Workflows API

Workflows are designed to manage and run multiple atomic jobs. Jobs are designed to execute specific tasks on the Ray cluster, for example performing inference of a model using a dataset that was uploaded. Workflows are designed to be a way to link these jobs together.

For example, in order to evaluate how a model performs on a summarization task, there are two jobs that must be executed. First, an inference job runs in order to run inference with a model and collect the outputs. Next, an evaluation job runs in order to evaluate the outputs of the inference job against the ground truth stored in the dataset.

Endpoints

POST /api/v1/workflows/

Create Workflow

A workflow is a single execution for an experiment. A workflow is a collection of 1 or more jobs. It must be associated with an experiment id, which means you must already have created an experiment and have that ID in the request.

Status Codes:
GET /api/v1/workflows/{workflow_id}

Get Workflow

Get a workflow by ID.

Parameters:

  • workflow_id (string)

Status Codes:
DELETE /api/v1/workflows/{workflow_id}

Delete Workflow

Delete a workflow by ID.

Parameters:

  • workflow_id (string)

Status Codes:
GET /api/v1/workflows/{workflow_id}/logs

Get Workflow Logs

Get the logs for a workflow.

Parameters:

  • workflow_id (string)

Status Codes: