DART：将自回归推理蒸馏为静默思维
DART: Distilling Autoregressive Reasoning to Silent Thought
AI总结：
本文提出自蒸馏框架 DART，通过 REM 将自回归 CoT 推理对齐并压缩为少量静默思维 token，使模型推理时无需额外延迟即可直接作答并显著优于非自回归基线。
摘要：
思维链（Chain-of-Thought，CoT）推理显著提升了大型语言模型（LLM）解决复杂任务的能力。然而，其自回归范式会带来显著的计算开销，阻碍其在延迟敏感应用中的部署。为解决这一问题，我们提出 DART（Distilling Autoregressive Reasoning to Silent Thought，将自回归推理蒸馏为静默思维），这是一个自蒸馏框架，使 LLM 能够用非自回归的静默思维（Silent Thought，ST）替代自回归 CoT。具体而言，DART 引入两条训练路径：用于传统推理的 CoT 路径，以及用于从少量 ST token 直接生成答案的 ST 路径。ST 路径利用轻量级的推理演化模块（Reasoning Evolvement Module，REM），将其隐藏状态与 CoT 路径对齐，使 ST token 演化为信息丰富的嵌入。推理期间仅激活 ST 路径，利用不断演化的 ST token 直接给出答案。大量实验结果表明，与现有非自回归基线相比，DART 在不增加额外推理延迟的情况下带来显著性能提升，是高效推理的一种可行替代方案。
英文摘要：
Chain-of-Thought (CoT) reasoning has significantly advanced Large Language Models (LLMs) in solving complex tasks. However, its autoregressive paradigm leads to significant computational overhead, hindering its deployment in latency-sensitive applications. To address this, we propose \textbf{DART} (\textbf{D}istilling \textbf{A}utoregressive \textbf{R}easoning to Silent \textbf{T}hought), a self-distillation framework that enables LLMs to replace autoregressive CoT with non-autoregressive Silent Thought (ST). Specifically, DART introduces two training pathways: the CoT pathway for traditional reasoning and the ST pathway for generating answers directly from a few ST tokens. The ST pathway utilizes a lightweight Reasoning Evolvement Module (REM) to align its hidden states with the CoT pathway, enabling the ST tokens to evolve into informative embeddings. During inference, only the ST pathway is activated, leveraging evolving ST tokens to deliver the answer directly. Extensive experimental results demonstrate that DART offers significant performance gains compared with existing non-autoregressive baselines without extra inference latency, serving as a feasible alternative for efficient reasoning.
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。
