Exploring Knowledge Purification in Multi-Teacher Knowledge Distillation for LLMs

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Jin, Ruihan, Shao, Pengpeng, Wen, Zhengqi, Wu, Jinyang, Feng, Mingkuan, Yang, Shuo, Zhang, Chu Yuan, Tao, Jianhua
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866913023607701504
author Jin, Ruihan
Shao, Pengpeng
Wen, Zhengqi
Wu, Jinyang
Feng, Mingkuan
Yang, Shuo
Zhang, Chu Yuan
Tao, Jianhua
author_facet Jin, Ruihan
Shao, Pengpeng
Wen, Zhengqi
Wu, Jinyang
Feng, Mingkuan
Yang, Shuo
Zhang, Chu Yuan
Tao, Jianhua
contents Knowledge distillation has emerged as a pivotal technique for transferring knowledge from stronger large language models (LLMs) to smaller, more efficient models. However, traditional distillation approaches face challenges related to knowledge conflicts and high resource demands, particularly when leveraging multiple teacher models. In this paper, we introduce the concept of \textbf{Knowledge Purification}, which consolidates the rationales from multiple teacher LLMs into a single rationale, thereby mitigating conflicts and enhancing efficiency. To investigate the effectiveness of knowledge purification, we further propose five purification methods from various perspectives. Our experiments demonstrate that these methods not only improve the performance of the distilled model but also effectively alleviate knowledge conflicts. Moreover, router-based methods exhibit robust generalization capabilities, underscoring the potential of innovative purification techniques in optimizing multi-teacher distillation and facilitating the practical deployment of powerful yet lightweight models.
format Preprint
id arxiv_https___arxiv_org_abs_2602_01064
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Exploring Knowledge Purification in Multi-Teacher Knowledge Distillation for LLMs
Jin, Ruihan
Shao, Pengpeng
Wen, Zhengqi
Wu, Jinyang
Feng, Mingkuan
Yang, Shuo
Zhang, Chu Yuan
Tao, Jianhua
Computation and Language
Knowledge distillation has emerged as a pivotal technique for transferring knowledge from stronger large language models (LLMs) to smaller, more efficient models. However, traditional distillation approaches face challenges related to knowledge conflicts and high resource demands, particularly when leveraging multiple teacher models. In this paper, we introduce the concept of \textbf{Knowledge Purification}, which consolidates the rationales from multiple teacher LLMs into a single rationale, thereby mitigating conflicts and enhancing efficiency. To investigate the effectiveness of knowledge purification, we further propose five purification methods from various perspectives. Our experiments demonstrate that these methods not only improve the performance of the distilled model but also effectively alleviate knowledge conflicts. Moreover, router-based methods exhibit robust generalization capabilities, underscoring the potential of innovative purification techniques in optimizing multi-teacher distillation and facilitating the practical deployment of powerful yet lightweight models.
title Exploring Knowledge Purification in Multi-Teacher Knowledge Distillation for LLMs
topic Computation and Language
url https://arxiv.org/abs/2602.01064