tensogram 0.15.0

Fast binary N-tensor message format for scientific data

Tensogram Python Bindings

Python bindings for Tensogram N-tensor message format.

Native extension with PyO3 and maturin. Supports NumPy arrays, async I/O, and GIL-free operation on Python 3.13t.

Installation

pip install tensogram

Usage

import numpy as np
import tensogram

data = np.random.randn(100, 200).astype(np.float32)
msg = tensogram.encode(
    {"version": 2},
    [({"type": "ntensor", "shape": [100, 200], "dtype": "float32",
       "compression": "szip"}, data)],
)
result = tensogram.decode(msg)
arr = result.objects[0][1]

Features

NumPy integration, async I/O, GIL-free parallel ops, partial decode, full codec support (szip, zstd, lz4, blosc2, zfp, sz3).

Documentation

• Full docs: https://sites.ecmwf.int/docs/tensogram/main/
• Repository: https://github.com/ecmwf/tensogram

License

Copyright 2026- ECMWF. Licensed under Apache-2.0. See LICENSE.