World Models Python Packages

causal-worlds

Generate fictional-but-coherent causal operations worlds (executable sim + time-series + ground-truth causal answer-key) from a natural-language description — for benchmarking causal-discovery and control agents.

x-transformers-rl

Implementation of a transformer for reinforcement learning using `x-transformers`

improving-transformers-world-model

Implementation of the new SOTA for model based RL, from the paper "Improving Transformer World Models for Data-Efficient RL", in Pytorch

openwam

Official code for "OpenWAM: An Open Framework for Composable World-Action Models" (Stanford University)

gaia2-pytorch

Implementation of the world model architecture for self driving out of Wayve

openwam-sdk

Official code for "OpenWAM: An Open Framework for Composable World-Action Models" (Stanford University)

trio-retina

Model-agnostic state layer for world models — turn any detector (YOLO · VLM · DINO) into one standard, queryable stream of events + latent state. numpy-only, runs on CPU at the edge.

mstar-ai

A high-performance, universal serving framework for any-to-any models.

mstar-project

A high-performance, universal serving framework for any-to-any models.

genie2-pytorch

Implementation of a framework for Genie2 in Pytorch

diffav

Physics-informed, RL-aligned evaluation engine for autonomous driving: JAX/Flax NNX trajectory-diffusion world models on the Waymo Open Dataset, DPO and reward-guided adversarial steering toward the scenarios that break AV stacks, physics feasibility, occupancy flow, WOSAC-style realism metrics and sensor simulation.

worldgap

Reusable world-model-based domain-gap quantification — for perception pipelines and actuator/mechanism models — before any hardware is touched.

grilly

GPU-accelerated neural network operations using Vulkan compute shaders.

dreamerv2

Mastering Atari with Discrete World Models

worldbench

Checkpoint regression testing for robotics world models

dreamerv3

Mastering Diverse Domains through World Models

dreamer

Mastering Diverse Domains through World Models

ewm-core

Observable agent trajectories, auditable decisions, and Claude in the agent slot. Infrastructure for AI systems that improve through architecture, not retraining.

pixyzrl

A Bayesian RL Framework with Probabilistic Generative Models

flashdreams

high-performance inference and serving library for interactive autoregressive video and world models

geno-lewm

An action-conditioned JEPA world model for DNA, built on top of Carbon.

worldproof

A reality check for world models: evaluate rollout predictions against ground truth and physical invariants, and get a report card showing where and why the imagination breaks. Runs on a laptop

gemla

Γ–EML–α transport architecture SDK for lifted-phase diagnostics, action-based inference, and adversarial-control gates.

mmlm

Toward Multi Modality Language Model - implementation of GPT-4o/Project Astra