For AI agents: The documentation index is at https://docs.digitalocean.com/llms.txt. Markdown versions of pages use the same URL with index.html.md in place of the HTML page (for example, append index.html.md to the directory path instead of opening the HTML document).

You can compare two evaluations to identify differences in evaluation scores, configurations, metric results, token usage, latency, and candidate outputs. This can help you determine how changes to a candidate or evaluation configuration affect performance.

Compare Evaluations

To compare two evaluations, go to the DigitalOcean Control Panel, in the left menu, click DATA & LEARNING, and then click Evaluations.

In the Evaluations tab, find and select the evaluation you want to compare, and then in the top-right corner, click Compare Evaluation.

In the Evaluation A dropdown menu, select an evaluation. The evaluation you opened is selected by default.

In the Evaluation B dropdown menu, select the evaluation you want to compare against Evaluation A.

Review Comparison

The comparison shows the following sections:

• Overall Score: Compares each evaluation’s overall score, star metric, and pass threshold.

• Configuration: Compares the candidate, judge model, dataset, temperature, maximum tokens, and runtime used for each evaluation.

• Router task summary: For evaluations that use an Inference Router, breaks down the results by task category.

  Each task category includes the number of prompts, star metric pass rate, candidate token usage, and average latency.

  Click Show details for a task category to view its prompts.

• Metric Comparison: Compares the percentage of dataset entries that passed each selected metric.

• Performance: Compares candidate and judge token usage and the candidate’s average, P50, P90, and P95 latency.

• Per-Entry Results: When both evaluations use the same dataset, compares each dataset entry’s input and candidate outputs side by side.

  Note

  If the evaluations use different datasets, you cannot compare their per-entry results side by side.

When comparing changes between evaluation runs, use the same dataset when possible so differences in results are easier to attribute to changes in the candidate or evaluation configuration.

For guidance on designing representative test data, see Evaluation Dataset Design Best Practices.

For guidance on configuring and running dataset evaluations, see Dataset Evaluation Best Practices.