Token attribution methods provide intuitive explanations for language model outputs by identifying causally important input tokens. However, as modern LLMs increasingly rely on extended reasoning chains, existing schemes face two critical challenges: (1) efficiency bottleneck, where attributing a target span of M tokens within a context of length N requires O(M*N) operations, making long-context attribution prohibitively slow; and (2) faithfulness drop, where intermediate reasoning tokens absorb attribution mass, preventing importance from propagating back to the original input. To address these, we introduce FlashTrace, an efficient multi-token attribution method that employs span-wise aggregation to compute attribution over multi-token targets in a single pass, while maintaining faithfulness. Moreover, we design a recursive attribution mechanism that traces importance through intermediate reasoning chains back to source inputs. Extensive experiments on long-context retrieval (RULER) and multi-step reasoning (MATH, MorehopQA) tasks demonstrate that FlashTrace achieves over 130x speedup over existing baselines while maintaining superior faithfulness. We further analyze the dynamics of recursive attribution, showing that even a single recursive hop improves faithfulness by tracing importance through the reasoning chain.

翻译：词元归因方法通过识别具有因果重要性的输入词元，为语言模型的输出提供直观解释。然而，随着现代大语言模型日益依赖扩展推理链，现有方案面临两大关键挑战：(1) 效率瓶颈：在长度为 N 的上下文中对 M 个词元的目标片段进行归因需要 O(M*N) 次运算，使得长上下文归因过程极其缓慢；(2) 忠实性下降：中间推理词元会吸收归因权重，阻碍重要性回传至原始输入。为解决这些问题，我们提出了 FlashTrace，一种高效的多词元归因方法。该方法采用片段级聚合策略，单次前向传播即可计算多词元目标的归因，同时保持忠实性。此外，我们设计了一种递归归因机制，能够沿中间推理链将重要性追溯回源输入。在长上下文检索（RULER）与多步推理（MATH、MorehopQA）任务上的大量实验表明，FlashTrace 在保持更优忠实性的同时，相比现有基线实现了超过 130 倍的加速。我们进一步分析了递归归因的动态特性，结果表明即使仅进行单次递归跳转，也能通过沿推理链追溯重要性来提升忠实性。