Need PyTorch tool base on python 3.8 version on JetPack 4.6.4 with cuda 10.2

I want a pytorch version run on jetpack 4.6.4 with cuda 10.2.
Because I am running the cellpose environment, which requires a minimum version of Python 3.8, I need to recompile a compatible PyTorch to enable GPU acceleration.

I noticed that someone in the forum has already done the same thing, but unfortunately, the installation package links they posted have expired.
Jetson Nano Pytorch Wheel files for latest Python 3.8/3.9 versions with CUDA 10.2 support - Jetson & Embedded Systems / Jetson Nano - NVIDIA Developer Forums

I am trying to compile according to this guide. But this is a method based on Python 3.6, and I am concerned that this guide may be outdated.:
PyTorch for Jetson - Jetson & Embedded Systems / Announcements - NVIDIA Developer Forums

What should I pay attention to? Especially some problems that others have encountered, I want to get the correct way from you.


We only build PyTorch with the default Python version.
For JetPack 4.6.4, it is Python 3.6.

If you need the Python 3.8 package on JetPack 4.6.4, please compile it from the source directly.
You can find some instructions in the below topic:


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