---

Shaggy is a lightweight package that implements autoencoder models in PyTorch. It provides modular encoder–decoder architectures, the SOAP optimizer, gradient-checkpointing utilities, and save/load tools. Basically everything needed to go from raw data to a trained latent representation with minimal boilerplate.

---

C O N T R I B U T O R S

---

We build on the work of François Rozet, Gerome Andry, and Sacha Lewin as well as to the entire Science with AI Laboratory (SAIL) team. Thanks !

---

T U T O R I A L

---

A self-contained tutorial is available as a Jupyter notebook. It walks through dataset loading, model configuration, training with a live loss plot, and reconstruction visualization on CIFAR-10.

➜ notebook/demo.ipynb

---

I N S T A L L A T I O N

---

• If you want the latest version, install it directly from GitHub:

  pip install git+https://github.com/VikVador/shaggy
  

• If you want a local editable install with all optional dependencies (training, notebooks, linting):

  conda create -n shaggy python=3.11
  conda activate shaggy
  

  then

  pip install --editable '.[all]' --extra-index-url https://download.pytorch.org/whl/cu121
  

  Optionally, install the pre-commit hooks to automatically detect code issues before each commit:

  pre-commit install --config pre-commit.yml