1Z0-184-25 Question 103
Single answerYou are building a recommendation platform for an e-commerce site that relies on semantic search. You need to generate vector embeddings of product descriptions externally (outside the Oracle database) using a custom deep learning model. After generation, these embeddings must be integrated into your Autonomous Database for fast vector queries. Which approach should you adopt to efficiently manage the embedding generation and ensure they are properly ingested into the database?
- A
Set up a scalable GPU-accelerated environment (such as OCI Data Science or OCI Container Engine for Kubernetes) to run your model, store embeddings in Object Storage, and import them into the database periodically using data ingestion tools.
- B
Embed the model directly inside the Autonomous Database by installing external Python libraries for on-the-fly embedding generation within stored procedures.
- C
Continuously stream product descriptions to a load balancer that triggers Oracle Functions to generate embeddings, and rely solely on manual copy commands to synchronize them with the database.
- D
Use Oracle GoldenGate for Big Data to replicate embed operations directly from an online transaction processing system (OLTP) into the Autonomous Database.
Show answer and explanation
Correct answer: A
Explanation
Generating resource-intensive vector embeddings is best handled by compute resources outside the database, such as GPU instances in OCI Data Science or an OCI Container Engine for Kubernetes cluster. Oracle� recommended approach is to store intermediate results (embeddings) in Object Storage or similar storage, then ingest them into the Autonomous Database for advanced vector queries. This workflow adheres to Oracle� best practices for scalable AI/ML pipelines and ensures efficient embedding generation without overloading the database environment. Refer to the Oracle Cloud Infrastructure Data Science documentation for guidance on GPU-enabled notebook sessions and job scheduling.
- A. Correct.
Option 1 is correct. Leveraging a scalable environment with GPU acceleration (e.g., OCI Data Science, OCI Container Engine for Kubernetes) allows efficient generation of embeddings using your custom model. Storing them in Object Storage, then loading them in batches or through data ingestion pipelines into the database, helps maintain performance and synchronization. Oracle documentation recommends generating resource-intensive embeddings outside the database and then ingesting the results for vector queries.
- B. Incorrect.
Option 2 is incorrect. Installing external deep learning libraries and running the generation within stored procedures in Autonomous Database is not recommended due to potential performance overhead, maintenance complexity, and limitations in library support. The scenario explicitly requires generation outside the database.
- C. Incorrect.
Option 3 is incorrect. While Oracle Functions can be used for certain tasks, relying solely on triggered functions and manual copy commands to synchronize the embeddings with the database is inefficient and prone to errors. This approach lacks effective batching and might not scale for high volumes of data.
- D. Incorrect.
Option 4 is incorrect. Oracle GoldenGate for Big Data is useful for data replication between heterogeneous systems, but it is not typically designed for generating embeddings or handling large-scale compute-intensive inference tasks. It also wouldn't solve the requirement for external GPU-based processing.