SOLAR: Communication-Efficient Model Adaptation via Subspace-Oriented Latent Adapter Reparametrization

TL;DR

SOLAR is a novel framework that compresses PEFT adapters by leveraging subspace similarity, significantly reducing communication costs while maintaining task performance across language and vision models.

Contribution

It introduces a subspace-oriented reparameterization method that decouples adapter size from PEFT structure, enabling efficient model adaptation with minimal communication overhead.

Findings

Reduces communication costs of PEFT adapters without performance loss.

Compatible with multiple PEFT methods like LoRA and AdaLoRA.

Demonstrates effectiveness on language and vision tasks with large models.

Abstract

Parameter-efficient fine-tuning (PEFT) methods, such as LoRA, enable scalable adaptation of foundation models by injecting low-rank adapters. However, their communication and storage costs remain a major bottleneck in resource-constrained settings. We propose SOLAR (Subspace-Oriented Latent Adapter Reparameterization), a post-training compression framework that substantially reduces the communication cost (i.e., the number of parameters to transmit or store) of PEFT adapters. SOLAR expresses each PEFT update as a linear combination of basis vectors formed from the foundation model's singular vectors with controlled random perturbations. By exploiting the subspace similarity (the alignment of principal directions) between the foundation model and task-specific fine-tuned updates, SOLAR decouples the adapter size from PEFT structure and ensures compact yet expressive representations. It is model-agnostic and compatible with existing PEFT methods, including LoRA, AdaLoRA, and other adapter modules. We theoretically establish a bound on the reconstruction error. Experiments on language and vision tasks using LLaMA, GPT, and ViT models demonstrate that SOLAR preserves task performance while significantly reducing model representation sizes, offering an effective and communication-efficient solution for deployment in distributed systems and edge devices.