JAX Framework for Parameterized Stochastic Circuits (PSCs) and Directed Factor Graphs (DFGs) of Kernels
-
Updated
Aug 5, 2026 - Python
JAX Framework for Parameterized Stochastic Circuits (PSCs) and Directed Factor Graphs (DFGs) of Kernels
Software platform for Thermodynamic Sampling Units - probabilistic computing emulator
This repo is an educational POC showing how thermodynamic sampling units can produce sparse polynomial approximations. It’s intended for demonstration, not production.
Experimental validation workspace for Extropic THRML: thermodynamic computing with JAX-accelerated block Gibbs sampling
Architecture "full stack" post-Von Neumann basée sur la géométrie de Clifford, les réseaux d'interaction, le calcul thermodynamique et l'instantiation directe sans parsing. Une refonte intégrale de la pile informatique pour l'ère de l'efficacité énergétique.
Thermodynamic Hypergraphical Model Library in JAX - Part of Genesis Conductor by Igor Holt
ThermoCompLab — Interactive Thermodynamic Computing & Stochastic AI Laboratory. 6 modules: SPU simulator, Boltzmann machine, Langevin dynamics, thermodynamic diffusion, energy arena, fluctuation theorems. Zero dependencies.
Wave-function collapse as annealed Gibbs sampling on Extropic's THRML: real Gumin tilesets as Potts energies, an annealing hardware requirement, a first-order 'component ocean' phase transition, and measured GPU vs projected TSU energy
S-unit/p-unit thermodynamic AI hardware experiments for LLM inference and agentic EDA planning
Thermodynamic Attention Sampling Bridge — No-retrain inference bridge between frozen transformers and stochastic-compute hardware
Replicating Extropic's Fashion-MNIST Denoising Thermodynamic Model and measuring the K_mix=250 constant behind its energy claim: FID vs Gibbs sweeps, measured GPU joules vs projected TSU energy, same-protocol DDPM baseline
Extropic is a hardware company building thermodynamic (probabilistic) computing systems designed to run generative AI workloads at dramatically lower energy cost than GPUs. Its core hardware is the Thermodynamic Sampling Unit (TSU) — an inherently probabilistic chip whose physics natively performs the sampling that probabilistic AI algorithms…
Investigating a post-silicon future. A hybrid-manufactured 3D topological architecture for sustainable, room-temperature spintronics and thermodynamic inference.
Add a description, image, and links to the thermodynamic-computing topic page so that developers can more easily learn about it.
To associate your repository with the thermodynamic-computing topic, visit your repo's landing page and select "manage topics."