ShadowPEFT: Shadow Network for Parameter-Efficient Fine-Tuning

TL;DR

ShadowPEFT introduces a layer-level refinement framework for parameter-efficient fine-tuning of large language models, offering a flexible alternative to traditional weight perturbation methods like LoRA.

Contribution

It proposes a centralized shadow module for layer-space adaptation, enabling reuse, pretraining, and deployment flexibility across various scenarios.

Findings

ShadowPEFT matches or outperforms LoRA and DoRA with similar parameter budgets.

The shadow module can be pretrained independently and reused across layers.

Experiments show advantages in inference latency and system-level deployment.

Abstract

Parameter-efficient fine-tuning (PEFT) reduces the training cost of full-parameter fine-tuning for large language models (LLMs) by training only a small set of task-specific parameters while freezing the pretrained backbone. However, existing approaches, such as Low-Rank Adaptation (LoRA), achieve adaptation by inserting independent low-rank perturbations directly to individual weights, resulting in a local parameterization of adaptation. We propose ShadowPEFT, a centralized PEFT framework that instead performs layer-level refinement through a depth-shared shadow module. At each transformer layer, ShadowPEFT maintains a parallel shadow state and evolves it repeatedly for progressively richer hidden states. This design shifts adaptation from distributed weight-space perturbations to a shared layer-space refinement process. Since the shadow module is decoupled from the backbone, it can be reused across depth, independently pretrained, and optionally deployed in a detached mode, benefiting edge computing scenarios. Experiments on generation and understanding benchmarks show that ShadowPEFT matches or outperforms LoRA and DoRA under comparable trainable-parameter budgets. Additional analyses on shadow pretraining, cross-dataset transfer, parameter scaling, inference latency, and system-level evaluation suggest that centralized layer-space adaptation is a competitive and flexible alternative to conventional low-rank PEFT.