Saliency-Aware Model Merging
About
Model merging aims to consolidate multiple task-specific models fine-tuned on different datasets into a unified architecture that performs cross-domain proficiency. Current data-free model merging methods often struggle to scale as they rely on simple parameter-level heuristics that ignore inter-layer dependencies and non-uniform distribution of expertise. This work proposes SA-Merging, which is built upon connectivity-based saliency formulations from structural pruning (e.g., SynFlow) and extends them to the data-free model merging setting. We define a saliency score over task vectors relative to a shared base model, and further introduce merge-aware modulation that incorporates agreement across experts to mitigate task interference. Based on this formulation, an iterative saliency-aware merging procedure progressively removes non-informative updates while preserving end-to-end connectivity. Furthermore, we extend SA-Merging to introduce rank-wise saliency decomposition for LoRAs without compromising their structural integrity. Extensive experiments on vision and language tasks demonstrate the effectiveness of our saliency-based approach, further reducing the gap between data-free and test-time adaptation methods.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Image Classification | Stanford Cars | Accuracy71.8 | 660 | |
| Image Classification | DTD | Accuracy71 | 599 | |
| Image Classification | RESISC45 | Accuracy86.5 | 472 | |
| Image Classification | Vision Multi-task Suite (SUN397, Cars, RESISC45, EuroSAT, SVHN, GTSRB, MNIST, DTD) | Average Accuracy89.8 | 104 | |
| Natural Language Understanding | GLUE (test) | QNLI91.5 | 47 | |
| Natural Language Understanding | GLUE | Average GLUE Score90.2 | 30 | |
| Image Classification | GTSRB | Accuracy95 | 21 | |
| Image Classification | MNIST | Accuracy99.6 | 20 | |
| Image Classification | 8-task vision suite CLIP ViT-L/14 (test) | SUN397 Accuracy82 | 14 | |
| Model Merging | LLM Benchmark Family MMLU, TruthfulQA, BBQ, CNN/DailyMail | MMLU Score69.87 | 5 |