Calibrating Large Language Models with Sample Consistency

TL;DR

This paper introduces a novel calibration method for Large Language Models based on consistency across multiple samples, demonstrating improved confidence estimation and potential performance gains across various models and datasets.

Contribution

The study proposes a consistency-based calibration approach for LLMs, outperforming existing methods and providing practical guidance for effective confidence estimation.

Findings

Consistency measures outperform traditional calibration methods.

Larger sample sizes and explanations improve calibration.

Instruction tuning complicates confidence calibration.

Abstract

Accurately gauging the confidence level of Large Language Models' (LLMs) predictions is pivotal for their reliable application. However, LLMs are often uncalibrated inherently and elude conventional calibration techniques due to their proprietary nature and massive scale. In this work, we explore the potential of deriving confidence from the distribution of multiple randomly sampled model generations, via three measures of consistency. We perform an extensive evaluation across various open and closed-source models on nine reasoning datasets. Results show that consistency-based calibration methods outperform existing post-hoc approaches. Meanwhile, we find that factors such as intermediate explanations, model scaling, and larger sample sizes enhance calibration, while instruction-tuning makes calibration more difficult. Moreover, confidence scores obtained from consistency have the potential to enhance model performance. Finally, we offer practical guidance on choosing suitable consistency metrics for calibration, tailored to the characteristics of various LMs.