Prepare and test your skills
Prepare and test your skills
Free sample question, answered and explained below. The Professional Cloud Architect practice bank has 2,335 more like it. Try 10 free: instant score, no account.
An enterprise legal department wants to adapt an open-weights foundation model from Vertex AI Model Garden to draft case briefs adhering to strict firm-specific formatting rules and specialized legal terminology.
The organization has identified the following requirements and constraints:
Which model customization strategy should the cloud architect recommend?
Prompt engineering involves designing input prompts, context headers, system instructions, and few-shot input-output demonstrations at runtime to guide a foundation model's generation without modifying any underlying model parameters.
The organization had zero training data, needed to rapidly prototype without training pipelines, or had short context inputs with straightforward, flexible formatting guidelines that easily fit inside standard token limits.
Full fine-tuning is an adaptation approach in which all weights and layers of a pre-trained deep learning network are unlocked and updated via backpropagation against a custom training dataset.
The organization had hundreds of thousands of diverse domain-specific examples, required foundational shifts in core reasoning across vast new languages or modalities, and possessed unlimited compute and storage budgets to maintain bespoke foundation models.
Training from scratch involves randomly initializing model weights and training a deep neural network on massive, web-scale corpora (often trillions of tokens) using distributed GPU/TPU clusters managed by Vertex AI Custom Training.
The enterprise was building a completely proprietary architecture for an unrepresented modality or a novel language lacking any existing foundation model baseline in Model Garden.
Parameter-Efficient Fine-Tuning (PEFT), commonly implemented through Low-Rank Adaptation (LoRA) or adapter-based layers, is a domain adaptation technique that freezes the majority of a pre-trained foundation model's parameters and trains only a small set of auxiliary parameters (adapter weights). In Vertex AI Model Garden, adapter-based tuning allows organizations to tailor open-weights and foundation models to specialized domain vocabularies, reasoning styles, and task-specific response formats without retraining the underlying base architecture.
Compared to full fine-tuning or training from scratch, PEFT / LoRA provides the ideal balance of deep stylistic adaptation, low infrastructure overhead, and cost efficiency for medium-scale enterprise customization.
Answer them yourself and get your score the moment you finish. No account needed.
Full access includes
Keep the momentum going with these hand-picked practice scenarios
Last updated Validated