Reproduce “SpliTEE: Fast and Private LLM Inference by Coupling GPU-Assisted Trusted Execution Environments with Differential Privacy” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shashie Dilhara Batan Arachchige, Robin Carpentier, Hassan Jameel Asghar, Dali Kaafar
SpliTEE keeps prompts and nonlinear computations inside a trusted execution environment while outsourcing matrix multiplications to a GPU, adding Gaussian noise and correcting it inside the TEE. With a Qwen model and an 800-token limit, it reduced average inference time from 82.78 seconds for full TDX to 35.29 seconds. Evidence covers two models, public prompts, and one reconstruction attack, not universal privacy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.