让大模型用工具调用方式高效推理，准确率更高、速度更快。
CoLT: Reasoning with Chain of Latent Tool Calls
- 把隐式推理转为可调用的工具，通过种子令牌触发外部小模型解码。
- 在4个数学数据集上准确率更高，推理长度更短，比基线快30%以上。
- 无需结构修改，兼容强化学习和不同解码器，适合部署优化场景。
链式思维（CoT）是提升大语言模型推理能力的关键技术，而隐式推理方法被提出以加速低效的逐标记推理过程。然而，现有隐式推理方法通常需要模型结构修改和大量训练，限制了其广泛应用。本文提出CoLT框架，将隐式推理实现为“工具调用”。不同于完全在隐空间中推理，CoLT生成包含推理步骤信息的种子令牌；当触发隐式工具调用时，一个小型外部模型以种子令牌的隐藏状态为输入，将其还原为完整的推理步骤。该方法确保主模型始终在显式标记空间中推理，保持原有能力的同时显著提升效率。在四个数学数据集上的实验表明，CoLT在准确率和推理长度上均优于基线隐式模型，并兼容强化学习算法与不同解码器结构。
原文摘要 · Abstract (English)
Chain-of-Thought (CoT) is a critical technique in enhancing the reasoning ability of Large Language Models (LLMs), and latent reasoning methods have been proposed to accelerate the inefficient token-level reasoning chain. We notice that existing latent reasoning methods generally require model structure augmentation and exhaustive training, limiting their broader applicability. In this paper, we propose CoLT, a novel framework that implements latent reasoning as ``tool calls''. Instead of reasoning entirely in the latent space, CoLT generates seed tokens that contain information of a reasoning step. When a latent tool call is triggered, a smaller external model will take the hidden states of seed tokens as its input, and unpack the seed tokens back to a full reasoning step. In this way, we can ensure that the main model reasons in the explicit token space, preserving its ability while improving efficiency. Experimental results on four mathematical datasets demonstrate that CoLT achieves higher accuracy and shorter reasoning length than baseline latent models, and is compatible with reinforcement learning algorithms and different decoder structures.
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。