AWS Certified AI Practitioner Practice Exam

Which SageMaker inference option is suitable for a company needing near real-time latency with large data inputs?

1. Real-time inference

2. Serverless inference

3. Asynchronous inference

4. Batch transform

The correct answer is: Asynchronous inference

The option that best fits the requirement for near real-time latency with large data inputs is real-time inference. This approach enables you to perform inference requests and receive responses quickly, making it suitable for applications where low latency is critical. Real-time inference allows for a direct interaction between the application and the SageMaker model endpoint, enabling responses to be obtained in milliseconds, which is essential for near real-time scenarios. The other choices serve different purposes: serverless inference is designed to auto-scale based on demand but may not guarantee the same level of low-latency response required for real-time applications. Asynchronous inference, while beneficial for handling large datasets and complex processing tasks, can introduce delays since it processes requests in the background and requires triggering a callback once the processing is complete. Batch transform is intended for scenarios where high throughput is prioritized over latency, as it processes large datasets in bulk rather than on a per-request basis, leading to longer processing times. Therefore, real-time inference is the appropriate approach for the company’s need for handling large data inputs while maintaining the low latency required for near real-time requirements.