Thought Miners

thoughtminers provide a set of receipes and convenient functions for the graphical and probabilistic representation and manipulation of reasoning spaces from large language models (LLMs). Current functionalities include visualisation of latent activation of reasoning traces and inference time scalling methods based on uncertainty minimisation.

Installation

The last official release can be downloaded from PIP:

pip install thoughtminers

The current version under development can be installed from the master branch of the GitHub folder:

pip install “git+https://github.com/centre-for-humanities-computing/thoughtminers.git”

Features

Visualising reasoning traces, densities and dynamics

See this tutorial for more details.

Acknowledgments

This package is developed and maintained by Nicolas Legrand with the support of the Centre for Humanities Computing.