchvision from the source to get the .whl package.

Below is the build script for your reference:
Torchvision: jetson-containers/packages/pytorch/torchvision/build.sh at master · dusty-nv/jetson-containers · GitHub
Torchaudio: jetson-containers/packages/pytorch/torchaudio/build.sh at master · dusty-nv/jetson-containers · GitHub

Thanks.

[mfatihaydogdu]

Hello

Thank you for the answers.

What about the .whl’s in

https://download.pytorch.org/whl/torchvision/
https://download.pytorch.org/whl/torchaudio/

Are the .whl’s compatible with Jetpack 5.1.2?

For our own builts, is it OK to select the torchvision and torchaudio versions with respect to the compatibility matrices below?

torchvision

https://pytorch.org/audio/stable/installation.html

Best Regards

[AastaLLL]

Hi,

You can give it a try or check with the package provided to see if these files can work on the Jetson environment.

But in general, these files only include CPU implement and no CUDA support.

Thanks.