API Reference

The released predictions are read with the Hugging Face datasets library or plain JSON, see Data Format. The Python API covers producing them.

Module	Content
retexo.export	raw predictions to the released format
retexo.paths	where retexo looks for data, caches and runs

The rest of the package is organised by what a piece does:

Package	Content
retexo.core	edit operations, the builder that makes a variant from a source, and the scribe that replays and verifies a script
retexo.datasets	the synthetic-pair generator and the records it writes
retexo.aligners	scoring and decoding of source-word choices
retexo.baselines	the systems compared in the paper, the decoder, and the scorer
retexo.unified	the stretch-and-link pointer
retexo.edit_typing	the symbolic typer that names a link from the two words
retexo.resources	Latin morphology, WordNet, named entities and word vectors
retexo.pretraining	continued pretraining of the encoder
retexo.llm	the language-model baselines