What changed
Amazon SageMaker HyperPod now provides Data Capture, a built-in capability that records live inference inputs and outputs to Amazon S3 at three independent tiers. This feature allows teams to capture domain-specific jargon, unexpected formats, new products, and changing user behavior that may not be covered by pre-curated datasets.
What this means in practice
The Data Capture feature enables teams to detect drift, identify failure patterns, and use real user interactions to improve models. In a demonstrated example, a home-appliance support router captures production traffic, and Amazon Athena queries reveal a new-product blind spot. A SQL policy labels the captured records, and a fine-tune on that data teaches the router about the new product. The inference operator serves the model on one node, and the training operator runs the fine-tuning on another. Synthetic data is used for the walkthrough, with scripts, manifests, SQL, datasets, and expected outputs available in a GitHub repository.
Limitations and availability
Data Capture is available on SageMaker HyperPod clusters orchestrated by Amazon EKS with the inference and training operators installed. Two NVIDIA GPUs are required, one for serving and one for training. The serving pod holds its GPU for the life of the endpoint, so with only one GPU, the training pod remains pending. The feature is available in the same region as the S3 bucket. The dataCapture.modelPod.enabled: true block is required to activate Tier 3 capture. The supplied announcement does not specify the exact release date or availability beyond the described setup.
