| 1 | import type * as sagemaker from "@distilled.cloud/aws/sagemaker-runtime"; |
| 2 | import type * as Effect from "effect/Effect"; |
| 3 | import * as Binding from "../../Binding.ts"; |
| 4 | |
| 5 | /** |
| 6 | * The `InvokeEndpointWithResponseStream` request with the binding-injected |
| 7 | * `EndpointName` made optional: it defaults to the first bound endpoint name |
| 8 | * and may be overridden per call with any of the bound endpoint names. |
| 9 | * |
| 10 | * `Body` is the raw, model-specific inference payload — there is no |
| 11 | * auto-marshalling. The response `Body` is an event `Stream` whose |
| 12 | * `PayloadPart` events carry raw model-specific bytes. |
| 13 | */ |
| 14 | export interface InvokeEndpointWithResponseStreamRequest extends Omit< |
| 15 | sagemaker.InvokeEndpointWithResponseStreamInput, |
| 16 | "EndpointName" |
| 17 | > { |
| 18 | /** |
| 19 | * The endpoint to run inference on for this call. Must be one of the |
| 20 | * endpoint names the binding was created with. |
| 21 | * @default the first bound endpoint name |
| 22 | */ |
| 23 | EndpointName?: string; |
| 24 | } |
| 25 | |
| 26 | /** |
| 27 | * Runtime binding for `sagemaker:InvokeEndpointWithResponseStream` — the |
| 28 | * streaming variant of {@link InvokeEndpoint}. The inference response arrives |
| 29 | * incrementally as an event `Stream` of `PayloadPart` events carrying raw |
| 30 | * model-specific bytes; the container behind the endpoint must support |
| 31 | * inference streaming. |
| 32 | * |
| 33 | * The binding takes one or more endpoint names and grants the function |
| 34 | * `sagemaker:InvokeEndpoint` (the IAM action the streaming operation |
| 35 | * authorizes against) scoped to exactly those endpoint ARNs. |
| 36 | * |
| 37 | * ### Streaming an Endpoint Response |
| 38 | * **Example:** Aggregate Streamed Payload Parts |
| 39 | * ```typescript |
| 40 | * // init |
| 41 | * const invokeStream = yield* AWS.SageMakerRuntime.InvokeEndpointWithResponseStream( |
| 42 | * "my-streaming-endpoint", |
| 43 | * ); |
| 44 | * |
| 45 | * // runtime — Body is the raw payload the container expects |
| 46 | * const result = yield* invokeStream({ |
| 47 | * ContentType: "application/json", |
| 48 | * Body: JSON.stringify({ inputs: "Say hello." }), |
| 49 | * }); |
| 50 | * const events = yield* Stream.runCollect(result.Body); |
| 51 | * // each PayloadPart's Bytes is a model-specific chunk |
| 52 | * ``` |
| 53 | * |
| 54 | * @binding |
| 55 | */ |
| 56 | export interface InvokeEndpointWithResponseStream extends Binding.Service< |
| 57 | InvokeEndpointWithResponseStream, |
| 58 | "AWS.SageMakerRuntime.InvokeEndpointWithResponseStream", |
| 59 | ( |
| 60 | endpoint: string, |
| 61 | ...additionalEndpoints: string[] |
| 62 | ) => Effect.Effect< |
| 63 | ( |
| 64 | request: InvokeEndpointWithResponseStreamRequest, |
| 65 | ) => Effect.Effect< |
| 66 | sagemaker.InvokeEndpointWithResponseStreamOutput, |
| 67 | sagemaker.InvokeEndpointWithResponseStreamError |
| 68 | > |
| 69 | > |
| 70 | > {} |
| 71 | export const InvokeEndpointWithResponseStream = |
| 72 | Binding.Service<InvokeEndpointWithResponseStream>( |
| 73 | "AWS.SageMakerRuntime.InvokeEndpointWithResponseStream", |
| 74 | ); |