← All questions
EvaluationMedium· Benchmarking· Normalization

Why does MLE-Bench use medal rate rather than raw leade...

MLE-Bench uses medal rate to normalize performance across various competitions that utilize incompatible scoring scales. While this provides a useful metric, it is a step function that can suppress evidence of continuous progress. Therefore, it is best practice to report both the medal rate and the percentile distribution to provide a complete picture of the model’s performance.

Code playground (Python)