R1-Compress: Long Chain-of-Thought Compression via Chunk Compression and Search

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wang, Yibo, Luo, Haotian, Yao, Huanjin, Huang, Tiansheng, He, Haiying, Liu, Rui, Tan, Naiqiang, Huang, Jiaxing, Cao, Xiaochun, Tao, Dacheng, Shen, Li
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914155202609152
author Wang, Yibo
Luo, Haotian
Yao, Huanjin
Huang, Tiansheng
He, Haiying
Liu, Rui
Tan, Naiqiang
Huang, Jiaxing
Cao, Xiaochun
Tao, Dacheng
Shen, Li
author_facet Wang, Yibo
Luo, Haotian
Yao, Huanjin
Huang, Tiansheng
He, Haiying
Liu, Rui
Tan, Naiqiang
Huang, Jiaxing
Cao, Xiaochun
Tao, Dacheng
Shen, Li
contents Chain-of-Thought (CoT) reasoning enhances large language models (LLMs) by enabling step-by-step problem-solving, yet its extension to Long-CoT introduces substantial computational overhead due to increased token length. Existing compression approaches -- instance-level and token-level -- either sacrifice essential local reasoning signals like reflection or yield incoherent outputs. To address these limitations, we propose R1-Compress, a two-stage chunk-level compression framework that preserves both local information and coherence. Our method segments Long-CoT into manageable chunks, applies LLM-driven inner-chunk compression, and employs an inter-chunk search mechanism to select the short and coherent sequence. Experiments on Qwen2.5-Instruct models across MATH500, AIME24, and GPQA-Diamond demonstrate that R1-Compress significantly reduces token usage while maintaining comparable reasoning accuracy. On MATH500, R1-Compress achieves an accuracy of 92.4%, with only a 0.6% drop compared to the Long-CoT baseline, while reducing token usage by about 20%. Source code will be available at https://github.com/w-yibo/R1-Compress
format Preprint
id arxiv_https___arxiv_org_abs_2505_16838
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle R1-Compress: Long Chain-of-Thought Compression via Chunk Compression and Search
Wang, Yibo
Luo, Haotian
Yao, Huanjin
Huang, Tiansheng
He, Haiying
Liu, Rui
Tan, Naiqiang
Huang, Jiaxing
Cao, Xiaochun
Tao, Dacheng
Shen, Li
Computation and Language
Chain-of-Thought (CoT) reasoning enhances large language models (LLMs) by enabling step-by-step problem-solving, yet its extension to Long-CoT introduces substantial computational overhead due to increased token length. Existing compression approaches -- instance-level and token-level -- either sacrifice essential local reasoning signals like reflection or yield incoherent outputs. To address these limitations, we propose R1-Compress, a two-stage chunk-level compression framework that preserves both local information and coherence. Our method segments Long-CoT into manageable chunks, applies LLM-driven inner-chunk compression, and employs an inter-chunk search mechanism to select the short and coherent sequence. Experiments on Qwen2.5-Instruct models across MATH500, AIME24, and GPQA-Diamond demonstrate that R1-Compress significantly reduces token usage while maintaining comparable reasoning accuracy. On MATH500, R1-Compress achieves an accuracy of 92.4%, with only a 0.6% drop compared to the Long-CoT baseline, while reducing token usage by about 20%. Source code will be available at https://github.com/w-yibo/R1-Compress
title R1-Compress: Long Chain-of-Thought Compression via Chunk Compression and Search
topic Computation and Language
url https://arxiv.org/abs/2505.16838