LMAgent is an innovative platform designed to create a large-scale and multimodal agent society capable of simulating complex multi-user behaviors. Agents in the LMAgent society can engage in a variety of activities, including chatting, browsing, purchasing, reviewing products, and even conducting live e-commerce streams. This environment can serves as a testbed for research in areas such as urban planning, social dynamics, and e-commerce.
Multimodal Interaction: Agents in the LMAgent society can interact using various modalities, including text, images, and potentially audio and video.
Scalability: Designed to efficiently simulate over 10,000 agents, enabling large-scale experiments.
Self-consistency Prompting: A novel mechanism that ensures consistent and coherent agent behaviors.
Fast Memory Mechanism: The caching function of the memory bank greatly accelerates the efficiency of basic operations.
Multi-user Simulation: Simulates complex social interactions among multiple users, providing a rich environment for research.
Make sure you have Python >= 3.10
Clone the repository:
git clone <repository-url>
Navigate to the cloned directory and install the required dependencies:
pip install -r requirements.txt
Using conda to install faiss:
conda install faiss-cpu -c pytorch
Set your OpenAI API key and adjust other parameters in config/config.py.
To run a Command Line Interface (CLI) based simulation, use the following command:
python -u simulator.py --config_file config/config.yaml --output_file messages.json --log_file simulation.log
This will start the simulation using the configuration specified in config/config.yaml, output the messages to messages.json, and log the simulation details to simulation.log.
For a Website-based demonstration, execute the following:
python run_demo.py
This command will start a local web server that serves the simulation environment. You can interact with the simulation through your web browser.
The LMAgent system is built on YuLan-Rec, many thanks to RUC-GSAI for the open-source resource.
We welcome contributions from the community. If you'd like to contribute, please fork the repository and use a feature branch. Pull requests are warmly welcome.
LMAgent is licensed under the MIT License. Feel free to use and adapt it for your own projects.
More details will be released soon.
@article{liu2024lmagent,
title={LMAgent: A Large-scale Multimodal Agents Society for Multi-user Simulation},
author={Liu, Yijun and Liu, Wu and Gu, Xiaoyan and Rui, Yong and He, Xiaodong and Zhang, Yongdong},
journal={arXiv preprint arXiv:2412.09237},
year={2024}
}
LMAgent is an innovative platform designed to create a large-scale and multimodal agent society capable of simulating complex multi-user behaviors. Agents in the LMAgent society can engage in a variety of activities, including chatting, browsing, purchasing, reviewing products, and even conducting live e-commerce streams. This environment can serves as a testbed for research in areas such as urban planning, social dynamics, and e-commerce.
Multimodal Interaction: Agents in the LMAgent society can interact using various modalities, including text, images, and potentially audio and video.
Scalability: Designed to efficiently simulate over 10,000 agents, enabling large-scale experiments.
Self-consistency Prompting: A novel mechanism that ensures consistent and coherent agent behaviors.
Fast Memory Mechanism: The caching function of the memory bank greatly accelerates the efficiency of basic operations.
Multi-user Simulation: Simulates complex social interactions among multiple users, providing a rich environment for research.
Make sure you have Python >= 3.10
Clone the repository:
git clone <repository-url>
Navigate to the cloned directory and install the required dependencies:
pip install -r requirements.txt
Using conda to install faiss:
conda install faiss-cpu -c pytorch
Set your OpenAI API key and adjust other parameters in config/config.py.
To run a Command Line Interface (CLI) based simulation, use the following command:
python -u simulator.py --config_file config/config.yaml --output_file messages.json --log_file simulation.log
This will start the simulation using the configuration specified in config/config.yaml, output the messages to messages.json, and log the simulation details to simulation.log.
For a Website-based demonstration, execute the following:
python run_demo.py
This command will start a local web server that serves the simulation environment. You can interact with the simulation through your web browser.
The LMAgent system is built on YuLan-Rec, many thanks to RUC-GSAI for the open-source resource.
We welcome contributions from the community. If you'd like to contribute, please fork the repository and use a feature branch. Pull requests are warmly welcome.
LMAgent is licensed under the MIT License. Feel free to use and adapt it for your own projects.
More details will be released soon.
@article{liu2024lmagent,
title={LMAgent: A Large-scale Multimodal Agents Society for Multi-user Simulation},
author={Liu, Yijun and Liu, Wu and Gu, Xiaoyan and Rui, Yong and He, Xiaodong and Zhang, Yongdong},
journal={arXiv preprint arXiv:2412.09237},
year={2024}
}