Purifying Large Language Models by Ensembling a Small Language Model

TL;DR

This paper introduces a straightforward ensembling method combining large and small language models to reduce risks like copyright infringement and data poisoning while maintaining LLM performance.

Contribution

It proposes an easy-to-implement ensembling technique with theoretical guarantees to purify LLMs from untrusted data effects.

Findings

Ensembling preserves LLM performance.

Effectively mitigates copyright, poisoning, privacy issues.

Validated through comprehensive experiments.

Abstract

The emerging success of large language models (LLMs) heavily relies on collecting abundant training data from external (untrusted) sources. Despite substantial efforts devoted to data cleaning and curation, well-constructed LLMs have been reported to suffer from copyright infringement, data poisoning, and/or privacy violations, which would impede practical deployment of LLMs. In this study, we propose a simple and easily implementable method for purifying LLMs from the negative effects caused by uncurated data, namely, through ensembling LLMs with benign and small language models (SLMs). Aside from theoretical guarantees, we perform comprehensive experiments to empirically confirm the efficacy of ensembling LLMs with SLMs, which can effectively preserve the performance of LLMs while mitigating issues such as copyright infringement, data poisoning, and privacy violations.