arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2605.09346cs.CLcs.AI
收藏

RuPLaR : 通过基于规则的先验概率实现LLM推理链的高效潜在压缩

RuPLaR : Efficient Latent Compression of LLM Reasoning Chains with Rule-Based Priors From Multi-Step to One-Step

School of Computer Science and Technology(计算机科学与技术学院);Department of Electrical and Computer Engineering(电气与计算机工程系)

Xiaocheng Luo, Kang Wang, Zaifu Zhan, Yuechi Zhou, Xiangyu Duan

AI总结:

本文提出RuPLaR框架,通过基于规则的先验概率指导LLM生成单阶段潜在推理令牌,提升推理效率与准确性,实验表明其比现有方法提升11.1%的准确率且消耗更少token。

摘要:

Chain-of-Thought (CoT)范式虽然增强了大型语言模型(LLM)的可解释性,但受限于自然语言的效率和表达限制。Latent Chain-of-Thought(latent CoT)推理在连续潜在空间中运行,提供了一种有前景的替代方案,但面临现有多步骤或多模型范式中的结构性复杂性挑战,如误差传播和协调开销。本文引入One-Model One-Step,一种新的压缩框架用于具有基于规则先验的概率的潜在推理(RuPLaR),以解决这一挑战。我们的方法训练LLM在单个训练阶段中自主生成潜在推理令牌,受基于规则的先验概率分布指导,从而消除级联过程和跨模型依赖。为确保推理质量,我们设计了一个联合训练目标,通过交叉熵强制答案一致性,通过KL散度(Soft Thinking约束)对软令牌与基于规则的先验进行对齐,并在表示空间中添加问题-思考语义对齐约束。广泛的实验表明,我们的压缩框架不仅在现有latent CoT方法上提升了11.1%的准确率,而且以最少的token使用实现了这一点,凸显了其有效性和可扩展性。代码:https://github.com/xiaocen-luo/RuPLaR。

英文摘要:

The Chain-of-Thought (CoT) paradigm, while enhancing the interpretability of Large Language Models (LLMs), is constrained by the inefficiencies and expressive limits of natural language. Latent Chain-of-Thought (latent CoT) reasoning, which operates in a continuous latent space, offers a promising alternative but faces challenges from structural complexities in existing multi-step or multi-model paradigms, such as error propagation and coordination overhead. In this paper, we introduce One-Model One-Step, a novel compression framework for Latent Reasoning with Rule-Based Priors(RuPLaR) to address this challenge. Our method trains an LLM to autonomously generate latent reasoning tokens in a single training stage, guided by rule-based prior probability distributions, thereby eliminating cascaded processes and inter-model dependencies. To ensure reasoning quality, we design a joint training objective that enforces answer consistency via cross-entropy, aligns soft token distributions with rule-based priors via KL divergence (the Soft Thinking constraint), and adds a problem-thought semantic alignment constraint in the representation space. Extensive experiments show that our compression framework not only improves accuracy by 11.03% over existing latent CoT methods on GSM8k-Aug but also achieves this with minimal token usage, underscoring its effectiveness and extensibility. Code: https://github.com/xiaocen-luo/RuPLaR.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