Efficient multi-prompt evaluation of LLMs

TL;DR

This paper introduces PromptEval, a method for efficiently estimating the performance distribution of large language models across many prompt variations, improving robustness and reproducibility of evaluations.

Contribution

PromptEval provides a novel approach to estimate performance distributions over numerous prompts, enabling more reliable and comprehensive LLM evaluation under limited budgets.

Findings

Accurately estimates performance quantiles across 100 prompts on MMLU

Consistently estimates performance distribution across benchmarks

Effective in LLM-as-a-judge and prompt selection applications

Abstract

Most popular benchmarks for comparing LLMs rely on a limited set of prompt templates, which may not fully capture the LLMs' abilities and can affect the reproducibility of results on leaderboards. Many recent works empirically verify prompt sensitivity and advocate for changes in LLM evaluation. In this paper, we consider the problem of estimating the performance distribution across many prompt variants instead of finding a single prompt to evaluate with. We introduce PromptEval, a method for estimating performance across a large set of prompts borrowing strength across prompts and examples to produce accurate estimates under practical evaluation budgets. The resulting distribution can be used to obtain performance quantiles to construct various robust performance metrics (e.g., top 95% quantile or median). We prove that PromptEval consistently estimates the performance distribution and demonstrate its efficacy empirically on three prominent LLM benchmarks: MMLU, BIG-bench Hard, and LMentry; for example, PromptEval can accurately estimate performance quantiles across 100 prompt templates on MMLU with a budget equivalent to two single-prompt evaluations. Moreover, we show how PromptEval can be useful in LLM-as-a-judge and best prompt identification applications.