Reproduce “ROSETTA: Efficient and Accurate Privacy-Preserving LLM Decoding via Hybrid CKKS/TFHE Evaluation” 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
Jiangrui Yu, Baosheng Zhang, Liang Kong, Lin Ding, Yi Chen, Ye Yu, Mingzhe Zhang, Meng Li
ROSETTA combines CKKS parallel arithmetic with TFHE lookup tables to assign nonlinear operations to different encryption schemes during private LLM decoding. Its more accurate lookup method and cost-aware selector account for scheme conversions and CKKS computation levels. In one-token decode measurements, it is 1.5–2.1× faster than CacheMir; average PPL degradation is +1.4% versus PEGASUS, not a universal PPL improvement over CacheMir.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.