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

Source: [RuPLaR : Efficient Latent Compression of LLM Reasoning Chains with Rule-Based Priors From Multi-Step to One-Step](<https://www.sota2.com/research/paper/6a02b465dbc836b87d824ac0>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-05-13T05:00:12.693Z

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

Published: 2026-05-10T05:55:07.000Z

## Abstract

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 tokens 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 11.1% over existing latent CoT methods but also achieves this with minimal token usage, underscoring its effectiveness and extensibility. Code: https://github.com/xiaocen-luo/RuPLaR.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Mathematical Reasoning on GSM Hard (Accuracy, Line Count, Accuracy per Line)](<https://www.sota2.com/research/sota/mathematical-reasoning-on-gsm-hard-accuracy-line-count-accuracy-per-line.md>): Mathematical Reasoning; GSM-Hard; Accuracy: 10.8.

- [Mathematical Reasoning on SVAMP (Acc, # L, Acc./# L)](<https://www.sota2.com/research/sota/mathematical-reasoning-on-svamp-acc-sharp-l-acc-sharp-l.md>): Mathematical Reasoning; SVAMP; Accuracy: 56.8.

- [Mathematical Reasoning on MultiArith (Acc, #L, Acc/#L)](<https://www.sota2.com/research/sota/mathematical-reasoning-on-multiarith-acc-sharp-l-acc-sharp-l.md>): Mathematical Reasoning; MultiArith; Accuracy: 95.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2605.09346>)

[PDF](<https://arxiv.org/pdf/2605.09346.pdf>)
