Research map / Architectures and efficiency / Pruning and efficient tuning
Demystifying Mergeability: Interpretable Properties to Predict Model Merging Success
Luca Zhou, Bo Zhao, Rose Yu, Emanuele Rodolà
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 gradient-based metrics (ablation) | L1 linear mergeability predictor · LOTO validation, ViT-B/16 | avg change in Pearson r: -0.149 | Improved |
| add gradient-based metrics (ablation) | L1 linear mergeability predictor · LOTO validation, ViT-B/32 | avg change in Pearson r: -0.131 | Improved |
| add subspace-overlap metrics (ablation) | L1 linear mergeability predictor · LOTO validation, ViT-B/16 | avg change in Pearson r: -0.045 | Improved |
| add effective-rank metrics (ablation) | L1 linear mergeability predictor · LOTO validation, ViT-B/16 | avg change in Pearson r: +0.020 | No clear effect |
| add activation metrics (ablation) | L1 linear mergeability predictor · LOTO validation, ViT-B/32 | avg change in Pearson r: -0.059 | Improved |
| add L1 regularization (lambda=1.0) (ablation) | linear mergeability predictor · LOTO validation, ViT-B/16 | Pearson r: 0.525 -> 0.544 | Improved |
| add L1 regularization (lambda=1.0) (ablation) | linear mergeability predictor · LOTO validation, ViT-B/32 | Pearson r: 0.676 -> 0.648 | Hurt |
| add L1 regularization (lambda=1.0) (ablation) | linear mergeability predictor · fold-to-fold std of validation r, ViT-B/16 | std of Pearson r: 0.254 -> 0.159 | Improved |
| use linear L1 predictor instead of MLP (ablation) | mergeability predictor · LOTO validation, ViT-B/16 | Pearson r: 0.493 -> 0.544 | Improved |
| use Pearson-correlation objective instead of MSE (ablation) | L1 linear mergeability predictor · LOTO validation, ViT-B/16 | Pearson r: 0.147 -> 0.544 | Improved |
| use Pearson-correlation objective instead of MSE (ablation) | L1 linear mergeability predictor · LOTO validation, ViT-B/32 | Pearson r: 0.143 -> 0.648 | Improved |
| fine-tune with gradient-magnitude penalty | task-specific ViT-B/16 models · post-merge normalized accuracy | average delta: +0.41 | Improved |
Closest papers by meaning
- MergeBench: A Benchmark for Merging Domain-Specialized LLMs (NeurIPS 2025 · similarity 0.88)
- MergeNet: Knowledge Migration Across Heterogeneous Models, Tasks, and Modalities (AAAI 2025 · similarity 0.88)
- Whoever Started the interference Should End It: Guiding Data-Free Model Merging via Task Vectors (ICML 2025 · similarity 0.87)
- Parameter Competition Balancing for Model Merging (NeurIPS 2024 · similarity 0.86)
- Beyond Parameter Arithmetic: Sparse Complementary Fusion for Distribution-Aware Model Merging (NeurIPS 2026 · similarity 0.86)
- Rethinking Layer-wise Model Merging through Chain of Merges (NeurIPS 2026 · similarity 0.86)
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.
