Pulling from Google Artifact Registry with Fuzzball

This quick demo pulls software down from Google Artifact Registry, one of the many places Fuzzball can fetch containers. Fuzzball supports most artifact registries, public or private, including Docker Hub and NVIDIA NGC, using whatever credentials each requires. Because Fuzzball was designed to run behind air gaps, it also integrates with air-gapped container registries.

The container image in question is a Python 3.12 environment the presenter assembled with MPI tooling, Llama model tooling, and the standard PyTorch stack used for these models: torch, torchtune, tokenizers, and Transformers. It is a compact look at how AI workloads get their software into Fuzzball.

Key takeaways

• Fuzzball can pull containers from Google Artifact Registry and most other public or private registries, including Docker Hub and NGC.
• Fuzzball was designed to run behind air gaps and integrates with air-gapped container registries.
• The demo container bundles Python 3.12 with MPI tooling, Llama model tooling, and PyTorch packages like torch, torchtune, tokenizers, and Transformers.

Questions this video answers

Which container registries can Fuzzball pull from?

Fuzzball supports most artifact registries, including Google Artifact Registry as shown here, Docker Hub, NVIDIA NGC, and private registries, using the credentials you supply. Because it was built for air-gapped sites, it also works with any air-gapped container registry you operate internally.

This video is part of the Fuzzball playlist. Browse every CIQ video by product and topic.