desilofhe-cu128 1.0.0

The DESILO Fully Homomorphic Encryption (FHE) library

The DESILO FHE Library

Welcome to the DESILO FHE library!

The DESILO FHE library is a homomorphic encryption library that is fast and intuitive. It delivers state-of-the-art performance, maximum accuracy, and versatile usability. Designed to be accessible and efficient for everyone, from beginners unfamiliar with homomorphic encryption to experts developing advanced algorithms.

Built on C++ and CUDA, the DESILO FHE library is optimized for both CPUs and NVIDIA GPUs. It includes a Python wrapper for easy use and seamless integration into existing AI workflows. Additionally, it supports CPU parallelization for environments without GPU access.

Currently, the DESILO FHE library supports real and complex number operations based on the RNS-CKKS scheme, with plans to expand to various integer-based schemes in the future. It provides a powerful solution adaptable to diverse environments and requirements.

Documentation

Please refer to the official documentation for detailed installation instructions, examples, and documentation.

Quickstart

Installation

CPU-only version

pip install desilofhe

CUDA 12.6

pip install desilofhe-cu126

CUDA 12.8

pip install desilofhe-cu128

Encrypt / Decrypt

Encryption and decryption require a public key and a secret key, respectively. The public key is generated from the secret key. You can encrypt and decrypt data as shown below. Note that CKKS is a real-number-based scheme, so a very small error is introduced during encryption and decryption.

from desilofhe import Engine

engine = Engine()

secret_key = engine.create_secret_key()
public_key = engine.create_public_key(secret_key)

data = [1, 2, 3]
encrypted = engine.encrypt(data, public_key)
decrypted = engine.decrypt(encrypted, secret_key)

print(decrypted[:3]) # [1. 2. 3.]

Add / Subtract

To add or subtract ciphertext, no additional keys are required beyond the public and secret keys. You can encrypt data separately and then perform addition and subtraction as follows:

from desilofhe import Engine

engine = Engine()

secret_key = engine.create_secret_key()
public_key = engine.create_public_key(secret_key)

data1 = [1, 2, 3]
encrypted1 = engine.encrypt(data1, public_key)

data2 = [4, 5, 6]
encrypted2 = engine.encrypt(data2, public_key)

added = engine.add(encrypted1, encrypted2)
subtracted = engine.subtract(encrypted1, encrypted2)

decrypted1 = engine.decrypt(added, secret_key)
decrypted2 = engine.decrypt(subtracted, secret_key)

print(decrypted1[:3], decrypted2[:3]) # [5. 7. 9.] [-3. -3. -3.]

Multiply

To multiply ciphertexts, a special key called the relinearization key is required due to the nature of homomorphic encryption. You can encrypt data and perform multiplication as shown below. Note that ciphertexts in homomorphic encryption have a maximum multiplication depth, which will decrease after multiplication.

from desilofhe import Engine

engine = Engine()

secret_key = engine.create_secret_key()
public_key = engine.create_public_key(secret_key)
relinearization_key = engine.create_relinearization_key(secret_key)

data1 = [1, 2, 3]
encrypted1 = engine.encrypt(data1, public_key)

data2 = [4, 5, 6]
encrypted2 = engine.encrypt(data2, public_key)

multiplied = engine.multiply(encrypted1, encrypted2, relinearization_key)

decrypted = engine.decrypt(multiplied, secret_key)

print(decrypted[:3]) # [~4 ~10 ~18]
print(encrypted1.level, encrypted2.level, multiplied.level) # 7 7 6

Contact

If you have any questions, please contact us at [e-mail].