About

Nutpie is part of the PyMC organization. The PyMC organization develops and maintains tools for Bayesian statistical modeling and probabilistic machine learning.

Nutpie provides a high-performance implementation of the No-U-Turn Sampler (NUTS) that can be used with models defined in PyMC, Stan and other frameworks. It was created to enable faster and more efficient Bayesian inference while maintaining compatibility with existing probabilistic programming tools.

For more information about the PyMC organization, visit the following links:

• PyMC Website
• PyMC GitHub Organization

Paper

The algorithms behind nutpie’s mass matrix adaptation are described in the following paper:

Adrian Seyboldt, Eliot L. Carlson, Bob Carpenter (2026). Preconditioning Hamiltonian Monte Carlo by minimizing Fisher Divergence. arXiv:2603.18845 [stat.CO]. https://arxiv.org/abs/2603.18845

Citation

If you use nutpie in your research, please cite the following paper:

@article{seyboldt2026preconditioning,
  title   = {Preconditioning {Hamiltonian Monte Carlo} by minimizing
             {Fisher} Divergence},
  author  = {Adrian Seyboldt and Eliot L. Carlson and Bob Carpenter},
  year    = {2026},
  eprint  = {2603.18845},
  archivePrefix = {arXiv},
  primaryClass  = {stat.CO},
  url     = {https://arxiv.org/abs/2603.18845}
}