Atom of Thoughts (AOT) Reasoning

An implementation of the Atom of Thoughts (AOT) reasoning technique that helps language models break down complex questions into simpler components for more accurate answers.

Requirements

• Python 3.8+
• Ollama installed and running locally

Installation

# Clone the repository
git clone https://github.com/agi-dude/AtomOfThoughts
cd AtomOfThoughts

# Install dependencies
pip install -r requirements.txt

Usage

Chat

python main.py --model llama3

This starts an interactive terminal chat where you can ask questions and get responses using the AOT reasoning technique.

API Server

python main.py --mode api --port 8000 --model llama3

This starts an OpenAI-compatible API server on port 8000 that you can use with any OpenAI client library.

Command Line Arguments

• --model: Ollama model to use (default: llama3)
• --mode: Run mode - "terminal" or "api" (default: terminal)
• --port: Port for API server (default: 8000)

References

arXiv:2502.12018