OAT-Rephrase: Optimization-Aware Training Data Rephrasing for Zeroth-Order LLM Fine-Tuning

TL;DR

OAT-Rephrase introduces an optimization-aware data rephrasing method that enhances zeroth-order fine-tuning of large language models, reducing convergence issues and improving performance across multiple tasks.

Contribution

The paper presents a novel rephrasing strategy leveraging an LLM to improve zeroth-order optimization in LLM fine-tuning, addressing convergence and stability challenges.

Findings

Consistent performance improvements across five classification tasks

Reduction or elimination of the gap with first-order methods

Effective enhancement across three LLM architectures

Abstract

Fine-tuning large language models (LLMs) using zeroth-order optimization (ZO) offers a memory-efficient alternative to gradient-based methods but suffers from slower convergence and unstable optimization due to noisy gradient estimates. This paper introduces OAT-Rephrase, an Optimization-Aware Training data rephrasing strategy that leverages an LLM to rephrase training instances based on its understanding of the ZO dynamics, specifically MeZO, derived directly from its paper. The approach incorporates a dual-stage pipeline featuring a rewriter LLM and a semantic judge, ensuring all rephrasings retain task relevance and logical consistency. Evaluations across five classification tasks and three LLM architectures demonstrate that OAT-Rephrase consistently improves MeZO fine-tuning performance, often narrowing or eliminating the gap with first-order methods. Our findings suggest that optimization-aware rephrasing serves as a reusable and low-overhead enhancement for zeroth-order tuning regimes.