Databricks Generative AI Engineer Associate Question 236
Select 3You are building a Retrieval-Augmented Generation (RAG) application on Databricks that involves querying a vector database and serving relevant features to a large language model (LLM). Which resources are essential to serve the features effectively during inference?
- A
A vector database with high retrieval speed and scalability
- B
A feature store to preprocess and store structured data features for inference
- C
A real-time streaming service for continuously updating the model weights
- D
A compute cluster capable of running low-latency queries on the feature store
- E
A data visualization tool to monitor the outputs of the LLM
Show answer and explanation
Correct answers: A, B, D
Explanation
Serving features for a RAG application requires a combination of tools that support efficient storage and retrieval of embeddings, as well as structured data features. A vector database is vital for storing embeddings, a feature store handles structured data, and a compute cluster enables low-latency access. Tools like streaming services or visualization platforms, while useful for other aspects of the workflow, are not directly involved in feature serving.
- A. Correct.
A vector database is required to store and retrieve embeddings efficiently for the RAG application. High retrieval speed is critical for low-latency inference.
- B. Correct.
A feature store is essential for preprocessing and storing structured data features that can be retrieved and integrated with embeddings during inference.
- C. Incorrect.
A real-time streaming service for updating model weights is not necessary for serving features during inference. Model weights are typically updated offline during training.
- D. Correct.
A compute cluster ensures low-latency execution of queries on the feature store, which is important for serving features in real-time.
- E. Incorrect.
A data visualization tool is useful for monitoring and debugging but is not directly required to serve features for a RAG application.