I want user a base image that building by myself, but I can’t pull it when I create a project and using custom contaner. I was used privately docker registry that setuping by myself.
AI workbench send a message “got error: registry ‘test-docker.io’ is not supported”. Is that means only ‘docker.io’ & ‘nvcr.io’ supported now?
Hi. Great question.
If your containers are public, then any registry should work. However you will need to add image labels for Workbench to be able to build the container. See below.
For private containers, we only support the following registries:
- NVIDIA NGC
- GitHub
- GitLab.
If you move the container image to one of those, the authentication should go through as long as you’ve got it configured in Workbench.
But, the container needs the proper image labels on it for it to be interpreted and built.
You can see the image label specification here: Use Your Own Base Container - NVIDIA Docs
1 Like
For reference, here is a guide to building your own AI Workbench-compatible base containers from scratch.
Ultimately, these container image labels are there to help AI Workbench build out a new .project/spec.yaml file around your desired project container, which is the file that helps AI Workbench translate between a GitHub repository and an AI Workbench project.
If you are feeling more adventurous, you can also take an existing project, open its existing .project/spec.yaml file, and swap out the container image bits according to the reference page here.
Building from scratch is recommended for custom base container images you would like to re-use across many projects, while the latter approach can be done for more one-off, per-project use cases. Hope this helps!