LMAgents

Small demo for a bot group chat using LMStudio.

Features

• Group chat with multiple AI agents.
• Agents respond based on predefined personalities and prompts.
• Real-time chat interface using Flask and Socket.IO.
• Save agent responses to a CSV file.
• Extract embeddings from agent responses.

Getting Started

Prerequisites

• LM Studio
• Python 3.6+
• Flask
• Flask-SocketIO
• Flask-CORS
• OpenAI API

Installation

1. Clone the repository:

git clone https://github.com/vschetinger/LMAgents.git

2. Create a virtual environment:

python -m venv LMAgents

3. Activate the virtual environment:

   • On Windows:

   LMAgents\Scripts\activate

   • On macOS and Linux:

   source LMAgents/bin/activate

4. Install the required dependencies:

pip install -r requirements.txt

5. Install LM Studio: https://lmstudio.ai/

6. Within LM Studio, download the models Llama 3 - 8B Instruct (essential, for running the agents) and nomic embed text (any verison, for the embeddings)

Usage

1. Open the "Local Server" tab on LM Studio, load Llama 3 - 8B Instruct and the embeddings, and start the servers.

2. Start the Flask application:

python app.py

3. Open your web browser and navigate to http://127.0.0.1:5000.

4. Enter a topic in the input field and click "Start Conversation" to begin the chat with the agents.

Extracting embeddings

The extract_embeddings.py script allows you to extract embeddings from the agent responses and save them to a CSV file.

Run the script:

python extract_embeddings.py

The embeddings will be saved in agent_responses_embeddings.csv.

NOTE: the embeddings server must also have been started for this to work.

Configuration

The agents are configured in the agents_config.json file. Each agent has the following properties:

• name: The name of the agent.
• api_url: The API URL for the OpenAI server.
• api_key: The API key for the OpenAI server.
• model: The model identifier.
• temperature: The temperature setting for the model.
• starting_prompt: The initial prompt for the agent.

Example configuration:

 [ { 
    "name": "Goethe", 
    "api_url": "http://127.0.0.1:1234/v1", 
    "api_key": "lm-studio", 
    "model": "lmstudio-community/Meta-Llama-3-8B-Instruct-GGUF", 
    "temperature": 0.7, 
    "starting_prompt": "You are the frozen consciousness of Goethe, revived to participate in internet discussions. You will provide short, at max tweet-sized sassy answer"
     }, ... ]

Testing

You can test the LM Studio setup using the agent_test.py script:

python agent_test.py

Contributing

Contributions are welcome! If you find any issues or have suggestions for improvements, please open an issue or submit a pull request.

License

This project is licensed under the MIT License.