推测式RAG中保证忠实证据抽取的约束混合解码
原标题：Guaranteeing Faithful Evidence Extraction in Speculative Retrieval-Augmented Generation
AI 摘要
该论文提出 Constrained Hybrid Decoding (CHyD)，一种面向推测式检索增强生成（speculative RAG）的「忠实优先」解码范式。传统推测解码以推理效率为目标，而 CHyD 在模型进入抽取模式时施加硬解码约束，将生成限制为检索文档中连续且逐字存在的片段，从而保证输出中任何引用片段都逐字来自上下文。作者在多种抽象式、抽取式和半抽取式 QA 基准（含航空维修等技术数据集）上评估，发现现有混合方法常出现引用片段幻觉，技术领域精确抽取准确率大幅下降，而 CHyD 在不同模型下均接近完美的抽取忠实度，虽在流畅度指标上有取舍，但提升了精确答案正确率，适合安全关键场景。
正文节选
Guaranteeing Faithful Evidence Extraction in Speculative Retrieval-Augmented Generation Abstract. Large Language Models (LLMs) are increasingly used as interfaces for information retrieval, but they remain prone to hallucinations and faithfulness errors, in which the generated answers diverge from the retrieved evidence. While Retrieval-Augmented Generation (RAG) and recent hybrid or semi-extractive approaches mitigate this issue, they do not guarantee that quoted or extracted spans are verbatim