Amazon SageMaker HyperPod now supports data capture for inference workloads

This article announces data capture support for inference workloads in Amazon SageMaker HyperPod, enabling organizations to record and analyze model inputs and outputs for monitoring and compliance.

• Record inference request and response payloads for model monitoring and debugging
• Capture data at endpoint, load balancer, or model pod levels for layered observability
• Asynchronous delivery to Amazon S3 with configurable sampling and KMS encryption
• Non-blocking design preserves production availability and performance
• Available for HyperPod clusters using EKS orchestrator across all supported AWS regions

Data capture enables organizations to gain visibility into inference workloads for drift detection, regulatory compliance, troubleshooting, and dataset creation without building custom logging infrastructure.