Jetson containers building pip server hosting

I already messaged Dustin, but maybe other people could chime in:

I saw @dusty_nv using the custom pip server in the AI Lab meet. I’m wondering if self hosting that is a best solution. Are you hosting the server yourself? If I’m pulling packages from you directly, it will get bogged down pretty fast:

Collecting pyarrow>=12.0.0 (from datasets)
  Downloading (35.7 MB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╸━━━━━━━━━ 27.3/35.7 MB 21.8 kB/s eta 0:06:26
ERROR: Operation cancelled by user

I guess you are using website to point to your local server?

Could it be hosted on legit nvidia servers/architecture?

I guess Jetson developers aren’t that numerous group of people, but it would be nice if you could address this somehow (best by using Nvidia’s own servers with big bandwidth and availability)

@i_love_nvidia yes this is a prototype server hosted on my build farm here (like the container builds are), we hope to transition it to official hosting at some point. I have gigabit uplink and while I keep an eye on it, it barely gets used currently and I think should be fine for our uses while we try this out. I wonder if your bandwidth issues are more geographically related to the connect and the fact that it’s not on a CDN (which would require a whole other level of mirroring should this be deployed more widely).

Also you can try navigating your browser to that server address and downloading the pip wheels right from there if it improves the speed. And you can disable the use of this in jetson-containers, and it will fall back to building them from source.

1 Like

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