In order to access and use the NGC Deep Learning containers, you must have an NGC account. Only NGC account holders are authorized to run the containers.
When you are ready to pull the NGC containers to your computer, you first do a ‘docker login nvcr.io’, give the username $oauthtoken and paste in your NGC APIKey as the password. Once you are logged in to nvcr.io, the docker pull command will work.
If I run “docker pull nvcr.io/nvidia/pytorch:18.11-py3” right now, I get about 1/3 of the layers downloaded and then it gives me “unauthorized: authentication required” and the pull stops. I tried twice and it seems to get to a different spot each time; maybe that has to do with threading/sessions.
I don’t know if showing the layers helps, but here it is anyhow:
If I run “docker pull nvcr.io/nvidia/pytorch:18.11-py3” right now, I get about 1/3 of the layers downloaded and then it gives me “unauthorized: authentication required” and the pull stops.
This symptom sounds a bit like something I’ve see on other docker registries when a push or a pull is taking a long time – there’s an authorization token that can time out during the pull (or push). I haven’t seen that on nvcr.io myself, though. Can you give us an idea of how much time has elapsed after docker pull starts until the error is returned? time docker pull nvcr.io/nvidia/pytorch:18.11-py3 would probably be sufficient to test that. Also, are you able to tell us where you’re pulling from?
I am encoutering the same issue, here is the output
time sudo docker pull nvcr.io/nvidia/pytorch:18.12-py3
unauthorized: authentication required
I’m having trouble pulling your tenrorrt and tensorflow images (similar to this thread)
I login properly (i get “login suceedeed”). But around half of the download i get “authentication required”. It would seem that oauth session expires too early causing log out and interrupting my docker pull.
However, after reading your statement “You no longer need to be logged in to pull tensorrt or tensorflow from nvcr.io” I expected this to run without an error. Am I misunderstanding something or doing something wrong?