Databricks Machine Learning Associate Question 379
Single answerYou are tasked with deploying a machine learning model on Databricks to serve predictions for a real-time application with low latency requirements. The model is trained, and you need to distribute it for inference. Which of the following approaches is the most appropriate for minimizing latency while ensuring scalability?
- A
Deploy the model as a real-time endpoint using Databricks Model Serving.
- B
Save the model as a Delta table and serve predictions by running batch inference jobs.
- C
Export the model to a cloud service like AWS SageMaker for real-time inference.
- D
Deploy the model using a scheduled Databricks Job to process incoming requests.
Show answer and explanation
Correct answer: A
Explanation
Databricks Model Serving provides a managed solution to deploy machine learning models for real-time inference with low latency. It is highly scalable and well-integrated within the Databricks ecosystem, making it the best choice for real-time applications. Other options, such as batch inference or scheduled jobs, are better suited for offline or periodic processing tasks and do not meet the low-latency requirement.
- A. Correct.
This is the correct approach because Databricks Model Serving is designed to serve low-latency, real-time predictions, and it supports scalable deployments for such use cases.
- B. Incorrect.
This approach is unsuitable for real-time applications as batch inference is not designed for low-latency requirements. It is better suited for periodic or offline prediction tasks.
- C. Incorrect.
While exporting to a cloud service like AWS SageMaker can achieve real-time inference, it is not the most efficient option when deploying directly within Databricks is possible. This adds unnecessary complexity to the solution.
- D. Incorrect.
Using a scheduled Databricks Job is not appropriate for real-time applications since it is designed for batch processing or periodic tasks rather than handling individual requests with low latency.