Enhancing Long-Chain Reasoning Distillation through Error-Aware Self-Reflection

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wu, Zhuoyang, Li, Xinze, Liu, Zhenghao, Yan, Yukun, Liu, Zhiyuan, Yu, Minghe, Yang, Cheng, Gu, Yu, Yu, Ge, Sun, Maosong
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912645382144000
author Wu, Zhuoyang
Li, Xinze
Liu, Zhenghao
Yan, Yukun
Liu, Zhiyuan
Yu, Minghe
Yang, Cheng
Gu, Yu
Yu, Ge
Sun, Maosong
author_facet Wu, Zhuoyang
Li, Xinze
Liu, Zhenghao
Yan, Yukun
Liu, Zhiyuan
Yu, Minghe
Yang, Cheng
Gu, Yu
Yu, Ge
Sun, Maosong
contents Large Language Models (LLMs) have exhibited strong reasoning capabilities and achieved remarkable performance in mathematical problem-solving tasks. Recently, distilling reasoning ability from long-form Chains-of-Thought (CoTs) has emerged as a promising approach for enhancing Small Language Models (SLMs). Existing studies typically treat SLMs as student models and use long-form CoTs as supervision signals for Supervised Fine-Tuning (SFT) to transfer reasoning ability. However, such long-form CoT teachers are usually unaware of the student model's capacity, which limits the effective utilization of the provided reasoning traces. To overcome this limitation, we propose errOr-aware self-ReflectION (ORION), a framework that refines teacher CoTs through an Error-Aware Reflection process. ORION enables the student model to construct more tailored teacher CoTs by refining teacher CoTs and incorporating its own reasoning errors. Experiments on multiple mathematical reasoning benchmarks demonstrate that ORION consistently improves performance by more than 2% over all baselines. Further analysis reveals that the CoTs constructed by ORION exhibit higher coherence and logical consistency, thereby serving as more effective supervision signals for SFT. All codes are available at https://github.com/NEUIR/ORION.git.
format Preprint
id arxiv_https___arxiv_org_abs_2505_22131
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Enhancing Long-Chain Reasoning Distillation through Error-Aware Self-Reflection
Wu, Zhuoyang
Li, Xinze
Liu, Zhenghao
Yan, Yukun
Liu, Zhiyuan
Yu, Minghe
Yang, Cheng
Gu, Yu
Yu, Ge
Sun, Maosong
Computation and Language
Large Language Models (LLMs) have exhibited strong reasoning capabilities and achieved remarkable performance in mathematical problem-solving tasks. Recently, distilling reasoning ability from long-form Chains-of-Thought (CoTs) has emerged as a promising approach for enhancing Small Language Models (SLMs). Existing studies typically treat SLMs as student models and use long-form CoTs as supervision signals for Supervised Fine-Tuning (SFT) to transfer reasoning ability. However, such long-form CoT teachers are usually unaware of the student model's capacity, which limits the effective utilization of the provided reasoning traces. To overcome this limitation, we propose errOr-aware self-ReflectION (ORION), a framework that refines teacher CoTs through an Error-Aware Reflection process. ORION enables the student model to construct more tailored teacher CoTs by refining teacher CoTs and incorporating its own reasoning errors. Experiments on multiple mathematical reasoning benchmarks demonstrate that ORION consistently improves performance by more than 2% over all baselines. Further analysis reveals that the CoTs constructed by ORION exhibit higher coherence and logical consistency, thereby serving as more effective supervision signals for SFT. All codes are available at https://github.com/NEUIR/ORION.git.
title Enhancing Long-Chain Reasoning Distillation through Error-Aware Self-Reflection
topic Computation and Language
url https://arxiv.org/abs/2505.22131