SageMaker HyperPod inference adds multi-tier data capture, direct Hugging Face Hub deployment, and NVMe caching

AWS Machine Learning Blog·low signal

AWS shipped five capabilities for enterprise inference on SageMaker HyperPod: multi-tier data capture for auditing and model improvement, direct deployment from the Hugging Face Hub, local NVMe model caching, and Route 53 integration. The data-capture tier is the notable one — it makes inference request/response logging a first-class primitive rather than something teams bolt on for eval pipelines.