Mixture-of-Subspaces in Low-Rank Adaptation

TL;DR

This paper introduces MoSLoRA, a novel low-rank adaptation method that mixes subspaces to improve performance across various large models and tasks, demonstrating robustness and efficiency.

Contribution

The paper proposes a new subspace-based Low-Rank Adaptation method, MoSLoRA, which jointly learns a mixer to enhance model performance across multiple modalities.

Findings

MoSLoRA outperforms standard LoRA on diverse tasks

Jointly learning the mixer improves adaptation effectiveness

The method is computationally efficient and easy to implement

Abstract

In this paper, we introduce a subspace-inspired Low-Rank Adaptation (LoRA) method, which is computationally efficient, easy to implement, and readily applicable to large language, multimodal, and diffusion models. Initially, we equivalently decompose the weights of LoRA into two subspaces, and find that simply mixing them can enhance performance. To study such a phenomenon, we revisit it through a fine-grained subspace lens, showing that such modification is equivalent to employing a fixed mixer to fuse the subspaces. To be more flexible, we jointly learn the mixer with the original LoRA weights, and term the method Mixture-of-Subspaces LoRA (MoSLoRA). MoSLoRA consistently outperforms LoRA on tasks in different modalities, including commonsense reasoning, visual instruction tuning, and subject-driven text-to-image generation, demonstrating its effectiveness and robustness. Codes are available at https://github.com/wutaiqiang/MoSLoRA.