Cer-Eval: Certifiable and Cost-Efficient Evaluation Framework for LLMs

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Wang, Ganghua, Chen, Zhaorun, Li, Bo, Xu, Haifeng
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866918011985723392
author Wang, Ganghua
Chen, Zhaorun
Li, Bo
Xu, Haifeng
author_facet Wang, Ganghua
Chen, Zhaorun
Li, Bo
Xu, Haifeng
contents As foundation models continue to scale, the size of trained models grows exponentially, presenting significant challenges for their evaluation. Current evaluation practices involve curating increasingly large datasets to assess the performance of large language models (LLMs). However, there is a lack of systematic analysis and guidance on determining the sufficiency of test data or selecting informative samples for evaluation. This paper introduces a certifiable and cost-efficient evaluation framework for LLMs. Our framework adapts to different evaluation objectives and outputs confidence intervals that contain true values with high probability. We use ``test sample complexity'' to quantify the number of test points needed for a certifiable evaluation and derive tight bounds on test sample complexity. Based on the developed theory, we develop a partition-based algorithm, named Cer-Eval, that adaptively selects test points to minimize the cost of LLM evaluation. Real-world experiments demonstrate that Cer-Eval can save 20% to 40% test points across various benchmarks, while maintaining an estimation error level comparable to the current evaluation process and providing a 95% confidence guarantee.
format Preprint
id arxiv_https___arxiv_org_abs_2505_03814
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Cer-Eval: Certifiable and Cost-Efficient Evaluation Framework for LLMs
Wang, Ganghua
Chen, Zhaorun
Li, Bo
Xu, Haifeng
Machine Learning
Artificial Intelligence
Computation and Language
As foundation models continue to scale, the size of trained models grows exponentially, presenting significant challenges for their evaluation. Current evaluation practices involve curating increasingly large datasets to assess the performance of large language models (LLMs). However, there is a lack of systematic analysis and guidance on determining the sufficiency of test data or selecting informative samples for evaluation. This paper introduces a certifiable and cost-efficient evaluation framework for LLMs. Our framework adapts to different evaluation objectives and outputs confidence intervals that contain true values with high probability. We use ``test sample complexity'' to quantify the number of test points needed for a certifiable evaluation and derive tight bounds on test sample complexity. Based on the developed theory, we develop a partition-based algorithm, named Cer-Eval, that adaptively selects test points to minimize the cost of LLM evaluation. Real-world experiments demonstrate that Cer-Eval can save 20% to 40% test points across various benchmarks, while maintaining an estimation error level comparable to the current evaluation process and providing a 95% confidence guarantee.
title Cer-Eval: Certifiable and Cost-Efficient Evaluation Framework for LLMs
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2505.03814