Quantifying Variance in Evaluation Benchmarks

TL;DR

This paper investigates the sources and magnitude of variance in evaluation benchmarks for large language models, providing empirical measures, analysis of variance reduction techniques, and practical recommendations for more reliable model comparisons.

Contribution

It introduces metrics for quantifying benchmark variance, evaluates variance across models, and explores methods to reduce variance, aiding more accurate model evaluation.

Findings

Simple framing changes can reduce variance in smaller models.

Item analysis techniques have limited success in variance reduction.

Variance considerations are crucial for fair model comparisons.

Abstract

Evaluation benchmarks are the cornerstone of measuring capabilities of large language models (LLMs), as well as driving progress in said capabilities. Originally designed to make claims about capabilities (or lack thereof) in fully pretrained models, evaluation benchmarks are now also extensively used to decide between various training choices. Despite this widespread usage, we rarely quantify the variance in our evaluation benchmarks, which dictates whether differences in performance are meaningful. Here, we define and measure a range of metrics geared towards measuring variance in evaluation benchmarks, including seed variance across initialisations, and monotonicity during training. By studying a large number of models -- both openly available and pretrained from scratch -- we provide empirical estimates for a variety of variance metrics, with considerations and recommendations for practitioners. We also evaluate the utility and tradeoffs of continuous versus discrete performance measures and explore options for better understanding and reducing this variance. We find that simple changes, such as framing choice tasks (like MMLU) as completion tasks, can often reduce variance for smaller scale ($\sim$7B) models, while more involved methods inspired from human testing literature (such as item analysis and item response theory) struggle to meaningfully reduce variance. Overall, our work provides insights into variance in evaluation benchmarks, suggests LM-specific techniques to reduce variance, and more generally encourages practitioners to carefully factor in variance when comparing models.