LongFaith: Enhancing Long-Context Reasoning in LLMs with Faithful Synthetic Data

2025-02-18

AI Summary

Existing long-context large language models (LLMs) relying on synthetic data augmentation often suffer from diminished reasoning faithfulness due to information distortion, missing citations, and factual inconsistencies. To address this, we propose a truth-aligned and citation-driven faithful synthesis paradigm—the first framework integrating ground-truth constraints with citation-aware prompting. Our approach systematically models three core dimensions of faithfulness: verifiability, attributability, and context anchoring. It combines truth-guided instruction synthesis, multi-stage filtering, and structured annotation techniques. We release two high-quality, open-source datasets—LongFaith-SFT and LongFaith-PO—and fine-tune LLMs using them. Empirical evaluation shows substantial improvements on LongBench and multi-hop reasoning benchmarks. Ablation studies confirm the effectiveness and scalability of our method, demonstrating that grounding synthesis in verifiable facts and explicit citations significantly enhances faithful long-context reasoning.

Technology Category

• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Machine Learning: Large Multimodal Models (LMMs)
• Knowledge Representation and Reasoning: Reasoning with Beliefs

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI

Abstract

Despite the growing development of long-context large language models (LLMs), data-centric approaches relying on synthetic data have been hindered by issues related to faithfulness, which limit their effectiveness in enhancing model performance on tasks such as long-context reasoning and question answering (QA). These challenges are often exacerbated by misinformation caused by lack of verification, reasoning without attribution, and potential knowledge conflicts. We propose LongFaith, a novel pipeline for synthesizing faithful long-context reasoning instruction datasets. By integrating ground truth and citation-based reasoning prompts, we eliminate distractions and improve the accuracy of reasoning chains, thus mitigating the need for costly verification processes. We open-source two synthesized datasets, LongFaith-SFT and LongFaith-PO, which systematically address multiple dimensions of faithfulness, including verified reasoning, attribution, and contextual grounding. Extensive experiments on multi-hop reasoning datasets and LongBench demonstrate that models fine-tuned on these datasets significantly improve performance. Our ablation studies highlight the scalability and adaptability of the LongFaith pipeline, showcasing its broad applicability in developing long-context LLMs.

Problem

• Improving long-context reasoning in LLMs
• Mitigating misinformation in synthetic data
• Enhancing faithfulness in reasoning datasets

Innovation

• Synthesizes faithful long-context reasoning datasets
• Integrates ground truth and citation-based prompts
• Open-sources LongFaith-SFT and LongFaith-PO datasets