paper-with-me

홈 › Papers

Memory-Efficient LLM Training by Various-Grained Low-Rank Projection of Gradients

2025-05-03 · Yezhen Wang, Zhouhao Yang, Brian K Chen, Fanyi Pu, Bo Li, Tianyu Gao, Kenji Kawaguchi

Building upon the success of low-rank adapter (LoRA), low-rank gradient projection (LoRP) has emerged as a promising solution for memory-efficient fine-tuning. However, existing LoRP methods typically treat each row of the gradient matrix as the default projection unit, leaving the role of projection granularity underexplored. In this work, we propose a novel framework, VLoRP, that extends low-rank gradient projection by introducing an additional degree of freedom for controlling the trade-off between memory efficiency and performance, beyond the rank hyper-parameter. Through this framework, we systematically explore the impact of projection granularity, demonstrating that finer-grained projections lead to enhanced stability and efficiency even under a fixed memory budget. Regarding the optimization for VLoRP, we present ProjFactor, an adaptive memory-efficient optimizer, that significantly reduces memory requirement while ensuring competitive performance, even in the presence of gradient accumulation. Additionally, we provide a theoretical analysis of VLoRP, demonstrating the descent and convergence of its optimization trajectory under both SGD and ProjFactor. Extensive experiments are conducted to validate our findings, covering tasks such as commonsense reasoning, MMLU, and GSM8K.

📄 PDF Abstract BibTeX arXiv:2505.01744

Code (0)

등록된 구현이 없습니다.

Tasks

GSM8KMMLU

Methods 이 논문이 사용한 방법론

SGD Stochastic Gradient Descent is an iterative optimization technique that uses minibatches of data to form an expectation of the gradient, rather than the full gradient using…
Adapter 설명 없음

Similar Papers 제목 키워드 기반

COAP: Memory-Efficient Training with Correlation-Aware Gradient Projection

2024-11-26 · CVPR 2025 1 · Jinqi Xiao, Shen Sang, Tiancheng Zhi, Jing Liu 외

Training large-scale neural networks in vision, and multimodal domains demands substantial memory resources, primarily due to the storage of optimizer states. While LoRA, a popular parameter-efficient method, reduces mem…

Quantization

OASIS: Online Activation Subspace Learning for Memory-Efficient Training

2026-04-10 · Sakshi Choudhary, Utkarsh Saxena, Kaushik Roy arxiv

Training large language models (LLMs) is constrained by memory requirements, with activations accounting for a substantial fraction of the total footprint. Existing approaches reduce memory using low-rank weight paramete…

OwLore: Outlier-weighed Layerwise Sampled Low-Rank Projection for Memory-Efficient LLM Fine-tuning

2024-05-28 · Pengxiang Li, Lu Yin, Xiaowei Gao, Shiwei Liu

The rapid advancements in Large Language Models (LLMs) have revolutionized various natural language processing tasks. However, the substantial size of LLMs presents significant challenges in training or fine-tuning. Whil…

MMLU

Unbiased Gradient Low-Rank Projection

2025-10-20 · Rui Pan, Yang Luo, Yuxing Liu, Yang You 외 arxiv

Memory-efficient optimization is critical for training increasingly large language models (LLMs). A popular strategy involves gradient low-rank projection, storing only the projected optimizer states, with GaLore being a…

Gradient Weight-normalized Low-rank Projection for Efficient LLM Training

2024-12-27 · Jia-Hong Huang, Yixian Shen, Hongyi Zhu, Stevan Rudinac 외

Large Language Models (LLMs) have shown remarkable performance across various tasks, but the escalating demands on computational resources pose significant challenges, particularly in the extensive utilization of full fi…

parameter-efficient fine-tuning