arazzo: 1.0.1
info:
  title: Hugging Face Embeddings Pair
  summary: Confirm an embedding model is available, then embed a query and a document for comparison.
  description: >-
    A semantic-search building block over the Inference Providers router. The
    workflow lists the available router models to confirm an embedding model is
    servable, then creates an embedding vector for a query string and a separate
    embedding vector for a document string so the two can be compared downstream
    (for example by cosine similarity). Every step spells out its request inline
    so the flow can be read and executed without opening the underlying OpenAPI
    description.
  version: 1.0.0
sourceDescriptions:
- name: inferenceProvidersApi
  url: ../openapi/hugging-face-inference-providers-api.yml
  type: openapi
workflows:
- workflowId: embeddings-pair
  summary: Embed a query and a document with the same model for similarity comparison.
  description: >-
    Confirms an embedding model is available on the router, then produces an
    embedding vector for a query and for a document.
  inputs:
    type: object
    required:
    - hfToken
    - modelId
    - query
    - document
    properties:
      hfToken:
        type: string
        description: Hugging Face access token used as a Bearer credential.
      modelId:
        type: string
        description: The embedding model id to use.
      query:
        type: string
        description: The query text to embed.
      document:
        type: string
        description: The document text to embed for comparison against the query.
  steps:
  - stepId: listRouterModels
    description: >-
      List the models available through the router to confirm the requested
      embedding model can be served before embedding.
    operationId: listModels
    parameters:
    - name: Authorization
      in: header
      value: Bearer $inputs.hfToken
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      models: $response.body#/data
  - stepId: embedQuery
    description: >-
      Create an embedding vector for the query text using the requested model.
    operationId: createEmbeddings
    parameters:
    - name: Authorization
      in: header
      value: Bearer $inputs.hfToken
    requestBody:
      contentType: application/json
      payload:
        model: $inputs.modelId
        input: $inputs.query
        encoding_format: float
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      queryVector: $response.body#/data/0/embedding
  - stepId: embedDocument
    description: >-
      Create an embedding vector for the document text using the same model so
      it can be compared to the query vector.
    operationId: createEmbeddings
    parameters:
    - name: Authorization
      in: header
      value: Bearer $inputs.hfToken
    requestBody:
      contentType: application/json
      payload:
        model: $inputs.modelId
        input: $inputs.document
        encoding_format: float
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      documentVector: $response.body#/data/0/embedding
      totalTokens: $response.body#/usage/total_tokens
  outputs:
    queryVector: $steps.embedQuery.outputs.queryVector
    documentVector: $steps.embedDocument.outputs.documentVector