GCP Professional Cloud Architect Practice Question

Your media company wants to launch a generative text-summarization feature on its global news portal within three months. Data scientists must start from a proven large-language model, fine-tune it with 50 000 proprietary articles, and expose a low-latency, highly available REST endpoint that multiple GKE clusters can call. The team prefers serverless operations, built-in experiment tracking, and IAM-based access control. Which Google Cloud approach best satisfies these needs?

Containerize a custom BERT model, deploy it on Cloud Run with autoscaling, and manage experiments through a separately hosted MLflow server.
Fine-tune an open-source transformer model on a self-managed TensorFlow cluster running on preemptible Compute Engine VMs behind an external HTTP(S) load balancer.
Select a foundation model from Vertex AI Model Garden, perform tuning with Vertex AI custom training, and deploy the resulting model to a Vertex AI online prediction endpoint secured by Cloud IAM.
Use BigQuery ML to train a text-summarization model and export it to Cloud Functions for real-time inference.

Report Issue

Answer Description

Vertex AI Model Garden provides curated foundation models that can be tuned with custom data using Vertex AI training pipelines. After tuning, the model can be deployed to a managed online prediction endpoint that automatically scales and delivers low-latency synchronous inference. The endpoint integrates with Vertex AI Experiments for tracking and is secured by Cloud IAM. Running a self-managed TensorFlow cluster on Compute Engine or GKE would require significant operational management and lacks integrated experiment tracking. BigQuery ML cannot export large language models directly to Cloud Functions for serving, and while Cloud Run supports GPUs, it does not offer built-in experiment tracking and still leaves infrastructure management to the team. Therefore, selecting and fine-tuning a foundation model from Vertex AI Model Garden and serving it with Vertex AI online prediction best meets the requirements.

Ask Bash

Bash is our AI bot, trained to help you pass your exam. AI Generated Content may display inaccurate information, always double-check anything important.