Low-Rank Curvature for Zeroth-Order Optimization in LLM Fine-Tuning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Seung, Hyunseok, Lee, Jaewoo, Ko, Hyunsuk
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909897912745984
author Seung, Hyunseok
Lee, Jaewoo
Ko, Hyunsuk
author_facet Seung, Hyunseok
Lee, Jaewoo
Ko, Hyunsuk
contents We introduce LOREN, a curvature-aware zeroth-order (ZO) optimization method for fine-tuning large language models (LLMs). Existing ZO methods, which estimate gradients via finite differences using random perturbations, often suffer from high variance and suboptimal search directions. Our approach addresses these challenges by: (i) reformulating the problem of gradient preconditioning as that of adaptively estimating an anisotropic perturbation distribution for gradient estimation, (ii) capturing curvature through a low-rank block diagonal preconditioner using the framework of natural evolution strategies, and (iii) applying a REINFORCE leave-one-out (RLOO) gradient estimator to reduce variance. Experiments on standard LLM benchmarks show that our method outperforms state-of-the-art ZO methods by achieving higher accuracy and faster convergence, while cutting peak memory usage by up to 27.3% compared with MeZO-Adam.
format Preprint
id arxiv_https___arxiv_org_abs_2511_07971
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Low-Rank Curvature for Zeroth-Order Optimization in LLM Fine-Tuning
Seung, Hyunseok
Lee, Jaewoo
Ko, Hyunsuk
Machine Learning
We introduce LOREN, a curvature-aware zeroth-order (ZO) optimization method for fine-tuning large language models (LLMs). Existing ZO methods, which estimate gradients via finite differences using random perturbations, often suffer from high variance and suboptimal search directions. Our approach addresses these challenges by: (i) reformulating the problem of gradient preconditioning as that of adaptively estimating an anisotropic perturbation distribution for gradient estimation, (ii) capturing curvature through a low-rank block diagonal preconditioner using the framework of natural evolution strategies, and (iii) applying a REINFORCE leave-one-out (RLOO) gradient estimator to reduce variance. Experiments on standard LLM benchmarks show that our method outperforms state-of-the-art ZO methods by achieving higher accuracy and faster convergence, while cutting peak memory usage by up to 27.3% compared with MeZO-Adam.
title Low-Rank Curvature for Zeroth-Order Optimization in LLM Fine-Tuning
topic Machine Learning
url https://arxiv.org/abs/2511.07971