CoT2ToT: Chain-of-Thought to Tree-of-Thoughts

CoT2ToT is a Python package that processes and visualizes chain-of-thought (CoT) outputs from language models by converting them into structured graphs (trees). This tool helps you analyze and explore the reasoning steps generated by an LLM.

Features

• CoT Parsing: Convert raw LLM outputs into structured ChainOfThoughts objects.

  • LLM Extraction: Automatically parse CoT outputs using the language model.
  • Manual Extraction: Use predefined delimiters to segment and extract reasoning steps.

• Graph Construction: Transform a ChainOfThoughts object into a GraphOfThoughts directed graph, representing individual thoughts as nodes and their relationships as edges.

• Visualization: Visualize graphs using:

  • Static Tree Plot: A hierarchical display of the reasoning graph.
  • Animated Tree: An animated visualization that reveals nodes and edges gradually, with optional reasoning text. Animations can be saved as GIFs.

• Command-Line Interface (CLI): A CLI tool (quick_start.py) that allows you to run the different steps—parsing, graph conversion, and visualization—from the command line.

Installation and Usage

pip install cot2tot

You can create a CoT2ToT instance by using your own LLM enpoint and key which will be used with the OpenAI python library. Once a graph is created, you can plot it or animate it.

from cot2tot import CoT2ToT, CoT2ToTConfig

config = CoT2ToTConfig(
    llm_endpoint="<YOUR ENDPOINT>",
    llm_key="<YOUR KEY>",
    llm_model="<CHOSEN LLM MODEL>"
)

cot2tot_instance = CoT2ToT(config)

example_reasoning = "<|begin_of_thought|>[....]<|end_of_solution|>"

cot2tot_instance.run_pipeline(example_reasoning, verbose=True, plot=False)

cot2tot_instance.plot()

cot2tot_instance.animate(save_file_name="example_video.gif")

Repository Structure

• cot2tot/ The core package directory containing:

  • models.py: Data models (e.g., Thought, ChainOfThoughts, GraphOfThoughts).
  • parser.py: Functions to parse LLM outputs into CoT objects and convert them into graphs.
  • utils.py: Utility functions for text processing and tree layout computations.
  • visualize.py: Functions for static and animated visualization of graphs.

• tests/ Unit tests for the package:

  • test_models.py
  • test_parser.py
  • test_utils.py
  • test_visualize.py
  • test/fixtures/ Sample data files used in tests and demonstrations.

• quick_start.py A command-line interface tool for running parsing, graph conversion, and visualization tasks.

• pyproject.toml Poetry configuration file that manages dependencies and package metadata.

How to contribute

CoT2ToT uses Poetry for dependency management and packaging.

git clone https://github.com/your_username/cot2tot.git
cd cot2tot
poetry install

CLI

The package includes an CLI tool (cli_quick_start.py) for running different processing steps.

1. Parsing an LLM Output

Convert an LLM output file into a ChainOfThoughts object. You can choose between LLM extraction (default) or manual extraction.

poetry run python quick_start.py parse --input path/to/llm_output.json --output parsed_cot.json

For manual extraction:

poetry run python quick_start.py parse --input path/to/llm_output.json --method manual --output parsed_cot.json

2. Converting a CoT to a Graph

Convert a parsed ChainOfThoughts JSON file into a GraphOfThoughts.

poetry run python quick_start.py graph --input parsed_cot.json --output graph.json

3. Visualizing a Graph

Visualize a graph from a JSON file either as a static plot or an animated visualization.

Static tree plot:

poetry run python quick_start.py visualize --input graph.json

Animated visualization (with reasoning text and GIF saving):

poetry run python quick_start.py visualize --input graph.json --animate --show_reasoning --speed 0.3 --save tree_animation.gif

Running Tests

To run the tests with Poetry:

poetry run pytest

Contributing

Contributions are welcome! Please open issues or submit pull requests for improvements and bug fixes.

License

This project is licensed under the Apache-2.0 License. See the LICENSE file for details.