Runs

Customizing run reports

The run report is a single source of truth for you to understand exactly how your AI system is performing during your experiments or pre-release testing. In order to see exactly the data you need to figure out insights and take decisions, we allow customisation of the report in multiple ways.

Toggle columns

For prompt/workflow runs, by default we only show the input from the dataset and the retrieved context (if applicable) and output from the run. However, there might be cases where you want to see other dataset columns to analyse the output. Similarly, you may want to hide some already visible columns in order to see limited data while analysing evaluations. To show/hide columns, follow the below steps:

On the run report table header, you will see a button called ‘Toggle columns’.

Clicking on this will open a dropdown with options of all columns from the dataset and the run result

Select columns that you want to be visible. Use search if you have a lot of columns.

Test run summary

You can also do this via the column header cell by hovering and clicking the ‘three dot button > Hide column’

Test run summary

Pinning columns

While analysing the report, you can horizontally scroll to see all columns. Incase you have certain columns that you would want to always have as reference while looking at rest of the data, you can pin them. Eg. Pin inputs columns to the left and a certain evaluation column to the right while analysing the retrieved context, output, etc.

To pin a column, click the 3 dots buttons on the row header and choose ‘Pin to left’ or ‘Pin to right’. You can also unpin the column in the same way.

Test run summary

Re-ordering columns

You can easily re-order all columns of the table by holding down the button shown in the below screenshot and dragging the column

Test run summary

Search and filter

In case of large reports with a lot of entries, you can use the search or filters to easily reach the relevant entries you care about. Filtering allows you to put a combination of criteria. These could be performance metrics or evaluation scores.

You can also directly filter out the results that are failing on a particular metric by clicking the filter icon next to its score in the summary card.

Test run summary

Test run summary

On this page