Predibase Inference Streaming (HTTP + SSE)

asyncapi: '2.6.0'
id: 'urn:com:predibase:inference:sse'
info:
  title: Predibase Inference Streaming (HTTP + SSE)
  version: '1.0.0'
  description: |
    AsyncAPI 2.6 description of Predibase's **inference streaming** surface.

    Predibase does not publish a WebSocket API. The only asynchronous /
    event-style transport documented at
    https://docs.predibase.com/user-guide/inference/rest_api and
    https://docs.predibase.com/user-guide/inference/migrate-openai is **HTTP
    Server-Sent Events (SSE)**, delivered two ways:

    1. Over the OpenAI-compatible endpoint `POST /v1/chat/completions` when the
       request body sets `stream: true`.
    2. Over the native `POST /generate_stream` endpoint, which always streams
       generated tokens.

    SSE is a one-way, server-to-client HTTP streaming channel; it is **not**
    WebSocket. The request bodies are modeled in the companion OpenAPI document
    at `openapi/predibase-openapi.yml`.
  contact:
    name: API Evangelist
    email: kin@apievangelist.com
    url: https://apievangelist.com
  license:
    name: API documentation - Predibase Terms of Service
    url: https://predibase.com/terms-of-service
  x-transport-notes:
    transport: HTTP Server-Sent Events (SSE)
    protocol: https
    direction: server-to-client (one-way)
    mediaType: text/event-stream
    triggeredBy: 'POST .../v1/chat/completions with { "stream": true }, or POST .../generate_stream'
    notWebSocket: true
    source: https://docs.predibase.com/user-guide/inference/rest_api
defaultContentType: text/event-stream
servers:
  serving:
    url: serving.app.predibase.com/{tenant}/deployments/v2/llms/{model}
    protocol: https
    description: |
      Predibase inference (serving) base. Streaming is delivered as HTTP
      Server-Sent Events over this base, either via the OpenAI-compatible
      `/v1/chat/completions` endpoint with `stream: true` or via the native
      `/generate_stream` endpoint. AsyncAPI 2.6 has no dedicated SSE protocol
      identifier; `https` is used and the SSE transport is documented in
      `info.x-transport-notes` and on each channel.
    security:
      - bearerAuth: []
    variables:
      tenant:
        default: TENANT_ID
        description: Predibase tenant ID.
      model:
        default: DEPLOYMENT_NAME
        description: Deployment name.
channels:
  /v1/chat/completions:
    description: |
      OpenAI-compatible chat completion SSE stream. The client opens this
      channel by issuing `POST /v1/chat/completions` with a JSON body containing
      `stream: true`. The server responds with `Content-Type: text/event-stream`
      and emits `data:` lines, each carrying one JSON `chat.completion.chunk`,
      followed by a final `data: [DONE]` line.
    bindings:
      http:
        type: request
        method: POST
        bindingVersion: '0.3.0'
      x-sse:
        mediaType: text/event-stream
        eventField: 'data'
        terminator: '[DONE]'
    subscribe:
      operationId: streamChatCompletionChunks
      summary: Subscribe to streamed chat completion chunks (SSE).
      bindings:
        http:
          type: response
          bindingVersion: '0.3.0'
      message:
        oneOf:
          - $ref: '#/components/messages/ChatCompletionChunk'
          - $ref: '#/components/messages/StreamDone'
  /generate_stream:
    description: |
      Native token stream. The client opens this channel by issuing
      `POST /generate_stream` with a JSON body containing `inputs` and optional
      `parameters` (including `adapter_id` and `adapter_source`). The server
      emits one SSE `data:` event per generated token.
    bindings:
      http:
        type: request
        method: POST
        bindingVersion: '0.3.0'
      x-sse:
        mediaType: text/event-stream
        eventField: 'data'
    subscribe:
      operationId: streamGeneratedTokens
      summary: Subscribe to streamed generated tokens (SSE).
      bindings:
        http:
          type: response
          bindingVersion: '0.3.0'
      message:
        $ref: '#/components/messages/TokenChunk'
components:
  securitySchemes:
    bearerAuth:
      type: http
      scheme: bearer
      bearerFormat: 'Predibase API token'
      description: |
        Set `Authorization: Bearer <PREDIBASE_API_TOKEN>` on the request that
        opens the SSE stream.
  messages:
    ChatCompletionChunk:
      name: ChatCompletionChunk
      title: Streamed chat completion chunk
      contentType: application/json
      summary: A single SSE `data:` event carrying one JSON `chat.completion.chunk` object.
      payload:
        $ref: '#/components/schemas/ChatCompletionChunk'
    TokenChunk:
      name: TokenChunk
      title: Streamed generated-token chunk
      contentType: application/json
      summary: A single SSE `data:` event carrying one generated token and its metadata.
      payload:
        $ref: '#/components/schemas/TokenChunk'
    StreamDone:
      name: StreamDone
      title: Stream terminator
      contentType: text/plain
      summary: 'The literal SSE event `data: [DONE]` marking end of the chat completion stream.'
      payload:
        $ref: '#/components/schemas/StreamDoneSentinel'
  schemas:
    StreamDoneSentinel:
      type: string
      enum:
        - '[DONE]'
      description: 'End-of-stream sentinel. The full SSE line is `data: [DONE]`.'
    ChatCompletionChunk:
      type: object
      required:
        - id
        - object
        - choices
      properties:
        id:
          type: string
        object:
          type: string
          enum:
            - chat.completion.chunk
        created:
          type: integer
        model:
          type: string
        choices:
          type: array
          items:
            type: object
            properties:
              index:
                type: integer
              delta:
                type: object
                properties:
                  role:
                    type: string
                  content:
                    type: string
                    nullable: true
              finish_reason:
                type: string
                nullable: true
    TokenChunk:
      type: object
      properties:
        token:
          type: object
          properties:
            id:
              type: integer
            text:
              type: string
            logprob:
              type: number
        generated_text:
          type: string
          nullable: true
          description: Present only on the final event; the full generated string.
        details:
          type: object
          nullable: true