Torus ML

Machine Learning framework for encrypted data using Fully Homomorphic Encryption.

Overview

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

Features

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: native CUDA / Metal / WebGPU via the Torus compiler over the luxfi/gpu runtime

Installation

pip install torus-ml

Or from source:

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

Quick Start

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)

Integration

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

License

BSD 3-Clause Clear License

1.6 KiB Raw Permalink Blame History