tinyBenchmarks: evaluating LLMs with fewer examples

TL;DR

This paper introduces tiny benchmarks and evaluation strategies that significantly reduce the number of examples needed to reliably assess large language models, making evaluations more efficient and cost-effective.

Contribution

It proposes methods to accurately estimate LLM performance using fewer examples and releases tiny benchmark datasets and tools for efficient evaluation.

Findings

100 curated examples suffice for MMLU performance estimation

Tiny benchmarks reliably reproduce original evaluation results

Evaluation tools enable cost-effective LLM assessment

Abstract

The versatility of large language models (LLMs) led to the creation of diverse benchmarks that thoroughly test a variety of language models' abilities. These benchmarks consist of tens of thousands of examples making evaluation of LLMs very expensive. In this paper, we investigate strategies to reduce the number of evaluations needed to assess the performance of an LLM on several key benchmarks. For example, we show that to accurately estimate the performance of an LLM on MMLU, a popular multiple-choice QA benchmark consisting of 14K examples, it is sufficient to evaluate this LLM on 100 curated examples. We release evaluation tools and tiny versions of popular benchmarks: Open LLM Leaderboard, MMLU, HELM, and AlpacaEval 2.0. Our empirical analysis demonstrates that these tools and tiny benchmarks are sufficient to reliably and efficiently reproduce the original evaluation results.