Skip to main content

View workflow

Overview​

This page explains how to view a workflow you created and what appears in the workflow dashboard. For instructions on creating a workflow, see Create a new workflow.

Instructions​

To view a workflow:

  1. Click Workflows in the left navigation bar.
  2. In the Workflows table, select the workflow you want to view.

The workflow dashboard displays the following details:

  • Model host: The model host selected during workflow creation.
  • LLM model: The LLM model selected during workflow creation.
  • Document / Collection: The document or collection selected during workflow creation.
  • Created test: Displays once tests are created.
  • Created evaluation: Displays once evaluations are created. View workflow details

A workflow consists of the following eight steps. Each step becomes available once the previous step is complete:

  1. Topic modeling: Identifies the most relevant topics in the provided text.
  2. Text generation: Generates test cases based on the text and identified topics.
  3. Generated test validation: Validates the generated test suite.
  4. Automated evaluation: Evaluates the model using the generated test cases.
  5. Human evaluation calibration: Calibrates evaluation metrics through human input to improve accuracy and reliability.
  6. Weakness detection: Detects weak areas by evaluating performance across topics and query types.
  7. Robustness testing (input variations): Perturbs test cases to simulate different scenarios and evaluate model robustness.
  8. Robustness testing (adversarial inputs): Tests model robustness using adversarial inputs.

Clone workflow​

To create a new workflow by cloning an existing one:

  1. Click Clone. Click clone
  2. Update the Workflow name, Description, Model host, LLM model, or Parameters fields as needed.
  3. Click Create.

Delete workflow​

warning

This action permanently deletes the workflow from the server. It cannot be undone.

To delete a workflow:

  1. Click Delete. Click delete
  2. Click Delete again to confirm.
Feedback