Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Bi-LoRA: Efficient Sharpness-Aware Minimization for Fine-Tuning Large-Scale Models

About

Fine-tuning large-scale pre-trained models with limited data presents significant challenges for generalization. While Sharpness-Aware Minimization (SAM) has proven effective in improving generalization by seeking flat minima, its substantial extra memory and computation overhead make it impractical for large models. Integrating SAM with parameter-efficient fine-tuning methods like Low-Rank Adaptation (LoRA) is a promising direction. However, we find that directly applying SAM to LoRA parameters limits the sharpness optimization to a restricted subspace, hindering its effectiveness. To address this limitation, we propose Bi-directional Low-Rank Adaptation (Bi-LoRA), which introduces an auxiliary LoRA module to model SAM's adversarial weight perturbations. It decouples SAM's weight perturbations from LoRA optimization: the primary LoRA module adapts to specific tasks via standard gradient descent, while the auxiliary module captures the sharpness of the loss landscape through gradient ascent. Such dual-module design enables Bi-LoRA to capture broader sharpness for achieving flatter minima while remaining memory-efficient. Another important benefit is that the dual design allows for simultaneous optimization and perturbation, eliminating SAM's doubled training costs. Extensive experiments across diverse tasks and architectures demonstrate Bi-LoRA's efficiency and effectiveness in enhancing generalization.

Yuhang Liu, Tao Li, Zhehao Huang, Zuopeng Yang, Xiaolin Huang• 2025

Related benchmarks

TaskDatasetResultRank
Mathematical ReasoningGSM8K
Accuracy60.32
1398
DialogueMT-Bench
MT-Bench Score6.26
41
Instruction FollowingBBH--
40
Code GenerationHumanEval
Pass@127.2
36
Instruction FollowingDROP
DROP Score51.53
20
Instruction FollowingMMLU
MMLU Accuracy63.67
20
Instruction FollowingHEval
PASS@146.12
12
Instruction FollowingInstruction-following Evaluation Suite (MMLU, DROP, HEval, BBH) (test)
MMLU79.67
11
Natural Language UnderstandingGLUE
MNLI Accuracy86.33
4
Natural Language UnderstandingSuperGLUE (val)
BoolQ Accuracy72.25
4
Showing 10 of 11 rows

Other info

Follow for update