Pyro
Pyro is an open-source probabilistic programming library built on PyTorch, developed primarily by researchers at Uber AI Labs and now maintained by a community of contributors.
Profile
Pyro is an open-source library for probabilistic programming in Python, enabling users to build and infer complex statistical models using deep learning and Bayesian methods.
Pyro is an open-source probabilistic programming library built on PyTorch, developed primarily by researchers at Uber AI Labs and now maintained by a community of contributors. First released in 2017, Pyro enables flexible and scalable probabilistic modeling, including deep learning, Bayesian inference, and reinforcement learning. The library is hosted on GitHub under the pyro-ppl organization, with over 9,000 stars and 1,000 forks as of June 2026.
Pyro's development is led by a core team including Fritz Obermeyer, Eli Bingham, and others, with the most recent commit on June 5, 2026. The project is licensed under the Apache License 2.0. Pyro competes with other probabilistic programming frameworks like TensorFlow Probability, Stan, and PyMC.
Its key differentiator is tight integration with PyTorch, allowing users to combine neural networks with probabilistic reasoning. The library supports a range of inference algorithms, including variational inference, Markov chain Monte Carlo, and importance sampling. Pyro has been used in academic research and industry applications, including by companies like Uber, but specific customer names are not publicly disclosed.
The project has not disclosed any funding rounds or revenue, as it is an open-source project rather than a commercial entity. The most recent activity includes fixes for lint formatting and mypy checks, and updates to Docker configurations for PyTorch 2.7.0 compatibility.
Who buys this
- Academic researchers in machine learning, statistics, and artificial intelligence
- Data scientists and engineers at technology companies building probabilistic models
- Graduate students and educators teaching probabilistic programming and Bayesian inference
- Researchers in fields like computational biology, finance, and robotics applying Bayesian methods
Strengths and what to watch
Strengths
- Deep integration with PyTorch allows seamless combination of neural networks and probabilistic models
- Active open-source community with regular commits and over 2,500 commits as of June 2026
- Supports a wide range of inference algorithms including variational inference, MCMC, and importance sampling
Watch for
- As an open-source project, Pyro has no disclosed funding or revenue, making long-term maintenance dependent on volunteer contributions
- Competition from other probabilistic programming frameworks like TensorFlow Probability and PyMC may limit adoption
- The project's reliance on PyTorch means changes in PyTorch's API or direction could require significant adaptation
Recent moves
Key Information
- Industry
- Stat Languages
- Founded
- 2017
Frequently Asked Questions
What is Pyro probabilistic programming library?
Pyro is an open-source probabilistic programming library for Python, built on PyTorch. It allows users to build and infer complex statistical models using deep learning and Bayesian methods. Developed by researchers at Uber AI Labs, it supports flexible and scalable probabilistic modeling.
How does Pyro integrate with PyTorch?
Pyro is deeply integrated with PyTorch, enabling seamless combination of neural networks with probabilistic reasoning. This tight integration allows users to leverage PyTorch's deep learning capabilities within probabilistic models, making it a key differentiator from other frameworks like TensorFlow Probability or Stan.
What inference algorithms does Pyro support?
Pyro supports a range of inference algorithms including variational inference, Markov chain Monte Carlo (MCMC), and importance sampling. These algorithms allow users to perform Bayesian inference on complex models, making Pyro suitable for both academic research and industry applications.
Who maintains Pyro and is it actively developed?
Pyro is maintained by a community of contributors, originally developed by researchers at Uber AI Labs. The core team includes Fritz Obermeyer and Eli Bingham. As of June 2026, the project has over 2,500 commits and recent activity includes fixes for lint formatting and Docker updates.
What are the main use cases for Pyro?
Pyro is used by academic researchers in machine learning and statistics, data scientists building probabilistic models, and educators teaching Bayesian inference. It also finds applications in fields like computational biology, finance, and robotics, where combining neural networks with probabilistic reasoning is valuable.
How does Pyro compare to TensorFlow Probability or PyMC?
Pyro competes with TensorFlow Probability, Stan, and PyMC. Its key differentiator is tight integration with PyTorch, allowing users to combine neural networks with probabilistic reasoning. However, as an open-source project with no disclosed funding, long-term maintenance depends on volunteer contributions, and changes in PyTorch may require adaptation.
Sources
- github.com — GitHub repository details including stars, forks, recent commits, and license
- ir.pyrogenesis.com — Financial results for PyroGenesis Inc., a separate company; not relevant to Pyro PPL
- finance.yahoo.com — Yahoo Finance article on PyroGenesis Inc. financial results; not relevant to Pyro PPL
- ir.revvity.com — Revvity quarterly results; not relevant to Pyro PPL