extropic-ai/thrml

Health v2 · maintenance only

Adoption not part of the score

1144 stars · 141 forks observed · 2026-08-28

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

THRML is a JAX library for building and sampling probabilistic graphical models, focused on efficient block Gibbs sampling of energy-based models on sparse, heterogeneous graphs. It is developed by Extropic as a software counterpart to their probabilistic sampling hardware, providing GPU-accelerated sampling tools today.

Use cases

• sample from probabilistic graphical models with block Gibbs in JAX
• build and sample Ising models and discrete energy-based models on GPU
• prototype sampling algorithms for future Extropic probabilistic hardware
• run parallel Gibbs sweeps over graph-coloured blocks of a factor graph
• simulate large spin models for research on diffusion-like probabilistic models

When to choose

• you need fast, GPU-accelerated block Gibbs sampling of discrete PGMs or EBMs
• your models are sparse, heterogeneous factor graphs expressible in JAX
• you want to experiment with Extropic's probabilistic computing paradigm in software

When to avoid

• you need general-purpose probabilistic programming with inference beyond Gibbs sampling
• you work with continuous variables or non-JAX ML stacks like PyTorch or TensorFlow
• you need a mature, battle-tested MCMC library with broad model support

Facets

library · maturity active

machine-learning simulation gpu-computing machine-learning deep-learning gpu-computing python cross-platform jax probabilistic-graphical-models gibbs-sampling energy-based-models ising-model probabilistic-computing mcmc algorithms gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="extropic-ai/thrml")