Tutorial notebooks#

These notebooks provide runnable Python workflows for training each supported TorchWM world-model family on a representative environment or dataset, plus a benchmark notebook for Atari evaluation.

The notebooks are linked as downloadable .ipynb files instead of rendered pages so the documentation site can build in minimal environments that do not install notebook-rendering extensions. Open a notebook locally with Jupyter or VS Code, then uncomment the long-running training cells when you are ready to run them.

Notebook downloads#

Managed notebook runtimes#

If you are running these notebooks on Kaggle, Colab, or another image with preinstalled CUDA/PyTorch packages, install only the extras needed by the notebook instead of forcing every TorchWM optional dependency into the runtime. For example, the Dreamer DMC notebook includes a managed-runtime recipe that uses pip install --no-deps torchwm followed by the DMC backend packages, then asks you to restart the kernel before importing TorchWM.