Research map / Large language models / Instruction and fine-tuning
FREE-Merging: Fourier Transform for Efficient Model Merging
Shenghe Zheng, Hongzhi Wang
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 FR-Merging high-pass filtering (ablation) | backbone · ViT-B/32 8 tasks | avg accuracy: 70.10 -> 76.96 | Improved |
| add FR-Merging high-pass filtering (ablation) | backbone · ViT-L/14 8 tasks | avg accuracy: 84.50 -> 88.21 | Improved |
| add Eq.3 merging coefficients (ablation) | FR-Merging · ViT-B/32 8 tasks | avg accuracy: 76.96 -> 78.10 | Improved |
| add top-K experts without rescaling (ablation) | backbone · ViT-B/32 8 tasks | avg accuracy: 70.10 -> 65.24 | Hurt |
| add rescaling to top-K experts (ablation) | FREE-Merging · ViT-B/32 8 tasks | avg accuracy: 79.38 -> 89.68 | Improved |
| add FR-Merging (ablation) | Ties-Merging · ViT-B/32 8 tasks | avg accuracy: 73.6 -> 78.3 | Improved |
| add FR-Merging (ablation) | Breadcrumbs · ViT-B/32 8 tasks | avg accuracy: 75.0 -> 78.8 | Improved |
| add FR-Merging (ablation) | TSVM · ViT-L/14 8 tasks | avg accuracy: 91.9 -> 92.5 | Improved |
| add FR-Merging (ablation) | TSVM · ViT-B/32 8 tasks | avg accuracy: 83.9 -> 84.8 | Improved |
| replace PCB-Merging with FR-Merging | cost-free merging · ViT-B/32 8 tasks | avg accuracy: 75.8 -> 78.1 | Improved |
| replace Twin-Merging with FREE-Merging | expert-based merging · ViT-L/14 8 tasks | avg accuracy: 92.7 -> 93.7 | Improved |
| replace Twin-Merging with FREE-Merging | expert-based merging · 30 vision tasks | avg accuracy: 75.52 -> 79.67 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Twin-Merging — Twin-Merging: Dynamic Integration of Modular Expertise in Model Merging (NeurIPS 2024) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Mitigating Parameter Interference in Model Merging via Sharpness-Aware Fine-Tuning (ICLR 2025 · similarity 0.90)
- Whoever Started the interference Should End It: Guiding Data-Free Model Merging via Task Vectors (ICML 2025 · similarity 0.90)
- EMR-Merging: Tuning-Free High-Performance Model Merging (NeurIPS 2024 · similarity 0.90)
- Merging on the Fly Without Retraining: A Sequential Approach to Scalable Continual Model Merging (NeurIPS 2025 · similarity 0.89)
- MergeBench: A Benchmark for Merging Domain-Specialized LLMs (NeurIPS 2025 · similarity 0.89)
- Contribution-Aware Structured Sparsity for Model Merging (NeurIPS 2026 · similarity 0.88)
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.
