LoRA vs Full Fine-tuning: An Illusion of Equivalence

2024-10-28

arXiv.org

AI Summary

Despite comparable downstream performance, LoRA and full-parameter fine-tuning exhibit fundamental differences in representation learning—particularly in weight spectral structure and generalization behavior. Method: We conduct singular value spectrum analysis, comparative weight structural characterization, and multi-task sequential adaptation experiments to investigate how LoRA’s low-rank updates interact with pretrained representations. Contribution/Results: We identify, for the first time, “intrusive dimensions”—novel singular vector directions activated by LoRA outside the pretrained distribution—which degrade pretrained knowledge retention and impair cross-distribution generalization and robustness in continual multi-task learning. We show that LoRA and full fine-tuning explore distinct parameter subspaces. High-rank LoRA approximates the spectral characteristics of full fine-tuning, while rank-stabilized LoRA effectively suppresses intrusive dimensions, significantly improving knowledge preservation and transfer robustness. These findings provide both theoretical grounding and practical guidance for designing efficient, robust adaptation methods.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Safety and Robustness
• Computer Vision: Representation Learning for Vision

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Fine-tuning is a crucial paradigm for adapting pre-trained large language models to downstream tasks. Recently, methods like Low-Rank Adaptation (LoRA) have been shown to match the performance of fully fine-tuned models on various tasks with an extreme reduction in the number of trainable parameters. Even in settings where both methods learn similarly accurate models, emph{are their learned solutions really equivalent?} We study how different fine-tuning methods change pre-trained models by analyzing the model's weight matrices through the lens of their spectral properties. We find that full fine-tuning and LoRA yield weight matrices whose singular value decompositions exhibit very different structure; moreover, the fine-tuned models themselves show distinct generalization behaviors when tested outside the adaptation task's distribution. More specifically, we first show that the weight matrices trained with LoRA have new, high-ranking singular vectors, which we call emph{intruder dimensions}. Intruder dimensions do not appear during full fine-tuning. Second, we show that LoRA models with intruder dimensions, despite achieving similar performance to full fine-tuning on the target task, become worse models of the pre-training distribution and adapt less robustly to multiple tasks sequentially. Higher-rank, rank-stabilized LoRA models closely mirror full fine-tuning, even when performing on par with lower-rank LoRA models on the same tasks. These results suggest that models updated with LoRA and full fine-tuning access different parts of parameter space, even when they perform equally on the fine-tuned distribution. We conclude by examining why intruder dimensions appear in LoRA fine-tuned models, why they are undesirable, and how their effects can be minimized.

Problem

• Comparing LoRA and full fine-tuning equivalence in LLMs
• Analyzing spectral properties of weight matrices post-fine-tuning
• Investigating intruder dimensions' impact on forgetting in LoRA

Innovation

• Analyzes spectral properties of weight matrices
• Identifies intruder dimensions in LoRA
• Scales intruder dimensions to reduce forgetting