Google Professional Machine Learning Engineer Question 46
Select 3Google Cloud PlatformYou are tasked with building a retrieval-augmented generation (RAG) application for customer support using Vertex AI. The application must retrieve relevant knowledge base documents stored in a BigQuery table and use a pre-trained large language model (LLM) to generate accurate customer responses. Which of the following steps should you include in your implementation?
- A
Use Vertex AI Feature Store to store and retrieve the embeddings for the knowledge base documents.
- B
Create a custom embedding model and deploy it to Vertex AI to generate embeddings for the knowledge base documents.
- C
Use Vertex AI Matching Engine to perform similarity searches over the embeddings of the knowledge base documents.
- D
Leverage a pre-trained embedding model from Vertex AI Model Garden to generate embeddings for the knowledge base documents.
- E
Use a pre-trained large language model from Vertex AI Model Garden to generate customer responses based on retrieved documents.
Show answer and explanation
Correct answers: C, D, E
Explanation
To implement a retrieval-augmented generation (RAG) application using Vertex AI, you need to generate embeddings for the knowledge base documents using a pre-trained embedding model (Vertex AI Model Garden), store and query these embeddings for similarity searches using Vertex AI Matching Engine, and utilize a pre-trained large language model to generate responses based on the retrieved documents (Vertex AI Model Garden). These steps ensure an efficient and scalable RAG implementation without requiring custom model development.
- A. Incorrect.
Vertex AI Feature Store is designed for managing and serving features for ML models, not for storing document embeddings or performing similarity searches. This is not the correct tool for a RAG workflow.
- B. Incorrect.
Creating and deploying a custom embedding model is unnecessary in this scenario because pre-trained embedding models are already available in Vertex AI Model Garden.
- C. Correct.
Vertex AI Matching Engine is optimized for performing similarity searches on large-scale embedding datasets, making it a key component in retrieving relevant knowledge base documents.
- D. Correct.
Using a pre-trained embedding model from Vertex AI Model Garden is an efficient way to generate high-quality embeddings for the knowledge base documents without the need to train a custom model.
- E. Correct.
Using a pre-trained large language model from Vertex AI Model Garden is appropriate for generating customer responses based on the retrieved documents, as it simplifies the implementation and leverages state-of-the-art NLP capabilities.