{"repo":"cipherflow-fhe/latti-ai","free":true,"listed":false,"github":"https://github.com/cipherflow-fhe/latti-ai","clone":"git clone https://github.com/cipherflow-fhe/latti-ai.git","description":"A framework for performing AI model inference on encrypted data.","language":"C++","stars":27,"topics":["ai-inference","cryptography","deep-learning","fhe","fully-homomorphic-encryption","homomorphic-encryption","privacy","torch"],"license":"Apache-2.0","category":"machine-learning","readme_excerpt":"LattiAI LattiAI is a development platform for privacy-preserving AI model inference, built on top of the LattiSense Fully Homomorphic Encryption (FHE) framework developed by CipherFlow. LattiAI covers the complete pipeline from plaintext models trained with frameworks like PyTorch to encrypted inference deployment. Through model adaptation, model compilation, and a high-performance HE operator library, it automatically converts standard AI models into encrypted inference services based on the CKKS fully homomorphic encryption scheme. Throughout the entire inference process, data remains encrypted — the server cannot access the user's raw data, and the user cannot access the model parameters, achieving bidirectional privacy protection for both data and model. AI developers can complete end-to-end encrypted inference deployment without understanding the underlying cryptographic primitives. Main Features - Model Adaptation : Provides plug-and-play polynomial approximation operators that replace non-polynomial activation functions (e.g., ReLU, SiLU) and MaxPooling with FHE-friendly polynomial activations and AvgPool. After fine-tuning or retraining, the adapted model achieves accuracy on par with the original. Validated on ResNet-18, ResNet-44, MobileNetV2, YOLOv5 and more, with additional conversion strategies under active development. - Model Compiler : Takes adapted model files ( .pth , .onnx ) and, through operator mapping and computation graph optimization, automatically gen","default_branch":null,"files":null,"tree":[],"storefront":"/r/cipherflow-fhe","claimed":false,"request_supported":{"post":"https://gitbuyer.com/r/cipherflow-fhe/latti-ai/request-supported","requests":0},"note":"indexed from public GitHub; nothing is for sale on this page. Clone it from GitHub. Paid listings live at /search."}