A team wants to start prototyping from a strong pre-trained foundation rather than training a model from scratch in their notebooks. They plan to browse, evaluate and deploy first-party and selected open and partner models, then call or fine-tune them from PyTorch or JAX inside Colab Enterprise. Which Google Cloud capability is designed to be that catalogue and starting point for such models?
- AColab Enterprise runtime templates, which exist mainly to define a curated catalogue of pre-trained foundation models for browsing and deployment.
- BAgent Platform Workbench instances, whose core function is to act as the searchable catalogue from which first-party and partner models are discovered and deployed.
- CVPC Service Controls perimeters, which provide the catalogue and one-click deployment of pre-trained foundation models for notebook prototyping.
- DModel Garden, which lets teams discover, evaluate and deploy first-party, open and partner pre-trained models, then call or fine-tune them from frameworks such as PyTorch or JAX. Correct
Why A is wrong: This is tempting because runtime templates are a real Colab Enterprise feature, but they configure compute and environment settings, not a catalogue of models to evaluate and deploy, so they do not serve this purpose.
Why B is wrong: This is tempting because Workbench is part of the same platform, but it provides notebook compute environments rather than a model catalogue, so it is not the discovery and deployment surface described.
Why C is wrong: This is tempting because security controls surround these workflows, but VPC Service Controls enforce a data-exfiltration boundary and do not catalogue or deploy models, so they are unrelated to model discovery.
Why D is correct: Correct: Model Garden is the curated catalogue for discovering, evaluating and deploying first-party, open and partner models, after which teams can call or fine-tune them from frameworks like PyTorch or JAX, matching the team's goal.