Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics
This work addresses a practical issue in parameter-efficient fine-tuning for machine learning practitioners, offering an incremental improvement to LoRA's stability.
The paper investigates the impact of initializing LoRA matrices to non-zero values instead of zero, finding that this improves robustness to suboptimal learning rates without harming fine-tuning performance, as validated through experiments on various models and datasets.
Low-rank adaptation (LoRA) is a widely used parameter-efficient fine-tuning method. In standard LoRA layers, one of the matrices, $A$ or $B$, is initialized to zero, ensuring that fine-tuning starts from the pretrained model. However, there is no theoretical support for this practice. In this paper, we investigate the impact of non-zero initialization on LoRA's fine-tuning dynamics from an infinite-width perspective. Our analysis reveals that, compared to zero initialization, simultaneously initializing $A$ and $B$ to non-zero values improves LoRA's robustness to suboptimal learning rates, particularly smaller ones. Further analysis indicates that although the non-zero initialization of $AB$ introduces random noise into the pretrained weight, it generally does not affect fine-tuning performance. In other words, fine-tuning does not need to strictly start from the pretrained model. The validity of our findings is confirmed through extensive experiments across various models and datasets. The code is available at https://github.com/Leopold1423/non_zero_lora-icml25.