RbFT: Robust Fine-tuning for Retrieval-Augmented Generation against Retrieval Defects

TL;DR

This paper introduces RbFT, a fine-tuning approach that significantly improves the robustness of retrieval-augmented generation systems against retrieval errors and noise, ensuring more reliable language model outputs.

Contribution

The paper proposes RbFT, a novel fine-tuning method that enhances LLM resilience to retrieval defects, outperforming existing approaches in robustness while maintaining efficiency.

Findings

RbFT improves robustness across various retrieval conditions.

It surpasses existing methods in handling retrieval noise.

Maintains high inference efficiency and compatibility.

Abstract

Retrieval-augmented generation (RAG) enhances large language models (LLMs) by integrating external knowledge retrieved from a knowledge base. However, its effectiveness is fundamentally constrained by the reliability of both the retriever and the knowledge base. In real-world scenarios, imperfections in these components often lead to the retrieval of noisy, irrelevant, or misleading counterfactual information, ultimately undermining the trustworthiness of RAG systems. To address this challenge, we propose Robust Fine-Tuning (RbFT), a method designed to enhance the resilience of LLMs against retrieval defects through two targeted fine-tuning tasks. Experimental results demonstrate that RbFT significantly improves the robustness of RAG systems across diverse retrieval conditions, surpassing existing methods while maintaining high inference efficiency and compatibility with other robustness techniques.