Amazon SageMaker HyperPod now supports data capture for inference workloads, allowing customers to record request and response payloads for model monitoring, compliance, debugging, and offline analysis. Data can be captured at the endpoint, load balancer, or model pod level, with captured data delivered asynchronously to Amazon S3. The feature supports configurable sampling, encryption via customer-managed AWS KMS keys, and is designed to never block inference traffic. It is available for HyperPod clusters using the EKS orchestrator across all supported AWS regions.
142 Impressions