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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add Mixture of Rank-Wise Experts (ablation) | SmartFed · LLaMA2-7B average | accuracy: 46.52 -> 50.75 | Improved |
| add Elastic Expert Quota Allocation (ablation) | SmartFed · LLaMA2-7B average | accuracy: 47.30 -> 50.75 | Improved |
| add Mixture of Rank-Wise Experts (ablation) | SmartFed · LLaMA2-13B average | accuracy: 58.00 -> 62.95 | Improved |
| add Elastic Expert Quota Allocation (ablation) | SmartFed · LLaMA2-13B average | accuracy: 59.64 -> 62.95 | Improved |
| reuse existing LoRA modules with rank-wise routing instead of training LoRA from scratch (FedIT) | SmartFed · LLaMA2-7B average | accuracy: 44.03 -> 50.75 | Improved |
| reuse LoRA via parameter merging (Linear Arithmetic) instead of rank-wise routing | Linear Arithmetic · LLaMA2-7B average | accuracy: 47.26 -> 42.71 | Hurt |
| replace coarse module-level routing (MoLE) with rank-wise experts | MoLE · LLaMA2-7B average | accuracy: 47.26 -> 50.75 | Improved |
| replace LoRA-Flow with SmartFed | LoRA-Flow · LLaMA2-7B average | accuracy: 47.59 -> 50.75 | Improved |
| use SmartFed on newer backbone LLaMA3-3B | FedIT · LLaMA3-3B average | accuracy: 41.70 -> 53.56 | Improved |
| use SmartFed on LLaMA3-8B | FedIT · LLaMA3-8B average | accuracy: 56.20 -> 70.02 | Improved |
| use SmartFed on Qwen3-8B | FedIT · Qwen3-8B average | accuracy: 58.43 -> 72.25 | Improved |
| increase Query expert quota from 16 to 64 at fixed Value quota 16 | MoRE · Chinese+Math, LLaMA2-7B | accuracy: 30.58 -> 36.98 | Improved |
Closest papers by meaning
- Adaptive LoRA Experts Allocation and Selection for Federated Fine-Tuning (NeurIPS 2025 · similarity 0.91)
- Fed-SB: A Silver Bullet for Extreme Communication Efficiency and Performance in (Private) Federated LoRA Fine-Tuning (NeurIPS 2026 · similarity 0.90)
- Robust Federated Finetuning of LLMs via Alternating Optimization of LoRA (NeurIPS 2025 · similarity 0.90)
- Ravan: Multi-Head Low-Rank Adaptation for Federated Fine-Tuning (NeurIPS 2025 · similarity 0.90)
- FLoRA: Federated Fine-Tuning Large Language Models with Heterogeneous Low-Rank Adaptations (NeurIPS 2024 · similarity 0.90)
- Federated Full-Parameter Tuning of Billion-Sized Language Models with Communication Cost under 18 Kilobytes (ICML 2024 · similarity 0.89)
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.
