Getting Started

Repository Structure

BioNeMo Recipes is organized around three root directories:

• models/: TransformerEngine-backed model implementations and export utilities.
• recipes/: training, fine-tuning, inference, and convergence recipes.
• interpretability/: interpretability workflows such as sparse autoencoders.

Documentation source is stored in docs/.

Local Development

Start from the README in the model or recipe directory you are modifying:

cd recipes/evo2_megatron
bash .ci_build.sh
pytest -v .

Many recipes are self-contained packages. Shared helper code used by the megatron recipes lives inside each package as bionemo.common.

Next Steps

• For model implementations, start in models/.
• For runnable workflows, start in recipes/.
• For interpretability workflows, start in interpretability/.