pyro-ppl/numpyro

Health v2 · maintenance only

Adoption not part of the score

2746 stars · 301 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

NumPyro is a lightweight probabilistic programming library providing a NumPy backend for Pyro, built on JAX for automatic differentiation and JIT compilation to GPU/TPU/CPU. It focuses on fast MCMC inference algorithms such as Hamiltonian Monte Carlo and the No U-Turn Sampler.

Use cases

• fit bayesian models with mcmc in python
• run hamiltonian monte carlo fast on gpu
• port pyro models to a numpy/jax backend
• sample posterior distributions with NUTS
• do bayesian inference with jit-compiled kernels
• handle discrete latent variables in mcmc with MixedHMC
• scale hmc to large datasets with HMCECS

When to choose

• you need fast, JIT-compiled MCMC/HMC inference in Python
• you want Pyro-style primitives with a NumPy/JAX backend
• you need GPU or TPU acceleration for Bayesian modeling
• you want a lightweight, hackable probabilistic programming substrate

When to avoid

• you need PyTorch-based Pyro features or the full Pyro ecosystem
• you require variational inference or features not yet implemented in NumPyro
• you need a stable, frozen API - the project is under active development

Facets

library · maturity active

machine-learning math sdk machine-learning data-science python cross-platform probabilistic-programming bayesian-inference mcmc jax hmc nuts pyro algorithms gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="pyro-ppl/numpyro")