发现等待前的内部状态能调控推理模式，解释模型为何选择回溯或重新开始。
Internal states before wait modulate reasoning patterns
- 通过跨编码器分析等待前的模型内部状态，定位影响推理的关键特征
- 识别出少量特征可显著提升或抑制等待令牌概率，与不同推理行为相关
- 适用于研究大模型推理机制、改进自修正能力的科研人员
先前研究表明，推理模型的表现很大程度上取决于其推理与自我修正的能力。一个显著特征是‘等待令牌’（token wait），常标志回溯等推理行为。然而，为何模型选择以这种复杂方式推理仍不清楚，限制了对高效推理机制的理解。本文探究等待令牌前的模型内部状态是否包含调控后续推理过程的信息。我们在 DeepSeek-R1-Distill-Llama-8B 及其基础版本的多个层级训练跨编码器，并引入一种跨编码器设定下的潜在属性分析技术。定位到一组小规模特征，可促进或抑制等待令牌的概率。通过分析最大激活样本及因果干预实验，发现这些特征确实与不同推理模式相关，如从头重启、回忆先前知识、表达不确定性及双重检查。
原文摘要 · Abstract (English)
Prior work has shown that a significant driver of performance in reasoning models is their ability to reason and self-correct. A distinctive marker in these reasoning traces is the token wait, which often signals reasoning behavior such as backtracking. Despite being such a complex behavior, little is understood of exactly why models do or do not decide to reason in this particular manner, which limits our understanding of what makes a reasoning model so effective. In this work, we address the question whether model's latents preceding wait tokens contain relevant information for modulating the subsequent reasoning process. We train crosscoders at multiple layers of DeepSeek-R1-Distill-Llama-8B and its base version, and introduce a latent attribution technique in the crosscoder setting. We locate a small set of features relevant for promoting/suppressing wait tokens' probabilities. Finally, through a targeted series of experiments analyzing max activating examples and causal interventions, we show that many of our identified features indeed are relevant for the reasoning process and give rise to different types of reasoning patterns such as restarting from the beginning, recalling prior knowledge, expressing uncertainty, and double-checking.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。