zero

A zero-overhead library for PyTorch users.

Data and Tensors

Stream

Smart wrapper for data loaders and iterables.

collate	Almost an alias for torch.utils.data.dataloader.default_collate.
concat	Concatenate items (tensors, numpy-arrays, tuples, dicts etc.) along the first dimension.
iter_batches	Efficiently iterate over data (tensor, tuple of tensors, dict of tensors etc.) in a batchwise manner.

Utilities

ProgressTracker	Tracks the best score, helps with early stopping.
Timer	Measures time.

evaluation

Context-manager & decorator for models evaluation.

improve_reproducibility

Set seeds in random, numpy and torch and make some cuDNN operations deterministic.

Submodules

See the sidebar on the left.