Parameter-Efficient Fine-Tuning with Differential Privacy for Robust Instruction Adaptation in Large Language Models

TL;DR

This paper introduces a parameter-efficient fine-tuning method with differential privacy for large language models, enhancing privacy, stability, and performance in instruction adaptation tasks.

Contribution

It presents a novel framework combining gradient clipping, adaptive noise allocation, and parameter projection to improve privacy and efficiency in instruction fine-tuning.

Findings

Outperforms baseline models in accuracy and privacy budget

Maintains stable performance across diverse data conditions

Reduces privacy budget consumption and training instability

Abstract

This study addresses the issues of privacy protection and efficiency in instruction fine-tuning of large-scale language models by proposing a parameter-efficient method that integrates differential privacy noise allocation with gradient clipping in a collaborative optimization framework. The method keeps the backbone model frozen and updates parameters through a low-dimensional projection subspace, while introducing clipping and adaptive noise allocation during gradient computation. This design reduces privacy budget consumption and ensures training stability and robustness. The unified framework combines gradient constraints, noise allocation, and parameter projection, effectively mitigating performance fluctuations and privacy risks in multi-task instruction scenarios. Experiments are conducted across hyperparameter, environment, and data sensitivity dimensions. Results show that the method outperforms baseline models in accuracy, privacy budget, and parameter efficiency, and maintains stable performance under diverse and uncertain data conditions. The findings enrich the theoretical integration of differential privacy and parameter-efficient fine-tuning and demonstrate its practical adaptability in instruction tasks, providing a feasible solution for secure training in complex instruction environments.