Saved in:
Bibliographic Details
Main Authors: Zhang, Ziqi, Shamsabadi, Ali Shahin, Lu, Hanxiao, Cai, Yifeng, Haddadi, Hamed
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2508.07054
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911100523511808
author Zhang, Ziqi
Shamsabadi, Ali Shahin
Lu, Hanxiao
Cai, Yifeng
Haddadi, Hamed
author_facet Zhang, Ziqi
Shamsabadi, Ali Shahin
Lu, Hanxiao
Cai, Yifeng
Haddadi, Hamed
contents Recent advances in Knowledge Distillation (KD) aim to mitigate the high computational demands of Large Language Models (LLMs) by transferring knowledge from a large ''teacher'' to a smaller ''student'' model. However, students may inherit the teacher's privacy when the teacher is trained on private data. In this work, we systematically characterize and investigate membership and memorization privacy risks inherent in six LLM KD techniques. Using instruction-tuning settings that span seven NLP tasks, together with three teacher model families (GPT-2, LLAMA-2, and OPT), and various size student models, we demonstrate that all existing LLM KD approaches carry membership and memorization privacy risks from the teacher to its students. However, the extent of privacy risks varies across different KD techniques. We systematically analyse how key LLM KD components (KD objective functions, student training data and NLP tasks) impact such privacy risks. We also demonstrate a significant disagreement between memorization and membership privacy risks of LLM KD techniques. Finally, we characterize per-block privacy risk and demonstrate that the privacy risk varies across different blocks by a large margin.
format Preprint
id arxiv_https___arxiv_org_abs_2508_07054
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Membership and Memorization in LLM Knowledge Distillation
Zhang, Ziqi
Shamsabadi, Ali Shahin
Lu, Hanxiao
Cai, Yifeng
Haddadi, Hamed
Machine Learning
Artificial Intelligence
Recent advances in Knowledge Distillation (KD) aim to mitigate the high computational demands of Large Language Models (LLMs) by transferring knowledge from a large ''teacher'' to a smaller ''student'' model. However, students may inherit the teacher's privacy when the teacher is trained on private data. In this work, we systematically characterize and investigate membership and memorization privacy risks inherent in six LLM KD techniques. Using instruction-tuning settings that span seven NLP tasks, together with three teacher model families (GPT-2, LLAMA-2, and OPT), and various size student models, we demonstrate that all existing LLM KD approaches carry membership and memorization privacy risks from the teacher to its students. However, the extent of privacy risks varies across different KD techniques. We systematically analyse how key LLM KD components (KD objective functions, student training data and NLP tasks) impact such privacy risks. We also demonstrate a significant disagreement between memorization and membership privacy risks of LLM KD techniques. Finally, we characterize per-block privacy risk and demonstrate that the privacy risk varies across different blocks by a large margin.
title Membership and Memorization in LLM Knowledge Distillation
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2508.07054