PMLE - Collaborating Within and Across Teams to Manage Data and Models (16% of the exam) - Section 2.2

Prototype models in notebooks using Agent Platform Workbench and Colab Enterprise, applying collaboration and security best practices with common frameworks such as PyTorch, scikit-learn, and JAX, and using Model Garden models.

Prototype models in Agent Platform Workbench and Colab Enterprise using frameworks such as PyTorch, scikit-learn, and JAX, applying collaboration and access-control best practices suited to shared notebook environments. Recognise when to call a Model Garden foundation model from a notebook versus training a custom model from scratch.

Agent Platform WorkbenchColab EnterprisePyTorch and JAXModel Garden

Practice question for this objective

Free sampleCollaborating Within and Across Teams to Manage Data and Modelsmedium

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

Identify Model Garden as the catalogue for discovering, evaluating and deploying pre-trained models that teams then call or fine-tune from notebook frameworks. Model Garden centralises discovery, evaluation and deployment of first-party, open and partner pre-trained models, giving teams a vetted starting point they can then integrate, call or fine-tune from frameworks such as PyTorch or JAX inside a notebook, which the other listed components do not provide.

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.

See more PMLE practice questions, answers explained.

Exam traps in Collaborating Within and Across Teams to Manage Data and Models

Answers that look right on this material and are not. Each one is a distractor from a different question in the PMLE bank for this domain.

Copy the model's public demo URL from an external hosting site and send the notebook's sample inputs to it for quick comparison.
Why it is wrong: It seems like the fastest way to try the model, but sending data to an external hosting endpoint leaves the project boundary and violates the stated security policy.
Colab Enterprise runs only on the public colab.research.google.com domain and cannot inherit a project's IAM or VPC Service Controls boundary.
Why it is wrong: This is tempting because the consumer Colab product is public and weakly governed, but the Enterprise variant runs within the customer's Google Cloud project and does honour IAM and VPC Service Controls, so the claim is wrong.
Download the model weights from a public community hub over the open internet into the notebook and load them locally, since prototyping is temporary.
Why it is wrong: Pulling weights from an arbitrary public hub introduces an unvetted external source and supply-chain risk, which directly conflicts with the policy of keeping access inside the project boundary even for short-lived prototypes.

Examworthy is not affiliated with or endorsed by Google Cloud. Original, blueprint-aligned practice material only.