1Z0-184-25 Question 127
Select 3Your organization is creating a chatbot on Oracle Cloud Infrastructure (OCI) using a large language model (LLM) to provide answers based on a large set of product manuals stored in Oracle Object Storage. They decide to use a retrieval-augmented generation (RAG) architecture so that each query references up-to-date, relevant content. Which three recommended steps should you implement to achieve an optimal RAG solution in OCI, balancing performance and maintainability?
- A
Store your entire LLM model in an Edge Service so it can be retrieved with minimal latency on demand
- B
Break your manual documents into smaller chunks and store vector embeddings in an OCI-based vector index for quick lookups
- C
Regularly refresh embeddings to incorporate new or updated documents so that your knowledge base remains current
- D
Use an ephemeral instance to reconstruct and embed your entire dataset upon every single query to guarantee freshness
- E
Retrieve relevant chunks from your vector store on each user query and attach them as context in the LLM prompt
- F
Depend on a single monolithic data file for the LLM to parse at runtime without separate retrieval steps
Show answer and explanation
Correct answers: B, C, E
Explanation
Retrieval-augmented generation (RAG) in OCI involves breaking your data into manageable pieces, creating embeddings, and storing these vectors in an index that supports fast similarity searches. Whenever a query arrives, the most relevant segments are retrieved and provided to the LLM as context. This ensures that the response is grounded in the latest information without overburdening the model. For additional details on implementing a scalable RAG architecture on OCI, refer to Oracle Cloud Infrastructure documentation on vector search, data chunking, and ML deployment best practices.
- A. Incorrect.
Option 1 is incorrect: Storing the entire LLM in an Edge Service is neither a standard OCI RAG practice nor necessary. Typically, you deploy models on dedicated compute or ML services, not in an edge cache.
- B. Correct.
Option 2 is correct: Splitting documents into chunks and storing their vector embeddings in an OCI-based vector index (such as a suitable search index) is a core RAG practice for efficient retrieval.
- C. Correct.
Option 3 is correct: Keeping embeddings up to date for newly added or updated documents is essential to maintain relevancy in your RAG system.
- D. Incorrect.
Option 4 is incorrect: Re-embedding your entire dataset on every query is extremely inefficient and generally not how RAG solutions are designed to scale in OCI.
- E. Correct.
Option 5 is correct: An LLM typically needs only the most relevant context. Dynamically retrieving the top-scoring chunks from your vector index and supplying them within the LLM prompt is fundamental to RAG.
- F. Incorrect.
Option 6 is incorrect: Using a single large file and forcing the LLM to parse it each time ignores the advantages of retrieval-based context filtering and does not optimize performance.