Benchmarking Uncertainty Quantification Methods for Large Language Models with LM-Polygraph

TL;DR

This paper introduces LM-Polygraph, a comprehensive benchmark for evaluating uncertainty quantification methods in large language models, enabling consistent comparison and analysis of various techniques across multiple tasks.

Contribution

It provides a unified benchmark environment with state-of-the-art UQ baselines and evaluation protocols for large language models, addressing fragmentation in current research.

Findings

Identified the most effective UQ techniques across eleven tasks.

Demonstrated the importance of confidence normalization for interpretability.

Provided a scalable framework for future UQ research in LLMs.

Abstract

The rapid proliferation of large language models (LLMs) has stimulated researchers to seek effective and efficient approaches to deal with LLM hallucinations and low-quality outputs. Uncertainty quantification (UQ) is a key element of machine learning applications in dealing with such challenges. However, research to date on UQ for LLMs has been fragmented in terms of techniques and evaluation methodologies. In this work, we address this issue by introducing a novel benchmark that implements a collection of state-of-the-art UQ baselines and offers an environment for controllable and consistent evaluation of novel UQ techniques over various text generation tasks. Our benchmark also supports the assessment of confidence normalization methods in terms of their ability to provide interpretable scores. Using our benchmark, we conduct a large-scale empirical investigation of UQ and normalization techniques across eleven tasks, identifying the most effective approaches. Code: https://github.com/IINemo/lm-polygraph Benchmark: https://huggingface.co/LM-Polygraph