用精简草稿代替冗长推理，让大模型更快更省地解决复杂问题。
Chain of Draft: Thinking Faster by Writing Less
- 模仿人类思维，生成极简但关键的中间推理步骤。
- 仅用7.6%的词元量，准确率媲美甚至超过传统链式思考。
- 适合追求高效推理、低延迟和低成本部署的场景。
大型语言模型在复杂推理任务中表现出色，通常依赖链式思考（CoT）提示，即通过详尽的逐步推理。然而，人类通常采用更高效的方式：生成简洁的中间思考草稿，仅保留必要信息。本文提出链式草稿（Chain of Draft, CoD），一种受人类认知过程启发的新范式，让模型在解题时生成最小化但富有信息量的中间输出。通过减少冗余，聚焦核心洞察，CoD 在多种推理任务中达到或超越CoT的准确率，同时仅使用约7.6%的词元量，显著降低计算成本与延迟。代码与数据已公开于 https://github.com/sileix/chain-of-draft。
原文摘要 · Abstract (English)
Large Language Models (LLMs) have demonstrated remarkable performance in solving complex reasoning tasks through mechanisms like Chain-of-Thought (CoT) prompting, which emphasizes verbose, step-by-step reasoning. However, humans typically employ a more efficient strategy: drafting concise intermediate thoughts that capture only essential information. In this work, we propose Chain of Draft (CoD), a novel paradigm inspired by human cognitive processes, where LLMs generate minimalistic yet informative intermediate reasoning outputs while solving tasks. By reducing verbosity and focusing on critical insights, CoD matches or surpasses CoT in accuracy while using as little as only 7.6% of the tokens, significantly reducing cost and latency across various reasoning tasks. Our code and data are available at https://github.com/sileix/chain-of-draft.
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。