MLS-C01 Question 360
Single answerYou are a Machine Learning engineer tasked with deploying a custom-trained model to production using Amazon SageMaker. The model requires real-time inference and must handle unpredictable traffic patterns. Additionally, you need a solution that minimizes downtime during model updates. Which deployment approach should you use?
- A
Deploy the model using Amazon SageMaker Hosting Services with an endpoint and enable Auto Scaling.
- B
Deploy the model using Amazon SageMaker Batch Transform for real-time inference.
- C
Deploy the model as a Lambda function and use API Gateway to handle requests.
- D
Deploy the model using Amazon SageMaker Hosting Services with multi-model endpoints.
Show answer and explanation
Correct answer: A
Explanation
Amazon SageMaker Hosting Services with an endpoint is the best choice for deploying a model for real-time inference, especially when handling unpredictable traffic patterns. With Auto Scaling enabled, the endpoint can dynamically adjust to traffic changes. Additionally, SageMaker supports deployment strategies like blue/green deployments to minimize downtime during model updates. Other options, such as Batch Transform or Lambda with API Gateway, are either not designed for real-time inference or lack the ability to efficiently handle unpredictable traffic patterns.
- A. Correct.
This is correct. Amazon SageMaker Hosting Services with an endpoint supports real-time inference and can handle unpredictable traffic patterns. Enabling Auto Scaling ensures that the service scales dynamically to meet demand, and SageMaker endpoints support strategies for minimizing downtime during model updates, such as blue/green deployments.
- B. Incorrect.
This is incorrect. Amazon SageMaker Batch Transform is designed for batch inference, not real-time inference. It is not suitable for scenarios requiring immediate responses to unpredictable traffic.
- C. Incorrect.
This is incorrect. While deploying the model as a Lambda function with API Gateway can provide real-time inference, it is not well-suited for handling large-scale unpredictable traffic patterns compared to SageMaker's built-in capabilities, like Auto Scaling.
- D. Incorrect.
This is incorrect. Multi-model endpoints are used to host multiple models on a single endpoint to optimize cost. However, they are more suitable for predictable workloads and do not specifically address unpredictable traffic patterns or minimize downtime during updates.