arXiv:2602.08324v5 Announce Type: replace 
Abstract: Chain-of-Thought (CoT) reasoning successfully enhances the reasoning capabilities of Large Language Models (LLMs), yet it incurs substantial computational overhead for inference. Existing CoT compression methods often suffer from a critical loss of logical fidelity at high compression ratios, resulting in significant performance degradation. To achieve high-fidelity, fast reasoning, we propose a novel EXTreme-RAtio Chain-of-Thought Compression framework, termed Extra-CoT, which aggressively reduces the token budget while preserving answer accuracy. To generate reliable, high-fidelity supervision, we first train a dedicated semantically-preserved compressor on mathematical CoT data with fine-grained annotations. An LLM is then fine-tuned on these compressed pairs via a mixed-ratio supervised fine-tuning (SFT), teaching it to follow a spectrum of compression budgets and providing a stable initialization for reinforcement learning (RL). We further propose Constrained and Hierarchical Ratio Policy Optimization (CHRPO) to explicitly incentivize question-solving ability under lower budgets by a hierarchical reward. Experiments on three mathematical reasoning benchmarks show the superiority of Extra-CoT. For example, on MATH-500 using Qwen3-1.7B, Extra-CoT achieves over 73\% token reduction with an accuracy improvement of 0.6\%, significantly outperforming state-of-the-art (SOTA) methods. Our source codes have been released at https://github.com/Mwie1024/Extra-CoT.

تم تقديم إطار عمل جديد يسمى Extra-CoT لتحسين كفاءة نماذج اللغة الكبيرة (LLMs) من خلال تنفيذ ضغط سلسلة التفكير بنسبة قصوى. يهدف هذا الأسلوب إلى تقليل الحمل الحسابي بشكل كبير أثناء الاستدلال مع الحفاظ على دقة منطقية عالية ودقة في الإجابات، مما يعالج قيود تقنيات الضغط الحالية.

Se ha introducido un nuevo marco llamado Extra-CoT para mejorar la eficiencia de los Modelos de Lenguaje Grande (LLMs) mediante la implementación de Compresión de Cadena de Pensamiento de Ratio Extremo. Este método tiene como objetivo reducir significativamente la sobrecarga computacional durante la inferencia, manteniendo al mismo tiempo una alta fidelidad lógica y precisión en las respuestas, abordando las limitaciones de las técnicas de compresión existentes.

Un nouveau cadre nommé Extra-CoT a été introduit pour améliorer l'efficacité des grands modèles de langage (LLMs) en mettant en œuvre une compression de chaîne de pensée à ratio extrême. Cette méthode vise à réduire considérablement la surcharge computationnelle lors de l'inférence tout en maintenant une haute fidélité logique et une précision des réponses, répondant ainsi aux limites des techniques de compression existantes.

A novel framework named Extra-CoT has been introduced to enhance the efficiency of Large Language Models (LLMs) by implementing Extreme-Ratio Chain-of-Thought Compression. This method aims to significantly reduce computational overhead during inference while maintaining high logical fidelity and answer accuracy, addressing the limitations of existing compression techniques.