Databricks Generative AI Engineer Associate Question 220
Select 4You are tasked with deploying an endpoint for a basic Retrieval-Augmented Generation (RAG) application in Databricks. What are the correct steps to properly deploy the endpoint?
- A
Prepare the knowledge base by indexing documents into a vector database.
- B
Train a custom large language model (LLM) for the RAG application.
- C
Create a Databricks MLflow model serving endpoint for the RAG pipeline.
- D
Integrate the retrieval step and the LLM into a single inference pipeline.
- E
Enable real-time monitoring for the endpoint to track performance and errors.
Show answer and explanation
Correct answers: A, C, D, E
Explanation
To deploy an endpoint for a basic RAG application in Databricks, you need to first prepare the knowledge base by indexing documents into a vector database. Then, create an inference pipeline that combines the retrieval step and the LLM. Deploy this pipeline using a Databricks MLflow model serving endpoint to handle requests. Finally, enable real-time monitoring for performance tracking and diagnostics. Training a custom LLM is unnecessary for a basic setup as pre-trained models are sufficient.
- A. Correct.
Correct: Indexing documents into a vector database is a key step in preparing the knowledge base for retrieval in a RAG application.
- B. Incorrect.
Incorrect: Training a custom LLM is not required for a basic RAG application, as pre-trained LLMs are typically used.
- C. Correct.
Correct: Creating a model serving endpoint is essential for deploying the RAG pipeline to handle real-time inference requests.
- D. Correct.
Correct: Integrating the retrieval step and the LLM into a single inference pipeline ensures the RAG application can seamlessly provide responses.
- E. Correct.
Correct: Enabling real-time monitoring ensures the endpoint's performance and errors can be tracked and addressed promptly.