How to install torchvision with torch1.14.0 with CUDA 11.4

Hello,

I’ve an AGX Orin Dev kit with JetPack 5.1 (Rev 1).
I tried to install Torch and Torchvision working with Cuda 11.4 (which is the Cuda version available with Jetpack 5.1).

So following this link (Installing PyTorch for Jetson Platform) I was able to install torch-1.14.0a0+44dac51c.nv23.02-cp38-cp38-linux_aarch64, but I can not find the equivalent torchvision package.
Where can I find it ?

My aim is to be able to work with Yolo (from ultralytics) using CUDA and not CPU.

Thanks for your help

1 Like

Hi @dgotfroi ,
Usually if the packages are not readilly available is best just to build from source. And the bonus is that sometimes it gives an extra bump in performance.
Here is the source repo and according to them you just need to have Torch installed and do:

python setup.py install

Before doing so I suggest to enable jetson-clocks to speed things up and enable all cores.
There are some ways to do that, here is one using cmd, and another is using jtop.

Regards,
Andres
Embedded SW Engineer at RidgeRun
Contact us: support@ridgerun.com
Developers wiki: https://developer.ridgerun.com/
Website: www.ridgerun.com

2 Likes

Hi @andres.artavia
Thanks a lot ! It works perfectly !

could you please tell us the steps you followed to overcome this issue

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