LightThinker: Thinking Step-by-Step Compression

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhang, Jintian, Zhu, Yuqi, Sun, Mengshu, Luo, Yujie, Qiao, Shuofei, Du, Lun, Zheng, Da, Chen, Huajun, Zhang, Ningyu
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908553377218560
author Zhang, Jintian
Zhu, Yuqi
Sun, Mengshu
Luo, Yujie
Qiao, Shuofei
Du, Lun
Zheng, Da
Chen, Huajun
Zhang, Ningyu
author_facet Zhang, Jintian
Zhu, Yuqi
Sun, Mengshu
Luo, Yujie
Qiao, Shuofei
Du, Lun
Zheng, Da
Chen, Huajun
Zhang, Ningyu
contents Large language models (LLMs) have shown remarkable performance in complex reasoning tasks, but their efficiency is hindered by the substantial memory and computational costs associated with generating lengthy tokens. In this paper, we propose LightThinker, a novel method that enables LLMs to dynamically compress intermediate thoughts during reasoning. Inspired by human cognitive processes, LightThinker compresses verbose thought steps into compact representations and discards the original reasoning chains, thereby significantly reducing the number of tokens stored in the context window. This is achieved by training the model on when and how to perform compression through data construction, mapping hidden states to condensed gist tokens, and creating specialized attention masks. Additionally, we introduce the Dependency (Dep) metric to quantify the degree of compression by measuring the reliance on historical tokens during generation. Extensive experiments on four datasets and two models show that LightThinker reduces peak memory usage and inference time, while maintaining competitive accuracy. Our work provides a new direction for improving the efficiency of LLMs in complex reasoning tasks without sacrificing performance. Code is released at https://github.com/zjunlp/LightThinker.
format Preprint
id arxiv_https___arxiv_org_abs_2502_15589
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle LightThinker: Thinking Step-by-Step Compression
Zhang, Jintian
Zhu, Yuqi
Sun, Mengshu
Luo, Yujie
Qiao, Shuofei
Du, Lun
Zheng, Da
Chen, Huajun
Zhang, Ningyu
Computation and Language
Artificial Intelligence
Information Retrieval
Machine Learning
Multimedia
Large language models (LLMs) have shown remarkable performance in complex reasoning tasks, but their efficiency is hindered by the substantial memory and computational costs associated with generating lengthy tokens. In this paper, we propose LightThinker, a novel method that enables LLMs to dynamically compress intermediate thoughts during reasoning. Inspired by human cognitive processes, LightThinker compresses verbose thought steps into compact representations and discards the original reasoning chains, thereby significantly reducing the number of tokens stored in the context window. This is achieved by training the model on when and how to perform compression through data construction, mapping hidden states to condensed gist tokens, and creating specialized attention masks. Additionally, we introduce the Dependency (Dep) metric to quantify the degree of compression by measuring the reliance on historical tokens during generation. Extensive experiments on four datasets and two models show that LightThinker reduces peak memory usage and inference time, while maintaining competitive accuracy. Our work provides a new direction for improving the efficiency of LLMs in complex reasoning tasks without sacrificing performance. Code is released at https://github.com/zjunlp/LightThinker.
title LightThinker: Thinking Step-by-Step Compression
topic Computation and Language
Artificial Intelligence
Information Retrieval
Machine Learning
Multimedia
url https://arxiv.org/abs/2502.15589