Sparse data, clearly specified

Binsparse is an open, portable binary storage format for sparse matrices and tensors. It lets tools exchange structured sparse data without inventing another file format.

Read version 0.1.0

View the specification on GitHub

Portable

Binsparse provides a shared on-disk representation designed for interchange across languages and tools.

Versioned

Immutable specification releases are available alongside the latest working draft.

Open

The specification and its development process are public and community-driven.

Specification

Parsers

Here is a table listing the current tensor frameworks that support the format:

Framework (Parser) Language Status
Scipy (binsparse-reference-python) Python Compliant
PyData/sparse (binsparse-reference-python) Python Compliant
PyTorch (binsparse-reference-python) Python Compliant
TACO (binsparse-taco-parser) C/C++ Compliant
Finch.jl Julia Compliant
Matlab (binsparse-reference-c) Matlab Functional