Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CoMoL: Efficient Mixture of LoRA Experts via Dynamic Core Space Merging

Jie Cao, Zhenxuan Fan, Zhuonan Wang, Tianwei Lin, Ziyuan Zhao, Rolan Yan, Wenqiao Zhang, Feifei Shao, Hongwei Wang, Jun Xiao, Siliang Tang

TL;DR

Extensive experiments demonstrate that CoMoL retains the adaptability of MoE-LoRA architectures while achieving parameter efficiency comparable to standard LoRA, consistently outperforming existing methods across multiple tasks.

Abstract

Large language models (LLMs) achieve remarkable performance on diverse downstream and domain-specific tasks via parameter-efficient fine-tuning (PEFT). However, existing PEFT methods, particularly MoE-LoRA architectures, suffer from limited parameter efficiency and coarse-grained adaptation due to the proliferation of LoRA experts and instance-level routing. To address these issues, we propose Core Space Mixture of LoRA (\textbf{CoMoL}), a novel MoE-LoRA framework that incorporates expert diversity, parameter efficiency, and fine-grained adaptation. Specifically, CoMoL introduces two key components: core space experts and core space routing. Core space experts store each expert in a compact core matrix, preserving diversity while controlling parameter growth. Core space routing dynamically selects and activates the appropriate core experts for each token, enabling fine-grained, input-adaptive routing. Activated core experts are then merged via a soft-merging strategy into a single core expert, which is combined with a shared LoRA to form a specialized LoRA module. Besides, the routing network is projected into the same low-rank space as the LoRA matrices, further reducing parameter overhead without compromising expressiveness. Extensive experiments demonstrate that CoMoL retains the adaptability of MoE-LoRA architectures while achieving parameter efficiency comparable to standard LoRA, consistently outperforming existing methods across multiple tasks.

CoMoL: Efficient Mixture of LoRA Experts via Dynamic Core Space Merging

TL;DR

Extensive experiments demonstrate that CoMoL retains the adaptability of MoE-LoRA architectures while achieving parameter efficiency comparable to standard LoRA, consistently outperforming existing methods across multiple tasks.

Abstract

Large language models (LLMs) achieve remarkable performance on diverse downstream and domain-specific tasks via parameter-efficient fine-tuning (PEFT). However, existing PEFT methods, particularly MoE-LoRA architectures, suffer from limited parameter efficiency and coarse-grained adaptation due to the proliferation of LoRA experts and instance-level routing. To address these issues, we propose Core Space Mixture of LoRA (\textbf{CoMoL}), a novel MoE-LoRA framework that incorporates expert diversity, parameter efficiency, and fine-grained adaptation. Specifically, CoMoL introduces two key components: core space experts and core space routing. Core space experts store each expert in a compact core matrix, preserving diversity while controlling parameter growth. Core space routing dynamically selects and activates the appropriate core experts for each token, enabling fine-grained, input-adaptive routing. Activated core experts are then merged via a soft-merging strategy into a single core expert, which is combined with a shared LoRA to form a specialized LoRA module. Besides, the routing network is projected into the same low-rank space as the LoRA matrices, further reducing parameter overhead without compromising expressiveness. Extensive experiments demonstrate that CoMoL retains the adaptability of MoE-LoRA architectures while achieving parameter efficiency comparable to standard LoRA, consistently outperforming existing methods across multiple tasks.
Paper Structure (30 sections, 15 equations, 2 figures, 5 tables)

This paper contains 30 sections, 15 equations, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Revisiting MoE-LoRA Methods
  3. LoRA Basics
  4. MoE-LoRA Methods
  5. Soft routing mechanism.
  6. Sparse routing mechanism.
  7. Soft-merging MoE-LoRA Method
  8. Comparison of MoE-LoRA Methods
  9. Technical Details of CoMoL
  10. Core Space of LoRA
  11. Core Space Mixture of LoRA
  12. Parameter Efficiency
  13. Token-level soft-merging of experts in Core Space
  14. Core Space Routing
  15. Experiments
  16. ...and 15 more sections

Figures (2)

  • Figure 1: Comparison of LoRA-based methods.(a) LoRA injects trainable low-rank matrices $\boldsymbol{A}$ and $\boldsymbol{B}$ into transformer layers to approximate weight updates. (b) MoE-LoRA employs multiple LoRA experts with a routing network to select and activate experts based on input tokens. (c) CoMoL fuses expert parameters in the low-rank core space, preserving token-level routing while maintaining expert-specific expressiveness with minimal parameters.
  • Figure 2: Performance and trainable parameter comparison between CoMoL and LoRA across different ranks on the code generation benchmark.