pyro

Why we included this project

Pyro puts Bayesian modeling directly inside PyTorch, so you can write generative models in ordinary Python and let it handle the variational inference or MCMC underneath. The core stays small: a few composable abstractions that automate inference for standard cases while still leaving specialists room to customize the internals when a problem demands it. It grew out of Uber AI and is maintained with a dedicated team at the Broad Institute, which gives it a track record in real deployments rather than just research demos. If your work already lives in PyTorch and you need uncertainty estimates, Bayesian layers, or time-to-event and experimental-design modeling, Pyro is a solid, mature place to start.