ARM: Role-Conditioned Neuron Transplantation for Training-Free Generalist LLM Agent Merging

TL;DR

ARM introduces a training-free neuron transplantation method for merging multiple large language model experts, enhancing multi-environment adaptability and out-of-domain generalization without gradient updates.

Contribution

It presents a novel role-conditioned neuron transplantation framework that improves model merging for interactive LLM agents across diverse environments.

Findings

Outperforms prior merging methods and domain-specific models

Enhances cross-benchmark and out-of-domain generalization

Operates efficiently without gradient-based optimization

Abstract

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.