Evaluation Datasets

Evaluation Datasets are collections of test data used to measure the performance of your AI workflows.

Overview

The Datasets page provides a centralized view of all your evaluation datasets:

Managing Datasets

In the Datasets list, you can:

• View all available evaluation datasets
• See when each dataset was created
• Filter datasets by their active status
• Access detailed information for any dataset

Dataset Details

When you select a dataset, you'll see its detailed information:

The Dataset Details page includes:

• Dataset name and description
• Last updated timestamp
• Dataset contents overview
• List of dataset entries with their inputs
• Reference outputs (expected results)
• History of evaluation runs using this dataset

Dataset Entries

Each dataset contains entries that will be used to evaluate your workflows:

• ID: A unique identifier for each entry
• Inputs: The data that will be fed into your workflow (e.g., product descriptions, queries)
• Reference Outputs: The expected outputs to compare against your workflow's results

Table Settings

You can customize how dataset entries are displayed using the Table Settings:

Available display options include:

• Show/Hide Inputs: Toggle visibility of input data
• Show/Hide State: Toggle visibility of intermediate state information
• Show/Hide Reference Outputs: Toggle visibility of expected outputs

Latest Evaluation Runs

The Dataset Details page also shows a history of recent evaluation runs performed with this dataset:

• Timestamp of each evaluation run
• Workflow name that was evaluated
• Brief description of the workflow
• Options to open the run details or workflow

Running Evaluations

To run an evaluation against a dataset:

1. Navigate to the Dataset Details page
2. Select the workflow you want to evaluate
3. Configure any evaluation parameters
4. Click "Run Evaluation"

Best Practices

For optimal evaluation dataset management:

• Include diverse test cases that cover edge cases
• Maintain reference outputs that reflect desired behavior
• Version datasets when significant changes are made
• Use consistent formatting for inputs and reference outputs

Next Steps

• View Evaluation Runs - Check the results of workflow evaluations
• Return to Dashboard