【Paper】

【English | Chinese】

AgentVerse is designed to facilitate the deployment of multiple LLM-based agents in various applications. AgentVerse primarily provides two frameworks: task-solving and simulation.

• Task-solving: This framework assembles multiple agents as an automatic multi-agent system (AgentVerse-Tasksolving, Multi-agent as system) to collaboratively accomplish the corresponding tasks. Applications: software development system, consulting system, etc.

• Simulation: This framework allows users to set up custom environments to observe behaviors among, or interact with, multiple agents. ⚠️⚠️⚠️ We're refactoring the code. If you require a stable version that exclusively supports simulation framework, you can use release-0.1 branch. Applications: game, social behavior research of LLM-based agents, etc.

• [2024/3/17] AgentVerse was introduced in NVIDIA's blog - Building Your First LLM Agent Application.

• [2024/1/17] We're super excited to announce that our paper got accepted at ICLR 2024. More updates will be coming soon!

• [2023/10/17] We're super excited to share our open-source AI community hugging face: AgentVerse. You are able to try out the two simulation applications, NLP Classroom and Prisoner's Dilemma,with your code of the openai API key and the openai organization. Have fun!

• [2023/10/5] Re-factor our codebase to enable the deployment of both simulation and task-solving framework! We have placed the code for Minecraft example in the paper at the minecraft branch. Our tool-using example will soon be updated to the main branch. Stay tuned!

• [2023/8/22] We're excited to share our paper AgentVerse: Facilitating Multi-Agent Collaboration and Exploring Emergent Behaviors in Agents that illustrate the task-solving framework in detail of AgentVerse.

• [2023/6/5] We are thrilled to present an array of demos, including NLP Classroom, Prisoner Dilemma, Software Design, Database Administrator, and a simple H5 Pokemon Game that enables the interaction with the characters in Pokemon! Try out these demos and have fun!

• [2023/5/1] 🚀 AgentVerse is officially launched!

• Code release of our paper
• Add support for local LLM (LLaMA, Vicunna, etc.)
• Add documentation
• Support more sophisticated memory for conversation history

• 📰 What's New
• 🗓 Coming Soon
• Contents
• 🚀 Getting Started
  • Installation
  • Environment Variables
  • Simulation
    • Framework Required Modules
    • CLI Example
    • GUI Example
  • Task-Solving
    • Framework Required Modules
    • CLI Example
  • Local Model Support
  • vLLM Support
  • FSChat Support
    • 1. Install the Additional Dependencies
    • 2. Launch the Local Server
    • 3. Modify the Config File
• AgentVerse Showcases
  • Simulation Showcases
  • Task-Solving Showcases
• 🌟 Join Us!
  • Leaders
  • Contributors
  • How Can You Contribute?
  • Social Media and Community
• Star History
  • Citation
• Contact

Manually Install (Recommended!)

Make sure you have Python >= 3.9

git clone https://github.com/OpenBMB/AgentVerse.git --depth 1
cd AgentVerse
pip install -e .

If you want to use AgentVerse with local models such as LLaMA, you need to additionally install some other dependencies:

pip install -r requirements_local.txt

Install with pip

Or you can install through pip

pip install -U agentverse

You need to export your OpenAI API key as follows：

# Export your OpenAI API key
export OPENAI_API_KEY="your_api_key_here"

If you want use Azure OpenAI services, please export your Azure OpenAI key and OpenAI API base as follows：

export AZURE_OPENAI_API_KEY="your_api_key_here"
export AZURE_OPENAI_API_BASE="your_api_base_here"

- agentverse 
  - agents
    - simulation_agent
  - environments
    - simulation_env

You can create a multi-agent environments provided by us. Using the classroom scenario as an example. In this scenario, there are nine agents, one playing the role of a professor and the other eight as students.

agentverse-simulation --task simulation/nlp_classroom_9players

We also provide a local website demo for this environment. You can launch it with

agentverse-simulation-gui --task simulation/nlp_classroom_9players

After successfully launching the local server, you can visit http://127.0.0.1:7860/ to view the classroom environment.

If you want to run the simulation cases with tools (e.g., simulation/nlp_classroom_3players_withtool), you need to install BMTools as follows:

git clone git+https://github.com/OpenBMB/BMTools.git
cd BMTools
pip install -r requirements.txt
python setup.py develop

This is optional. If you do not install BMTools, the simulation cases without tools can still run normally.

