xnn.common.data.hub.base

Dataset registry and the load_dataset entry point.

A DatasetBuilder knows how to download and preprocess one upstream dataset into xnn’ native structure dicts (the list-of-dicts format consumed by AtomicDataset). Builders self-register under a short name via register_dataset(); load_dataset() looks one up and drives it, giving a HuggingFace load_dataset()-style one-liner.

Functions

get_builder(name)	Look up a registered builder by name.
list_datasets()	Return the names of all registered datasets, sorted.
load_dataset(name, *[, split, cutoff, cache_dir])	Download and preprocess an upstream dataset in one call.
register_dataset(builder)	Register a builder instance under its name.

Classes

DatasetBuilder()

Base class for a downloadable, preprocessable dataset.