Skip to content

Documentation for Fine-Tuning Task Management API

POST /fine-tuning/task — create_fine_tuning_task

Description

Creates a fine-tuning task using a batch of training data and optional metadata. It checks for duplicate tasks using the idempotency_key.

Request Parameters

  • embedding_model_id (str): ID of the base embedding model.
  • batch_id (str, optional): Training data batch ID.
  • metadata (dict, optional): Custom metadata to associate with the task.
  • idempotency_key (str, optional): Unique key to prevent task duplication.
  • deploy_as_blue (bool, optional): Auto-deploy as active model after training.
  • wait_on_conflict (bool, optional): Wait if deployment conflicts occur.

Request JSON Example

{
  "embedding_model_id": "embed-123",
  "batch_id": "batch-xyz",
  "metadata": { "initiated_by": "admin" },
  "idempotency_key": "unique-key-abc",
  "deploy_as_blue": true,
  "wait_on_conflict": false
}

Response JSON Example

{
  "id": "task-001",
  "batch_id": "batch-xyz",
  "best_model_url": "https://storage/models/ft-001",
  "best_model_id": "model-ft-001",
  "metadata": { "initiated_by": "admin" },
  "idempotency_key": "unique-key-abc",
  "deploy_as_blue": true,
  "wait_on_conflict": false,
  "status": "pending"
}

GET /fine-tuning/task — get_fine_tuning_tasks

Description

Returns a list of fine-tuning tasks. Supports filtering by status and pagination.

Query Parameters

  • offset (int): Number of tasks to skip (default is 0).
  • limit (int): Maximum number of tasks to return (default is 100).
  • status (TaskStatus, optional): Filter tasks by current status (pending, processing, completed, etc.).

Response JSON Example

[
  {
    "id": "task-001",
    "batch_id": "batch-xyz",
    "best_model_url": "https://storage/models/ft-001",
    "best_model_id": "model-ft-001",
    "metadata": { "initiated_by": "admin" },
    "status": "completed"
  }
]

PUT /fine-tuning/task/{id}/restart — restart_fine_tuning_task

Description

Restarts a task by resetting its status and pushing it to the task queue.

Path Parameter

  • id (str): ID of the task to restart.

Response JSON Example

{
  "id": "task-001",
  "batch_id": "batch-xyz",
  "status": "pending"
}

PUT /fine-tuning/task/{id}/cancel — cancel_fine_tuning_task

Description

Cancels a running fine-tuning task. Terminates the associated background job and sets the task status to "canceled".

Path Parameter

  • id (str): ID of the task to cancel.

Response JSON Example

{
  "id": "task-001",
  "batch_id": "batch-xyz",
  "status": "canceled"
}