zk-dtypes 0.0.8

zk_dtypes is a stand-alone implementation of several NumPy dtype extensions used in Zero Knowledge.

zk_dtypes

zk_dtypes is a stand-alone implementation of several NumPy dtype extensions used in Zero Knowledge libraries inspired by ml_dtypes, including:

• Narrow integer:

  • int2
  • int4
  • uint2
  • uint4

• Prime Field:

  • babybear
  • babybear_mont
  • goldilocks
  • goldilocks_mont
  • koalabear
  • koalabear_mont
  • mersenne31

• Extension Field:

  • babybearx4
  • babybearx4_mont
  • goldilocksx3
  • goldilocksx3_mont
  • koalabearx4
  • koalabearx4_mont
  • mersenne31x2

• Binary field:

  • binary_field_t0
  • binary_field_t1
  • binary_field_t2
  • binary_field_t3
  • binary_field_t4
  • binary_field_t5
  • binary_field_t6
  • binary_field_t7

• Elliptic curve:

  • bn254_sf
  • bn254_sf_mont
  • bn254_g1_affine
  • bn254_g1_affine_mont
  • bn254_g1_jacobian
  • bn254_g1_jacobian_mont
  • bn254_g1_xyzz
  • bn254_g1_xyzz_mont
  • bn254_g2_affine
  • bn254_g2_affine_mont
  • bn254_g2_jacobian
  • bn254_g2_jacobian_mont
  • bn254_g2_xyzz
  • bn254_g2_xyzz_mont

Prerequisite

1. Follow the bazel installation guide.

Build instructions

1. Clone the zk_dtypes repo

   git clone https://github.com/fractalyze/zk_dtypes
   

2. Build zk_dtypes

   bazel build //...
   

3. Test zk_dtypes

   bazel test //...
   

Installation

The zk_dtypes package is tested with Python versions 3.11-3.13, and can be installed with the following command:

pip install zk_dtypes

To test your installation, you can run the following:

pip install absl-py pytest
pytest zk_dtypes/tests

Installation from source

To build and install the package from source, run:

pip install .

Installation from prebuilt binary

Use USE_BAZEL_OUTPUT=1 for a faster installation that uses pre-built Bazel artifacts. This is the recommended path for development.

• On Linux / macOS

  # Build the shared library (.so)
  bazel build //zk_dtypes:_zk_dtypes_ext.so
  
  # Install using the Bazel output
  USE_BAZEL_OUTPUT=1 pip install .
  

• On Windows

  # Build the Python extension (.pyd)
  bazel build //zk_dtypes:_zk_dtypes_ext.pyd
  
  # Install using the Bazel output (PowerShell syntax)
  $env:USE_BAZEL_OUTPUT=1; pip install .
  

Example Usage

>>> from zk_dtypes import babybear_mont
>>> import numpy as np
>>> a = np.array([-1, -3, 2**30, 7], dtype=babybear_mont)
>>> b = np.array([5, 2, 4, 10], dtype=babybear_mont)
>>> a + b
array([4, 2013265920, 1073741828, 17], dtype=babybear_mont)

Importing zk_dtypes also registers the data types with numpy, so that they may be referred to by their string name:

>>> np.dtype('babybear_mont')
dtype(babybear_mont)

See examples/zk_dtypes_examples.ipynb for more examples.

Benchmarks

Benchmarks are disabled by default (though CI verifies their validity). To execute a specific benchmark (e.g., field_mul_benchmark), run the following command manually:

bazel run -c opt //zk_dtypes:field_mul_benchmark

License

The zk_dtypes source code is a modified derivative of the ml_dtypes project and inherits the original Apache 2.0 License (see LICENSE). All subsequent modifications comply with and are released under the same license.

Pre-compiled Wheels Dependencies

Note that pre-compiled wheels utilize the following dependencies:

• The EIGEN project, licensed under the MPL 2.0 license (see LICENSE.eigen).
• Chromium, licensed under the Free-BSD 3-Clause license (see LICENSE.chromium).