RLKD: Distilling LLMs' Reasoning via Reinforcement Learning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Xu, Shicheng, Pang, Liang, Zhu, Yunchang, Gu, Jia, Wei, Zihao, Deng, Jingcheng, Pan, Feiyang, Shen, Huawei, Cheng, Xueqi
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908801293090816
author Xu, Shicheng
Pang, Liang
Zhu, Yunchang
Gu, Jia
Wei, Zihao
Deng, Jingcheng
Pan, Feiyang
Shen, Huawei
Cheng, Xueqi
author_facet Xu, Shicheng
Pang, Liang
Zhu, Yunchang
Gu, Jia
Wei, Zihao
Deng, Jingcheng
Pan, Feiyang
Shen, Huawei
Cheng, Xueqi
contents Distilling reasoning paths from teacher to student models via supervised fine-tuning (SFT) provides a shortcut for improving the reasoning ability of smaller Large Language Models (LLMs). However, the reasoning paths generated by teacher models often reflect only surface-level traces of their underlying authentic reasoning. Insights from cognitive neuroscience suggest that authentic reasoning involves a complex interweaving between meta-reasoning (which selects appropriate sub-problems from multiple candidates) and solving (which addresses the sub-problem). This implies authentic reasoning has an implicit multi-branch structure. Supervised fine-tuning collapses this rich structure into a flat sequence of token prediction in the teacher's reasoning path, preventing effective distillation of this structure to students. To address this limitation, we propose RLKD, a reinforcement learning (RL)-based distillation framework guided by a novel Generative Structure Reward Model (GSRM). Our GSRM converts reasoning paths into multiple meta-reasoning-solving steps and computes rewards to measure structural alignment between student and teacher reasoning. RLKD combines this reward with RL, enabling student LLMs to internalize the teacher's implicit multi-branch reasoning structure rather than merely mimicking fixed output paths. Experiments show RLKD surpasses standard SFT-RL pipelines even when trained on 0.1% of data under an RL-only regime, unlocking greater student reasoning potential than SFT-based distillation. Code is available at https://github.com/xsc1234/RLKD.
format Preprint
id arxiv_https___arxiv_org_abs_2505_16142
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle RLKD: Distilling LLMs' Reasoning via Reinforcement Learning
Xu, Shicheng
Pang, Liang
Zhu, Yunchang
Gu, Jia
Wei, Zihao
Deng, Jingcheng
Pan, Feiyang
Shen, Huawei
Cheng, Xueqi
Computation and Language
Distilling reasoning paths from teacher to student models via supervised fine-tuning (SFT) provides a shortcut for improving the reasoning ability of smaller Large Language Models (LLMs). However, the reasoning paths generated by teacher models often reflect only surface-level traces of their underlying authentic reasoning. Insights from cognitive neuroscience suggest that authentic reasoning involves a complex interweaving between meta-reasoning (which selects appropriate sub-problems from multiple candidates) and solving (which addresses the sub-problem). This implies authentic reasoning has an implicit multi-branch structure. Supervised fine-tuning collapses this rich structure into a flat sequence of token prediction in the teacher's reasoning path, preventing effective distillation of this structure to students. To address this limitation, we propose RLKD, a reinforcement learning (RL)-based distillation framework guided by a novel Generative Structure Reward Model (GSRM). Our GSRM converts reasoning paths into multiple meta-reasoning-solving steps and computes rewards to measure structural alignment between student and teacher reasoning. RLKD combines this reward with RL, enabling student LLMs to internalize the teacher's implicit multi-branch reasoning structure rather than merely mimicking fixed output paths. Experiments show RLKD surpasses standard SFT-RL pipelines even when trained on 0.1% of data under an RL-only regime, unlocking greater student reasoning potential than SFT-based distillation. Code is available at https://github.com/xsc1234/RLKD.
title RLKD: Distilling LLMs' Reasoning via Reinforcement Learning
topic Computation and Language
url https://arxiv.org/abs/2505.16142