DenseLoRA: Dense Low-Rank Adaptation of Large Language Models

TL;DR

DenseLoRA is a novel adaptation method for large language models that improves parameter efficiency and performance by using a dense low-rank matrix and a unified encoder-decoder structure.

Contribution

It introduces DenseLoRA, a new approach that enhances parameter utilization and performance over traditional LoRA by employing a dense low-rank matrix and representation fine-tuning.

Findings

DenseLoRA achieves 83.8% accuracy with only 0.01% trainable parameters.

It outperforms LoRA, which achieves 80.8% accuracy with 0.70% trainable parameters.

Extensive experiments validate the effectiveness of DenseLoRA components.

Abstract

Low-rank adaptation (LoRA) has been developed as an efficient approach for adapting large language models (LLMs) by fine-tuning two low-rank matrices, thereby reducing the number of trainable parameters. However, prior research indicates that many of the weights in these matrices are redundant, leading to inefficiencies in parameter utilization. To address this limitation, we introduce Dense Low-Rank Adaptation (DenseLoRA), a novel approach that enhances parameter efficiency while achieving superior performance compared to LoRA. DenseLoRA builds upon the concept of representation fine-tuning, incorporating a single Encoder-Decoder to refine and compress hidden representations across all adaptation layers before applying adaptation. Instead of relying on two redundant low-rank matrices as in LoRA, DenseLoRA adapts LLMs through a dense low-rank matrix, improving parameter utilization and adaptation efficiency. We evaluate DenseLoRA on various benchmarks, showing that it achieves 83.8% accuracy with only 0.01% of trainable parameters, compared to LoRA's 80.8% accuracy with 0.70% of trainable parameters on LLaMA3-8B. Additionally, we conduct extensive experiments to systematically assess the impact of DenseLoRA's components on overall model performance. Code is available at https://github.com/mulin-ahu/DenseLoRA.