Saved in:
Bibliographic Details
Main Authors: Yang, Hui, Ren, Tao, Jiang, Jinyang, Tian, Wan, Peng, Yijie
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2603.05960
Tags: Add Tag
No Tags, Be the first to tag this record!
Table of Contents:
  • Memory-efficient optimization methods have recently gained increasing attention for scaling full-parameter training of large language models under the GPU-memory bottleneck. Existing approaches either lack clear convergence guarantees, or only achieve the standard ${\mathcal{O}}(ε^{-4})$ iteration complexity in the nonconvex settings. We propose Omni-Masked Gradient Descent (OMGD), an optimization method based on mask traversal for memory efficient training, and provide a nonconvex convergence analysis that establishes a strictly improved iteration complexity of $\tilde{\mathcal{O}}(ε^{-3})$ for finding an $ε$-approximate stationary point. Empirically, OMGD is a lightweight, plug-and-play approach that integrates seamlessly into most mainstream optimizers, yielding consistent improvements over competitive baselines in both fine-tuning and pre-training tasks.