Databricks Machine Learning Associate Question 652
Single answerYou are deploying a machine learning model on Databricks for real-time inference. To improve scalability, you want to split incoming data between multiple model endpoints based on specific criteria, such as customer region or request type. Which of the following methods would be most appropriate for this scenario?
- A
Use a routing function that directs requests to specific model endpoints based on predefined rules.
- B
Deploy a single model endpoint and scale it horizontally to handle all requests.
- C
Use Databricks' AutoML to automatically split the incoming data across multiple endpoints.
- D
Configure a Databricks job to batch process requests and send them to a single endpoint.
Show answer and explanation
Correct answer: A
Explanation
In real-time inference scenarios where data needs to be split between multiple endpoints based on specific criteria, using a routing function is the most effective solution. It allows for dynamic and scalable distribution of requests to the appropriate models, ensuring efficient utilization of resources and meeting latency requirements.
- A. Correct.
This is correct because a routing function can efficiently direct requests to the appropriate model endpoint based on predefined logic, such as customer attributes or request type. It ensures scalability and flexibility in real-time inference.
- B. Incorrect.
This is incorrect because scaling a single endpoint horizontally does not address the need to split data based on specific criteria. It only increases capacity without adding routing logic.
- C. Incorrect.
This is incorrect because Databricks AutoML is a tool for automating model training and hyperparameter tuning, not for routing data to multiple endpoints.
- D. Incorrect.
This is incorrect because batch processing is not suitable for real-time inference, which requires low-latency responses. Additionally, sending all requests to a single endpoint does not address the need for splitting data.