Self-Training Elicits Concise Reasoning in Large Language Models

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Munkhbat, Tergel, Ho, Namgyu, Kim, Seo Hyun, Yang, Yongjin, Kim, Yujin, Yun, Se-Young
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866908401570676736
author Munkhbat, Tergel
Ho, Namgyu
Kim, Seo Hyun
Yang, Yongjin
Kim, Yujin
Yun, Se-Young
author_facet Munkhbat, Tergel
Ho, Namgyu
Kim, Seo Hyun
Yang, Yongjin
Kim, Yujin
Yun, Se-Young
contents Chain-of-thought (CoT) reasoning has enabled large language models (LLMs) to utilize additional computation through intermediate tokens to solve complex tasks. However, we posit that typical reasoning traces contain many redundant tokens, incurring extraneous inference costs. Upon examination of the output distribution of current LLMs, we find evidence on their latent ability to reason more concisely, relative to their default behavior. To elicit this capability, we propose simple fine-tuning methods which leverage self-generated concise reasoning paths obtained by best-of-N sampling and few-shot conditioning, in task-specific settings. Our combined method achieves a 30% reduction in output tokens on average, across five model families on GSM8K and MATH, while maintaining average accuracy. By exploiting the fundamental stochasticity and in-context learning capabilities of LLMs, our self-training approach robustly elicits concise reasoning on a wide range of models, including those with extensive post-training. Code is available at https://github.com/TergelMunkhbat/concise-reasoning
format Preprint
id arxiv_https___arxiv_org_abs_2502_20122
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Self-Training Elicits Concise Reasoning in Large Language Models
Munkhbat, Tergel
Ho, Namgyu
Kim, Seo Hyun
Yang, Yongjin
Kim, Yujin
Yun, Se-Young
Computation and Language
Artificial Intelligence
Machine Learning
Chain-of-thought (CoT) reasoning has enabled large language models (LLMs) to utilize additional computation through intermediate tokens to solve complex tasks. However, we posit that typical reasoning traces contain many redundant tokens, incurring extraneous inference costs. Upon examination of the output distribution of current LLMs, we find evidence on their latent ability to reason more concisely, relative to their default behavior. To elicit this capability, we propose simple fine-tuning methods which leverage self-generated concise reasoning paths obtained by best-of-N sampling and few-shot conditioning, in task-specific settings. Our combined method achieves a 30% reduction in output tokens on average, across five model families on GSM8K and MATH, while maintaining average accuracy. By exploiting the fundamental stochasticity and in-context learning capabilities of LLMs, our self-training approach robustly elicits concise reasoning on a wide range of models, including those with extensive post-training. Code is available at https://github.com/TergelMunkhbat/concise-reasoning
title Self-Training Elicits Concise Reasoning in Large Language Models
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2502.20122