Buffer of Thoughts: Thought-Augmented Reasoning with Large Language Models
Ling Yang1∗🖂, Zhaochen Yu1, Tianjun Zhang2, Shiyi Cao2, Minkai Xu3, Wentao Zhang1, Joseph E. Gonzalez2, Bin Cui1 1Peking University, 2UC Berkeley, 3Stanford University Project: https://github.com/YangLing0818/buffer-of-thought-llm Equal Contribution. 🖂
Abstract
We introduce Buffer of Thoughts (BoT), a novel and versatile thought-augmented reasoning approach for enhancing accuracy, efficiency and robustness of large language models (LLMs). Specifically, we propose meta-buffer to store a series of informative high-level thoughts, namely thought-template, distilled from the problem-solving processes across various tasks. Then for each problem, we retrieve a relevant thought-template and adaptively instantiate it with specific reasoning structures to conduct efficient reasoning. To guarantee the scalability and stability, we further propose buffer-manager to dynamically update the meta-buffer, thus enhancing the capacity of meta-buffer as more tasks are solved. We conduct extensive experiments on 10 challenging reasoning-intensive tasks, and achieve significant performance improvements over previous SOTA methods: 11% on Game of 24, 20% on Geometric Shapes and 51% on Checkmate-in-One. Further analysis demonstrate the superior generalization ability and model robustness of our BoT, while requiring only 12% of the cost of multi-query prompting methods (e.g., tree/graph of thoughts) on average. Notably, we find that our Llama3-8B + BoT has the po
中文速览
大语言模型在推理任务中常常要么依赖人工设计的示例、要么反复调用模型导致成本高昂，且无法积累跨任务的通用经验。Buffer of Thoughts（BoT）提出了一套"思维模板库"机制：先用一个问题精炼器提取任务的核心信息，再从一个名为 meta-buffer 的轻量级库中检索最相关的高层思维模板，然后将其实例化为具体推理步骤；每解决一个新问题，buffer-manager 就会把解题过程中归纳出的通用思路蒸馏成新模板存回库中，让库随使用不断扩充。在 10 个高难度推理任务上，BoT 相比此前最优方法分别提升了 11%（24点游戏）、20%（几何形状）和 51%（一步将死），同时平均只消耗多次查询方法（如思维树/思维图）约 12% 的计算成本，甚至让 Llama3-8B 有潜力超越 Llama3-70B。这项工作的意义在于，它首次将"积累并复用高层推理经验"的机制引入提示框架，让大模型像人一样举一反三，在不牺牲准确率的前提下大幅降低推理开销。
原文 arXiv:2406.04271;中英对照 + 大白话阅读 https://aha.fim.ai/paper/2406.04271v2