---
title: "Sketch-of-Thought: Efficient LLM Reasoning with Adaptive Cognitive-Inspired Sketching"
canonical_url: "https://www.modelscope.cn/papers/124263"
md_url: "https://www.modelscope.cn/papers/124263.md"
arxiv_id: 2503.05179
published: 2025-03-07
last_updated: 2025-03-07
authors:
  - "Simon A. Aytes"
  - "Jinheon Baek"
  - "Sung Ju Hwang"
model_name: "Sketch-of-Thought (SoT)"
model_developer: "KAIST, 深度自动有限公司"
domain:
  - "自然语言处理"
  - "深度学习"
  - "计算机视觉"
type:
  - "自然语言处理"
  - "深度学习"
  - "计算机视觉"
  - "Computation and Language (cs.CL)"
  - "Artificial Intelligence (cs.AI)"
  - "Machine Learning (cs.LG)"
arxiv_url: "https://arxiv.org/abs/2503.05179"
pdf_url: "https://arxiv.org/pdf/2503.05179.pdf"
code_link: "https://www.github.com/SimonAytes/SoT"
---

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

> 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…

「Sketch-of-Thought: Efficient LLM Reasoning with Adaptive Cognitive-Inspired Sketching」是 ModelScope 魔搭社区收录的论文，arXiv 2503.05179，作者为 Simon A. Aytes, Jinheon Baek, Sung Ju Hwang，发表于 2025-03-07，属于 自然语言处理、深度学习、计算机视觉 领域。

- **ArXiv**: 2503.05179
- **Published**: 2025-03-07
- **Authors**: Simon A. Aytes, Jinheon Baek, Sung Ju Hwang
- **Model**: Sketch-of-Thought (SoT)
- **Developer**: KAIST, 深度自动有限公司
- **Domain**: 自然语言处理, 深度学习, 计算机视觉
- **ArXiv URL**: https://arxiv.org/abs/2503.05179
- **PDF**: https://arxiv.org/pdf/2503.05179.pdf
- **Code**: https://www.github.com/SimonAytes/SoT

Source: https://www.modelscope.cn/papers/124263

---

> Sketch-of-Thought：精简推理过程，大幅提升大型语言模型效率

## 摘要

本文提出了一种名为Sketch-of-Thought (SoT)的新颖提示框架，旨在通过结合认知科学启发的推理范式和语言约束，减少大型语言模型（LLMs）在推理过程中的冗长输出，从而降低计算开销并保持推理准确性。SoT框架包括三个特定的推理范式：概念链接（Conceptual Chaining）、分块符号主义（Chunked Symbolism）和专家词典（Expert Lexicons），分别适用于不同的推理任务。这些范式由一个轻量级路由模型动态选择，以确保每个问题使用最合适的推理方法。实验结果表明，SoT在15个推理数据集上减少了76%的token使用量，并且在某些领域如数学和多步推理中提高了准确性。此外，SoT还展示了在多语言和多模态任务中的有效性。

## Abstract

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: this https URL.