PrefixNLI：在事实不一致出现时即刻检测
PrefixNLI: Detecting Factual Inconsistencies as Soon as They Arise
AI总结：
本文将蕴含检测任务泛化至任意文本前缀，提出专用模型 MiniTruePrefixes，在检测事实不一致上显著优于基线，并集成至解码框架大幅提升摘要事实一致性，使小模型性能媲美大模型。
摘要：
自然语言推理（NLI）模型已通过多种方式被用于提升 LLM 输出的事实性。这通常通过应用 NLI 模型来判断模型输出是否由假定证据所蕴含，从而触发某些纠正措施来实现，例如在推理时进行束重排序或在训练期间提供 RL 奖励。尽管 NLI 模型被训练用于检测完整句子上的事实不一致，但在常见的自回归生成架构中，针对不断演变的文本前缀进行决策是在解码期间完成的。针对这一设定，我们将蕴含检测任务泛化以适用于任意文本前缀，并提出其在提升生成忠实度方面的效用。我们为该任务提供了合适的评估与训练数据，并训练了 MiniTruePrefixes，这是一种新颖的专用模型，能更好地检测文本前缀上的事实不一致，在前缀级蕴含任务上比类似的基线 NLI 模型高出 5-14 个 F1 点。我们进一步证明，将 MiniTruePrefixes 集成到受控解码框架中，能显著改善抽象摘要中的事实一致性。在 MiniTruePrefixes 的引导下，LLaMA-3.2-3B-Instruct 在忠实度和运行时间上匹配了同模型家族中的 8B 模型，且仅使用一半的内存。
英文摘要：
Natural Language Inference (NLI) models have been used in various ways to improve the factuality of LLM outputs. This is typically done by applying an NLI model to judge whether the model output is entailed from the supposed evidence, triggering some corrective actions, such as beam reranking at inference time or RL rewards during training. While NLI models are trained to detect factual inconsistencies over complete sentences, decisions in the common autoregressive generation architecture are made for each evolving text prefix, during decoding. Addressing this setting, we generalize the entailment detection task to apply over arbitrary text prefixes, and suggest its utility for improving generation faithfulness. Providing suitable evaluation and training datasets for this task, we train MiniTruePrefixes, a novel specialized model that better detects factual inconsistencies over text prefixes, outperforming comparable baseline NLI models by 5-14 F1 points in prefix-level entailment. We further demonstrate that integrating MiniTruePrefixes into a controlled decoding framework substantially improves factual consistency in abstractive summarization. When guided by MiniTruePrefixes, LLaMA-3.2-3B-Instruct matches the faithfulness and runtime of the 8B model from the same model family, while using only half the memory.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
