不用额外数据，通过提取特征引导模型推理，提升数学与逻辑题解题能力。
Feature Extraction and Steering for Enhanced Chain-of-Thought Reasoning in Language Models
- 用稀疏自编码器提取可解释的思维特征，指导模型生成更优推理过程。
- 在多个基准测试中，推理准确率提升15%-23%，尤其在复杂问题上表现显著。
- 提出无需预训练自编码器的新方法，适配无现成特征工具的模型使用。
大型语言模型（LLMs）通过思维链（CoT）技术解决推理与数学问题已展现出潜力。如DeepSeek-R1所示，扩展思维链长度能显著提升复杂问题的推理能力，但需昂贵且高质量的长思维链数据及微调。受DeepSeek-R1深度思考范式的启发，本文提出一种无需外部数据的引导技术，以增强LLM的推理能力。首先，利用稀疏自编码器（SAEs）从原始思维链中提取可解释特征，并用于引导模型生成过程中的内部状态。针对多数模型缺乏预训练SAE的问题，进一步提出一种新型无SAE引导算法，直接从模型残差激活中计算引导方向，无需显式SAE。实验表明，基于SAE和无SAE的引导方法均显著提升了多种模型的推理能力，在多个基准测试中准确率提升15%-23%。
原文摘要 · Abstract (English)
Large Language Models (LLMs) demonstrate the ability to solve reasoning and mathematical problems using the Chain-of-Thought (CoT) technique. Expanding CoT length, as seen in models such as DeepSeek-R1, significantly enhances this reasoning for complex problems, but requires costly and high-quality long CoT data and fine-tuning. This work, inspired by the deep thinking paradigm of DeepSeek-R1, utilizes a steering technique to enhance the reasoning ability of an LLM without external datasets. Our method first employs Sparse Autoencoders (SAEs) to extract interpretable features from vanilla CoT. These features are then used to steer the LLM's internal states during generation. Recognizing that many LLMs do not have corresponding pre-trained SAEs, we further introduce a novel SAE-free steering algorithm, which directly computes steering directions from the residual activations of an LLM, obviating the need for an explicit SAE. Experimental results demonstrate that both our SAE-based and subsequent SAE-free steering algorithms significantly enhance the reasoning capabilities of LLMs.
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。