Skip to content

fine_tuning

Documentation for Fine-Tuning Task Management

create_fine_tuning_task

Functionality

Creates a new fine-tuning task based on the provided request body. It first retrieves the corresponding model iteration by the given embedding_model_id, validates the associated plugin, and checks for an existing task using an idempotency key. If an existing task is found, it returns that task; otherwise, a new task is created and dispatched via the fine-tuning worker.

Parameters

  • body: A FineTuningTaskRunRequest object containing task details such as embedding_model_id and an optional idempotency_key.

Usage

  • Purpose: Initiates a fine-tuning process by creating and sending a task that triggers model training or adjustments based on fine-tuning iteration data.

Example

Request payload:

{
  "embedding_model_id": "model123",
  "idempotency_key": "unique-key-456"
}

Expected response:

{
  "id": "task-id",
  "status": "PENDING"
}

get_fine_tuning_tasks

Functionality

Retrieves a list of fine-tuning tasks from the backend with optional filtering. It supports pagination via offset and limit and allows filtering by task status.

Parameters

  • offset: Number of tasks to skip.
  • limit: Number of tasks to return.
  • status: Optional task status filter to select specific tasks.

Usage

  • Purpose: Fetch fine-tuning tasks to monitor and manage processing, pending, or completed tasks.

Example

A sample HTTP GET request:

GET /task?offset=0&limit=20&status=processing

This will return a list of fine-tuning tasks matching the criteria.

restart_fine_tuning_task

Functionality

Restarts a fine-tuning task by resetting its status to pending. It checks for the existence of the task and its state. For tasks not in a processing state, it updates the status and dispatches the task with a new broker id.

Parameters

  • id: Identifier of the task to be restarted.

Usage

  • Purpose: Restart a fine-tuning task that might be stuck or failed, setting it to pending.

Example

from embedding_studio.api.api_v1.endpoints.fine_tuning import restart_fine_tuning_task
result = restart_fine_tuning_task('TASK_ID')

cancel_fine_tuning_task

Functionality

Cancels a fine-tuning task by aborting its ongoing execution using dramatiq_abort and updating its status to "canceled". It returns the updated task details on success.

Parameters

  • id: A string representing the unique identifier of the task.

Usage

This endpoint is used to cancel a fine-tuning task. It retrieves the task by its id, aborts the task processing if it exists, updates its status, and returns the updated task information.

Example

Assuming a task with ID "12345" exists, a client can cancel it using:

PUT /task/12345/cancel

The response will include the task details with status set to "canceled".