Recommended Reading

We use a lot of external libraries in Undertale - if you’re not already pretty familiar with the following, it’s worth reading through their documentation and possibly completing their tutorial(s) before contributing.

datatrove

The dataset building pipeline library from the folks at HuggingFace. We use this to codify all of our dataset building pipelines and parallelize them across compute infrastructure.

PyTorch

The deep learning library. If you’re not already deeply familiar with PyTorch, the textbook Deep Learning with Pytorch is an excellent resource.

PyTorch Lightning

All of our models are written in PyTorch and wrapped in Lightning modules. We largely let Lightning handle the complexities of multi-node, multi-GPU training, validation, and integration with tensorboard for monitoring training.

Tensorboard

The visualization tool we use for tracking training runs.

Sphinx

A software documentation library. All of our documentation is written in reStructuredText and built with Sphinx. We also use autodoc with Google-style Python docstrings for automatically generated reference documentation.

pyinstrument

The statistical profiler for Python. We use this occasionally for performance testing.