SVD-Free Low-Rank Adaptive Gradient Optimization for Large Language Models

• URL: http://arxiv.org/abs/2505.17967v1
• Date: Fri, 23 May 2025 14:37:00 GMT
• Title: SVD-Free Low-Rank Adaptive Gradient Optimization for Large Language Models
• Authors: Ionut-Vlad Modoranu, Mher Safaryan, Erik Schultheis, Dan Alistarh,
• Abstract summary: We propose a two-step procedure to approximate SVD-based gradient projections into lower-dimensional spaces.<n>Our experiments on both pre-training and fine-tuning tasks demonstrate the effectiveness of our dual strategy.
• Score: 37.60342078872549
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Low-rank optimization has emerged as a promising direction in training large language models (LLMs) to reduce the memory usage of adaptive optimizers by constraining learning to a lower-dimensional space. Prior work typically projects gradients of linear layers using approaches based on Singular Value Decomposition (SVD). However, applying SVD-based procedures individually to each layer in large models is computationally expensive and incurs additional memory costs due to storing the projection matrices. In this work, we propose a computationally efficient and conceptually simple two-step procedure to approximate SVD-based gradient projections into lower-dimensional spaces. First, we construct a complete orthogonal basis using predefined orthogonal matrices of the Discrete Cosine Transform (DCT). Second, we adaptively select basis columns based on their alignment with the gradient of each layer. Each projection matrix in our method is obtained via a single matrix multiplication followed by a lightweight sorting step to identify the most relevant basis vectors. Due to the predefined nature of the orthogonal bases, they are computed once at the start of training. During training, we store only the indices of the selected columns, avoiding the need to store full projection matrices for each layer. Our numerical experiments on both pre-training and fine-tuning tasks demonstrate the effectiveness of our dual strategy in approximating optimal low-rank projections, matching the performance of costly SVD-based methods while achieving faster runtime and reduced memory usage.

Related papers

• A Minimalist Optimizer Design for LLM Pretraining [31.996047271119156]
  Training large language models typically relies on adaptives such as Adam.<n>Recent works such as GaLore Fira, and APOLLO have proposed state-compressed variants to reduce memory consumption.<n>In this work, we investigate what is the minimal amount of state that is truly necessary to retain state-of-the-art performance in LLM pretraining.
  arXiv  Detail & Related papers  (2025-06-20T00:10:35Z)
• Efficient Adaptation of Pre-trained Vision Transformer via Householder Transformation [53.88562288388169]
  A common strategy for. Efficient Fine-Tuning (PEFT) of pre-trained Vision Transformers (ViTs) involves adapting the model to downstream tasks. We propose a novel PEFT approach inspired by Singular Value Decomposition (SVD) for representing the adaptation matrix. SVD decomposes a matrix into the product of a left unitary matrix, a diagonal matrix of scaling values, and a right unitary matrix.
  arXiv  Detail & Related papers  (2024-10-30T12:08:30Z)
• PMaF: Deep Declarative Layers for Principal Matrix Features [37.662505982849844]
  We explore two differentiable deep declarative layers, namely least squares on sphere (LESS) and implicit eigen decomposition (IED) LESS can be used to represent data features with a low-dimensional vector containing dominant information from a high-dimensional matrix. IED can be used to represent data features with a low-dimensional vector containing dominant information from a high-dimensional matrix.
  arXiv  Detail & Related papers  (2023-06-26T15:13:36Z)
• Layer-wise Adaptive Step-Sizes for Stochastic First-Order Methods for Deep Learning [8.173034693197351]
  We propose a new per-layer adaptive step-size procedure for first-order optimization methods in deep learning. The proposed approach exploits the layer-wise curvature information contained in the diagonal blocks of the Hessian in deep neural networks (DNNs) to compute adaptive step-sizes (i.e., LRs) for each layer. Numerical experiments show that SGD with momentum and AdamW combined with the proposed per-layer step-sizes are able to choose effective LR schedules.
  arXiv  Detail & Related papers  (2023-05-23T04:12:55Z)
• Memory-Efficient Backpropagation through Large Linear Layers [107.20037639738433]
  In modern neural networks like Transformers, linear layers require significant memory to store activations during backward pass. This study proposes a memory reduction approach to perform backpropagation through linear layers.
  arXiv  Detail & Related papers  (2022-01-31T13:02:41Z)
• Unfolding Projection-free SDP Relaxation of Binary Graph Classifier via GDPA Linearization [59.87663954467815]
  Algorithm unfolding creates an interpretable and parsimonious neural network architecture by implementing each iteration of a model-based algorithm as a neural layer. In this paper, leveraging a recent linear algebraic theorem called Gershgorin disc perfect alignment (GDPA), we unroll a projection-free algorithm for semi-definite programming relaxation (SDR) of a binary graph. Experimental results show that our unrolled network outperformed pure model-based graph classifiers, and achieved comparable performance to pure data-driven networks but using far fewer parameters.
  arXiv  Detail & Related papers  (2021-09-10T07:01:15Z)
• Why Approximate Matrix Square Root Outperforms Accurate SVD in Global Covariance Pooling? [59.820507600960745]
  We propose a new GCP meta-layer that uses SVD in the forward pass, and Pad'e Approximants in the backward propagation to compute the gradients. The proposed meta-layer has been integrated into different CNN models and achieves state-of-the-art performances on both large-scale and fine-grained datasets.
  arXiv  Detail & Related papers  (2021-05-06T08:03:45Z)
• On the Efficient Implementation of the Matrix Exponentiated Gradient Algorithm for Low-Rank Matrix Optimization [26.858608065417663]
  Convex optimization over the spectrahedron has important applications in machine learning, signal processing and statistics. We propose efficient implementations of MEG, which are tailored for optimization with low-rank matrices, and only use a single low-rank SVD on each iteration. We also provide efficiently-computable certificates for the correct convergence of our methods.
  arXiv  Detail & Related papers  (2020-12-18T19:14:51Z)
• Learning Low-rank Deep Neural Networks via Singular Vector Orthogonality Regularization and Singular Value Sparsification [53.50708351813565]
  We propose SVD training, the first method to explicitly achieve low-rank DNNs during training without applying SVD on every step. We empirically show that SVD training can significantly reduce the rank of DNN layers and achieve higher reduction on computation load under the same accuracy.
  arXiv  Detail & Related papers  (2020-04-20T02:40:43Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.