When Benchmarks are Targets: Revealing the Sensitivity of Large Language Model Leaderboards

TL;DR

This paper demonstrates that small changes in benchmark setups can significantly alter LLM rankings, highlighting the need for more robust evaluation methods beyond simple leaderboard scores.

Contribution

It systematically analyzes the sensitivity of LLM benchmark rankings to minor perturbations and proposes best practices for more reliable evaluation.

Findings

Minor perturbations can change rankings by up to 8 positions

Hybrid scoring methods improve robustness of evaluations

Current benchmarks are highly sensitive to small changes

Abstract

Large Language Model (LLM) leaderboards based on benchmark rankings are regularly used to guide practitioners in model selection. Often, the published leaderboard rankings are taken at face value - we show this is a (potentially costly) mistake. Under existing leaderboards, the relative performance of LLMs is highly sensitive to (often minute) details. We show that for popular multiple-choice question benchmarks (e.g., MMLU), minor perturbations to the benchmark, such as changing the order of choices or the method of answer selection, result in changes in rankings up to 8 positions. We explain this phenomenon by conducting systematic experiments over three broad categories of benchmark perturbations and identifying the sources of this behavior. Our analysis results in several best-practice recommendations, including the advantage of a hybrid scoring method for answer selection. Our study highlights the dangers of relying on simple benchmark evaluations and charts the path for more robust evaluation schemes on the existing benchmarks. The code for this paper is available at https://github.com/National-Center-for-AI-Saudi-Arabia/lm-evaluation-harness.