Google Professional Machine Learning Engineer Question 45
Select 3Google Cloud PlatformYou are building a Retrieval-Augmented Generation (RAG) application on Google Cloud to create a conversational AI model for customer support. The application needs to retrieve relevant documents from a large dataset and generate human-like responses. Which of the following steps should you take to implement this application using Vertex AI?
- A
Use Vertex AI Matching Engine to perform similarity search on the document embeddings.
- B
Train a custom sequence-to-sequence model on Vertex AI to generate responses based on retrieved documents.
- C
Store the document embeddings in BigQuery for efficient retrieval during runtime.
- D
Use a pre-trained generative model from the Vertex AI Model Garden to generate responses after retrieving relevant documents.
- E
Deploy a Bigtable instance to serve as the vector database for storing document embeddings.
Show answer and explanation
Correct answers: A, B, D
Explanation
Retrieval-Augmented Generation (RAG) applications combine retrieval and generation phases. Vertex AI Matching Engine is used for efficient retrieval of relevant documents, which forms the retrieval phase. For the generation phase, you can either train a custom sequence-to-sequence model using Vertex AI or leverage pre-trained generative models from the Vertex AI Model Garden. Other tools like BigQuery and Bigtable are not optimized for handling embeddings or performing similarity search, making them unsuitable for this use case.
- A. Correct.
Correct. Vertex AI Matching Engine is designed for efficient similarity search and retrieval of relevant documents, making it a key component for implementing the retrieval phase in a RAG application.
- B. Correct.
Correct. A custom sequence-to-sequence model can be trained on Vertex AI for generating high-quality responses based on the retrieved documents, enabling the generation phase of the RAG workflow.
- C. Incorrect.
Incorrect. BigQuery is optimized for analytical queries, not for similarity search or retrieval of embeddings. Using BigQuery for this purpose would not be efficient.
- D. Correct.
Correct. Using a pre-trained generative model from the Vertex AI Model Garden is a practical and scalable approach to implement the generation phase without the need for training a model from scratch.
- E. Incorrect.
Incorrect. Bigtable is a NoSQL database optimized for large-scale structured data, but it is not designed to serve as a vector database for storing and retrieving document embeddings.