Reference#

Lookup tables, syntax references, and generated API docs for RFD4-Proteina. Reach for this section once you already know roughly what you’re doing and need to look something up; for task-oriented walkthroughs, start at Workflow instead.

Glossary Conditioning and inference terminology used throughout the docs. Annotations Every condition annotation the model reads, its CIF column, values, and training definition. See the Conditioning concept page for why they behave the way they do. Checkpoints The model-weight checkpoints RFD4-Proteina loads, and how to override them. CLI options Every CLI option for rfd4-generate, rfd4-evaluate, rfd4-train, and rfd4-train-ae, with types, defaults, and examples. YAML/JSON condition spec Every field and condition annotation accepted in a YAML/JSON condition-spec input file. Metrics Every metric, oracle, pass criterion, and pre-oracle filter that ships, and the tag that turns each one on. Search parameters Every parameter accepted under the search block for reward guided (beam) search, including per-checkpoint fields, reward terms, and threshold criteria. Output format How generated designs and refolds are addressed, tagged, and annotated. Troubleshooting An index of every FAQ and troubleshooting table across the docs, linking to each one in place.

See API reference below for the reusable Python components behind the Hydra and CLI workflows.

API reference#

These pages are generated from the package’s Google style docstrings. You configure RFD4-Proteina mostly through Hydra and its command line entry points, so what follows documents the reusable Python components sitting behind those workflows, grouped by pipeline stage rather than alphabetically.

Data and conditioning#

Parsing structures into tokens, annotating and cropping them, and resolving conditions onto a batch.

Generation, search, and evaluation#

Composing inputs, running reward-guided search, scoring trajectories, and aggregating metrics once samples are written.