CaLM: Contrasting Large and Small Language Models to Verify Grounded Generation

TL;DR

CaLM introduces a verification framework that uses contrasting large and small language models to improve grounded generation by ensuring responses align with cited sources, enhancing credibility without fine-tuning.

Contribution

The paper proposes CaLM, a novel framework that leverages contrasting small and large language models to verify and refine grounded responses based on cited sources.

Findings

Achieves 1.5% to 7% improvement in QA accuracy

Does not require model fine-tuning

Effective across multiple open-domain datasets

Abstract

Grounded generation aims to equip language models (LMs) with the ability to produce more credible and accountable responses by accurately citing verifiable sources. However, existing methods, by either feeding LMs with raw or preprocessed materials, remain prone to errors. To address this, we introduce CaLM, a novel verification framework. CaLM leverages the insight that a robust grounded response should be consistent with information derived solely from its cited sources. Our framework empowers smaller LMs, which rely less on parametric memory and excel at processing relevant information given a query, to validate the output of larger LMs. Larger LM responses that closely align with the smaller LMs' output, which relies exclusively on cited documents, are verified. Responses showing discrepancies are iteratively refined through a feedback loop. Experiments on three open-domain question-answering datasets demonstrate significant performance gains of 1.5% to 7% absolute average without any required model fine-tuning.