AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Don't Reinvent the Wheel, Just Realign the Spokes: Resource-Efficient Federated Fine-Tuning via Rank-Wise Expert Assembly

Yebo Wu, Jingguang Li, Zhijiang Guo, Li Li

ICML 2026 · Large language models › Instruction and fine-tuning · approach: Efficient LLM fine-tuning systems

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add Mixture of Rank-Wise Experts (ablation)SmartFed · LLaMA2-7B averageaccuracy: 46.52 -> 50.75Improved
add Elastic Expert Quota Allocation (ablation)SmartFed · LLaMA2-7B averageaccuracy: 47.30 -> 50.75Improved
add Mixture of Rank-Wise Experts (ablation)SmartFed · LLaMA2-13B averageaccuracy: 58.00 -> 62.95Improved
add Elastic Expert Quota Allocation (ablation)SmartFed · LLaMA2-13B averageaccuracy: 59.64 -> 62.95Improved
reuse existing LoRA modules with rank-wise routing instead of training LoRA from scratch (FedIT)SmartFed · LLaMA2-7B averageaccuracy: 44.03 -> 50.75Improved
reuse LoRA via parameter merging (Linear Arithmetic) instead of rank-wise routingLinear Arithmetic · LLaMA2-7B averageaccuracy: 47.26 -> 42.71Hurt
replace coarse module-level routing (MoLE) with rank-wise expertsMoLE · LLaMA2-7B averageaccuracy: 47.26 -> 50.75Improved
replace LoRA-Flow with SmartFedLoRA-Flow · LLaMA2-7B averageaccuracy: 47.59 -> 50.75Improved
use SmartFed on newer backbone LLaMA3-3BFedIT · LLaMA3-3B averageaccuracy: 41.70 -> 53.56Improved
use SmartFed on LLaMA3-8BFedIT · LLaMA3-8B averageaccuracy: 56.20 -> 70.02Improved
use SmartFed on Qwen3-8BFedIT · Qwen3-8B averageaccuracy: 58.43 -> 72.25Improved
increase Query expert quota from 16 to 64 at fixed Value quota 16MoRE · Chinese+Math, LLaMA2-7Baccuracy: 30.58 -> 36.98Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy