Google Professional Machine Learning Engineer Question 47
Select 3Google Cloud PlatformYou are tasked with building a retrieval-augmented generation (RAG) application using Google Cloud's Vertex AI for a document search system. The system needs to retrieve relevant documents from a large dataset stored in BigQuery and then generate a natural language response using a fine-tuned LLM. Which steps should you take to implement this solution in Vertex AI?
- A
Use Vertex AI Matching Engine to create an index for relevant document retrieval.
- B
Export the dataset from BigQuery to Cloud Storage and use a custom Python script for retrieval.
- C
Fine-tune a pre-trained language model in Vertex AI using relevant document embeddings.
- D
Integrate the retrieval step with a Generative AI Studio model in Vertex AI for response generation.
- E
Set up a pipeline using Vertex AI Pipelines to orchestrate retrieval and generation tasks.
Show answer and explanation
Correct answers: A, D, E
Explanation
In a RAG application, the retrieval step is critical to fetch relevant documents, and Vertex AI Matching Engine is ideal for this purpose. The pre-trained LLMs in Generative AI Studio can then generate responses based on the retrieved content. Vertex AI Pipelines can be used to streamline the workflow, making it easier to manage and execute retrieval and generation tasks in a scalable manner. Exporting datasets or fine-tuning models unnecessarily adds complexity and is not aligned with best practices for RAG applications on Google Cloud.
- A. Correct.
Correct: Vertex AI Matching Engine is optimized for large-scale, low-latency retrieval and is a recommended tool for indexing and retrieving relevant documents in a RAG application.
- B. Incorrect.
Incorrect: Exporting the dataset and using a custom Python script is inefficient and unnecessary when powerful tools like Vertex AI Matching Engine and BigQuery ML are available.
- C. Incorrect.
Incorrect: Fine-tuning the language model using document embeddings is not required. Instead, the pre-trained LLM can be used directly with retrieved documents for generation.
- D. Correct.
Correct: Generative AI Studio in Vertex AI provides pre-trained LLMs that can be integrated with the retrieval step to generate coherent responses based on retrieved documents.
- E. Correct.
Correct: Vertex AI Pipelines can be used to automate and orchestrate the entire RAG workflow, ensuring the retrieval and generation steps are seamlessly executed.