Machine Learning framework for encrypted data using Fully Homomorphic Encryption.

Torus ML enables training and inference on encrypted data without decryption. Part of the Lux FHE ecosystem.

• Privacy-preserving ML: Train models on encrypted data
• Quantization-aware training: Optimized for FHE computation
• scikit-learn compatible: Familiar API for ML practitioners
• GPU acceleration: Leverages luxfi/fhe-gpu

pip install torus-ml

Or from source:

git clone https://github.com/luxfi/torus-ml
cd torus-ml
pip install -e .

from torus.ml import FHEModelClient, FHEModelServer
from sklearn.datasets import make_classification

# Train a model
X, y = make_classification(n_samples=1000)
model = FHEModelClient()
model.fit(X, y)

# Compile for FHE
model.compile(X)

# Encrypted inference
encrypted_X = model.encrypt(X[:1])
encrypted_pred = model.predict(encrypted_X)
pred = model.decrypt(encrypted_pred)

• luxfi/fhe - Core FHE library (Go)
• luxfi/torus - Python FHE framework
• luxfi/fhe-compiler - LLVM compiler
• luxfi/lattice - Lattice primitives

BSD 3-Clause Clear License

• Lux Network
• Documentation