Mixture of Agents (MoA) Chatbot

This project implements the concept of a Mixture of Agents (MoA), where several preliminary agents generate basic responses to a given prompt, and one final agent combines these responses to generate the final answer. The project also includes a console chat with the bot, which maintains the dialogue history. UPD: Added the ability to input the llm response language. The selected language will be used as part of the hint when generating the response.
Added the ability to navigate through the query history using navigation arrows.

Features

• Model Listing: Retrieve and display a list of available models from ollama.
• Multi-Agent Response Generation: Generate responses from multiple agents and combine them into a final unified response.
• Interactive CLI: Command-line interface for user interaction, including prompts and model selection.

Requirements

• Python 3.7+
• ollama library
• typer library
• InquirerPy library

Installation

1. Clone the repository:

   git clone https://github.com/MX-Goliath/MoA-Ollama.git
   cd MoA-Ollama

2. Create and activate a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows, use `venv\Scripts\activate`

3. Install the required packages:

   pip install ollama typer InquirerPy

Update

• Since ollama has added support for parallel query execution, preliminary responses are now processed in parallel, which significantly increases the speed of generating the first token of the final response

Usage

1. Run the application:

   python MoA_Ollama.py

2. Follow the prompts:

   • Select a model from the list of available models.
   • Enter your prompts when asked. Type 'exit' to quit the application.

3. Interaction Flow:

   • The application will generate responses using multiple agents.
   • A final unified response will be generated based on the preliminary responses.
   • The conversation history will be maintained throughout the session.