I can install PyTorch from the wheel from here:
https://developer.download.nvidia.com/compute/redist/jp/v51/pytorch/
which I found here:
Installing PyTorch for Jetson Platform - NVIDIA Docs
This is the version in the model zoo.
But, then I get errors like this:
raise ValueError(f"Unknown CUDA arch ({arch}) or GPU not supported")
ValueError: Unknown CUDA arch (8.7+PTX) or GPU not supported
This seems to be a know error:
CUDA 8.7+PTX isn’t in the list of supported arches · Issue #94438 · pytorch/pytorch (github.com)
If I instead use a docker container:
NVIDIA L4T ML | NVIDIA NGC
Then everything works OK. If I check the Pytorch version, I see its ‘2.0.0a0+ec3941ad.nv23.02’.
Is there a link to the correct Pytorch wheel somewhere, so I don’t have to compile it myself?