Data module (AtomicData, Batch, readers/writers)

Core classes

AtomicData	Graph-structured container for a single atomic system.
Batch	Graph-aware batch built on MultiLevelStorage.

I/O and pipelines

AtomicDataZarrWriter	Writer for serializing AtomicData into Zarr stores.
AtomicDataZarrReader	Reader for loading AtomicData from Zarr stores.
Dataset	AtomicData-native, map-style dataset that bypasses TensorDict conversion.
InMemoryDataset	Resident dataset that holds the whole dataset in memory as one Batch.
DataLoader	Batch-iterating data loader that yields Batch.
Reader	Abstract base class for data readers.

Dataset composition and sampling

MultiDataset	Compose multiple batch-loadable datasets behind one global index space.
MultiDatasetSampler	Sample global indices from a MultiDataset at dataset-level rates.
MultiDatasetBatchSampler	Sample full global-index batches from a MultiDataset.

Write configuration

ZarrArrayConfig	Per-array storage settings for compression, chunking, and sharding.
ZarrWriteConfig	Top-level storage plan handed to AtomicDataZarrWriter.