Author: Kareem Syed-Mohammed

Today, Amazon SageMaker AI introduces capacity aware instance pool for new and existing inference endpoints. You define a prioritized list of instance types, and SageMaker AI automatically works through your list whenever capacity is constrained at creation, during scale-out, and during scale-in. Your endpoint provisions on available AI Infrastructure without manual intervention. This capability is available for Single Model Endpoints, Inference Component-based endpoints, and Asynchronous Inference endpoints.

In this post, we walk through how to search for available p-family GPU capacity, create a training plan reservation for inference, and deploy a SageMaker AI inference endpoint on that reserved capacity. We follow a data scientist’s journey as they reserve capacity for model evaluation and manage the endpoint throughout the reservation lifecycle.

Amazon SageMaker AI now supports EAGLE-based adaptive speculative decoding, a technique that accelerates large language model inference by up to 2.5x while maintaining output quality. In this post, we explain how to use EAGLE 2 and EAGLE 3 speculative decoding in Amazon SageMaker AI, covering the solution architecture, optimization workflows using your own datasets or SageMaker’s built-in data, and benchmark results demonstrating significant improvements in throughput and latency.