Databricks Generative AI Engineer Associate Question 219
Select 4You are tasked with deploying a basic Retrieval-Augmented Generation (RAG) application on Databricks. Which of the following steps must be performed in the correct order to successfully deploy and operationalize an endpoint?
- A
Prepare and pre-process the dataset for indexing (e.g., embedding generation).
- B
Deploy a vector database to store the embeddings.
- C
Train a new large language model (LLM) from scratch.
- D
Build and test the retrieval and generation pipeline.
- E
Create a Databricks Model Serving endpoint for deployment.
Show answer and explanation
Correct answers: A, B, D, E
Explanation
To deploy a basic RAG application, you must first prepare the dataset by generating embeddings, store those embeddings in a vector database, build and test the retrieval-generation pipeline, and then deploy the application using a Databricks Model Serving endpoint. Training a new LLM is not required in this workflow, as pre-trained models are typically leveraged in RAG applications.
- A. Correct.
Correct: Preparing and pre-processing the dataset is a fundamental step, as embeddings must be generated before they can be stored in a vector database for retrieval.
- B. Correct.
Correct: A vector database is required to efficiently store and query the embeddings generated during pre-processing.
- C. Incorrect.
Incorrect: Training a new LLM from scratch is unnecessary for a basic RAG application. Pre-trained models are typically used, saving time and resources.
- D. Correct.
Correct: Building and testing the retrieval and generation pipeline ensures that the application logic is functional and integrates the vector database with the language model.
- E. Correct.
Correct: Creating a Databricks Model Serving endpoint is the final step to deploy the RAG application for end-user access.