Inside a dustynv/pytorch:2.7-r36.4.0-cu128-24.04 container, when trying to pip install any package, i’m getting errors after a few retries of Looking in indexes: https://pypi.jetson-ai-lab.dev/jp6/cu128, https://pypi.ngc.nvidia.com .
I see that this has happened before. Is it the same issue again ?
@sam244 This is the third outage of the year. If you rely your production on to it, you might want to consider alternatives. Hosting your own buildfarm might be much safer if your production depends on it. You can build many of the required packages from source for the Jetson. There are many helpful scripts here:
It’s 2.8. The individual readme files for the packages are not up to date. The information you are looking for is in this file:
There is also a docker container for PyTorch 2.7. If you have no other hardware specific dependencies you could use this docker and install all other package via a different pip index.
I just recently onboarded to the jetson index as it was super helpful to enable my CI/CD and I am now somewhat stuck as well.
I gave jetston-containers a shot but wasn’t able to work how to build CI/CD around it using ubuntu in the cloud. I got the feeling the toolkit is aimed for use directly on L4T environments which I don’t actually have ready access to.
Anyone have a guide on running a jetson-containers build farm in the cloud?
Since both are down, is there a backup to the backup? I assume now the only way to get torch and torchvision for jetpack 6.1 with cuda 12.6 is to build from source?
While both are down or in general for the future, is there a good guide or information on how to use a different server for pip or host your own compatible with jetson-containers