Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Finetune-RAG: Fine-Tuning Language Models to Resist Hallucination in Retrieval-Augmented Generation

Zhan Peng Lee, Andre Lin, Calvin Tan

TL;DR

<3-5 sentence high-level summary>

Abstract

Retrieval-Augmented Generation (RAG) has emerged as a powerful framework to improve factuality in large language models (LLMs) by grounding their outputs in retrieved documents. However, ensuring perfect retrieval of relevant information remains challenging, and when irrelevant content is passed downstream to an LLM, it can lead to hallucinations. In this work, we propose Finetune-RAG, a simple and effective fine-tuning approach that features the first-of-its-kind RAG training dataset constructed to mimic real-world imperfections. Experimental results show that Finetune-RAG improves factual accuracy by 21.2% over the base model. We also propose Bench-RAG, an LLM-as-a-judge evaluation pipeline that stress tests models under realistic imperfect retrieval scenarios. Our codebase and dataset are fully open sourced for community use.

Finetune-RAG: Fine-Tuning Language Models to Resist Hallucination in Retrieval-Augmented Generation

TL;DR

<3-5 sentence high-level summary>

Abstract

Retrieval-Augmented Generation (RAG) has emerged as a powerful framework to improve factuality in large language models (LLMs) by grounding their outputs in retrieved documents. However, ensuring perfect retrieval of relevant information remains challenging, and when irrelevant content is passed downstream to an LLM, it can lead to hallucinations. In this work, we propose Finetune-RAG, a simple and effective fine-tuning approach that features the first-of-its-kind RAG training dataset constructed to mimic real-world imperfections. Experimental results show that Finetune-RAG improves factual accuracy by 21.2% over the base model. We also propose Bench-RAG, an LLM-as-a-judge evaluation pipeline that stress tests models under realistic imperfect retrieval scenarios. Our codebase and dataset are fully open sourced for community use.
Paper Structure (41 sections, 6 equations, 2 figures, 1 table)

This paper contains 41 sections, 6 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Background
  3. Retrieval-Augmented Generation
  4. Hallucination in Language Models
  5. Related Works
  6. Mitigating Hallucination with Synthetic Prompt Tuning
  7. Refusal-Aware Fine-Tuning
  8. Constrained Reasoning with Decompose-and-Query (D&Q)
  9. Methodology
  10. Problem Setup
  11. Prompt Construction
  12. System Message
  13. User Message
  14. Experimental Setup
  15. Model
  16. ...and 26 more sections

Figures (2)

  • Figure 1: Evaluation results across training steps (Baseline format). Accuracy is plotted on the right y-axis, and other metrics use the left y-axis.
  • Figure 2: Evaluation results across training steps (XML format). Accuracy is plotted on the right y-axis, and other metrics use the left y-axis.