ROSETTA: Efficient and Accurate Privacy-Preserving LLM Decoding via Hybrid CKKS/TFHE Evaluation

arXiv:2609.16915 · cs.CR · Submitted 2026-09-15 · Read on arXiv

cs.CR

Submitted: 2026-09-15

Updated: 2026-09-15

Comments: 15 pages. Accepted at ACM CCS 2026

Code: https://github.com/tuneinsight/lattigo

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Transformer — 19× in this paper · explained in 2 episodes
• Trace — 16× in this paper · explained in 2 episodes
• Decode — 11× in this paper · explained in 2 episodes
• Accuracy — 7× in this paper · explained in 1 episode
• Benchmark — 6× in this paper · explained in 2 episodes
• Large Language Models (LLMs) — 3× in this paper · explained in 6 episodes
• Perplexity — 3× in this paper · explained in 3 episodes
• Fully Homomorphic Encryption (FHE) — 2× in this paper · explained in 5 episodes
• Generative AI — 2× in this paper · explained in 2 episodes
• Hybrid Approach — 2× in this paper · explained in 4 episodes
• KV cache — 2× in this paper · explained in 3 episodes
• Learning with Errors — 2× in this paper
• Quantization — 2× in this paper · explained in 12 episodes
• Activation Function — 1× in this paper · explained in 1 episode
• Core — 1× in this paper · explained in 2 episodes
• Mixture of Experts — 1× in this paper · explained in 2 episodes
• Ring Learning with Errors — 1× in this paper
• Transformer Models — 1× in this paper · explained in 2 episodes
• Verifier — 1× in this paper · explained in 2 episodes

Sources

• Training Verifiers to Solve Math Word Problems
• Trinity: A General Purpose FHE Accelerator
• Large Language Model Agent in Financial Trading: A Survey
• Orion: A Fully Homomorphic Encryption Framework for Deep Learning
• The Llama 3 Herd of Models
• NeuJeans: Private Neural Network Inference with Joint Optimization of Convolution and FHE Bootstrapping
• Gazelle: A Low Latency Framework for Secure Neural Network Inference
• MPCFormer: fast, performant and private Transformer inference with MPC
• Pointer Sentinel Mixture Models
• Llama 2: Open Foundation and Fine-Tuned Chat Models
• Efficient Streaming Language Models with Attention Sinks
• HEQuant: Marrying Homomorphic Encryption and Quantization for Communication-Efficient Private Inference
• Falcon: Accelerating Homomorphically Encrypted Convolutions for Efficient Private Mobile Network Inference
• PrivCirNet: Efficient Private Inference via Block Circulant Transformation
• PrivQuant: Communication-Efficient Private Inference with Quantized Network/Protocol Co-Optimization
• Cachemir: Fully Homomorphic Encrypted Inference of Generative Large Language Model with KV Cache
• EQO: Exploring Ultra-Efficient Private Inference with Winograd-Based Protocol and Quantization Co-Optimization
• TinyLlama: An Open-Source Small Language Model

Related papers

• SoK: AI-Augmented Binary Reversing
• Relaxed Sender Anonymity for CBDC Interbank Settlement: A Zero-Knowledge Approach on Permissioned EVM
• Calibration-Family Overfit: Why Trusted Sabotage Monitors Don't Transfer Across Lineages
• Efficient Fuzzy PSI under One-Sided Assumptions
• Sealing the Audit-Runtime Gap for LLM Skills
• Token Composition: A Graph Based on EVM Logs