OpenFHE-NumPy is a NumPy-like API for homomorphic encryption operations, built on top of OpenFHE. This library enables data scientists and machine learning practitioners to perform computations on encrypted data using familiar NumPy syntax.
OpenFHE-NumPy is organized as a hybrid C++/Python project with the following structure:
openfhe-numpy/
├── config/ # Configuration files
├── docs/ # Documentation
├── examples/ # Python examples
├── openfhe_numpy/ # Main Python package
│ ├── cpp/ # C++ implementation
│ ├── operations/ # Matrix operations
│ ├── tensor/ # Tensor implementations
│ └── utils/ # Utility functions
├── tests/ # Test suite
└── CMakeLists.txt # Build configuration
- C++ compiler: Supporting C++17 standard
- CMake: Version 3.16 or newer
- Python: Version 3.8 or newer
- NumPy: Any version
- OpenFHE: Any version
- OpenFHE Python: Any version
Before building, make sure you have the following dependencies installed:
- OpenFHE 1.4.0+ by following the instructions in OpenFHE Documentation
- OpenFHE Python Bindings by following the instructions in OpenFHE Python Documentation
We recommend following OpenFHE C++ and OpenFHE Python installation instructions first (which covers Linux, Windows and MacOS) and then getting back to this repo. If the some package cannot be found when running a Python example (occurs only for some environments), check the PYTHONPATH (OpenFHE Python) environment variable and the LD_LIBRARY_PATH (OpenFHE libraries). This ensures that the packages can be correctly located and imported.
# Clone the repository
git clone https://github.com/openfheorg/openfhe-numpy.git
cd openfhe-numpy
# Create build directory
mkdir build && cd build
# Configure with CMake
cmake ..
# Build the package
make
# Install
sudo make installOn Ubuntu, openfhe_numpy can be installed using pip. All available releases are listed at Python Package Index OpenFHE-Numpy Release History. Find the release for your version of Ubuntu and run
pip install openfhe_numpy==<openfhe_package_version>
Once installed, any python example at https://github.com/openfheorg/openfhe-numpy/tree/main/examples/python can be executed.
Note that Ubuntu LTS 20.04, 22.04, and 24.04 are currently supported. pip uninstall can be used to uninstall the openfhe package.
Run tests with unittest. See the testing readme for detailed instuctions.
import numpy as np
import openfhe_numpy as onp
from openfhe import *
# Initialize CKKS context
params = CCParamsCKKSRNS()
params.SetMultiplicativeDepth(7)
params.SetScalingModSize(59)
params.SetFirstModSize(60)
params.SetScalingTechnique(FIXEDAUTO)
params.SetSecretKeyDist(UNIFORM_TERNARY)
cc = GenCryptoContext(params)
cc.Enable(PKESchemeFeature.PKE)
cc.Enable(PKESchemeFeature.LEVELEDSHE)
cc.Enable(PKESchemeFeature.ADVANCEDSHE)
# Generate keys
keys = cc.KeyGen()
cc.EvalMultKeyGen(keys.secretKey)
cc.EvalSumKeyGen(keys.secretKey)
# Create matrix and encrypt it
A = np.array([[1, 2], [3, 4]])
ring_dim = cc.GetRingDimension()
total_slots = ring_dim // 2
# Encrypt with OpenFHE-NumPy
tensor_A = onp.array(
cc=cc,
data=A,
batch_size=batch_size,
order=onp.ROW_MAJOR,
fhe_type="C",
mode="zero",
public_key=keys.publicKey,
)
# Generate keys
onp.EvalSquareMatMultRotateKeyGen(keys.secretKey, tensor_A.ncols)
# Perform encrypted operations
tensor_product = tensor_A @ tensor_A # Matrix multiplication
tensor_sum = onp.add(tensor_A, tensor_A) # Element-wise addition
# Decrypt results
decrypted_product = tensor_product.decrypt(keys.secretKey, unpack_type="original")
decrypted_sum = tensor_sum.decrypt(keys.secretKey, unpack_type="original")
print("Result of A @ A:")
print(decrypted_product)
print("Result of A + A:")
print(decrypted_sum)OpenFHE-NumPy currently supports the following operations:
| Operation | Description | Example |
|---|---|---|
add |
Element-wise addition | onp.add(a, b) or a + b |
subtract |
Element-wise subtraction | onp.subtract(a, b) or a - b |
multiply |
Element-wise multiplication | onp.multiply(a, b) or a * b |
matmul |
Matrix multiplication | onp.matmul(a, b) or a @ b |
transpose |
Matrix transposition | onp.transpose(a) |
cumulative_sum |
Cumulative sum along axis | onp.cumulative_sum(a, axis) |
power |
Element-wise power | onp.power(a, exp) |
dot |
Dot product | onp.dot(a, b) |
sum |
Sum along axis | onp.sum(a, axis) |
In the current version, the OpenFHE-NumPy package supports operations on single-ciphertext vectors/matrices, where each encrypted array variable (which has type CTArray or PTArray) contains only a single encoding vector.
For example, we can consider a matrix:
1 2 3
4 5 6
7 8 9
As an encoding vector of the form: 1 2 3 0 4 5 6 0 7 8 9 0
The size of the encoded vector must be smaller than the number of available plaintext slots. Certain operations, such as matrix–vector multiplication, may require the ciphertext vector to be duplicated. In such cases, users should ensure that a sufficient number of slots are available for the function to execute correctly. We plan to release a future version with support for block ciphertexts, which will remove this limitation in the future.
For detailed documentation on the API, please visit our documentation site.
We provide several examples showcasing the library's functionality:
- 1D Convolution
- Matrix Accumulation
- Matrix Addition
- Matrix Summation
- Matrix Transpostion
- Matrix-Vector Multiplication
- Square Matrix Multiplication
- Vector Operations
OpenFHE Development - Contributing Guide
OpenFHE-NumPy is licensed under the BSD 2-Clause License. See the LICENSE file for details.
OpenFHE-NumPy is an independent project and is not officially affiliated with NumPy.
