Expanding Sparse Tuning for Low Memory Usage

TL;DR

SNELL introduces a low-memory sparse tuning method for vision models by decomposing matrices into low-rank forms and applying nonlinear kernels, achieving state-of-the-art results efficiently.

Contribution

The paper proposes SNELL, a novel sparse tuning approach that reduces memory usage by matrix decomposition and nonlinear merging, enabling effective large-scale model adaptation.

Findings

SNELL achieves state-of-the-art performance on multiple tasks.

It significantly reduces memory usage compared to existing sparse tuning methods.

The method effectively adapts large pre-trained models to downstream tasks.

Abstract

Parameter-efficient fine-tuning (PEFT) is an effective method for adapting pre-trained vision models to downstream tasks by tuning a small subset of parameters. Among PEFT methods, sparse tuning achieves superior performance by only adjusting the weights most relevant to downstream tasks, rather than densely tuning the whole weight matrix. However, this performance improvement has been accompanied by increases in memory usage, which stems from two factors, i.e., the storage of the whole weight matrix as learnable parameters in the optimizer and the additional storage of tunable weight indexes. In this paper, we propose a method named SNELL (Sparse tuning with kerNELized LoRA) for sparse tuning with low memory usage. To achieve low memory usage, SNELL decomposes the tunable matrix for sparsification into two learnable low-rank matrices, saving from the costly storage of the whole original matrix. A competition-based sparsification mechanism is further proposed to avoid the storage of tunable weight indexes. To maintain the effectiveness of sparse tuning with low-rank matrices, we extend the low-rank decomposition by applying nonlinear kernel functions to the whole-matrix merging. Consequently, we gain an increase in the rank of the merged matrix, enhancing the ability of SNELL in adapting the pre-trained models to downstream tasks. Extensive experiments on multiple downstream tasks show that SNELL achieves state-of-the-art performance with low memory usage, endowing PEFT with sparse tuning to large-scale models. Codes are available at https://github.com/ssfgunner/SNELL.