Leaner Training, Lower Leakage: Revisiting Memorization in LLM Fine-Tuning with LoRA

2025/06/25 by Fei Wang, Wang, Fei, Baochun Li +1 · 1 citation
Business, Management and Accounting · Decision Sciences · Engineering · #Advanced Data Processing Techniques #Business Process Modeling and Analysis #Divergence (linguistics) #Memorization #Scale (ratio) #Simulation Techniques and Applications #Task (project management) #Task analysis

paper · pdf · doi:10.48550/arxiv.2506.20856

published in arXiv (Cornell University) (Cornell University)

openalex publication_date 2025/06/25 · openalex created_date 2025/10/15 · openalex updated_date 2026/08/05

Abstract

Memorization in large language models (LLMs) makes them vulnerable to data extraction attacks. While pre-training memorization has been extensively studied, fewer works have explored its impact in fine-tuning, particularly for LoRA fine-tuning, a widely adopted parameter-efficient method. In this work, we re-examine memorization in fine-tuning and uncover a surprising divergence from prior findings across different fine-tuning strategies. Factors such as model scale and data duplication, which strongly influence memorization in pre-training and full fine-tuning, do not follow the same trend in LoRA fine-tuning. Using a more relaxed similarity-based memorization metric, we demonstrate that LoRA significantly reduces memorization risks compared to full fine-tuning, while still maintaining strong task performance.

Citations

• Rethinking LLM Memorization through the Lens of Adversarial Compression
• Alpaca against Vicuna: Using LLMs to Uncover Memorization of LLMs
• Exploring Memorization in Fine-tuned Language Models
• Ethicist: Targeted Training Data Extraction Through Loss Smoothed Soft Prompting and Calibrated Confidence Estimation
• ChatDoctor: A Medical Chat Model Fine-Tuned on a Large Language Model Meta-AI (LLaMA) Using Medical Domain Knowledge
• Quantifying Privacy Risks of Masked Language Models Using Membership Inference Attacks
• Deduplicating Training Data Mitigates Privacy Risks in Language Models
• Parameter-Efficient Transfer Learning for NLP
• The Secret Sharer: Evaluating and Testing Unintended Memorization in Neural Networks

Cited by

• LLM Harms: A Taxonomy and Discussion

Related

• Exploring Memorization in Fine-tuned Language Models
• Memorization in Fine-Tuned Large Language Models
• LoRA vs Full Fine-tuning: An Illusion of Equivalence
• Early Detection and Reduction of Memorisation for Domain Adaptation and Instruction Tuning
• Memorization in NLP Fine-tuning Methods