Make LoRA Great Again: Boosting LoRA with Adaptive Singular Values and Mixture-of-Experts Optimization Alignment

TL;DR

This paper introduces GOAT, a framework that enhances Low-Rank Adaptation (LoRA) for large models by adaptively integrating priors and aligning optimization, significantly improving performance across diverse tasks without changing existing architectures.

Contribution

GOAT proposes an adaptive SVD-based MoE framework and a theoretical scaling factor to improve LoRA's efficiency and performance, bridging the gap with full fine-tuning.

Findings

GOAT achieves state-of-the-art results on 25 datasets.

Proper scaling boosts LoRA MoE performance without architectural changes.

GOAT closes the performance gap with full fine-tuning.

Abstract

While Low-Rank Adaptation (LoRA) enables parameter-efficient fine-tuning for Large Language Models (LLMs), its performance often falls short of Full Fine-Tuning (Full FT). Current methods optimize LoRA by initializing with static singular value decomposition (SVD) subsets, leading to suboptimal leveraging of pre-trained knowledge. Another path for improving LoRA is incorporating a Mixture-of-Experts (MoE) architecture. However, weight misalignment and complex gradient dynamics make it challenging to adopt SVD prior to the LoRA MoE architecture. To mitigate these issues, we propose \underline{G}reat L\underline{o}R\underline{A} Mixture-of-Exper\underline{t} (GOAT), a framework that (1) adaptively integrates relevant priors using an SVD-structured MoE, and (2) aligns optimization with full fine-tuned MoE by deriving a theoretical scaling factor. We demonstrate that proper scaling, without modifying the architecture or training algorithms, boosts LoRA MoE's efficiency and performance. Experiments across 25 datasets, including natural language understanding, commonsense reasoning, image classification, and natural language generation, demonstrate GOAT's state-of-the-art performance, closing the gap with Full FT.