FLoE: Fisher-Based Layer Selection for Efficient Sparse Adaptation of Low-Rank Experts

TL;DR

FLoE introduces a Fisher-guided, sparse layer selection and Bayesian rank optimization for efficient, task-specific adaptation of large language models, reducing redundancy and improving resource efficiency.

Contribution

FLoE presents a novel importance scoring and automatic rank allocation method for sparse, efficient fine-tuning of LLMs, surpassing uniform adaptation approaches.

Findings

FLoE achieves better efficiency-accuracy trade-offs across multiple benchmarks.

It significantly reduces parameter redundancy compared to uniform PEFT methods.

FLoE adapts effectively in resource-constrained environments.

Abstract

Parameter-Efficient Fine-Tuning (PEFT) methods have emerged as a widely adopted strategy for adapting pre-trained Large Language Models (LLMs) to downstream tasks, significantly reducing memory and computational costs. However, most existing PEFT techniques uniformly deploy LoRA adapters across all layers, disregarding the intrinsic heterogeneity of layer contributions and task-specific rank requirements. This uniform paradigm leads to redundant parameter allocation and suboptimal adaptation efficiency. To address these limitations, we propose FLoE, a novel PEFT framework that introduces two key innovations: (i) a Fisher information-guided importance scoring mechanism to dynamically identify task-critical transformer layers for MoE-based low-rank adaptation, enabling sparse adapter deployment; and (ii) a Bayesian optimization-driven rank allocator that automatically determines optimal LoRA ranks on specific datasets without exhaustive grid search. Extensive experiments across diverse LLMs and benchmarks reveal that FLoE achieves impressive efficiency-accuracy trade-offs, making FLoE particularly advantageous in resource-constrained environments that necessitate rapid adaptation.