Viam Train and Monitor an ML Model

arazzo: 1.0.1
info:
  title: Viam Train and Monitor an ML Model
  summary: Submit a TFLite training job, poll its status, and branch to logs or cancel.
  description: >-
    Submits a built-in TFLite training job against a curated dataset, then reads
    the job back to inspect its status. When the job is still pending or running
    it pulls the training logs, and when it has failed it cancels the job to free
    resources. Each request body is inlined so the flow can be executed directly
    against the Viam ML Training API.
  version: 1.0.0
sourceDescriptions:
- name: mlTrainingApi
  url: ../openapi/viam-ml-training-api-openapi.yml
  type: openapi
workflows:
- workflowId: train-and-monitor-model
  summary: Submit a training job, check status, and branch to logs or cancellation.
  description: >-
    Submits a TFLite training job, retrieves it by the returned id, and branches
    on status — fetching logs while it runs or canceling it if it failed.
  inputs:
    type: object
    required:
    - apiKey
    - organizationId
    - datasetId
    - modelName
    - modelType
    properties:
      apiKey:
        type: string
        description: Viam API key value sent in the key header.
      organizationId:
        type: string
        description: The organization the training job runs under.
      datasetId:
        type: string
        description: The dataset id the model is trained against.
      modelName:
        type: string
        description: Name for the resulting model.
      modelType:
        type: string
        description: One of single_label_classification, multi_label_classification, object_detection.
  steps:
  - stepId: submitJob
    description: Submit a built-in TFLite training job against the dataset.
    operationId: submitTrainingJob
    parameters:
    - name: key
      in: header
      value: $inputs.apiKey
    requestBody:
      contentType: application/json
      payload:
        organization_id: $inputs.organizationId
        dataset_id: $inputs.datasetId
        model_name: $inputs.modelName
        model_type: $inputs.modelType
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      trainingJobId: $response.body#/id
  - stepId: checkStatus
    description: Retrieve the training job by the returned id to inspect its status.
    operationId: getTrainingJob
    parameters:
    - name: key
      in: header
      value: $inputs.apiKey
    requestBody:
      contentType: application/json
      payload:
        id: $steps.submitJob.outputs.trainingJobId
    successCriteria:
    - condition: $statusCode == 200
    onSuccess:
    - name: jobFailed
      type: goto
      stepId: cancelJob
      criteria:
      - context: $response.body
        condition: $.metadata.state == 'failed'
        type: jsonpath
    - name: jobRunning
      type: goto
      stepId: fetchLogs
      criteria:
      - context: $response.body
        condition: $.metadata.state != 'failed'
        type: jsonpath
  - stepId: fetchLogs
    description: Pull the training job logs while the job is pending or running.
    operationId: getTrainingJobLogs
    parameters:
    - name: key
      in: header
      value: $inputs.apiKey
    requestBody:
      contentType: application/json
      payload:
        id: $steps.submitJob.outputs.trainingJobId
    successCriteria:
    - condition: $statusCode == 200
    onSuccess:
    - name: done
      type: end
  - stepId: cancelJob
    description: Cancel the training job when it has failed to free resources.
    operationId: cancelTrainingJob
    parameters:
    - name: key
      in: header
      value: $inputs.apiKey
    requestBody:
      contentType: application/json
      payload:
        id: $steps.submitJob.outputs.trainingJobId
    successCriteria:
    - condition: $statusCode == 200
  outputs:
    trainingJobId: $steps.submitJob.outputs.trainingJobId