async-inference

To run SageMaker async inferences, you upload large payloads to S3 as input, invoke the async endpoint, and poll S3 for results. This decoupled S3 I/O approach prevents timeouts on long-running workloads.

Using SageMaker async endpoints with S3 I/O is an effective way to process long-running batch inferences. It applies to batch processing and enables cost-efficient scaling across regions without blocking the client.

Yes, you can use S3 to manage input and output for SageMaker asynchronous endpoints. The skill implements an end-to-end flow using S3 for data I/O, async endpoint invocation, and polling with retry logic to retrieve results.

You retrieve results from an asynchronous SageMaker endpoint invocation by polling the S3 output bucket. The skill implements polling with retry logic to automatically fetch results once the long-running inference completes.

Yes, this skill supports cost-efficient scaling across regions for SageMaker inferences. Asynchronous processing with S3 I/O enables non-real-time workloads to scale efficiently without maintaining idle real-time endpoints.