arazzo: 1.0.1
info:
  title: OpenAI Fine-Tuning Job
  summary: Upload a training file, start a fine-tuning job, poll until terminal, and read the result.
  description: >-
    Uploads a JSONL training file with the fine-tune purpose, enqueues a
    fine-tuning job against it, polls the job until it reaches a terminal status,
    and branches on the outcome to return the fine-tuned model name on success.
    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: openaiApi
  url: ../openapi/openai-openapi-master.yml
  type: openapi
workflows:
- workflowId: fine-tuning-job
  summary: Upload training data, run a fine-tuning job, and poll to completion.
  description: >-
    Uploads the training file, creates a fine-tuning job, polls the job status,
    and on success returns the resulting fine-tuned model identifier.
  inputs:
    type: object
    required:
    - apiKey
    - trainingFile
    - model
    properties:
      apiKey:
        type: string
        description: OpenAI API key used as a Bearer token.
      trainingFile:
        type: string
        description: The JSONL training file contents to upload (binary/multipart field).
      model:
        type: string
        description: The base model to fine-tune (e.g. gpt-4o-mini).
  steps:
  - stepId: uploadTrainingFile
    description: Upload the JSONL training file with the fine-tune purpose.
    operationId: createFile
    parameters:
    - name: Authorization
      in: header
      value: "Bearer $inputs.apiKey"
    requestBody:
      contentType: multipart/form-data
      payload:
        file: $inputs.trainingFile
        purpose: fine-tune
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      fileId: $response.body#/id
  - stepId: createJob
    description: Create a fine-tuning job using the uploaded training file.
    operationId: createFineTuningJob
    parameters:
    - name: Authorization
      in: header
      value: "Bearer $inputs.apiKey"
    requestBody:
      contentType: application/json
      payload:
        model: $inputs.model
        training_file: $steps.uploadTrainingFile.outputs.fileId
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      jobId: $response.body#/id
      status: $response.body#/status
  - stepId: pollJob
    description: >-
      Poll the fine-tuning job until it reaches a terminal status, branching to
      the result step when finished and looping while it is still running.
    operationId: retrieveFineTuningJob
    parameters:
    - name: Authorization
      in: header
      value: "Bearer $inputs.apiKey"
    - name: fine_tuning_job_id
      in: path
      value: $steps.createJob.outputs.jobId
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      status: $response.body#/status
      fineTunedModel: $response.body#/fine_tuned_model
    onSuccess:
    - name: jobSucceeded
      type: goto
      stepId: getResult
      criteria:
      - context: $response.body
        condition: $.status == "succeeded"
        type: jsonpath
    - name: jobStillRunning
      type: goto
      stepId: pollJob
      criteria:
      - context: $response.body
        condition: $.status == "validating_files" || $.status == "queued" || $.status == "running"
        type: jsonpath
  - stepId: getResult
    description: Retrieve the finished fine-tuning job and return the fine-tuned model.
    operationId: retrieveFineTuningJob
    parameters:
    - name: Authorization
      in: header
      value: "Bearer $inputs.apiKey"
    - name: fine_tuning_job_id
      in: path
      value: $steps.createJob.outputs.jobId
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      fineTunedModel: $response.body#/fine_tuned_model
      trainedTokens: $response.body#/trained_tokens
  outputs:
    jobId: $steps.createJob.outputs.jobId
    fineTunedModel: $steps.getResult.outputs.fineTunedModel