1Z0-184-25 Question 104
Select 2Your organization stores large volumes of unstructured text in Oracle Object Storage and plans to enable semantic search in an Oracle Autonomous Database. You have chosen a custom open-source model to generate vector embeddings, but you want to run these computations outside the database for performance and scalability reasons. Which two steps are recommended to ensure optimal performance and efficient integration of these externally generated embeddings into Oracle Autonomous Database?
- A
Use Oracle Data Science or GPU-enabled Compute shapes to process text from Object Storage, store the generated embeddings temporarily on block volumes, and then load these embeddings into Oracle Autonomous Database via scheduled or asynchronous jobs.
- B
Simplify the data pipeline by batching documents from Object Storage, producing embeddings externally, and writing the resulting embeddings directly to a vector column in the Autonomous Database table.
- C
Invoke an external API for embedding generation from within an Autonomous Database stored procedure, calling it for every row insertion to ensure real-time data consistency.
- D
Rely on built-in GPU support within Oracle Autonomous Database to generate embeddings inside the database and store them directly in vector columns.
Show answer and explanation
Correct answers: A, B
Explanation
Generating embeddings outside the database is recommended for large-scale or computationally intensive inference jobs. Oracle Cloud Infrastructure supports GPU-enabled Compute shapes and Data Science environments to handle external processing efficiently. Storing only the final embeddings in Oracle Autonomous Database reduces overhead, streamlines performance, and simplifies maintenance. For more details, consult Oracle documentation on handling vector data types in Autonomous Database and best practices for batch or asynchronous data pipelines.
- A. Correct.
Correct. Leveraging Oracle Data Science or GPU-enabled Compute shapes lets you offload the embedding generation workload from the database. Storing embeddings temporarily on block volumes and loading them via asynchronous or scheduled jobs helps avoid unnecessary overhead and maintains scalability.
- B. Correct.
Correct. Batching documents outside the database and writing the generated embeddings directly to the database is a best practice for handling large workloads. This approach reduces contention and ensures a clean separation of concerns between the inference pipeline and the database.
- C. Incorrect.
Incorrect. Running an external embedding API for each row insertion from a stored procedure would introduce significant overhead, degrade insert performance, and complicate transaction management, making it unsuitable for large-scale or high-throughput scenarios.
- D. Incorrect.
Incorrect. Oracle Autonomous Database does not currently offer built-in GPU support for generating vector embeddings inside the database. Embedding generation should occur outside, then be stored in the database for efficient queries and indexing.