Answer-first summary for fast verification
Answer: Deploy the model versions to a Vertex AI endpoint, leveraging the built-in traffic split feature to manage the distribution between versions., Combine the use of Vertex AI for model deployment and Dataflow for real-time prediction processing, ensuring minimal management overhead.
The BEST solution for managing the pipeline efficiently, while minimizing management overhead, is to deploy the model versions to a Vertex AI endpoint and use its built-in traffic split feature (Option C). This approach leverages a fully managed service for model deployment and traffic management, reducing the need for custom code or additional infrastructure management. Vertex AI simplifies the process of updating models and adjusting traffic splits, making it ideal for scenarios requiring frequent model iterations. Option E is also correct as it suggests a comprehensive approach that combines Vertex AI's deployment capabilities with Dataflow's real-time processing, ensuring efficiency and minimal management. However, the simplest and most straightforward solution is Option C, which directly addresses the requirement with the least complexity.
Author: LeetQuiz Editorial Team
Ultimate access to all questions.
No comments yet.
You are tasked with improving the sentiment analysis model for your company's social media page to ensure it remains accurate and up-to-date. The model processes real-time data from Pub/Sub using Dataflow for predictions. After each training iteration, the latest two versions of the model are kept live, with a traffic split favoring the newest version at an 80:20 ratio. Given the constraints of minimizing management overhead and ensuring seamless model updates, which of the following solutions is the BEST for managing this pipeline efficiently? Choose the correct option.
A
Implement a custom partitioning strategy in Dataflow using beam.Partition() to direct traffic to different Vertex AI endpoints based on the model version.
B
Utilize Cloud Run to host the model versions, configuring environment variables to manage the traffic split between revisions.
C
Deploy the model versions to a Vertex AI endpoint, leveraging the built-in traffic split feature to manage the distribution between versions.
D
Create an App Engine application that hosts the model versions, using deployment settings to control the traffic split between versions.
E
Combine the use of Vertex AI for model deployment and Dataflow for real-time prediction processing, ensuring minimal management overhead.