Very Large-Scale Multi-Agent Simulation in AgentScope

TL;DR

<3-5 sentence high-level summary> This paper addresses core challenges in large-scale, multi-agent simulations by extending AgentScope with an actor-based distributed mechanism, automatic workflow conversion, flexible inter-agent and agent-environment interactions, and easy-to-use heterogeneous configurations. It also introduces a web-based Agent-Manager for monitoring and managing agents across devices, and demonstrates scalability by simulating up to 1 million agents on a 4-device cluster using diverse LLMs. Through extensive experiments, the authors analyze the effects of prompts, LLM mixtures, and background diversity on agent behavior and convergence, highlighting the potential of their framework for realistic, scalable, and diverse social simulations. The work provides practical tooling and empirical insights to advance research and applications in large-scale, LLM-driven agent-based systems, with code released on GitHub for reproducibility and further exploration.

Abstract

Recent advances in large language models (LLMs) have opened new avenues for applying multi-agent systems in very large-scale simulations. However, there remain several challenges when conducting multi-agent simulations with existing platforms, such as limited scalability and low efficiency, unsatisfied agent diversity, and effort-intensive management processes. To address these challenges, we develop several new features and components for AgentScope, a user-friendly multi-agent platform, enhancing its convenience and flexibility for supporting very large-scale multi-agent simulations. Specifically, we propose an actor-based distributed mechanism as the underlying technological infrastructure towards great scalability and high efficiency, and provide flexible environment support for simulating various real-world scenarios, which enables parallel execution of multiple agents, automatic workflow conversion for distributed deployment, and both inter-agent and agent-environment interactions. Moreover, we integrate an easy-to-use configurable tool and an automatic background generation pipeline in AgentScope, simplifying the process of creating agents with diverse yet detailed background settings. Last but not least, we provide a web-based interface for conveniently monitoring and managing a large number of agents that might deploy across multiple devices. We conduct a comprehensive simulation to demonstrate the effectiveness of these proposed enhancements in AgentScope, and provide detailed observations and insightful discussions to highlight the great potential of applying multi-agent systems in large-scale simulations. The source code is released on GitHub at https://github.com/modelscope/agentscope/tree/main/examples/paper_large_scale_simulation to inspire further research and development in large-scale multi-agent simulations.

Figures (20)

• Figure 1: An example of automatic parallel execution, where circles represent agents and directed edges represent message passing flows.
• Figure 2: The example of the automatic conversation of a centralized workflow with four agents running sequentially (on the left) into a distributed workflow with agents running in parallel (on the right). Users only need to add a to_dist function during the initialization phase, without any further modifications required.
• Figure 3: Multi-layer environment structure for agent-based simulation.
• Figure 4:
• Figure 5: Agent-based simulations with varying scales of agents (a, b) and varying numbers of devices (c, d).