AWS.SageMakerRuntime reference

InvokeEndpoint

Source: src/AWS/SageMakerRuntime/InvokeEndpoint.ts

Runtime binding for sagemaker:InvokeEndpoint — run real-time inference against a deployed SageMaker endpoint with a model-native request body.

SageMaker Runtime is a pure pay-per-call data-plane API: the endpoint itself is provisioned out of band (a deployed SageMaker model/endpoint — costly to keep running). The binding takes one or more endpoint names and grants the function sagemaker:InvokeEndpoint scoped to exactly those endpoint ARNs. Pass the request/response bodies as raw bytes — the container behind the endpoint owns the payload shape, no marshalling.

InvokeEndpoint: Invoking an Endpoint

// init
const invokeEndpoint = yield* AWS.SageMakerRuntime.InvokeEndpoint(
  "my-model-endpoint",
);

// runtime — Body is the raw payload the container expects
const result = yield* invokeEndpoint({
  ContentType: "application/json",
  Accept: "application/json",
  Body: JSON.stringify({ instances: [[1, 2, 3, 4]] }),
});
const raw = new TextDecoder().decode(result.Body);

InvokeEndpointAsync

Source: src/AWS/SageMakerRuntime/InvokeEndpointAsync.ts

Runtime binding for sagemaker:InvokeEndpointAsync — enqueue an asynchronous inference request against a deployed SageMaker async endpoint. The request payload is read from S3 (InputLocation) and the result is written back to S3; the call returns immediately with an OutputLocation.

The binding takes one or more endpoint names and grants the function sagemaker:InvokeEndpointAsync scoped to exactly those endpoint ARNs.

InvokeEndpointAsync: Invoking an Async Endpoint

// init
const invokeAsync = yield* AWS.SageMakerRuntime.InvokeEndpointAsync(
  "my-async-endpoint",
);

// runtime
const result = yield* invokeAsync({
  ContentType: "application/json",
  InputLocation: "s3://my-bucket/input/request.json",
});
// result.OutputLocation — poll S3 for the written inference result

InvokeEndpointWithResponseStream

Source: src/AWS/SageMakerRuntime/InvokeEndpointWithResponseStream.ts

Runtime binding for sagemaker:InvokeEndpointWithResponseStream — the streaming variant of InvokeEndpoint. The inference response arrives incrementally as an event Stream of PayloadPart events carrying raw model-specific bytes; the container behind the endpoint must support inference streaming.

The binding takes one or more endpoint names and grants the function sagemaker:InvokeEndpoint (the IAM action the streaming operation authorizes against) scoped to exactly those endpoint ARNs.

InvokeEndpointWithResponseStream: Streaming an Endpoint Response

// init
const invokeStream = yield* AWS.SageMakerRuntime.InvokeEndpointWithResponseStream(
  "my-streaming-endpoint",
);

// runtime — Body is the raw payload the container expects
const result = yield* invokeStream({
  ContentType: "application/json",
  Body: JSON.stringify({ inputs: "Say hello." }),
});
const events = yield* Stream.runCollect(result.Body);
// each PayloadPart's Bytes is a model-specific chunk