The table view is the default view when you open the UI. It shows a list of evaluation rows which you can click to inspect. Every evaluation row corresponds to a single rollout.

​
How to open the table view

To know if you are in the table view, check that the Table tab is selected in the top left corner of the UI.
Table View

Make sure you are in the Table View by checking the top left corner of the UI for the Table tab.

​
Inspecting a row

In the table view, you will see a list of evaluation rows. For each row, you can see:
  • Name: the test function name
  • Status: either running, finished, stopped, or error
  • Invocation ID: auto-generated by EP for every invocation
  • Rollout ID: auto-generated by EP for every rollout
  • Model: the model used for the evaluation
  • Score: found in evaluation_result.score
  • Created: the timestamp of when the row was created
Table View

List of evaluation rows.

To inspect a row, hover over the row you want to inspect and click to expand. When you expand a row, you can see the trajectory of the rollout as well as other metadata like evaluation results, IDs, input metadata, and eval metadata.

Click on a row to inspect the evaluation.

​
Chat Interface

On the left side of an expanded row, you can see the chat interface. This is where you can see the trajectory of the rollout to inspect the model’s responses and tool calls.
Chat Interface

On the left side of an expanded row, you can see the chat interface.

​
Metadata

On the right side of an expanded row, you can see the metadata. This is where you can see the evaluation results, IDs, input metadata, and eval metadata.
Metadata

On the right side of an expanded row, you can see the metadata.

​
Filtering

Above the table, you can see a section for configuring filters. You can filter based on any attribute of the evaluation row.
Filtering

The filter section above the table.

​
Filtering based on Invocation ID

Often times you just want to see the rollouts for a single invocation. To do this, you can easily click on the funnel icon next to the invocation ID in the table. This will automatically add a filter for the invocation ID to the table.

Click on the funnel icon next to the invocation ID to filter the table by invocation ID.

​
Custom filters

You can also create custom filters by clicking on the + Add Filter Group button above the table. Then you can choose to filter by AND or OR and add filters to the group by clicking on the + Add Filter to Group button.

Click on the + Add Filter Group button above the table to create a custom filter. In this example, we filter for scores equal to 0, models with gpt in the name, and a specific run_id.

​
Viewing live rollouts

When it takes a long time to run an eval, it can be helpful to see the live rollouts so you can track the progress of an eval and catch unexpected errors or problems. Whenever you run an @evaluation_test, the UI automatically shows running tests and you can watch rollouts live in the chat interface. When a test finishes, detailed evaluation results appear to the right of the chat. Checkout this example of a test running in VSCode and the UI updating with the rollout.

Expand running rows to see the chat interface update with the rollout.

​
Next Steps

Often times you want to ask questions like “how did the model perform on this eval across this dataset?” or “which model should I use for my application?”. Creating and running evals helps you answer these questions, but answering these questions requires computing some aggregate metrics across a set of evaluation rows. To do this, you can use the Pivot View to pivot the data and see the data in a different way.