TorchWM Documentation#
TorchWM is a modular PyTorch library for world models, latent-dynamics planning, and representation learning. Train Dreamer, JEPA, IRIS, DiT, Genie, and DIAMOND agents with a unified API.
import torchwm
# Runs on ``pip install torchwm[gym]``.
agent = torchwm.create_model(
"dreamer", env="Pendulum-v1", env_backend="gym", total_steps=5_000
)
agent.train()
Get Started
User Guides
- Public API Quick Reference
- Using Operators for Inference
- Training Guide
- Inference Guide
- Evaluation Guide
- Memory & Replay Buffers
- Environments Guide
- Environment Backends
- NuPlan Dataset
- TorchWM CLI
- Package Overview
- Controllers and Policies
- World Models Study Guide
- Tutorial: Plug TorchWM world models into RL libraries
- Modular RSSM
- Vision Components
- Datasets
- Loss Functions
- Plugin Registry
- World Models Deep Dive (Ha & Schmidhuber, 2018)
Algorithms
Development