Gespeichert in:
| Hauptverfasser: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | https://arxiv.org/abs/2601.07309 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866911368891858944 |
|---|---|
| author | Feng, Zhuoka Chen, Kang Zhao, Sihan Xiong, Kai Wang, Yaoning Yu, Minshen Nian, Junjie Xiao, Changyi Cao, Yixin Jiang, Yugang |
| author_facet | Feng, Zhuoka Chen, Kang Zhao, Sihan Xiong, Kai Wang, Yaoning Yu, Minshen Nian, Junjie Xiao, Changyi Cao, Yixin Jiang, Yugang |
| contents | Interactive large language model agents have advanced rapidly, but most remain specialized to a single environment and fail to adapt robustly to other environments. Model merging offers a training-free alternative by integrating multiple experts into a single model. In this paper, we propose Agent-Role Merging (ARM), an activation-guided, role-conditioned neuron transplantation method for model merging in LLM agents. ARM improves existing merging methods from static natural language tasks to multi-turn agent scenarios, and over the generalization ability across various interactive environments. This is achieved with a well designed 3-step framework: 1) constructing merged backbones, 2) selection based on its role-conditioned activation analysis, and 3) neuron transplantation for fine-grained refinements. Without gradient-based optimization, ARM improves cross-benchmark generalization while enjoying efficiency. Across diverse domains, the model obtained via ARM merging outperforms prior model merging methods and domain-specific expert models, while demonstrating strong out-of-domain generalization. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_07309 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | ARM: Role-Conditioned Neuron Transplantation for Training-Free Generalist LLM Agent Merging Feng, Zhuoka Chen, Kang Zhao, Sihan Xiong, Kai Wang, Yaoning Yu, Minshen Nian, Junjie Xiao, Changyi Cao, Yixin Jiang, Yugang Artificial Intelligence Machine Learning Interactive large language model agents have advanced rapidly, but most remain specialized to a single environment and fail to adapt robustly to other environments. Model merging offers a training-free alternative by integrating multiple experts into a single model. In this paper, we propose Agent-Role Merging (ARM), an activation-guided, role-conditioned neuron transplantation method for model merging in LLM agents. ARM improves existing merging methods from static natural language tasks to multi-turn agent scenarios, and over the generalization ability across various interactive environments. This is achieved with a well designed 3-step framework: 1) constructing merged backbones, 2) selection based on its role-conditioned activation analysis, and 3) neuron transplantation for fine-grained refinements. Without gradient-based optimization, ARM improves cross-benchmark generalization while enjoying efficiency. Across diverse domains, the model obtained via ARM merging outperforms prior model merging methods and domain-specific expert models, while demonstrating strong out-of-domain generalization. |
| title | ARM: Role-Conditioned Neuron Transplantation for Training-Free Generalist LLM Agent Merging |
| topic | Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2601.07309 |