AlphaLoRA: Assigning LoRA Experts Based on Layer Training Quality

TL;DR

AlphaLoRA is a novel method that allocates LoRA experts across model layers based on training quality, reducing redundancy and improving performance in large language models without additional training.

Contribution

The paper introduces AlphaLoRA, a theoretically grounded, training-free approach for assigning LoRA experts based on layer training quality, addressing redundancy issues in MoE architectures.

Findings

AlphaLoRA achieves comparable or better performance than baselines.

It effectively reduces redundancy in LoRA experts.

Demonstrates versatility across multiple models and benchmarks.

Abstract

Parameter-efficient fine-tuning methods, such as Low-Rank Adaptation (LoRA), are known to enhance training efficiency in Large Language Models (LLMs). Due to the limited parameters of LoRA, recent studies seek to combine LoRA with Mixture-of-Experts (MoE) to boost performance across various tasks. However, inspired by the observed redundancy in traditional MoE structures, previous studies identify similar redundancy among LoRA experts within the MoE architecture, highlighting the necessity for non-uniform allocation of LoRA experts across different layers. In this paper, we leverage Heavy-Tailed Self-Regularization (HT-SR) Theory to design a fine-grained allocation strategy. Our analysis reveals that the number of experts per layer correlates with layer training quality, which exhibits significant variability across layers. Based on this, we introduce AlphaLoRA, a theoretically principled and training-free method for allocating LoRA experts to further mitigate redundancy. Experiments on three models across ten language processing and reasoning benchmarks demonstrate that AlphaLoRA achieves comparable or superior performance over all baselines. Our code is available at https://github.com/morelife2017/alphalora.