思维草图：通过自适应认知启发式草图实现高效大语言模型推理

Sketch-of-Thought: Efficient LLM Reasoning with Adaptive Cognitive-Inspired Sketching

March 7, 2025

作者: Simon A. Aytes, Jinheon Baek, Sung Ju Hwang

cs.AI

摘要

近期大型语言模型的进展通过思维链（Chain of Thought, CoT）提示展现了卓越的推理能力，但往往伴随着中间输出过于冗长的问题，这增加了计算开销。我们提出了思维草图（Sketch-of-Thought, SoT），一种新颖的提示框架，它结合了认知启发的推理范式与语言约束，旨在最小化标记使用的同时保持推理准确性。SoT被设计为一个灵活的框架，能够整合任何基于认知科学的自定义推理范式，并通过三种具体范式——概念链（Conceptual Chaining）、块状符号（Chunked Symbolism）和专家词汇（Expert Lexicons）——进行实例化，每种范式针对不同的推理任务，并通过轻量级路由模型动态选择。在涵盖15个推理数据集、多种语言及多模态场景的综合评估中，我们展示了SoT实现了76%的标记减少，且对准确性的影响微乎其微。在数学推理和多跳推理等特定领域，它甚至在使用显著更少标记的同时提升了准确性。我们的代码已公开：https://www.github.com/SimonAytes/SoT。

English

Recent advances in large language models have demonstrated remarkable reasoning capabilities through Chain of Thought (CoT) prompting, but often at the cost of excessive verbosity in their intermediate outputs, which increases computational overhead. We introduce Sketch-of-Thought (SoT), a novel prompting framework that combines cognitive-inspired reasoning paradigms with linguistic constraints to minimize token usage while preserving reasoning accuracy. SoT is designed as a flexible framework that can incorporate any custom reasoning paradigms based on cognitive science, and we instantiate it with three such paradigms - Conceptual Chaining, Chunked Symbolism, and Expert Lexicons - each tailored to different reasoning tasks and selected dynamically via a lightweight routing model. Through comprehensive evaluation across 15 reasoning datasets with multiple languages and multimodal scenarios, we demonstrate that SoT achieves token reductions of 76% with negligible accuracy impact. In certain domains like mathematical and multi-hop reasoning, it even improves accuracy while using significantly fewer tokens. Our code is publicly available: https://www.github.com/SimonAytes/SoT.