- agentverse 
  - agents
    - simulation_env
  - environments
    - tasksolving_env

To run the experiments with the task-solving environment proposed in our paper, you can use the following command:

To run AgentVerse on a benchmark dataset, you can try

# Run the Humaneval benchmark using gpt-3.5-turbo (config file `agentverse/tasks/tasksolving/humaneval/gpt-3.5/config.yaml`)
agentverse-benchmark --task tasksolving/humaneval/gpt-3.5 --dataset_path data/humaneval/test.jsonl --overwrite

To run AgentVerse on a specific problem, you can try

# Run a single query (config file `agentverse/tasks/tasksolving/brainstorming/gpt-3.5/config.yaml`). The task is specified in the config file.
agentverse-tasksolving --task tasksolving/brainstorming

To run the tool using cases presented in our paper, i.e., multi-agent using tools such as web browser, Jupyter notebook, bing search, etc., you can first build ToolsServer provided by XAgent. You can follow their instruction to build and run the ToolServer.

After building and launching the ToolServer, you can use the following command to run the task-solving cases with tools:

agentverse-tasksolving --task tasksolving/tool_using/24point

We have provided more tasks in agentverse/tasks/tasksolving/tool_using/ that show how multi-agent can use tools to solve problems.

Also, you can take a look at agentverse/tasks/tasksolving for more experiments we have done in our paper.

If you want to use vLLM, follow the guide here to install and setup the vLLM server which is used to handle larger inference workloads. Create the following environment variables to connect to the vLLM server:

export VLLM_API_KEY="your_api_key_here"
export VLLM_API_BASE="http://your_vllm_url_here"

Then modify the model in the task config file so that it matches the model name in the vLLM server. For example:

model_type: vllm
model: llama-2-7b-chat-hf

This section provides a step-by-step guide to integrate FSChat into AgentVerse. FSChat is a framework that supports local models such as LLaMA, Vicunna, etc. running on your local machine.

If you want to use local models such as LLaMA, you need to additionally install some other dependencies:

pip install -r requirements_local.txt

Then modify the MODEL_PATH and MODEL_NAME according to your need to launch the local server with the following command:

bash scripts/run_local_model_server.sh

The script will launch a service for Llama 7B chat model. The MODEL_NAME in AgentVerse currently supports several models including llama-2-7b-chat-hf, llama-2-13b-chat-hf, llama-2-70b-chat-hf, vicuna-7b-v1.5, and vicuna-13b-v1.5. If you wish to integrate additional models that are compatible with FastChat, you need to:

1. Add the new MODEL_NAME into the LOCAL_LLMS within agentverse/llms/__init__.py. Furthermore, establish
2. Add the mapping from the new MODEL_NAME to its corresponding Huggingface identifier in the LOCAL_LLMS_MAPPING within the agentverse/llms/__init__.py file.

In your config file, set the llm_type to local and model to the MODEL_NAME. For example

llm:
  llm_type: local
  model: llama-2-7b-chat-hf
  ...

You can refer to agentverse/tasks/tasksolving/commongen/llama-2-7b-chat-hf/config.yaml for a more detailed example.

Refer to simulation showcases

Refer to tasksolving showcases

AgentVerse is on a mission to revolutionize the multi-agent environment for large language models, and we're eagerly looking for passionate collaborators to join us on this exciting journey.

• Issue and Pull-Request: If you encounter any problems when use AgentVerse, you can propose the issue in English. Beisdes, you can also autonomously ask us to assign issue to you and send the PR (Please follow the PULL_REQUEST_TEMPLATE) after you solve it.

• Code Development: If you're an engineer, help us refine, optimize, and expand the current framework. We're always looking for talented developers to enhance our existing features and develop new modules.

• Documentation and Tutorials: If you have a knack for writing, help us improve our documentation, create tutorials, or write blog posts to make AgentVerse more accessible to the broader community.

• Application Exploration: If you're intrigued by multi-agent applications and are eager to experiment using AgentVerse, we'd be thrilled to support your journey and see what you create!

• Feedback and Suggestions: Use AgentVerse and provide us with feedback. Your insights can lead to potential improvements and ensure that our framework remains top-notch.

Also, if you're passionate about advancing the frontiers of multi-agent applications, become core AgentVerse team members, or are eager to dive deeper into agent research. Please reach out AgentVerse Team, and CC to Weize Chen and Yusheng Su. We're keen to welcome motivated individuals like you to our team!

• Twitter: https://twitter.com/Agentverse71134

• Discord: https://discord.gg/gDAXfjMw.

• Hugging Face: https://huggingface.co/spaces/AgentVerse/agentVerse.

If you find this repo helpful, feel free to cite us.

@article{chen2023agentverse,
  title={Agentverse: Facilitating multi-agent collaboration and exploring emergent behaviors in agents},
  author={Chen, Weize and Su, Yusheng and Zuo, Jingwei and Yang, Cheng and Yuan, Chenfei and Qian, Chen and Chan, Chi-Min and Qin, Yujia and Lu, Yaxi and Xie, Ruobing and others},
  journal={arXiv preprint arXiv:2308.10848},
  year={2023}
}

AgentVerse Team: [email protected]

Project leaders:

• Weize Chen: [email protected]

• Yusheng Su: [email protected]